FUSE: ignore setxattr for some keys from gsyncd aux mount

Context ------- gsyncd/geo-rep plans to rely on Rsync to sync extended attributes. When this is in place, all xattrs *visible* on the mount point would be candidate for syncing. This set could include gluster internal xattrs too (as xome xlators do not filter out in their cbks). Syncing these xattrs to the slave could result in unexpected functioning of the slave mount. Soln. ----- For gsyncd auxillary mounts (identified by client_pid -1), we only allow xtime related xattrs to go through and silently ignore (w/o propagating error back to the client) the rest of them. This provides a future proof solution as we need not worry about what xattrs show up on the mounts. Also, 'user' namespace xattrs are always passed through even if it's from a gsyncd aux mount. Signed-off-by: Venky Shankar <vshankar@redhat.com> Change-Id: I6fac5e03d2b25fa4cdece4b2897fb202617b3c23 BUG: 841062 Reviewed-on: http://review.gluster.com/3687 Tested-by: Gluster Build System <jenkins@build.gluster.com> Reviewed-by: Vijay Bellur <vbellur@redhat.com>
author: Venky Shankar <vshankar@redhat.com> 2012-07-17 22:54:21 +0530
committer: Vijay Bellur <vbellur@redhat.com> 2012-07-19 02:23:57 -0700
commit: 84a735558d2682446b2e45d97ff97e698ab2d256 (patch)
tree: 90733b6d292a81b3340e758187f1055d67472a4d /libglusterfs
parent: 20fedc3232e876917f963b28a4f4b3bf469b2e6f (diff)
0 files changed, 0 insertions, 0 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 3fcf53d8153..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"
@@ -45,4794 +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;
+    int i = 0;
 
-        priv   = this->private;
-
-        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;
-
-        GF_ASSERT (gfid_req);
+afr_fav_child_reset_sink_xattrs(void *opaque);
 
-        *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;
+
+    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);
+    }
+
+    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_rm_stale_children (afr_inode_ctx_t *ctx, int32_t *stale_children,
-                                 int32_t child_count)
+static int
+__afr_mark_pending_lk_heal(xlator_t *this, afr_private_t *priv, int child)
 {
-        int             i            = 0;
-        int32_t         read_child   = -1;
+    afr_lk_heal_info_t *info = NULL;
+    afr_lk_heal_info_t *tmp = NULL;
 
-        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]);
+    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;
 }
 
-void
-afr_inode_ctx_set_opendir_done (afr_inode_ctx_t *ctx)
+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;
+}
+
+gf_boolean_t
+afr_is_lock_mode_mandatory(dict_t *xdata)
 {
-        uint64_t        remaining_mask = 0;
-        uint64_t        mask = 0;
+    int ret = 0;
+    uint32_t lk_mode = GF_LK_ADVISORY;
 
-        remaining_mask = (~AFR_ICTX_OPENDIR_DONE_MASK & ctx->masks);
-        mask = (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_OPENDIR_DONE_MASK);
-        ctx->masks = remaining_mask | mask;
+    ret = dict_get_uint32(xdata, GF_LOCK_MODE, &lk_mode);
+    if (!ret && lk_mode == GF_LK_MANDATORY)
+        return _gf_true;
+
+    return _gf_false;
 }
 
-void
-afr_inode_set_ctx_params (xlator_t *this, inode_t *inode,
-                          afr_inode_params_t *params)
+call_frame_t *
+afr_copy_frame(call_frame_t *base)
 {
-        GF_ASSERT (inode);
-        GF_ASSERT (params);
+    afr_local_t *local = NULL;
+    call_frame_t *frame = NULL;
+    int op_errno = 0;
 
-        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;
+    frame = copy_frame(base);
+    if (!frame)
+        return NULL;
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local) {
+        AFR_STACK_DESTROY(frame);
+        return NULL;
+    }
 
-        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 frame;
 }
 
-void
-afr_set_split_brain (xlator_t *this, inode_t *inode, afr_spb_state_t mdata_spb,
-                     afr_spb_state_t data_spb)
-{
-        afr_inode_ctx_t *ctx = NULL;
+/* 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;
+    }
 
-        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;
+    return _gf_false;
 }
 
-void
-afr_set_opendir_done (xlator_t *this, inode_t *inode)
+static void
+afr_inode_ctx_destroy(afr_inode_ctx_t *ctx)
 {
-        afr_inode_params_t params = {0};
+    int i = 0;
+
+    if (!ctx)
+        return;
 
-        params.op = AFR_INODE_SET_OPENDIR_DONE;
-        afr_inode_set_ctx_params (this, inode, &params);
+    for (i = 0; i < AFR_NUM_CHANGE_LOGS; i++) {
+        GF_FREE(ctx->pre_op_done[i]);
+    }
+
+    GF_FREE(ctx);
 }
 
-void
-afr_inode_set_read_ctx (xlator_t *this, inode_t *inode, int32_t read_child,
-                        int32_t *fresh_children)
-{
-        afr_inode_params_t params = {0};
-        afr_private_t      *priv  = NULL;
+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;
+    }
 
-        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));
+    ictx = GF_CALLOC(1, sizeof(afr_inode_ctx_t), gf_afr_mt_inode_ctx_t);
+    if (!ictx)
+        goto out;
 
-        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);
+    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;
 }
 
-void
-afr_inode_rm_stale_children (xlator_t *this, inode_t *inode,
-                             int32_t *stale_children)
-{
-        afr_inode_params_t params = {0};
+/*
+ * 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);
 
-        GF_FREE (sh->buf);
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_ctx_get(this, inode, &ctx);
+        if (ret)
+            goto unlock;
 
-        GF_FREE (sh->parentbufs);
+        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);
+
+    if (event_gen1 != event_gen2)
+        need_refresh = _gf_true;
+out:
+    return need_refresh;
+}
+
+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;
+    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;
+    }
 
-        GF_FREE (local->internal_lock.inode_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;
+    }
+
+    local->refreshinode = inode_ref(inode);
 
-        return call_count;
+    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;
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int ret = 0;
 
-        for (i = 0; i < child_count; i++)
-                if (elems[i])
-                        ret++;
-        return ret;
-}
-
-/**
- * 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;
-
-        priv = this->private;
+    int ret = -1;
 
-        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_WARNING,
-                        "%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_WARNING, "%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_WARNING, "%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);
         }
-
-        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;
+    } else {
+    cant_interpret:
+        afr_attempt_readsubvol_set(frame, this, success_replies, readable,
+                                   &read_subvol);
+        if (read_subvol == -1) {
+            goto error;
         }
+    }
 
-out:
-        STACK_DESTROY(frame->root);
-        return 0;
-}
-
-static void
-afr_attempt_local_discovery (xlator_t *this, int32_t child_index)
-{
-        call_frame_t    *newframe = NULL;
-        loc_t            tmploc = {0,};
-        afr_private_t   *priv = this->private;
+    afr_handle_quota_size(frame, this);
 
-        newframe = create_frame(this,this->ctx->pool);
-        if (!newframe) {
-                return;
+    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;
         }
+    }
 
-        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);
-}
-
-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_private_t   *priv   = this->private;
+    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;
 
-        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);
+error:
+    AFR_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno, NULL, NULL,
+                     NULL, NULL);
+}
 
-        afr_lookup_cache_args (local, child_index, xattr,
-                               buf, postparent);
+/*
+ * 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
+ */
 
-        if (local->do_discovery && (priv->read_child == (-1))) {
-                afr_attempt_local_discovery(this,child_index);
-        }
+int
+afr_higher_errno(int32_t old_errno, int32_t new_errno)
+{
+    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;
 
-        local->cont.lookup.success_children[local->success_count] = child_index;
-        local->success_count++;
+    return new_errno;
 }
 
 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)
+afr_final_errno(afr_local_t *local, afr_private_t *priv)
 {
-        afr_local_t *   local = NULL;
-        int             call_count      = -1;
-        int             child_index     = -1;
+    int i = 0;
+    int op_errno = 0;
+    int tmp_errno = 0;
 
-         child_index = (long) cookie;
+    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);
+    }
 
-        LOCK (&frame->lock);
-        {
-                local = frame->local;
-
-                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);
-
-         }
-unlock:
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-        if (call_count == 0) {
-               afr_lookup_done (frame, this);
-        }
-
-         return 0;
+    return op_errno;
 }
 
-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 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);
 
-        iatts = GF_CALLOC (child_count, sizeof (*iatts), gf_afr_mt_iatt);
-        if (NULL == iatts)
-                goto out;
-        local->cont.lookup.postparents = iatts;
+        priv->read_child = child_index;
+    }
+out:
+    STACK_DESTROY(frame->root);
+    return 0;
+}
 
-        iatts = GF_CALLOC (child_count, sizeof (*iatts), gf_afr_mt_iatt);
-        if (NULL == iatts)
-                goto out;
-        local->cont.lookup.bufs = iatts;
+static void
+afr_attempt_local_discovery(xlator_t *this, int32_t child_index)
+{
+    call_frame_t *newframe = NULL;
+    loc_t tmploc = {
+        0,
+    };
+    afr_private_t *priv = this->private;
 
-        success_children = afr_children_create (child_count);
-        if (NULL == success_children)
-                goto out;
-        local->cont.lookup.success_children = success_children;
+    newframe = create_frame(this, this->ctx->pool);
+    if (!newframe) {
+        return;
+    }
 
-        local->fresh_children = afr_children_create (child_count);
-        if (NULL == local->fresh_children)
-                goto out;
+    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);
+}
 
-        sources = GF_CALLOC (sizeof (*sources), child_count, gf_afr_mt_int32_t);
-        if (NULL == sources)
-                goto out;
-        local->cont.lookup.sources = sources;
+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);
 
-        pending_matrix = afr_matrix_create (child_count, child_count);
-        if (NULL == pending_matrix)
-                goto out;
-        local->cont.lookup.pending_matrix = pending_matrix;
+    if (dict)
+        dict_unref(dict);
 
-        ret = 0;
-out:
-        return ret;
+    return 0;
 }
 
-int
-afr_lookup (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, dict_t *xattr_req)
+gf_boolean_t
+afr_is_pending_set(xlator_t *this, dict_t *xdata, int type)
 {
-        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;
-
-        priv = this->private;
+    int idx = -1;
+    afr_private_t *priv = NULL;
+    void *pending_raw = NULL;
+    int *pending_int = NULL;
+    int i = 0;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (local, out);
+    priv = this->private;
+    idx = afr_index_for_transaction_type(type);
 
-        local->op_ret = -1;
+    if (dict_get_ptr(xdata, AFR_DIRTY, &pending_raw) == 0) {
+        if (pending_raw) {
+            pending_int = pending_raw;
 
-        frame->local = local;
-        local->fop = GF_FOP_LOOKUP;
-
-        loc_copy (&local->loc, loc);
-        ret = loc_path (&local->loc, NULL);
-        if (ret < 0) {
-                op_errno = EINVAL;
-                goto out;
+            if (ntoh32(pending_int[idx]))
+                return _gf_true;
         }
+    }
 
-        ret = inode_ctx_get (local->loc.inode, this, &ctx);
-        if (ret == 0) {
-                /* lookup is a revalidate */
+    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;
 
-                local->read_child_index = afr_inode_get_read_ctx (this,
-                                                               local->loc.inode,
-                                                               NULL);
-        } 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 (ntoh32(pending_int[idx]))
+            return _gf_true;
+    }
 
-        local->child_up = memdup (priv->child_up,
-                                  sizeof (*local->child_up) * priv->child_count);
-        if (NULL == local->child_up) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    return _gf_false;
+}
 
-        ret = afr_lookup_cont_init (local, priv->child_count);
-        if (ret < 0) {
-                op_errno = -ret;
-                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;
 
-        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;
+    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;
         }
 
-        /* By default assume ENOTCONN. On success it will be set to 0. */
-        local->op_errno = ENOTCONN;
+        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;
+        }
 
-        ret = afr_lookup_xattr_req_prepare (local, this, xattr_req, &local->loc,
-                                            &gfid_req);
-        if (ret) {
-                local->op_errno = -ret;
-                goto out;
+        if (gf_uuid_compare(stbuf.ia_gfid, replies[i].poststat.ia_gfid)) {
+            start = _gf_false;
+            break;
         }
-        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 (!IA_EQUAL(stbuf, replies[i].poststat, type)) {
+            start = _gf_false;
+            break;
         }
-        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;
-                }
+
+        /*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;
         }
 
-        ret = 0;
-out:
-        if (ret)
-                AFR_STACK_UNWIND (lookup, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL);
+        /*Check if xattrs need heal*/
+        if (!afr_xattrs_are_equal(replies[first].xdata, replies[i].xdata))
+            start = _gf_true;
+    }
 
-        return 0;
+    return start;
 }
 
+int
+afr_lookup_metadata_heal_check(call_frame_t *frame, xlator_t *this)
 
-/* {{{ open */
+{
+    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:
+    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_fd_ctx_set (xlator_t *this, fd_t *fd)
+afr_lookup_selfheal_wrap(void *opaque)
 {
-        afr_private_t * priv   = NULL;
-        int             ret    = -1;
-        uint64_t        ctx    = 0;
-        afr_fd_ctx_t *  fd_ctx = NULL;
+    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,
+    };
 
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (fd, out);
+    local = frame->local;
+    this = frame->this;
+    loc_pargfid(&local->loc, pargfid);
 
-        priv = this->private;
+    ret = afr_selfheal_name(frame->this, pargfid, local->loc.name,
+                            &local->cont.lookup.gfid_req, local->xattr_req);
+    if (ret == -EIO)
+        goto unwind;
 
-        ret = __fd_ctx_get (fd, this, &ctx);
+    afr_local_replies_wipe(local, this->private);
 
-        if (ret == 0)
-                goto out;
+    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);
 
-        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_lookup_metadata_heal_check(frame, this);
+    return 0;
+
+unwind:
+    AFR_STACK_UNWIND(lookup, frame, -1, EIO, NULL, NULL, NULL, NULL);
+    return 0;
+}
+
+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 {
+            if ((replies[i].op_errno != ENOTCONN) &&
+                (replies[i].op_errno != ENOENT) &&
+                (replies[i].op_errno != ESTALE)) {
+                op_errno = replies[i].op_errno;
+            }
         }
 
-        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;
+        /*gfid is missing, needs heal*/
+        if ((replies[i].op_ret == -1) && (replies[i].op_errno == ENODATA)) {
+            goto name_heal;
         }
 
-        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 (first == -1) {
+            first = i;
+            continue;
         }
 
-        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;
+        if (replies[i].op_ret != replies[first].op_ret) {
+            name_state_mismatch = _gf_true;
         }
 
-        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;
+        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;
+            }
         }
+    }
 
-        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;
+    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 (!replies[i].valid)
+                continue;
+            if (par_readables[i] && replies[i].op_ret < 0 &&
+                replies[i].op_errno != ENOTCONN) {
+                goto name_heal;
+            }
         }
+    }
 
-        fd_ctx->up_count   = priv->up_count;
-        fd_ctx->down_count = priv->down_count;
+    goto metadata_heal;
 
-        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;
-        }
+name_heal:
+    heal = afr_frame_create(this, NULL);
+    if (!heal)
+        goto metadata_heal;
 
-	pthread_mutex_init (&fd_ctx->delay_lock, NULL);
-        INIT_LIST_HEAD (&fd_ctx->entries);
-        fd_ctx->call_child = -1;
+    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;
 
-        INIT_LIST_HEAD (&fd_ctx->eager_locked);
+metadata_heal:
+    ret = afr_lookup_metadata_heal_check(frame, this);
 
-        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;
+    return ret;
 }
 
-
 int
-afr_fd_ctx_set (xlator_t *this, fd_t *fd)
-{
-        int ret = -1;
-
-        LOCK (&fd->lock);
-        {
-                ret = __afr_fd_ctx_set (this, fd);
-        }
-        UNLOCK (&fd->lock);
-
-        return ret;
+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;
 }
 
-/* {{{ 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)
+static void
+afr_discover_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        int call_count  = -1;
+    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;
 
-        local = frame->local;
+    priv = this->private;
+    local = frame->local;
+    data_readable = alloca0(priv->child_count);
+    success_replies = alloca0(priv->child_count);
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                        }
-                        local->success_count++;
-                }
+    afr_fill_success_replies(local, priv, success_replies);
+    if (AFR_COUNT(success_replies, priv->child_count) > 0)
+        local->op_ret = 0;
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    if (local->op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(frame->local, this->private);
+        goto error;
+    }
 
-	call_count = afr_frame_return (frame);
+    if (!afr_has_quorum(success_replies, this, frame))
+        goto unwind;
 
-	if (call_count == 0)
-		AFR_STACK_UNWIND(flush, frame, local->op_ret,
-				 local->op_errno, NULL);
+    ret = afr_replies_interpret(frame, this, local->inode, NULL);
+    if (ret) {
+        afr_inode_need_refresh_set(local->inode, this);
+    }
 
-        return 0;
-}
+    read_subvol = afr_read_subvol_decide(local->inode, this, NULL,
+                                         data_readable);
 
-int
-afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-        afr_local_t   *local = NULL;
-        int            ret        = -1;
-        int            op_errno   = 0;
-	int	       call_count = -1;
-	int	       i = 0;
+unwind:
+    afr_attempt_readsubvol_set(frame, this, success_replies, data_readable,
+                               &read_subvol);
+    if (read_subvol == -1)
+        goto error;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    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);
+    }
 
-        priv = this->private;
+    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;
 
-	AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-	local = frame->local;
+error:
+    AFR_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno, NULL, NULL,
+                     NULL, NULL);
+}
 
-	ret = afr_local_init(local, priv, &op_errno);
-	if (ret < 0)
-		goto out;
+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);
+    }
 
-	local->fd = fd_ref(fd);
-        call_count = local->call_count;
+out:
+    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);
 
-	afr_delayed_changelog_wake_up (this, fd);
+    return 0;
+}
 
-        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;
-                }
-        }
+static int
+afr_ta_id_file_check_cbk(int ret, call_frame_t *ta_frame, void *opaque)
+{
+    return 0;
+}
 
-	ret = 0;
+static void
+afr_discover_done(call_frame_t *frame, xlator_t *this)
+{
+    int ret = 0;
+    afr_private_t *priv = NULL;
 
-out:
-	if (ret < 0)
-		AFR_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+    priv = this->private;
+    if (!priv->thin_arbiter_count)
+        goto unwind;
+    if (!gf_uuid_is_null(priv->ta_gfid))
+        goto unwind;
 
-        return 0;
+    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);
 }
 
-/* }}} */
-
-
 int
-afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd)
+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)
 {
-        uint64_t        ctx = 0;
-        afr_fd_ctx_t    *fd_ctx = NULL;
-        int             ret = 0;
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int child_index = -1;
+    GF_UNUSED int ret = 0;
+    int8_t need_heal = 1;
 
-        ret = fd_ctx_get (fd, this, &ctx);
-        if (ret < 0)
-                goto out;
+    child_index = (long)cookie;
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+    local = frame->local;
 
-        if (fd_ctx) {
-                GF_FREE (fd_ctx->pre_op_done);
+    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);
+    }
 
-                GF_FREE (fd_ctx->opened_on);
+    if (local->do_discovery && (op_ret == 0))
+        afr_attempt_local_discovery(this, child_index);
 
-                GF_FREE (fd_ctx->locked_on);
+    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;
+    }
 
-                GF_FREE (fd_ctx->pre_op_piggyback);
-                GF_FREE (fd_ctx->lock_piggyback);
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        afr_set_need_heal(this, local);
+        afr_lookup_metadata_heal_check(frame, this);
+    }
 
-                GF_FREE (fd_ctx->lock_acquired);
-
-		pthread_mutex_destroy (&fd_ctx->delay_lock);
+    return 0;
+}
 
-                GF_FREE (fd_ctx);
+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;
 out:
-        return 0;
+    AFR_STACK_UNWIND(lookup, frame, -1, local->op_errno, 0, 0, 0, 0);
+    return 0;
 }
 
-
 int
-afr_release (xlator_t *this, fd_t *fd)
+afr_discover(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-        afr_locked_fd_t *locked_fd = NULL;
-        afr_locked_fd_t *tmp       = NULL;
-        afr_private_t   *priv      = NULL;
+    int op_errno = ENOMEM;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int event = 0;
 
-        priv = this->private;
+    priv = this->private;
 
-        afr_cleanup_fd_ctx (this, fd);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        list_for_each_entry_safe (locked_fd, tmp, &priv->saved_fds,
-                                  list) {
+    if (!local->call_count) {
+        op_errno = ENOTCONN;
+        goto out;
+    }
 
-                if (locked_fd->fd == fd) {
-                        list_del_init (&locked_fd->list);
-                        GF_FREE (locked_fd);
-                }
+    if (__is_root_gfid(loc->inode->gfid)) {
+        if (!priv->root_inode)
+            priv->root_inode = inode_ref(loc->inode);
 
+        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;
         }
+    }
+
+    local->op = GF_FOP_LOOKUP;
 
+    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;
+}
 
-/* {{{ fsync */
+/*
+ * 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_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_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-        afr_local_t *local = NULL;
-        int call_count = -1;
-        int child_index = (long) cookie;
-        int read_child  = 0;
-	call_stub_t *stub = NULL;
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
+    int event = 0;
+    int ret = 0;
 
-        local = frame->local;
+    if (loc_is_nameless(loc)) {
+        if (xattr_req)
+            dict_del_sizen(xattr_req, "gfid-req");
+        afr_discover(frame, this, loc, xattr_req);
+        return 0;
+    }
 
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+    if (afr_is_private_directory(this->private, loc->parent->gfid, loc->name,
+                                 frame->root->pid)) {
+        op_errno = EPERM;
+        goto out;
+    }
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-                if (op_ret == 0) {
-                        local->op_ret = 0;
+    if (!local->call_count) {
+        op_errno = ENOTCONN;
+        goto out;
+    }
 
-                        if (local->success_count == 0) {
-                                local->cont.fsync.prebuf  = *prebuf;
-                                local->cont.fsync.postbuf = *postbuf;
-                        }
+    local->op = GF_FOP_LOOKUP;
 
-                        if (child_index == read_child) {
-                                local->cont.fsync.prebuf  = *prebuf;
-                                local->cont.fsync.postbuf = *postbuf;
-                        }
+    loc_copy(&local->loc, loc);
 
-                        local->success_count++;
-                }
+    local->inode = inode_ref(loc->inode);
 
-                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, default_fsync_cbk,
-                                           local->op_ret, local->op_errno,
-                                           &local->cont.fsync.prebuf,
-                                           &local->cont.fsync.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);
+    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");
+        }
+    }
 
-        return 0;
-}
-
+    afr_read_subvol_get(loc->parent, this, NULL, NULL, &event,
+                        AFR_DATA_TRANSACTION, NULL);
 
-int
-afr_fsync (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 ret = -1;
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_errno = 0;
+    afr_lookup_do(frame, this, 0);
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    return 0;
+out:
+    AFR_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
 
-        priv = this->private;
+    return 0;
+}
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+void
+_afr_cleanup_fd_ctx(xlator_t *this, afr_fd_ctx_t *fd_ctx)
+{
+    afr_private_t *priv = this->private;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    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;
+}
 
-        call_count = local->call_count;
+int
+afr_cleanup_fd_ctx(xlator_t *this, fd_t *fd)
+{
+    uint64_t ctx = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int ret = 0;
 
-        local->fd             = fd_ref (fd);
+    ret = fd_ctx_get(fd, this, &ctx);
+    if (ret < 0)
+        goto out;
 
-	if (afr_fd_has_witnessed_unstable_write (this, fd)) {
-		/* don't care. we only wanted to CLEAR the bit */
-	}
+    fd_ctx = (afr_fd_ctx_t *)(long)ctx;
 
-        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 (fd_ctx) {
+        _afr_cleanup_fd_ctx(this, fd_ctx);
+    }
 
-        ret = 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
+    return 0;
 }
 
-/* }}} */
+int
+afr_release(xlator_t *this, fd_t *fd)
+{
+    afr_cleanup_fd_ctx(this, fd);
 
-/* {{{ fallocate */
+    return 0;
+}
 
-static int
-afr_fallocate_unwind (call_frame_t *frame, xlator_t *this)
+afr_fd_ctx_t *
+__afr_fd_ctx_get(fd_t *fd, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    uint64_t ctx = 0;
+    int ret = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-        local = frame->local;
+    ret = __fd_ctx_get(fd, this, &ctx);
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    if (ret < 0) {
+        ret = __afr_fd_ctx_set(this, fd);
+        if (ret < 0)
+            goto out;
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (fallocate, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.fallocate.prebuf,
-                                  &local->cont.fallocate.postbuf,
-                                  NULL);
-        }
-        return 0;
+        ret = __fd_ctx_get(fd, this, &ctx);
+        if (ret < 0)
+            goto out;
+    }
+
+    fd_ctx = (afr_fd_ctx_t *)(long)ctx;
+out:
+    return fd_ctx;
 }
 
-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_fd_ctx_t *
+afr_fd_ctx_get(fd_t *fd, 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_fd_ctx_t *fd_ctx = NULL;
 
-        local = frame->local;
-        priv  = this->private;
+    LOCK(&fd->lock);
+    {
+        fd_ctx = __afr_fd_ctx_get(fd, this);
+    }
+    UNLOCK(&fd->lock);
 
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+    return fd_ctx;
+}
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+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;
+    int i = 0;
 
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
+    VALIDATE_OR_GOTO(this->private, out);
+    VALIDATE_OR_GOTO(fd, out);
 
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.fallocate.prebuf  = *prebuf;
-                                local->cont.fallocate.postbuf = *postbuf;
-                        }
+    priv = this->private;
 
-                        if (child_index == read_child) {
-                                local->cont.fallocate.prebuf  = *prebuf;
-                                local->cont.fallocate.postbuf = *postbuf;
-                        }
+    ret = __fd_ctx_get(fd, this, &ctx);
 
-                        local->success_count++;
+    if (ret == 0)
+        goto out;
 
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    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 (need_unwind)
-                local->transaction.unwind (frame, this);
+    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;
+    }
 
-        call_count = afr_frame_return (frame);
+    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;
+    }
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    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;
 }
 
-static int
-afr_fallocate_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;
+/* {{{ flush */
 
-        local = frame->local;
-        priv = this->private;
+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_local_t *local = NULL;
+    int call_count = -1;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    local = frame->local;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
+    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);
 
-        local->call_count = call_count;
+    if (call_count == 0)
+        AFR_STACK_UNWIND(flush, frame, local->op_ret, local->op_errno,
+                         local->xdata_rsp);
 
-        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;
-                }
+    return 0;
+}
+
+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;
         }
+    }
 
-        return 0;
+    return 0;
 }
 
-static int
-afr_fallocate_done (call_frame_t *frame, xlator_t *this)
+afr_local_t *
+afr_wakeup_same_fd_delayed_op(xlator_t *this, afr_lock_t *lock, fd_t *fd)
 {
-        afr_local_t *local = NULL;
-
-        local = frame->local;
+    afr_local_t *local = NULL;
 
-        local->transaction.unwind (frame, this);
+    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;
+        }
+    }
 
-        AFR_STACK_DESTROY (frame);
+    return local;
+}
 
-        return 0;
+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);
+    }
 }
 
-static int
-afr_do_fallocate(call_frame_t *frame, xlator_t *this)
+int
+afr_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, 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;
-        }
+    afr_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = ENOMEM;
 
-        transaction_frame->local = local;
-        frame->local = 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_FALLOCATE;
+    local->op = GF_FOP_FLUSH;
+    if (!afr_is_consistent_io_possible(local, this->private, &op_errno))
+        goto out;
 
-        local->transaction.fop    = afr_fallocate_wind;
-        local->transaction.done   = afr_fallocate_done;
-        local->transaction.unwind = afr_fallocate_unwind;
+    local->fd = fd_ref(fd);
 
-        local->transaction.main_frame = frame;
+    stub = fop_flush_stub(frame, afr_flush_wrapper, fd, xdata);
+    if (!stub)
+        goto out;
 
-        local->transaction.start   = local->cont.fallocate.offset;
-        local->transaction.len     = 0;
+    afr_delayed_changelog_wake_resume(this, fd->inode, stub);
 
-	/* 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;
-        }
-
-        op_ret = 0;
+    return 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);
-        }
-
-        return 0;
+    AFR_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+    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_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;
-        call_frame_t   *transaction_frame = NULL;
-        int ret = -1;
-        int 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;
-
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
+    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;
         }
-        QUORUM_CHECK(fallocate,out);
+        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;
+}
 
-	local->cont.fallocate.mode = mode;
-        local->cont.fallocate.offset  = offset;
-	local->cont.fallocate.len = len;
+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;
 
-        local->fd = fd_ref (fd);
+    priv = this->private;
 
-        afr_open_fd_fix (fd, this);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        afr_do_fallocate (frame, this);
+    local->op = GF_FOP_FSYNCDIR;
+    if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+        goto out;
 
-        ret = 0;
-out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+    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;
         }
+    }
 
-        return 0;
+    return 0;
+out:
+    AFR_STACK_UNWIND(fsyncdir, frame, -1, op_errno, NULL);
+
+    return 0;
 }
 
 /* }}} */
 
-/* {{{ discard */
-
 static int
-afr_discard_unwind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+afr_serialized_lock_wind(call_frame_t *frame, xlator_t *this);
 
-        local = frame->local;
+static gf_boolean_t
+afr_is_conflicting_lock_present(int32_t op_ret, int32_t op_errno)
+{
+    if (op_ret == -1 && op_errno == EAGAIN)
+        return _gf_true;
+    return _gf_false;
+}
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+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;
+    }
+}
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (discard, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.discard.prebuf,
-                                  &local->cont.discard.postbuf,
-                                  NULL);
-        }
-        return 0;
+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;
+    }
 }
 
-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)
+void
+afr_fop_lock_proceed(call_frame_t *frame)
 {
-        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;
+    afr_private_t *priv = NULL;
 
-        local = frame->local;
-        priv  = this->private;
+    local = frame->local;
+    priv = frame->this->private;
 
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+    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);
+}
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+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)
+
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = -1;
+    int child_index = (long)cookie;
+    uuid_t gfid = {0};
 
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
+    local = frame->local;
+    priv = this->private;
 
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.discard.prebuf  = *prebuf;
-                                local->cont.discard.postbuf = *postbuf;
-                        }
+    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));
+    }
 
-                        if (child_index == read_child) {
-                                local->cont.discard.prebuf  = *prebuf;
-                                local->cont.discard.postbuf = *postbuf;
-                        }
+    call_count = afr_frame_return(frame);
+    if (call_count == 0)
+        afr_fop_lock_proceed(frame);
 
-                        local->success_count++;
+    return 0;
+}
 
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+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;
+    }
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
 
-        call_count = afr_frame_return (frame);
+        if (local->replies[i].op_ret == -1)
+            continue;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+        afr_fop_lock_wind(frame, this, i, afr_unlock_partial_lock_cbk);
 
-        return 0;
+        if (!--call_count)
+            break;
+    }
+
+out:
+    return 0;
 }
 
-static int
-afr_discard_wind (call_frame_t *frame, xlator_t *this)
+int32_t
+afr_fop_lock_done(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
+    int i = 0;
+    int lock_count = 0;
+    unsigned char *success = NULL;
 
-        local = frame->local;
-        priv = this->private;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+    success = alloca0(priv->child_count);
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
+        if (local->replies[i].op_ret == 0) {
+            lock_count++;
+            success[i] = 1;
         }
 
-        local->call_count = call_count;
+        if (local->op_ret == -1 && local->op_errno == EAGAIN)
+            continue;
 
-        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;
-                }
+        if ((local->replies[i].op_ret == -1) &&
+            (local->replies[i].op_errno == EAGAIN)) {
+            local->op_ret = -1;
+            local->op_errno = EAGAIN;
+            continue;
         }
 
-        return 0;
-}
+        if (local->replies[i].op_ret == 0)
+            local->op_ret = 0;
 
-static int
-afr_discard_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
+        local->op_errno = local->replies[i].op_errno;
+    }
 
-        local->transaction.unwind (frame, this);
+    if (afr_fop_lock_is_unlock(frame))
+        goto unwind;
 
-        AFR_STACK_DESTROY (frame);
+    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;
+    return 0;
+unwind:
+    afr_fop_lock_unwind(frame, local->op, local->op_ret, local->op_errno,
+                        local->xdata_rsp);
+    return 0;
 }
 
 static int
-afr_do_discard(call_frame_t *frame, xlator_t *this)
+afr_common_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, 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;
+    int child_index = (long)cookie;
 
-        local = frame->local;
+    local = frame->local;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
+    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;
+}
 
-        transaction_frame->local = local;
-        frame->local = NULL;
+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)
 
-        local->op = GF_FOP_DISCARD;
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int child_index = (long)cookie;
+    int next_child = 0;
 
-        local->transaction.fop    = afr_discard_wind;
-        local->transaction.done   = afr_discard_done;
-        local->transaction.unwind = afr_discard_unwind;
+    local = frame->local;
+    priv = this->private;
 
-        local->transaction.main_frame = frame;
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
 
-        local->transaction.start   = local->cont.discard.offset;
-        local->transaction.len     = 0;
+    for (next_child = child_index + 1; next_child < priv->child_count;
+         next_child++) {
+        if (local->child_up[next_child])
+            break;
+    }
 
-        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
-        if (op_ret < 0) {
-            op_errno = -op_ret;
-            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);
+    }
 
-        op_ret = 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);
-        }
-
-        return 0;
+    return 0;
 }
 
-int
-afr_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
-	    size_t len, dict_t *xdata)
+static int
+afr_serialized_lock_wind(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;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
 
-        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;
-
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
+    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;
         }
-        QUORUM_CHECK(discard, out);
+    }
+    return 0;
+}
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+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)
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+{
+    int call_count = 0;
 
-        local->cont.discard.offset  = offset;
-	local->cont.discard.len = len;
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
 
-        local->fd = fd_ref (fd);
+    call_count = afr_frame_return(frame);
+    if (call_count == 0)
+        afr_fop_lock_done(frame, this);
 
-        afr_open_fd_fix (fd, this);
+    return 0;
+}
 
-        afr_do_discard(frame, this);
+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;
 
-        ret = 0;
-out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL);
-        }
+    priv = this->private;
+    local = frame->local;
+    call_count = local->call_count;
 
-        return 0;
+    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;
 }
 
-/* }}} */
-
-/* {{{ fsync */
-
-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)
+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 (fsyncdir, 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_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              int32_t datasync, 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_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_handle_inodelk(frame, this, GF_FOP_FINODELK, volume, NULL, fd, cmd,
+                       flock, xdata);
+    return 0;
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+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_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;
+}
 
-        priv = this->private;
+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;
+}
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+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;
+}
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+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;
 
-        call_count = local->call_count;
+    local = frame->local;
 
-        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;
+    LOCK(&frame->lock);
+    {
+        if (op_ret != 0) {
+            local->op_errno = op_errno;
+            goto unlock;
+        }
+
+        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 (fsyncdir, 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;
+        }
+    }
 
-/* {{{ xattrop */
+    return 0;
+out:
+    AFR_STACK_UNWIND(statfs, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
 
 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)
+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) {
-                        if (!local->cont.xattrop.xattr)
-                                local->cont.xattrop.xattr = dict_ref (xattr);
-                        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 (xattrop, frame, local->op_ret, local->op_errno,
-                local->cont.xattrop.xattr, 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_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             gf_xattrop_flags_t optype, dict_t *xattr, 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_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->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 (xattrop, frame, -1, op_errno, NULL, NULL);
-        return 0;
+    return 0;
 }
 
-/* }}} */
-
-/* {{{ fxattrop */
-
 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_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;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int child_index = -1;
 
-        int call_count = -1;
+    local = frame->local;
+    priv = this->private;
 
-        local = frame->local;
+    child_index = (long)cookie;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0) {
-                        if (!local->cont.fxattrop.xattr)
-                                local->cont.fxattrop.xattr = dict_ref (xattr);
+    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;
 
-                        local->op_ret = 0;
-                }
+        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);
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+        afr_lk_unlock(frame, this);
+    } else {
+        if (local->op_ret < 0)
+            local->op_errno = afr_final_errno(local, priv);
 
-        call_count = afr_frame_return (frame);
+        AFR_STACK_UNWIND(lk, frame, local->op_ret, local->op_errno,
+                         &local->cont.lk.ret_flock, local->xdata_rsp);
+    }
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (fxattrop, frame, local->op_ret, local->op_errno,
-                                  local->cont.fxattrop.xattr, xdata);
+    return 0;
+}
 
-        return 0;
+int
+afr_lk_transaction_cbk(int ret, call_frame_t *frame, void *opaque)
+{
+    return 0;
 }
 
+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;
+}
 
-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;
+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);
+    }
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    return 0;
 
-        priv = this->private;
+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;
+}
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+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;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    priv = this->private;
 
-        call_count = local->call_count;
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        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;
-                }
+    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 (fxattrop, frame, -1, op_errno, NULL, 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_inodelk_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;
+    afr_local_t *local = NULL;
+    int call_count = -1;
 
-        local = frame->local;
+    local = frame->local;
+    call_count = afr_frame_return(frame);
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+    if (call_count == 0)
+        AFR_STACK_UNWIND(lease, frame, local->op_ret, local->op_errno, lease,
+                         xdata);
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0)
-                AFR_STACK_UNWIND (inodelk, frame, local->op_ret,
-                                  local->op_errno, xdata);
-
-        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_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_inodelk_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->inodelk,
-                                    volume, loc, cmd, flock, 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 (inodelk, frame, -1, op_errno, NULL);
-        return 0;
+    return 0;
 }
 
-
 int32_t
-afr_finodelk_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno,
-                  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 = -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_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 (finodelk, frame, local->op_ret,
-                                  local->op_errno, 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_finodelk (call_frame_t *frame, xlator_t *this,
-              const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock,
-              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;
-        afr_local_t *local  = NULL;
-        int ret = -1;
-        int i = 0;
-        int32_t 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 (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;
+    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 < 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;
-                }
-        }
+    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 (finodelk, frame, -1, op_errno, NULL);
-        return 0;
+    AFR_STACK_UNWIND(lease, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
 }
 
+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;
+    }
 
-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_local_t *local = NULL;
-        int call_count = -1;
+    AFR_STACK_UNWIND(ipc, frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
 
-        local = frame->local;
+out:
+    return 0;
+}
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+int
+afr_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = -1;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int call_cnt = -1;
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
 
-        call_count = afr_frame_return (frame);
+    if (op != GF_IPC_TARGET_UPCALL)
+        goto wind_default;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (entrylk, frame, local->op_ret,
-                                  local->op_errno, xdata);
+    VALIDATE_OR_GOTO(this->private, err);
+    priv = this->private;
 
-        return 0;
-}
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto err;
 
+    call_cnt = local->call_count;
 
-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)
-{
-        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;
+    if (xdata) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (dict_set_int8(xdata, priv->pending_key[i], 0) < 0)
+                goto err;
+        }
+    }
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->child_up[i])
+            continue;
 
-        priv = this->private;
+        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;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+err:
+    if (op_errno == -1)
+        op_errno = errno;
+    AFR_STACK_UNWIND(ipc, frame, -1, op_errno, NULL);
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    return 0;
 
-        call_count = local->call_count;
+wind_default:
+    STACK_WIND(frame, default_ipc_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ipc, op, xdata);
+    return 0;
+}
 
-        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_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t ctx_int = 0;
+    afr_inode_ctx_t *ctx = NULL;
 
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (entrylk, frame, -1, op_errno, NULL);
+    afr_spb_choice_timeout_cancel(this, inode);
+    inode_ctx_del(inode, this, &ctx_int);
+    if (!ctx_int)
         return 0;
-}
 
+    ctx = (afr_inode_ctx_t *)(uintptr_t)ctx_int;
+    afr_inode_ctx_destroy(ctx);
+    return 0;
+}
 
+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;
+}
 
-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)
+/**
+ * find_child_index - find the child's index in the array of subvolumes
+ * @this: AFR
+ * @child: child
+ */
 
+static int
+afr_find_child_index(xlator_t *this, xlator_t *child)
 {
-        afr_local_t *local = NULL;
-        int call_count = -1;
+    afr_private_t *priv = NULL;
+    int child_count = -1;
+    int i = -1;
 
-        local = frame->local;
+    priv = this->private;
+    child_count = priv->child_count;
+    if (priv->thin_arbiter_count) {
+        child_count++;
+    }
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+    for (i = 0; i < child_count; i++) {
+        if ((xlator_t *)child == priv->children[i])
+            break;
+    }
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    return i;
+}
 
-        call_count = afr_frame_return (frame);
+int
+__afr_get_up_children_count(afr_private_t *priv)
+{
+    int up_children = 0;
+    int i = 0;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (fentrylk, frame, local->op_ret,
-                                  local->op_errno, xdata);
+    for (i = 0; i < priv->child_count; i++)
+        if (priv->child_up[i] == 1)
+            up_children++;
 
-        return 0;
+    return up_children;
 }
 
-
-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)
+static int
+__get_heard_from_all_status(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_private_t *priv = this->private;
+    int i;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    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;
+}
 
-        priv = this->private;
+glusterfs_event_t
+__afr_transform_event_from_state(xlator_t *this)
+{
+    int i = 0;
+    int up_children = 0;
+    afr_private_t *priv = this->private;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    if (__get_heard_from_all_status(this))
+        /* have_heard_from_all. Let afr_notify() do the propagation. */
+        return GF_EVENT_MAXVAL;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    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;
+    }
 
-        call_count = local->call_count;
+    if (up_children)
+        /* We received at least one child up */
+        return GF_EVENT_CHILD_UP;
+    else
+        return GF_EVENT_CHILD_DOWN;
 
-        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;
-                }
-        }
+    return GF_EVENT_MAXVAL;
+}
 
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (fentrylk, frame, -1, op_errno, NULL);
-        return 0;
+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);
 }
 
-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)
+static void
+__afr_launch_notify_timer(xlator_t *this, afr_private_t *priv)
 {
-        afr_local_t *local = NULL;
-        int call_count = 0;
+    struct timespec delay = {
+        0,
+    };
 
-        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;
-                        }
-                }
+    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");
+    }
+}
 
-                if (op_ret == -1)
-                        local->op_errno = op_errno;
+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;
+
+    priv = this->private;
 
+    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];
         }
-        UNLOCK (&frame->lock);
+    }
+    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;
+}
 
-        call_count = afr_frame_return (frame);
+int
+find_worst_up_child(xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    int i = -1;
+    int32_t worst_child = -1;
+    int64_t worst_latency = INT64_MIN;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (statfs, frame, local->op_ret, local->op_errno,
-                                  &local->cont.statfs.buf, xdata);
+    priv = this->private;
 
-        return 0;
+    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;
 }
 
-
-int32_t
-afr_statfs (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, dict_t *xdata)
+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              child_count = 0;
-        afr_local_t   *  local       = NULL;
-        int              i           = 0;
-        int              ret = -1;
-        int              call_count = 0;
-        int32_t          op_errno    = 0;
-
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (loc, out);
+    afr_private_t *priv = NULL;
+    int up_children = 0;
 
-        priv = this->private;
-        child_count = priv->child_count;
+    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;
+    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;
 
-        call_count = local->call_count;
+    up_children = __afr_get_up_children_count(priv);
 
-        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;
-                }
+    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);
         }
-
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (statfs, frame, -1, op_errno, NULL, NULL);
-        return 0;
+    }
 }
 
-
-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)
+static int64_t
+afr_get_halo_latency(xlator_t *this)
 {
-        afr_local_t * local = NULL;
-        int call_count = -1;
+    afr_private_t *priv = NULL;
+    int64_t halo_max_latency_msec = 0;
 
-        local = frame->local;
-        call_count = afr_frame_return (frame);
+    priv = this->private;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  lock, xdata);
+    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;
+}
 
-        return 0;
+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;
+    }
+
+    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;
+}
 
-int32_t
-afr_lk_unlock (call_frame_t *frame, xlator_t *this)
+void
+afr_ta_lock_release_synctask(xlator_t *this)
 {
-        afr_local_t   * local = NULL;
-        afr_private_t * priv  = NULL;
-        int i = 0;
-        int call_count = 0;
+    call_frame_t *ta_frame = NULL;
+    int ret = 0;
 
-        local = frame->local;
-        priv  = this->private;
+    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;
+    }
 
-        call_count = afr_locked_nodes_count (local->cont.lk.locked_nodes,
-                                             priv->child_count);
+    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.");
+    }
+}
 
-        if (call_count == 0) {
-                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  &local->cont.lk.ret_flock, NULL);
-                return 0;
-        }
+static void
+afr_handle_inodelk_contention(xlator_t *this, struct gf_upcall *upcall)
+{
+    struct gf_upcall_inodelk_contention *lc = NULL;
+    unsigned int inmem_count = 0;
+    unsigned int onwire_count = 0;
+    afr_private_t *priv = this->private;
 
-        local->call_count = call_count;
+    lc = upcall->data;
 
-        local->cont.lk.user_flock.l_type = F_UNLCK;
+    if (strcmp(lc->domain, AFR_TA_DOM_NOTIFY) != 0)
+        return;
 
-        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 (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;
 
-        return 0;
+    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_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;
-        afr_private_t *priv = NULL;
-        int child_index = -1;
-/*        int            ret  = 0; */
-
-
-        local = frame->local;
-        priv  = this->private;
+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;
 
-        child_index = (long) cookie;
+    /*
+     * 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);
+        }
+    }
 
-        if (!child_went_down (op_ret, op_errno) && (op_ret == -1)) {
-                local->op_ret   = -1;
-                local->op_errno = op_errno;
+    if (event == GF_EVENT_CHILD_PING) {
+        /* This is the only xlator that handles PING, no reason to
+         * propagate.
+         */
+        goto out;
+    }
 
-                afr_lk_unlock (frame, this);
-                return 0;
+    if (event == GF_EVENT_TRANSLATOR_OP) {
+        LOCK(&priv->lock);
+        {
+            had_heard_from_all = __get_heard_from_all_status(this);
         }
+        UNLOCK(&priv->lock);
 
-        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;
+        if (!had_heard_from_all) {
+            ret = -1;
+        } else {
+            input = data;
+            output = data2;
+            ret = afr_xl_op(this, input, output);
         }
+        goto out;
+    }
 
-        child_index++;
+    if (event == GF_EVENT_UPCALL) {
+        afr_handle_upcall_event(this, data);
+    }
 
-        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 */
+    LOCK(&priv->lock);
+    {
+        had_heard_from_all = __get_heard_from_all_status(this);
+        switch (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;
+                }
+                __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;
 
-                AFR_STACK_UNWIND (lk, frame, -1, ENOTCONN,
-                                  &local->cont.lk.ret_flock, NULL);
-        } else {
-                /* locking has succeeded on all nodes that are up */
+            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;
+                }
+                __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;
 
-                /* 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");
+            case GF_EVENT_CHILD_CONNECTING:
+                priv->last_event[idx] = event;
 
-                   ret = afr_save_locked_fd (this, local->fd);
-                   if (ret)
-                   gf_log (this->name, GF_LOG_DEBUG,
-                   "Could not save locked fd");
+                break;
 
-                */
-                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  &local->cont.lk.ret_flock, NULL);
+            case GF_EVENT_SOME_DESCENDENT_DOWN:
+                priv->last_event[idx] = event;
+                break;
+            default:
+                propagate = 1;
+                break;
         }
+        have_heard_from_all = __get_heard_from_all_status(this);
+        if (!had_heard_from_all && have_heard_from_all) {
+            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;
+                }
 
-        return 0;
+                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 -1;
+}
 
 int
-afr_lk (call_frame_t *frame, xlator_t *this,
-        fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+afr_internal_lock_init(afr_internal_lock_t *lk, size_t child_count)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local = NULL;
-        int i = 0;
-        int32_t op_errno = 0;
-        int     ret      = -1;
+    int ret = -ENOMEM;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    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;
 
-        priv = this->private;
+    lk->lock_op_ret = -1;
+    lk->lock_op_errno = EUCLEAN;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    ret = 0;
+out:
+    return ret;
+}
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+void
+afr_matrix_cleanup(int32_t **matrix, unsigned int m)
+{
+    int i = 0;
 
-        local->cont.lk.locked_nodes = GF_CALLOC (priv->child_count,
-                                                 sizeof (*local->cont.lk.locked_nodes),
-                                                 gf_afr_mt_char);
+    if (!matrix)
+        goto out;
+    for (i = 0; i < m; i++) {
+        GF_FREE(matrix[i]);
+    }
 
-        if (!local->cont.lk.locked_nodes) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    GF_FREE(matrix);
+out:
+    return;
+}
 
-        local->fd            = fd_ref (fd);
-        local->cont.lk.cmd   = cmd;
-        local->cont.lk.user_flock = *flock;
-        local->cont.lk.ret_flock = *flock;
+int32_t **
+afr_matrix_create(unsigned int m, unsigned int n)
+{
+    int32_t **matrix = NULL;
+    int i = 0;
 
-        STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) 0,
-                           priv->children[i],
-                           priv->children[i]->fops->lk,
-                           fd, cmd, flock, xdata);
+    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:
-        if (ret < 0)
-                AFR_STACK_UNWIND (lk, frame, -1, op_errno, NULL, NULL);
-        return 0;
+    afr_matrix_cleanup(matrix, m);
+    return NULL;
 }
 
 int
-afr_forget (xlator_t *this, inode_t *inode)
-{
-        uint64_t        ctx_addr = 0;
-        afr_inode_ctx_t *ctx     = NULL;
+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;
+    }
 
-        inode_ctx_get (inode, this, &ctx_addr);
+    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;
 
-        if (!ctx_addr)
-                goto out;
+    local->pending = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+    if (!local->pending)
+        goto out;
 
-        ctx = (afr_inode_ctx_t *)(long)ctx_addr;
-        GF_FREE (ctx->fresh_children);
-        GF_FREE (ctx);
+    ret = 0;
 out:
-        return 0;
+    return ret;
 }
 
-int
-afr_priv_dump (xlator_t *this)
+void
+afr_set_low_priority(call_frame_t *frame)
 {
-        afr_private_t *priv = NULL;
-        char  key_prefix[GF_DUMP_MAX_BUF_LEN];
-        char  key[GF_DUMP_MAX_BUF_LEN];
-        int   i = 0;
+    frame->root->pid = LOW_PRIO_PROC_PID;
+}
 
+void
+afr_priv_destroy(afr_private_t *priv)
+{
+    int i = 0;
+    int child_count = -1;
 
-        GF_ASSERT (this);
-        priv = this->private;
+    if (!priv)
+        goto out;
 
-        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);
+    GF_FREE(priv->sh_domain);
+    GF_FREE(priv->last_event);
 
-        return 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;
 }
 
-
-/**
- * find_child_index - find the child's index in the array of subvolumes
- * @this: AFR
- * @child: child
- */
-
-static int
-find_child_index (xlator_t *this, xlator_t *child)
+int **
+afr_mark_pending_changelog(afr_private_t *priv, unsigned char *pending,
+                           dict_t *xattr, ia_type_t iat)
 {
-        afr_private_t *priv = NULL;
-        int i = -1;
+    int i = 0;
+    int **changelog = NULL;
+    int idx = -1;
+    int m_idx = 0;
+    int d_idx = 0;
+    int ret = 0;
 
-        priv = this->private;
+    m_idx = afr_index_for_transaction_type(AFR_METADATA_TRANSACTION);
+    d_idx = afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if ((xlator_t *) child == priv->children[i])
-                        break;
-        }
+    idx = afr_index_from_ia_type(iat);
 
-        return i;
-}
+    changelog = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+    if (!changelog)
+        goto out;
 
-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;
-
-        had_heard_from_all = 1;
-        for (i = 0; i < priv->child_count; i++) {
-                if (!priv->last_event[i]) {
-                        had_heard_from_all = 0;
-                }
-        }
+    for (i = 0; i < priv->child_count; i++) {
+        if (!pending[i])
+            continue;
 
-        /* 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.
+        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.
          */
-        idx = find_child_index (this, data);
-        if (idx < 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Received child_up "
-                        "from invalid subvolume");
-                goto out;
-        }
-
-        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;
-                }
-                UNLOCK (&priv->lock);
+        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 changelog;
+}
 
-                break;
+static dict_t *
+afr_set_heal_info(char *status)
+{
+    dict_t *dict = NULL;
+    int ret = -1;
 
-        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;
-                }
-                UNLOCK (&priv->lock);
+    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);
+    }
 
-                break;
+    if (ret && dict) {
+        dict_unref(dict);
+        dict = NULL;
+    }
+    return dict;
+}
 
-        case GF_EVENT_CHILD_CONNECTING:
-                LOCK (&priv->lock);
-                {
-                        priv->last_event[idx] = event;
-                }
-                UNLOCK (&priv->lock);
+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;
 
-                break;
+    afr_selfheal_extract_xattr(this, replies, type, dirty, NULL);
+    for (i = 0; i < priv->child_count; i++) {
+        if (dirty[i] > 1)
+            return _gf_true;
+    }
 
-        case GF_EVENT_TRANSLATOR_OP:
-                input = data;
-                output = data2;
-                ret = afr_xl_op (this, input, output);
-                goto out;
-                break;
+    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:
-                propagate = 1;
-                break;
-        }
+            /*IA_IFBLK, IA_IFCHR, IA_IFLNK, IA_IFIFO, IA_IFSOCK*/
+            mdata_chk = _gf_true;
+            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 (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;
+    }
 
-        /* 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;
+    return _gf_false;
+}
 
-        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 */
-                                }
-                        }
-                }
-                UNLOCK (&priv->lock);
+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;
         }
-
-        ret = 0;
-        if (propagate)
-                ret = default_notify (this, event, data);
-        if (call_psh && priv->shd.iamshd)
-                afr_proactive_self_heal ((void*) (long) up_child);
-
-out:
-        return ret;
+    } 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_first_up_child (unsigned char *child_up, size_t child_count)
-{
-        int         ret      = -1;
-        int         i        = 0;
-
-        GF_ASSERT (child_up);
-
-        for (i = 0; i < child_count; i++) {
-                if (child_up[i]) {
-                        ret = i;
-                        break;
-                }
-        }
+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;
+    }
 
-        return ret;
+    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_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno)
-{
-        int     ret = -1;
+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;
+    }
 
-        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;
+    if (ret == -EIO) {
+        ret = gf_asprintf(&status, "split-brain%s", substr ? substr : "");
+        if (ret < 0) {
+            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;
+        dict = afr_set_heal_info(status);
+        if (!dict) {
+            ret = -1;
+            goto out;
         }
-
-        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;
+    } else if (ret == -EAGAIN) {
+        ret = gf_asprintf(&status, "possibly-healing%s", substr ? substr : "");
+        if (ret < 0) {
+            goto out;
         }
-
-        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;
+        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;
+            }
+            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;
 
-        ret = 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_internal_lock_init (afr_internal_lock_t *lk, size_t child_count,
-                        transaction_lk_type_t lk_type)
+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             ret = -ENOMEM;
+    int ret = -1;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
 
-        lk->inode_locked_nodes = GF_CALLOC (sizeof (*lk->inode_locked_nodes),
-                                            child_count, gf_afr_mt_char);
-        if (NULL == lk->inode_locked_nodes)
-                goto out;
+    priv = this->private;
 
-        lk->locked_nodes = GF_CALLOC (sizeof (*lk->locked_nodes),
-                                      child_count, gf_afr_mt_char);
-        if (NULL == lk->locked_nodes)
-                goto out;
+    replies = alloca0(sizeof(*replies) * priv->child_count);
 
-        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;
+    ret = afr_selfheal_unlocked_discover(frame, inode, gfid, replies);
+    if (ret)
+        goto out;
 
-        lk->lock_op_ret   = -1;
-        lk->lock_op_errno = EUCLEAN;
-        lk->transaction_lk_type = lk_type;
+    if (!afr_can_decide_split_brain_source_sinks(replies, priv->child_count)) {
+        ret = -EAGAIN;
+        goto out;
+    }
 
-        ret = 0;
+    ret = _afr_is_split_brain(frame, this, replies, AFR_DATA_TRANSACTION,
+                              d_spb);
+    if (ret)
+        goto out;
+
+    ret = _afr_is_split_brain(frame, this, replies, AFR_METADATA_TRANSACTION,
+                              m_spb);
 out:
-        return ret;
+    if (replies) {
+        afr_replies_wipe(replies, priv->child_count);
+        replies = NULL;
+    }
+    return ret;
 }
 
-void
-afr_matrix_cleanup (int32_t **matrix, unsigned int m)
+int
+afr_get_split_brain_status_cbk(int ret, call_frame_t *frame, void *opaque)
 {
-        int             i         = 0;
+    GF_FREE(opaque);
+    return 0;
+}
 
-        if (!matrix)
-                goto out;
-        for (i = 0; i < m; i++) {
-                GF_FREE (matrix[i]);
+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';
+
+        ret = gf_asprintf(&status,
+                          "data-split-brain:%s    "
+                          "metadata-split-brain:%s%s",
+                          (d_spb) ? "yes" : "no", (m_spb) ? "yes" : "no",
+                          choices);
+
+        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;
+        }
+    }
 
-        GF_FREE (matrix);
+    ret = 0;
 out:
-        return;
+    AFR_STACK_UNWIND(getxattr, frame, ret, op_errno, dict, NULL);
+    if (dict)
+        dict_unref(dict);
+    if (inode)
+        inode_unref(inode);
+    return ret;
 }
 
-int32_t**
-afr_matrix_create (unsigned int m, unsigned int n)
-{
-        int32_t         **matrix = NULL;
-        int             i       = 0;
-
-        matrix = GF_CALLOC (sizeof (*matrix), m, gf_afr_mt_int32_t);
-        if (!matrix)
-                goto out;
+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;
+        }
+    }
 
-        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:
-        afr_matrix_cleanup (matrix, m);
-        return NULL;
+    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;
 }
 
 int
-afr_transaction_local_init (afr_local_t *local, xlator_t *this)
+afr_get_child_index_from_name(xlator_t *this, char *name)
 {
-        int            child_up_count = 0;
-        int            ret = -ENOMEM;
-        afr_private_t *priv = NULL;
+    afr_private_t *priv = this->private;
+    int index = -1;
 
-        priv = this->private;
-        ret = afr_internal_lock_init (&local->internal_lock, priv->child_count,
-                                      AFR_TRANSACTION_LK);
-        if (ret < 0)
-                goto out;
+    for (index = 0; index < priv->child_count; index++) {
+        if (!strcmp(priv->children[index]->name, name))
+            goto out;
+    }
+    index = -1;
+out:
+    return index;
+}
 
-        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;
+void
+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);
+}
 
-        local->first_up_child = afr_first_up_child (local->child_up,
-                                                    priv->child_count);
+void
+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;
 
-        local->transaction.eager_lock =
-                GF_CALLOC (sizeof (*local->transaction.eager_lock),
-                           priv->child_count,
-                           gf_afr_mt_int32_t);
+    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;
+}
 
-        if (!local->transaction.eager_lock)
-                goto out;
+gf_boolean_t
+afr_get_need_heal(xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    gf_boolean_t need_heal = _gf_true;
 
-        local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children)
-                goto out;
+    LOCK(&priv->lock);
+    {
+        need_heal = priv->need_heal;
+    }
+    UNLOCK(&priv->lock);
+    return need_heal;
+}
 
-        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_msg_id(char *op_type)
+{
+    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;
+}
 
-        local->pending = afr_matrix_create (priv->child_count,
-                                            AFR_NUM_CHANGE_LOGS);
-        if (!local->pending)
-                goto out;
+int
+afr_fav_child_reset_sink_xattrs_cbk(int ret, call_frame_t *heal_frame,
+                                    void *opaque)
+{
+    call_frame_t *txn_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_local_t *heal_local = NULL;
+    xlator_t *this = NULL;
 
-        local->transaction.txn_changelog = afr_matrix_create (priv->child_count,
-                                                           AFR_NUM_CHANGE_LOGS);
-        if (!local->transaction.txn_changelog)
-                goto out;
+    heal_local = heal_frame->local;
+    txn_frame = heal_local->heal_frame;
+    local = txn_frame->local;
+    this = txn_frame->this;
 
-	INIT_LIST_HEAD (&local->transaction.eager_locked);
+    /* Refresh the inode agan and proceed with the transaction.*/
+    afr_inode_refresh(txn_frame, this, local->inode, NULL, local->refreshfn);
 
-        ret = 0;
-out:
-        return ret;
-}
+    AFR_STACK_DESTROY(heal_frame);
 
-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;
+    return 0;
 }
 
-int32_t*
-afr_children_create (int32_t child_count)
-{
-        int32_t           *children = NULL;
-        int               i               = 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);
+    }
 
-        GF_ASSERT (child_count > 0);
+    return ret;
+}
 
-        children = GF_CALLOC (child_count, sizeof (*children),
-                              gf_afr_mt_int32_t);
-        if (NULL == children)
+/*
+ * 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;
-        for (i = 0; i < child_count; i++)
-                children[i] = -1;
+            }
+            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;
+
 out:
-        return children;
+    return ret;
 }
 
-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;
-                }
-        }
+uint64_t
+afr_write_subvol_get(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    uint64_t write_subvol = 0;
 
-        if (!child_found) {
-                GF_ASSERT (i < child_count);
-                children[i] = child;
-        }
+    local = frame->local;
+    LOCK(&local->inode->lock);
+    write_subvol = local->inode_ctx->write_subvol;
+    UNLOCK(&local->inode->lock);
+
+    return write_subvol;
 }
 
-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;
-                }
-        }
+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;
 }
 
 int
-afr_get_children_count (int32_t *children, unsigned int child_count)
+afr_write_subvol_reset(call_frame_t *frame, xlator_t *this)
 {
-        int count = 0;
-        int i = 0;
+    afr_local_t *local = NULL;
 
-        for (i = 0; i < child_count; i++) {
-                if (children[i] == -1)
-                        break;
-                count++;
-        }
-        return count;
-}
+    local = frame->local;
+    LOCK(&local->inode->lock);
+    {
+        GF_ASSERT(local->inode_ctx->lock_count > 0);
+        local->inode_ctx->lock_count--;
 
-void
-afr_set_low_priority (call_frame_t *frame)
-{
-        frame->root->pid = LOW_PRIO_PROC_PID;
+        if (!local->inode_ctx->lock_count)
+            local->inode_ctx->write_subvol = 0;
+    }
+    UNLOCK(&local->inode->lock);
+
+    return 0;
 }
 
 int
-afr_child_fd_ctx_set (xlator_t *this, fd_t *fd, int32_t child,
-                      int flags)
+afr_set_inode_local(xlator_t *this, afr_local_t *local, inode_t *inode)
 {
-        int             ret = 0;
-        uint64_t        ctx = 0;
-        afr_fd_ctx_t    *fd_ctx      = NULL;
+    int ret = 0;
 
-        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;
-        }
-
-        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);
-                goto out;
-        }
-
-        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;
+    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_have_quorum (char *logname, afr_private_t *priv)
+afr_ta_is_fop_called_from_synctask(xlator_t *this)
 {
-        unsigned int        quorum = 0;
+    struct synctask *task = NULL;
+    gf_lkowner_t tmp_owner = {
+        0,
+    };
 
-        GF_VALIDATE_OR_GOTO(logname,priv,out);
+    task = synctask_get();
+    if (!task)
+        return _gf_false;
 
-        quorum = priv->quorum_count;
-        if (quorum != AFR_QUORUM_AUTO) {
-                return (priv->up_count >= (priv->down_count + quorum));
-        }
+    set_lk_owner_from_ptr(&tmp_owner, (void *)this);
 
-        quorum = priv->child_count / 2 + 1;
-        if (priv->up_count >= (priv->down_count + quorum)) {
-                return _gf_true;
-        }
+    if (!is_same_lkowner(&tmp_owner, &task->frame->root->lk_owner))
+        return _gf_false;
 
-        /*
-         * 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;
-                        }
-                }
+    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 ret;
+}
 
+int
+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 _gf_false;
+    return ret;
 }
 
-void
-afr_priv_destroy (afr_private_t *priv)
+call_frame_t *
+afr_ta_frame_create(xlator_t *this)
 {
-        int            i           = 0;
+    call_frame_t *frame = NULL;
+    void *lk_owner = NULL;
 
-        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);
-out:
-        return;
+    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;
 }
 
-int
-xlator_subvolume_count (xlator_t *this)
+gf_boolean_t
+afr_ta_has_quorum(afr_private_t *priv, afr_local_t *local)
 {
-        int i = 0;
-        xlator_list_t *list = NULL;
+    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;
+    }
 
-        for (list = this->children; list; list = list->next)
-                i++;
-        return i;
+    return _gf_false;
 }
 
-inline gf_boolean_t
-afr_is_errno_set (int *child_errno, int child)
+static gf_boolean_t
+afr_is_add_replica_mount_lookup_on_root(call_frame_t *frame)
 {
-        return child_errno[child];
+    afr_local_t *local = NULL;
+
+    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;
 }
 
-inline gf_boolean_t
-afr_is_errno_unset (int *child_errno, int child)
+gf_boolean_t
+afr_lookup_has_quorum(call_frame_t *frame, const unsigned int up_children_count)
 {
-        return !afr_is_errno_set (child_errno, child);
+    if (frame && (up_children_count > 0) &&
+        afr_is_add_replica_mount_lookup_on_root(frame))
+        return _gf_true;
+
+    return _gf_false;
 }
 
 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);
-                }
-        }
+afr_handle_replies_quorum(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    unsigned char *success_replies = NULL;
+
+    success_replies = alloca0(priv->child_count);
+    afr_fill_success_replies(local, priv, success_replies);
+
+    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;
+    }
 }
 
 gf_boolean_t
-afr_is_fd_fixable (fd_t *fd)
-{
-        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;
+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;
+            }
+        }
+    }
 
-        return _gf_true;
+    return _gf_false;
 }
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index 0e1718814ee..f8bf8340dab 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -8,508 +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"
+#include "afr-transaction.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)
+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_local_t *local  = NULL;
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int32_t child_index = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-        local = frame->local;
+    local = frame->local;
+    fd_ctx = local->fd_ctx;
+    child_index = (long)cookie;
 
-        afr_set_opendir_done (this, local->fd->inode);
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
 
-        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                          local->op_errno, local->fd, NULL);
+    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);
 
-        return 0;
-}
+    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;
+}
 
-gf_boolean_t
-__checksums_differ (uint32_t *checksum, int child_count,
-                    unsigned char *child_up)
+int
+afr_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+            dict_t *xdata)
 {
-        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];
-        }
+    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;
 
-        return ret;
-}
+    priv = this->private;
 
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-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_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;
+    local->op = GF_FOP_OPENDIR;
 
-        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;
-        }
+    if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) {
+        op_errno = afr_quorum_errno(priv);
+        goto out;
+    }
 
-        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;
-        }
+    if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+        goto out;
 
-        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;
-        }
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx)
+        goto out;
 
-        list_for_each_entry (entry, &entries->list, list) {
-                last_offset = entry->d_off;
-        }
+    loc_copy(&local->loc, loc);
 
-        /* read more entries */
+    local->fd = fd_ref(fd);
+    local->fd_ctx = fd_ctx;
 
-        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);
+    call_count = local->call_count;
 
-        return 0;
+    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);
 
-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);
-                }
+            if (!--call_count)
+                break;
         }
+    }
 
-        return 0;
+    return 0;
+out:
+    AFR_STACK_UNWIND(opendir, frame, -1, op_errno, fd, NULL);
+    return 0;
 }
 
-
-int
-afr_examine_dir (call_frame_t *frame, xlator_t *this)
+static int
+afr_validate_read_subvol(inode_t *inode, xlator_t *this, int par_read_subvol)
 {
-        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;
-                }
-        }
-
+    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;
-}
 
+    /* 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;
+}
 
-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)
+static void
+afr_readdir_transform_entries(call_frame_t *frame, gf_dirent_t *subvol_entries,
+                              int subvol, gf_dirent_t *entries, fd_t *fd)
 {
-        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) {
-
-                        /*
-                         * 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;
-                }
+    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;
         }
 
-        return 0;
+        list_del_init(&entry->list);
+        list_add_tail(&entry->list, &entries->list);
 
-out:
-        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                          local->op_errno, local->fd, NULL);
+        if (!validate_subvol)
+            continue;
 
-        return 0;
+        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_opendir (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, fd_t *fd)
+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_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);
-
-        priv = this->private;
-
-        child_count = priv->child_count;
+    afr_local_t *local = NULL;
+    gf_dirent_t entries;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    INIT_LIST_HEAD(&entries.list);
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    local = frame->local;
 
-        loc_copy (&local->loc, loc);
-
-        local->fd    = fd_ref (fd);
-
-        call_count = local->call_count;
-
-        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;
-                }
-        }
-
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (opendir, frame, -1, op_errno, fd, 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;
-}
-
+    }
 
-/**
- * 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
- */
+    if (op_ret >= 0)
+        afr_readdir_transform_entries(frame, subvol_entries, (long)cookie,
+                                      &entries, local->fd);
 
+    AFR_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, xdata);
 
-struct entry_name {
-        char *name;
-        struct list_head list;
-};
+    gf_dirent_free(&entries);
 
-static void
-afr_forget_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;
-        }
-
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-        list_for_each_entry_safe (entry, tmp, &fd_ctx->entries, list) {
-                GF_FREE (entry->name);
-                list_del (&entry->list);
-                GF_FREE (entry);
-        }
+    return 0;
 }
 
-
-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)
+int
+afr_readdir_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries, NULL);
-
+    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_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)
+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_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries, NULL);
-
-        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;
 }
 
 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)
+afr_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, dict_t *xdata)
 {
-        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;
-        }
+    afr_do_readdir(frame, this, fd, size, offset, GF_FOP_READDIR, xdata);
 
-        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;
-out:
-        AFR_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
-        return 0;
+    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_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_READDIR, xdata);
-        return 0;
-}
-
+    afr_do_readdir(frame, this, fd, size, offset, GF_FOP_READDIRP, dict);
 
-int32_t
-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;
+    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 5a22696ce39..b7cceb79158 100644
--- a/xlators/cluster/afr/src/afr-dir-write.c
+++ b/xlators/cluster/afr/src/afr-dir-write.c
@@ -8,521 +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;
-
-        local = frame->local;
-        priv  = this->private;
 
-        if (local->op_ret < 0)
-                goto out;
+        afr_mark_entry_pending_changelog(frame, this);
 
-        if (local->success_count == priv->child_count)
-                goto out;
+        afr_transaction_resume(frame, this);
+    }
 
-        if (!afr_is_new_entry_changelog_needed (local->op))
-                goto out;
-
-        afr_mark_new_entry_changelog (frame, this);
-
-out:
-        return;
+    return 0;
 }
 
-void
-afr_dir_fop_handle_all_fop_failures (call_frame_t *frame)
+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)
 {
-        xlator_t        *this = NULL;
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv = NULL;
+    int call_count = 0;
 
-        this = frame->this;
-        local = frame->local;
-        priv = this->private;
+    call_count = afr_frame_return(frame);
 
-        if (local->op_ret >= 0)
-                goto out;
+    if (call_count == 0)
+        AFR_STACK_DESTROY(frame);
 
-        __mark_all_success (local->pending, priv->child_count,
-                            local->transaction.type);
-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);
-        afr_dir_fop_handle_all_fop_failures (frame);
-        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_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    local = frame->local;
 
-        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_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;
 }
 
 /* }}} */
@@ -530,814 +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;
+    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_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);
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-                AFR_STACK_UNWIND (mkdir, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL);
-        }
-
-        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;
-
-        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;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        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);
+    local = frame->local;
+    priv = this->private;
 
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        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_STACK_DESTROY (frame);
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        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;
 }
 
 /* }}} */
@@ -1345,233 +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;
+    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_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;
 }
 
 /* }}} */
@@ -1579,405 +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;
+    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_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 051b7b1647d..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>
@@ -16,1858 +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);
-
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        children = priv->children;
+    local = frame->local;
+    priv = this->private;
 
-        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;
-
-        if (!strncmp (key, AFR_XATTR_PREFIX,
-                      strlen (AFR_XATTR_PREFIX))) {
+    struct list_head *list = data;
+    struct _xattr_key *xkey = NULL;
 
-                xkey = GF_CALLOC (1, sizeof (*xkey), gf_afr_mt_xattr_key);
-                if (!xkey)
-                        return -1;
+    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->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;
-
-        local = frame->local;
+    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);
+    }
 
-        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;
+    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;
-
-                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;
 
-                /* closing part */
-                *(xattr_serz + padding + tlen) = ')';
-                *(xattr_serz + padding + tlen + 1) = '\0';
+        xattr = gf_strdup(xattr);
 
-                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");
+        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;
+        }
 
-        unwind:
-                AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, nxattr,
-                                  xdata);
+        local->cont.getxattr.xattr_len += strlen(xattr) + 1;
+    }
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!callcnt) {
+        if (!local->cont.getxattr.xattr_len)
+            goto unwind;
 
-                if (nxattr)
-                        dict_unref (nxattr);
+        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;
         }
 
-        return ret;
+        /* 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);
+        }
+
+    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);
+
+        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 (getxattr, 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(getxattr, frame, local->op_ret, local->op_errno,
+                         nxattr, local->xdata_rsp);
+
+        if (nxattr)
+            dict_unref(nxattr);
+    }
+
+out:
+    return ret;
 }
 
-static gf_boolean_t
-afr_is_special_xattr (const char *name, fop_getxattr_cbk_t *cbk,
-                      gf_boolean_t is_fgetxattr)
+static int
+afr_aggregate_stime_xattr(dict_t *this, char *key, data_t *value, void *data)
+{
+    int ret = 0;
+
+    if (fnmatch(GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0)
+        ret = gf_get_max_stime(THIS, data, key, value);
+
+    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)
 {
-        gf_boolean_t    is_spl = _gf_true;
+    afr_local_t *local = NULL;
+    int32_t callcnt = 0;
+
+    if (!frame || !frame->local || !this) {
+        gf_msg("", GF_LOG_ERROR, 0, AFR_MSG_INVALID_ARG, "possible NULL deref");
+        goto out;
+    }
 
-        GF_ASSERT (cbk);
-        if (!cbk) {
-                is_spl = _gf_false;
-                goto out;
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+
+        if (!dict || (op_ret < 0)) {
+            local->op_errno = op_errno;
+            goto cleanup;
         }
 
-        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;
-                }
+        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);
+
+    if (!callcnt) {
+        AFR_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno,
+                         local->dict, xdata);
+    }
+
+out:
+    return 0;
+}
+
+static gf_boolean_t
+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 || !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 {
-                is_spl = _gf_false;
+            *cbk = afr_getxattr_pathinfo_cbk;
         }
+    } else if (!strncmp(name, GF_XATTR_CLRLK_CMD, SLEN(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,
+                        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;
 }
 
-int32_t
-afr_getxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, const char *name, dict_t *xdata)
+static int
+afr_handle_heal_xattrs(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       const char *heal_op)
 {
-        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 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;
+    }
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+out:
+    if (ret == 1) {
+        AFR_STACK_UNWIND(getxattr, frame, -1, ENOMEM, NULL, NULL);
+        if (data)
+            GF_FREE(data);
+        ret = 0;
+    }
+    return ret;
+}
 
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
+int32_t
+afr_getxattr(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;
+    xlator_t **children = NULL;
+    int i = 0;
+    int32_t op_errno = 0;
+    int ret = -1;
+    fop_getxattr_cbk_t cbk = NULL;
 
-        children = priv->children;
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    priv = this->private;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    children = priv->children;
 
-        loc_copy (&local->loc, loc);
-        if (!name)
-                goto no_name;
+    loc_copy(&local->loc, loc);
 
-        local->cont.getxattr.name = gf_strdup (name);
+    local->op = GF_FOP_GETXATTR;
 
-        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)) {
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-                local->marker.call_count = priv->child_count;
+    if (!name)
+        goto no_name;
 
-                sub_volumes = alloca ( priv->child_count * sizeof (xlator_t *));
-                for (i = 0, trav = this->children; trav ;
-                     trav = trav->next, i++) {
+    local->cont.getxattr.name = gf_strdup(name);
 
-                        *(sub_volumes + i)  = trav->xlator;
-                }
+    if (!local->cont.getxattr.name) {
+        op_errno = ENOMEM;
+        goto out;
+    }
 
-                if (cluster_getmarkerattr (frame, this, loc, name,
-                                           local, afr_getxattr_unwind,
-                                           sub_volumes,
-                                           priv->child_count,
-                                           MARKER_UUID_TYPE,
-                                           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;
-                }
+    if (!strncmp(name, AFR_XATTR_PREFIX, SLEN(AFR_XATTR_PREFIX))) {
+        op_errno = ENODATA;
+        goto out;
+    }
 
-                return 0;
-        }
+    if (cluster_handle_marker_getxattr(frame, loc, name, priv->vol_uuid,
+                                       afr_getxattr_unwind,
+                                       afr_marker_populate_args) == 0)
+        return 0;
 
-        /*
-         * 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;
-        }
+    ret = afr_handle_heal_xattrs(frame, this, &local->loc, name);
+    if (ret == 0)
+        return 0;
 
-        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;
-        }
+    /*
+     * 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;
+    }
 
-        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;
-
-                        }
-
-                        if (cluster_getmarkerattr (frame, this, loc,
-                                                   name, local,
-                                                   afr_getxattr_unwind,
-                                                   sub_volumes,
-                                                   priv->child_count,
-                                                   MARKER_XTIME_TYPE,
-                                                   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;
-                }
-        }
+    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:
-        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.getxattr.last_index);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
 
-        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;
+
+    // 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++) {
-                STACK_WIND_COOKIE (frame, cbk,
-                                   (void *) (long) i,
-                                   children[i], children[i]->fops->fgetxattr,
-                                   fd, name, NULL);
+    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;
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
 
-        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);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
+    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);
 
-        children = priv->children;
+    afr_fix_open(fd, this);
 
-        local = frame->local;
-
-        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;
 }
 
+/* }}} */
+
+/* {{{ seek */
 
-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)
+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;
+}
 
-        local->fd                    = fd_ref (fd);
+int
+afr_seek_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local->cont.readv.size       = size;
-        local->cont.readv.offset     = offset;
-        local->cont.readv.flags      = flags;
+    local = frame->local;
+    priv = this->private;
 
-        afr_open_fd_fix (fd, this);
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(seek, frame, local->op_ret, local->op_errno, 0, NULL);
+        return 0;
+    }
 
-        STACK_WIND_COOKIE (frame, afr_readv_cbk,
-                           (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->readv,
-                           fd, size, offset, flags, xdata);
+    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;
+}
 
-        ret = 0;
+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_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_SEEK;
+    local->fd = fd_ref(fd);
+    local->cont.seek.offset = offset;
+    local->cont.seek.what = what;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    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 ce4fbf22698..1d6e4f3570a 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -8,2146 +8,2558 @@
   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"
 
-/* {{{ writev */
+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;
+            } 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;
+            }
+        }
+    }
+
+    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;
+}
 
-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;
+
+    local = frame->local;
+    priv = this->private;
+
+    local->replies[child_index].valid = 1;
+
+    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);
+
+    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;
+}
+
+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;
+    int child_index = (long)cookie;
+    int call_count = -1;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    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);
 
-        src_local = src_frame->local;
-        dst_local = dst_frame->local;
+        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);
+        }
+
+        afr_transaction_resume(frame, this);
+    }
 
-        dst_local->op_ret = src_local->op_ret;
-        dst_local->op_errno = src_local->op_errno;
-        dst_local->cont.writev.prebuf = src_local->cont.writev.prebuf;
-        dst_local->cont.writev.postbuf = src_local->cont.writev.postbuf;
+    return 0;
 }
 
+/* {{{ writev */
+
 void
-afr_writev_unwind (call_frame_t *frame, xlator_t *this)
+afr_writev_copy_outvars(call_frame_t *src_frame, call_frame_t *dst_frame)
 {
-        afr_local_t *   local = NULL;
-        local = frame->local;
-
-        AFR_STACK_UNWIND (writev, frame,
-                          local->op_ret, local->op_errno,
-                          &local->cont.writev.prebuf,
-                          &local->cont.writev.postbuf,
-                          NULL);
+    afr_local_t *src_local = NULL;
+    afr_local_t *dst_local = NULL;
+
+    src_local = src_frame->local;
+    dst_local = dst_frame->local;
+
+    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);
 }
 
-call_frame_t*
-afr_transaction_detach_fop_frame (call_frame_t *frame)
+void
+afr_writev_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *fop_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = this->private;
 
-        local = frame->local;
+    local = frame->local;
 
-        LOCK (&frame->lock);
-        {
-                fop_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    if (priv->consistent_metadata)
+        afr_zero_fill_stat(local);
 
-        return fop_frame;
+    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_transaction_writev_unwind (call_frame_t *frame, xlator_t *this)
+afr_transaction_writev_unwind(call_frame_t *frame, xlator_t *this)
 {
-        call_frame_t *fop_frame = NULL;
+    call_frame_t *fop_frame = NULL;
 
-        fop_frame = afr_transaction_detach_fop_frame (frame);
+    fop_frame = afr_transaction_detach_fop_frame(frame);
 
-        if (fop_frame) {
-                afr_writev_copy_outvars (frame, fop_frame);
-                afr_writev_unwind (fop_frame, this);
-        }
-        return 0;
+    if (fop_frame) {
+        afr_writev_copy_outvars(frame, fop_frame);
+        afr_writev_unwind(fop_frame, this);
+    }
+    return 0;
 }
 
 static void
-afr_writev_handle_short_writes (call_frame_t *frame, xlator_t *this)
+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);
-        }
+    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);
+    }
 }
 
-int
-afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+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)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t    *fop_frame = 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;
-                }
-
-		local->replies[child_index].valid = 1;
-		local->replies[child_index].op_ret = op_ret;
-		local->replies[child_index].op_errno = op_errno;
-
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
-
-		/* 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 ((local->success_count == 0) ||
-			    (child_index == read_child)) {
-				local->cont.writev.prebuf  = *prebuf;
-				local->cont.writev.postbuf = *postbuf;
-			}
-			local->success_count++;
-		}
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-
-		if (!local->stable_write)
-			afr_fd_report_unstable_write (this, local->fd);
-
-                afr_writev_handle_short_writes (frame, this);
-                /*
-                 * 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);
+    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;
         }
-        return 0;
+
+        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;
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
 }
 
-int
-afr_writev_wind (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_private_t *priv = NULL;
-        int i = 0;
-        int call_count = -1;
-
-        local = frame->local;
-        priv = this->private;
+    afr_local_t *local = NULL;
+    afr_lock_t *lock = NULL;
+
+    local = frame->local;
+
+    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);
+
+    __afr_inode_write_finalize(frame, this);
+
+    afr_writev_handle_short_writes(frame, this);
+
+    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;
+    }
+}
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+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)
+{
+    call_frame_t *fop_frame = NULL;
+    int child_index = (long)cookie;
+    int call_count = -1;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    afr_inode_write_fill(frame, this, child_index, op_ret, op_errno, prebuf,
+                         postbuf, xdata);
 
-        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;
-	}
+    call_count = afr_frame_return(frame);
 
-        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,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    if (call_count == 0) {
+        afr_process_post_writev(frame, this);
 
-        return 0;
+        if (!afr_txn_nothing_failed(frame, this)) {
+            // Don't unwind until post-op is complete
+            afr_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);
+            afr_transaction_resume(frame, this);
+            afr_writev_unwind(fop_frame, this);
+        }
+    }
+    return 0;
 }
 
+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;
+
+    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);
+
+    return 0;
+}
 
 int
-afr_writev_done (call_frame_t *frame, xlator_t *this)
+afr_writev_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;
 
-        iobref_unref (local->cont.writev.iobref);
-        local->cont.writev.iobref = NULL;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
+    priv = this->private;
 
+    if (AFR_IS_ARBITER_BRICK(priv, subvol)) {
+        afr_arbiter_writev_wind(frame, this, subvol);
         return 0;
+    }
+
+    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_do_writev (call_frame_t *frame, xlator_t *this)
+afr_do_writev(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;
+    call_frame_t *transaction_frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        local = frame->local;
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    local = frame->local;
+    transaction_frame->local = local;
+    frame->local = NULL;
 
-        transaction_frame->local = local;
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+    if (!AFR_FRAME_INIT(frame, op_errno))
+        goto out;
 
-        local->op = GF_FOP_WRITE;
+    local->op = GF_FOP_WRITE;
 
-        local->success_count      = 0;
+    local->transaction.wind = afr_writev_wind;
+    local->transaction.unwind = afr_transaction_writev_unwind;
 
-        local->transaction.fop    = afr_writev_wind;
-        local->transaction.done   = afr_writev_done;
-        local->transaction.unwind = afr_transaction_writev_unwind;
+    local->transaction.main_frame = frame;
 
-        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;
-        } 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;
+    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 (op_ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (writev, frame, op_ret, op_errno, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
+    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;
+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:
+    AFR_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
 
-        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;
+    return 0;
+}
 
-        local->loc.inode = inode_ref (fd->inode);
-        ret = loc_path (&local->loc, NULL);
-        if (ret < 0)
-                goto out;
+/* }}} */
 
-        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);
-}
+/* {{{ truncate */
 
-void
-afr_open_fd_fix (fd_t *fd, xlator_t *this)
+int
+afr_truncate_unwind(call_frame_t *frame, xlator_t *this)
 {
-        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;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        priv  = this->private;
+    local = frame->local;
 
-        if (!afr_is_fd_fixable (fd))
-                goto out;
-
-        fd_ctx = afr_fd_ctx_get (fd, this);
-        if (!fd_ctx)
-                goto out;
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        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;
-                }
+    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;
+}
 
-                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;
+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;
 
-                        if (!priv->child_up[i])
-                                continue;
+    local = frame->local;
 
-                        fd_ctx->opened_on[i] = AFR_FD_OPENING;
+    if (op_ret == 0 && prebuf->ia_size != postbuf->ia_size)
+        local->stable_write = _gf_false;
 
-                        need_open[i] = 1;
-                        need_open_count++;
-                }
-        }
-        UNLOCK (&fd->lock);
-        if (ret)
-                goto out;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        if (need_self_heal)
-                afr_trigger_open_fd_self_heal (fd, this);
+int
+afr_truncate_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        if (!need_open_count)
-                goto out;
+    local = frame->local;
+    priv = this->private;
 
-        afr_fix_open (this, fd, need_open_count, need_open);
-out:
-        return;
+    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_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_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;
-        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);
+    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;
-        }
+    local->cont.truncate.offset = offset;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        QUORUM_CHECK(writev,out);
+    if (!local->xdata_req)
+        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;
+    local->transaction.wind = afr_truncate_wind;
+    local->transaction.unwind = afr_truncate_unwind;
 
-        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);
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-        local->fd                = fd_ref (fd);
+    local->op = GF_FOP_TRUNCATE;
 
-	/* 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));
+    local->transaction.main_frame = frame;
+    local->transaction.start = offset;
+    local->transaction.len = 0;
 
-        afr_open_fd_fix (fd, this);
+    /* Set it true speculatively, will get reset in afr_truncate_wind_cbk
+       if truncate was not a NOP */
+    local->stable_write = _gf_true;
 
-        afr_do_writev (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)
-                AFR_STACK_UNWIND (writev, 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;
 }
 
-
 /* }}} */
 
-/* {{{ truncate */
+/* {{{ ftruncate */
 
 int
-afr_truncate_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;
-
-        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.truncate.prebuf,
-                                  &local->cont.truncate.postbuf,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         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_truncate_wind_cbk (call_frame_t *frame, void *cookie, 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 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;
-                }
-
-                if (afr_fop_failed (op_ret, op_errno) && op_errno != EFBIG)
-                        afr_transaction_fop_failed (frame, this, child_index);
-
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.truncate.prebuf  = *prebuf;
-                                local->cont.truncate.postbuf = *postbuf;
-                        }
-
-                        if (child_index == read_child) {
-                                local->cont.truncate.prebuf  = *prebuf;
-                                local->cont.truncate.postbuf = *postbuf;
-                        }
-
-                        local->success_count++;
-
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    afr_local_t *local = NULL;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    local = frame->local;
 
-        call_count = afr_frame_return (frame);
+    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;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
 }
 
-
-int32_t
-afr_truncate_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_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_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;
-                }
-        }
-
-        return 0;
-}
-
-
 int
-afr_truncate_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_truncate (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, off_t offset, dict_t *xdata)
+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;
+    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;
 
-        QUORUM_CHECK(truncate,out);
+    local->cont.ftruncate.offset = offset;
+    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->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;
+    local->op = GF_FOP_FTRUNCATE;
 
-        local->cont.truncate.offset  = offset;
+    local->transaction.wind = afr_ftruncate_wind;
+    local->transaction.unwind = afr_ftruncate_unwind;
 
-        local->transaction.fop    = afr_truncate_wind;
-        local->transaction.done   = afr_truncate_done;
-        local->transaction.unwind = afr_truncate_unwind;
+    local->transaction.main_frame = frame;
 
-        loc_copy (&local->loc, loc);
+    local->transaction.start = local->cont.ftruncate.offset;
+    local->transaction.len = 0;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = offset;
-        local->transaction.len     = 0;
+    afr_fix_open(fd, this);
 
-        ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
+    /* Set it true speculatively, will get reset in afr_ftruncate_wind_cbk
+       if truncate was not a NOP */
+    local->stable_write = _gf_true;
 
-        ret = 0;
+    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 (truncate, frame, -1, op_errno, NULL, NULL, NULL);
-        }
+    AFR_STACK_UNWIND(ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 /* }}} */
 
-/* {{{ ftruncate */
-
+/* {{{ setattr */
 
 int
-afr_ftruncate_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;
+    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.ftruncate.prebuf,
-                                  &local->cont.ftruncate.postbuf,
-                                  NULL);
-        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         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;
+}
 
 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_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 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;
-                }
-
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
-
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.ftruncate.prebuf  = *prebuf;
-                                local->cont.ftruncate.postbuf = *postbuf;
-                        }
-
-                        if (child_index == read_child) {
-                                local->cont.ftruncate.prebuf  = *prebuf;
-                                local->cont.ftruncate.postbuf = *postbuf;
-                        }
-
-                        local->success_count++;
-
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                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;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, preop,
+                                 postop, NULL, xdata);
 }
 
-
 int
-afr_ftruncate_wind (call_frame_t *frame, xlator_t *this)
+afr_setattr_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;
-
-        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_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;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    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_ftruncate_done (call_frame_t *frame, xlator_t *this)
+afr_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *buf,
+            int32_t valid, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
+    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;
+
+    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();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->transaction.wind = afr_setattr_wind;
+    local->transaction.unwind = afr_setattr_unwind;
+
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
+
+    local->op = GF_FOP_SETATTR;
+
+    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;
+    }
+
+    return 0;
+out:
+    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_do_ftruncate (call_frame_t *frame, xlator_t *this)
+afr_fsetattr_unwind(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;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
-
-        transaction_frame->local = local;
-        frame->local = NULL;
-
-        local->op = GF_FOP_FTRUNCATE;
-
-        local->transaction.fop    = afr_ftruncate_wind;
-        local->transaction.done   = afr_ftruncate_done;
-        local->transaction.unwind = afr_ftruncate_unwind;
-
-        local->transaction.main_frame = frame;
-
-        local->transaction.start   = local->cont.ftruncate.offset;
-        local->transaction.len     = 0;
-
-        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
-        if (op_ret < 0) {
-            op_errno = -op_ret;
-            goto out;
-        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        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);
-        }
+    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;
+}
 
-        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)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, preop,
+                                 postop, NULL, xdata);
 }
 
+int
+afr_fsetattr_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;
+
+    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;
+}
 
 int
-afr_ftruncate (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, off_t offset, dict_t *xdata)
+afr_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, 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;
+    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(ftruncate,out);
+    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();
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = 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_fsetattr_wind;
+    local->transaction.unwind = afr_fsetattr_unwind;
 
-        local->cont.ftruncate.offset  = offset;
+    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_FSETATTR;
 
-        afr_open_fd_fix (fd, this);
+    afr_fix_open(fd, this);
 
-        afr_do_ftruncate (frame, this);
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        ret = 0;
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_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 (ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
-/* }}} */
-
-/* {{{ setattr */
+/* {{{ setxattr */
 
 int
-afr_setattr_unwind (call_frame_t *frame, xlator_t *this)
+afr_setxattr_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.setattr.preop_buf,
-                                  &local->cont.setattr.postop_buf,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
+
+    AFR_STACK_UNWIND(setxattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    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_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_setxattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        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;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    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;
+}
 
-        local = frame->local;
-        priv  = this->private;
+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)
 
-        read_child = afr_inode_get_read_ctx (this, local->loc.inode, NULL);
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i, ret = 0;
+    char *op_type = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    local = frame->local;
+    priv = this->private;
+    i = (long)cookie;
 
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
 
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.setattr.preop_buf  = *preop;
-                                local->cont.setattr.postop_buf = *postop;
-                        }
+    ret = dict_get_str_sizen(local->xdata_req, "replicate-brick-op", &op_type);
+    if (ret)
+        goto out;
 
-                        if (child_index == read_child) {
-                                local->cont.setattr.preop_buf  = *preop;
-                                local->cont.setattr.postop_buf = *postop;
-                        }
+    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);
 
-                        local->success_count++;
+out:
+    syncbarrier_wake(&local->barrier);
+    return 0;
+}
 
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+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;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    local = frame->local;
+    priv = this->private;
 
-        call_count = afr_frame_return (frame);
+    AFR_ONLIST(locked_nodes, frame, afr_emptyb_set_pending_changelog_cbk,
+               xattrop, &local->loc, GF_XATTROP_ADD_ARRAY, local->xattr_req,
+               NULL);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    /* It is sufficient if xattrop was successful on one child */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
 
-        return 0;
+        if (local->replies[i].op_ret == 0) {
+            ret = 0;
+            goto out;
+        } else {
+            ret = afr_higher_errno(ret, local->replies[i].op_errno);
+        }
+    }
+out:
+    return -ret;
 }
 
-
-int32_t
-afr_setattr_wind (call_frame_t *frame, xlator_t *this)
+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)
 {
-        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;
-        }
+    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;
+}
 
-        local->call_count = call_count;
+void
+afr_brick_args_cleanup(void *opaque)
+{
+    afr_empty_brick_args_t *data = NULL;
 
-        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;
-                }
-        }
+    data = opaque;
+    loc_wipe(&data->loc);
+    GF_FREE(data);
+}
 
-        return 0;
+int
+_afr_handle_empty_brick_cbk(int ret, call_frame_t *frame, void *opaque)
+{
+    afr_brick_args_cleanup(opaque);
+    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_setattr_done (call_frame_t *frame, xlator_t *this)
+afr_split_brain_resolve_do(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                           char *data)
 {
-        afr_local_t *local = NULL;
+    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;
+}
 
-        local = frame->local;
+int
+afr_get_split_brain_child_index(xlator_t *this, void *value, size_t len)
+{
+    int spb_child_index = -1;
+    char *spb_child_str = NULL;
 
-        local->transaction.unwind (frame, this);
+    spb_child_str = alloca0(len + 1);
+    memcpy(spb_child_str, value, len);
 
-        AFR_STACK_DESTROY (frame);
+    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_setattr (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, struct iatt *buf, int32_t valid, dict_t *xdata)
+afr_can_set_split_brain_choice(void *opaque)
 {
-        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;
-
-        QUORUM_CHECK(setattr,out);
+    afr_spbc_timeout_t *data = opaque;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    loc_t *loc = NULL;
+    int ret = -1;
+
+    frame = data->frame;
+    loc = data->loc;
+    this = frame->this;
+
+    ret = afr_is_split_brain(frame, this, loc->inode, loc->gfid, &data->d_spb,
+                             &data->m_spb);
+
+    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;
+}
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
+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;
+            }
         }
 
-        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.setattr.in_buf = *buf;
-        local->cont.setattr.valid  = valid;
-
-        local->transaction.fop    = afr_setattr_wind;
-        local->transaction.done   = afr_setattr_done;
-        local->transaction.unwind = afr_setattr_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 = 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;
-out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL, NULL);
+        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;
 }
 
-/* {{{ fsetattr */
-
 int
-afr_fsetattr_unwind (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 = NULL;
-        call_frame_t   *main_frame = NULL;
+    int ret = -1;
+    int op_errno = 0;
+    uint64_t timeout = 0;
+    afr_private_t *priv = NULL;
 
-        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 (fsetattr, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.fsetattr.preop_buf,
-                                  &local->cont.fsetattr.postop_buf,
-                                  NULL);
-        }
+    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_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_handle_empty_brick(xlator_t *this, call_frame_t *frame, loc_t *loc,
+                       dict_t *dict)
 {
-        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;
+    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;
+        }
+        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;
+out:
+    if (ret == 1) {
+        AFR_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
+        ret = 0;
+    }
+    return ret;
+}
 
-        local = frame->local;
-        priv  = this->private;
+static int
+afr_handle_special_xattr(xlator_t *this, call_frame_t *frame, loc_t *loc,
+                         dict_t *dict)
+{
+    int ret = -1;
 
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+    ret = afr_handle_split_brain_commands(this, frame, loc, dict);
+    if (ret == 0)
+        goto out;
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    ret = afr_handle_spb_choice_timeout(this, frame, dict);
+    if (ret == 0)
+        goto out;
 
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
+    /* Applicable for replace-brick and add-brick commands */
+    ret = afr_handle_empty_brick(this, frame, loc, dict);
+out:
+    return ret;
+}
 
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.fsetattr.preop_buf  = *preop;
-                                local->cont.fsetattr.postop_buf = *postop;
-                        }
+int
+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;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = EINVAL;
 
-                        if (child_index == read_child) {
-                                local->cont.fsetattr.preop_buf  = *preop;
-                                local->cont.fsetattr.postop_buf = *postop;
-                        }
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.afr.*", dict, op_errno, out);
 
-                        local->success_count++;
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.glusterfs.afr.*", dict, op_errno, out);
 
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                local->op_errno = op_errno;
-        }
-        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;
 
-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;
+    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_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;
-                }
-        }
+    AFR_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
+/* {{{ fsetxattr */
 
 int
-afr_fsetattr_done (call_frame_t *frame, xlator_t *this)
+afr_fsetxattr_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(fsetxattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    return 0;
+}
 
-        return 0;
+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)
+{
+    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_fsetxattr_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;
+
+    local = frame->local;
+    priv = this->private;
+
+    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;
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+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;
 
-        priv = this->private;
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.afr.*", dict, op_errno, out);
 
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.glusterfs.afr.*", dict, op_errno, out);
 
-        QUORUM_CHECK(fsetattr,out);
+    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) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
+    local->cont.fsetxattr.dict = dict_ref(dict);
+    local->cont.fsetxattr.flags = flags;
 
-        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.fsetattr.in_buf = *buf;
-        local->cont.fsetattr.valid  = valid;
+    if (!local->xdata_req)
+        goto out;
 
-        local->transaction.fop    = afr_fsetattr_wind;
-        local->transaction.done   = afr_fsetattr_done;
-        local->transaction.unwind = afr_fsetattr_unwind;
+    local->transaction.wind = afr_fsetxattr_wind;
+    local->transaction.unwind = afr_fsetxattr_unwind;
 
-        local->fd                 = fd_ref (fd);
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        afr_open_fd_fix (fd, this);
+    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 (fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(fsetxattr, frame, -1, op_errno, NULL);
+    return 0;
 }
 
+/* }}} */
 
-/* {{{ setxattr */
-
+/* {{{ removexattr */
 
 int
-afr_setxattr_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;
+    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 (setxattr, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  NULL);
-        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         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_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;
+    AFR_STACK_UNWIND(removexattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    return 0;
 }
 
-
 int
-afr_setxattr_wind (call_frame_t *frame, xlator_t *this)
+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_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;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, NULL, xdata);
 }
 
-
 int
-afr_setxattr_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_setxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, dict_t *dict, int32_t flags, 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          = EINVAL;
-
-        VALIDATE_OR_GOTO (this, out);
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.afr.*", dict,
-                                   op_errno, out);
+    GF_IF_NATIVE_XATTR_GOTO("trusted.afr.*", name, op_errno, out);
 
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.afr.*", dict,
-                                   op_errno, out);
+    GF_IF_NATIVE_XATTR_GOTO("trusted.glusterfs.afr.*", name, op_errno, out);
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this->private, 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;
 
-        QUORUM_CHECK(setxattr,out);
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    local->cont.removexattr.name = gf_strdup(name);
 
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    if (!local->xdata_req)
+        goto out;
 
-        local->cont.setxattr.dict  = dict_ref (dict);
-        local->cont.setxattr.flags = flags;
+    local->transaction.wind = afr_removexattr_wind;
+    local->transaction.unwind = afr_removexattr_unwind;
 
-        local->transaction.fop    = afr_setxattr_wind;
-        local->transaction.done   = afr_setxattr_done;
-        local->transaction.unwind = afr_setxattr_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 (setxattr, 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;
 }
 
-/* {{{ fsetxattr */
-
-
+/* ffremovexattr */
 int
-afr_fsetxattr_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;
+    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 (fsetxattr, 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,
+                          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_fsetxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+afr_fremovexattr_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              need_unwind = 0;
+    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);
-        {
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                        }
-                        local->success_count++;
+    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->child_count) {
-                                need_unwind = 1;
-                        }
-                }
+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;
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    GF_IF_NATIVE_XATTR_GOTO("trusted.afr.*", name, op_errno, out);
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    GF_IF_NATIVE_XATTR_GOTO("trusted.glusterfs.afr.*", name, op_errno, out);
 
-        call_count = afr_frame_return (frame);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        return 0;
-}
+    local->cont.removexattr.name = gf_strdup(name);
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
+    if (!local->xdata_req)
+        goto out;
 
-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;
+    local->transaction.wind = afr_fremovexattr_wind;
+    local->transaction.unwind = afr_fremovexattr_unwind;
 
-        local = frame->local;
-        priv = this->private;
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    local->op = GF_FOP_FREMOVEXATTR;
 
-        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_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;
-                }
-        }
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
-}
+    AFR_STACK_UNWIND(fremovexattr, frame, -1, op_errno, NULL);
 
+    return 0;
+}
 
 int
-afr_fsetxattr_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_fsetxattr (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, dict_t *dict, int32_t flags, 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_errno          = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.afr.*", dict,
-                                   op_errno, out);
+int
+afr_fallocate_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;
+
+    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;
+}
 
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.afr.*", dict,
-                                   op_errno, 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;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        QUORUM_CHECK(fsetxattr,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)
-                goto 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) {
-                goto out;
-        }
+    if (!local->xdata_req)
+        goto out;
 
-        transaction_frame->local = local;
+    local->op = GF_FOP_FALLOCATE;
 
-        local->op_ret = -1;
+    local->transaction.wind = afr_fallocate_wind;
+    local->transaction.unwind = afr_fallocate_unwind;
 
-        local->cont.fsetxattr.dict  = dict_ref (dict);
-        local->cont.fsetxattr.flags = flags;
+    local->transaction.main_frame = frame;
 
-        local->transaction.fop    = afr_fsetxattr_wind;
-        local->transaction.done   = afr_fsetxattr_done;
-        local->transaction.unwind = afr_fsetxattr_unwind;
+    local->transaction.start = local->cont.fallocate.offset;
+    local->transaction.len = 0;
 
-        local->fd                 = fd_ref (fd);
+    afr_fix_open(fd, this);
 
-        local->transaction.main_frame = frame;
-        local->transaction.start  = LLONG_MAX - 1;
-        local->transaction.len    = 0;
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_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(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
-
-/* {{{ removexattr */
-
+/* {{{ discard */
 
 int
-afr_removexattr_unwind (call_frame_t *frame, xlator_t *this)
+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 (removexattr, 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(discard, 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_removexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+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)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int call_count  = -1;
-        int need_unwind = 0;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        local = frame->local;
-        priv = this->private;
+int
+afr_discard_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;
+
+    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;
+}
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                        }
-                        local->success_count++;
+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;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-                        if (local->success_count == priv->wait_count) {
-                                need_unwind = 1;
-                        }
-                }
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    local->cont.discard.offset = offset;
+    local->cont.discard.len = len;
 
-        call_count = afr_frame_return (frame);
+    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);
-        }
+    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->op = GF_FOP_DISCARD;
 
-int32_t
-afr_removexattr_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->transaction.wind = afr_discard_wind;
+    local->transaction.unwind = afr_discard_unwind;
 
-        local = frame->local;
-        priv = this->private;
+    local->transaction.main_frame = frame;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    local->transaction.start = local->cont.discard.offset;
+    local->transaction.len = 0;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    afr_fix_open(fd, this);
 
-        local->call_count = call_count;
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_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_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;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(discard, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
+/* {{{ zerofill */
 
 int
-afr_removexattr_done (call_frame_t *frame, xlator_t *this)
+afr_zerofill_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(discard, 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_removexattr (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, const char *name, dict_t *xdata)
+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)
 {
-        afr_private_t   *priv              = NULL;
-        afr_local_t     *local             = NULL;
-        call_frame_t    *transaction_frame = NULL;
-        int              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_zerofill_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;
+
+    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;
+}
 
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.afr.*",
-                                 name, op_errno, out);
+int
+afr_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             size_t len, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.afr.*",
-                                 name, op_errno, out);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, 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.zerofill.offset = offset;
+    local->cont.zerofill.len = len;
 
-        QUORUM_CHECK(removexattr,out);
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    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->op = GF_FOP_ZEROFILL;
 
-        local->cont.removexattr.name = gf_strdup (name);
+    local->transaction.wind = afr_zerofill_wind;
+    local->transaction.unwind = afr_zerofill_unwind;
 
-        local->transaction.fop    = afr_removexattr_wind;
-        local->transaction.done   = afr_removexattr_done;
-        local->transaction.unwind = afr_removexattr_unwind;
+    local->transaction.main_frame = frame;
 
-        loc_copy (&local->loc, loc);
+    local->transaction.start = local->cont.zerofill.offset;
+    local->transaction.len = len;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
+    afr_fix_open(fd, this);
 
-        ret = afr_transaction (transaction_frame, this, AFR_METADATA_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 (removexattr, frame, -1, op_errno, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
-/* ffremovexattr */
-int
-afr_fremovexattr_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 (fremovexattr, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  NULL);
-        }
-        return 0;
+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);
 }
 
-
 int
-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_xattrop_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 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->wait_count) {
-                                need_unwind = 1;
-                        }
-                }
-
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    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;
+}
 
-        call_count = afr_frame_return (frame);
+int
+afr_xattrop_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(xattrop, main_frame, local->op_ret, local->op_errno,
+                     local->xattr_rsp, local->xdata_rsp);
+    return 0;
+}
 
 int32_t
-afr_fremovexattr_wind (call_frame_t *frame, xlator_t *this)
+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;
-        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;
+    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.xattrop.xattr = dict_ref(xattr);
+    local->cont.xattrop.optype = optype;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-        local->call_count = call_count;
+    local->transaction.wind = afr_xattrop_wind;
+    local->transaction.unwind = afr_xattrop_unwind;
 
-        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;
-                }
-        }
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-        return 0;
-}
+    local->op = GF_FOP_XATTROP;
 
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-int
-afr_fremovexattr_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        AFR_STACK_DESTROY (frame);
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(xattrop, frame, -1, op_errno, NULL, NULL);
+    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);
+}
 
 int
-afr_fremovexattr (call_frame_t *frame, xlator_t *this,
-                  fd_t *fd, const char *name, dict_t *xdata)
+afr_fxattrop_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_ret   = -1;
-        int op_errno = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    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;
+}
 
-        VALIDATE_OR_GOTO (this, out);
+int
+afr_fxattrop_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.afr.*",
-                                 name, op_errno, out);
+    local = frame->local;
 
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.afr.*",
-                                 name, op_errno, out);
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    AFR_STACK_UNWIND(fxattrop, main_frame, local->op_ret, local->op_errno,
+                     local->xattr_rsp, local->xdata_rsp);
+    return 0;
+}
 
-        priv = this->private;
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+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_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    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;
+
+    local->cont.xattrop.xattr = dict_ref(xattr);
+    local->cont.xattrop.optype = optype;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    local->transaction.wind = afr_fxattrop_wind;
+    local->transaction.unwind = afr_fxattrop_unwind;
+
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
+
+    local->op = GF_FOP_FXATTROP;
+
+    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;
+    }
+
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        QUORUM_CHECK(fremovexattr, out);
+    AFR_STACK_UNWIND(fxattrop, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
+int
+afr_fsync_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (local, out);
+    local = frame->local;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
-
-        transaction_frame->local = local;
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        local->op_ret = -1;
+    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);
 
-        local->cont.removexattr.name = gf_strdup (name);
+    return 0;
+}
 
-        local->transaction.fop    = afr_fremovexattr_wind;
-        local->transaction.done   = afr_fremovexattr_done;
-        local->transaction.unwind = afr_fremovexattr_unwind;
+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;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
+    local = frame->local;
+    priv = this->private;
 
-        op_ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
-        if (op_ret < 0) {
-            op_errno = -op_ret;
-            goto out;
-        }
+    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;
+}
 
-        op_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 (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(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 ed11079fd1b..a787069b7a1 100644
--- a/xlators/cluster/afr/src/afr-inode-write.h
+++ b/xlators/cluster/afr/src/afr-inode-write.h
@@ -12,60 +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);
+
+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);
+
+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 812609819f5..bc8eabe0f43 100644
--- a/xlators/cluster/afr/src/afr-lk-common.c
+++ b/xlators/cluster/afr/src/afr-lk-common.c
@@ -8,2136 +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)
+afr_lockees_cleanup(afr_internal_lock_t *int_lock)
 {
-        snprintf (str, size, "Basename=%s, lk-owner=%s",
-                  basename ? basename : "<nul>",
-                  lkowner_utoa (owner));
-}
+    int i = 0;
 
-static void
-afr_print_verdict (int op_ret, int op_errno, char *str)
-{
-        if (op_ret < 0) {
-                if (op_errno == EAGAIN)
-                        strcpy (str, "EAGAIN");
-                else
-                        strcpy (str, "FAILED");
-        }
-        else
-                strcpy (str, "GRANTED");
-}
-
-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;
-        }
-
-}
-
-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);
+    for (i = 0; i < int_lock->lockee_count; i++) {
+        afr_lockee_cleanup(&int_lock->lockee[i]);
+    }
 
+    return;
 }
-
-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;
-
-        int_lock = &local->internal_lock;
-
-        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");
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int32_t call_count = 0;
+    int i = 0;
 
-                ret = 0;
-        }
-
-        if (ret == -1)
-                gf_log (THIS->name, GF_LOG_DEBUG,
-                        "lk op is not set");
+    local = frame->local;
+    priv = this->private;
 
-        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;
-}
-
-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;
-        int i = 0;
+    int i = 0;
 
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    priv = this->private;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        int_lock->inodelk_lock_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;
 
-        for (i = 0; i < priv->child_count; i++) {
-                int_lock->inode_locked_nodes[i] = 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);
+    }
 
-        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)
+afr_locked_nodes_count(unsigned char *locked_nodes, int child_count)
 
 {
-        int i = 0;
-        int call_count = 0;
+    int i = 0;
+    int call_count = 0;
 
-        for (i = 0; i < child_count; i++) {
-                if (locked_nodes[i] & LOCKED_YES)
-                        call_count++;
-        }
+    for (i = 0; i < child_count; i++) {
+        if (locked_nodes[i] & LOCKED_YES)
+            call_count++;
+    }
 
-        return 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)
+static void
+afr_log_locks_failure(call_frame_t *frame, char *where, char *what,
+                      int op_errno)
 {
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        int call_count = 0;
-
-        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;
 
-        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);
-        }
+    fop = gf_fop_list[local->op];
 
-        return 0;
+    switch (local->transaction.type) {
+        case AFR_ENTRY_RENAME_TRANSACTION:
+        case AFR_ENTRY_TRANSACTION:
+            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 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_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;
-        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));
-        }
-
+    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;
 
-        int_lock->inode_locked_nodes[child_index] &= LOCKED_NO;
-        if (local->transaction.eager_lock)
-                local->transaction.eager_lock[child_index] = 0;
+    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;
 
-        afr_unlock_common_cbk (frame, cookie, this, op_ret, op_errno, xdata);
+    if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
+        afr_log_locks_failure(frame, priv->children[child_index]->name,
+                              "unlock", op_errno);
+    }
 
-        return 0;
+    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);
 
-}
+    LOCK(&frame->lock);
+    {
+        call_count = --int_lock->lk_call_count;
+    }
+    UNLOCK(&frame->lock);
 
-static int
-afr_unlock_inodelk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = 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;
-
-        flock.l_start = int_lock->lk_flock.l_start;
-        flock.l_len   = int_lock->lk_flock.l_len;
-        flock.l_type  = F_UNLCK;
-
-        full_flock.l_type = F_UNLCK;
-        call_count = afr_locked_nodes_count (int_lock->inode_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 (call_count == 0) {
+        int_lock->lock_cbk(frame, this);
+    }
 
-        if (local->fd)
-                fd_ctx = afr_fd_ctx_get (local->fd, this);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if ((int_lock->inode_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,
-                                           this->name, 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,
-                                           this->name, &local->loc,
-                                           F_SETLK, &flock, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-out:
-        return 0;
+    return ret;
 }
 
-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));
-
-        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);
+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:
+        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;
 
-        return 0;
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+            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 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;
-        }
-
-        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,
-                                           this->name,
-                                           &int_lock->lockee[lockee_no].loc,
-                                           int_lock->lockee[lockee_no].basename,
-                                           ENTRYLK_UNLOCK, ENTRYLK_WRLCK, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
+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;
         }
+    }
 
 out:
-        return 0;
-
+    return 0;
 }
 
 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_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++;
                 }
-                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)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        switch (local->transaction.type) {
-        case AFR_DATA_TRANSACTION:
-        case AFR_METADATA_TRANSACTION:
-                memcpy (int_lock->inode_locked_nodes,
-                        int_lock->locked_nodes,
-                        priv->child_count);
-                int_lock->inodelk_lock_count = int_lock->lock_count;
-                break;
-
-        case AFR_ENTRY_RENAME_TRANSACTION:
-        case AFR_ENTRY_TRANSACTION:
-                /*entrylk_count is being used in both non-blocking and blocking
-                 * modes */
-                break;
+                UNLOCK(&local->inode->lock);
+            }
         }
+        afr_lock_blocking(frame, this, cky + 1);
+    }
 
-        return 0;
-
+    return 0;
 }
 
-static inline gf_boolean_t
-afr_is_entrylk (afr_internal_lock_t *int_lock,
-                afr_transaction_type trans_type)
+static gf_boolean_t
+_is_lock_wind_needed(afr_local_t *local, int child_index)
 {
-        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) {
-
-                is_entrylk = _gf_true;
-
-        } else if ((int_lock->transaction_lk_type == AFR_TRANSACTION_LK) &&
-                 (trans_type == AFR_ENTRY_TRANSACTION ||
-                  trans_type == AFR_ENTRY_RENAME_TRANSACTION)) {
+    if (!local->child_up[child_index])
+        return _gf_false;
 
-                is_entrylk = _gf_true;
-
-        } else {
-                is_entrylk = _gf_false;
-        }
-
-        return is_entrylk;
+    return _gf_true;
 }
 
 static gf_boolean_t
-_is_lock_wind_needed (afr_local_t *local, int child_index)
-{
-        if (!local->child_up[child_index])
-                return _gf_false;
+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;
+        }
+        if (ret)
+            return ret;
+    }
+    if (!ret)
+        afr_log_locks_failure(frame, "all", "lock", int_lock->lock_op_errno);
 
-        return _gf_true;
+    return ret;
 }
 
 int
-afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
+afr_lock_blocking(call_frame_t *frame, xlator_t *this, int cookie)
 {
-        afr_internal_lock_t *int_lock    = 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;
-
-        flock.l_start = int_lock->lk_flock.l_start;
-        flock.l_len   = int_lock->lk_flock.l_len;
-        flock.l_type  = int_lock->lk_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) {
-                is_entrylk = afr_is_entrylk (int_lock, local->transaction.type);
-
-                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;
-                }
-        }
+    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;
 
-        if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
-                /* we're done locking */
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    priv = this->private;
+    child_index = cookie % priv->child_count;
+    lockee_num = cookie / priv->child_count;
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "we're done locking");
+    if (local->fd) {
+        ret = fd_ctx_get(local->fd, this, &ctx);
 
-                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,
-                                           this->name, 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,
-                                           this->name, &local->loc,
-                                           F_SETLKW, &flock, NULL);
-                }
+        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);
 
-                break;
+            local->op_ret = -1;
+            int_lock->lock_op_ret = -1;
 
-        case AFR_ENTRY_RENAME_TRANSACTION:
-        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,
-                                           this->name, 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,
-                                           this->name,
-                                           &int_lock->lockee[lockee_no].loc,
-                                           int_lock->lockee[lockee_no].basename,
-                                           ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
-                }
+            afr_unlock_now(frame, this);
 
-                break;
+            return 0;
         }
+    }
 
-        return 0;
-}
+    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;
 
-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;
-
-        switch (local->transaction.type) {
-        case AFR_DATA_TRANSACTION:
-        case AFR_METADATA_TRANSACTION:
-                initialize_inodelk_variables (frame, this);
-                break;
+            afr_unlock_now(frame, this);
 
-        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;
+            return 0;
         }
+    }
 
-        afr_lock_blocking (frame, this, 0);
+    if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
+        /* we're done locking */
 
-        return 0;
-}
-
-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);
-                }
-        }
+        gf_msg_debug(this->name, 0, "we're done locking");
 
+        int_lock->lock_op_ret = 0;
+        int_lock->lock_cbk(frame, this);
         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;
-
-                        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:
+    if (!_is_lock_wind_needed(local, child_index)) {
+        afr_lock_blocking(frame, this, cookie + 1);
         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_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;
-
-        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 {
-			int_lock->inode_locked_nodes[child_index]
-				|= LOCKED_YES;
-			int_lock->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 (int_lock->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);
-                }
-        }
+    afr_internal_lock_wind(frame, afr_lock_cbk, (void *)(long)cookie,
+                           child_index, lockee_num, _gf_true, _gf_false);
 
-        return 0;
-}
-
-int
-afr_nonblocking_inodelk (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;
-        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;
-
-        flock.l_start = int_lock->lk_flock.l_start;
-        flock.l_len   = int_lock->lk_flock.l_len;
-        flock.l_type  = int_lock->lk_flock.l_type;
-
-        full_flock.l_type = int_lock->lk_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;
-
-                        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 (!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,
-                                           this->name, 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,
-                                           this->name, &local->loc,
-                                           F_SETLK, &flock, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-out:
-        return ret;
+    return 0;
 }
 
 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);
-        }
-
-        return 0;
-}
-
-int
-afr_mark_locked_nodes (xlator_t *this, fd_t *fd,
-                       unsigned char *locked_nodes)
+afr_blocking_lock(call_frame_t *frame, xlator_t *this)
 {
-        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;
-}
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int up_count = 0;
 
-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;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        priv = this->private;
+    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);
 
-        list_for_each_entry (locked_fd, &priv->saved_fds, list) {
-                if (locked_fd->fd == fd) {
-                        found = 1;
-                        break;
-                }
-        }
+    afr_lock_blocking(frame, this, 0);
 
-        return found;
+    return 0;
 }
 
-static int
-__afr_save_locked_fd (xlator_t *this, fd_t *fd)
+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)
 {
-        afr_private_t   *priv      = NULL;
-        afr_locked_fd_t *locked_fd = NULL;
-        int              ret       = 0;
-
-        priv = this->private;
-
-        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);
+    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;
 
-        list_add_tail (&locked_fd->list, &priv->saved_fds);
+    priv = this->private;
 
-out:
-        return ret;
-}
-
-int
-afr_save_locked_fd (xlator_t *this, fd_t *fd)
-{
-        afr_private_t   *priv      = NULL;
-        int              ret       = 0;
+    child_index = ((long)cookie) % priv->child_count;
+    lockee_num = ((long)cookie) / priv->child_count;
 
-        priv = this->private;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        pthread_mutex_lock (&priv->mutex);
+    if (op_ret == 0 && local->transaction.type == AFR_DATA_TRANSACTION) {
+        LOCK(&local->inode->lock);
         {
-                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_local_t     *local     = NULL;
-        afr_locked_fd_t *locked_fd = NULL;
-
-        local = frame->local;
-
-        locked_fd = local->locked_fd;
-
-        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;
-
-        ret = fd_ctx_get (fd, this, &tmp);
-        if (ret)
-                goto out;
-
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
-
-        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;
-                }
+            local->inode_ctx->lock_count++;
         }
+        UNLOCK(&local->inode->lock);
+    }
 
-out:
-        return source_child;
-
-}
-
-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;
-        }
-
-        source_child = local->source_child;
-
-        memcpy (&flock, lock, sizeof (*lock));
-
-        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;
-
-cleanup:
-        afr_lock_recovery_cleanup (frame, this);
-        return 0;
+    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_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);
-
-        return 0;
-}
-
-static int
-is_afr_lock_eol (struct gf_flock *lock)
-{
-        int ret = 0;
-
-        if ((lock->l_type == GF_LK_EOL))
-                ret = 1;
-
-        return ret;
-}
-
-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;
+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;
         }
-
-        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_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;
-
-        fdctx->opened_on[child_index] = AFR_FD_OPENED;
-
-out:
-        return ret;
+    return ret;
 }
 
 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)
-{
-        int32_t child_index = (long )cookie;
-        int ret = 0;
-
-        if (op_ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Reopen during lock-recovery failed");
-                goto cleanup;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Open succeeded => proceed to recover locks");
-
-        ret = afr_lock_recovery (frame, this);
-        if (ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Lock recovery failed");
-                goto cleanup;
-        }
-
-        ret = afr_mark_fd_opened (this, fd, child_index);
-        if (ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Marking fd open failed");
-                goto cleanup;
-        }
-
-        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);
-
+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;
-}
-
-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;
-}
-
-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);
-                }
-        }
+    }
 
 out:
-        if ((ret < 0) && frame)
-                AFR_STACK_DESTROY (frame);
-        return ret;
-}
-
-void
-afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src,
-                          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;
-
-        dst_local = dst->local;
-        dst_lock  = &dst_local->internal_lock;
-        src_local = src->local;
-        src_lock  = &src_local->internal_lock;
-        if (src_lock->inode_locked_nodes) {
-                memcpy (dst_lock->inode_locked_nodes,
-                        src_lock->inode_locked_nodes,
-                        sizeof (*dst_lock->inode_locked_nodes) * child_count);
-                memset (src_lock->inode_locked_nodes, 0,
-                        sizeof (*src_lock->inode_locked_nodes) * child_count);
-        }
-
-        dst_lock->transaction_lk_type = src_lock->transaction_lk_type;
-        dst_lock->selfheal_lk_type    = src_lock->selfheal_lk_type;
-        dst_lock->inodelk_lock_count = src_lock->inodelk_lock_count;
-        src_lock->inodelk_lock_count = 0;
+    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 22e074571ed..00000000000
--- a/xlators/cluster/afr/src/afr-self-heal-algorithm.c
+++ /dev/null
@@ -1,827 +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)) {
-                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,
-                                    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_local_copy (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, sh_loop_lock_success, sh_loop_lock_failure);
-        return 0;
-out:
-        sh->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)) &&
-                       (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))
-                        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)) {
-                        // 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) {
-                sh->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));
-
-                sh->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);
-                sh->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) {
-                        sh->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));
-                sh->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)) {
-                        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;
-}
-
-void
-afr_sh_transfer_lock (call_frame_t *dst, call_frame_t *src,
-                      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;
-
-        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);
-        afr_lk_transfer_datalock (dst, src, child_count);
-        src_sh->data_lock_held = _gf_false;
-        dst_sh->data_lock_held = _gf_true;
-}
-
-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;
-        afr_sh_transfer_lock (first_loop_frame, sh_frame, priv->child_count);
-        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) {
-                sh->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 5f985374f29..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
@@ -8,2661 +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"
+#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);
-}
-
-void
-afr_sh_print_split_brain_log (int32_t *pending_matrix[], xlator_t *this,
-                              const char *loc)
-{
-        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) {
-                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:
-        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);
-        if (buf)
-                GF_FREE (buf);
-        return;
+    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_init_pending_matrix (int32_t **pending_matrix, size_t child_count)
+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)
 {
-        int             i   = 0;
-        int             j   = 0;
+    afr_local_t *local = NULL;
 
-        GF_ASSERT (pending_matrix);
+    local = frame->local;
 
-        for (i = 0; i < child_count; i++) {
-                for (j = 0; j < child_count; j++) {
-                        pending_matrix[i][j] = 0;
-                }
-        }
-}
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+    syncbarrier_wake(&local->barrier);
 
-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;
-                        }
-                }
-        }
+    return 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]);
-                }
-        }
+afr_selfheal_post_op(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     int subvol, dict_t *xattr, dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    loc_t loc = {
+        0,
+    };
+    int ret = 0;
 
-        return ret;
-}
+    priv = this->private;
+    local = frame->local;
 
-typedef enum {
-        AFR_NODE_INVALID,
-        AFR_NODE_INNOCENT,
-        AFR_NODE_FOOL,
-        AFR_NODE_WISE,
-} afr_node_type;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-typedef struct {
-        afr_node_type type;
-        int           wisdom;
-} afr_node_character;
+    local->op_ret = 0;
 
+    STACK_WIND(frame, afr_selfheal_post_op_cbk, priv->children[subvol],
+               priv->children[subvol]->fops->xattrop, &loc,
+               GF_XATTROP_ADD_ARRAY, xattr, xdata);
 
-static int
-afr_sh_is_innocent (int32_t *array, int child_count)
-{
-        int i   = 0;
-        int ret = 1;   /* innocent until proven guilty */
+    syncbarrier_wait(&local->barrier, 1);
+    if (local->op_ret < 0)
+        ret = -local->op_errno;
 
-        for (i = 0; i < child_count; i++) {
-                if (array[i]) {
-                        ret = 0;
-                        break;
-                }
-        }
+    loc_wipe(&loc);
+    local->op_ret = 0;
 
-        return ret;
+    return ret;
 }
 
-
-static int
-afr_sh_is_fool (int32_t *array, int i, int child_count)
-{
-        return array[i];   /* fool if accuses itself */
+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_wise (int32_t *array, int i, 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)
 {
-        return !array[i];  /* wise if does not accuse itself */
-}
+    int i = (long)cookie;
+    afr_local_t *local = NULL;
 
+    local = frame->local;
 
-static int
-afr_sh_all_nodes_innocent (afr_node_character *characters,
-                           int child_count)
-{
-        int i   = 0;
-        int ret = 1;
+    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);
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_INNOCENT) {
-                        ret = 0;
-                        break;
-                }
-        }
+    syncbarrier_wake(&local->barrier);
 
-        return ret;
+    return 0;
 }
 
-
-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;
-                }
+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;
         }
+    }
 
-        return ret;
+    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;
 
-/*
- * 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.
- */
-
-static void
-afr_sh_compute_wisdom (int32_t *pending_matrix[],
-                       afr_node_character characters[], int child_count)
-{
-        int i = 0;
-        int j = 0;
+    afr_selfheal_extract_xattr(this, replies, type, input_dirty, input_matrix);
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type == AFR_NODE_WISE) {
-                        characters[i].wisdom = 1;
+    if (local->need_full_crawl)
+        afr_selfheal_extract_xattr(this, replies, AFR_DATA_TRANSACTION, NULL,
+                                   full_heal_mtx_in);
 
-                        for (j = 0; j < child_count; j++) {
-                                if ((characters[j].type == AFR_NODE_WISE)
-                                    && pending_matrix[j][i]) {
+    for (i = 0; i < priv->child_count; i++)
+        if (sinks[i] && !healed_sinks[i])
+            pending[i] = 1;
 
-                                        characters[i].wisdom = 0;
-                                }
-                        }
-                }
+    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];
+            }
         }
-}
-
+    }
 
-static int
-afr_sh_wise_nodes_conflict (afr_node_character *characters,
-                            int child_count)
-{
-        int i   = 0;
-        int ret = 1;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!pending[i])
+            output_dirty[i] = -input_dirty[i];
+    }
 
-        for (i = 0; i < child_count; i++) {
-                if ((characters[i].type == AFR_NODE_WISE)
-                    && characters[i].wisdom == 1) {
+    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;
 
-                        /* There is atleast one bona-fide wise node */
-                        ret = 0;
-                        break;
-                }
+        xattr = afr_selfheal_output_xattr(this, local->need_full_crawl, type,
+                                          output_dirty, output_matrix, i,
+                                          full_heal_mtx_out);
+        if (!xattr) {
+            continue;
         }
 
-        return ret;
-}
-
+        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);
+        }
 
-static int
-afr_sh_mark_wisest_as_sources (int sources[],
-                               afr_node_character *characters,
-                               int child_count)
-{
-        int nsources = 0;
-        int i        = 0;
+        afr_selfheal_post_op(frame, this, inode, i, xattr, xdata);
+        dict_unref(xattr);
+    }
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].wisdom == 1) {
-                        sources[i] = 1;
-                        nsources++;
-                }
-        }
+    if (xdata)
+        dict_unref(xdata);
 
-        return nsources;
+    return 0;
 }
 
-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;
-        }
+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 int32_t
-afr_find_biggest_witness_among_fools (int32_t *witnesses,
-                                      afr_node_character *characters,
-                                      int32_t child_count)
+void
+afr_replies_copy(struct afr_reply *dst, struct afr_reply *src, int count)
 {
-        int i               = 0;
-        int biggest_witness = -1;
-
-        GF_ASSERT (witnesses);
-        GF_ASSERT (characters);
-        GF_ASSERT (child_count > 0);
+    int i = 0;
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_FOOL)
-                        continue;
+    if (dst == src)
+        return;
 
-                if (biggest_witness < witnesses[i])
-                        biggest_witness = witnesses[i];
-        }
-        return biggest_witness;
+    for (i = 0; i < count; i++) {
+        afr_reply_copy(&dst[i], &src[i]);
+    }
 }
 
 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)
-{
-        int i        = 0;
-        int nsources = 0;
-
-        GF_ASSERT (sources);
-        GF_ASSERT (witnesses);
-        GF_ASSERT (characters);
-        GF_ASSERT (child_count > 0);
-
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_FOOL)
-                        continue;
-
-                if (witness == witnesses[i]) {
-                        sources[i] = 1;
-                        nsources++;
-                }
-        }
-        return nsources;
-}
-
-static int
-afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix,
-                                     afr_node_character *characters,
-                                     int child_count)
+afr_selfheal_fill_dirty(xlator_t *this, int *dirty, int subvol, int idx,
+                        dict_t *xdata)
 {
-        int32_t       biggest_witness = 0;
-        int           nsources        = 0;
-        int32_t       *witnesses      = NULL;
+    void *pending_raw = NULL;
+    int pending[3] = {
+        0,
+    };
 
-        GF_ASSERT (child_count > 0);
-
-        witnesses = GF_CALLOC (child_count, sizeof (*witnesses),
-                               gf_afr_mt_int32_t);
-        if (NULL == witnesses) {
-                nsources = -1;
-                goto out;
-        }
+    if (!dirty)
+        return 0;
 
-        afr_compute_witness_of_fools (witnesses, pending_matrix, characters,
-                                      child_count);
-        biggest_witness = afr_find_biggest_witness_among_fools (witnesses,
-                                                                characters,
-                                                                child_count);
-        nsources = afr_mark_fool_as_source_by_witness (sources, witnesses,
-                                                       characters, child_count,
-                                                       biggest_witness);
-out:
-        GF_FREE (witnesses);
-        return nsources;
-}
+    if (dict_get_ptr(xdata, AFR_DIRTY, &pending_raw))
+        return -1;
 
-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++;
-                }
-        }
-        return nsources;
-}
+    if (!pending_raw)
+        return -1;
 
-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;
-}
+    memcpy(pending, pending_raw, sizeof(pending));
 
-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;
+    dirty[subvol] = ntoh32(pending[idx]);
 
-        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;
+    return 0;
 }
 
 int
-afr_get_no_xattr_dir_read_child (xlator_t *this, int32_t *success_children,
-                                 struct iatt *bufs)
+afr_selfheal_fill_matrix(xlator_t *this, int **matrix, int subvol, int idx,
+                         dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-        int            i = 0;
-        int            child = -1;
-        int            read_child = -1;
+    int i = 0;
+    void *pending_raw = NULL;
+    int pending[3] = {
+        0,
+    };
+    afr_private_t *priv = NULL;
 
-        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;
 
-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;
-}
+    if (!matrix)
+        return 0;
 
-char *
-afr_get_character_str (afr_node_type type)
-{
-        char *character = NULL;
+    for (i = 0; i < priv->child_count; i++) {
+        if (dict_get_ptr(xdata, priv->pending_key[i], &pending_raw))
+            continue;
 
-        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;
-}
+        if (!pending_raw)
+            continue;
 
-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;
+        memcpy(pending, pending_raw, sizeof(pending));
 
-        GF_ASSERT ((child >= 0) && (child < child_count));
+        matrix[subvol][i] = ntoh32(pending[idx]);
+    }
 
-        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 0;
 }
 
 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_selfheal_extract_xattr(xlator_t *this, struct afr_reply *replies,
+                           afr_transaction_type type, int *dirty, int **matrix)
 {
-        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_private_t *priv = NULL;
+    int i = 0;
+    dict_t *xdata = NULL;
+    int idx = -1;
 
-        priv = this->private;
-        child_count = priv->child_count;
+    idx = afr_index_for_transaction_type(type);
 
-        if (afr_get_children_count (success_children, priv->child_count) == 0)
-                goto out;
-
-        if (!ignore_ignorant) {
-                ignorant_subvols = GF_CALLOC (sizeof (*ignorant_subvols),
-                                              child_count, gf_afr_mt_char);
-                if (NULL == ignorant_subvols)
-                        goto out;
-        }
+    priv = this->private;
 
-        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 (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
 
-        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;
+        if (!replies[i].xdata)
+            continue;
 
-        afr_sh_print_pending_matrix (pending_matrix, this);
+        xdata = replies[i].xdata;
 
-        nsources = afr_mark_sources (this, sources, pending_matrix, bufs,
-                                     sh_type, success_children, subvol_status);
-out:
-        GF_FREE (ignorant_subvols);
-        return nsources;
-}
+        afr_selfheal_fill_dirty(this, dirty, i, idx, xdata);
+        afr_selfheal_fill_matrix(this, matrix, i, idx, xdata);
+    }
 
-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;
-        }
+    return 0;
 }
 
-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;
-        }
-}
-/**
- * mark_sources: Mark all 'source' nodes and return number of source
- * nodes found
- *
- * A node (a row in the pending matrix) belongs to one of
- * three categories:
- *
- * M is the pending matrix.
- *
- * '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.
- *
- * All 'innocent' nodes are sinks. If all nodes are innocent, no self-heal is
- * needed.
- *
- * 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 by chance there are multiple sources with differing sizes, select
+ * the largest file as the source.
  *
- * All fools are sinks, unless there are no 'wise' nodes. In that case,
- * one of the fools is made a 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;
 
-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;
+    /* 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 (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,
-                                                                child_count);
+        if (size <= replies[i].poststat.ia_size) {
+            size = replies[i].poststat.ia_size;
         }
+    }
 
-out:
-        if (nsources == 0)
-                afr_mark_success_children_sources (sources, success_children,
-                                                   child_count);
-        GF_FREE (characters);
-
-        gf_log (this->name, GF_LOG_DEBUG, "Number of sources: %d", nsources);
-        return nsources;
+    /* 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_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_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;
+        }
+    }
 }
 
-
-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;
-}
-
-
-int
-afr_sh_missing_entries_done (call_frame_t *frame, xlator_t *this)
+void
+afr_mark_active_sinks(xlator_t *this, unsigned char *sources,
+                      unsigned char *locked_on, unsigned char *sinks)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
+    int i = 0;
+    afr_private_t *priv = NULL;
 
-        local = frame->local;
-        sh = &local->self_heal;
+    priv = this->private;
 
-        afr_sh_reset (frame, this);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i] && locked_on[i])
+            sinks[i] = 1;
+        else
+            sinks[i] = 0;
+    }
+}
 
-        if (local->unhealable) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "split brain found, aborting selfheal of %s",
-                        local->loc.path);
-        }
+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;
 
-        if (is_self_heal_failed (sh)) {
-                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;
+    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);
 
-        return 0;
+    return _gf_true;
 }
 
-
-static int
-afr_sh_missing_entries_finish (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_can_decide_split_brain_source_sinks(struct afr_reply *replies,
+                                        int child_count)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local = frame->local;
-        int_lock = &local->internal_lock;
+    int i = 0;
 
-        int_lock->lock_cbk = afr_sh_missing_entries_done;
-        afr_unlock (frame, this);
+    for (i = 0; i < child_count; i++)
+        if (replies[i].valid != 1 || replies[i].op_ret != 0)
+            return _gf_false;
 
-        return 0;
+    return _gf_true;
 }
 
 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)
+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;
-        sh->parentbufs = GF_CALLOC (child_count, sizeof (*sh->parentbufs),
-                                    gf_afr_mt_iatt);
-        if (!sh->parentbufs)
+            }
+    }
+    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;
-        sh->child_errno = GF_CALLOC (child_count, sizeof (*sh->child_errno),
-                                     gf_afr_mt_int);
-        if (!sh->child_errno)
+            }
+            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;
-        sh->success_children = afr_children_create (child_count);
-        if (!sh->success_children)
+            }
+            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;
-        sh->fresh_children = afr_children_create (child_count);
-        if (!sh->fresh_children)
+            }
+            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;
-        sh->xattr = GF_CALLOC (child_count, sizeof (*sh->xattr),
-                               gf_afr_mt_dict_t);
-        if (!sh->xattr)
+            }
+            break;
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_BRICK:
+            ret = dict_get_str_sizen(xdata_req, "child-name", &name);
+            if (ret)
                 goto out;
-        ret = 0;
+            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:
-        return ret;
+    if (ret < 0)
+        memset(sources, 0, sizeof(*sources) * priv->child_count);
+    return ret;
 }
 
-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;
+int
+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;
+            }
         }
-        UNLOCK (&frame->lock);
-        return;
+    }
+    return fav_child;
 }
 
-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;
+/*
+ * afr_sh_fav_by_mtime: Choose favorite child by mtime.
+ */
+int
+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;
 }
 
+/*
+ * afr_sh_fav_by_ctime: Choose favorite child by ctime.
+ */
 int
-afr_impunge_frame_create (call_frame_t *frame, xlator_t *this,
-                          int active_source, call_frame_t **impunge_frame)
-{
-        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;
-
-        op_errno = ENOMEM;
-        priv = this->private;
-        new_frame = copy_frame (frame);
-        if (!new_frame) {
-                goto out;
-        }
-
-        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;
-
-        impunge_local->pending = afr_matrix_create (priv->child_count,
-                                                    AFR_NUM_CHANGE_LOGS);
-        if (!impunge_local->pending)
-                goto out;
-
-        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;
+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;
 }
 
-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));
-                sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        }
-        afr_sh_missing_entries_finish (frame, this);
+/*
+ * 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_sh_create_entry_cbk (call_frame_t *frame, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno)
+afr_sh_get_fav_by_policy(xlator_t *this, struct afr_reply *replies,
+                         inode_t *inode, char **policy_str)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        if (op_ret < 0)
-                sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        afr_sh_missing_entries_finish (frame, this);
-        return 0;
-}
+    afr_private_t *priv = NULL;
+    int fav_child = -1;
 
-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;
-        }
+    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;
+    }
 
-        afr_sh_missing_entry_call_impunge_recreate (frame, this,
-                                                    buf, postparent,
-                                                    afr_sh_create_entry_cbk);
-out:
-        return 0;
+    return fav_child;
 }
 
-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;
-        }
+int
+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;
+}
 
-        //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;
-                }
-        }
+gf_boolean_t
+afr_is_file_empty_on_all_children(afr_private_t *priv,
+                                  struct afr_reply *replies)
+{
+    int i = 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;
-        }
+    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;
+    }
 
-        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:
-        sh->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;
+    return _gf_true;
 }
 
-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 ((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;
-        }
-
-        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;
+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;
 
-done:
-        sh->lookup_done (frame, this, op_ret, op_errno);
-out:
-        return 0;
+    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;
+        }
+    }
+
+    return source;
 }
 
 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);
-                        sh->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);
+_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;
-}
 
-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;
-        }
+    xdata = dict_new();
+    if (!xdata)
+        return -1;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (expunge_local, 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);
 
-        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);
-}
+    afr_selfheal_extract_xattr(this, replies, type, input_dirty, input_matrix);
 
-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;
-                }
-        }
-}
+    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];
+    }
 
-int
-afr_sh_purge_stale_entries_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-        afr_private_t       *priv     = NULL;
+    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);
 
-        local    = frame->local;
-        sh       = &local->self_heal;
-        priv     = this->private;
+        afr_selfheal_post_op(frame, this, inode, i, xattr, xdata);
 
-        if (is_self_heal_failed (sh)) {
-                afr_sh_missing_entries_finish (frame, this);
-        } 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);
-                }
-        }
-        return 0;
-}
-
-gf_boolean_t
-afr_sh_purge_entry_condition (afr_local_t *local, afr_private_t *priv,
-                              int child)
-{
-        afr_self_heal_t *sh = NULL;
-
-        sh = &local->self_heal;
+        undid_pending[i] = 1;
+        dict_unref(xattr);
+    }
 
-        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;
+    if (xdata)
+        dict_unref(xdata);
 
-        return _gf_false;
+    return 0;
 }
 
 gf_boolean_t
-afr_sh_purge_stale_entry_condition (afr_local_t *local, afr_private_t *priv,
-                                    int child)
+afr_does_witness_exist(xlator_t *this, uint64_t *witness)
 {
-        afr_self_heal_t *sh = NULL;
+    int i = 0;
+    afr_private_t *priv = NULL;
 
-        sh = &local->self_heal;
+    priv = this->private;
 
-        if (local->child_up[child] &&
-            (!afr_is_child_present (sh->fresh_children, priv->child_count,
-                                    child))
-             && (sh->child_errno[child] != ENOENT))
-                return _gf_true;
+    for (i = 0; i < priv->child_count; i++) {
+        if (witness[i])
+            return _gf_true;
+    }
+    return _gf_false;
+}
 
-        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_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_selfheal_post_op_failure_accounting(afr_private_t *priv, char *accused,
+                                        unsigned char *sources,
+                                        unsigned char *locked_on)
 {
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv = NULL;
-        afr_self_heal_t *sh = NULL;
-        int             i = 0;
-        int             call_count = 0;
+    int i = 0;
+    unsigned int quorum_count = 0;
 
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
+    if (AFR_COUNT(sources, priv->child_count) != 0)
+        return;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (purge_condition (local, priv, i))
-                        call_count++;
+    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;
+}
 
-        if (call_count == 0) {
-                sh->post_remove_call (frame, this);
-                goto out;
-        }
+/*
+ * This function determines if a self-heal is required for a given inode,
+ * and if needed, in what direction.
+ *
+ * locked_on[] is the array representing servers which have been locked and
+ * from which xattrs have been fetched for analysis.
+ *
+ * The output of the function is by filling the arrays sources[] and sinks[].
+ *
+ * sources[i] is set if i'th server is an eligible source for a selfheal.
+ *
+ * sinks[i] is set if i'th server needs to be healed.
+ *
+ * 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.
+ *
+ */
 
-        local->call_count = call_count;
+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 *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++) {
-                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);
+            for (j = 0; j < priv->child_count; j++)
+                if (matrix[i][j])
+                    *pflag |= PFLAG_PENDING;
+            if (*pflag)
+                break;
         }
-out:
-        return;
+    }
+
+    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];
+        }
+    }
+
+    /* count the number of dirty fops witnessed */
+    for (i = 0; i < priv->child_count; i++)
+        witness[i] += dirty[i];
+
+    return 0;
 }
 
 void
-afr_sh_purge_entry (call_frame_t *frame, xlator_t *this)
+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_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;
-        afr_self_heal_t *sh = NULL;
+    afr_local_t *local = NULL;
+    int i = -1;
+    GF_UNUSED int ret = -1;
+    int8_t need_heal = 1;
+
+    local = frame->local;
+    i = (long)cookie;
 
-        local = frame->local;
-        sh = &local->self_heal;
-        sh->post_remove_call = afr_sh_missing_entries_finish;
+    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);
+    }
 
-        afr_sh_purge_entry_common (frame, this, afr_sh_purge_entry_condition);
+    local->replies[i].need_heal = need_heal;
+    syncbarrier_wake(&local->barrier);
+
+    return 0;
 }
 
-void
-afr_sh_purge_stale_entry (call_frame_t *frame, xlator_t *this)
+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)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int             i = 0;
+    loc_t loc = {
+        0,
+    };
+    dict_t *xattr_req = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    inode_t *inode = NULL;
 
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
+    local = frame->local;
+    priv = frame->this->private;
 
-        sh->post_remove_call = afr_sh_purge_stale_entries_done;
+    xattr_req = dict_new();
+    if (!xattr_req)
+        return NULL;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (afr_is_child_present (sh->fresh_children,
-                                          priv->child_count, i))
-                        continue;
+    if (xattr)
+        dict_copy(xattr, xattr_req);
 
-                if ((!local->child_up[i]) || sh->child_errno[i] != 0)
-                        continue;
+    if (afr_xattr_req_prepare(frame->this, xattr_req) != 0) {
+        dict_unref(xattr_req);
+        return NULL;
+    }
 
-                GF_ASSERT (!uuid_is_null (sh->entrybuf.ia_gfid) ||
-                           uuid_is_null (sh->buf[i].ia_gfid));
+    inode = inode_new(parent->table);
+    if (!inode) {
+        dict_unref(xattr_req);
+        return NULL;
+    }
 
-                if ((sh->entrybuf.ia_type != sh->buf[i].ia_type) ||
-                    (uuid_compare (sh->buf[i].ia_gfid,
-                                   sh->entrybuf.ia_gfid)))
-                        continue;
+    loc.parent = inode_ref(parent);
+    gf_uuid_copy(loc.pargfid, parent->gfid);
+    loc.name = name;
+    loc.inode = inode_ref(inode);
 
-                afr_children_add_child (sh->fresh_children, i,
-                                        priv->child_count);
+    AFR_ONLIST(lookup_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               xattr_req);
 
-        }
-        afr_sh_purge_entry_common (frame, this,
-                                   afr_sh_purge_stale_entry_condition);
-}
+    afr_replies_copy(replies, local->replies, priv->child_count);
 
-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);
+    loc_wipe(&loc);
+    dict_unref(xattr_req);
+
+    return inode;
 }
 
-void
-afr_get_children_of_fresh_parent_dirs (afr_self_heal_t *sh,
-                                       unsigned int child_count)
+static int
+afr_set_multi_dom_lock_count_request(xlator_t *this, dict_t *dict)
 {
-        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);
+    int ret = 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));
+
+    ret = dict_set_uint32(dict, GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS, 1);
+    if (ret)
+        return ret;
+
+    sprintf(key1, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, this->name);
+    ret = dict_set_uint32(dict, key1, 1);
+    if (ret)
+        return ret;
+
+    sprintf(key2, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, priv->sh_domain);
+    ret = dict_set_uint32(dict, key2, 1);
+    if (ret)
+        return ret;
+
+    return 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);
-                sh->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;
-        }
+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)
+{
+    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;
+    }
 
-        return;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, gfid);
 
-fail:
-        sh->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(discover_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               xattr_req);
+
+    afr_replies_copy(replies, local->replies, priv->child_count);
+
+    loc_wipe(&loc);
+    dict_unref(xattr_req);
+
+    return 0;
 }
 
-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;
-        }
+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;
 
-        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;
-        }
+    local = frame->local;
 
-        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;
+    if (local->xattr_req)
+        dict = local->xattr_req;
 
-out:
-        afr_sh_set_error (sh, op_errno);
-        sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-	afr_sh_missing_entries_finish (frame, this);
-        return;
+    return afr_selfheal_unlocked_discover_on(frame, inode, gfid, replies,
+                                             local->child_up, dict);
 }
 
-void
-afr_sh_common_reset (afr_self_heal_t *sh, unsigned int child_count)
+unsigned int
+afr_success_count(struct afr_reply *replies, unsigned int count)
 {
-        int             i = 0;
+    int i = 0;
+    unsigned int success = 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);
+    for (i = 0; i < count; i++)
+        if (replies[i].valid && replies[i].op_ret == 0)
+            success++;
+    return success;
 }
 
-/* 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);
-                }
-        }
+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;
 
-        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;
-                }
-        }
+    local = frame->local;
+    i = (long)cookie;
 
-        if (xattr_req)
-                dict_unref (xattr_req);
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
 
-        return 0;
+    syncbarrier_wake(&local->barrier);
+
+    return 0;
 }
 
+int
+afr_locked_fill(call_frame_t *frame, xlator_t *this, unsigned char *locked_on)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int count = 0;
 
+    local = frame->local;
+    priv = this->private;
 
-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.");
-                sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                afr_sh_missing_entries_done (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 {
-
-                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);
+            locked_on[i] = 0;
         }
+    }
 
-        return 0;
+    return count;
 }
 
 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_tryinodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                        char *dom, off_t off, size_t size,
+                        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,
+    };
+    struct gf_flock flock = {
+        0,
+    };
 
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
-        int_lock->selfheal_lk_type    = AFR_ENTRY_SELF_HEAL_LK;
+    flock.l_type = F_WRLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-        afr_set_lock_number (frame, this);
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLK, &flock,
+              NULL);
 
-        int_lock->lk_basename = base_name;
-        int_lock->lk_loc      = loc;
-        int_lock->lock_cbk    = lock_cbk;
+    loc_wipe(&loc);
 
-        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);
-
-        return 0;
+    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_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_internal_lock_t *int_lock = NULL;
-        int                 ret       = -1;
-        int32_t             op_errno  = 0;
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
+    afr_local_t *local = NULL;
+    int i = 0;
+    afr_private_t *priv = NULL;
 
-        local    = frame->local;
-        sh       = &local->self_heal;
+    priv = this->private;
+    local = frame->local;
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "attempting to recreate missing entries for path=%s",
-                local->loc.path);
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        ret = afr_build_parent_loc (&sh->parent_loc, &local->loc, &op_errno);
-        if (ret)
-                goto out;
+    flock.l_type = F_WRLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-        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;
-}
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLK, &flock,
+              NULL);
 
-static int
-afr_self_heal_missing_entries (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = 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_uninodelk(frame, this, inode, dom, off, size,
+                                   locked_on);
 
-        local = frame->local;
-        sh = &local->self_heal;
+            AFR_SEQ(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLKW,
+                    &flock, NULL);
+            break;
+        }
+    }
 
-        sh->afr_set_self_heal_status  = afr_set_gfid_or_missing_entry_sh_status;
-        sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
+    loc_wipe(&loc);
 
-        afr_self_heal_parent_entrylk (frame, this,
-                                      afr_sh_post_nb_entrylk_missing_entry_sh_cbk);
-        return 0;
+    return afr_locked_fill(frame, this, locked_on);
+}
+
+static void
+afr_get_lock_and_eagain_counts(afr_private_t *priv, struct afr_reply *replies,
+                               int *lock_count, int *eagain_count)
+{
+    int i = 0;
+
+    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)++;
+        }
+    }
 }
 
-afr_local_t*
-afr_local_copy (afr_local_t *l, 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_private_t *priv = NULL;
-        afr_local_t   *lc     = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_self_heal_t *shc = 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;
 
-        priv = this->private;
+    priv = this->private;
+    local = frame->local;
 
-        sh = &l->self_heal;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        lc = mem_get0 (this->local_pool);
-        if (!lc)
-                goto out;
+    flock.l_type = F_WRLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-        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;
-
-        uuid_copy (shc->sh_gfid_req, sh->sh_gfid_req);
-        if (l->loc.path)
-                loc_copy (&lc->loc, &l->loc);
-
-        lc->child_up  = memdup (l->child_up,
-                                sizeof (*lc->child_up) * priv->child_count);
-        if (l->xattr_req)
-                lc->xattr_req = dict_ref (l->xattr_req);
-
-        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);
-        if (l->internal_lock.inode_locked_nodes)
-                lc->internal_lock.inode_locked_nodes =
-                        memdup (l->internal_lock.inode_locked_nodes,
-                                sizeof (*lc->internal_lock.inode_locked_nodes) * priv->child_count);
-        else
-                lc->internal_lock.inode_locked_nodes =
-                        GF_CALLOC (sizeof (*l->internal_lock.inode_locked_nodes),
-                                   priv->child_count,
-                                   gf_afr_mt_char);
-
-        if (l->internal_lock.locked_nodes)
-                lc->internal_lock.locked_nodes =
-                        memdup (l->internal_lock.locked_nodes,
-                                sizeof (*lc->internal_lock.locked_nodes) * priv->child_count);
-        else
-                lc->internal_lock.locked_nodes =
-                        GF_CALLOC (sizeof (*l->internal_lock.locked_nodes),
-                                   priv->child_count,
-                                   gf_afr_mt_char);
-
-        for (i = 0; i < l->internal_lock.lockee_count; i++) {
-                loc_copy (&lc->internal_lock.lockee[i].loc,
-                          &l->internal_lock.lockee[i].loc);
-
-                lc->internal_lock.lockee[i].locked_count =
-                        l->internal_lock.lockee[i].locked_count;
-
-                if (l->internal_lock.lockee[i].basename)
-                        lc->internal_lock.lockee[i].basename =
-                                gf_strdup (l->internal_lock.lockee[i].basename);
-
-                if (l->internal_lock.lockee[i].locked_nodes) {
-                        lc->internal_lock.lockee[i].locked_nodes =
-                                memdup (l->internal_lock.lockee[i].locked_nodes,
-                                        sizeof (*lc->internal_lock.lockee[i].locked_nodes) *
-                                        priv->child_count);
-                } else {
-                        lc->internal_lock.lockee[i].locked_nodes =
-                                GF_CALLOC (priv->child_count,
-                                           sizeof (*lc->internal_lock.lockee[i].locked_nodes),
-                                           gf_afr_mt_char);
-                }
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLK, &flock,
+              NULL);
 
-        }
-        lc->internal_lock.lockee_count = l->internal_lock.lockee_count;
+    afr_get_lock_and_eagain_counts(priv, local->replies, &lock_count,
+                                   &eagain_count);
 
-        lc->internal_lock.inodelk_lock_count =
-                l->internal_lock.inodelk_lock_count;
-        lc->internal_lock.entrylk_lock_count =
-                l->internal_lock.entrylk_lock_count;
+    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_SEQ(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLKW,
+                &flock, NULL);
+    }
 
-out:
-        return lc;
+    loc_wipe(&loc);
+
+    return afr_locked_fill(frame, this, locked_on);
 }
 
 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_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)
+{
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
 
-        afr_self_heal_type_str_get (sh, sh_type_str,
-                                    sizeof(sh_type_str));
-        if (is_self_heal_failed (sh) && !priv->shd.iamshd) {
-                loglevel = GF_LOG_ERROR;
-        } else {
-                loglevel = GF_LOG_DEBUG;
-        }
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        afr_log_self_heal_completion_status (local, loglevel);
+    flock.l_type = F_UNLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-        FRAME_SU_UNDO (bgsh_frame, afr_local_t);
+    AFR_ONLIST(locked_on, frame, afr_selfheal_lock_cbk, inodelk, dom, &loc,
+               F_SETLK, &flock, 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));
-        }
+    loc_wipe(&loc);
 
-        if (sh->background) {
-                LOCK (&priv->lock);
-                {
-                        priv->background_self_heals_started--;
-                }
-                UNLOCK (&priv->lock);
-        }
+    return 0;
+}
 
-        AFR_STACK_DESTROY (bgsh_frame);
+int
+afr_selfheal_tryentrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                        char *dom, const char *name, unsigned char *locked_on)
+{
+    loc_t loc = {
+        0,
+    };
 
-        return 0;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+              ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL);
+
+    loc_wipe(&loc);
+
+    return afr_locked_fill(frame, this, locked_on);
 }
 
 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);
+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;
 
-        sh_local        = afr_local_copy (local, this);
-        if (!sh_local)
-                goto out;
-        sh_frame->local = sh_local;
-        sh              = &sh_local->self_heal;
+    priv = this->private;
+    local = frame->local;
 
-        sh->inode       = inode_ref (inode);
-        sh->orig_frame  = frame;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        sh->completion_cbk = afr_self_heal_completion_cbk;
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+              ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL);
 
-        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;
+    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);
 
-        sh->pending_matrix = afr_matrix_create (priv->child_count,
-                                                priv->child_count);
-        if (!sh->pending_matrix)
-                goto out;
+            AFR_SEQ(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+                    ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
+            break;
+        }
+    }
 
-        sh->delta_matrix = afr_matrix_create (priv->child_count,
-                                              priv->child_count);
-        if (!sh->delta_matrix)
-                goto out;
+    loc_wipe(&loc);
 
-        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;
-        }
+    return afr_locked_fill(frame, this, locked_on);
+}
 
-        if (local->self_heal.background) {
-                LOCK (&priv->lock);
-                {
-                        if (priv->background_self_heals_started
-                            < priv->background_self_heal_count) {
-                                priv->background_self_heals_started++;
+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)
+{
+    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;
 
-                        } else {
-                                local->self_heal.background = _gf_false;
-                                sh->background = _gf_false;
-                        }
-                }
-                UNLOCK (&priv->lock);
-        }
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        if (!local->loc.parent) {
-                sh->do_missing_entry_self_heal = _gf_false;
-                sh->do_gfid_self_heal = _gf_false;
-        }
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+              ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL);
 
-        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);
+    afr_get_lock_and_eagain_counts(priv, local->replies, &lock_count,
+                                   &eagain_count);
 
-                afr_sh_missing_entries_done (sh_frame, this);
-        }
-        op_errno = 0;
+    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);
 
-out:
-        if (op_errno) {
-                orig_sh->unwind (frame, this, -1, op_errno, 1);
-                if (sh_frame)
-                        AFR_STACK_DESTROY (sh_frame);
-        }
-        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);
 }
 
-void
-afr_self_heal_type_str_get (afr_self_heal_t *self_heal_p, char *str,
-                            size_t size)
+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)
 {
-        GF_ASSERT (str && (size > strlen (" missing-entry gfid "
-                                          "meta-data data entry")));
+    loc_t loc = {
+        0,
+    };
 
-        if (self_heal_p->do_metadata_self_heal) {
-                snprintf (str, size, " meta-data");
-        }
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        if (self_heal_p->do_data_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str), " data");
-        }
+    AFR_ONLIST(locked_on, frame, afr_selfheal_lock_cbk, entrylk, dom, &loc,
+               name, ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
 
-        if (self_heal_p->do_entry_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str), " entry");
-        }
+    loc_wipe(&loc);
 
-        if (self_heal_p->do_missing_entry_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str),
-                         " missing-entry");
-        }
+    return 0;
+}
 
-        if (self_heal_p->do_gfid_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str), " gfid");
-        }
+gf_boolean_t
+afr_is_data_set(xlator_t *this, dict_t *xdata)
+{
+    return afr_is_pending_set(this, xdata, AFR_DATA_TRANSACTION);
 }
 
-afr_self_heal_type
-afr_self_heal_type_for_transaction (afr_transaction_type type)
+gf_boolean_t
+afr_is_metadata_set(xlator_t *this, dict_t *xdata)
 {
-        afr_self_heal_type sh_type = AFR_SELF_HEAL_INVALID;
+    return afr_is_pending_set(this, xdata, AFR_METADATA_TRANSACTION);
+}
 
-        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;
+gf_boolean_t
+afr_is_entry_set(xlator_t *this, dict_t *xdata)
+{
+    return afr_is_pending_set(this, xdata, AFR_ENTRY_TRANSACTION);
 }
 
+/*
+ * 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.
+ */
+
 int
-afr_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name)
+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);
+
+    return ret;
+}
+
+inode_t *
+afr_inode_find(xlator_t *this, uuid_t gfid)
 {
-        int   ret = -1;
-        uuid_t pargfid = {0};
+    inode_table_t *table = NULL;
+    inode_t *inode = NULL;
 
-        if (!child)
-                goto out;
+    table = this->itable;
+    if (!table)
+        return NULL;
 
-        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);
+    inode = inode_find(table, gfid);
+    if (inode)
+        return inode;
 
-        if (uuid_is_null (pargfid))
-                goto out;
+    inode = inode_new(table);
+    if (!inode)
+        return NULL;
 
-        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);
+    gf_uuid_copy(inode->gfid, gfid);
 
-        if (-1 == ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "asprintf failed while setting child path");
-        }
+    return inode;
+}
 
-        child->name = strrchr (child->path, '/');
-        if (child->name)
-                child->name++;
+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;
 
-        child->parent = inode_ref (parent->inode);
-        child->inode = inode_new (parent->inode->table);
-        uuid_copy (child->pargfid, pargfid);
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        return NULL;
+    }
 
-        if (!child->inode) {
-                ret = -1;
-                goto out;
-        }
+    local = AFR_FRAME_INIT(frame, (*op_errno));
+    if (!local) {
+        STACK_DESTROY(frame->root);
+        return NULL;
+    }
 
-        ret = 0;
-out:
-        if ((ret == -1) && child)
-                loc_wipe (child);
+    syncopctx_setfspid(&pid);
 
-        return ret;
-}
+    frame->root->pid = pid;
 
-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;
+    afr_set_lk_owner(frame, this, frame->root);
 
-        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;
-                }
-        }
+    return frame;
+}
 
-        afr_sh_delta_to_xattr (this, sh->delta_matrix, erase_xattr,
-                               priv->child_count, type);
+int
+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;
 
-        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;
-        }
+    priv = this->private;
 
-        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);
-                }
-        }
+    gf_uuid_copy(inode->gfid, replies[source].poststat.ia_gfid);
 
-        ret = 0;
-out:
-        if (erase_xattr) {
-                for (i = 0; i < priv->child_count; i++) {
-                        if (erase_xattr[i]) {
-                                dict_unref (erase_xattr[i]);
-                        }
-                }
-        }
+    xattr = dict_new();
+    if (!xattr)
+        return -ENOMEM;
 
-        GF_FREE (erase_xattr);
+    changelog = afr_mark_pending_changelog(priv, newentry, xattr,
+                                           replies[source].poststat.ia_type);
 
-        if (ret < 0) {
-                sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                finish (frame, this);
-        }
+    if (!changelog) {
+        ret = -ENOMEM;
+        goto out;
+    }
 
-        return 0;
+    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;
 }
 
-void
-afr_set_data_sh_status (afr_self_heal_t *sh, afr_self_heal_status status)
-{
-        xlator_t              *this = NULL;
-
-        this = THIS;
+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;
 
-        if (sh)
-                sh->afr_all_sh_status.data_self_heal = status;
-        else
-                gf_log_callingfn (this->name, GF_LOG_ERROR,
-                                  "Null self heal struct");
+out:
+    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_set_metadata_sh_status (afr_self_heal_t *sh, afr_self_heal_status status)
+int
+afr_selfheal(xlator_t *this, uuid_t gfid)
 {
-        xlator_t              *this = NULL;
+    int ret = -1;
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
 
-        this = THIS;
+    frame = afr_frame_create(this, NULL);
+    if (!frame)
+        return ret;
 
-        if (sh)
-                sh->afr_all_sh_status.metadata_self_heal = status;
-        else
-                gf_log_callingfn (this->name, GF_LOG_ERROR,
-                                  "Null self heal struct");
-}
+    local = frame->local;
+    local->xdata_req = dict_new();
 
-void
-afr_set_entry_sh_status (afr_self_heal_t *sh, afr_self_heal_status status)
-{
-        xlator_t              *this = NULL;
+    ret = afr_selfheal_do(frame, this, gfid);
 
-        this = THIS;
+    if (frame)
+        AFR_STACK_DESTROY(frame);
 
-        if (sh)
-                sh->afr_all_sh_status.entry_self_heal = status;
-        else
-                gf_log_callingfn (this->name, GF_LOG_ERROR,
-                                  "Null self heal struct");
+    return ret;
 }
-void
-afr_set_gfid_or_missing_entry_sh_status (afr_self_heal_t *sh,
-                                         afr_self_heal_status status)
+
+afr_local_t *
+__afr_dequeue_heals(afr_private_t *priv)
 {
-        xlator_t              *this = NULL;
+    afr_local_t *local = NULL;
 
-        this = THIS;
+    if (list_empty(&priv->heal_waiting))
+        goto none;
+    if ((priv->background_self_heal_count > 0) &&
+        (priv->healers >= priv->background_self_heal_count))
+        goto none;
 
-        if (sh)
-                sh->afr_all_sh_status.gfid_or_missing_entry_self_heal = status;
-        else
-                gf_log_callingfn (this->name, GF_LOG_ERROR,
-                                  "Null self heal struct");
+    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;
 }
 
-void
-afr_set_local_for_unhealable (afr_local_t *local)
+int
+afr_refresh_selfheal_wrap(void *opaque)
 {
-        afr_self_heal_t  *sh = NULL;
+    call_frame_t *heal_frame = opaque;
+    afr_local_t *local = heal_frame->local;
+    int ret = 0;
 
-        sh = &local->self_heal;
-
-        local->unhealable = 1;
-        if (sh->afr_set_self_heal_status)
-                sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+    ret = afr_selfheal(heal_frame->this, local->refreshinode->gfid);
+    return ret;
 }
 
 int
-is_self_heal_failed (afr_self_heal_t *sh)
+afr_refresh_heal_done(int ret, call_frame_t *frame, void *opaque)
 {
-        afr_sh_status_for_all_type  sh_status = sh->afr_all_sh_status;
+    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;
+
+    LOCK(&priv->lock);
+    {
+        list_del_init(&local->healer);
+        priv->healers--;
+        GF_ASSERT(priv->healers >= 0);
+        local = __afr_dequeue_heals(priv);
+    }
+    UNLOCK(&priv->lock);
 
-        int sh_failed = 0;
-        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;
+    AFR_STACK_DESTROY(heal_frame);
 
-        return sh_failed;
+    if (local)
+        afr_heal_synctask(this, local);
+    return 0;
 }
 
-char *
-get_sh_completion_status (afr_self_heal_status status)
+void
+afr_heal_synctask(xlator_t *this, afr_local_t *local)
 {
+    int ret = 0;
+    call_frame_t *heal_frame = NULL;
 
-        char *not_attempted       = " is not attempted";
-        char *failed              = " failed";
-        char *successfull_complt  = " is successfully completed";
-        char *result              = " has unknown status";
+    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);
+}
 
-        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 = successfull_complt;
-                        break;
+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);
 
-        return result;
+    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.");
+    }
 
+    return can_heal;
 }
 
-void
-afr_log_self_heal_completion_status (afr_local_t *local, gf_loglevel_t loglvl)
+int
+afr_choose_source_by_policy(afr_private_t *priv, unsigned char *sources,
+                            afr_transaction_type type)
 {
+    int source = -1;
+    int i = 0;
 
-        char *gfid_or_missing_entry_sh = NULL;
-        char *metadata_sh              = NULL;
-        char *data_sh                  = NULL;
-        char *entry_sh                 = NULL;
-
-        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;
+    /* 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;
 
-        this = THIS;
-
-        gfid_or_missing_entry_sh = get_sh_completion_status
-                                   (all_status.gfid_or_missing_entry_self_heal);
+            source = i;
+            goto out;
+        }
+    }
 
-        metadata_sh = get_sh_completion_status (all_status.metadata_self_heal);
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i]) {
+            source = i;
+            goto out;
+        }
+    }
+out:
+    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;
+}
 
-        data_sh = get_sh_completion_status (all_status.data_self_heal);
+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++;
+        }
+    }
 
-        entry_sh = get_sh_completion_status (all_status.entry_self_heal);
+    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;
+    }
 
-        gf_log (this->name, loglvl, "%s "
-                "gfid or missing entry self heal %s,"
-                " medatadata self heal %s,"
-                " data self heal %s,"
-                " entry self heal %s on  %s",
-                (sh->background ? "background" : "foreground"),
-                gfid_or_missing_entry_sh, metadata_sh, data_sh, entry_sh,
-                local->loc.path);
+    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 {
+            op_errno = ENOMEM;
+        }
+        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 329bb2f1ed0..00000000000
--- a/xlators/cluster/afr/src/afr-self-heal-common.h
+++ /dev/null
@@ -1,158 +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_SELF_HEAL_ENTRY,
-        AFR_SELF_HEAL_METADATA,
-        AFR_SELF_HEAL_DATA,
-        AFR_SELF_HEAL_INVALID = -1,
-} afr_self_heal_type;
-
-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,
-                    afr_lock_cbk_t lock_cbk);
-afr_local_t *
-afr_local_copy (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,
-                  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);
-
-void
-afr_set_data_sh_status (afr_self_heal_t *sh, afr_self_heal_status status);
-
-void
-afr_set_metadata_sh_status (afr_self_heal_t *sh, afr_self_heal_status staus);
-
-void
-afr_set_entry_sh_status (afr_self_heal_t *sh, afr_self_heal_status status);
-
-void
-afr_set_gfid_or_missing_entry_sh_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);
-#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 fc7f5e7ac4b..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
@@ -8,1500 +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"
+#include <glusterfs/byte-order.h>
+#include "protocol-common.h"
+#include "afr-messages.h"
+#include <glusterfs/events.h>
 
-int
-afr_sh_data_fail (call_frame_t *frame, xlator_t *this);
+#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 {
+            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;
 
-static inline gf_boolean_t
-afr_sh_data_proceed (unsigned int success_count)
-{
-        return (success_count >= AFR_SH_MIN_PARTICIPANTS);
+    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
+            return _gf_true;
+    }
+out:
+    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:
+    if (iovec)
+        GF_FREE(iovec);
+    if (iobref)
+        iobref_unref(iobref);
+    return zero_filled;
 }
 
-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;
+static int
+__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;
 
-        local = frame->local;
-        sh = &local->self_heal;
+    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;
 
-        sh->completion_cbk (frame, this);
+    for (i = 0; i < priv->child_count; i++) {
+        if (healed_sinks[i] && locked_on[i])
+            return _gf_true;
+    }
 
-        return 0;
+    return _gf_false;
 }
 
-
-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)
+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_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);
+    int ret = -1;
+    afr_private_t *priv = NULL;
+    unsigned char *data_lock = NULL;
 
-        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;
+    priv = this->private;
+    data_lock = alloca0(priv->child_count);
 
-        if (call_count == 0) {
-                afr_sh_data_done (frame, this);
-                return 0;
+    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;
         }
 
-        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;
+        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_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)
+static int
+afr_selfheal_data_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                        unsigned char *healed_sinks)
 {
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
 
-        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);
-        }
+    local = frame->local;
+    priv = this->private;
 
+    if (!priv->ensure_durability)
         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;
+    AFR_ONLIST(healed_sinks, frame, afr_sh_generic_fop_cbk, fsync, fd, 0, NULL);
 
-                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;
+    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_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)
+static int
+afr_data_self_heal_type_get(afr_private_t *priv, unsigned char *healed_sinks,
+                            int source, struct afr_reply *replies)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        int child_index = (long) cookie;
-
-        local = frame->local;
-        sh = &local->self_heal;
+    int type = AFR_SELFHEAL_DATA_FULL;
+    int i = 0;
 
-        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);
+    if (priv->data_self_heal_algorithm == AFR_SELFHEAL_DATA_DYNAMIC) {
+        type = AFR_SELFHEAL_DATA_FULL;
+        for (i = 0; i < priv->child_count; i++) {
+            if (!healed_sinks[i] && i != source)
+                continue;
+            if (replies[i].poststat.ia_size) {
+                type = AFR_SELFHEAL_DATA_DIFF;
+                break;
+            }
         }
-
-        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;
-}
-
-//Fun fact, lock_cbk is being used for both lock & unlock
-int
-afr_sh_data_unlock (call_frame_t *frame, xlator_t *this,
-                    afr_lock_cbk_t lock_cbk)
-{
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_self_heal_t     *sh       = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
-
-        GF_ASSERT (sh->data_lock_held);
-
-        sh->data_lock_held = _gf_false;
-        int_lock->lock_cbk = lock_cbk;
-        afr_unlock (frame, this);
-
-        return 0;
+    } else {
+        type = priv->data_self_heal_algorithm;
+    }
+    return type;
 }
 
-int
-afr_sh_data_finish (call_frame_t *frame, 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_local_t   *local = NULL;
-        afr_self_heal_t *sh = NULL;
+    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;
 
-        local = frame->local;
-        sh = &local->self_heal;
+    gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+           "performing data selfheal on %s", uuid_utoa(fd->inode->gfid));
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "finishing data selfheal of %s", local->loc.path);
-
-        if (sh->data_lock_held)
-                afr_sh_data_unlock (frame, this, afr_sh_data_close);
-        else
-                afr_sh_data_close (frame, this);
+    priv = this->private;
+    if (priv->arbiter_count) {
+        arbiter_sink_status = healed_sinks[ARBITER_BRICK_INDEX];
+        healed_sinks[ARBITER_BRICK_INDEX] = 0;
+    }
 
-        return 0;
-}
+    block = 128 * 1024 * priv->data_self_heal_window_size;
 
-int
-afr_sh_data_fail (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local = NULL;
-        afr_self_heal_t *sh = NULL;
+    type = afr_data_self_heal_type_get(priv, healed_sinks, source, replies);
 
-        local = frame->local;
-        sh = &local->self_heal;
+    iter_frame = afr_copy_frame(frame);
+    if (!iter_frame) {
+        ret = -ENOMEM;
+        goto out;
+    }
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "finishing failed data selfheal of %s", local->loc.path);
+    for (off = 0; off < replies[source].poststat.ia_size; off += block) {
+        if (AFR_COUNT(healed_sinks, priv->child_count) == 0) {
+            ret = -ENOTCONN;
+            goto out;
+        }
 
-        sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        if (sh->data_lock_held)
-                afr_sh_data_unlock (frame, this, afr_sh_data_close);
-        else
-                afr_sh_data_close (frame, this);
-        return 0;
-}
+        ret = afr_selfheal_data_block(iter_frame, this, fd, source,
+                                      healed_sinks, off, block, type, replies);
+        if (ret < 0)
+            goto out;
 
-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));
-                sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        AFR_STACK_RESET(iter_frame);
+        if (iter_frame->local == NULL) {
+            ret = -ENOTCONN;
+            goto out;
         }
+    }
+
+    ret = afr_selfheal_data_fsync(frame, this, fd, healed_sinks);
 
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if (is_self_heal_failed (sh)) {
-                        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,
-                                  afr_post_sh_big_lock_success,
-                                  afr_post_sh_big_lock_failure);
-        }
 out:
-        return 0;
-}
+    if (arbiter_sink_status)
+        healed_sinks[ARBITER_BRICK_INDEX] = arbiter_sink_status;
 
-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;
+    if (iter_frame)
+        AFR_STACK_DESTROY(iter_frame);
+    return ret;
 }
 
-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)
+static int
+__afr_selfheal_truncate_sinks(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                              unsigned char *healed_sinks, uint64_t size)
 {
-        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);
-                {
-                        sh->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;
-        }
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
 
-        call_count = afr_frame_return (frame);
-        if (call_count == 0) {
-                if (is_self_heal_failed (sh))
-                        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 = frame->local;
+    priv = this->private;
 
-        local->call_count = call_count;
-        for (i = 0; i < priv->child_count; i++) {
-                if (!sh->success[i] || sh->sources[i])
-                        continue;
+    /* 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);
 
-                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);
-        }
+    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;
 
-        return 0;
+    return 0;
 }
 
-static struct afr_sh_algorithm *
-sh_algo_from_name (xlator_t *this, char *name)
+gf_boolean_t
+afr_has_source_witnesses(xlator_t *this, unsigned char *sources,
+                         uint64_t *witness)
 {
-        int i = 0;
-
-        if (name == NULL)
-                goto out;
+    int i = 0;
+    afr_private_t *priv = NULL;
 
-        while (afr_self_heal_algorithms[i].name) {
-                if (!strcmp (name, afr_self_heal_algorithms[i].name)) {
-                        return &afr_self_heal_algorithms[i];
-                }
-
-                i++;
-        }
+    priv = this->private;
 
-out:
-        return NULL;
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i] && witness[i])
+            return _gf_true;
+    }
+    return _gf_false;
 }
 
-
-static int
-sh_zero_byte_files_exist (afr_local_t *local, int child_count)
+static gf_boolean_t
+afr_does_size_mismatch(xlator_t *this, unsigned char *sources,
+                       struct afr_reply *replies)
 {
-        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;
-                }
-        }
+    int i = 0;
+    afr_private_t *priv = NULL;
+    struct iatt *min = NULL;
+    struct iatt *max = NULL;
 
-        return ret;
-}
+    priv = this->private;
 
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
 
-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");
-                }
-        }
+        if (replies[i].op_ret < 0)
+            continue;
 
-        return algo;
-}
+        if (!sources[i])
+            continue;
 
+        if (AFR_IS_ARBITER_BRICK(priv, i) && (replies[i].poststat.ia_size == 0))
+            continue;
 
-int
-afr_sh_data_sync_prepare (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        struct afr_sh_algorithm *sh_algo = NULL;
+        if (!min)
+            min = &replies[i].poststat;
 
-        local = frame->local;
-        sh = &local->self_heal;
+        if (!max)
+            max = &replies[i].poststat;
 
-        sh->algo_completion_cbk = afr_sh_data_fsync;
-        sh->algo_abort_cbk      = afr_sh_data_fail;
+        if (min->ia_size > replies[i].poststat.ia_size)
+            min = &replies[i].poststat;
 
-        sh_algo = afr_sh_data_pick_algo (frame, this);
+        if (max->ia_size < replies[i].poststat.ia_size)
+            max = &replies[i].poststat;
+    }
 
-        sh->algo = sh_algo;
-        sh_algo->fn (frame, this);
+    if (min && max) {
+        if (min->ia_size != max->ia_size)
+            return _gf_true;
+    }
 
-        return 0;
+    return _gf_false;
 }
 
-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)
+static void
+afr_mark_biggest_witness_as_source(xlator_t *this, unsigned char *sources,
+                                   uint64_t *witness)
 {
-        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));
-                        sh->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);
-                }
-        }
-        UNLOCK (&frame->lock);
+    int i = 0;
+    afr_private_t *priv = NULL;
+    uint64_t biggest_witness = 0;
 
-        call_count = afr_frame_return (frame);
+    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];
+    }
 
-        if (call_count == 0) {
-                if (is_self_heal_failed (sh))
-                        afr_sh_data_fail (frame, this);
-                else
-                        afr_sh_data_sync_prepare (frame, this);
-        }
+    /* 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;
+    }
 
-        return 0;
+    return;
 }
 
-
-int
-afr_sh_data_trim_sinks (call_frame_t *frame, xlator_t *this)
+/* 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)
 {
-        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;
+    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;
 
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sources = sh->sources;
-        call_count = sh->active_sinks;
-
-        local->call_count = call_count;
-
-        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;
+        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_inode_set_read_ctx (afr_self_heal_t *sh, xlator_t *this)
-{
-        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;
-        }
-
-        /* detect changes not visible through pending flags -- JIC */
-        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;
-        }
+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);
 
-        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;
-}
-
-void
-afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
-{
-        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));
-                sh->unwound = _gf_true;
-        }
+    afr_mark_active_sinks(this, sources, locked_on, healed_sinks);
+    source = afr_choose_source_by_policy(priv, sources, AFR_DATA_TRANSACTION);
 
-        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;
-        }
-
-        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);
-
-        sh->actual_sh_started = _gf_true;
-        afr_sh_data_trim_sinks (frame, this);
+    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_data_fxattrop_fstat_done (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;
-        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));
-        }
-
-        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)) {
-
-                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);
-
-                sh->sources[priv->favorite_child] = 1;
+    int ret = -1;
+    int source = -1;
+    afr_private_t *priv = NULL;
+    uint64_t *witness = NULL;
 
-                nsources = afr_sh_source_count (sh->sources,
-                                                priv->child_count);
-        }
+    priv = this->private;
 
-        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);
+    ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies);
 
-                afr_sh_data_fail (frame, this);
-                return 0;
-        }
+    if (ret)
+        return ret;
 
-        afr_set_split_brain (this, sh->inode, DONT_KNOW, NO_SPB);
+    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;
 
-        ret = afr_sh_inode_set_read_ctx (sh, this);
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "No active sources found.");
+    /* 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);
 
-                afr_sh_data_fail (frame, this);
-                return 0;
-        }
-
-        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 {
-                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;
-}
+    source = __afr_selfheal_data_finalize_source(
+        frame, this, inode, sources, sinks, healed_sinks, locked_on,
+        undid_pending, replies, witness);
+    if (source < 0)
+        return -EIO;
 
-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 source;
 }
 
-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)
-{
-        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);
-        }
+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:
-        return 0;
-}
 
+    if (did_sh)
+        afr_log_selfheal(fd->inode->gfid, this, ret, "data", source, sources,
+                         healed_sinks);
+    else
+        ret = 1;
 
-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;
-
-        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;
-}
+    if (locked_replies)
+        afr_replies_wipe(locked_replies, priv->child_count);
 
-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 ret;
 }
 
 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)
+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)
 {
-        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;
-}
+    afr_local_t *local = NULL;
+    int i = (long)cookie;
 
+    local = frame->local;
 
-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;
-                }
-        }
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
 
-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);
-        }
+    syncbarrier_wake(&local->barrier);
 
-        return 0;
+    return 0;
 }
 
 int
-afr_sh_data_big_lock_success (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;
-
-        sh->data_lock_held = _gf_true;
-        afr_sh_data_fxattrop (frame, this);
-        return 0;
-}
+    priv = this->private;
 
-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;
+    fd_tmp = fd_create(inode, 0);
+    if (!fd_tmp)
+        return -ENOMEM;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        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));
+    frame = afr_frame_create(this, &ret);
+    if (!frame) {
+        ret = -ret;
+        fd_unref(fd_tmp);
+        goto out;
+    }
+    local = frame->local;
 
-                sh->data_lock_failure_handler (frame, this);
-        } else {
+    AFR_ONLIST(local->child_up, frame, afr_selfheal_data_open_cbk, open, &loc,
+               O_RDWR | O_LARGEFILE, fd_tmp, NULL);
 
-                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));
+    ret = -ENOTCONN;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
 
-                sh->data_lock_success_handler (frame, this);
+        if (local->replies[i].op_ret < 0) {
+            ret = -local->replies[i].op_errno;
+            continue;
         }
 
-        return 0;
-}
+        ret = 0;
+        break;
+    }
 
-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 {
+    if (ret < 0) {
+        fd_unref(fd_tmp);
+        goto out;
+    } else {
+        fd_bind(fd_tmp);
+    }
 
-                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);
-        }
-
-        return 0;
-}
-
-int
-afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this, off_t start, off_t len)
-{
-        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_SELFHEAL_LK;
-        int_lock->selfheal_lk_type    = AFR_DATA_SELF_HEAL_LK;
-
-        afr_set_lock_number (frame, this);
-
-        int_lock->lk_flock.l_start = start;
-        int_lock->lk_flock.l_len   = len;
-        int_lock->lk_flock.l_type  = F_WRLCK;
-        int_lock->lock_cbk         = afr_sh_data_post_nonblocking_inodelk_cbk;
-
-        afr_nonblocking_inodelk (frame, this);
-
-        return 0;
-}
-
-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;
-}
-
-int
-afr_post_sh_big_lock_failure (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;
-        afr_sh_set_timestamps (frame, this);
-        return 0;
-}
-
-
-int
-afr_sh_data_lock (call_frame_t *frame, xlator_t *this,
-                  off_t start, off_t len, gf_boolean_t block,
-                  afr_lock_cbk_t success_handler,
-                  afr_lock_cbk_t failure_handler)
-{
-        afr_local_t *   local = NULL;
-        afr_self_heal_t * sh  = NULL;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        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, start, len);
-}
-
-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;
-	gf_boolean_t	 block = _gf_true;
-
-        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));
-                        sh->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);
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if (is_self_heal_failed (sh)) {
-                        afr_sh_data_fail (frame, this);
-                        return 0;
-                }
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "fd for %s opened, commencing sync",
-                        local->loc.path);
-
-                /*
-                 * The read and write self-heal trigger codepaths do not provide
-                 * an unwind callback. We run a trylock in these codepaths
-                 * because we are sensitive to locking latency.
-                 */
-
-		block = sh->unwind ? _gf_true : _gf_false;
-                afr_sh_data_lock (frame, this, 0, 0, block,
-                                  afr_sh_data_big_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;
+    afr_private_t *priv = NULL;
+    unsigned char *locked_on = NULL;
+    int ret = 0;
+    inode_t *inode = fd->inode;
 
-        fd = fd_create (local->loc.inode, frame->root->pid);
-        sh->healing_fd = fd;
+    priv = this->private;
 
-        /* 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;
-        }
-
-        return 0;
-}
+    locked_on = alloca0(priv->child_count);
 
-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;
+    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;
         }
 
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        for (i = 0; i < priv->child_count ; i++) {
-                if (1 == local->child_up[i])
-                        sh->success[i] = 1;
-        }
+        ret = __afr_selfheal_data(frame, this, fd, locked_on);
+    }
+unlock:
+    afr_selfheal_uninodelk(frame, this, inode, priv->sh_domain, 0, 0,
+                           locked_on);
 
-        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;
-}
-
-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;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sh->afr_set_self_heal_status = afr_set_data_sh_status;
-
-        if (afr_can_start_data_self_heal (sh, priv)) {
-                sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
-                if (IA_ISREG (sh->type)) {
-                        afr_sh_data_open (frame, this);
-                } else {
-                        afr_sh_data_lock (frame, this, 0, 0, _gf_true,
-                                          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 14ccca21b8b..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
@@ -8,2330 +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;
-
-        if (sh->entries_skipped) {
-                sh->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)) {
+    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);
+    }
 
-                        next_active_source = i;
-                        break;
-                }
-        }
 out:
-        return next_active_source;
-}
-
-
-
-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_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);
+    loc_wipe(&loc);
+    loc_wipe(&loc2);
+    if (frame) {
+        AFR_STACK_DESTROY(frame);
+    }
 
-        return 0;
+    return ret;
 }
 
-
 int
-afr_sh_entry_expunge_unlink (call_frame_t *expunge_frame, xlator_t *this,
-                             int active_src)
+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;
-
-        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)
+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;
-
-        priv = this->private;
-        expunge_local = expunge_frame->local;
+    int ret = 0;
+    afr_private_t *priv = NULL;
+    int i = 0;
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "looking up %s on %s",
-                expunge_local->loc.path, priv->children[active_src]->name);
+    priv = this->private;
 
-        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)
-{
-        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 (!replies[source].valid)
+        return -EIO;
 
-        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);
+    /* 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;
 
-                if (postparent)
-                        expunge_sh->parentbuf = *postparent;
-
-                afr_sh_entry_expunge_purge (expunge_frame, this, active_src);
-
-                return 0;
-        }
-
-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;
-}
-
-
-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 ((strcmp (name, ".") == 0)
-            || (strcmp (name, "..") == 0)) {
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "skipping inspection of %s under %s",
-                        name, local->loc.path);
-                op_ret = 0;
-                goto out;
-        }
-
-        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);
-        }
-
-        return 0;
-}
-
-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;
+            if (!gf_uuid_compare(replies[i].poststat.ia_gfid,
+                                 replies[source].poststat.ia_gfid))
+                continue;
 
-        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;
+            ret = afr_selfheal_recreate_entry(frame, i, source, sources,
+                                              fd->inode, name, inode, replies);
         }
+        if (ret < 0)
+            break;
+    }
 
-        active_src = next_active_sink (frame, this, sh->active_source);
-        sh->active_source = active_src;
-
-        if (is_self_heal_failed (sh)) {
-                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)
-{
-        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;
+    return ret;
 }
 
-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_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)
 {
-        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;
-
-        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);
+    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;
 }
 
-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)
+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)
 {
-        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);
-        }
+    int ret = 0;
+    int i = 0;
+    int source = -1;
+    int src = -1;
+    afr_private_t *priv = NULL;
 
-        return 0;
-}
+    priv = this->private;
 
-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));
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid && replies[i].op_ret == 0) {
+            source = i;
+            break;
         }
+    }
 
-        call_count = afr_frame_return (setattr_frame);
-        if (call_count == 0)
-                AFR_STACK_DESTROY (setattr_frame);
+    if (source == -1) {
+        /* entry got deleted in the mean time? */
         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;
-
-        priv          = this->private;
-        impunge_local = impunge_frame->local;
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source || !healed_sinks[i])
+            continue;
 
-        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;
-        }
+        ret |= afr_selfheal_recreate_entry(frame, i, source, sources, fd->inode,
+                                           name, inode, replies);
+    }
 
-        afr_set_pending_dict (priv, xattr, impunge_local->pending, active_src,
-                              LOCAL_LAST);
-
-        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;
-        }
-
-        call_count = afr_frame_return (impunge_frame);
-        if (call_count == 0)
-                afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                                op_ret, op_errno);
-
-        return 0;
+    int i = 0;
+    int pending[3] = {
+        0,
+    };
+    void *pending_raw = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (!xdata)
+        return _gf_false;
+
+    /* 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;
+
+        if (!pending_raw)
+            continue;
+
+        memcpy(pending, pending_raw, sizeof(pending));
+        if (ntoh32(pending[idx]))
+            return _gf_true;
+    }
+
+    return _gf_false;
 }
 
-int
-afr_sh_entry_impunge_hardlink (call_frame_t *impunge_frame, xlator_t *this,
-                               int child_index)
+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;
-        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);
+    int i = 0;
+    int idx = 0;
+    afr_private_t *priv = NULL;
 
-        return 0;
-}
+    priv = this->private;
 
-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;
-}
+    if (!priv->esh_granular)
+        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;
-}
-
-int
-afr_sh_entry_impunge_mknod (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;
-        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);
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "creating missing file %s on %s",
-                impunge_local->loc.path,
-                priv->children[child_index]->name);
+    /* 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.
+     */
 
-        dict = dict_new ();
-        if (!dict)
-                gf_log (this->name, GF_LOG_ERROR, "Out of memory");
+    if (source != -1) {
+        if (is_full_heal_marker_present(this, replies[source].xdata, idx))
+            return _gf_true;
+    }
 
-        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);
+    /* 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->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);
+    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_mkdir (call_frame_t *impunge_frame, xlator_t *this,
-                            int child_index, struct iatt *stbuf)
+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;
-
-        int ret = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
+    afr_private_t *priv = NULL;
+    int source = -1;
+    int sources_count = 0;
+    int i = 0;
 
-        dict = dict_new ();
-        if (!dict) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                return 0;
-        }
-
-        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);
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "creating missing directory %s on %s",
-                impunge_local->loc.path,
-                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->mkdir,
-                           &impunge_local->loc,
-                           st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type),
-                           0, dict);
+    priv = this->private;
 
-        if (dict)
-                dict_unref (dict);
-
-        return 0;
-}
+    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;
+    }
 
-int
-afr_sh_entry_impunge_symlink (call_frame_t *impunge_frame, xlator_t *this,
-                              int child_index, const char *linkname)
-{
-        afr_private_t   *priv          = NULL;
-        afr_local_t     *impunge_local = NULL;
-        dict_t          *dict          = NULL;
-        struct iatt     *buf           = NULL;
-        int              ret           = 0;
+    source = afr_choose_source_by_policy(priv, sources, AFR_ENTRY_TRANSACTION);
 
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-
-        buf = &impunge_local->cont.dir_fop.buf;
-
-        dict = dict_new ();
-        if (!dict) {
-                afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                                -1, ENOMEM);
-                goto out;
+    /*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;
+            }
         }
+        return -1;
+    }
 
-        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;
+    return source;
 }
 
-
 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_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;
-        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;
-        }
-
-        afr_sh_entry_impunge_symlink (impunge_frame, this, child_index,
-                                      impunge_sh->linkname);
-
-        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 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_symlink_unlink (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,
-                "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);
-
+    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_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)
+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) && (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;
-        }
+    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_sink (call_frame_t *impunge_frame, xlator_t *this,
-                                    int child_index)
-{
-        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);
-
-        return 0;
-}
-
-
-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 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              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;
-
-out:
-        LOCK (&impunge_frame->lock);
+    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)
         {
-                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;
-}
+            offset = entry->d_off;
 
+            if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
+                continue;
 
-int
-afr_sh_entry_impunge_readlink (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;
-        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);
-
-        return 0;
-}
-
-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;
+    loc_wipe(&loc);
+    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;
-}
-
-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 ((strcmp (entry->d_name, ".") == 0)
-            || (strcmp (entry->d_name, "..") == 0)) {
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "skipping inspection of %s under %s",
-                        entry->d_name, local->loc.path);
-                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));
-                        sh->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_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;
-}
-
-
-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));
-                        sh->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_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;
+    if (args.frame)
+        AFR_STACK_DESTROY(args.frame);
+    return ret;
 }
 
-
-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_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) {
-                sh->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);
-                sh->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->afr_set_self_heal_status = afr_set_entry_sh_status;
-        if (local->self_heal.do_entry_self_heal && priv->entry_self_heal) {
-                sh->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 ac2d7fcc668..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
@@ -8,624 +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"
+#include <glusterfs/byte-order.h>
+#include "protocol-common.h"
+#include <glusterfs/events.h>
 
+#define AFR_HEAL_ATTR (GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE)
 
-int
-afr_sh_metadata_done (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_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);
-
-        return 0;
-}
-
-int
-afr_sh_metadata_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;
-
-        sh->afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        afr_sh_metadata_finish (frame, this);
-        return 0;
-}
-
-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;
+    return afr_is_xattr_ignorable(k);
 }
 
-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;
-}
-
+        ret = syncop_setxattr(priv->children[i], &loc, xattr, 0, NULL, NULL);
+        if (ret)
+            healed_sinks[i] = 0;
+    }
+    ret = 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);
-
-        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;
+    uint64_t ret;
 
-        priv = this->private;
-        local = frame->local;
-
-        if (op_ret < 0) {
-                afr_sh_metadata_sync_cbk (frame, cookie,
-                                          this, -1, op_errno, xdata);
-                goto out;
-        }
+    ret = (((uint64_t)(ia->ia_mtime)) * 1000000000) +
+          (uint64_t)(ia->ia_mtime_nsec);
 
-        i = (long) cookie;
-
-        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);
-
-        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);
+    afr_replies_wipe(replies, priv->child_count);
+    ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies);
 
- out:
-        return 0;
+out:
+    if (xattr)
+        dict_unref(xattr);
+    return ret;
 }
 
-int
-afr_sh_metadata_sync (call_frame_t *frame, xlator_t *this, dict_t *xattr)
+/*
+ * 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_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--;
+    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;
         }
 
-        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);
+        /* 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;
         }
 
-        return 0;
-}
-
-
-int
-afr_sh_metadata_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 (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;
+        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;
         }
 
-        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;
-        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) {
-                sh->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;
+        /* Metadata split brain, select one subvol
+           arbitrarily */
+        for (i = 0; i < priv->child_count; i++) {
+            if (locked_on[i] && healed_sinks[i]) {
+                sources[i] = 1;
+                healed_sinks[i] = 0;
+                break;
+            }
         }
-        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);
+    }
+
+    /* 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;
         }
-
-        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;
+    }
+
+    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;
         }
+    }
+    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;
+}
 
-        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;
+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 *pflag)
+{
+    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;
         }
+    }
 
-        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);
-        }
-
-        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;
-}
+    source = __afr_selfheal_metadata_finalize_source(
+        frame, this, inode, sources, sinks, healed_sinks, undid_pending,
+        locked_on, replies);
 
-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_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        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;
+        }
 
-        int_lock->lk_flock.l_start = LLONG_MAX - 1;
-        int_lock->lk_flock.l_len   = 0;
-        int_lock->lk_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->afr_set_self_heal_status = afr_set_metadata_sh_status;
-
-        if (afr_can_start_metadata_self_heal (sh, priv)) {
-                sh->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 8f3673acee9..a51f79b1f43 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -8,1815 +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;
+    }
+
+    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);
 
-        j = afr_index_for_transaction_type (type);
-        for (i = 0; i < child_count; i++) {
-                if (!participants[i])
-                        pending[i][j] = 0;
+            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;
-
-                break;
+    return fop_frame;
+}
 
-        case AFR_ENTRY_TRANSACTION:
-        case AFR_ENTRY_RENAME_TRANSACTION:
-                if (priv->entry_change_log)
-                        ret = 1;
+static void
+afr_save_lk_owner(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
 
-                break;
-        }
+    local = frame->local;
 
-        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;
+    afr_local_t *local = NULL;
 
-        priv  = this->private;
-        local = frame->local;
-        type  = local->transaction.type;
+    local = frame->local;
 
-        if (__changelog_enabled (priv, type)) {
-                switch (local->op) {
+    frame->root->lk_owner = local->saved_lk_owner;
+}
 
-                case GF_FOP_WRITE:
-                case GF_FOP_FTRUNCATE:
-                        op_ret = 1;
-                        break;
+void
+__mark_all_success(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i;
 
-                case GF_FOP_FLUSH:
-                        op_ret = 0;
-                        break;
+    local = frame->local;
+    priv = this->private;
 
-                default:
-                        op_ret = 1;
-                }
-        }
-
-        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) {
-			AFR_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;
-        }
-
-        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));
+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);
 
-        if (!rm_stale_children)
-                goto out;
-
-        afr_inode_rm_stale_children (this, inode, stale_children);
-out:
-        GF_FREE (stale_children);
+    if (entry) {
+        afr_ta_post_op_synctask(this, entry);
         return;
-}
-
-unsigned char*
-afr_locked_nodes_get (afr_transaction_type type, afr_internal_lock_t *int_lock)
-{
-        unsigned char *locked_nodes = NULL;
-        switch (type) {
-        case AFR_DATA_TRANSACTION:
-        case AFR_METADATA_TRANSACTION:
-                locked_nodes = int_lock->inode_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;
+    } 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);
+            }
         }
-        return 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_post_op_done(call_frame_t *frame, xlator_t *this)
 {
-        int           call_count = 0;
-        unsigned char *locked_nodes = NULL;
+    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;
+}
 
-        locked_nodes = afr_locked_nodes_get (type, int_lock);
-        GF_ASSERT (locked_nodes);
+static void
+afr_changelog_post_op_fail(call_frame_t *frame, xlator_t *this, int op_errno)
+{
+    afr_local_t *local = frame->local;
+    local->op_ret = -1;
+    local->op_errno = op_errno;
 
-        call_count = afr_locked_children_count (locked_nodes, child_count);
-        if (type == AFR_ENTRY_RENAME_TRANSACTION)
-                call_count *= 2;
+    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);
 
-        return call_count;
+    afr_changelog_post_op_done(frame, this);
 }
 
-int
-afr_changelog_post_op_call_count (afr_transaction_type type,
-                                  unsigned char *pre_op,
-                                  unsigned int child_count)
+unsigned char *
+afr_locked_nodes_get(afr_transaction_type type, afr_internal_lock_t *int_lock)
 {
-        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;
+    /*Because same set of subvols participate in all lockee
+     * entities*/
+    return int_lock->lockee[0].locked_nodes;
 }
 
-void
-afr_compute_txn_changelog (afr_local_t *local, afr_private_t *priv)
+int
+afr_changelog_call_count(afr_transaction_type type,
+                         unsigned char *pre_op_subvols,
+                         unsigned char *failed_subvols,
+                         unsigned int child_count)
 {
-        int             i = 0;
-        int             index = 0;
-        int32_t         postop = 0;
-        int32_t         preop = 1;
-        int32_t         **txn_changelog = NULL;
+    int i = 0;
+    int call_count = 0;
 
-        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);
+    for (i = 0; i < child_count; i++) {
+        if (pre_op_subvols[i] && !failed_subvols[i]) {
+            call_count++;
         }
+    }
+
+    if (type == AFR_ENTRY_RENAME_TRANSACTION)
+        call_count *= 2;
+
+    return call_count;
 }
 
-afr_xattrop_type_t
-afr_get_postop_xattrop_type (int32_t **pending, int optimized, int child,
-                             afr_transaction_type type)
+gf_boolean_t
+afr_txn_nothing_failed(call_frame_t *frame, xlator_t *this)
 {
-        int                     index = 0;
-        afr_xattrop_type_t      op = LOCAL_LAST;
-
-        index = afr_index_for_transaction_type (type);
-        if (optimized && !pending[child][index])
-                op = LOCAL_FIRST;
-        return op;
+    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_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");
+afr_handle_symmetric_errors(call_frame_t *frame, xlator_t *this)
+{
+    if (afr_is_symmetric_error(frame, this))
+        __mark_all_success(frame, this);
 }
 
-
 gf_boolean_t
-afr_txn_nothing_failed (call_frame_t *frame, xlator_t *this)
+afr_has_quorum(unsigned char *subvols, xlator_t *this, call_frame_t *frame)
 {
-	afr_private_t *priv = NULL;
-	afr_local_t *local = NULL;
-	int index = -1;
-	int i = 0;
+    unsigned int quorum_count = 0;
+    afr_private_t *priv = NULL;
+    unsigned int up_children_count = 0;
 
-	local = frame->local;
-	priv = this->private;
+    priv = this->private;
+    up_children_count = AFR_COUNT(subvols, priv->child_count);
 
-        index = afr_index_for_transaction_type (local->transaction.type);
+    if (afr_lookup_has_quorum(frame, up_children_count))
+        return _gf_true;
 
-	for (i = 0; i < priv->child_count; i++) {
-                if (local->pending[i][index] == 0)
-			return _gf_false;
+    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];
         }
+    }
 
-	return _gf_true;
-}
+    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;
 
-int
-afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this)
+    return _gf_false;
+}
+
+static gf_boolean_t
+afr_has_fop_quorum(call_frame_t *frame)
 {
-        afr_private_t * priv = this->private;
-        afr_internal_lock_t *int_lock = NULL;
-        int i          = 0;
-        int call_count = 0;
+    xlator_t *this = frame->this;
+    afr_local_t *local = frame->local;
+    unsigned char *locked_nodes = NULL;
 
-        afr_local_t *  local = NULL;
-        afr_fd_ctx_t  *fdctx = NULL;
-        dict_t        **xattr = NULL;
-        int            piggyback = 0;
-        int            nothing_failed = 1;
+    locked_nodes = afr_locked_nodes_get(local->transaction.type,
+                                        &local->internal_lock);
+    return afr_has_quorum(locked_nodes, this, NULL);
+}
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+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);
+}
 
-        __mark_non_participant_children (local->pending, priv->child_count,
-                                         local->transaction.pre_op,
-                                         local->transaction.type);
+gf_boolean_t
+afr_need_dirty_marking(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *local = NULL;
+    gf_boolean_t need_dirty = _gf_false;
 
-        if (local->fd)
-                afr_transaction_rm_stale_children (frame, this,
-                                                   local->fd->inode,
-                                                   local->transaction.type);
+    local = frame->local;
 
-        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 ();
-        }
+    if (!priv->quorum_count || !local->optimistic_change_log)
+        return _gf_false;
 
-        call_count = afr_changelog_post_op_call_count (local->transaction.type,
-                                                       local->transaction.pre_op,
-                                                       priv->child_count);
-        local->call_count = call_count;
+    if (local->transaction.type == AFR_DATA_TRANSACTION ||
+        local->transaction.type == AFR_METADATA_TRANSACTION)
+        return _gf_false;
 
-        if (local->fd)
-                fdctx = afr_fd_ctx_get (local->fd, this);
+    if (AFR_COUNT(local->transaction.failed_subvols, priv->child_count) ==
+        priv->child_count)
+        return _gf_false;
 
-        if (call_count == 0) {
-                /* no child is up */
-                int_lock->lock_cbk = local->transaction.done;
-                afr_unlock (frame, this);
-                goto out;
-        }
+    if (!afr_has_fop_cbk_quorum(frame))
+        need_dirty = _gf_true;
 
-	nothing_failed = afr_txn_nothing_failed (frame, this);
+    return need_dirty;
+}
 
-        afr_compute_txn_changelog (local , priv);
+void
+afr_handle_quorum(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    const char *file = NULL;
+    uuid_t gfid = {0};
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (!local->transaction.pre_op[i])
-                        continue;
+    local = frame->local;
+    priv = frame->this->private;
 
-                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;
+    if (priv->quorum_count == 0)
+        return;
 
-                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--;
-                }
+    /* If the fop already failed return right away to preserve errno */
+    if (local->op_ret == -1)
+        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
-                */
+    /*
+     * 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;
 
-                afr_set_postop_dict (local, this, xattr[i],
-                                     local->optimistic_change_log, i);
+    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;
+    }
+}
 
-                /* fall through */
+int
+afr_fill_ta_loc(xlator_t *this, loc_t *loc, gf_boolean_t is_gfid_based_fop)
+{
+    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;
+}
 
-                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 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;
+}
 
-                if (!--call_count)
-                        break;
+int **
+afr_set_changelog_xattr(afr_private_t *priv, unsigned char *pending,
+                        dict_t *xattr, afr_local_t *local)
+{
+    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;
         }
-
-out:
         for (i = 0; i < priv->child_count; i++) {
-                dict_unref (xattr[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;
         }
+    }
 
-        return 0;
+out:
+    return changelog;
 }
 
-
-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 void
+afr_ta_locked_xattrop_validate(afr_private_t *priv, afr_local_t *local,
+                               gf_boolean_t *valid)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = this->private;
-        int call_count  = -1;
-        int child_index = (long) cookie;
+    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;
+}
 
-        local = frame->local;
+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. */
+        }
 
-        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;
-                }
+        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);
 
-                call_count = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
+    if (changelog)
+        afr_matrix_cleanup(changelog, priv->child_count);
 
-        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);
-                }
-        }
+    loc_wipe(&loc);
 
-        return 0;
+    return ret;
 }
 
-int
-afr_changelog_pre_op (call_frame_t *frame, xlator_t *this)
+static int
+afr_ta_post_op_synctask(xlator_t *this, afr_local_t *local)
 {
-        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;
-
-        local = frame->local;
-        int_lock = &local->internal_lock;
-
-        xattr = alloca (priv->child_count * sizeof (*xattr));
-        memset (xattr, 0, (priv->child_count * sizeof (*xattr)));
+    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;
+}
 
-        for (i = 0; i < priv->child_count; i++) {
-                xattr[i] = dict_new ();
+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);
+}
 
-        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;
-        }
+static void
+afr_ta_fill_failed_subvol(afr_private_t *priv, afr_local_t *local)
+{
+    int i = 0;
 
-        local->call_count = call_count;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.failed_subvols[i]) {
+            local->ta_failed_subvol = i;
+            break;
+        }
+    }
+}
 
-        __mark_all_pending (local->pending, priv->child_count,
-                            local->transaction.type);
+static void
+afr_post_op_handle_success(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
 
-        if (local->fd)
-                fdctx = afr_fd_ctx_get (local->fd, this);
+    local = frame->local;
+    if (local->is_new_entry == _gf_true) {
+        afr_mark_new_entry_changelog(frame, this);
+    }
+    afr_changelog_post_op_do(frame, 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;
+    return;
+}
 
-                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--;
-                }
+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);
 
+    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_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;
+}
 
-                ret = afr_set_pending_dict (priv, xattr[i], local->pending,
-                                            i, LOCAL_FIRST);
+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 (ret < 0)
-                        gf_log (this->name, GF_LOG_INFO,
-                                "failed to set pending entry");
+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;
+        }
 
-                /* fall through */
+        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:
+    if (xattr)
+        dict_unref(xattr);
 
-                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;
-                }
+    return;
+}
 
-                if (!--call_count)
-                        break;
-        }
-out:
-        for (i = 0; i < priv->child_count; i++) {
-                dict_unref (xattr[i]);
+static int
+afr_changelog_post_op_now(call_frame_t *frame, xlator_t *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_post_blocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_changelog_pre_op_uninherit(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;
+    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;
+
+    /* This function must be idempotent. So check if we
+       were called before and return the same answer again.
+
+       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++) {
+            if (local->transaction.pre_op[i] != ctx->pre_op_done[type][i]) {
+                ret = !local->transaction.dirtied;
+                goto unlock;
+            }
+        }
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Blocking inodelks failed.");
-                local->transaction.done (frame, this);
+        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;
+        }
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Blocking inodelks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
+        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);
 
-        return 0;
-}
+    local->transaction.uninherit_done = _gf_true;
+    local->transaction.uninherit_value = ret;
 
+    return ret;
+}
 
-int
-afr_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_changelog_pre_op_inherit(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;
+    priv = this->private;
+
+    if (local->transaction.type != AFR_DATA_TRANSACTION)
+        return _gf_false;
+
+    type = afr_index_for_transaction_type(local->transaction.type);
+
+    LOCK(&local->inode->lock);
+    {
+        if (!local->inode_ctx->on_disk[type]) {
+            /* nothing to inherit yet */
+            ret = _gf_false;
+            goto unlock;
+        }
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+        for (i = 0; i < priv->child_count; 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;
+            }
+        }
 
-        /* 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 {
+        local->inode_ctx->inherited[type]++;
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking inodelks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
-        }
+        ret = _gf_true;
 
-        return 0;
-}
+        local->transaction.inherited = _gf_true;
+    }
+unlock:
+    UNLOCK(&local->inode->lock);
 
+    return ret;
+}
 
-int
-afr_post_blocking_entrylk_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;
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Blocking entrylks failed.");
-                local->transaction.done (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,
-                        "Blocking entrylks 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_nonblocking_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;
 
-        /* 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);
+    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,
-                        "Non 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_blocking_rename_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;
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Blocking entrylks failed.");
-                local->transaction.done (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,
-                        "Blocking entrylks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
-        }
-        return 0;
-}
+    if (xattr)
+        local->transaction.changelog_xdata[child_index] = dict_ref(xattr);
 
+    call_count = afr_frame_return(frame);
 
-int
-afr_post_lower_unlock_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    if (call_count == 0) {
+        local->transaction.changelog_resume(frame, this);
+    }
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    return 0;
+}
 
-        GF_ASSERT (!int_lock->higher_locked);
+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;
 
-        int_lock->lock_cbk = afr_post_blocking_rename_cbk;
-        afr_blocking_lock (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_set_transaction_flock (afr_local_t *local)
+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_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        int_lock = &local->internal_lock;
+    local = frame->local;
+    priv = this->private;
 
-        int_lock->lk_flock.l_len   = local->transaction.len;
-        int_lock->lk_flock.l_start = local->transaction.start;
-        int_lock->lk_flock.l_type  = F_WRLCK;
+    *call_count = afr_changelog_call_count(
+        local->transaction.type, local->transaction.pre_op,
+        local->transaction.failed_subvols, priv->child_count);
 
-        return 0;
+    if (*call_count == 0) {
+        changelog_resume(frame, this);
+        return -1;
+    }
+
+    afr_changelog_populate_xdata(frame, op, xdata, newloc_xdata);
+    local->call_count = *call_count;
+
+    local->transaction.changelog_resume = changelog_resume;
+    return 0;
 }
 
 int
-afr_lock_rec (call_frame_t *frame, xlator_t *this)
+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_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    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;
+        }
+    }
 
-        int_lock->transaction_lk_type = AFR_TRANSACTION_LK;
+    ret = afr_changelog_prepare(this, frame, &call_count, changelog_resume, op,
+                                &xdata, &newloc_xdata);
 
-        switch (local->transaction.type) {
-        case AFR_DATA_TRANSACTION:
-        case AFR_METADATA_TRANSACTION:
-                afr_set_transaction_flock (local);
+    if (ret)
+        return 0;
 
-                int_lock->lock_cbk = afr_post_nonblocking_inodelk_cbk;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.pre_op[i] ||
+            local->transaction.failed_subvols[i])
+            continue;
 
-                afr_nonblocking_inodelk (frame, this);
+        switch (local->transaction.type) {
+            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
-is_afr_delayed_changelog_post_op_needed (call_frame_t *frame, xlator_t *this)
+int
+afr_post_nonblocking_lock_cbk(call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t      *local = NULL;
-	gf_boolean_t      res = _gf_false;
+    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;
+}
 
-	local = frame->local;
-	if (!local)
-		goto out;
+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;
 
-	if (!local->delayed_post_op)
-		goto out;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-	res = _gf_true;
-out:
-	return res;
-}
+    if (int_lock->lock_op_ret < 0) {
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_INTERNAL_LKS_FAILED,
+               "Blocking entrylks failed.");
 
+        afr_transaction_done(frame, this);
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "Blocking entrylks done. Proceeding to FOP");
 
-void
-afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
-			       call_stub_t *stub);
+        afr_internal_lock_finish(frame, this);
+    }
+    return 0;
+}
 
-void
-afr_delayed_changelog_wake_up_cbk (void *data)
+int
+afr_post_lower_unlock_cbk(call_frame_t *frame, xlator_t *this)
 {
-	fd_t           *fd = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
 
-	fd = data;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-	afr_delayed_changelog_wake_up (THIS, fd);
-}
+    GF_ASSERT(!int_lock->higher_locked);
 
+    int_lock->lock_cbk = afr_post_blocking_rename_cbk;
+    afr_blocking_lock(frame, this);
 
-/*
-  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)
-{
-	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;
-
-	priv = frame->this->private;
-	local = frame->local;
-	fdctx = afr_fd_ctx_get (fd, frame->this);
-
-	LOCK(&fd->lock);
-	{
-		piggyback = _gf_true;
-
-		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]--;
-			}
-		}
-	}
-	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;
-        }
-
-        return piggyback;
+    return 0;
 }
 
+int
+afr_set_transaction_flock(xlator_t *this, afr_local_t *local,
+                          afr_lockee_t *lockee)
+{
+    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;
+}
 
-/* SET operation */
 int
-afr_fd_report_unstable_write (xlator_t *this, fd_t *fd)
+afr_lock(call_frame_t *frame, xlator_t *this)
 {
-	afr_fd_ctx_t *fdctx = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+
+    int_lock->lock_cbk = afr_post_nonblocking_lock_cbk;
+    int_lock->domain = this->name;
+
+    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]);
+            }
 
-	fdctx = afr_fd_ctx_get (fd, this);
+            break;
 
-	LOCK(&fd->lock);
-	{
-		fdctx->witnessed_unstable_write = _gf_true;
-	}
-	UNLOCK(&fd->lock);
+        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;
+    return 0;
 }
 
-/* TEST and CLEAR operation */
-gf_boolean_t
-afr_fd_has_witnessed_unstable_write (xlator_t *this, fd_t *fd)
+static gf_boolean_t
+afr_locals_overlap(afr_local_t *local1, afr_local_t *local2)
 {
-	afr_fd_ctx_t *fdctx = NULL;
-	gf_boolean_t witness = _gf_false;
+    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));
+}
 
-	fdctx = afr_fd_ctx_get (fd, this);
+gf_boolean_t
+afr_has_lock_conflict(afr_local_t *local, gf_boolean_t waitlist_check)
+{
+    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;
+        }
+    }
+
+    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;
+}
 
-	LOCK(&fd->lock);
-	{
-		if (fdctx->witnessed_unstable_write) {
-			witness = _gf_true;
-			fdctx->witnessed_unstable_write = _gf_false;
-		}
-	}
-	UNLOCK (&fd->lock);
+/* }}} */
+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));
+}
 
-	return witness;
+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);
+    }
 }
 
+static void
+afr_lock_resume_shared(struct list_head *list)
+{
+    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);
+    }
+}
 
 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_internal_lock_finish(call_frame_t *frame, xlator_t *this)
 {
-	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_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);
+        }
+    }
 
-		afr_transaction_fop_failed (frame, this, child_index);
-	}
+    return 0;
+}
 
-        call_count = afr_frame_return (frame);
+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;
+        }
+    } else if (local->transaction.type == AFR_DATA_TRANSACTION) {
+        if (lock->num_inodelks > 1) {
+            return _gf_true;
+        }
+    }
 
-        if (call_count == 0)
-		afr_changelog_post_op_now (frame, this);
+    return _gf_false;
+}
 
-	return 0;
+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_changelog_fsync (call_frame_t *frame, xlator_t *this)
+afr_fd_report_unstable_write(xlator_t *this, afr_local_t *local)
 {
-	afr_local_t *local = NULL;
-	int i = 0;
-	int call_count = 0;
-	afr_private_t *priv = NULL;
-
-	local = frame->local;
-	priv = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    LOCK(&local->inode->lock);
+    {
+        local->inode_ctx->witnessed_unstable_write = _gf_true;
+    }
+    UNLOCK(&local->inode->lock);
 
-	if (!call_count) {
-		/* will go straight to unlock */
-		afr_changelog_post_op_now (frame, this);
-		return 0;
-	}
+    return 0;
+}
 
-	local->call_count = call_count;
+/* TEST and CLEAR operation */
+gf_boolean_t
+afr_fd_has_witnessed_unstable_write(xlator_t *this, inode_t *inode)
+{
+    afr_inode_ctx_t *ctx = NULL;
+    gf_boolean_t witness = _gf_false;
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (!local->transaction.pre_op[i])
-			continue;
+    LOCK(&inode->lock);
+    {
+        (void)__afr_inode_ctx_get(this, inode, &ctx);
 
-		STACK_WIND_COOKIE (frame, afr_changelog_fsync_cbk,
-				   (void *) (long) i, priv->children[i],
-				   priv->children[i]->fops->fsync, local->fd,
-				   1, NULL);
-		if (!--call_count)
-			break;
-	}
+        if (ctx->witnessed_unstable_write) {
+            witness = _gf_true;
+            ctx->witnessed_unstable_write = _gf_false;
+        }
+    }
+    UNLOCK(&inode->lock);
 
-	return 0;
+    return witness;
 }
 
-
 int
-afr_changelog_post_op_safe (call_frame_t *frame, xlator_t *this)
+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_local_t    *local = NULL;
+    afr_private_t *priv = NULL;
+    int child_index = (long)cookie;
+    int call_count = -1;
+    afr_local_t *local = NULL;
 
-	local = frame->local;
+    priv = this->private;
+    local = frame->local;
 
-	if (!local->fd || local->transaction.type != AFR_DATA_TRANSACTION) {
-		afr_changelog_post_op_now (frame, this);
-		return 0;
-	}
+    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]);
 
-	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;
-	}
+        afr_transaction_fop_failed(frame, this, child_index);
+    }
 
-	/* Calling afr_changelog_post_op_now() now will result in
-	   issuing ->[f]xattrop().
+    call_count = afr_frame_return(frame);
 
-	   Performing a hard POST-OP (->[f]xattrop() FOP) is a more
-	   responsible operation that what it might appear on the surface.
+    if (call_count == 0)
+        afr_changelog_post_op_now(frame, this);
 
-	   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.
+    return 0;
+}
 
-	   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".
+int
+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;
 
-	   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.
+    local = frame->local;
+    priv = this->private;
 
-	   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.
-	*/
+    call_count = AFR_COUNT(local->transaction.pre_op, priv->child_count);
 
-	if (!afr_fd_has_witnessed_unstable_write (this, local->fd)) {
-		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;
+    }
 
-	/* Time to fsync() */
+    local->call_count = call_count;
 
-	afr_changelog_fsync (frame, this);
+    xdata = dict_new();
+    if (xdata) {
+        ret = dict_set_int32_sizen(xdata, "batch-fsync", 1);
+        ret = dict_set_str(xdata, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+    }
 
-	return 0;
-}
+    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;
+    }
 
-void
-afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
-			       call_stub_t *stub)
-{
-	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)
-		return;
-
-	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;
-	}
-unlock:
-	pthread_mutex_unlock (&fd_ctx->delay_lock);
+    if (xdata)
+        dict_unref(xdata);
 
-	if (prev_frame) {
-		local = prev_frame->local;
-		local->transaction.resume_stub = stub;
-		afr_changelog_post_op_safe (prev_frame, this);
-	} else if (stub) {
-		AFR_CALL_RESUME (stub);
-	}
+    return 0;
 }
 
-
-void
-afr_changelog_post_op (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_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-	local = frame->local;
+    local = frame->local;
+    priv = this->private;
 
-	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);
+    if (!local->fd || local->transaction.type != AFR_DATA_TRANSACTION) {
+        afr_changelog_post_op_now(frame, this);
+        return 0;
+    }
+
+    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.
+           */
+        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_changelog_post_op(call_frame_t *frame, xlator_t *this)
+{
+    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;
+        }
 
+        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:
+    UNLOCK(&local->inode->lock);
 
-/* Wake up the sleeping/delayed post-op, and also register
-   a stub to have it resumed after this transaction
-   completely finishes.
+    if (!list_empty(&shared)) {
+        afr_lock_resume_shared(&shared);
+    }
 
-   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);
+out:
+    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_delayed_changelog_wake_up (xlator_t *this, fd_t *fd)
+int
+afr_transaction_resume(call_frame_t *frame, xlator_t *this)
 {
-	afr_delayed_changelog_post_op (this, NULL, fd, NULL);
-}
+    afr_local_t *local = NULL;
 
+    local = frame->local;
 
-int
-afr_transaction_resume (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-	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);
-	}
-
-	afr_restore_lk_owner (frame);
-
-        if (__fop_changelog_needed (frame, this)) {
-                afr_changelog_post_op (frame, this);
-        } 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);
-                }
-        }
+    afr_restore_lk_owner(frame);
 
-        return 0;
-}
+    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);
 
+    afr_changelog_post_op(frame, this);
+
+    return 0;
+}
 
 /**
  * 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_fop_failed(call_frame_t *frame, xlator_t *this, int child_index)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
+    afr_local_t *local = NULL;
 
-        local = frame->local;
-        priv  = this->private;
+    local = frame->local;
 
-        __mark_child_dead (local->pending, priv->child_count,
-                           child_index, local->transaction.type);
+    local->transaction.failed_subvols[child_index] = 1;
 }
 
-
-
 static gf_boolean_t
-afr_locals_overlap (afr_local_t *local1, afr_local_t *local2)
+__need_previous_lock_unlocked(afr_local_t *local)
 {
-	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_lock_t *lock = NULL;
+
+    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;
 }
 
-
 void
-afr_transaction_eager_lock_init (afr_local_t *local, xlator_t *this)
+__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_private_t *priv = NULL;
-	afr_fd_ctx_t  *fdctx = NULL;
-	afr_local_t   *each = NULL;
-
-	priv = this->private;
-
-	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;
+    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;
+            }
+        }
+    }
+
+    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 {
+            *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;
+    }
 
-        /*
-         * 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;
-			}
-		}
-
-		local->transaction.eager_lock_on = _gf_true;
-		list_add_tail (&local->transaction.eager_locked,
-			       &fdctx->eager_locked);
-	}
-unlock:
-	UNLOCK (&local->fd->lock);
+    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;
 }
 
+void
+afr_transaction_start(afr_local_t *local, xlator_t *this)
+{
+    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);
+}
 
 int
-afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type)
+afr_write_txn_refresh_done(call_frame_t *frame, xlator_t *this, int err)
 {
-        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_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;
+}
 
-	afr_transaction_eager_lock_init (local, this);
+int
+afr_transaction_lockee_init(call_frame_t *frame)
+{
+    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 && local->transaction.eager_lock_on)
-                afr_set_lk_owner (frame, this, local->fd);
-        else
-                afr_set_lk_owner (frame, this, frame->root);
+    switch (local->transaction.type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+            ret = afr_add_inode_lockee(local, priv->child_count);
+            break;
 
-        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);
+        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;
+                }
 
-                if (fd) {
-                        afr_delayed_changelog_wake_up (this, fd);
-                        fd_unref (fd);
+                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;
+}
 
-        if (afr_lock_server_count (priv, local->transaction.type) == 0) {
-                afr_internal_lock_finish (frame, this);
-        } else {
-                afr_lock (frame, this);
-        }
+int
+afr_transaction(call_frame_t *frame, xlator_t *this, afr_transaction_type type)
+{
+    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 55e8bbcca08..beefa26f4a6 100644
--- a/xlators/cluster/afr/src/afr-transaction.h
+++ b/xlators/cluster/afr/src/afr-transaction.h
@@ -11,33 +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);
 
-int32_t
-afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type);
+void
+afr_delayed_changelog_wake_up(xlator_t *this, fd_t *fd);
+
+void
+__mark_all_success(call_frame_t *frame, xlator_t *this);
+
+gf_boolean_t
+afr_txn_nothing_failed(call_frame_t *frame, xlator_t *this);
+
+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_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);
+afr_read_txn_continue(call_frame_t *frame, xlator_t *this, int subvol);
+
 void
-afr_set_delayed_post_op (call_frame_t *frame, xlator_t *this);
+afr_pending_read_increment(afr_private_t *priv, int child_index);
 
 void
-afr_delayed_changelog_wake_up (xlator_t *this, fd_t *fd);
+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_needs_changelog_update(afr_local_t *local);
 void
-__mark_all_success (int32_t *pending[], int child_count,
-                    afr_transaction_type type);
+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 6fabc00171f..df7366f0a65 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -15,757 +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) {
+        return ret;
+    }
 
-        ret = xlator_mem_acct_init (this, gf_afr_mt_end + 1);
+    return ret;
+}
 
-        if (ret != 0) {
-                gf_log(this->name, GF_LOG_ERROR, "Memory accounting init"
-                       "failed");
-                return ret;
+int
+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++;
+    }
 
-        return ret;
+    return index;
 }
 
+static void
+fix_quorum_options(xlator_t *this, afr_private_t *priv, char *qtype,
+                   dict_t *options)
+{
+    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
-xlator_subvolume_index (xlator_t *this, xlator_t *subvol)
+afr_set_favorite_child_policy(afr_private_t *priv, char *policy)
 {
-        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;
+
+    index = gf_get_index_by_elem(afr_favorite_child_policies, policy);
+    if (index < 0 || index >= AFR_FAV_CHILD_POLICY_MAX)
+        return -1;
+
+    priv->fav_child_policy = index;
+
+    return 0;
+}
 
-        return index;
+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;
+    }
 }
 
 void
-fix_quorum_options (xlator_t *this, afr_private_t *priv, char *qtype)
+afr_handle_anon_inode_options(afr_private_t *priv, 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;
-        }
+    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);
+    }
 }
 
 int
-reconfigure (xlator_t *this, dict_t *options)
+reconfigure(xlator_t *this, dict_t *options)
 {
-        afr_private_t *priv        = NULL;
-        xlator_t      *read_subvol = NULL;
-        int            read_subvol_index = -1;
-        int            ret         = -1;
-        int            index       = -1;
-        char          *qtype       = NULL;
+    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;
 
-        priv = this->private;
+    priv = this->private;
 
-        GF_OPTION_RECONF ("background-self-heal-count",
-                          priv->background_self_heal_count, options, uint32,
-                          out);
+    GF_OPTION_RECONF("metadata-splitbrain-forced-heal",
+                     priv->metadata_splitbrain_forced_heal, options, bool, out);
 
-        GF_OPTION_RECONF ("metadata-self-heal",
-                          priv->metadata_self_heal, options, bool, out);
+    GF_OPTION_RECONF("background-self-heal-count",
+                     priv->background_self_heal_count, options, uint32, out);
 
-        GF_OPTION_RECONF ("data-self-heal", priv->data_self_heal, options, str,
-                          out);
+    GF_OPTION_RECONF("heal-wait-queue-length", priv->heal_wait_qlen, options,
+                     uint32, out);
 
-        GF_OPTION_RECONF ("entry-self-heal", priv->entry_self_heal, options,
-                          bool, out);
+    GF_OPTION_RECONF("metadata-self-heal", priv->metadata_self_heal, options,
+                     bool, out);
 
-        GF_OPTION_RECONF ("strict-readdir", priv->strict_readdir, 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 ("data-self-heal-window-size",
-                          priv->data_self_heal_window_size, options,
-                          uint32, out);
+    GF_OPTION_RECONF("entry-self-heal", priv->entry_self_heal, options, bool,
+                     out);
 
-        GF_OPTION_RECONF ("data-change-log", priv->data_change_log, options,
-                          bool, out);
+    GF_OPTION_RECONF("data-self-heal-window-size",
+                     priv->data_self_heal_window_size, options, uint32, out);
 
-        GF_OPTION_RECONF ("metadata-change-log",
-                          priv->metadata_change_log, options, bool, out);
+    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 ("entry-change-log", priv->entry_change_log, options,
-                          bool, out);
+    GF_OPTION_RECONF("halo-enabled", priv->halo_enabled, options, bool, out);
 
-        GF_OPTION_RECONF ("data-self-heal-algorithm",
-                          priv->data_self_heal_algorithm, options, str, out);
+    GF_OPTION_RECONF("halo-shd-max-latency", priv->shd.halo_max_latency_msec,
+                     options, uint32, out);
 
-        GF_OPTION_RECONF ("self-heal-daemon", priv->shd.enabled, options, bool, out);
+    GF_OPTION_RECONF("halo-nfsd-max-latency", priv->nfsd.halo_max_latency_msec,
+                     options, uint32, out);
 
-        GF_OPTION_RECONF ("read-subvolume", read_subvol, options, xlator, out);
+    GF_OPTION_RECONF("halo-max-latency", priv->halo_max_latency_msec, options,
+                     uint32, out);
 
-        GF_OPTION_RECONF ("read-hash-mode", priv->hash_mode,
-                          options, uint32, out);
+    GF_OPTION_RECONF("halo-max-replicas", priv->halo_max_replicas, 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("halo-min-replicas", priv->halo_min_replicas, options,
+                     uint32, out);
+
+    GF_OPTION_RECONF("read-subvolume", read_subvol, options, xlator, out);
 
-        GF_OPTION_RECONF ("read-subvolume-index",read_subvol_index, options,int32,out);
+    choose_local_old = priv->choose_local;
+    GF_OPTION_RECONF("choose-local", priv->choose_local, 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;
+    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;
         }
+        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;
+}
 
-        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);
+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. ");
 
-	GF_OPTION_RECONF ("post-op-delay-secs", priv->post_op_delay_secs, options,
-			  uint32, out);
+        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;
+    }
 
-        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);
-        priv->did_discovery = _gf_false;
+    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;
 
-        ret = 0;
 out:
-        return ret;
-
+    GF_FREE(ptr1);
+    return ret;
 }
 
-
-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.";
-
+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);
+}
 
 int32_t
-init (xlator_t *this)
+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;
+    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;
         }
-
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Volume is dangling.");
+    }
+    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);
 
-	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);
+    priv->pending_reads = GF_CALLOC(sizeof(*priv->pending_reads),
+                                    priv->child_count, gf_afr_mt_atomic_t);
 
-        child_count = xlator_subvolume_count (this);
+    GF_OPTION_INIT("read-hash-mode", priv->hash_mode, uint32, out);
 
-        priv->child_count = child_count;
-
-        priv->read_child = -1;
+    priv->favorite_child = -1;
 
-        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;
-        }
-        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("favorite-child-policy", fav_child_policy, str, out);
+    if (afr_set_favorite_child_policy(priv, fav_child_policy) == -1)
+        goto out;
 
+    GF_OPTION_INIT("shd-max-threads", priv->shd.max_threads, uint32, out);
 
-        GF_OPTION_INIT ("background-self-heal-count",
-                        priv->background_self_heal_count, uint32, out);
+    GF_OPTION_INIT("shd-wait-qlength", priv->shd.wait_qlength, uint32, out);
 
-        GF_OPTION_INIT ("data-self-heal", priv->data_self_heal, str, out);
+    GF_OPTION_INIT("background-self-heal-count",
+                   priv->background_self_heal_count, uint32, out);
 
-        GF_OPTION_INIT ("data-self-heal-algorithm",
-                        priv->data_self_heal_algorithm, str, out);
+    GF_OPTION_INIT("heal-wait-queue-length", priv->heal_wait_qlen, uint32, out);
 
-        GF_OPTION_INIT ("data-self-heal-window-size",
-                        priv->data_self_heal_window_size, uint32, 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 ("metadata-self-heal", priv->metadata_self_heal, 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 ("entry-self-heal", priv->entry_self_heal, bool, out);
+    GF_OPTION_INIT("data-self-heal-window-size",
+                   priv->data_self_heal_window_size, uint32, out);
 
-        GF_OPTION_INIT ("self-heal-daemon", priv->shd.enabled, bool, out);
+    GF_OPTION_INIT("metadata-self-heal", priv->metadata_self_heal, bool, out);
 
-        GF_OPTION_INIT ("iam-self-heal-daemon", priv->shd.iamshd, bool, out);
+    GF_OPTION_INIT("entry-self-heal", priv->entry_self_heal, bool, out);
 
-        GF_OPTION_INIT ("data-change-log", priv->data_change_log, bool, out);
+    GF_OPTION_INIT("halo-shd-max-latency", priv->shd.halo_max_latency_msec,
+                   uint32, out);
 
-        GF_OPTION_INIT ("metadata-change-log", priv->metadata_change_log, 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 ("entry-change-log", priv->entry_change_log, bool, out);
+    GF_OPTION_INIT("halo-enabled", priv->halo_enabled, bool, out);
 
-        GF_OPTION_INIT ("optimistic-change-log", priv->optimistic_change_log,
-                        bool, out);
+    GF_OPTION_INIT("halo-nfsd-max-latency", priv->nfsd.halo_max_latency_msec,
+                   uint32, out);
 
-        GF_OPTION_INIT ("inodelk-trace", priv->inodelk_trace, bool, out);
+    GF_OPTION_INIT("iam-nfs-daemon", priv->nfsd.iamnfsd, bool, out);
 
-        GF_OPTION_INIT ("entrylk-trace", priv->entrylk_trace, bool, out);
+    GF_OPTION_INIT("optimistic-change-log", priv->optimistic_change_log, bool,
+                   out);
 
-        GF_OPTION_INIT ("strict-readdir", priv->strict_readdir, bool, 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);
 
-        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("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);
 
-	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("post-op-delay-secs", priv->post_op_delay_secs, uint32, out);
+    GF_OPTION_INIT("ensure-durability", priv->ensure_durability, bool, out);
 
-        priv->wait_count = 1;
+    GF_OPTION_INIT("self-heal-daemon", priv->shd.enabled, 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("iam-self-heal-daemon", priv->shd.iamshd, bool, out);
+    GF_OPTION_INIT("heal-timeout", priv->shd.timeout, int32, 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("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->children = GF_CALLOC (sizeof (xlator_t *), child_count,
-                                    gf_afr_mt_xlator_t);
-        if (!priv->children) {
-                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;
 
-        priv->pending_key = GF_CALLOC (sizeof (*priv->pending_key),
-                                       child_count,
-                                       gf_afr_mt_char);
-        if (!priv->pending_key) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    priv->wait_count = 1;
 
-        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");
-                        ret = -ENOMEM;
-                        goto out;
-                }
-
-                trav = trav->next;
-                i++;
-        }
+    priv->local = GF_CALLOC(sizeof(unsigned char), child_count, gf_afr_mt_char);
+    if (!priv->local) {
+        ret = -ENOMEM;
+        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->anon_inode = GF_CALLOC(sizeof(unsigned char), child_count,
+                                 gf_afr_mt_char);
 
-        /* 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->child_up = GF_CALLOC(sizeof(unsigned char), child_count,
+                               gf_afr_mt_char);
 
-        priv->first_lookup = 1;
-        priv->root_inode = NULL;
+    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);
 
-        if (!priv->shd.iamshd) {
-                ret = 0;
-                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->last_event = GF_CALLOC(child_count, sizeof(*priv->last_event),
+                                 gf_afr_mt_int32_t);
+    if (!priv->last_event) {
         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;
+        goto out;
+    }
 
-        priv->shd.pending = GF_CALLOC (sizeof (*priv->shd.pending), child_count,
-                                       gf_afr_mt_int32_t);
-        if (!priv->shd.pending)
-                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.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;
+    /* 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;
+    }
 
-        priv->shd.healed = eh_new (AFR_EH_HEALED_LIMIT, _gf_false);
-        if (!priv->shd.healed)
-                goto out;
+    priv->root_inode = NULL;
 
-        priv->shd.heal_failed = eh_new (AFR_EH_HEAL_FAIL_LIMIT, _gf_false);
-        if (!priv->shd.heal_failed)
-                goto out;
+    ret = 0;
+out:
+    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);
+}
 
-        priv->shd.split_brain = eh_new (AFR_EH_SPLIT_BRAIN_LIMIT, _gf_false);
-        if (!priv->shd.split_brain)
-                goto out;
+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;
 
-        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 = this->private;
 
-        ret = 0;
-out:
-        return ret;
-}
+    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);
 
-int
-fini (xlator_t *this)
-{
-        afr_private_t *priv = NULL;
+    if (this->local_pool != NULL) {
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
 
-        priv = this->private;
-        this->private = NULL;
-        afr_priv_destroy (priv);
-        if (this->itable);//I dont see any destroy func
+    this->private = NULL;
+    afr_priv_destroy(priv);
+    if (this->itable) {
+        inode_table_destroy(this->itable);
+        this->itable = NULL;
+    }
 
-        return 0;
+    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  = {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 cbe6b339d08..d62f9a9caf2 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -8,1158 +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
-
-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;
-} 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;
-} afr_private_t;
+    int lockee_count;
 
-typedef enum {
-        AFR_SELF_HEAL_NOT_ATTEMPTED,
-        AFR_SELF_HEAL_STARTED,
-        AFR_SELF_HEAL_FAILED,
-} afr_self_heal_status;
+    int32_t lk_call_count;
+    int32_t lk_expected_count;
+    int32_t lk_attempted_count;
 
-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;
-
-
-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 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;
+    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;
 
-        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;
-
-	void (*afr_set_self_heal_status) (struct afr_self_heal_ *sh,
-                                          afr_self_heal_status status);
-        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;
+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 struct afr_self_heal_ afr_self_heal_t;
-
 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;
+    AFR_FD_NOT_OPENED,
+    AFR_FD_OPENED,
+    AFR_FD_OPENING
+} afr_fd_open_status_t;
 
-typedef enum {
-        AFR_TRANSACTION_LK,
-        AFR_SELFHEAL_LK,
-} transaction_lk_type_t;
+typedef struct {
+    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_LOCK_OP,
-        AFR_UNLOCK_OP,
-} afr_lock_op_type_t;
+} afr_fd_ctx_t;
 
 typedef enum {
-        AFR_DATA_SELF_HEAL_LK,
-        AFR_METADATA_SELF_HEAL_LK,
-        AFR_ENTRY_SELF_HEAL_LK,
-}selfheal_lk_type_t;
+    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;
 
-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;
+typedef struct _afr_local {
+    glusterfs_fop_t op;
+    unsigned int call_count;
 
-/*
-  xattr format: trusted.afr.volume = [x y z]
-  x - data pending
-  y - metadata pending
-  z - entry pending
-*/
+    /* @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;
 
-static inline int
-afr_index_for_transaction_type (afr_transaction_type type)
-{
-        switch (type) {
+    uint32_t open_fd_count;
+    int32_t num_inodelks;
 
-        case AFR_DATA_TRANSACTION:
-                return 0;
+    int32_t op_ret;
+    int32_t op_errno;
 
-        case AFR_METADATA_TRANSACTION:
-                return 1;
+    int dirty[AFR_NUM_CHANGE_LOGS];
 
-        case AFR_ENTRY_TRANSACTION:
-        case AFR_ENTRY_RENAME_TRANSACTION:
-                return 2;
-        }
+    int32_t **pending;
 
-        return -1;  /* make gcc happy */
-}
+    loc_t loc;
+    loc_t newloc;
 
-typedef struct {
-        loc_t                   loc;
-        char                    *basename;
-        unsigned char           *locked_nodes;
-        int                     locked_count;
+    fd_t *fd;
+    afr_fd_ctx_t *fd_ctx;
 
-} afr_entry_lockee_t;
+    /* @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;
 
-int
-afr_entry_lockee_cmp (const void *l1, const void *l2);
+    /* @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;
 
-typedef struct {
-        loc_t *lk_loc;
-        struct gf_flock lk_flock;
+    /* @readfn:
 
-        int                     lockee_count;
-        afr_entry_lockee_t      lockee[AFR_LOCKEE_COUNT_MAX];
+       pointer to function which will perform the read operation on a given
+       subvolume. Used in read transactions.
+    */
 
-        const char *lk_basename;
-        const char *lower_basename;
-        const char *higher_basename;
-        char lower_locked;
-        char higher_locked;
+    afr_read_txn_wind_t readfn;
 
-        unsigned char *locked_nodes;
-        unsigned char *lower_locked_nodes;
-        unsigned char *inode_locked_nodes;
+    /* @inode:
 
-        selfheal_lk_type_t selfheal_lk_type;
-        transaction_lk_type_t transaction_lk_type;
+       the inode on which the read txn is performed on. ref'ed and copied
+       from either fd->inode or loc.inode
+    */
 
-        int32_t lock_count;
-        int32_t inodelk_lock_count;
-        int32_t entrylk_lock_count;
+    inode_t *inode;
 
-        uint64_t lock_number;
-        int32_t lk_call_count;
-        int32_t lk_expected_count;
-        int32_t lk_attempted_count;
+    /* @parent[2]:
 
-        int32_t lock_op_ret;
-        int32_t lock_op_errno;
-        afr_lock_cbk_t lock_cbk;
-} afr_internal_lock_t;
+       parent inode[s] on which directory transactions are performed.
+    */
 
-typedef struct _afr_locked_fd {
-        fd_t  *fd;
-        struct list_head list;
-} afr_locked_fd_t;
+    inode_t *parent;
+    inode_t *parent2;
 
-struct afr_reply {
-	int	valid;
-	int32_t	op_ret;
-	int32_t	op_errno;
-};
+    /* @readable:
 
-typedef struct _afr_local {
-        int     uid;
-        int     gid;
-        unsigned int call_count;
-        unsigned int success_count;
-        unsigned int enoent_count;
+       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*/
 
+    afr_inode_refresh_cbk_t refreshfn;
 
-        unsigned int unhealable;
+    /* @refreshinode:
 
-        unsigned int read_child_index;
-        unsigned char read_child_returned;
-        unsigned int first_up_child;
+       Inode currently getting refreshed.
+    */
+    inode_t *refreshinode;
 
-	gf_lkowner_t  saved_lk_owner;
+    dict_t *xattr_req;
 
-        int32_t op_ret;
-        int32_t op_errno;
+    dict_t *dict;
 
-        int32_t **pending;
+    int read_subvol; /* Current read subvolume */
 
-        loc_t loc;
-        loc_t newloc;
+    int optimistic_change_log;
 
-        fd_t *fd;
+    afr_internal_lock_t internal_lock;
 
-        glusterfs_fop_t fop;
+    /*To handle setattr/setxattr on yet to be linked inode from dht*/
+    uuid_t refreshgfid;
 
-        unsigned char *child_up;
-        int32_t       *fresh_children; //in the order of response
+    /* @refreshed:
 
-        int32_t *child_errno;
+       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;
 
-        dict_t  *xattr_req;
+    gf_boolean_t update_num_inodelks;
+    gf_boolean_t update_open_fd_count;
 
-        int32_t  inodelk_count;
-        int32_t  entrylk_count;
+    /*
+      @pre_op_compat:
 
-        afr_internal_lock_t internal_lock;
+      compatibility mode of pre-op. send a separate pre-op and
+      op operations as part of transaction, rather than combining
+    */
 
-        afr_locked_fd_t *locked_fd;
-        int32_t          source_child;
-        int32_t          lock_recovery_child;
+    gf_boolean_t pre_op_compat;
 
-        dict_t  *dict;
-        int      optimistic_change_log;
-	gf_boolean_t      delayed_post_op;
+    /* 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;
 
-	/* 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;
+    /* 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;
 
-        /*
-          This struct contains the arguments for the "continuation"
-          (scheme-like) of fops
-        */
+    /*
+      This struct contains the arguments for the "continuation"
+      (scheme-like) of fops
+    */
 
-        int   op;
+    struct {
         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;
-
-                        int32_t op_ret;
-
-                        struct iovec *vector;
-                        struct iobref *iobref;
-                        int32_t count;
-                        off_t offset;
-                        uint32_t flags;
-                } writev;
-
-                struct {
-                        struct iatt prebuf;
-                        struct iatt postbuf;
-                } fsync;
-
-                struct {
-                        off_t offset;
-                        struct iatt prebuf;
-                        struct iatt postbuf;
-                } truncate;
-
-                struct {
-                        off_t offset;
-                        struct iatt prebuf;
-                        struct iatt postbuf;
-                } ftruncate;
-
-                struct {
-                        struct iatt in_buf;
-                        int32_t valid;
-                        struct iatt preop_buf;
-                        struct iatt postop_buf;
-                } setattr;
-
-                struct {
-                        struct iatt in_buf;
-                        int32_t valid;
-                        struct iatt preop_buf;
-                        struct iatt postop_buf;
-                } 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;
-			struct iatt prebuf;
-			struct iatt postbuf;
-		} fallocate;
-
-		struct {
-			off_t offset;
-			size_t len;
-			struct iatt prebuf;
-			struct iatt postbuf;
-		} discard;
-
-        } cont;
+            struct statvfs buf;
+            unsigned char buf_set;
+        } statfs;
 
         struct {
-                off_t start, len;
+            fd_t *fd;
+            int32_t flags;
+        } open;
 
-                gf_boolean_t    eager_lock_on;
-                int *eager_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 */
 
-                char *basename;
-                char *new_basename;
+        struct {
+            int32_t mask;
+            int last_index; /* index of the child we tried previously */
+        } access;
 
-                loc_t parent_loc;
-                loc_t new_parent_loc;
+        struct {
+            int last_index;
+        } stat;
 
-                afr_transaction_type type;
+        struct {
+            int last_index;
+        } fstat;
 
-		/* pre-compute the post piggyback status before
-		   entering POST-OP phase
-		*/
-		int              *postop_piggybacked;
+        struct {
+            size_t size;
+            int last_index;
+        } readlink;
 
-		/* stub to resume on destruction
-		   of the transaction frame */
-		call_stub_t      *resume_stub;
+        struct {
+            char *name;
+            long xattr_len;
+            int last_index;
+        } getxattr;
 
-		struct list_head  eager_locked;
+        struct {
+            size_t size;
+            off_t offset;
+            int last_index;
+            uint32_t flags;
+        } readv;
 
-                int32_t         **txn_changelog;//changelog after pre+post ops
-                unsigned char   *pre_op;
+        /* dir read */
 
-                call_frame_t *main_frame;
+        struct {
+            uint32_t *checksum;
+            int success_count;
+            int32_t op_ret;
+            int32_t op_errno;
+        } opendir;
 
-                int (*fop) (call_frame_t *frame, xlator_t *this);
+        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 */
 
-                int (*done) (call_frame_t *frame, xlator_t *this);
+        struct {
+            struct iatt prebuf;
+            struct iatt postbuf;
+        } inode_wfop;  // common structure for all inode-write-fops
 
-                int (*resume) (call_frame_t *frame, xlator_t *this);
+        struct {
+            struct iovec *vector;
+            struct iobref *iobref;
+            off_t offset;
+            int32_t op_ret;
+            int32_t count;
+            uint32_t flags;
+        } writev;
 
-                int (*unwind) (call_frame_t *frame, xlator_t *this);
+        struct {
+            off_t offset;
+        } truncate;
 
-                /* post-op hook */
-        } transaction;
+        struct {
+            off_t offset;
+        } ftruncate;
 
-        afr_self_heal_t self_heal;
+        struct {
+            struct iatt in_buf;
+            int32_t valid;
+        } setattr;
 
-        struct marker_str     marker;
+        struct {
+            struct iatt in_buf;
+            int32_t valid;
+        } fsetattr;
 
-        /* extra data for fops */
-        dict_t         *xdata_req;
-        dict_t         *xdata_rsp;
+        struct {
+            dict_t *dict;
+            int32_t flags;
+        } setxattr;
 
-        mode_t          umask;
-        int             xflag;
-        gf_boolean_t    do_discovery;
-	struct afr_reply *replies;
-} afr_local_t;
+        struct {
+            dict_t *dict;
+            int32_t flags;
+        } fsetxattr;
 
-typedef enum {
-        AFR_FD_NOT_OPENED,
-        AFR_FD_OPENED,
-        AFR_FD_OPENING
-} afr_fd_open_status_t;
+        struct {
+            char *name;
+        } removexattr;
 
-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;
+        struct {
+            dict_t *xattr;
+            gf_xattrop_flags_t optype;
+        } xattrop;
 
-        unsigned int *lock_piggyback;
-        unsigned int *lock_acquired;
+        /* dir write */
 
-        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 */
+        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
 
-        int32_t last_tried;
+        struct {
+            fd_t *fd;
+            dict_t *params;
+            int32_t flags;
+            mode_t mode;
+        } create;
 
-        int  hit, miss;
-        gf_boolean_t failed_over;
-        struct list_head entries; /* needed for readdir failover */
+        struct {
+            dict_t *params;
+            dev_t dev;
+            mode_t mode;
+        } mknod;
 
-        unsigned char *locked_on; /* which subvolumes locks have been successful */
+        struct {
+            dict_t *params;
+            int32_t mode;
+        } mkdir;
 
-	/* used for delayed-post-op optimization */
-	pthread_mutex_t    delay_lock;
-	gf_timer_t        *delay_timer;
-	call_frame_t      *delay_frame;
-        int               call_child;
+        struct {
+            dict_t *params;
+            char *linkpath;
+        } symlink;
 
-	/* 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;
+        struct {
+            off_t offset;
+            size_t len;
+            int32_t mode;
+        } fallocate;
 
-	/* list of frames currently in progress */
-	struct list_head  eager_locked;
-} afr_fd_ctx_t;
+        struct {
+            off_t offset;
+            size_t len;
+        } discard;
 
+        struct {
+            off_t offset;
+            off_t len;
+            struct iatt prebuf;
+            struct iatt postbuf;
+        } zerofill;
 
-/* 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);
+        struct {
+            char *volume;
+            int32_t cmd;
+            int32_t in_cmd;
+            struct gf_flock in_flock;
+            struct gf_flock flock;
+            void *xdata;
+        } inodelk;
 
+        struct {
+            char *volume;
+            char *basename;
+            void *xdata;
+            entrylk_cmd in_cmd;
+            entrylk_cmd cmd;
+            entrylk_type type;
+        } entrylk;
 
-/* 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)))
+        struct {
+            off_t offset;
+            gf_seek_what_t what;
+        } seek;
 
-#define afr_fop_failed(op_ret, op_errno) ((op_ret) == -1)
+        struct {
+            struct gf_lease user_lease;
+            struct gf_lease ret_lease;
+            unsigned char *locked_nodes;
+        } lease;
 
-/* have we tried all children? */
-#define all_tried(i, count)  ((i) == (count) - 1)
+        struct {
+            int flags;
+        } rmdir;
 
-int32_t
-afr_set_dict_gfid (dict_t *dict, uuid_t gfid);
+        struct {
+            int32_t datasync;
+        } fsync;
 
-int
-pump_command_reply (call_frame_t *frame, xlator_t *this);
+        struct {
+            uuid_t gfid_req;
+            gf_boolean_t needs_fresh_lookup;
+        } lookup;
 
-int32_t
-afr_notify (xlator_t *this, int32_t event, void *data, void *data2);
+    } cont;
 
-int
-afr_init_entry_lockee (afr_entry_lockee_t *lockee, afr_local_t *local,
-                       loc_t *loc, char *basename, int child_count);
+    struct {
+        char *basename;
+        char *new_basename;
 
-void
-afr_entry_lockee_cleanup (afr_internal_lock_t *int_lock);
+        loc_t parent_loc;
+        loc_t new_parent_loc;
+
+        /* stub to resume on destruction
+           of the transaction frame */
+        call_stub_t *resume_stub;
+
+        struct list_head owner_list;
+        struct list_head wait_list;
+
+        unsigned char *pre_op;
+
+        /* Changelog xattr dict for [f]xattrop*/
+        dict_t **changelog_xdata;
+        unsigned char *pre_op_sources;
+
+        /* @failed_subvols: subvolumes on which a pre-op or a
+            FOP failed. */
+        unsigned char *failed_subvols;
+
+        call_frame_t *main_frame; /*Fop frame*/
+        call_frame_t *frame;      /*Transaction frame*/
+
+        int (*wind)(call_frame_t *frame, xlator_t *this, int subvol);
+
+        int (*unwind)(call_frame_t *frame, xlator_t *this);
+
+        off_t start, len;
+
+        afr_transaction_type type;
+
+        int32_t in_flight_sb_errno; /* This is where the cause of the
+                                       failure on the last good copy of
+                                       the file is stored.
+                                       */
+
+        /* @changelog_resume: function to be called after changlogging
+           (either pre-op or post-op) is done
+        */
+        afr_changelog_resume_t changelog_resume;
+
+        gf_boolean_t eager_lock_on;
+        gf_boolean_t do_eager_unlock;
+
+        /* @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
+
+           A 'true' value set in @dirtied flag means an 'undirtying'
+           has to be done in POST-OP phase.
+        */
+        gf_boolean_t dirtied;
+
+        /* @inherited: flag which indicates that the dirty flags
+           of the previous transaction were inherited
+        */
+        gf_boolean_t inherited;
+
+        /*
+          @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);
+
+int32_t
+afr_notify(xlator_t *this, int32_t event, void *data, void *data2);
 
 int
-afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this);
+xattr_is_equal(dict_t *this, char *key1, data_t *value1, void *data);
 
 int
-afr_blocking_lock (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_internal_lock_finish (call_frame_t *frame, xlator_t *this);
+afr_add_inode_lockee(afr_local_t *local, int child_count);
 
 void
-afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src,
-                          unsigned int child_count);
-
-int pump_start (call_frame_t *frame, xlator_t *this);
+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_CALL_RESUME(stub)					\
-        do {                                                    \
-                afr_local_t *__local = NULL;                    \
-                xlator_t    *__this = NULL;                     \
-								\
-		__local = stub->frame->local;			\
-		__this = stub->frame->this;			\
-		stub->frame->local = NULL;			\
-								\
-		call_resume (stub);				\
-                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_inode_rm_stale_children (xlator_t *this, inode_t *inode,
-                             int32_t *stale_children);
+afr_handle_inconsistent_fop(call_frame_t *frame, int32_t *op_ret,
+                            int32_t *op_errno);
+
 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_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_fix_open (xlator_t *this, fd_t *fd, size_t need_open_count, int *need_open);
+afr_process_post_writev(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_unwind(call_frame_t *frame, xlator_t *this);
+
+void
+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_ta_dict_contains_pending_xattr(dict_t *dict, afr_private_t *priv,
+                                   int child);
+
+void
+afr_selfheal_childup(xlator_t *this, afr_private_t *priv);
 
 gf_boolean_t
-afr_fd_has_witnessed_unstable_write (xlator_t *this, fd_t *fd);
+afr_is_lock_mode_mandatory(dict_t *xdata);
+
+void
+afr_dom_lock_release(call_frame_t *frame);
 
 void
-afr_delayed_changelog_wake_resume (xlator_t *this, fd_t *fd, call_stub_t *stub);
+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 db83410c74d..00000000000
--- a/xlators/cluster/afr/src/pump.c
+++ /dev/null
@@ -1,2638 +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++;
-	}
-
-        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 = syncenv_new (0);
-        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;
-
-        if (pump_priv->env)
-                syncenv_destroy (pump_priv->env);
-
-        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 ec54e438e35..8ba0cc4c732 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -8,90 +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
author	Venky Shankar <vshankar@redhat.com>	2012-07-17 22:54:21 +0530
committer	Vijay Bellur <vbellur@redhat.com>	2012-07-19 02:23:57 -0700
commit	84a735558d2682446b2e45d97ff97e698ab2d256 (patch)
tree	90733b6d292a81b3340e758187f1055d67472a4d /libglusterfs
parent	20fedc3232e876917f963b28a4f4b3bf469b2e6f (diff)