525 files changed, 182246 insertions, 112539 deletions

diff --git a/xlators/Makefile.am b/xlators/Makefile.am
index b1643d26c..f60fa85ce 100644
--- a/xlators/Makefile.am
+++ b/xlators/Makefile.am
@@ -1,3 +1,4 @@
-SUBDIRS = cluster storage protocol performance debug features encryption mount nfs mgmt system
+SUBDIRS = cluster storage protocol performance debug features encryption mount nfs mgmt system \
+          playground
 
 CLEANFILES =
diff --git a/xlators/bindings/Makefile.am b/xlators/bindings/Makefile.am
deleted file mode 100644
index f77665802..000000000
--- 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 c0b9141c6..000000000
--- 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 = $(xlatordir)/python
-
-python_so_SOURCES = python.c
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall \
-	-I$(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 ee0eb1310..000000000
--- a/xlators/bindings/python/src/gluster.py
+++ /dev/null
@@ -1,47 +0,0 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
-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 ba24c8165..000000000
--- a/xlators/bindings/python/src/glusterstack.py
+++ /dev/null
@@ -1,55 +0,0 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
-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 e9069d07c..000000000
--- a/xlators/bindings/python/src/glustertypes.py
+++ /dev/null
@@ -1,167 +0,0 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
-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 3310a2115..000000000
--- a/xlators/bindings/python/src/python.c
+++ /dev/null
@@ -1,232 +0,0 @@
-/*
-   Copyright (c) 2007-2010 Chris AtLee <chris@atlee.ca>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#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 507455c85..000000000
--- a/xlators/bindings/python/src/testxlator.py
+++ /dev/null
@@ -1,56 +0,0 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
-
-"""
-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 0990822a7..6e883e565 100644
--- a/xlators/cluster/Makefile.am
+++ b/xlators/cluster/Makefile.am
@@ -1,3 +1,3 @@
-SUBDIRS = stripe afr dht
+SUBDIRS = stripe afr dht nsr-server nsr-recon nsr-client
 
 CLEANFILES = 
diff --git a/xlators/cluster/afr/src/Makefile.am b/xlators/cluster/afr/src/Makefile.am
index e192b599b..ea5a90abb 100644
--- a/xlators/cluster/afr/src/Makefile.am
+++ b/xlators/cluster/afr/src/Makefile.am
@@ -1,27 +1,38 @@
 xlator_LTLIBRARIES = afr.la pump.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 $(top_builddir)/xlators/lib/src/libxlator.c
+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-lk-common.c \
+	afr-read-txn.c \
+	$(top_builddir)/xlators/lib/src/libxlator.c
 
-afr_la_LDFLAGS = -module -avoidversion
-afr_la_SOURCES = $(afr_common_source) afr.c
+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
+
+afr_la_LDFLAGS = -module -avoid-version
+afr_la_SOURCES = $(afr_common_source) $(AFR_SELFHEAL_SOURCES) afr.c
 afr_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-pump_la_LDFLAGS = -module -avoidversion
-pump_la_SOURCES =  $(afr_common_source) pump.c
+pump_la_LDFLAGS = -module -avoid-version
+pump_la_SOURCES =  $(afr_common_source) $(AFR_SELFHEAL_SOURCES) pump.c
 pump_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 $(top_builddir)/xlators/lib/src/libxlator.h
+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-mem-types.h \
+	afr-common.c afr-self-heald.h pump.h \
+	$(top_builddir)/xlators/lib/src/libxlator.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) \
+	-I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/xlators/lib/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	    -I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/contrib/md5 -shared -nostartfiles $(GF_CFLAGS) \
-	    -I$(top_srcdir)/xlators/lib/src
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 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 d6b358e2b..164a651ba 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include <libgen.h>
@@ -44,6 +35,7 @@
 #include "compat.h"
 #include "byte-order.h"
 #include "statedump.h"
+#include "inode.h"
 
 #include "fd.h"
 
@@ -53,662 +45,863 @@
 #include "afr-dir-write.h"
 #include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-#include "pump.h"
+#include "afr-self-heald.h"
 
-#define AFR_ICTX_OPENDIR_DONE_MASK     0x0000000200000000ULL
-#define AFR_ICTX_SPLIT_BRAIN_MASK      0x0000000100000000ULL
-#define AFR_ICTX_READ_CHILD_MASK       0x00000000FFFFFFFFULL
 
-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)
+call_frame_t *
+afr_copy_frame (call_frame_t *base)
 {
-        int     i = 0;
+	afr_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	int op_errno = 0;
+
+	frame = copy_frame (base);
+	if (!frame)
+		return NULL;
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local) {
+		AFR_STACK_DESTROY (frame);
+		return NULL;
+	}
 
-        for (i = 0; i < child_count; i++)
-                dst[i] = src[i];
+	return frame;
 }
 
-void
-afr_xattr_req_prepare (xlator_t *this, dict_t *xattr_req, const char *path)
+/*
+ * 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.
+ */
+
+int
+__afr_inode_read_subvol_get_small (inode_t *inode, xlator_t *this,
+				   unsigned char *data, unsigned char *metadata,
+				   int *event_p)
 {
-        int             i           = 0;
-        afr_private_t   *priv       = NULL;
-        int             ret         = 0;
+	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;
 
-        priv   = this->private;
+	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 = __inode_ctx_get (inode, this, &val);
+	if (ret < 0)
+		return ret;
+
+	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_errno_count (int32_t *children, int *child_errno,
-                 unsigned int child_count, int32_t op_errno)
+__afr_inode_read_subvol_set_small (inode_t *inode, xlator_t *this,
+				   unsigned char *data, unsigned char *metadata,
+				   int event)
 {
-        int i = 0;
-        int errno_count = 0;
-        int child = 0;
+	afr_private_t *priv = NULL;
+	uint16_t datamap = 0;
+	uint16_t metadatamap = 0;
+	uint64_t val = 0;
+	int i = 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++;
-        }
-        return errno_count;
+	priv = this->private;
+
+	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);
+
+	return __inode_ctx_set (inode, this, &val);
 }
 
-int32_t
-afr_set_dict_gfid (dict_t *dict, uuid_t gfid)
+
+int
+__afr_inode_read_subvol_reset_small (inode_t *inode, xlator_t *this)
 {
-        int ret       = 0;
-        uuid_t *pgfid = NULL;
+	int ret = -1;
+	uint16_t datamap = 0;
+	uint16_t metadatamap = 0;
+	uint32_t event = 0;
+	uint64_t val = 0;
 
-        GF_ASSERT (gfid);
+	ret = __inode_ctx_get (inode, this, &val);
+	(void) ret;
 
-        pgfid = GF_CALLOC (1, sizeof (uuid_t), gf_common_mt_char);
-        if (!pgfid) {
-                ret = -1;
-                goto out;
-        }
+	metadatamap = (val & 0x000000000000ffff) >> 0;
+	datamap =     (val & 0x00000000ffff0000) >> 16;
+	event = 0;
 
-        uuid_copy (*pgfid, gfid);
+	val = ((uint64_t) metadatamap) |
+		(((uint64_t) datamap) << 16) |
+		(((uint64_t) event) << 32);
 
-        ret = dict_set_dynptr (dict, "gfid-req", pgfid, sizeof (uuid_t));
-        if (ret)
-                gf_log (THIS->name, GF_LOG_DEBUG, "gfid set failed");
+	return __inode_ctx_set (inode, this, &val);
+}
 
-out:
-        if (ret && pgfid)
-                GF_FREE (pgfid);
 
-        return ret;
+int
+__afr_inode_read_subvol_get (inode_t *inode, xlator_t *this,
+			     unsigned char *data, unsigned char *metadata,
+			     int *event_p)
+{
+	afr_private_t *priv = NULL;
+	int ret = -1;
+
+	priv = this->private;
+
+	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 ret;
 }
 
-afr_inode_ctx_t*
-afr_inode_ctx_get_from_addr (uint64_t addr, int32_t child_count)
+
+int
+__afr_inode_read_subvol_set (inode_t *inode, xlator_t *this, unsigned char *data,
+			     unsigned char *metadata, int event)
 {
-        int             ret  = -1;
-        afr_inode_ctx_t *ctx = NULL;
-        size_t          size = 0;
+	afr_private_t *priv = NULL;
+	int ret = -1;
 
-        GF_ASSERT (child_count > 0);
+	priv = this->private;
 
-        if (!addr) {
-                ctx = GF_CALLOC (1, sizeof (*ctx),
-                                 gf_afr_mt_inode_ctx_t);
-                if (!ctx)
-                        goto out;
-                size = sizeof (*ctx->fresh_children);
-                ctx->fresh_children = GF_CALLOC (child_count, size,
-                                                 gf_afr_mt_int32_t);
-                if (!ctx->fresh_children)
-                        goto out;
-        } else {
-                ctx = (afr_inode_ctx_t*) (long) addr;
-        }
-        ret = 0;
-out:
-        if (ret && ctx) {
-                if (ctx->fresh_children)
-                        GF_FREE (ctx->fresh_children);
-                GF_FREE (ctx);
-                ctx = NULL;
-        }
-        return ctx;
+	if (priv->child_count <= 16)
+		ret = __afr_inode_read_subvol_set_small (inode, this, data,
+							 metadata, event);
+	else
+		ret = -1;
+
+	return ret;
 }
 
-void
-afr_inode_get_ctx (xlator_t *this, inode_t *inode, afr_inode_params_t *params)
+
+int
+__afr_inode_read_subvol_reset (inode_t *inode, xlator_t *this)
 {
-        GF_ASSERT (inode);
-        GF_ASSERT (params);
+	afr_private_t *priv = NULL;
+	int ret = -1;
 
-        int             ret = 0;
-        afr_inode_ctx_t *ctx = NULL;
-        afr_private_t   *priv = NULL;
-        int             i = 0;
-        uint64_t        ctx_addr   = 0;
-        int32_t         read_child = -1;
-        int32_t         *fresh_children = NULL;
+	priv = this->private;
 
-        priv = this->private;
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx_addr);
-                if (ret < 0)
-                        goto unlock;
-                ctx = afr_inode_ctx_get_from_addr (ctx_addr, priv->child_count);
-                if (!ctx)
-                        goto unlock;
-                switch (params->mask_type) {
-                case AFR_ICTX_READ_CHILD_MASK:
-                        fresh_children = params->u.read_ctx.fresh_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_ICTX_OPENDIR_DONE_MASK:
-                        params->u.value = ctx->masks &
-                                          AFR_ICTX_OPENDIR_DONE_MASK;
-                        break;
-                case AFR_ICTX_SPLIT_BRAIN_MASK:
-                        params->u.value = ctx->masks & AFR_ICTX_SPLIT_BRAIN_MASK;
-                        break;
-                }
-        }
-unlock:
-        UNLOCK (&inode->lock);
+	if (priv->child_count <= 16)
+		ret = __afr_inode_read_subvol_reset_small (inode, this);
+	else
+		ret = -1;
+
+	return ret;
 }
 
-uint64_t
-afr_is_split_brain (xlator_t *this, inode_t *inode)
+
+int
+afr_inode_read_subvol_get (inode_t *inode, xlator_t *this, unsigned char *data,
+			   unsigned char *metadata, int *event_p)
 {
-        afr_inode_params_t params = {0};
+	int ret = -1;
+
+	LOCK(&inode->lock);
+	{
+		ret = __afr_inode_read_subvol_get (inode, this, data,
+						   metadata, event_p);
+	}
+	UNLOCK(&inode->lock);
 
-        params.mask_type = AFR_ICTX_SPLIT_BRAIN_MASK;
-        afr_inode_get_ctx (this, inode, &params);
-        return params.u.value;
+	return ret;
 }
 
-gf_boolean_t
-afr_is_opendir_done (xlator_t *this, inode_t *inode)
+
+int
+afr_inode_read_subvol_set (inode_t *inode, xlator_t *this, unsigned char *data,
+			   unsigned char *metadata, int event)
 {
-        afr_inode_params_t params = {0};
+	int ret = -1;
+
+	LOCK(&inode->lock);
+	{
+		ret = __afr_inode_read_subvol_set (inode, this, data, metadata,
+						   event);
+	}
+	UNLOCK(&inode->lock);
 
-        params.mask_type = AFR_ICTX_OPENDIR_DONE_MASK;
-        afr_inode_get_ctx (this, inode, &params);
-        return params.u.value;
+	return ret;
 }
 
 
-int32_t
-afr_inode_get_read_ctx (xlator_t *this, inode_t *inode, int32_t *fresh_children)
+int
+afr_inode_read_subvol_reset (inode_t *inode, xlator_t *this)
 {
-        afr_inode_params_t      params = {0};
+	int ret = -1;
 
-        params.mask_type                  = AFR_ICTX_READ_CHILD_MASK;
-        params.u.read_ctx.fresh_children = fresh_children;
-        afr_inode_get_ctx (this, inode, &params);
-        return params.u.read_ctx.read_child;
+	LOCK(&inode->lock);
+	{
+		ret = __afr_inode_read_subvol_reset (inode, this);
+	}
+	UNLOCK(&inode->lock);
+
+	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)
+
+int
+afr_accused_fill (xlator_t *this, dict_t *xdata, unsigned char *accused,
+		  afr_transaction_type type)
 {
-        uint64_t        rest_of_mask = 0;
-        uint64_t        mask         = 0;
-        int             i            = 0;
+	afr_private_t *priv = NULL;
+	int i = 0;
+	int idx = afr_index_for_transaction_type (type);
+	void *pending_raw = NULL;
+	int pending[3];
+	int ret = 0;
 
-        rest_of_mask = (~AFR_ICTX_READ_CHILD_MASK & ctx->masks);
-        mask = (AFR_ICTX_READ_CHILD_MASK & read_child);
-        ctx->masks = rest_of_mask | mask;
+	priv = this->private;
 
-        /* avoid memcpy as int, int32_t are used interchangeably
-         */
-        for (i = 0; i < child_count; i++) {
-                if (fresh_children)
-                        ctx->fresh_children[i] = fresh_children[i];
-                else
-                        ctx->fresh_children[i] = -1;
-        }
+	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));
+
+		if (ntoh32 (pending[idx]))
+			accused[i] = 1;
+	}
+
+	return 0;
 }
 
-void
-afr_inode_ctx_set_opendir_done (afr_inode_ctx_t *ctx)
+
+int
+afr_accuse_smallfiles (xlator_t *this, struct afr_reply *replies,
+		       unsigned char *data_accused)
 {
-        uint64_t        rest_of_mask = 0;
-        uint64_t        mask = 0;
+	int i = 0;
+	afr_private_t *priv = NULL;
+	uint64_t maxsize = 0;
+
+	priv = this->private;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (data_accused[i])
+			continue;
+		if (replies[i].poststat.ia_size > maxsize)
+			maxsize = replies[i].poststat.ia_size;
+	}
 
-        rest_of_mask = (~AFR_ICTX_OPENDIR_DONE_MASK & ctx->masks);
-        mask = (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_OPENDIR_DONE_MASK);
-        ctx->masks = rest_of_mask | mask;
+	for (i = 0; i < priv->child_count; i++) {
+		if (data_accused[i])
+			continue;
+		if (replies[i].poststat.ia_size < maxsize)
+			data_accused[i] = 1;
+	}
+
+	return 0;
 }
 
-void
-afr_inode_ctx_set_splitbrain (afr_inode_ctx_t *ctx, gf_boolean_t set)
-{
-        uint64_t        rest_of_mask = 0;
-        uint64_t        mask = 0;
 
-        if (set) {
-                rest_of_mask = (~AFR_ICTX_SPLIT_BRAIN_MASK & ctx->masks);
-                mask = (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_SPLIT_BRAIN_MASK);
-                ctx->masks = rest_of_mask | mask;
-        } else {
-                ctx->masks = (~AFR_ICTX_SPLIT_BRAIN_MASK & ctx->masks);
-        }
+int
+afr_replies_interpret (call_frame_t *frame, xlator_t *this, inode_t *inode)
+{
+	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);
+
+	for (i = 0; i < priv->child_count; i++) {
+		data_readable[i] = 1;
+		metadata_readable[i] = 1;
+	}
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].valid) {
+			data_readable[i] = 0;
+			metadata_readable[i] = 0;
+			continue;
+		}
+
+		if (replies[i].op_ret == -1) {
+			data_readable[i] = 0;
+			metadata_readable[i] = 0;
+			continue;
+		}
+
+		afr_accused_fill (this, replies[i].xdata, data_accused,
+				  (inode->ia_type == IA_IFDIR) ?
+				   AFR_ENTRY_TRANSACTION : AFR_DATA_TRANSACTION);
+
+		afr_accused_fill (this, replies[i].xdata,
+				  metadata_accused, AFR_METADATA_TRANSACTION);
+
+	}
+
+	if (inode->ia_type != IA_IFDIR)
+		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;
+		}
+	}
+
+	afr_inode_read_subvol_set (inode, this, data_readable,
+				   metadata_readable, event_generation);
+	return ret;
 }
 
-void
-afr_inode_set_ctx (xlator_t *this, inode_t *inode, afr_inode_params_t *params)
-{
-        GF_ASSERT (inode);
-        GF_ASSERT (params);
 
-        int             ret = 0;
-        afr_inode_ctx_t *ctx            = NULL;
-        afr_private_t   *priv           = NULL;
-        uint64_t        ctx_addr        = 0;
-        gf_boolean_t    set             = _gf_false;
-        int32_t         read_child      = -1;
-        int32_t         *fresh_children = NULL;
 
-        priv = this->private;
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx_addr);
-                if (ret < 0)
-                        ctx_addr = 0;
-                ctx = afr_inode_ctx_get_from_addr (ctx_addr, priv->child_count);
-                if (!ctx)
-                        goto unlock;
-                switch (params->mask_type) {
-                case AFR_ICTX_READ_CHILD_MASK:
-                        read_child = params->u.read_ctx.read_child;
-                        fresh_children = params->u.read_ctx.fresh_children;
-                        afr_inode_ctx_set_read_ctx (ctx, read_child,
-                                                    fresh_children,
-                                                    priv->child_count);
-                        break;
-                case AFR_ICTX_OPENDIR_DONE_MASK:
-                        afr_inode_ctx_set_opendir_done (ctx);
-                        break;
-                case AFR_ICTX_SPLIT_BRAIN_MASK:
-                        set = params->u.value;
-                        afr_inode_ctx_set_splitbrain (ctx, set);
-                        break;
-                }
-                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));
-                }
-        }
-unlock:
-        UNLOCK (&inode->lock);
+int
+afr_refresh_selfheal_done (int ret, call_frame_t *heal, void *opaque)
+{
+	if (heal)
+		STACK_DESTROY (heal->root);
+	return 0;
 }
 
-void
-afr_set_split_brain (xlator_t *this, inode_t *inode, gf_boolean_t set)
+int
+afr_inode_refresh_err (call_frame_t *frame, xlator_t *this)
 {
-        afr_inode_params_t      params = {0};
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	int i = 0;
+	int err = 0;
+
+	local = frame->local;
+	priv = this->private;
 
-        params.mask_type        = AFR_ICTX_SPLIT_BRAIN_MASK;
-        params.u.value          = set;
-        afr_inode_set_ctx (this, inode, &params);
+	for (i = 0; i < priv->child_count; i++) {
+		if (local->replies[i].valid && !local->replies[i].op_ret) {
+			err = 0;
+			goto ret;
+		}
+	}
+
+	err = afr_final_errno (local, priv);
+ret:
+	return -err;
 }
 
-void
-afr_set_opendir_done (xlator_t *this, inode_t *inode)
+
+int
+afr_refresh_selfheal_wrap (void *opaque)
 {
-        afr_inode_params_t params = {0};
+	call_frame_t *frame = opaque;
+	afr_local_t *local = NULL;
+	xlator_t *this = NULL;
+	int err = 0;
+
+	local = frame->local;
+	this = frame->this;
+
+	afr_selfheal (frame->this, local->refreshinode->gfid);
 
-        params.mask_type = AFR_ICTX_OPENDIR_DONE_MASK;
-        afr_inode_set_ctx (this, inode, &params);
+	afr_selfheal_unlocked_discover (frame, local->refreshinode,
+					local->refreshinode->gfid,
+					local->replies);
+
+	afr_replies_interpret (frame, this, local->refreshinode);
+
+	err = afr_inode_refresh_err (frame, this);
+
+	afr_replies_wipe (local, this->private);
+
+	local->refreshfn (frame, this, err);
+
+	return 0;
 }
 
-void
-afr_inode_set_read_ctx (xlator_t *this, inode_t *inode, int32_t read_child,
-                        int32_t *fresh_children)
+
+gf_boolean_t
+afr_selfheal_enabled (xlator_t *this)
 {
-        afr_inode_params_t params = {0};
+	afr_private_t *priv = NULL;
+	gf_boolean_t data = _gf_false;
 
-        GF_ASSERT (read_child >= 0);
-        GF_ASSERT (fresh_children);
+	priv = this->private;
 
-        params.mask_type                  = AFR_ICTX_READ_CHILD_MASK;
-        params.u.read_ctx.read_child     = read_child;
-        params.u.read_ctx.fresh_children = fresh_children;
-        afr_inode_set_ctx (this, inode, &params);
+	gf_string2boolean (priv->data_self_heal, &data);
+
+	return data || priv->metadata_self_heal || priv->entry_self_heal;
 }
 
-gf_boolean_t
-afr_is_source_child (int32_t *sources, int32_t child_count, int32_t child)
+
+
+int
+afr_inode_refresh_done (call_frame_t *frame, xlator_t *this)
 {
-        gf_boolean_t             source_xattrs = _gf_false;
+	call_frame_t *heal = NULL;
+	afr_local_t *local = NULL;
+	int ret = 0;
+	int err = 0;
 
-        GF_ASSERT (child < child_count);
+	local = frame->local;
 
-        if ((child >= 0) && (child < child_count) &&
-             sources[child]) {
-                source_xattrs = _gf_true;
-        }
-        return source_xattrs;
+	ret = afr_replies_interpret (frame, this, local->refreshinode);
+
+	err = afr_inode_refresh_err (frame, this);
+
+	afr_replies_wipe (local, this->private);
+
+	if (ret && afr_selfheal_enabled (this)) {
+		heal = copy_frame (frame);
+		if (heal)
+			heal->root->pid = -1;
+		ret = synctask_new (this->ctx->env, afr_refresh_selfheal_wrap,
+				    afr_refresh_selfheal_done, heal, frame);
+		if (ret)
+			goto refresh_done;
+	} else {
+	refresh_done:
+		local->refreshfn (frame, this, err);
+	}
+
+	return 0;
 }
 
-gf_boolean_t
-afr_is_child_present (int32_t *success_children, int32_t child_count,
-                      int32_t child)
+
+int
+afr_inode_refresh_subvol_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)
 {
-        gf_boolean_t             success_child = _gf_false;
-        int                      i = 0;
+	afr_local_t *local = NULL;
+	int call_child = (long) cookie;
+	int call_count = 0;
 
-        GF_ASSERT (child < child_count);
+	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;
+	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;
+		local->replies[call_child].postparent = *par;
+		local->replies[call_child].xdata = dict_ref (xdata);
+	}
+
+	call_count = afr_frame_return (frame);
+
+	if (call_count == 0)
+		afr_inode_refresh_done (frame, this);
+
+	return 0;
 }
 
-gf_boolean_t
-afr_is_read_child (int32_t *success_children, int32_t *sources,
-                   int32_t child_count, int32_t child)
+
+int
+afr_inode_refresh_subvol (call_frame_t *frame, xlator_t *this, int i,
+			  inode_t *inode, dict_t *xdata)
 {
-        gf_boolean_t             success_child = _gf_false;
-        gf_boolean_t             source        = _gf_false;
+	loc_t loc = {0, };
+	afr_private_t *priv = NULL;
 
-        GF_ASSERT (success_children);
-        GF_ASSERT (child_count > 0);
+	priv = this->private;
 
-        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);
-out:
-        return (success_child && source);
+	loc.inode = inode;
+	uuid_copy (loc.gfid, inode->gfid);
+
+	STACK_WIND_COOKIE (frame, afr_inode_refresh_subvol_cbk,
+			   (void *) (long) i, priv->children[i],
+			   priv->children[i]->fops->lookup, &loc, xdata);
+	return 0;
 }
 
-/* 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)
+afr_inode_refresh_do (call_frame_t *frame, xlator_t *this)
 {
-        int32_t                  read_child   = -1;
-        int                      i            = 0;
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	int call_count = 0;
+	int i = 0;
+	dict_t *xdata = NULL;
 
-        GF_ASSERT (success_children);
+	priv = this->private;
+	local = frame->local;
 
-        read_child = prev_read_child;
-        if (afr_is_read_child (success_children, sources, child_count,
-                               read_child))
-                goto out;
+	afr_replies_wipe (local, priv);
 
-        read_child = config_read_child;
-        if (afr_is_read_child (success_children, sources, child_count,
-                               read_child))
-                goto out;
+	xdata = dict_new ();
+	if (!xdata) {
+		afr_inode_refresh_done (frame, this);
+		return 0;
+	}
 
-        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 (afr_xattr_req_prepare (this, xdata) != 0) {
+		dict_unref (xdata);
+		afr_inode_refresh_done (frame, this);
+		return 0;
+	}
 
-out:
-        return read_child;
+	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])
+			continue;
+
+		afr_inode_refresh_subvol (frame, this, i, local->refreshinode,
+					  xdata);
+
+		if (!--call_count)
+			break;
+	}
+
+	dict_unref (xdata);
+
+	return 0;
 }
 
-/* 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)
-{
-        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);
-        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_refresh (call_frame_t *frame, xlator_t *this, inode_t *inode,
+		   afr_inode_refresh_cbk_t refreshfn)
 {
-        int             next_index      = 0;
-        int32_t         next_call_child = -1;
+	afr_local_t *local = NULL;
 
-        GF_ASSERT (last_index);
+	local = frame->local;
 
-        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];
-out:
-        return next_call_child;
+	local->refreshfn = refreshfn;
+
+	if (local->refreshinode) {
+		inode_unref (local->refreshinode);
+		local->refreshinode = NULL;
+	}
+
+	local->refreshinode = inode_ref (inode);
+
+	afr_inode_refresh_do (frame, this);
+
+	return 0;
 }
 
- /* 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_xattr_req_prepare (xlator_t *this, dict_t *xattr_req)
 {
-        int             ret   = 0;
-        afr_private_t   *priv = NULL;
-        int             i     = 0;
+        int             i           = 0;
+        afr_private_t   *priv       = NULL;
+        int             ret         = 0;
 
-        GF_ASSERT (child_up);
-        GF_ASSERT (call_child);
-        GF_ASSERT (last_index);
-        GF_ASSERT (fresh_children);
-        GF_ASSERT (read_child >= 0);
+        priv   = this->private;
 
-        priv = this->private;
-        *call_child = -1;
-        *last_index = -1;
+        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_log (this->name, GF_LOG_WARNING,
+                                "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_log (this->name, GF_LOG_DEBUG, "failed to set dirty "
+                        "query flag");
+        }
 
-        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;
-                        }
-                }
+	return ret;
+}
 
-                if (*call_child == -1) {
-                        ret = -ENOTCONN;
-                        goto out;
-                }
+int
+afr_lookup_xattr_req_prepare (afr_local_t *local, xlator_t *this,
+                              dict_t *xattr_req, loc_t *loc)
+{
+        int     ret = -ENOMEM;
+
+        local->xattr_req = dict_new ();
+        if (!local->xattr_req)
+                goto out;
+        if (xattr_req)
+                dict_copy (xattr_req, local->xattr_req);
 
-                *last_index = i;
+        ret = afr_xattr_req_prepare (this, local->xattr_req);
+	if (ret < 0) {
+		gf_log (this->name, GF_LOG_WARNING,
+			"%s: Unable to prepare 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);
+        }
+
+        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);
+        }
+
+        ret = 0;
 out:
-        gf_log (this->name, GF_LOG_DEBUG, "Returning %d, call_child: %d, "
-                "last_index: %d", ret, *call_child, *last_index);
         return ret;
 }
 
-void
-afr_reset_xattr (dict_t **xattr, unsigned int child_count)
+
+int
+afr_hash_child (inode_t *inode, int32_t child_count, int hashmode)
 {
-        unsigned int i = 0;
+        uuid_t gfid_copy = {0,};
+        pid_t pid;
 
-        if (!xattr)
-                goto out;
-        for (i = 0; i < child_count; i++) {
-                if (xattr[i]) {
-                        dict_unref (xattr[i]);
-                        xattr[i] = NULL;
-                }
+        if (!hashmode) {
+                return -1;
         }
-out:
-        return;
+
+        if (inode) {
+               uuid_copy (gfid_copy, inode->gfid);
+        }
+
+        if (hashmode > 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));
+        }
+
+        return SuperFastHash((char *)gfid_copy,
+                             sizeof(gfid_copy)) % child_count;
 }
 
-void
-afr_local_sh_cleanup (afr_local_t *local, xlator_t *this)
+
+int
+afr_read_subvol_select_by_policy (inode_t *inode, xlator_t *this,
+				  unsigned char *readable)
 {
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              i = 0;
+	afr_private_t *priv = NULL;
+	int read_subvol = -1;
+	int i = 0;
 
+	priv = this->private;
 
-        sh = &local->self_heal;
-        priv = this->private;
+	/* first preference - explicitly specified or local subvolume */
+	if (priv->read_child >= 0 && readable[priv->read_child])
+		return priv->read_child;
 
-        if (sh->buf)
-                GF_FREE (sh->buf);
+	/* second preference - use hashed mode */
+	read_subvol = afr_hash_child (inode, priv->child_count,
+				      priv->hash_mode);
+	if (read_subvol >= 0 && readable[read_subvol])
+		return read_subvol;
 
-        if (sh->parentbufs)
-                GF_FREE (sh->parentbufs);
+	for (i = 0; i < priv->child_count; i++) {
+		if (readable[i])
+			return i;
+	}
 
-        if (sh->inode)
-                inode_unref (sh->inode);
+	/* no readable subvolumes, either split brain or all subvols down */
 
-        if (sh->xattr) {
-                afr_reset_xattr (sh->xattr, priv->child_count);
-                GF_FREE (sh->xattr);
-        }
+	return -1;
+}
 
-        if (sh->child_errno)
-                GF_FREE (sh->child_errno);
 
-        if (sh->pending_matrix) {
-                for (i = 0; i < priv->child_count; i++) {
-                        GF_FREE (sh->pending_matrix[i]);
-                }
-                GF_FREE (sh->pending_matrix);
-        }
-
-        if (sh->delta_matrix) {
-                for (i = 0; i < priv->child_count; i++) {
-                        GF_FREE (sh->delta_matrix[i]);
-                }
-                GF_FREE (sh->delta_matrix);
-        }
+int
+afr_inode_read_subvol_type_get (inode_t *inode, xlator_t *this,
+				unsigned char *readable, int *event_p,
+				int type)
+{
+	int ret = -1;
 
-        if (sh->sources)
-                GF_FREE (sh->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;
+}
 
-        if (sh->success)
-                GF_FREE (sh->success);
 
-        if (sh->locked_nodes)
-                GF_FREE (sh->locked_nodes);
+int
+afr_read_subvol_get (inode_t *inode, xlator_t *this, int *subvol_p,
+		     int *event_p, afr_transaction_type type)
+{
+	afr_private_t *priv = NULL;
+	unsigned char *data_readable = NULL;
+	unsigned char *metadata_readable = NULL;
+	unsigned char *readable = NULL;
+	unsigned char *intersection = NULL;
+	int subvol = -1;
+	int event = 0;
 
-        if (sh->healing_fd && !sh->healing_fd_opened) {
-                fd_unref (sh->healing_fd);
-                sh->healing_fd = NULL;
-        }
+	priv = this->private;
 
-        if (sh->linkname)
-                GF_FREE ((char *)sh->linkname);
+	readable = alloca0 (priv->child_count);
+	data_readable = alloca0 (priv->child_count);
+	metadata_readable = alloca0 (priv->child_count);
+	intersection = alloca0 (priv->child_count);
 
-        if (sh->success_children)
-                GF_FREE (sh->success_children);
+	afr_inode_read_subvol_type_get (inode, this, readable, &event, type);
 
-        if (sh->fresh_children)
-                GF_FREE (sh->fresh_children);
+	afr_inode_read_subvol_get (inode, this, data_readable, metadata_readable,
+				   &event);
 
-        if (sh->fresh_parent_dirs)
-                GF_FREE (sh->fresh_parent_dirs);
+	AFR_INTERSECT (intersection, data_readable, metadata_readable,
+		       priv->child_count);
 
-        loc_wipe (&sh->parent_loc);
+	if (AFR_COUNT (intersection, priv->child_count) > 0)
+		subvol = afr_read_subvol_select_by_policy (inode, this,
+							   intersection);
+	else
+		subvol = afr_read_subvol_select_by_policy (inode, this,
+							   readable);
+	if (subvol_p)
+		*subvol_p = subvol;
+	if (event_p)
+		*event_p = event;
+	return subvol;
 }
 
 
 void
 afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this)
 {
-        int             i = 0;
-        afr_private_t * priv = NULL;
+        afr_private_t *priv    = NULL;
+        int           i        = 0;
 
         priv = this->private;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->pending && local->pending[i])
-                        GF_FREE (local->pending[i]);
-        }
-
-        GF_FREE (local->pending);
+        afr_matrix_cleanup (local->pending, priv->child_count);
 
-        if (local->internal_lock.locked_nodes)
-                GF_FREE (local->internal_lock.locked_nodes);
+        GF_FREE (local->internal_lock.locked_nodes);
 
-        if (local->internal_lock.inode_locked_nodes)
-                GF_FREE (local->internal_lock.inode_locked_nodes);
-
-        if (local->internal_lock.entry_locked_nodes)
-                GF_FREE (local->internal_lock.entry_locked_nodes);
+        for (i = 0; local->internal_lock.inodelk[i].domain; i++) {
+                GF_FREE (local->internal_lock.inodelk[i].locked_nodes);
+        }
 
-        if (local->internal_lock.lower_locked_nodes)
-                GF_FREE (local->internal_lock.lower_locked_nodes);
+        GF_FREE (local->internal_lock.lower_locked_nodes);
 
+        afr_entry_lockee_cleanup (&local->internal_lock);
 
-        GF_FREE (local->transaction.child_errno);
-        GF_FREE (local->child_errno);
+        GF_FREE (local->transaction.pre_op);
+        GF_FREE (local->transaction.eager_lock);
+        GF_FREE (local->transaction.fop_subvols);
+        GF_FREE (local->transaction.failed_subvols);
 
         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_replies_wipe (afr_local_t *local, afr_private_t *priv)
+{
+	int i;
+
+	if (!local->replies)
+		return;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (local->replies[i].xdata) {
+			dict_unref (local->replies[i].xdata);
+			local->replies[i].xdata = NULL;
+		}
+	}
+
+	memset (local->replies, 0, sizeof(*local->replies) * priv->child_count);
+}
+
+void
+afr_remove_eager_lock_stub (afr_local_t *local)
+{
+        LOCK (&local->fd->lock);
+        {
+                list_del_init (&local->transaction.eager_locked);
+        }
+        UNLOCK (&local->fd->lock);
+}
+
+void
 afr_local_cleanup (afr_local_t *local, xlator_t *this)
 {
         afr_private_t * priv = NULL;
@@ -716,7 +909,11 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this)
         if (!local)
                 return;
 
-        afr_local_sh_cleanup (local, this);
+	syncbarrier_destroy (&local->barrier);
+
+        if (local->transaction.eager_lock_on &&
+            !list_empty (&local->transaction.eager_locked))
+                afr_remove_eager_lock_stub (local);
 
         afr_local_transaction_cleanup (local, this);
 
@@ -731,49 +928,36 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this)
         if (local->xattr_req)
                 dict_unref (local->xattr_req);
 
-        if (local->child_up)
-                GF_FREE (local->child_up);
+        if (local->dict)
+                dict_unref (local->dict);
 
-        if (local->fresh_children)
-                GF_FREE (local->fresh_children);
+	afr_replies_wipe (local, priv);
+	GF_FREE(local->replies);
 
-        { /* 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;
-                }
+        GF_FREE (local->child_up);
 
-                if (local->cont.lookup.xattr) {
-                        dict_unref (local->cont.lookup.xattr);
-                }
+        GF_FREE (local->read_attempted);
 
-                if (local->cont.lookup.inode) {
-                        inode_unref (local->cont.lookup.inode);
-                }
+        GF_FREE (local->readable);
 
-                if (local->cont.lookup.postparents)
-                        GF_FREE (local->cont.lookup.postparents);
+	if (local->inode)
+		inode_unref (local->inode);
 
-                if (local->cont.lookup.bufs)
-                        GF_FREE (local->cont.lookup.bufs);
+	if (local->parent)
+		inode_unref (local->parent);
 
-                if (local->cont.lookup.success_children)
-                        GF_FREE (local->cont.lookup.success_children);
+	if (local->parent2)
+		inode_unref (local->parent2);
 
-                if (local->cont.lookup.sources)
-                        GF_FREE (local->cont.lookup.sources);
-        }
+	if (local->refreshinode)
+		inode_unref (local->refreshinode);
 
         { /* getxattr */
-                if (local->cont.getxattr.name)
-                        GF_FREE (local->cont.getxattr.name);
+                GF_FREE (local->cont.getxattr.name);
         }
 
         { /* lk */
-                if (local->cont.lk.locked_nodes)
-                        GF_FREE (local->cont.lk.locked_nodes);
+                GF_FREE (local->cont.lk.locked_nodes);
         }
 
         { /* create */
@@ -800,6 +984,8 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this)
 
         { /* writev */
                 GF_FREE (local->cont.writev.vector);
+                if (local->cont.writev.iobref)
+                        iobref_unref (local->cont.writev.iobref);
         }
 
         { /* setxattr */
@@ -807,18 +993,40 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this)
                         dict_unref (local->cont.setxattr.dict);
         }
 
+        { /* fsetxattr */
+                if (local->cont.fsetxattr.dict)
+                        dict_unref (local->cont.fsetxattr.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);
         }
 
         { /* opendir */
-                if (local->cont.opendir.checksum)
-                        GF_FREE (local->cont.opendir.checksum);
+                GF_FREE (local->cont.opendir.checksum);
         }
+
+        { /* readdirp */
+                if (local->cont.readdir.dict)
+                        dict_unref (local->cont.readdir.dict);
+        }
+
+        if (local->xdata_req)
+                dict_unref (local->xdata_req);
+
+        if (local->xdata_rsp)
+                dict_unref (local->xdata_rsp);
 }
 
 
@@ -840,1302 +1048,1060 @@ afr_frame_return (call_frame_t *frame)
 }
 
 
-/**
- * up_children_count - return the number of children that are up
- */
-
-int
-afr_up_children_count (int child_count, unsigned char *child_up)
-{
-        int i   = 0;
-        int ret = 0;
-
-        for (i = 0; i < child_count; i++)
-                if (child_up[i])
-                        ret++;
-        return ret;
-}
-
 gf_boolean_t
-afr_is_fresh_lookup (loc_t *loc, xlator_t *this)
+afr_is_entry_possibly_under_txn (afr_local_t *local, xlator_t *this)
 {
-        uint64_t          ctx = 0;
-        int32_t           ret = 0;
+	int i = 0;
+	int tmp = 0;
+	afr_private_t *priv = NULL;
 
-        GF_ASSERT (loc);
-        GF_ASSERT (this);
-        GF_ASSERT (loc->inode);
+	priv = this->private;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!local->replies[i].xdata)
+			continue;
+		if (dict_get_int32 (local->replies[i].xdata,
+				    GLUSTERFS_PARENT_ENTRYLK,
+				    &tmp) == 0)
+			if (tmp)
+				return _gf_true;
+	}
 
-        ret = inode_ctx_get (loc->inode, this, &ctx);
-        if (0 == ret)
-                return _gf_false;
-        return _gf_true;
+	return _gf_false;
 }
 
-void
-afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *postparent)
-{
-        GF_ASSERT (loc);
-        GF_ASSERT (buf);
 
-        uuid_copy (loc->gfid, buf->ia_gfid);
-        if (postparent)
-                uuid_copy (loc->pargfid, postparent->ia_gfid);
+/*
+ * 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'.
+ * */
+
+static void
+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;
+	uint64_t size = 0;
+	uint64_t max_size = 0;
+	int readable_cnt = 0;
+
+	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;
+		if (dict_get_uint64 (replies[i].xdata, QUOTA_SIZE_KEY, &size))
+			continue;
+		if (size > max_size)
+			max_size = size;
+	}
+
+	if (!max_size)
+		return;
+
+	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;
+		if (dict_set_uint64 (replies[i].xdata, QUOTA_SIZE_KEY, max_size))
+			continue;
+	}
 }
 
-void
-afr_lookup_build_response_params (afr_local_t *local, xlator_t *this)
+
+static void
+afr_lookup_done (call_frame_t *frame, xlator_t *this)
 {
-        int32_t         read_child = -1;
-        struct iatt     *buf = NULL;
-        struct iatt     *postparent = NULL;
-        dict_t          **xattr = NULL;
+        afr_private_t       *priv  = NULL;
+        afr_local_t         *local = NULL;
+	int                 i = -1;
+	int                 op_errno = 0;
+	int                 read_subvol = 0;
+	unsigned char      *readable = NULL;
+	int                 event = 0;
+	struct afr_reply   *replies = NULL;
+	uuid_t              read_gfid = {0, };
+	gf_boolean_t        locked_entry = _gf_false;
+	gf_boolean_t        can_interpret = _gf_true;
 
-        GF_ASSERT (local);
+        priv  = this->private;
+        local = frame->local;
+	replies = local->replies;
 
-        buf = &local->cont.lookup.buf;
-        postparent = &local->cont.lookup.postparent;
-        xattr = &local->cont.lookup.xattr;
+	locked_entry = afr_is_entry_possibly_under_txn (local, this);
 
-        read_child = afr_inode_get_read_ctx (this, local->cont.lookup.inode,
-                                             NULL);
-        gf_log (this->name, GF_LOG_DEBUG, "Building lookup response from %d",
-                read_child);
-        *xattr = dict_ref (local->cont.lookup.xattrs[read_child]);
-        *buf = local->cont.lookup.bufs[read_child];
-        *postparent = local->cont.lookup.postparents[read_child];
+	readable = alloca0 (priv->child_count);
 
-        if (IA_INVAL == local->cont.lookup.inode->ia_type) {
-                /* fix for RT #602 */
-                local->cont.lookup.inode->ia_type = buf->ia_type;
-        }
-}
+	afr_inode_read_subvol_get (local->loc.parent, this, readable,
+				   NULL, &event);
 
-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;
+	/* First, check if we have an ESTALE from somewhere,
+	   If so, propagate that so that a revalidate can be
+	   issued
+	*/
+	op_errno = afr_final_errno (frame->local, this->private);
+	local->op_errno = op_errno;
+        if (op_errno == ESTALE) {
+		local->op_errno = op_errno;
+		local->op_ret = -1;
+                goto unwind;
+	}
+
+	read_subvol = -1;
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].valid)
+			continue;
+
+		if (locked_entry && replies[i].op_ret == -1 &&
+		    replies[i].op_errno == ENOENT) {
+			/* Second, check entry is still
+			   "underway" in creation */
+			local->op_ret = -1;
+			local->op_errno = ENOENT;
+			read_subvol = i;
+			goto unwind;
+		}
+
+		if (replies[i].op_ret == -1)
+			continue;
+
+		if (read_subvol == -1 || !readable[read_subvol]) {
+			read_subvol = i;
+			uuid_copy (read_gfid, replies[i].poststat.ia_gfid);
+			local->op_ret = 0;
+		}
+	}
+
+	if (read_subvol == -1)
+		goto unwind;
+	/* 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) {
+			if (priv->child_up[i])
+				can_interpret = _gf_false;
+			continue;
+		}
+
+		if (!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;
+
+		/* LOG ERROR */
+		local->op_ret = -1;
+		local->op_errno = EIO;
+		goto unwind;
+	}
+
+	/* Forth, for the finalized GFID, pick the best subvolume
+	   to return stats from.
+	*/
+	if (can_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
+		*/
+		if (afr_replies_interpret (frame, this, local->inode)) {
+			read_subvol = afr_data_subvol_get (local->inode, this,
+							   0, 0);
+			afr_inode_read_subvol_reset (local->inode, this);
+			goto cant_interpret;
+		} else {
+			read_subvol = afr_data_subvol_get (local->inode, this,
+							   0, 0);
+		}
+	} else {
+	cant_interpret:
+		if (read_subvol == -1)
+			dict_del (replies[0].xdata, GF_CONTENT_KEY);
+		else
+			dict_del (replies[read_subvol].xdata, GF_CONTENT_KEY);
+	}
+
+	afr_handle_quota_size (frame, this);
+
+unwind:
+	if (read_subvol == -1)
+		read_subvol = 0;
+
+	AFR_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
+			  local->inode, &local->replies[read_subvol].poststat,
+			  local->replies[read_subvol].xdata,
+			  &local->replies[read_subvol].postparent);
+}
 
-        GF_ASSERT (local);
-        GF_ASSERT (this);
-        GF_ASSERT (xattr);
-        GF_ASSERT (child_index >= 0);
+/*
+ * 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 > ENOENT > others
+ */
 
-        ret = dict_get_uint32 (xattr, GLUSTERFS_INODELK_COUNT,
-                               &inodelk_count);
-        if (ret == 0)
-                local->inodelk_count += inodelk_count;
+int
+afr_higher_errno (int32_t old_errno, int32_t new_errno)
+{
+	if (old_errno == ENODATA || new_errno == ENODATA)
+		return ENODATA;
+	if (old_errno == ESTALE || new_errno == ESTALE)
+		return ESTALE;
+	if (old_errno == ENOENT || new_errno == ENOENT)
+		return ENOENT;
 
-        ret = dict_get_uint32 (xattr, GLUSTERFS_ENTRYLK_COUNT,
-                               &entrylk_count);
-        if (ret == 0)
-                local->entrylk_count += entrylk_count;
+	return new_errno;
 }
 
-static void
-afr_lookup_set_self_heal_data_by_xattr (afr_local_t *local, xlator_t *this,
-                                        dict_t *xattr)
-{
-        GF_ASSERT (local);
-        GF_ASSERT (this);
-        GF_ASSERT (xattr);
 
-        if (afr_sh_has_metadata_pending (xattr, this)) {
-                local->self_heal.need_metadata_self_heal = _gf_true;
-                gf_log(this->name, GF_LOG_DEBUG,
-                       "metadata self-heal is pending for %s.",
-                       local->loc.path);
-        }
+int
+afr_final_errno (afr_local_t *local, afr_private_t *priv)
+{
+	int i = 0;
+	int op_errno = 0;
+	int tmp_errno = 0;
 
-        if (afr_sh_has_entry_pending (xattr, this)) {
-                local->self_heal.need_entry_self_heal = _gf_true;
-                gf_log(this->name, GF_LOG_DEBUG,
-                       "entry self-heal is pending for %s.", local->loc.path);
-        }
+	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);
+	}
 
-        if (afr_sh_has_data_pending (xattr, this)) {
-                local->self_heal.need_data_self_heal = _gf_true;
-                gf_log(this->name, GF_LOG_DEBUG,
-                       "data self-heal is pending for %s.", local->loc.path);
-        }
+	return op_errno;
 }
 
-static void
-afr_detect_self_heal_by_iatt (afr_local_t *local, xlator_t *this,
-                            struct iatt *buf, struct iatt *lookup_buf)
+static int
+get_pathinfo_host (char *pathinfo, char *hostname, size_t size)
 {
-        if (PERMISSION_DIFFERS (buf, lookup_buf)) {
-                /* mismatching permissions */
-                gf_log (this->name, GF_LOG_INFO,
-                        "permissions differ for %s ", local->loc.path);
-                local->self_heal.need_metadata_self_heal = _gf_true;
-        }
+        char    *start = NULL;
+        char    *end = NULL;
+        int     ret  = -1;
+        int     i    = 0;
 
-        if (OWNERSHIP_DIFFERS (buf, lookup_buf)) {
-                /* mismatching permissions */
-                local->self_heal.need_metadata_self_heal = _gf_true;
-                gf_log (this->name, GF_LOG_INFO,
-                        "ownership differs for %s ", local->loc.path);
-        }
+        if (!pathinfo)
+                goto out;
 
-        if (SIZE_DIFFERS (buf, lookup_buf)
-            && IA_ISREG (buf->ia_type)) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "size differs for %s ", local->loc.path);
-                local->self_heal.need_data_self_heal = _gf_true;
-        }
+        start = strchr (pathinfo, ':');
+        if (!start)
+                goto out;
+        end = strrchr (pathinfo, ':');
+        if (start == end)
+                goto out;
 
-        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);
-        }
+        memset (hostname, 0, size);
+        i = 0;
+        while (++start != end)
+                hostname[i++] = *start;
+        ret = 0;
+out:
+        return ret;
 }
 
-static void
-afr_detect_self_heal_by_lookup_status (afr_local_t *local, xlator_t *this)
+int
+afr_local_pathinfo (char *pathinfo, gf_boolean_t *local)
 {
-        GF_ASSERT (local);
-        GF_ASSERT (this);
+        int             ret   = 0;
+        char            pathinfohost[1024] = {0};
+        char            localhost[1024] = {0};
+        xlator_t        *this = THIS;
 
-        if ((local->success_count > 0) && (local->enoent_count > 0)) {
-                local->self_heal.need_metadata_self_heal = _gf_true;
-                local->self_heal.need_data_self_heal     = _gf_true;
-                local->self_heal.need_entry_self_heal    = _gf_true;
-                local->self_heal.need_gfid_self_heal    = _gf_true;
-                local->self_heal.need_missing_entry_self_heal    = _gf_true;
-                gf_log(this->name, GF_LOG_INFO,
-                       "entries are missing in lookup of %s.",
-                       local->loc.path);
-                //If all self-heals are needed no need to check for other rules
+        *local = _gf_false;
+        ret = get_pathinfo_host (pathinfo, pathinfohost, sizeof (pathinfohost));
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Invalid pathinfo: %s",
+                        pathinfo);
                 goto out;
         }
 
-        if ((local->success_count > 0) &&
-            afr_is_split_brain (this, local->cont.lookup.inode) &&
-            IA_ISREG (local->cont.lookup.inode->ia_type)) {
-                local->self_heal.need_data_self_heal = _gf_true;
-                local->self_heal.need_gfid_self_heal    = _gf_true;
-                local->self_heal.need_missing_entry_self_heal    = _gf_true;
-                gf_log (this->name, GF_LOG_WARNING,
-                        "split brain detected during lookup of %s.",
-                        local->loc.path);
+        ret = gethostname (localhost, sizeof (localhost));
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "gethostname() failed, "
+                        "reason: %s", strerror (errno));
+                goto out;
         }
 
+        if (!strcmp (localhost, pathinfohost))
+                *local = _gf_true;
 out:
-        return;
+        return ret;
 }
 
-gf_boolean_t
-afr_can_self_heal_proceed (afr_self_heal_t *sh, afr_private_t *priv)
+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)
 {
-        GF_ASSERT (sh);
-        GF_ASSERT (priv);
-
-        return (sh->need_gfid_self_heal
-                || sh->need_missing_entry_self_heal
-                || (priv->data_self_heal && sh->need_data_self_heal)
-                || (priv->metadata_self_heal && sh->need_metadata_self_heal)
-                || (priv->entry_self_heal && sh->need_entry_self_heal));
-}
+        int              ret            = 0;
+        char            *pathinfo       = NULL;
+        gf_boolean_t     is_local        = _gf_false;
+        afr_private_t   *priv           = NULL;
+        int32_t          child_index    = -1;
 
-afr_transaction_type
-afr_transaction_type_get (ia_type_t ia_type)
-{
-        afr_transaction_type    type = AFR_METADATA_TRANSACTION;
+        if (op_ret != 0) {
+                goto out;
+        }
 
-        GF_ASSERT (ia_type != IA_INVAL);
+	priv = this->private;
+	child_index = (int32_t)(long)cookie;
 
-        if (IA_ISDIR (ia_type)) {
-                type = AFR_ENTRY_TRANSACTION;
-        } else if (IA_ISREG (ia_type)) {
-                type = AFR_DATA_TRANSACTION;
+        ret = dict_get_str (dict, GF_XATTR_PATHINFO_KEY, &pathinfo);
+        if (ret != 0) {
+                goto out;
         }
-        return type;
-}
-
-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;
-        struct iatt             *bufs          = NULL;
-        afr_transaction_type    type           = AFR_METADATA_TRANSACTION;
-
-        GF_ASSERT (local);
-        GF_ASSERT (this);
-        GF_ASSERT (local->success_count > 0);
 
-        bufs = local->cont.lookup.bufs;
-        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;
-        source = afr_lookup_select_read_child_by_txn_type (this, local, xattrs,
-                                                           type);
-        if (source < 0) {
-                gf_log (this->name, GF_LOG_DEBUG, "failed to select source "
-                        "for %s", local->loc.path);
+        ret = afr_local_pathinfo (pathinfo, &is_local);
+        if (ret) {
                 goto out;
         }
 
-        gf_log (this->name, GF_LOG_DEBUG, "Source selected as %d for %s",
-                source, local->loc.path);
-        *read_child = source;
-        ret = 0;
+        /*
+         * 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) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "selecting local read_child %s",
+                        priv->children[child_index]->name);
+                priv->read_child = child_index;
+        }
 out:
-        return ret;
-}
-
-static inline gf_boolean_t
-afr_is_self_heal_running (afr_local_t *local)
-{
-        GF_ASSERT (local);
-        return ((local->inodelk_count > 0) || (local->entrylk_count > 0));
+        STACK_DESTROY(frame->root);
+        return 0;
 }
 
 static void
-afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode,
-                      gf_boolean_t is_background, ia_type_t ia_type,
-                      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))
+afr_attempt_local_discovery (xlator_t *this, int32_t child_index)
 {
-        afr_local_t             *local = NULL;
-        char                    sh_type_str[256] = {0,};
-
-        GF_ASSERT (frame);
-        GF_ASSERT (this);
-        GF_ASSERT (inode);
+        call_frame_t    *newframe = NULL;
+        loc_t            tmploc = {0,};
+        afr_private_t   *priv = this->private;
 
-        local = frame->local;
-        local->self_heal.background = is_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));
-
-        gf_log (this->name, GF_LOG_INFO,
-                "background %s self-heal triggered. path: %s",
-                sh_type_str, local->loc.path);
+        newframe = create_frame(this,this->ctx->pool);
+        if (!newframe) {
+                return;
+        }
 
-        afr_self_heal (frame, this, inode);
+        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);
 }
 
+
 int
-afr_gfid_missing_count (const char *xlator_name, int32_t *success_children,
-                        struct iatt *bufs, unsigned int child_count,
-                        const char *path)
+afr_lookup_selfheal_wrap (void *opaque)
 {
-        int             gfid_miss_count   = 0;
-        int             i              = 0;
-        struct iatt     *child1        = NULL;
+	call_frame_t *frame = opaque;
+	afr_local_t *local = NULL;
+	xlator_t *this = NULL;
+	inode_t *inode = 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++;
-                }
-        }
+	local = frame->local;
+	this = frame->this;
 
-        return gfid_miss_count;
-}
+	afr_selfheal_name (frame->this, local->loc.pargfid, local->loc.name);
 
-static int
-afr_lookup_gfid_missing_count (afr_local_t *local, xlator_t *this)
-{
-        int32_t         *success_children = NULL;
-        afr_private_t   *priv          = NULL;
-        struct iatt     *bufs          = NULL;
-        int             miss_count     = 0;
+	afr_replies_wipe (local, this->private);
 
-        priv = this->private;
-        bufs = local->cont.lookup.bufs;
-        success_children = local->cont.lookup.success_children;
-
-        miss_count =  afr_gfid_missing_count (this->name, success_children,
-                                              bufs, priv->child_count,
-                                              local->loc.path);
-        return miss_count;
-}
+	inode = afr_selfheal_unlocked_lookup_on (frame, local->loc.parent,
+						 local->loc.name, local->replies,
+						 local->child_up);
+	if (inode)
+		inode_unref (inode);
+	afr_lookup_done (frame, this);
 
-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;
+	return 0;
 }
 
-/* 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:
-        return;
+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 need_heal = _gf_false;
+	struct afr_reply *replies = NULL;
+	int ret = 0;
+
+	local = frame->local;
+	replies = local->replies;
+	priv = this->private;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].valid)
+			continue;
+
+		if (first == -1) {
+			first = i;
+			continue;
+		}
+
+		if (replies[i].op_ret != replies[first].op_ret) {
+			need_heal = _gf_true;
+			break;
+		}
+
+		if (uuid_compare (replies[i].poststat.ia_gfid,
+				  replies[first].poststat.ia_gfid)) {
+			need_heal = _gf_true;
+			break;
+		}
+	}
+
+	if (need_heal) {
+		heal = copy_frame (frame);
+		if (heal)
+			heal->root->pid = -1;
+		ret = synctask_new (this->ctx->env, afr_lookup_selfheal_wrap,
+				    afr_refresh_selfheal_done, heal, frame);
+		if (ret)
+			goto lookup_done;
+	} else {
+	lookup_done:
+		afr_lookup_done (frame, this);
+	}
+
+	return ret;
 }
 
-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;
 
-        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;
-}
-
-static void
-afr_lookup_set_self_heal_data (afr_local_t *local, xlator_t *this)
+int
+afr_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+		int op_ret, int op_errno, inode_t *inode, struct iatt *buf,
+		dict_t *xdata, struct iatt *postparent)
 {
-        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_local_t *   local = NULL;
+        int             call_count      = -1;
+        int             child_index     = -1;
 
-        priv  = this->private;
-        afr_detect_self_heal_by_lookup_status (local, this);
+	child_index = (long) cookie;
 
-        if (afr_lookup_gfid_missing_count (local, this))
-                local->self_heal.need_gfid_self_heal    = _gf_true;
+	local = frame->local;
 
-        if (_gf_true == afr_lookup_conflicting_entries (local, this))
-                local->self_heal.need_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);
+	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);
+	}
 
-        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]);
+        call_count = afr_frame_return (frame);
+        if (call_count == 0) {
+		afr_lookup_entry_heal (frame, this);
         }
 
-        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_data_by_xattr (local, this,
-                                                        xattr[child1]);
-        }
+	return 0;
 }
 
-int
-afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno)
+
+
+static void
+afr_discover_done (call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
+        afr_private_t       *priv  = NULL;
+        afr_local_t         *local = NULL;
+	int                 i = -1;
+	int                 op_errno = 0;
+	int                 read_subvol = 0;
 
+        priv  = this->private;
         local = frame->local;
 
-        if (op_ret == -1) {
-                local->op_ret = -1;
-                if (afr_error_more_important (local->op_errno, op_errno))
-                        local->op_errno = op_errno;
+	for (i = 0; i < priv->child_count; i++) {
+		if (!local->replies[i].valid)
+			continue;
+		if (local->replies[i].op_ret == 0)
+			local->op_ret = 0;
+	}
 
-                goto out;
-        } else {
-                local->op_ret = 0;
-        }
+	op_errno = afr_final_errno (frame->local, this->private);
 
-        afr_lookup_done_success_action (frame, this, _gf_true);
-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);
+        if (local->op_ret < 0) {
+		local->op_errno = op_errno;
+		local->op_ret = -1;
+                goto unwind;
+	}
 
-        return 0;
-}
+	afr_replies_interpret (frame, this, local->inode);
 
-//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;
-
-        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;
+	read_subvol = afr_data_subvol_get (local->inode, this, 0, 0);
+	if (read_subvol == -1) {
+		gf_log (this->name, GF_LOG_WARNING, "no read subvols for %s",
+			local->loc.path);
 
-        memcpy (lookup_bufs, sh->buf, priv->child_count * sizeof (*sh->buf));
-        memcpy (lookup_parentbufs, sh->parentbufs,
-                priv->child_count * sizeof (*sh->parentbufs));
+		for (i = 0; i < priv->child_count; i++) {
+			if (!local->replies[i].valid ||
+			    local->replies[i].op_ret == -1)
+				continue;
+			read_subvol = i;
+			break;
+		}
+	}
 
-        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;
-        }
+unwind:
+	if (read_subvol == -1)
+		read_subvol = 0;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->xattr[i])
-                        local->cont.lookup.xattrs[i] = dict_ref (sh->xattr[i]);
-        }
-
-        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_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);
 }
 
-static void
-afr_lookup_perform_self_heal_if_needed (call_frame_t *frame, xlator_t *this,
-                                        gf_boolean_t *sh_launched)
+
+int
+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)
 {
-        size_t              up_count = 0;
-        afr_private_t       *priv    = NULL;
-        afr_local_t         *local   = NULL;
+        afr_local_t *   local = NULL;
+        int             call_count      = -1;
+        int             child_index     = -1;
 
-        GF_ASSERT (sh_launched);
-        *sh_launched = _gf_false;
-        priv         = this->private;
-        local        = frame->local;
+	child_index = (long) cookie;
 
-        up_count  = afr_up_children_count (priv->child_count, local->child_up);
-        if (up_count == 1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Only 1 child up - do not attempt to detect self heal");
-                goto out;
-        }
+	local = frame->local;
 
-        afr_lookup_set_self_heal_data (local, this);
-        if (afr_can_self_heal_proceed (&local->self_heal, priv)) {
-                if  (afr_is_self_heal_running (local))
-                        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 != -1) {
+		local->replies[child_index].poststat = *buf;
+		local->replies[child_index].postparent = *postparent;
+		if (xdata)
+			local->replies[child_index].xdata = dict_ref (xdata);
+	}
 
-                afr_launch_self_heal (frame, this, local->cont.lookup.inode,
-                                      _gf_true, local->cont.lookup.buf.ia_type,
-                                      afr_post_gfid_sh_success,
-                                      afr_self_heal_lookup_unwind);
-                *sh_launched = _gf_true;
-        }
-out:
-        return;
-}
+        if (local->do_discovery && (op_ret == 0))
+                afr_attempt_local_discovery (this, child_index);
 
-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++;
-                }
+        call_count = afr_frame_return (frame);
+        if (call_count == 0) {
+               afr_discover_done (frame, this);
         }
+
+	return 0;
 }
 
-static int
-afr_lookup_set_read_ctx (afr_local_t *local, xlator_t *this, int32_t read_child)
+
+int
+afr_discover_do (call_frame_t *frame, xlator_t *this, int err)
 {
-        afr_private_t           *priv = NULL;
+	int ret = 0;
+	int i = 0;
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	int call_count = 0;
 
-        GF_ASSERT (read_child >= 0);
+	local = frame->local;
+	priv = this->private;
 
-        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);
+	if (err) {
+		local->op_errno = -err;
+		ret = -1;
+		goto out;
+	}
 
-        return 0;
-}
+	call_count = local->call_count = AFR_COUNT (local->child_up,
+						    priv->child_count);
 
-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;
-        afr_private_t       *priv      = NULL;
-
-        local   = frame->local;
-        priv    = this->private;
-
-        if (local->loc.parent == NULL)
-                fail_conflict = _gf_true;
-
-        if (afr_conflicting_iattrs (local->cont.lookup.bufs,
-                                     local->cont.lookup.success_children,
-                                     priv->child_count, local->loc.path,
-                                     this->name)) {
-                if (fail_conflict == _gf_false) {
-                        ret = 0;
-                } else {
-                        local->op_ret = -1;
-                        local->op_errno = EIO;
-                }
-                goto out;
-        }
-
-        ret = afr_lookup_select_read_child (local, this, &read_child);
+        ret = afr_lookup_xattr_req_prepare (local, this, local->xattr_req,
+					    &local->loc);
         if (ret) {
-                local->op_ret = -1;
-                local->op_errno = EIO;
+                local->op_errno = -ret;
+		ret = -1;
                 goto out;
         }
 
-        ret = afr_lookup_set_read_ctx (local, this, read_child);
-        if (ret) {
-                local->op_ret = -1;
-                local->op_errno = EIO;
-                goto out;
-        }
-
-        afr_lookup_build_response_params (local, this);
-        if (afr_is_fresh_lookup (&local->loc, this)) {
-                afr_update_loc_gfids (&local->loc,
-                                      &local->cont.lookup.buf,
-                                      &local->cont.lookup.postparent);
+        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;
+                }
         }
 
-        ret = 0;
+	return 0;
 out:
-        return ret;
+	AFR_STACK_UNWIND (lookup, frame, -1, local->op_errno, 0, 0, 0, 0);
+	return 0;
 }
 
-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;
-        int                 gfid_miss_count = 0;
-        int                 enotconn_count = 0;
-        int                 up_children_count = 0;
 
-        priv  = this->private;
-        local = frame->local;
-
-        if (local->op_ret < 0)
-                goto unwind;
-        gfid_miss_count = afr_lookup_gfid_missing_count (local, this);
-        up_children_count = afr_up_children_count (priv->child_count,
-                                                   local->child_up);
-        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;
-        }
+int
+afr_discover (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
+{
+	int op_errno = ENOMEM;
+	afr_private_t *priv = NULL;
+	afr_local_t *local = NULL;
+	int event = 0;
 
-        ret = afr_lookup_done_success_action (frame, this, _gf_false);
-        if (ret)
-                goto unwind;
-        uuid_copy (local->self_heal.sh_gfid_req, local->cont.lookup.gfid_req);
+	priv = this->private;
 
-        afr_lookup_perform_self_heal_if_needed (frame, this, &sh_launched);
-        if (sh_launched) {
-                unwind = 0;
-                goto unwind;
-        }
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
- 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 (!local->call_count) {
+                op_errno = ENOTCONN;
+                goto out;
         }
-}
 
-/*
- * 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 > ENOENT > others
- *
- */
-
-gf_boolean_t
-afr_error_more_important (int32_t old_errno, int32_t new_errno)
-{
-        gf_boolean_t ret = _gf_true;
+	if (__is_root_gfid (loc->inode->gfid)) {
+		if (!this->itable)
+			this->itable = loc->inode->table;
+		if (!priv->root_inode)
+			priv->root_inode = inode_ref (loc->inode);
 
-        /* Nothing should ever overwrite ESTALE */
-        if (old_errno == ESTALE)
-                ret = _gf_false;
-
-        /* Nothing should overwrite ENOENT, except ESTALE */
-        else if ((old_errno == ENOENT) && (new_errno != ESTALE))
-                ret = _gf_false;
+		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;
+                }
+	}
 
-        return ret;
-}
+        local->op = GF_FOP_LOOKUP;
 
-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;
+        loc_copy (&local->loc, loc);
 
-        for (i = 0; i < child_count; i++) {
-                if (children) {
-                        child = children[i];
-                        if (child == -1)
-                                break;
-                } else {
-                        child = i;
-                }
-                if (afr_error_more_important (op_errno, child_errno[child]))
-                                op_errno = child_errno[child];
-        }
-        return op_errno;
-}
+	local->inode = inode_ref (loc->inode);
 
-static void
-afr_lookup_handle_error (afr_local_t *local, int32_t op_ret,  int32_t op_errno)
-{
-        GF_ASSERT (local);
-        if (op_errno == ENOENT)
-                local->enoent_count++;
+	if (xattr_req)
+		/* If xattr_req was null, afr_lookup_xattr_req_prepare() will
+		   allocate one for us */
+		local->xattr_req = dict_ref (xattr_req);
 
-        if (afr_error_more_important (local->op_errno, op_errno))
-                local->op_errno = op_errno;
+	if (uuid_is_null (loc->inode->gfid)) {
+		afr_discover_do (frame, this, 0);
+		return 0;
+	}
 
-        if (local->op_errno == ESTALE) {
-                local->op_ret = -1;
-        }
-}
+	afr_read_subvol_get (loc->inode, this, NULL, &event,
+			     AFR_DATA_TRANSACTION);
 
-static void
-afr_set_root_inode_on_first_lookup (afr_local_t *local, xlator_t *this,
-                                    inode_t *inode)
-{
-        afr_private_t           *priv = NULL;
-        GF_ASSERT (inode);
+	if (event != local->event_generation)
+		afr_inode_refresh (frame, this, loc->inode, afr_discover_do);
+	else
+		afr_discover_do (frame, this, 0);
 
-        if (inode->ino != 1)
-                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;
-        }
+	return 0;
 out:
-        return;
+	AFR_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+	return 0;
 }
 
-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;
-}
-
-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);
-}
-
-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)
-{
-        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);
-
-        afr_lookup_cache_args (local, child_index, xattr,
-                               buf, postparent);
-        local->cont.lookup.success_children[local->success_count] = child_index;
-        local->success_count++;
-}
 
 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_lookup_do (call_frame_t *frame, xlator_t *this, int err)
 {
-        afr_local_t *   local = NULL;
-        int             call_count      = -1;
-        int             child_index     = -1;
+	int ret = 0;
+	int i = 0;
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	int call_count = 0;
 
-         child_index = (long) cookie;
+	local = frame->local;
+	priv = this->private;
 
-        LOCK (&frame->lock);
-        {
-                local = frame->local;
+	if (err < 0) {
+		local->op_errno = -err;
+		ret = -1;
+		goto out;
+	}
 
-                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 = local->call_count = AFR_COUNT (local->child_up,
+						    priv->child_count);
 
-        call_count = afr_frame_return (frame);
-        if (call_count == 0) {
-               afr_lookup_done (frame, this);
+        ret = afr_lookup_xattr_req_prepare (local, this, local->xattr_req,
+					    &local->loc);
+        if (ret) {
+                local->op_errno = -ret;
+		ret = -1;
+                goto out;
         }
 
-         return 0;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND_COOKIE (frame, afr_lookup_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->lookup,
+                                           &local->loc, local->xattr_req);
+                        if (!--call_count)
+                                break;
+                }
+        }
+	return 0;
+out:
+	AFR_STACK_UNWIND (lookup, frame, -1, local->op_errno, 0, 0, 0, 0);
+	return 0;
 }
 
+/*
+ * afr_lookup()
+ *
+ * The goal here is to figure out what the element getting looked up is.
+ * i.e what is the GFID, inode type and a conservative estimate of the
+ * inode attributes are.
+ *
+ * As we lookup, operations may be underway on the entry name and the
+ * inode. In lookup() we are primarily concerned only with the entry
+ * operations. If the entry is getting unlinked or renamed, we detect
+ * what operation is underway by querying for on-going transactions and
+ * pending self-healing on the entry through xdata.
+ *
+ * If the entry is a file/dir, it may need self-heal and/or in a
+ * split-brain condition. Lookup is not the place to worry about these
+ * conditions. Outcast marking will naturally handle them in the read
+ * paths.
+ *
+ * Here is a brief goal of what we are trying to achieve:
+ *
+ * - LOOKUP on all subvolumes concurrently, querying on-going transaction
+ *   and pending self-heal info from the servers.
+ *
+ * - If all servers reply the same inode type and GFID, the overall call
+ *   MUST be a success.
+ *
+ * - If inode types or GFIDs mismatch, and there IS either an on-going
+ *   transaction or pending self-heal, inspect what the nature of the
+ *   transaction or pending heal is, and select the appropriate subvolume's
+ *   reply as the winner.
+ *
+ * - If inode types or GFIDs mismatch, and there are no on-going transactions
+ *   or pending self-heal on the entry name on any of the servers, fail the
+ *   lookup with EIO. Something has gone wrong beyond reasonable action.
+ */
+
 int
-afr_lookup_cont_init (afr_local_t *local, unsigned int child_count)
+afr_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-        int               ret            = -ENOMEM;
-        struct iatt       *iatts         = NULL;
-        int32_t           *success_children = NULL;
+        afr_local_t   *local = NULL;
+        int32_t        op_errno = 0;
+	int            event = 0;
 
-        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;
+	if (!loc->parent) {
+		afr_discover (frame, this, loc, xattr_req);
+		return 0;
+	}
 
-        iatts = GF_CALLOC (child_count, sizeof (*iatts), gf_afr_mt_iatt);
-        if (NULL == iatts)
-                goto out;
-        local->cont.lookup.postparents = iatts;
+	if (__is_root_gfid (loc->parent->gfid)) {
+		if (!strcmp (loc->name, GF_REPLICATE_TRASH_DIR)) {
+			op_errno = EPERM;
+			goto out;
+		}
+	}
 
-        iatts = GF_CALLOC (child_count, sizeof (*iatts), gf_afr_mt_iatt);
-        if (NULL == iatts)
-                goto out;
-        local->cont.lookup.bufs = iatts;
-
-        success_children = afr_fresh_children_create (child_count);
-        if (NULL == success_children)
-                goto out;
-        local->cont.lookup.success_children = success_children;
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
-        local->fresh_children = afr_fresh_children_create (child_count);
-        if (NULL == local->fresh_children)
+        if (!local->call_count) {
+                op_errno = ENOTCONN;
                 goto out;
+        }
 
-        ret = 0;
-out:
-        return ret;
-}
+        local->op = GF_FOP_LOOKUP;
 
-int
-afr_lookup (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, dict_t *xattr_req)
-{
-        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;
+        loc_copy (&local->loc, loc);
 
-        priv = this->private;
+	local->inode = inode_ref (loc->inode);
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+	if (xattr_req)
+		/* If xattr_req was null, afr_lookup_xattr_req_prepare() will
+		   allocate one for us */
+		local->xattr_req = dict_ref (xattr_req);
 
-        local->op_ret = -1;
+	afr_read_subvol_get (loc->parent, this, NULL, &event,
+			     AFR_DATA_TRANSACTION);
 
-        frame->local = local;
+	if (event != local->event_generation)
+		afr_inode_refresh (frame, this, loc->parent, afr_lookup_do);
+	else
+		afr_lookup_do (frame, this, 0);
 
-        if (!strcmp (loc->path, "/" GF_REPLICATE_TRASH_DIR)) {
-                op_errno = ENOENT;
-                goto out;
-        }
+	return 0;
+out:
+	AFR_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
 
-        loc_copy (&local->loc, loc);
+        return 0;
+}
 
-        ret = inode_ctx_get (loc->inode, this, &ctx);
-        if (ret == 0) {
-                /* lookup is a revalidate */
 
-                local->read_child_index = afr_inode_get_read_ctx (this,
-                                                                  loc->inode,
-                                                                  NULL);
-        } else {
-                LOCK (&priv->read_child_lock);
-                {
-                        local->read_child_index = (++priv->read_child_rr)
-                                % (priv->child_count);
-                }
-                UNLOCK (&priv->read_child_lock);
-        }
+/* {{{ open */
 
-        if (loc->parent)
-                local->cont.lookup.parent_ino = loc->parent->ino;
+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;
 
-        local->child_up = memdup (priv->child_up, priv->child_count);
-        if (NULL == local->child_up) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+        ret = __fd_ctx_get (fd, this, &ctx);
 
-        ret = afr_lookup_cont_init (local, priv->child_count);
         if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
-
-        local->call_count = afr_up_children_count (priv->child_count,
-                                                   local->child_up);
-        call_count = local->call_count;
+                ret = __afr_fd_ctx_set (this, fd);
+                if (ret < 0)
+                        goto out;
 
-        if (local->call_count == 0) {
-                ret      = -1;
-                op_errno = ENOTCONN;
-                goto out;
+                ret = __fd_ctx_get (fd, this, &ctx);
+                if (ret < 0)
+                        goto out;
         }
 
-        /* By default assume ENOTCONN. On success it will be set to 0. */
-        local->op_errno = ENOTCONN;
+        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+out:
+        return fd_ctx;
+}
 
-        if (xattr_req == NULL)
-                local->xattr_req = dict_new ();
-        else
-                local->xattr_req = dict_ref (xattr_req);
 
-        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);
-        }
-
-        ret = dict_get_ptr (xattr_req, "gfid-req", &gfid_req);
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "failed to get the gfid from dict");
-        } else {
-                uuid_copy (local->cont.lookup.gfid_req, gfid_req);
-        }
-        if (local->loc.parent != NULL)
-                dict_del (xattr_req, "gfid-req");
+afr_fd_ctx_t *
+afr_fd_ctx_get (fd_t *fd, xlator_t *this)
+{
+        afr_fd_ctx_t  *fd_ctx = NULL;
 
-        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,
-                                           loc, local->xattr_req);
-                        if (!--call_count)
-                                break;
-                }
+        LOCK(&fd->lock);
+        {
+                fd_ctx = __afr_fd_ctx_get (fd, this);
         }
+        UNLOCK(&fd->lock);
 
-        ret = 0;
-out:
-        if (ret == -1)
-                AFR_STACK_UNWIND (lookup, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL);
-
-        return 0;
+        return fd_ctx;
 }
 
 
-/* {{{ open */
-
 int
-afr_fd_ctx_set (xlator_t *this, fd_t *fd)
+__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;
 
         VALIDATE_OR_GOTO (this->private, out);
         VALIDATE_OR_GOTO (fd, out);
 
         priv = this->private;
 
-        LOCK (&fd->lock);
-        {
-                ret = __fd_ctx_get (fd, this, &ctx);
+        ret = __fd_ctx_get (fd, this, &ctx);
 
-                if (ret == 0)
-                        goto unlock;
+        if (ret == 0)
+                goto out;
 
-                fd_ctx = GF_CALLOC (1, sizeof (afr_fd_ctx_t),
-                                    gf_afr_mt_afr_fd_ctx_t);
-                if (!fd_ctx) {
-                        ret = -ENOMEM;
-                        goto unlock;
-                }
+        fd_ctx = GF_CALLOC (1, sizeof (afr_fd_ctx_t),
+                            gf_afr_mt_afr_fd_ctx_t);
+        if (!fd_ctx) {
+                ret = -ENOMEM;
+                goto out;
+        }
 
-                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 unlock;
-                }
+	for (i = 0; i < AFR_NUM_CHANGE_LOGS; i++) {
+		fd_ctx->pre_op_done[i] = GF_CALLOC (sizeof (*fd_ctx->pre_op_done[i]),
+						    priv->child_count,
+						    gf_afr_mt_int32_t);
+		if (!fd_ctx->pre_op_done[i]) {
+			ret = -ENOMEM;
+			goto out;
+		}
+	}
+
+        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;
+        }
 
-                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 unlock;
-                }
+	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;
+	}
 
-                fd_ctx->opened_on = GF_CALLOC (sizeof (*fd_ctx->opened_on),
-                                               priv->child_count,
-                                               gf_afr_mt_char);
-                if (!fd_ctx->opened_on) {
-                        ret = -ENOMEM;
-                        goto unlock;
-                }
+        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;
+        }
 
-                fd_ctx->up_count   = priv->up_count;
-                fd_ctx->down_count = priv->down_count;
+        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;
+        }
 
-                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 unlock;
-                }
+	pthread_mutex_init (&fd_ctx->delay_lock, NULL);
 
-                ret = __fd_ctx_set (fd, this, (uint64_t)(long) fd_ctx);
-                if (ret)
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "failed to set fd ctx (%p)", fd);
+        INIT_LIST_HEAD (&fd_ctx->eager_locked);
 
-                INIT_LIST_HEAD (&fd_ctx->entries);
-        }
-unlock:
-        UNLOCK (&fd->lock);
+        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;
 }
 
-/* {{{ flush */
 
 int
-afr_flush_unwind (call_frame_t *frame, xlator_t *this)
+afr_fd_ctx_set (xlator_t *this, fd_t *fd)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
-
-        local = frame->local;
+        int ret = -1;
 
-        LOCK (&frame->lock);
+        LOCK (&fd->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 (flush, main_frame,
-                                  local->op_ret, local->op_errno);
+                ret = __afr_fd_ctx_set (this, fd);
         }
+        UNLOCK (&fd->lock);
 
-        return 0;
+        return ret;
 }
 
+/* {{{ flush */
 
 int
-afr_flush_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno)
+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;
-        afr_private_t * priv  = NULL;
-        int call_count  = -1;
-        int child_index = (long) cookie;
-        int need_unwind = 0;
+        afr_local_t *local = NULL;
+        int call_count = -1;
 
         local = frame->local;
-        priv  = this->private;
 
         LOCK (&frame->lock);
         {
-                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->success_count++;
-
-                        if (local->success_count == priv->wait_count) {
-                                need_unwind = 1;
-                        }
-                }
-
-                local->op_errno = op_errno;
+			local->op_ret = op_ret;
+			if (!local->xdata_rsp && xdata)
+				local->xdata_rsp = dict_ref (xdata);
+		} else {
+			local->op_errno = op_errno;
+		}
         }
         UNLOCK (&frame->lock);
 
-        if (need_unwind)
-                afr_flush_unwind (frame, this);
+	call_count = afr_frame_return (frame);
 
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+	if (call_count == 0)
+		AFR_STACK_UNWIND (flush, frame, local->op_ret,
+				  local->op_errno, local->xdata_rsp);
 
         return 0;
 }
 
-
-int
-afr_flush_wind (call_frame_t *frame, xlator_t *this)
+static int
+afr_flush_wrapper (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int i = 0;
-        int call_count = -1;
+        int           i      = 0;
+        afr_local_t   *local = NULL;
+        afr_private_t *priv  = NULL;
+        int call_count       = -1;
 
-        local = frame->local;
         priv = this->private;
-
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
+        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_wind_cbk,
+                        STACK_WIND_COOKIE (frame, afr_flush_cbk,
                                            (void *) (long) i,
                                            priv->children[i],
                                            priv->children[i]->fops->flush,
-                                           local->fd);
-
+                                           local->fd, xdata);
                         if (!--call_count)
                                 break;
+
                 }
         }
 
         return 0;
 }
 
-
 int
-afr_flush_done (call_frame_t *frame, xlator_t *this)
+afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
+        afr_local_t   *local = NULL;
+        call_stub_t   *stub = NULL;
+        int            op_errno   = ENOMEM;
 
-        local = frame->local;
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
-        local->transaction.unwind (frame, this);
+	if (!local->call_count) {
+		op_errno = ENOTCONN;
+		goto out;
+	}
 
-        AFR_STACK_DESTROY (frame);
+	local->fd = fd_ref(fd);
 
-        return 0;
-}
-
-
-int
-afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
-{
-        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;
-        int call_count = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv = this->private;
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
+        stub = fop_flush_stub (frame, afr_flush_wrapper, fd, xdata);
+        if (!stub)
                 goto out;
-        }
-
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
 
-        transaction_frame->local = local;
+        afr_delayed_changelog_wake_resume (this, fd, stub);
 
-        local->op = GF_FOP_FLUSH;
-
-        local->transaction.fop    = afr_flush_wind;
-        local->transaction.done   = afr_flush_done;
-        local->transaction.unwind = afr_flush_unwind;
-
-        local->fd                 = fd_ref (fd);
-
-        local->transaction.main_frame = frame;
-        local->transaction.start  = 0;
-        local->transaction.len    = 0;
-
-        afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
-
-
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-
-                AFR_STACK_UNWIND (flush, frame, op_ret, op_errno);
-        }
-
+	AFR_STACK_UNWIND (flush, frame, -1, op_errno, NULL);
         return 0;
 }
 
@@ -2148,6 +2114,7 @@ afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd)
         uint64_t        ctx = 0;
         afr_fd_ctx_t    *fd_ctx = NULL;
         int             ret = 0;
+	int             i = 0;
 
         ret = fd_ctx_get (fd, this, &ctx);
         if (ret < 0)
@@ -2156,17 +2123,22 @@ afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd)
         fd_ctx = (afr_fd_ctx_t *)(long) ctx;
 
         if (fd_ctx) {
-                if (fd_ctx->pre_op_done)
-                        GF_FREE (fd_ctx->pre_op_done);
+                //no need to take any locks
+                if (!list_empty (&fd_ctx->eager_locked))
+                        gf_log (this->name, GF_LOG_WARNING, "%s: Stale "
+                                "Eager-lock stubs found",
+                                uuid_utoa (fd->inode->gfid));
+
+		for (i = 0; i < AFR_NUM_CHANGE_LOGS; i++)
+			GF_FREE (fd_ctx->pre_op_done[i]);
+
+                GF_FREE (fd_ctx->opened_on);
 
-                if (fd_ctx->opened_on)
-                        GF_FREE (fd_ctx->opened_on);
+                GF_FREE (fd_ctx->lock_piggyback);
 
-                if (fd_ctx->locked_on)
-                        GF_FREE (fd_ctx->locked_on);
+                GF_FREE (fd_ctx->lock_acquired);
 
-                if (fd_ctx->pre_op_piggyback)
-                        GF_FREE (fd_ctx->pre_op_piggyback);
+		pthread_mutex_destroy (&fd_ctx->delay_lock);
 
                 GF_FREE (fd_ctx);
         }
@@ -2179,24 +2151,8 @@ out:
 int
 afr_release (xlator_t *this, fd_t *fd)
 {
-        afr_locked_fd_t *locked_fd = NULL;
-        afr_locked_fd_t *tmp       = NULL;
-        afr_private_t   *priv      = NULL;
-
-        priv = this->private;
-
         afr_cleanup_fd_ctx (this, fd);
 
-        list_for_each_entry_safe (locked_fd, tmp, &priv->saved_fds,
-                                  list) {
-
-                if (locked_fd->fd == fd) {
-                        list_del_init (&locked_fd->list);
-                        GF_FREE (locked_fd);
-                }
-
-        }
-
         return 0;
 }
 
@@ -2204,51 +2160,87 @@ afr_release (xlator_t *this, fd_t *fd)
 /* {{{ fsync */
 
 int
+afr_fsync_unwind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+        AFR_STACK_UNWIND (fsync, frame, op_ret, op_errno, prebuf, postbuf,
+                          xdata);
+        return 0;
+}
+
+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)
+               struct iatt *postbuf, dict_t *xdata)
 {
         afr_local_t *local = NULL;
         int call_count = -1;
         int child_index = (long) cookie;
-        int read_child  = 0;
+	int read_subvol = 0;
+	call_stub_t *stub = NULL;
 
         local = frame->local;
 
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+	read_subvol = afr_data_subvol_get (local->inode, this, 0, 0);
 
         LOCK (&frame->lock);
         {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
-
                 if (op_ret == 0) {
-                        local->op_ret = 0;
+                        if (local->op_ret == -1) {
+				local->op_ret = 0;
 
-                        if (local->success_count == 0) {
-                                local->cont.fsync.prebuf  = *prebuf;
-                                local->cont.fsync.postbuf = *postbuf;
-                        }
+                                local->cont.inode_wfop.prebuf  = *prebuf;
+                                local->cont.inode_wfop.postbuf = *postbuf;
 
-                        if (child_index == read_child) {
-                                local->cont.fsync.prebuf  = *prebuf;
-                                local->cont.fsync.postbuf = *postbuf;
+				if (xdata)
+					local->xdata_rsp = dict_ref (xdata);
                         }
 
-                        local->success_count++;
-                }
-
-                local->op_errno = op_errno;
+                        if (child_index == read_subvol) {
+                                local->cont.inode_wfop.prebuf  = *prebuf;
+                                local->cont.inode_wfop.postbuf = *postbuf;
+				if (xdata) {
+					if (local->xdata_rsp)
+						dict_unref (local->xdata_rsp);
+					local->xdata_rsp = dict_ref (xdata);
+				}
+                        }
+                } else {
+			local->op_errno = op_errno;
+		}
         }
         UNLOCK (&frame->lock);
 
         call_count = afr_frame_return (frame);
 
         if (call_count == 0) {
-                AFR_STACK_UNWIND (fsync, frame, local->op_ret, local->op_errno,
-                                  &local->cont.fsync.prebuf,
-                                  &local->cont.fsync.postbuf);
+		/* Make a stub out of the frame, and register it
+		   with the waking up post-op. When the call-stub resumes,
+		   we are guaranteed that there was no post-op pending
+		   (i.e changelogs were unset in the server). This is an
+		   essential "guarantee", that fsync() returns only after
+		   completely finishing EVERYTHING, including the delayed
+		   post-op. This guarantee is expected by FUSE graph switching
+		   for example.
+		*/
+		stub = fop_fsync_cbk_stub (frame, afr_fsync_unwind_cbk,
+                                           local->op_ret, local->op_errno,
+                                           &local->cont.inode_wfop.prebuf,
+                                           &local->cont.inode_wfop.postbuf,
+                                           local->xdata_rsp);
+		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);
         }
 
         return 0;
@@ -2256,36 +2248,34 @@ afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 
 int
-afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd,
-           int32_t datasync)
+afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+	   dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
+	afr_private_t *priv = NULL;
         afr_local_t *local = NULL;
-        int ret = -1;
         int i = 0;
         int32_t call_count = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
+        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;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+        call_count = local->call_count;
+	if (!call_count) {
+		op_errno = ENOTCONN;
+		goto out;
+	}
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+        local->fd = fd_ref (fd);
 
-        call_count = local->call_count;
-        frame->local = local;
+	if (afr_fd_has_witnessed_unstable_write (this, fd)) {
+		/* don't care. we only wanted to CLEAR the bit */
+	}
 
-        local->fd             = fd_ref (fd);
-        local->cont.fsync.ino = fd->inode->ino;
+	local->inode = inode_ref (fd->inode);
 
         for (i = 0; i < priv->child_count; i++) {
                 if (local->child_up[i]) {
@@ -2293,17 +2283,16 @@ afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd,
                                            (void *) (long) i,
                                            priv->children[i],
                                            priv->children[i]->fops->fsync,
-                                           fd, datasync);
+                                           fd, datasync, xdata);
                         if (!--call_count)
                                 break;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (fsync, frame, op_ret, op_errno, NULL, NULL);
-        }
+	AFR_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
+
         return 0;
 }
 
@@ -2311,9 +2300,9 @@ out:
 
 /* {{{ fsync */
 
-int32_t
-afr_fsyncdir_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno)
+int
+afr_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+		  int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
         afr_local_t *local = NULL;
         int call_count = -1;
@@ -2322,10 +2311,13 @@ afr_fsyncdir_cbk (call_frame_t *frame, void *cookie,
 
         LOCK (&frame->lock);
         {
-                if (op_ret == 0)
+                if (op_ret == 0) {
                         local->op_ret = 0;
-
-                local->op_errno = op_errno;
+			if (!local->xdata_rsp && xdata)
+				local->xdata_rsp = dict_ref (xdata);
+		} else {
+			local->op_errno = op_errno;
+		}
         }
         UNLOCK (&frame->lock);
 
@@ -2333,57 +2325,49 @@ afr_fsyncdir_cbk (call_frame_t *frame, void *cookie,
 
         if (call_count == 0)
                 AFR_STACK_UNWIND (fsyncdir, frame, local->op_ret,
-                                  local->op_errno);
+				  local->op_errno, local->xdata_rsp);
 
         return 0;
 }
 
 
-int32_t
-afr_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              int32_t datasync)
+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_private_t *priv = NULL;
         afr_local_t *local = NULL;
-        int ret = -1;
         int i = 0;
         int32_t call_count = 0;
-        int32_t op_ret   = -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;
+        int32_t op_errno = ENOMEM;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+	priv = this->private;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+        local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
         call_count = local->call_count;
-        frame->local = local;
+	if (!call_count) {
+		op_errno = ENOTCONN;
+		goto out;
+	}
 
         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);
+                                    fd, datasync, xdata);
                         if (!--call_count)
                                 break;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (fsyncdir, frame, op_ret, op_errno);
-        }
+	AFR_STACK_UNWIND (fsyncdir, frame, -1, op_errno, NULL);
+
         return 0;
 }
 
@@ -2394,7 +2378,7 @@ out:
 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 *xattr, dict_t *xdata)
 {
         afr_local_t *local = NULL;
         int call_count = -1;
@@ -2403,8 +2387,15 @@ afr_xattrop_cbk (call_frame_t *frame, void *cookie,
 
         LOCK (&frame->lock);
         {
-                if (op_ret == 0)
+                if (op_ret == 0) {
+                        if (!local->cont.xattrop.xattr)
+                                local->cont.xattrop.xattr = dict_ref (xattr);
+
+			if (!local->xdata_rsp && xdata)
+				local->xdata_rsp = dict_ref (xdata);
+
                         local->op_ret = 0;
+                }
 
                 local->op_errno = op_errno;
         }
@@ -2414,7 +2405,7 @@ afr_xattrop_cbk (call_frame_t *frame, void *cookie,
 
         if (call_count == 0)
                 AFR_STACK_UNWIND (xattrop, frame, local->op_ret, local->op_errno,
-                                  xattr);
+                local->cont.xattrop.xattr, local->xdata_rsp);
 
         return 0;
 }
@@ -2422,49 +2413,41 @@ afr_xattrop_cbk (call_frame_t *frame, void *cookie,
 
 int32_t
 afr_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             gf_xattrop_flags_t optype, dict_t *xattr)
+             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_ret   = -1;
-        int32_t op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        int32_t op_errno = ENOMEM;
 
         priv = this->private;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
         call_count = local->call_count;
-        frame->local = local;
+	if (!call_count) {
+		op_errno = ENOTCONN;
+		goto out;
+	}
 
         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);
+                                    loc, optype, xattr, xdata);
                         if (!--call_count)
                                 break;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (xattrop, frame, op_ret, op_errno, NULL);
-        }
+	AFR_STACK_UNWIND (xattrop, frame, -1, op_errno, NULL, NULL);
+
         return 0;
 }
 
@@ -2475,7 +2458,7 @@ out:
 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 *xattr, dict_t *xdata)
 {
         afr_local_t *local = NULL;
 
@@ -2485,8 +2468,14 @@ afr_fxattrop_cbk (call_frame_t *frame, void *cookie,
 
         LOCK (&frame->lock);
         {
-                if (op_ret == 0)
+                if (op_ret == 0) {
+                        if (!local->cont.fxattrop.xattr)
+                                local->cont.fxattrop.xattr = dict_ref (xattr);
+
+			if (!local->xdata_rsp && xdata)
+				local->xdata_rsp = dict_ref (xdata);
                         local->op_ret = 0;
+                }
 
                 local->op_errno = op_errno;
         }
@@ -2496,7 +2485,7 @@ afr_fxattrop_cbk (call_frame_t *frame, void *cookie,
 
         if (call_count == 0)
                 AFR_STACK_UNWIND (fxattrop, frame, local->op_ret, local->op_errno,
-                                  xattr);
+                                  local->cont.fxattrop.xattr, local->xdata_rsp);
 
         return 0;
 }
@@ -2504,49 +2493,41 @@ afr_fxattrop_cbk (call_frame_t *frame, void *cookie,
 
 int32_t
 afr_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              gf_xattrop_flags_t optype, dict_t *xattr)
+              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_ret   = -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;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
                 goto out;
-        }
 
         call_count = local->call_count;
-        frame->local = local;
+	if (!call_count) {
+		op_errno = ENOTCONN;
+		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);
+                                    fd, optype, xattr, xdata);
                         if (!--call_count)
                                 break;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (fxattrop, frame, op_ret, op_errno, NULL);
-        }
+	AFR_STACK_UNWIND (fxattrop, frame, -1, op_errno, NULL, NULL);
+
         return 0;
 }
 
@@ -2554,8 +2535,8 @@ out:
 
 
 int32_t
-afr_inodelk_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno)
+afr_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+		 int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
 {
         afr_local_t *local = NULL;
@@ -2576,7 +2557,7 @@ afr_inodelk_cbk (call_frame_t *frame, void *cookie,
 
         if (call_count == 0)
                 AFR_STACK_UNWIND (inodelk, frame, local->op_ret,
-                                  local->op_errno);
+                                  local->op_errno, xdata);
 
         return 0;
 }
@@ -2584,57 +2565,50 @@ afr_inodelk_cbk (call_frame_t *frame, void *cookie,
 
 int32_t
 afr_inodelk (call_frame_t *frame, xlator_t *this,
-             const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *flock)
+             const char *volume, loc_t *loc, 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_ret   = -1;
-        int32_t op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        int32_t op_errno = ENOMEM;
 
         priv = this->private;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
+        local = AFR_FRAME_INIT (frame, op_errno);
+        if (!local)
                 goto out;
-        }
 
         call_count = local->call_count;
-        frame->local = local;
+	if (!call_count) {
+		op_errno = ENOMEM;
+		goto out;
+	}
 
         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);
+                                    volume, loc, cmd, flock, xdata);
 
                         if (!--call_count)
                                 break;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (inodelk, frame, op_ret, op_errno);
-        }
+	AFR_STACK_UNWIND (inodelk, frame, -1, op_errno, NULL);
+
         return 0;
 }
 
 
 int32_t
-afr_finodelk_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno)
+afr_finodelk_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;
@@ -2655,66 +2629,57 @@ afr_finodelk_cbk (call_frame_t *frame, void *cookie,
 
         if (call_count == 0)
                 AFR_STACK_UNWIND (finodelk, frame, local->op_ret,
-                                  local->op_errno);
+                                  local->op_errno, xdata);
 
         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)
+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_ret   = -1;
-        int32_t op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        int32_t op_errno = ENOMEM;
 
         priv = this->private;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
         call_count = local->call_count;
-        frame->local = local;
+	if (!call_count) {
+		op_errno = ENOTCONN;
+		goto out;
+	}
 
         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);
+                                    volume, fd, cmd, flock, xdata);
 
                         if (!--call_count)
                                 break;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (finodelk, frame, op_ret, op_errno);
-        }
+	AFR_STACK_UNWIND (finodelk, frame, -1, op_errno, NULL);
+
         return 0;
 }
 
 
 int32_t
-afr_entrylk_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno)
-
+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;
@@ -2734,67 +2699,59 @@ afr_entrylk_cbk (call_frame_t *frame, void *cookie,
 
         if (call_count == 0)
                 AFR_STACK_UNWIND (entrylk, frame, local->op_ret,
-                                  local->op_errno);
+                                  local->op_errno, xdata);
 
         return 0;
 }
 
 
-int32_t
-afr_entrylk (call_frame_t *frame, xlator_t *this,
-             const char *volume, loc_t *loc,
-             const char *basename, entrylk_cmd cmd, entrylk_type type)
+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_private_t *priv = NULL;
         afr_local_t *local  = NULL;
-        int ret = -1;
         int i = 0;
         int32_t call_count = 0;
-        int32_t op_ret   = -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;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
         call_count = local->call_count;
-        frame->local = local;
+	if (!call_count) {
+		op_errno = ENOTCONN;
+		goto out;
+	}
 
         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);
+                                    volume, loc, basename, cmd, type, xdata);
 
                         if (!--call_count)
                                 break;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (entrylk, frame, op_ret, op_errno);
-        }
+	AFR_STACK_UNWIND (entrylk, frame, -1, op_errno, NULL);
+
         return 0;
 }
 
 
 
-int32_t
-afr_fentrylk_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno)
+int
+afr_fentrylk_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;
@@ -2815,156 +2772,148 @@ afr_fentrylk_cbk (call_frame_t *frame, void *cookie,
 
         if (call_count == 0)
                 AFR_STACK_UNWIND (fentrylk, frame, local->op_ret,
-                                  local->op_errno);
+                                  local->op_errno, xdata);
 
         return 0;
 }
 
 
-int32_t
-afr_fentrylk (call_frame_t *frame, xlator_t *this,
-              const char *volume, fd_t *fd,
-              const char *basename, entrylk_cmd cmd, entrylk_type type)
+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_private_t *priv = NULL;
         afr_local_t *local  = NULL;
-        int ret = -1;
         int i = 0;
         int32_t call_count = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        int32_t op_errno = ENOMEM;
 
         priv = this->private;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+        local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
         call_count = local->call_count;
-        frame->local = local;
+	if (!call_count) {
+		op_errno = ENOTCONN;
+		goto out;
+	}
 
         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);
+                                    volume, fd, basename, cmd, type, xdata);
 
                         if (!--call_count)
                                 break;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (fentrylk, frame, op_ret, op_errno);
-        }
+	AFR_STACK_UNWIND (fentrylk, frame, -1, op_errno, NULL);
+
         return 0;
 }
 
-int32_t
-afr_statfs_cbk (call_frame_t *frame, void *cookie,
-                xlator_t *this, int32_t op_ret, int32_t op_errno,
-                struct statvfs *statvfs)
+
+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;
 
         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;
-                        }
-                }
-
-                if (op_ret == -1)
+                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:
         UNLOCK (&frame->lock);
 
         call_count = afr_frame_return (frame);
 
         if (call_count == 0)
                 AFR_STACK_UNWIND (statfs, frame, local->op_ret, local->op_errno,
-                                  &local->cont.statfs.buf);
+                                  &local->cont.statfs.buf, local->xdata_rsp);
 
         return 0;
 }
 
 
-int32_t
-afr_statfs (call_frame_t *frame, xlator_t *this,
-            loc_t *loc)
+int
+afr_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        afr_private_t *  priv        = NULL;
-        int              child_count = 0;
         afr_local_t   *  local       = NULL;
+	afr_private_t   *priv        = NULL;
         int              i           = 0;
-        int              ret = -1;
         int              call_count = 0;
-        int32_t          op_ret      = -1;
-        int32_t          op_errno    = 0;
+        int32_t          op_errno    = ENOMEM;
 
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (loc, out);
+	priv = this->private;
 
-        priv = this->private;
-        child_count = priv->child_count;
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
-
-        frame->local = local;
         call_count = local->call_count;
+	if (!call_count) {
+		op_errno = ENOTCONN;
+		goto out;
+	}
 
-        for (i = 0; i < child_count; i++) {
+        for (i = 0; i < priv->child_count; i++) {
                 if (local->child_up[i]) {
                         STACK_WIND (frame, afr_statfs_cbk,
                                     priv->children[i],
                                     priv->children[i]->fops->statfs,
-                                    loc);
+                                    loc, xdata);
                         if (!--call_count)
                                 break;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (statfs, frame, op_ret, op_errno, NULL);
-        }
+	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)
+                   int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                   dict_t *xdata)
 {
         afr_local_t * local = NULL;
         int call_count = -1;
@@ -2974,7 +2923,7 @@ afr_lk_unlock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         if (call_count == 0)
                 AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  lock);
+                                  lock, xdata);
 
         return 0;
 }
@@ -2996,7 +2945,7 @@ afr_lk_unlock (call_frame_t *frame, xlator_t *this)
 
         if (call_count == 0) {
                 AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  &local->cont.lk.ret_flock);
+                                  &local->cont.lk.ret_flock, NULL);
                 return 0;
         }
 
@@ -3010,7 +2959,7 @@ afr_lk_unlock (call_frame_t *frame, xlator_t *this)
                                     priv->children[i],
                                     priv->children[i]->fops->lk,
                                     local->fd, F_SETLK,
-                                    &local->cont.lk.user_flock);
+                                    &local->cont.lk.user_flock, NULL);
 
                         if (!--call_count)
                                 break;
@@ -3023,7 +2972,7 @@ afr_lk_unlock (call_frame_t *frame, xlator_t *this)
 
 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)
+            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;
@@ -3058,30 +3007,15 @@ afr_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                    priv->children[child_index],
                                    priv->children[child_index]->fops->lk,
                                    local->fd, local->cont.lk.cmd,
-                                   &local->cont.lk.user_flock);
+                                   &local->cont.lk.user_flock, xdata);
         } else if (local->op_ret == -1) {
                 /* all nodes have gone down */
 
                 AFR_STACK_UNWIND (lk, frame, -1, ENOTCONN,
-                                  &local->cont.lk.ret_flock);
+                                  &local->cont.lk.ret_flock, NULL);
         } else {
-                /* locking has succeeded on all nodes that are up */
-
-                /* 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");
-
-                   ret = afr_save_locked_fd (this, local->fd);
-                   if (ret)
-                   gf_log (this->name, GF_LOG_DEBUG,
-                   "Could not save locked fd");
-
-                */
                 AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  &local->cont.lk.ret_flock);
+                                  &local->cont.lk.ret_flock, NULL);
         }
 
         return 0;
@@ -3090,24 +3024,18 @@ afr_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 int
 afr_lk (call_frame_t *frame, xlator_t *this,
-        fd_t *fd, int32_t cmd, struct gf_flock *flock)
+        fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
         afr_private_t *priv = NULL;
         afr_local_t *local = NULL;
         int i = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        int32_t op_errno = ENOMEM;
 
         priv = this->private;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        AFR_LOCAL_INIT (local, priv);
-
-        frame->local  = local;
+        local = AFR_FRAME_INIT (frame, op_errno);
+        if (!local)
+                goto out;
 
         local->cont.lk.locked_nodes = GF_CALLOC (priv->child_count,
                                                  sizeof (*local->cont.lk.locked_nodes),
@@ -3126,32 +3054,18 @@ afr_lk (call_frame_t *frame, xlator_t *this,
         STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) 0,
                            priv->children[i],
                            priv->children[i]->fops->lk,
-                           fd, cmd, flock);
+                           fd, cmd, flock, xdata);
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (lk, frame, op_ret, op_errno, NULL);
-        }
+	AFR_STACK_UNWIND (lk, frame, -1, op_errno, NULL, NULL);
+
         return 0;
 }
 
 int
 afr_forget (xlator_t *this, inode_t *inode)
 {
-        uint64_t        ctx_addr = 0;
-        afr_inode_ctx_t *ctx     = NULL;
-
-        inode_ctx_get (inode, this, &ctx_addr);
-
-        if (!ctx_addr)
-                goto out;
-
-        ctx = (afr_inode_ctx_t *)(long)ctx_addr;
-        if (ctx->fresh_children)
-                GF_FREE (ctx->fresh_children);
-        GF_FREE (ctx);
-out:
         return 0;
 }
 
@@ -3170,41 +3084,22 @@ afr_priv_dump (xlator_t *this)
         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_build_key(key, key_prefix, "child_count");
-        gf_proc_dump_write(key, "%u", priv->child_count);
-        gf_proc_dump_build_key(key, key_prefix, "read_child_rr");
-        gf_proc_dump_write(key, "%u", priv->read_child_rr);
+        gf_proc_dump_write("child_count", "%u", priv->child_count);
         for (i = 0; i < priv->child_count; i++) {
-                gf_proc_dump_build_key(key, key_prefix, "child_up[%d]", i);
+                sprintf (key, "child_up[%d]", i);
                 gf_proc_dump_write(key, "%d", priv->child_up[i]);
-                gf_proc_dump_build_key(key, key_prefix,
-                                       "pending_key[%d]", i);
+                sprintf (key, "pending_key[%d]", i);
                 gf_proc_dump_write(key, "%s", priv->pending_key[i]);
         }
-        gf_proc_dump_build_key(key, key_prefix, "data_self_heal");
-        gf_proc_dump_write(key, "%d", priv->data_self_heal);
-        gf_proc_dump_build_key(key, key_prefix, "metadata_self_heal");
-        gf_proc_dump_write(key, "%d", priv->metadata_self_heal);
-        gf_proc_dump_build_key(key, key_prefix, "entry_self_heal");
-        gf_proc_dump_write(key, "%d", priv->entry_self_heal);
-        gf_proc_dump_build_key(key, key_prefix, "data_change_log");
-        gf_proc_dump_write(key, "%d", priv->data_change_log);
-        gf_proc_dump_build_key(key, key_prefix, "metadata_change_log");
-        gf_proc_dump_write(key, "%d", priv->metadata_change_log);
-        gf_proc_dump_build_key(key, key_prefix, "entry_change_log");
-        gf_proc_dump_write(key, "%d", priv->entry_change_log);
-        gf_proc_dump_build_key(key, key_prefix, "read_child");
-        gf_proc_dump_write(key, "%d", priv->read_child);
-        gf_proc_dump_build_key(key, key_prefix, "favorite_child");
-        gf_proc_dump_write(key, "%d", priv->favorite_child);
-        gf_proc_dump_build_key(key, key_prefix, "data_lock_server_count");
-        gf_proc_dump_write(key, "%u", priv->data_lock_server_count);
-        gf_proc_dump_build_key(key, key_prefix, "metadata_lock_server_count");
-        gf_proc_dump_write(key, "%u", priv->metadata_lock_server_count);
-        gf_proc_dump_build_key(key, key_prefix, "entry_lock_server_count");
-        gf_proc_dump_write(key, "%u", priv->entry_lock_server_count);
-        gf_proc_dump_build_key(key, key_prefix, "wait_count");
-        gf_proc_dump_write(key, "%u", priv->wait_count);
+        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);
 
         return 0;
 }
@@ -3234,24 +3129,35 @@ find_child_index (xlator_t *this, xlator_t *child)
 
 int32_t
 afr_notify (xlator_t *this, int32_t event,
-            void *data, ...)
+            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            = -1;
+        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]) {
@@ -3262,7 +3168,7 @@ afr_notify (xlator_t *this, int32_t event,
         /* 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 optimises revalidate lookup by sending
+         * 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
@@ -3279,9 +3185,20 @@ afr_notify (xlator_t *this, int32_t 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->event_generation++;
+                        }
                         priv->child_up[idx] = 1;
-                        priv->up_count++;
 
+                        call_psh = 1;
+                        up_child = idx;
                         for (i = 0; i < priv->child_count; i++)
                                 if (priv->child_up[i] == 1)
                                         up_children++;
@@ -3302,8 +3219,23 @@ afr_notify (xlator_t *this, int32_t event,
         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->event_generation++;
+                        }
                         priv->child_up[idx] = 0;
-                        priv->down_count++;
 
                         for (i = 0; i < priv->child_count; i++)
                                 if (priv->child_up[i] == 0)
@@ -3328,7 +3260,20 @@ afr_notify (xlator_t *this, int32_t event,
                         priv->last_event[idx] = event;
                 }
                 UNLOCK (&priv->lock);
+
                 break;
+
+        case GF_EVENT_TRANSLATOR_OP:
+                input = data;
+                output = data2;
+                if (!had_heard_from_all) {
+                        ret = -1;
+                        goto out;
+                }
+                ret = afr_xl_op (this, input, output);
+                goto out;
+                break;
+
         default:
                 propagate = 1;
                 break;
@@ -3355,6 +3300,7 @@ afr_notify (xlator_t *this, int32_t event,
 
                 LOCK (&priv->lock);
                 {
+                        up_children = AFR_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;
@@ -3375,52 +3321,69 @@ afr_notify (xlator_t *this, int32_t event,
         if (propagate)
                 ret = default_notify (this, event, data);
 
+        if (call_psh && priv->shd.iamshd) {
+                afr_selfheal_childup (this, up_child);
+	}
 out:
         return ret;
 }
 
-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;
-                }
-        }
-
-        return ret;
-}
 
 int
-AFR_LOCAL_INIT (afr_local_t *local, afr_private_t *priv)
+afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno)
 {
         local->op_ret = -1;
         local->op_errno = EUCLEAN;
-        local->call_count = afr_up_children_count (priv->child_count,
-                                                   priv->child_up);
-        if (local->call_count == 0) {
-                gf_log (THIS->name, GF_LOG_INFO, "no subvolumes up");
-                return -ENOTCONN;
-        }
 
+	syncbarrier_init (&local->barrier);
 
-        local->child_up = GF_CALLOC (sizeof (*local->child_up),
-                                     priv->child_count,
+        local->child_up = GF_CALLOC (priv->child_count,
+                                     sizeof (*local->child_up),
                                      gf_afr_mt_char);
         if (!local->child_up) {
-                return -ENOMEM;
+                if (op_errno)
+                        *op_errno = ENOMEM;
+                goto out;
         }
 
         memcpy (local->child_up, priv->child_up,
                 sizeof (*local->child_up) * priv->child_count);
-
-        return 0;
+        local->call_count = AFR_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;
+        }
+	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->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;
+	}
+
+	return 0;
+out:
+        return -1;
 }
 
 int
@@ -3429,16 +3392,6 @@ afr_internal_lock_init (afr_internal_lock_t *lk, size_t child_count,
 {
         int             ret = -ENOMEM;
 
-        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;
-
-        lk->entry_locked_nodes = GF_CALLOC (sizeof (*lk->entry_locked_nodes),
-                                            child_count, gf_afr_mt_char);
-        if (NULL == lk->entry_locked_nodes)
-                goto out;
-
         lk->locked_nodes = GF_CALLOC (sizeof (*lk->locked_nodes),
                                       child_count, gf_afr_mt_char);
         if (NULL == lk->locked_nodes)
@@ -3458,119 +3411,219 @@ out:
         return ret;
 }
 
+void
+afr_matrix_cleanup (int32_t **matrix, unsigned int m)
+{
+        int             i         = 0;
+
+        if (!matrix)
+                goto out;
+        for (i = 0; i < m; i++) {
+                GF_FREE (matrix[i]);
+        }
+
+        GF_FREE (matrix);
+out:
+        return;
+}
+
+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;
+
+        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;
+}
+
+int
+afr_inodelk_init (afr_inodelk_t *lk, char *dom, size_t child_count)
+{
+        int             ret = -ENOMEM;
+
+        lk->domain = dom;
+        lk->locked_nodes = GF_CALLOC (sizeof (*lk->locked_nodes),
+                                      child_count, gf_afr_mt_char);
+        if (NULL == lk->locked_nodes)
+                goto out;
+        ret = 0;
+out:
+        return ret;
+}
+
 int
-afr_transaction_local_init (afr_local_t *local, afr_private_t *priv)
+afr_transaction_local_init (afr_local_t *local, xlator_t *this)
 {
-        int i;
-        int child_up_count = 0;
-        int ret = -ENOMEM;
+        int            child_up_count = 0;
+        int            ret = -ENOMEM;
+        afr_private_t *priv = NULL;
 
+        priv = this->private;
         ret = afr_internal_lock_init (&local->internal_lock, priv->child_count,
                                       AFR_TRANSACTION_LK);
         if (ret < 0)
                 goto out;
 
+        if ((local->transaction.type == AFR_DATA_TRANSACTION) ||
+            (local->transaction.type == AFR_METADATA_TRANSACTION)) {
+                ret = afr_inodelk_init (&local->internal_lock.inodelk[0],
+                                        this->name, priv->child_count);
+                if (ret < 0)
+                        goto out;
+        }
+
         ret = -ENOMEM;
-        child_up_count = afr_up_children_count (priv->child_count, local->child_up);
+        child_up_count = AFR_COUNT (local->child_up, priv->child_count);
         if (priv->optimistic_change_log && child_up_count == priv->child_count)
                 local->optimistic_change_log = 1;
 
-        local->first_up_child = afr_first_up_child (local->child_up,
-                                                    priv->child_count);
+	local->pre_op_compat = priv->pre_op_compat;
+
+        local->transaction.eager_lock =
+                GF_CALLOC (sizeof (*local->transaction.eager_lock),
+                           priv->child_count,
+                           gf_afr_mt_int32_t);
 
-        local->child_errno = GF_CALLOC (sizeof (*local->child_errno),
-                                        priv->child_count,
-                                        gf_afr_mt_int32_t);
-        if (!local->child_errno)
+        if (!local->transaction.eager_lock)
                 goto out;
 
-        local->pending = GF_CALLOC (sizeof (*local->pending),
-                                    priv->child_count,
-                                    gf_afr_mt_int32_t);
+        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;
 
-        if (!local->pending)
+        local->transaction.fop_subvols = GF_CALLOC (sizeof (*local->transaction.fop_subvols),
+						    priv->child_count,
+						    gf_afr_mt_char);
+        if (!local->transaction.fop_subvols)
                 goto out;
 
-        local->fresh_children = afr_fresh_children_create (priv->child_count);
-        if (!local->fresh_children)
+        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;
 
-        for (i = 0; i < priv->child_count; i++) {
-                local->pending[i] = GF_CALLOC (sizeof (*local->pending[i]),
-                                               3, /* data + metadata + entry */
-                                               gf_afr_mt_int32_t);
-                if (!local->pending[i])
-                        goto out;
-        }
+        local->pending = afr_matrix_create (priv->child_count,
+                                            AFR_NUM_CHANGE_LOGS);
+        if (!local->pending)
+                goto out;
 
-        local->transaction.child_errno =
-                GF_CALLOC (sizeof (*local->transaction.child_errno),
-                           priv->child_count,
-                           gf_afr_mt_int32_t);
-        local->transaction.erase_pending = 1;
+	INIT_LIST_HEAD (&local->transaction.eager_locked);
 
         ret = 0;
 out:
         return ret;
 }
 
+
 void
-afr_reset_children (int32_t *fresh_children, int32_t child_count)
+afr_set_low_priority (call_frame_t *frame)
 {
-        unsigned int i = 0;
-        for (i = 0; i < child_count; i++)
-                fresh_children[i] = -1;
+        frame->root->pid = LOW_PRIO_PROC_PID;
 }
 
-int32_t*
-afr_fresh_children_create (int32_t child_count)
+
+gf_boolean_t
+afr_have_quorum (char *logname, afr_private_t *priv)
 {
-        int32_t           *fresh_children = NULL;
-        int               i               = 0;
+        unsigned int        quorum = 0;
 
-        GF_ASSERT (child_count > 0);
+        GF_VALIDATE_OR_GOTO(logname,priv,out);
+
+        quorum = priv->quorum_count;
+        if (quorum != AFR_QUORUM_AUTO) {
+                return (priv->up_count >= (priv->down_count + quorum));
+        }
+
+        quorum = priv->child_count / 2 + 1;
+        if (priv->up_count >= (priv->down_count + quorum)) {
+                return _gf_true;
+        }
+
+        /*
+         * 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;
+                        }
+                }
+        }
 
-        fresh_children = GF_CALLOC (child_count, sizeof (*fresh_children),
-                                    gf_afr_mt_int32_t);
-        if (NULL == fresh_children)
-                goto out;
-        for (i = 0; i < child_count; i++)
-                fresh_children[i] = -1;
 out:
-        return fresh_children;
+        return _gf_false;
 }
 
 void
-afr_fresh_children_add_child (int32_t *fresh_children, int32_t child,
-                              int32_t child_count)
+afr_priv_destroy (afr_private_t *priv)
 {
-        gf_boolean_t child_found = _gf_false;
-        int          i               = 0;
+        int            i           = 0;
 
-        for (i = 0; i < child_count; i++) {
-                if (fresh_children[i] == -1)
-                        break;
-                if (fresh_children[i] == child) {
-                        child_found = _gf_true;
-                        break;
-                }
-        }
-        if (!child_found) {
-                GF_ASSERT (i < child_count);
-                fresh_children[i] = child;
-        }
+        if (!priv)
+                goto out;
+        inode_unref (priv->root_inode);
+        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);
+
+        GF_FREE (priv);
+out:
+        return;
 }
 
 int
-afr_get_children_count (int32_t *fresh_children, unsigned int child_count)
+xlator_subvolume_count (xlator_t *this)
 {
-        int count = 0;
         int i = 0;
+        xlator_list_t *list = NULL;
 
-        for (i = 0; i < child_count; i++) {
-                if (fresh_children[i] == -1)
-                        break;
-                count++;
-        }
-        return count;
+        for (list = this->children; list; list = list->next)
+                i++;
+        return i;
+}
+
+
+void
+afr_handle_open_fd_count (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        afr_fd_ctx_t    *fd_ctx   = NULL;
+
+        local = frame->local;
+
+        if (!local->fd)
+		return;
+
+	fd_ctx = afr_fd_ctx_get (local->fd, this);
+	if (!fd_ctx)
+		return;
+
+	fd_ctx->open_fd_count = local->open_fd_count;
 }
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index 64b14f88f..fa1da3958 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -46,709 +37,384 @@
 #include "checksum.h"
 
 #include "afr.h"
-#include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-
-int
-afr_examine_dir_sh_unwind (call_frame_t *frame, xlator_t *this, int32_t op_ret,
-                           int32_t op_errno)
-{
-        afr_local_t *local  = NULL;
-
-        local = frame->local;
-
-        afr_set_opendir_done (this, local->fd->inode);
-
-        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                          local->op_errno, local->fd);
-
-        return 0;
-}
-
-
-gf_boolean_t
-__checksums_differ (uint32_t *checksum, int child_count,
-                    unsigned char *child_up)
-{
-        int          ret            = _gf_false;
-        int          i              = 0;
-        uint32_t     cksum          = 0;
-        gf_boolean_t activate_check = _gf_false;
-
-        for (i = 0; i < child_count; i++) {
-                if (!child_up[i])
-                        continue;
-                if (_gf_false == activate_check) {
-                        cksum          = checksum[i];
-                        activate_check = _gf_true;
-                        continue;
-                }
-
-                if (cksum != checksum[i]) {
-                        ret = _gf_true;
-                        break;
-                }
-
-                cksum = checksum[i];
-        }
-
-        return ret;
-}
-
-
-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)
-{
-        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;
-        int               child_index = 0;
-        uint32_t          entry_cksum = 0;
-        int               call_count  = 0;
-        off_t             last_offset = 0;
-        char              sh_type_str[256] = {0,};
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        child_index = (long) cookie;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "%s: failed to do opendir on %s",
-                        local->loc.path, priv->children[child_index]->name);
-                local->op_ret = -1;
-                local->op_ret = op_errno;
-                goto out;
-        }
-
-        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;
-        }
-
-        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-                entry_cksum = gf_rsync_weak_checksum (entry->d_name,
-                                                      strlen (entry->d_name));
-                local->cont.opendir.checksum[child_index] ^= entry_cksum;
-        }
-
-        list_for_each_entry (entry, &entries->list, list) {
-                last_offset = entry->d_off;
-        }
-
-        /* read more entries */
-
-        STACK_WIND_COOKIE (frame, afr_examine_dir_readdir_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->readdir,
-                           local->fd, 131072, last_offset);
-
-        return 0;
-
-out:
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if (__checksums_differ (local->cont.opendir.checksum,
-                                        priv->child_count,
-                                        local->child_up)) {
-
-                        sh->need_entry_self_heal  = _gf_true;
-                        sh->forced_merge          = _gf_true;
-                        sh->type                  = local->fd->inode->ia_type;
-                        sh->background            = _gf_false;
-                        sh->unwind                = afr_examine_dir_sh_unwind;
-
-                        afr_self_heal_type_str_get(&local->self_heal,
-                                                   sh_type_str,
-                                                   sizeof(sh_type_str));
-                        gf_log (this->name, GF_LOG_INFO,
-                                "%s self-heal triggered. path: %s, "
-                                "reason: checksums of directory differ,"
-                                " forced merge option set",
-                                sh_type_str, local->loc.path);
-
-                        afr_self_heal (frame, this, local->fd->inode);
-                } else {
-                        afr_set_opendir_done (this, local->fd->inode);
-
-                        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                                          local->op_errno, local->fd);
-                }
-        }
-
-        return 0;
-}
-
-
-int
-afr_examine_dir (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t * priv       = NULL;
-        afr_local_t *   local      = NULL;
-        int             i          = 0;
-        int             call_count = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        local->cont.opendir.checksum = GF_CALLOC (priv->child_count,
-                                                  sizeof (*local->cont.opendir.checksum),
-                                                  gf_afr_mt_int32_t);
-
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-        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);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
-}
+#include "afr-transaction.h"
 
 
 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)
+                 fd_t *fd, dict_t *xdata)
 {
-        afr_private_t *priv              = NULL;
         afr_local_t   *local             = NULL;
-        int32_t        up_children_count = 0;
-        int            ret               = -1;
         int            call_count        = -1;
+        int32_t        child_index       = 0;
+	afr_fd_ctx_t  *fd_ctx = NULL;
 
-        priv  = this->private;
         local = frame->local;
-
-        up_children_count = afr_up_children_count (priv->child_count,
-                                                   local->child_up);
+	fd_ctx = local->fd_ctx;
+        child_index = (long) cookie;
 
         LOCK (&frame->lock);
         {
-                if (op_ret >= 0)
+                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;
-
-                local->op_errno = op_errno;
+			fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+			if (!local->xdata_rsp && xdata)
+				local->xdata_rsp = dict_ref (xdata);
+                }
         }
         UNLOCK (&frame->lock);
 
         call_count = afr_frame_return (frame);
 
-        if (call_count == 0) {
-                if (local->op_ret != 0)
-                        goto out;
-
-                ret = afr_fd_ctx_set (this, local->fd);
-                if (ret) {
-                        local->op_ret = -1;
-                        local->op_errno = -1;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "failed to set fd ctx for fd %p",
-                                local->fd);
-                        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 anamolies).
-                         */
-
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "reading contents of directory %s looking for mismatch",
-                                local->loc.path);
-
-                        afr_examine_dir (frame, this);
-
-                } else {
-                        /* do the unwind */
-                        goto out;
-                }
-        }
-
-        return 0;
-
-out:
-        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                          local->op_errno, local->fd);
-
+        if (call_count == 0)
+		AFR_STACK_UNWIND (opendir, frame, local->op_ret,
+				  local->op_errno, local->fd, NULL);
         return 0;
 }
 
 
-int32_t
-afr_opendir (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, fd_t *fd)
+int
+afr_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
 {
         afr_private_t * priv        = NULL;
         afr_local_t   * local       = NULL;
-        int             child_count = 0;
         int             i           = 0;
-        int             ret         = -1;
         int             call_count  = -1;
-        int32_t         op_ret      = -1;
-        int32_t         op_errno    = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        int32_t         op_errno    = ENOMEM;
+	afr_fd_ctx_t *fd_ctx = NULL;
 
         priv = this->private;
 
-        child_count = priv->child_count;
+        local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+	fd_ctx = afr_fd_ctx_get (fd, this);
+	if (!fd_ctx)
+		goto out;
 
         loc_copy (&local->loc, loc);
 
-        frame->local = local;
         local->fd    = fd_ref (fd);
+	local->fd_ctx = fd_ctx;
 
         call_count = local->call_count;
 
-        for (i = 0; i < child_count; i++) {
+        for (i = 0; i < priv->child_count; i++) {
                 if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_opendir_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->opendir,
-                                    loc, fd);
+                        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;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (opendir, frame, op_ret, op_errno, fd);
-        }
-
+	AFR_STACK_UNWIND (opendir, frame, -1, op_errno, fd, NULL);
         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
- */
+#define BACKEND_D_OFF_BITS 63
+#define PRESENT_D_OFF_BITS 63
 
+#define ONE 1ULL
+#define MASK (~0ULL)
+#define PRESENT_MASK (MASK >> (64 - PRESENT_D_OFF_BITS))
+#define BACKEND_MASK (MASK >> (64 - BACKEND_D_OFF_BITS))
 
-struct entry_name {
-        char *name;
-        struct list_head list;
-};
+#define TOP_BIT (ONE << (PRESENT_D_OFF_BITS - 1))
+#define SHIFT_BITS (max (0, (BACKEND_D_OFF_BITS - PRESENT_D_OFF_BITS + 1)))
 
-
-static gf_boolean_t
-remembered_name (const char *name, struct list_head *entries)
+static uint64_t
+afr_bits_for (uint64_t num)
 {
-        struct entry_name *e   = NULL;
-        gf_boolean_t       ret = _gf_false;
+	uint64_t bits = 0, ctrl = 1;
 
-        list_for_each_entry (e, entries, list) {
-                if (!strcmp (name, e->name)) {
-                        ret = _gf_true;
-                        goto out;
-                }
-        }
+	while (ctrl < num) {
+		ctrl *= 2;
+		bits ++;
+	}
 
-out:
-        return ret;
+	return bits;
 }
 
-
-static void
-afr_remember_entries (gf_dirent_t *entries, fd_t *fd)
+int
+afr_itransform (xlator_t *this, int subvol, uint64_t x, uint64_t *y_p)
 {
-        struct entry_name *n      = NULL;
-        gf_dirent_t       *entry  = 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;
+        afr_private_t *conf = NULL;
+        int         cnt = 0;
+        int         max = 0;
+        uint64_t    y = 0;
+        uint64_t    hi_mask = 0;
+        uint64_t    off_mask = 0;
+        int         max_bits = 0;
+
+        if (x == ((uint64_t) -1)) {
+                y = (uint64_t) -1;
+                goto out;
         }
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-        list_for_each_entry (entry, &entries->list, list) {
-                n = GF_CALLOC (1, sizeof (*n), gf_afr_mt_entry_name);
-                n->name = gf_strdup (entry->d_name);
-                INIT_LIST_HEAD (&n->list);
+        conf = this->private;
+        if (!conf)
+                goto out;
 
-                list_add (&n->list, &fd_ctx->entries);
-        }
-}
+        max = conf->child_count;
+        cnt = subvol;
 
+	if (max == 1) {
+		y = x;
+		goto out;
+	}
 
-static off_t
-afr_filter_entries (gf_dirent_t *entries, fd_t *fd)
-{
-        gf_dirent_t  *entry  = NULL;
-        gf_dirent_t  *tmp    = NULL;
-        int           ret    = 0;
-        uint64_t      ctx    = 0;
-        afr_fd_ctx_t *fd_ctx = NULL;
-        off_t         offset = 0;
-
-        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 -1;
-        }
+        max_bits = afr_bits_for (max);
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+        hi_mask = ~(PRESENT_MASK >> (max_bits + 1));
 
-        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-                offset = entry->d_off;
-
-                if (remembered_name (entry->d_name, &fd_ctx->entries)) {
-                        list_del (&entry->list);
-                        GF_FREE (entry);
-                }
+        if (x & hi_mask) {
+                /* HUGE d_off */
+                off_mask = MASK << max_bits;
+                y = TOP_BIT | ((x >> SHIFT_BITS) & off_mask) | cnt;
+        } else {
+                /* small d_off */
+                y = ((x * max) + cnt);
         }
 
-        return offset;
+out:
+        if (y_p)
+                *y_p = y;
+
+        return 0;
 }
 
 
-static void
-afr_forget_entries (fd_t *fd)
+int
+afr_deitransform (xlator_t *this, uint64_t y, int *subvol_p,
+                  uint64_t *x_p)
 {
-        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);
-        }
-}
+        afr_private_t *conf = NULL;
+        int         cnt = 0;
+        int         max = 0;
+        uint64_t    x = 0;
+        int         subvol = 0;
+        int         max_bits = 0;
+        uint64_t    off_mask = 0;
+        uint64_t    host_mask = 0;
+
+        if (!this->private)
+                return -1;
 
+        conf = this->private;
+        max = conf->child_count;
 
-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)
-{
-        afr_private_t * priv        = NULL;
-        afr_local_t *   local       = NULL;
-        gf_dirent_t *   entry       = NULL;
-        gf_dirent_t *   tmp         = NULL;
-        int             child_index = -1;
+	if (max == 1) {
+		x = y;
+		cnt = 0;
+		goto out;
+	}
 
-        priv     = this->private;
-        local = frame->local;
-        child_index = (long) cookie;
+        if (y & TOP_BIT) {
+                /* HUGE d_off */
+                max_bits = afr_bits_for (max);
+                off_mask = (MASK << max_bits);
+                host_mask = ~(off_mask);
 
-        if (op_ret == -1)
-                goto out;
+                x = ((y & ~TOP_BIT) & off_mask) << SHIFT_BITS;
 
-        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-                if ((local->fd->inode == local->fd->inode->table->root)
-                    && !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) {
-                        list_del_init (&entry->list);
-                        GF_FREE (entry);
-                }
+                cnt = y & host_mask;
+	} else {
+                /* small d_off */
+                cnt = y % max;
+                x = y / max;
         }
 
 out:
-        AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries);
+        subvol = cnt;
+
+        if (subvol_p)
+                *subvol_p = subvol;
+
+        if (x_p)
+                *x_p = x;
 
         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)
+static void
+afr_readdir_transform_entries (gf_dirent_t *subvol_entries, int subvol,
+			       gf_dirent_t *entries, fd_t *fd)
 {
-        afr_private_t *  priv            = NULL;
-        afr_local_t *    local           = NULL;
-        xlator_t **      children        = NULL;
-        int32_t          next_call_child = -1;
-        int              ret             = 0;
-        gf_dirent_t *    entry           = NULL;
-        gf_dirent_t *    tmp             = NULL;
-        int32_t          *last_index     = NULL;
-        int32_t          read_child      = -1;
-        int32_t         *fresh_children   = NULL;
-        uint64_t         ctx             = 0;
-        afr_fd_ctx_t    *fd_ctx          = NULL;
-        off_t            offset          = 0;
-        int32_t         call_child       = -1;
-
-        priv     = this->private;
-        children = priv->children;
-
-        local = frame->local;
-
-        read_child = (long) cookie;
-        last_index = &local->cont.readdir.last_index;
-        fresh_children = local->fresh_children;
-
-        /* the value of the last_index changes if afr_next_call_child is
-         * called. So to find the call_child of this callback use last_index
-         * before the next_call_child call.
-         */
-        if (*last_index == -1)
-                call_child = read_child;
-        else
-                call_child = fresh_children[*last_index];
-
-        if (priv->strict_readdir) {
-                ret = fd_ctx_get (local->fd, this, &ctx);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "could not get fd ctx for fd=%p", local->fd);
-                        op_ret   = -1;
-                        op_errno = -ret;
-                        goto out;
+	afr_private_t *priv = NULL;
+        gf_dirent_t *entry = NULL;
+        gf_dirent_t *tmp = NULL;
+	unsigned char *data_readable = NULL;
+	unsigned char *metadata_readable = NULL;
+	int gen = 0;
+
+	priv = THIS->private;
+
+	data_readable = alloca0 (priv->child_count);
+	metadata_readable = alloca0 (priv->child_count);
+
+        list_for_each_entry_safe (entry, tmp, &subvol_entries->list, list) {
+                if (__is_root_gfid (fd->inode->gfid) &&
+                    !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) {
+			continue;
                 }
 
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+		list_del_init (&entry->list);
+		afr_itransform (THIS, subvol, entry->d_off, &entry->d_off);
+		list_add_tail (&entry->list, &entries->list);
 
-                if (op_ret == -1) {
-                        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;
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "starting readdir afresh on child %d, offset %"PRId64,
-                                next_call_child, (uint64_t) 0);
-
-                        fd_ctx->failed_over = _gf_true;
-
-                        STACK_WIND_COOKIE (frame, afr_readdirp_cbk,
-                                           (void *) (long) read_child,
-                                           children[next_call_child],
-                                           children[next_call_child]->fops->readdirp,
-                                           local->fd,
-                                           local->cont.readdir.size, 0);
-                        return 0;
-                }
-        }
+		if (entry->inode) {
+			gen = 0;
+			afr_inode_read_subvol_get (entry->inode, THIS,
+						   data_readable,
+						   metadata_readable, &gen);
 
-        if (op_ret != -1) {
-                list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-                        if ((local->fd->inode == local->fd->inode->table->root)
-                            && !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) {
-                                list_del_init (&entry->list);
-                                GF_FREE (entry);
-                        }
-                }
-        }
+			if (gen != priv->event_generation ||
+				!data_readable[subvol] ||
+				!metadata_readable[subvol]) {
 
-        if (priv->strict_readdir) {
-                if (fd_ctx->failed_over) {
-                        if (list_empty (&entries->list)) {
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "no entries found");
-                                goto out;
-                        }
-
-                        offset = afr_filter_entries (entries, local->fd);
-
-                        afr_remember_entries (entries, local->fd);
-
-                        if (list_empty (&entries->list)) {
-                                /* All the entries we got were duplicate. We
-                                   shouldn't send an empty list now, because
-                                   that'll make the application stop reading. So
-                                   try to get more entries */
-
-                                gf_log (this->name, GF_LOG_TRACE,
-                                        "trying to fetch non-duplicate entries "
-                                        "from offset %"PRId64", child %s",
-                                        offset, children[call_child]->name);
-
-                                STACK_WIND_COOKIE (frame, afr_readdirp_cbk,
-                                                   (void *) (long) read_child,
-                                                   children[call_child],
-                                                   children[call_child]->fops->readdirp,
-                                                   local->fd, local->cont.readdir.size, offset);
-                                return 0;
-                        }
-                } else {
-                        afr_remember_entries (entries, local->fd);
-                }
+				inode_unref (entry->inode);
+				entry->inode = NULL;
+			}
+		}
         }
-
-out:
-        AFR_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries);
-
-        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)
+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;
-        xlator_t **      children   = NULL;
-        int              call_child = 0;
-        afr_local_t     *local      = NULL;
-        uint64_t         ctx        = 0;
-        afr_fd_ctx_t    *fd_ctx     = NULL;
-        int              ret        = -1;
-        int32_t          op_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;
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        frame->local = local;
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+        afr_local_t *local = NULL;
+	gf_dirent_t  entries;
 
-        local->fresh_children = afr_fresh_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+	INIT_LIST_HEAD (&entries.list);
 
-        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.readdir.last_index);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                op_ret = -1;
-                goto out;
-        }
+        local = frame->local;
 
-        local->fd                  = fd_ref (fd);
-        local->cont.readdir.size   = size;
+        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;
 
-        if (priv->strict_readdir) {
-                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);
-                        op_errno = -ret;
-                        goto out;
-                }
+		afr_read_txn_continue (frame, this, (long) cookie);
+		return 0;
+	}
 
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+	if (op_ret >= 0)
+		afr_readdir_transform_entries (subvol_entries, (long) cookie,
+					       &entries, local->fd);
 
-                if (fd_ctx->last_tried != call_child) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "first up child has changed from %d to %d, "
-                                "restarting readdir from offset 0",
-                                fd_ctx->last_tried, call_child);
+        AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, &entries, xdata);
 
-                        fd_ctx->failed_over = _gf_true;
-                        offset = 0;
-                }
+        return 0;
+}
 
-                fd_ctx->last_tried = call_child;
-        }
 
-        if (whichop == GF_FOP_READDIR)
+int
+afr_readdir_wind (call_frame_t *frame, xlator_t *this, int subvol)
+{
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+
+	priv = this->private;
+	local = frame->local;
+
+	if (subvol == -1) {
+		AFR_STACK_UNWIND (readdir, frame, local->op_ret,
+				  local->op_errno, 0, 0);
+		return 0;
+	}
+
+        if (local->op == 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);
+                                   (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_readdirp_cbk,
-                                   (void *) (long) call_child,
-                                   children[call_child],
-                                   children[call_child]->fops->readdirp, fd,
-                                   size, offset);
+                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;
+}
 
-        op_ret = 0;
+
+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_local_t   *local     = NULL;
+        int32_t       op_errno   = 0;
+	int           subvol = -1;
+
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		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;
+
+	if (offset == 0) {
+		/* 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_deitransform (this, offset, &subvol,
+				  (uint64_t *)&local->cont.readdir.offset);
+		afr_readdir_wind (frame, this, subvol);
+	}
+
+        return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, NULL);
-        }
+        AFR_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
         return 0;
 }
 
 
 int32_t
 afr_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t offset)
+             off_t offset, dict_t *xdata)
 {
-        afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIR);
+	afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIR, xdata);
+
         return 0;
 }
 
 
 int32_t
 afr_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-              off_t offset)
+              off_t offset, dict_t *dict)
 {
-        afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIRP);
+        afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIRP, dict);
+
         return 0;
 }
 
@@ -756,7 +422,6 @@ afr_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
 int32_t
 afr_releasedir (xlator_t *this, fd_t *fd)
 {
-        afr_forget_entries (fd);
         afr_cleanup_fd_ctx (this, fd);
 
         return 0;
diff --git a/xlators/cluster/afr/src/afr-dir-read.h b/xlators/cluster/afr/src/afr-dir-read.h
index 3143cb973..09456d159 100644
--- a/xlators/cluster/afr/src/afr-dir-read.h
+++ b/xlators/cluster/afr/src/afr-dir-read.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  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 __DIR_READ_H__
@@ -23,23 +14,23 @@
 
 int32_t
 afr_opendir (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, fd_t *fd);
+	     loc_t *loc, fd_t *fd, dict_t *xdata);
 
 int32_t
 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);
+	     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);
+              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);
+	      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 58e979791..465dde54f 100644
--- a/xlators/cluster/afr/src/afr-dir-write.c
+++ b/xlators/cluster/afr/src/afr-dir-write.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -43,72 +34,218 @@
 #include "common-utils.h"
 #include "compat-errno.h"
 #include "compat.h"
+#include "byte-order.h"
 
 #include "afr.h"
 #include "afr-transaction.h"
 
-
 void
-afr_build_parent_loc (loc_t *parent, loc_t *child)
+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)
 {
-        char *tmp = NULL;
+        int     ret = -1;
+        char    *child_path = NULL;
 
         if (!child->parent) {
-                //this should never be called with root as the child
-                GF_ASSERT (0);
-                loc_copy (parent, child);
-                return;
+                if (op_errno)
+                        *op_errno = EINVAL;
+                goto out;
         }
 
-        tmp = gf_strdup (child->path);
-        parent->path   = gf_strdup (dirname (tmp));
-        GF_FREE (tmp);
+        child_path = gf_strdup (child->path);
+        if (!child_path) {
+                if (op_errno)
+                        *op_errno = ENOMEM;
+                goto out;
+        }
 
-        parent->name   = strrchr (parent->path, '/');
-        if (parent->name)
-                parent->name++;
+        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);
 
-        parent->inode  = inode_ref (child->parent);
-        parent->parent = inode_parent (parent->inode, 0, NULL);
-        parent->ino    = parent->inode->ino;
+        ret = 0;
+out:
+        GF_FREE (child_path);
 
-        if (!uuid_is_null (child->pargfid))
-                uuid_copy (parent->gfid, child->pargfid);
+        return ret;
 }
 
-/* {{{ create */
 
-int
-afr_create_unwind (call_frame_t *frame, xlator_t *this)
+static void
+__afr_dir_write_finalize (call_frame_t *frame, xlator_t *this)
 {
-        call_frame_t *main_frame = NULL;
-        afr_local_t  *local = NULL;
-        struct iatt  *unwind_buf = NULL;
+	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;
+
+	local = frame->local;
+	priv = this->private;
+
+	if (local->inode) {
+		afr_replies_interpret (frame, this, local->inode);
+		inode_read_subvol = afr_data_subvol_get (local->inode, this,
+							 NULL, NULL);
+	}
+	if (local->parent)
+		parent_read_subvol = afr_data_subvol_get (local->parent, this,
+							  NULL, NULL);
+	if (local->parent2)
+		parent2_read_subvol = afr_data_subvol_get (local->parent2, this,
+							   NULL, NULL);
+
+	local->op_ret = -1;
+	local->op_errno = afr_final_errno (local, priv);
+
+	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_read_subvol_reset (local->inode,
+							     this);
+			if (local->parent)
+				afr_inode_read_subvol_reset (local->parent,
+							     this);
+			if (local->parent2)
+				afr_inode_read_subvol_reset (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->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;
+		}
+	}
+}
+
+
+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)
+{
+        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 (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 (xdata)
+			local->replies[child_index].xdata = dict_ref (xdata);
+
+		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;
+}
 
-        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) {
-                if (local->cont.create.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.create.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.create.buf;
-                }
-
-                AFR_STACK_UNWIND (create, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  local->cont.create.fd,
-                                  local->cont.create.inode,
-                                  unwind_buf, &local->cont.create.preparent,
-                                  &local->cont.create.postparent);
+
+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)
+{
+        afr_local_t *local = NULL;
+        int child_index = (long) cookie;
+        int call_count = -1;
+
+        local = frame->local;
+
+	LOCK (&frame->lock);
+	{
+		__afr_dir_write_fill (frame, this, child_index, op_ret,
+				      op_errno, buf, preparent, postparent,
+				      preparent2, postparent2, xdata);
+	}
+	UNLOCK (&frame->lock);
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+		__afr_dir_write_finalize (frame, this);
+
+		if (afr_txn_nothing_failed (frame, this))
+			local->transaction.unwind (frame, this);
+
+		afr_mark_entry_pending_changelog (frame, this);
+
+                local->transaction.resume (frame, this);
         }
 
         return 0;
@@ -116,223 +253,269 @@ afr_create_unwind (call_frame_t *frame, xlator_t *this)
 
 
 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)
+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)
 {
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv  = NULL;
-        uint64_t        ctx = 0;
-        afr_fd_ctx_t    *fd_ctx = NULL;
-        int             ret = 0;
-        int             call_count = -1;
-        int             child_index = -1;
-        int32_t         *fresh_children = NULL;
+        int call_count = 0;
 
-        local = frame->local;
-        priv  = this->private;
+        call_count = afr_frame_return (frame);
 
-        child_index = (long) cookie;
+        if (call_count == 0)
+                AFR_STACK_DESTROY (frame);
 
-        LOCK (&frame->lock);
-        {
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
+        return 0;
+}
 
-                if (op_ret != -1) {
-                        local->op_ret = op_ret;
 
-                        ret = afr_fd_ctx_set (this, fd);
+void
+afr_mark_new_entry_changelog (call_frame_t *frame, xlator_t *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           idx         = 0;
+        int           op_errno    = ENOMEM;
+	unsigned char *pending    = NULL;
+	int           call_count   = 0;
 
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "could not set ctx on fd=%p", fd);
+        local = frame->local;
+        priv = this->private;
 
-                                local->op_ret   = -1;
-                                local->op_errno = -ret;
-                                goto unlock;
-                        }
+        new_frame = copy_frame (frame);
+        if (!new_frame)
+                goto out;
 
-                        ret = fd_ctx_get (fd, this, &ctx);
+	new_local = AFR_FRAME_INIT (new_frame, op_errno);
+	if (!new_local)
+		goto out;
 
-                        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;
-                        }
+        changelog = afr_matrix_create (priv->child_count, AFR_NUM_CHANGE_LOGS);
+        if (!changelog)
+                goto out;
 
-                        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+        xattr = dict_new ();
+	if (!xattr)
+		goto out;
 
-                        fd_ctx->opened_on[child_index] = 1;
-                        fd_ctx->flags                  = local->cont.create.flags;
+	idx = afr_index_for_transaction_type (AFR_DATA_TRANSACTION);
 
-                        if (local->success_count == 0)
-                                local->cont.create.buf        = *buf;
+	pending = alloca0 (priv->child_count);
 
-                        if (child_index == local->read_child_index) {
-                                local->cont.create.read_child_buf = *buf;
-                                local->cont.create.preparent      = *preparent;
-                                local->cont.create.postparent     = *postparent;
-                        }
+	for (i = 0; i < priv->child_count; i++) {
+		if (local->transaction.pre_op[i] &&
+		    !local->transaction.failed_subvols[i]) {
+			call_count ++;
+			continue;
+		}
 
-                        local->cont.create.inode = inode;
+		changelog[i][idx] = hton32(1);
+		pending[i] = 1;
+	}
 
-                        fresh_children = local->fresh_children;
-                        fresh_children[local->success_count] = child_index;
-                        local->success_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->inode);
 
-                local->op_errno = op_errno;
-        }
 
-unlock:
-        UNLOCK (&frame->lock);
+	afr_set_pending_dict (priv, xattr, changelog);
 
-        call_count = afr_frame_return (frame);
+        new_local->call_count = call_count;
 
-        if (call_count == 0) {
-                afr_set_read_ctx_from_policy (this, inode,
-                                              local->fresh_children,
-                                              local->read_child_index,
-                                              priv->read_child);
-                local->transaction.unwind (frame, this);
+        for (i = 0; i < priv->child_count; i++) {
+		if (pending[i])
+                        continue;
 
-                local->transaction.resume (frame, this);
+                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;
         }
 
-        return 0;
+        new_frame = NULL;
+out:
+        if (new_frame)
+                AFR_STACK_DESTROY (new_frame);
+	if (xattr)
+		dict_unref (xattr);
+        return;
 }
 
 
-int
-afr_create_wind (call_frame_t *frame, xlator_t *this)
+void
+afr_mark_entry_pending_changelog (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 pre_op_count = 0;
+	int failed_count = 0;
 
         local = frame->local;
-        priv = this->private;
+        priv  = this->private;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+        if (local->op_ret < 0)
+		return;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+	if (local->op != GF_FOP_CREATE && local->op != GF_FOP_MKNOD)
+		return;
 
-        local->call_count = call_count;
+	pre_op_count = AFR_COUNT (local->transaction.pre_op, priv->child_count);
+	failed_count = AFR_COUNT (local->transaction.failed_subvols,
+				  priv->child_count);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[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->cont.create.fd,
-                                           local->cont.create.params);
-                        if (!--call_count)
-                                break;
-                }
-        }
+	if (pre_op_count == priv->child_count && !failed_count)
+		return;
 
-        return 0;
+        afr_mark_new_entry_changelog (frame, this);
+
+        return;
 }
 
 
+/* {{{ create */
+
 int
-afr_create_done (call_frame_t *frame, xlator_t *this)
+afr_create_unwind (call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t * local = NULL;
+        call_frame_t *main_frame = NULL;
+        afr_local_t  *local = NULL;
 
         local = frame->local;
 
-        local->transaction.unwind (frame, this);
+	main_frame = afr_transaction_detach_fop_frame (frame);
 
-        AFR_STACK_DESTROY (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 (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, int32_t flags, mode_t mode,
-            fd_t *fd, dict_t *params)
+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_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t  * transaction_frame = NULL;
-        int ret = -1;
-        int op_ret   = -1;
-        int op_errno = 0;
+        return __afr_dir_write_cbk (frame, cookie, this, op_ret, op_errno, buf,
+				    preparent, postparent, NULL, NULL, xdata);
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
 
+int
+afr_create_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;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
+	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;
+}
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
+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_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                = ENOMEM;
+
+        priv = this->private;
+
+        QUORUM_CHECK(create,out);
+
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame)
                 goto out;
-        }
 
-        transaction_frame->local = local;
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		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->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->cont.create.mode  = mode;
         local->cont.create.fd    = fd_ref (fd);
-        if (params)
-                local->cont.create.params = dict_ref (params);
+        local->umask  = umask;
+
+        if (xdata)
+                local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        if (loc->parent)
-                local->cont.create.parent_ino = loc->parent->ino;
+	if (!local->xdata_req)
+		goto out;
 
-        local->transaction.fop    = afr_create_wind;
-        local->transaction.done   = afr_create_done;
+        local->transaction.wind   = afr_create_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
         local->transaction.unwind = afr_create_unwind;
 
-        afr_build_parent_loc (&local->transaction.parent_loc, loc);
+        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;
 
-        afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (create, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL);
-        }
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
 
+	AFR_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+			  NULL, NULL);
         return 0;
 }
 
@@ -345,33 +528,17 @@ afr_mknod_unwind (call_frame_t *frame, xlator_t *this)
 {
         call_frame_t *main_frame = NULL;
         afr_local_t  *local = NULL;
-        struct iatt *unwind_buf = 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) {
-                if (local->cont.mknod.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.mknod.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.mknod.buf;
-                }
-
-                AFR_STACK_UNWIND (mknod, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  local->cont.mknod.inode,
-                                  unwind_buf, &local->cont.mknod.preparent,
-                                  &local->cont.mknod.postparent);
-        }
+	main_frame = afr_transaction_detach_fop_frame (frame);
+        if (!main_frame)
+		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;
 }
 
@@ -380,183 +547,107 @@ 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)
+                    struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t     *local          = NULL;
-        afr_private_t   *priv           = NULL;
-        int             call_count      = -1;
-        int             child_index     = -1;
-        int32_t         *fresh_children = NULL;
-
-        local = frame->local;
-        priv = this->private;
-
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                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)
-                                local->cont.mknod.buf   = *buf;
-
-                        if (child_index == local->read_child_index) {
-                                local->cont.mknod.read_child_buf = *buf;
-                                local->cont.mknod.preparent      = *preparent;
-                                local->cont.mknod.postparent     = *postparent;
-                        }
-
-                        local->cont.mknod.inode = inode;
-
-                        fresh_children = local->fresh_children;
-                        fresh_children[local->success_count] = child_index;
-                        local->success_count++;
-                }
-
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                afr_set_read_ctx_from_policy (this, inode,
-                                              local->fresh_children,
-                                              local->read_child_index,
-                                              priv->read_child);
-                local->transaction.unwind (frame, this);
-
-                local->transaction.resume (frame, this);
-        }
-
-        return 0;
+	return __afr_dir_write_cbk (frame, cookie, this, op_ret, op_errno, buf,
+				    preparent, postparent, NULL, NULL, xdata);
 }
 
 
-int32_t
-afr_mknod_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_mknod_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_up_children_count (priv->child_count, local->child_up);
-
-        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_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->cont.mknod.params);
-                        if (!--call_count)
-                                break;
-                }
-        }
-
+	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_done (call_frame_t *frame, xlator_t *this)
+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_local_t * local = NULL;
-
-        local = frame->local;
-
-        local->transaction.unwind (frame, this);
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
-}
-
-
-int
-afr_mknod (call_frame_t *frame, xlator_t *this,
-           loc_t *loc, mode_t mode, dev_t dev, dict_t *params)
-{
-        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;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        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                = ENOMEM;
 
         priv = this->private;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+        QUORUM_CHECK(mknod,out);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame)
                 goto out;
-        }
 
-        transaction_frame->local = local;
+	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);
 
-        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;
-        if (params)
-                local->cont.mknod.params = dict_ref (params);
+        local->umask = umask;
+
+        if (xdata)
+                local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        if (loc->parent)
-                local->cont.mknod.parent_ino = loc->parent->ino;
+	if (!local->xdata_req)
+		goto out;
 
-        local->transaction.fop    = afr_mknod_wind;
-        local->transaction.done   = afr_mknod_done;
+        local->transaction.wind   = afr_mknod_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
         local->transaction.unwind = afr_mknod_unwind;
 
-        afr_build_parent_loc (&local->transaction.parent_loc, loc);
+        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;
 
-        afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
+        }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (mknod, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL);
-        }
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
 
+	AFR_STACK_UNWIND (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+			  NULL);
         return 0;
 }
 
@@ -570,33 +661,17 @@ afr_mkdir_unwind (call_frame_t *frame, xlator_t *this)
 {
         call_frame_t *main_frame = NULL;
         afr_local_t  *local = NULL;
-        struct iatt *unwind_buf = 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) {
-                if (local->cont.mkdir.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.mkdir.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.mkdir.buf;
-                }
-
-                AFR_STACK_UNWIND (mkdir, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  local->cont.mkdir.inode,
-                                  unwind_buf, &local->cont.mkdir.preparent,
-                                  &local->cont.mkdir.postparent);
-        }
+	main_frame = afr_transaction_detach_fop_frame (frame);
+        if (!main_frame)
+		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;
 }
 
@@ -605,184 +680,106 @@ 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)
+                    struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t     *local          = NULL;
-        afr_private_t   *priv           = NULL;
-        int             call_count      = -1;
-        int             child_index     = -1;
-        int32_t         *fresh_children = NULL;
-
-        local = frame->local;
-        priv = this->private;
-
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                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)
-                                local->cont.mkdir.buf   = *buf;
-
-                        if (child_index == local->read_child_index) {
-                                local->cont.mkdir.read_child_buf = *buf;
-                                local->cont.mkdir.preparent      = *preparent;
-                                local->cont.mkdir.postparent     = *postparent;
-                        }
-
-                        local->cont.mkdir.inode = inode;
-
-                        fresh_children = local->fresh_children;
-                        fresh_children[local->success_count] = child_index;
-                        local->success_count++;
-                }
-
-                local->op_errno         = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                afr_set_read_ctx_from_policy (this, inode,
-                                              local->fresh_children,
-                                              local->read_child_index,
-                                              priv->read_child);
-                local->transaction.unwind (frame, this);
-
-                local->transaction.resume (frame, this);
-        }
-
-        return 0;
+	return __afr_dir_write_cbk (frame, cookie, this, op_ret, op_errno, buf,
+				    preparent, postparent, NULL, NULL, xdata);
 }
 
 
 int
-afr_mkdir_wind (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_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
 
         local = frame->local;
         priv  = this->private;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-        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_mkdir_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->mkdir,
-                                           &local->loc, local->cont.mkdir.mode,
-                                           local->cont.mkdir.params);
-                        if (!--call_count)
-                                break;
-                }
-        }
-
+	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_done (call_frame_t *frame, xlator_t *this)
+afr_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+	   mode_t umask, dict_t *xdata)
 {
-        afr_local_t * local = NULL;
-
-        local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
-}
-
-
-int
-afr_mkdir (call_frame_t *frame, xlator_t *this,
-           loc_t *loc, mode_t mode, dict_t *params)
-{
-        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;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        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                = ENOMEM;
 
         priv = this->private;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+        QUORUM_CHECK(mkdir,out);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame)
                 goto out;
-        }
 
-        transaction_frame->local = local;
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		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->inode = inode_ref (loc->inode);
+	local->parent = inode_ref (loc->parent);
 
         local->cont.mkdir.mode  = mode;
-        if (params)
-                local->cont.mkdir.params = dict_ref (params);
+        local->umask = umask;
+
+        if (xdata)
+                local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        if (loc->parent)
-                local->cont.mkdir.parent_ino = loc->parent->ino;
+	if (!local->xdata_req)
+		goto out;
 
-        local->transaction.fop    = afr_mkdir_wind;
-        local->transaction.done   = afr_mkdir_done;
+        local->op = GF_FOP_MKDIR;
+        local->transaction.wind   = afr_mkdir_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
         local->transaction.unwind = afr_mkdir_unwind;
 
-        afr_build_parent_loc (&local->transaction.parent_loc, loc);
+        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;
 
-        afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
+        }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-
-                AFR_STACK_UNWIND (mkdir, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL);
-        }
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
 
+	AFR_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+			  NULL);
         return 0;
 }
 
@@ -796,33 +793,17 @@ afr_link_unwind (call_frame_t *frame, xlator_t *this)
 {
         call_frame_t *main_frame = NULL;
         afr_local_t  *local = NULL;
-        struct iatt *unwind_buf = 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) {
-                if (local->cont.link.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.link.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.link.buf;
-                }
-
-                AFR_STACK_UNWIND (link, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  local->cont.link.inode,
-                                  unwind_buf, &local->cont.link.preparent,
-                                  &local->cont.link.postparent);
-        }
+	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;
 }
 
@@ -831,182 +812,105 @@ 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)
+                   struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t     *local          = NULL;
-        afr_private_t   *priv           = NULL;
-        int             call_count      = -1;
-        int             child_index     = -1;
-        int32_t         *fresh_children = NULL;
-
-        local = frame->local;
-        priv = this->private;
-
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                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) {
-                                local->cont.link.buf        = *buf;
-                        }
-
-                        if (child_index == local->read_child_index) {
-                                local->cont.link.read_child_buf = *buf;
-                                local->cont.link.preparent      = *preparent;
-                                local->cont.link.postparent     = *postparent;
-                        }
-
-                        local->cont.link.inode    = inode;
-
-                        fresh_children = local->fresh_children;
-                        fresh_children[local->success_count] = child_index;
-                        local->success_count++;
-                }
-
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                afr_set_read_ctx_from_policy (this, inode,
-                                              local->fresh_children,
-                                              local->read_child_index,
-                                              priv->read_child);
-                local->transaction.unwind (frame, this);
-
-                local->transaction.resume (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_up_children_count (priv->child_count, local->child_up);
-
-        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_link_wind_cbk, (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->link,
-                                           &local->loc,
-                                           &local->newloc);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
+	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);
-
-        return 0;
-}
-
-
-int
-afr_link (call_frame_t *frame, xlator_t *this,
-          loc_t *oldloc, loc_t *newloc)
-{
-        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;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        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                = ENOMEM;
 
         priv = this->private;
 
+        QUORUM_CHECK(link,out);
+
         transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
+        if (!transaction_frame)
                 goto out;
-        }
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
-
-        transaction_frame->local = local;
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		goto out;
 
         loc_copy (&local->loc,    oldloc);
         loc_copy (&local->newloc, newloc);
 
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr)
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
+	local->inode = inode_ref (oldloc->inode);
+	local->parent = inode_ref (newloc->parent);
 
-        local->cont.link.ino = oldloc->inode->ino;
+        if (xdata)
+                local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        if (oldloc->parent)
-                local->cont.link.parent_ino = newloc->parent->ino;
+	if (!local->xdata_req)
+		goto out;
 
-        local->transaction.fop    = afr_link_wind;
-        local->transaction.done   = afr_link_done;
+        local->op = GF_FOP_LINK;
+
+        local->transaction.wind   = afr_link_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
         local->transaction.unwind = afr_link_unwind;
 
-        afr_build_parent_loc (&local->transaction.parent_loc, oldloc);
+        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 (oldloc->path);
-        local->transaction.new_basename = AFR_BASENAME (newloc->path);
+        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;
 
-        afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
+        }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (link, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL);
-        }
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
 
+	AFR_STACK_UNWIND (link, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+			  NULL);
         return 0;
 }
 
@@ -1020,33 +924,17 @@ afr_symlink_unwind (call_frame_t *frame, xlator_t *this)
 {
         call_frame_t *main_frame = NULL;
         afr_local_t  *local = NULL;
-        struct iatt *unwind_buf = 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) {
-                if (local->cont.symlink.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.symlink.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.symlink.buf;
-                }
-
-                AFR_STACK_UNWIND (symlink, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  local->cont.symlink.inode,
-                                  unwind_buf, &local->cont.symlink.preparent,
-                                  &local->cont.symlink.postparent);
-        }
+	main_frame = afr_transaction_detach_fop_frame (frame);
+        if (!main_frame)
+		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;
 }
 
@@ -1055,184 +943,106 @@ 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)
+                      struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t     *local          = NULL;
-        afr_private_t   *priv           = NULL;
-        int             call_count      = -1;
-        int             child_index     = -1;
-        int32_t         *fresh_children = NULL;
-
-        local = frame->local;
-        priv = this->private;
-
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                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)
-                                local->cont.symlink.buf        = *buf;
-
-                        if (child_index == local->read_child_index) {
-                                local->cont.symlink.read_child_buf = *buf;
-                                local->cont.symlink.preparent      = *preparent;
-                                local->cont.symlink.postparent     = *postparent;
-                        }
-
-                        local->cont.symlink.inode    = inode;
-
-                        fresh_children = local->fresh_children;
-                        fresh_children[local->success_count] = child_index;
-                        local->success_count++;
-                }
-
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                afr_set_read_ctx_from_policy (this, inode,
-                                              local->fresh_children,
-                                              local->read_child_index,
-                                              priv->read_child);
-                local->transaction.unwind (frame, this);
-
-                local->transaction.resume (frame, this);
-        }
-
-        return 0;
+        return __afr_dir_write_cbk (frame, cookie, this, op_ret, op_errno, buf,
+				    preparent, postparent, NULL, NULL, xdata);
 }
 
 
 int
-afr_symlink_wind (call_frame_t *frame, xlator_t *this)
+afr_symlink_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_up_children_count (priv->child_count, local->child_up);
-
-        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_symlink_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->symlink,
-                                           local->cont.symlink.linkpath,
-                                           &local->loc,
-                                           local->cont.symlink.params);
-
-                        if (!--call_count)
-                                break;
-
-                }
-        }
-
-        return 0;
-}
-
-
-int
-afr_symlink_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
-
+	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, 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;
-        call_frame_t  * transaction_frame = NULL;
-        int ret = -1;
-        int op_ret   = -1;
-        int op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        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                = ENOMEM;
 
         priv = this->private;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
+        QUORUM_CHECK(symlink,out);
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame)
                 goto out;
-        }
 
-        transaction_frame->local = local;
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		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->inode = inode_ref (loc->inode);
+	local->parent = inode_ref (loc->parent);
 
         local->cont.symlink.linkpath = gf_strdup (linkpath);
-        if (params)
-                local->cont.symlink.params = dict_ref (params);
+        local->umask = umask;
 
-        if (loc->parent)
-                local->cont.symlink.parent_ino = loc->parent->ino;
+        if (xdata)
+                local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        local->transaction.fop    = afr_symlink_wind;
-        local->transaction.done   = afr_symlink_done;
+	if (!local->xdata_req)
+		goto out;
+
+        local->op = GF_FOP_SYMLINK;
+        local->transaction.wind   = afr_symlink_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
         local->transaction.unwind = afr_symlink_unwind;
 
-        afr_build_parent_loc (&local->transaction.parent_loc, loc);
+        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;
 
-        afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
+        }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (symlink, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL);
-        }
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
 
+	AFR_STACK_UNWIND (symlink, frame, -1, op_errno, NULL, NULL, NULL,
+			  NULL, NULL);
         return 0;
 }
 
@@ -1245,35 +1055,19 @@ afr_rename_unwind (call_frame_t *frame, xlator_t *this)
 {
         call_frame_t *main_frame = NULL;
         afr_local_t  *local = NULL;
-        struct iatt *unwind_buf = 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) {
-                if (local->cont.rename.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.rename.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.rename.buf;
-                }
-
-                AFR_STACK_UNWIND (rename, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  unwind_buf,
-                                  &local->cont.rename.preoldparent,
-                                  &local->cont.rename.postoldparent,
-                                  &local->cont.rename.prenewparent,
-                                  &local->cont.rename.postnewparent);
-        }
+	main_frame = afr_transaction_detach_fop_frame (frame);
+        if (!main_frame)
+		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;
 }
 
@@ -1282,174 +1076,135 @@ 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)
+                     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);
-
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-
-                                if (buf) {
-                                        local->cont.rename.buf = *buf;
-                                }
-
-                                local->success_count++;
-                        }
-
-                        if (child_index == local->read_child_index) {
-                                local->cont.rename.read_child_buf = *buf;
-
-                                local->cont.rename.preoldparent  = *preoldparent;
-                                local->cont.rename.postoldparent = *postoldparent;
-                                local->cont.rename.prenewparent  = *prenewparent;
-                                local->cont.rename.postnewparent = *postnewparent;
-                        }
-                }
-
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                local->transaction.unwind (frame, this);
-                local->transaction.resume (frame, this);
-        }
-
-        return 0;
+        return __afr_dir_write_cbk (frame, cookie, this, op_ret, op_errno, buf,
+				    preoldparent, postoldparent, prenewparent,
+				    postnewparent, xdata);
 }
 
 
-int32_t
-afr_rename_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_rename_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_up_children_count (priv->child_count, local->child_up);
-
-        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_rename_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->rename,
-                                           &local->loc,
-                                           &local->newloc);
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
-}
-
-
-int
-afr_rename_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
+	local = frame->local;
+	priv = this->private;
 
+	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)
+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;
-        call_frame_t  * transaction_frame = NULL;
-        int ret = -1;
-        int op_ret   = -1;
-        int op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        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                = ENOMEM;
+        int                     nlockee                 = 0;
 
         priv = this->private;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+        QUORUM_CHECK(rename,out);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame)
+                op_errno = ENOMEM;
 
-        transaction_frame->local = local;
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		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->inode = inode_ref (oldloc->inode);
+	local->parent = inode_ref (oldloc->parent);
+	local->parent2 = inode_ref (newloc->parent);
 
-        local->cont.rename.ino = oldloc->inode->ino;
+        if (xdata)
+                local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        if (oldloc->parent)
-                local->cont.rename.oldparent_ino = oldloc->parent->ino;
-        if (newloc->parent)
-                local->cont.rename.newparent_ino = newloc->parent->ino;
+	if (!local->xdata_req)
+		goto out;
 
-        local->transaction.fop    = afr_rename_wind;
-        local->transaction.done   = afr_rename_done;
+        local->op = GF_FOP_RENAME;
+        local->transaction.wind   = afr_rename_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
         local->transaction.unwind = afr_rename_unwind;
 
-        afr_build_parent_loc (&local->transaction.parent_loc, oldloc);
-        afr_build_parent_loc (&local->transaction.new_parent_loc, newloc);
+        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;
 
-        afr_transaction (transaction_frame, this, AFR_ENTRY_RENAME_TRANSACTION);
+        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;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
+        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;
 
-                AFR_STACK_UNWIND (rename, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL);
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_RENAME_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
         }
 
+	return 0;
+out:
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
+
+	AFR_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+			  NULL, NULL);
         return 0;
 }
 
@@ -1465,22 +1220,13 @@ afr_unlink_unwind (call_frame_t *frame, xlator_t *this)
 
         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 (unlink, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  &local->cont.unlink.preparent,
-                                  &local->cont.unlink.postparent);
-        }
+	main_frame = afr_transaction_detach_fop_frame (frame);
+        if (!main_frame)
+		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;
 }
 
@@ -1488,163 +1234,103 @@ afr_unlink_unwind (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)
+                     struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int call_count  = -1;
-        int child_index = (long) cookie;
-
-        local = frame->local;
-        priv  = this->private;
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == local->read_child_index) {
-                        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.unlink.preparent  = *preparent;
-                                local->cont.unlink.postparent = *postparent;
-                        }
-
-                        if (child_index == local->read_child_index) {
-                                local->cont.unlink.preparent  = *preparent;
-                                local->cont.unlink.postparent = *postparent;
-                        }
-
-                        local->success_count++;
-                }
-
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                local->transaction.unwind (frame, this);
-
-                local->transaction.resume (frame, this);
-        }
-
-        return 0;
+        return __afr_dir_write_cbk (frame, cookie, this, op_ret, op_errno, NULL,
+				    preparent, postparent, NULL, NULL, xdata);
 }
 
 
-int32_t
-afr_unlink_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_unlink_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_up_children_count (priv->child_count, local->child_up);
-
-        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_unlink_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->unlink,
-                                           &local->loc);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
+	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_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
-afr_unlink (call_frame_t *frame, xlator_t *this,
-            loc_t *loc)
+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;
-        call_frame_t  * transaction_frame = NULL;
-        int ret = -1;
-        int op_ret   = -1;
-        int op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        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                = ENOMEM;
 
         priv = this->private;
 
+        QUORUM_CHECK(unlink,out);
+
         transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
+        if (!transaction_frame)
                 goto out;
-        }
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		goto out;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+        loc_copy (&local->loc, loc);
+        local->xflag = xflag;
 
-        transaction_frame->local = local;
+	local->inode = inode_ref (loc->inode);
+	local->parent = inode_ref (loc->parent);
 
-        loc_copy (&local->loc, loc);
+        if (xdata)
+                local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        if (loc->parent)
-                local->cont.unlink.parent_ino = loc->parent->ino;
+	if (!local->xdata_req)
+		goto out;
 
-        local->transaction.fop    = afr_unlink_wind;
-        local->transaction.done   = afr_unlink_done;
+        local->op = GF_FOP_UNLINK;
+        local->transaction.wind   = afr_unlink_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
         local->transaction.unwind = afr_unlink_unwind;
 
-        afr_build_parent_loc (&local->transaction.parent_loc, loc);
+        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;
 
-        afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
+        }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (unlink, frame, op_ret, op_errno,
-                                  NULL, NULL);
-        }
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
 
+	AFR_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
@@ -1662,22 +1348,13 @@ afr_rmdir_unwind (call_frame_t *frame, xlator_t *this)
 
         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.rmdir.preparent,
-                                  &local->cont.rmdir.postparent);
-        }
+	main_frame = afr_transaction_detach_fop_frame (frame);
+	if (!main_frame)
+		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;
 }
 
@@ -1685,165 +1362,117 @@ afr_rmdir_unwind (call_frame_t *frame, xlator_t *this)
 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)
+                    struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int call_count  = -1;
-        int child_index = (long) cookie;
-        int read_child  = 0;
-
-        local = frame->local;
-        priv = this->private;
-
-        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);
-
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.rmdir.preparent  = *preparent;
-                                local->cont.rmdir.postparent = *postparent;
-
-                        }
-
-                        if (child_index == read_child) {
-                                local->cont.rmdir.preparent  = *preparent;
-                                local->cont.rmdir.postparent = *postparent;
-                        }
-
-                        local->success_count++;
-                }
-
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                local->transaction.unwind (frame, this);
-                local->transaction.resume (frame, this);
-        }
-
-        return 0;
+        return __afr_dir_write_cbk (frame, cookie, this, op_ret, op_errno, NULL,
+				    preparent, postparent, NULL, NULL, xdata);
 }
 
 
 int
-afr_rmdir_wind (call_frame_t *frame, xlator_t *this)
+afr_rmdir_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_up_children_count (priv->child_count, local->child_up);
-
-        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_rmdir_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->rmdir,
-                                           &local->loc, local->cont.rmdir.flags);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
-}
-
-
-int
-afr_rmdir_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
-
+	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)
+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;
-        call_frame_t  * transaction_frame = NULL;
-        int ret = -1;
-        int op_ret   = -1;
-        int op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        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                = ENOMEM;
+        int                     nlockee                 = 0;
 
         priv = this->private;
 
+        QUORUM_CHECK(rmdir,out);
+
         transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
+        if (!transaction_frame)
                 goto out;
-        }
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		goto out;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
 
-        transaction_frame->local = local;
+        loc_copy (&local->loc, loc);
+	local->inode = inode_ref (loc->inode);
+	local->parent = inode_ref (loc->parent);
 
         local->cont.rmdir.flags = flags;
-        loc_copy (&local->loc, loc);
 
-        if (loc->parent)
-                local->cont.rmdir.parent_ino = loc->parent->ino;
+        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.fop    = afr_rmdir_wind;
-        local->transaction.done   = afr_rmdir_done;
+        local->op = GF_FOP_RMDIR;
+        local->transaction.wind   = afr_rmdir_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
         local->transaction.unwind = afr_rmdir_unwind;
 
-        afr_build_parent_loc (&local->transaction.parent_loc, loc);
+        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;
 
-        afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        nlockee++;
+        ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
+                                     &local->loc,
+                                     NULL,
+                                     priv->child_count);
+        if (ret)
+                goto out;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (rmdir, frame, op_ret, op_errno,
-                                  NULL, NULL);
+        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;
         }
 
+	return 0;
+out:
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
+
+	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 e589efa37..02f0a3682 100644
--- a/xlators/cluster/afr/src/afr-dir-write.h
+++ b/xlators/cluster/afr/src/afr-dir-write.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  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 __DIR_WRITE_H__
@@ -23,38 +14,34 @@
 int32_t
 afr_create (call_frame_t *frame, xlator_t *this,
             loc_t *loc, int32_t flags, mode_t mode,
-            fd_t *fd, dict_t *params);
+            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, dict_t *params);
+	   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, dict_t *params);
+	   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);
+	    loc_t *loc, int xflag, dict_t *xdata);
 
 int32_t
 afr_rmdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, int flags);
+	   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);
+	  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);
+	    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, dict_t *params);
-
-int32_t
-afr_setdents (call_frame_t *frame, xlator_t *this,
-	      fd_t *fd, int32_t flags, dir_entry_t *entries, int32_t count);
+	     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 caac56f65..4cb219246 100644
--- a/xlators/cluster/afr/src/afr-inode-read.c
+++ b/xlators/cluster/afr/src/afr-inode-read.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -44,247 +35,153 @@
 #include "compat-errno.h"
 #include "compat.h"
 
-#include "afr.h"
+#include "afr-transaction.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
- */
-
 /* {{{ access */
 
-int32_t
-afr_access_cbk (call_frame_t *frame, void *cookie,
-                xlator_t *this, int32_t op_ret, int32_t op_errno)
+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;
 
-        read_child = (long) cookie;
+	if (op_ret < 0) {
+		local->op_ret = op_ret;
+		local->op_errno = op_errno;
 
-        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;
+		afr_read_txn_continue (frame, this, (long) cookie);
+		return 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);
-        }
-
-out:
-        if (unwind) {
-                AFR_STACK_UNWIND (access, frame, op_ret, op_errno);
-        }
+	AFR_STACK_UNWIND (access, frame, op_ret, op_errno, xdata);
 
         return 0;
 }
 
 
-int32_t
-afr_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask)
+int
+afr_access_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_ret     = -1;
-        int32_t         op_errno   = 0;
-        int32_t         read_child = -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);
-
-        children = priv->children;
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        frame->local = local;
-
-        op_ret = AFR_LOCAL_INIT (local, priv);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                goto out;
-        }
-
-        local->fresh_children = afr_fresh_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+	afr_private_t *priv = NULL;
+	afr_local_t *local = NULL;
+
+	priv = this->private;
+	local = frame->local;
+
+	if (subvol == -1) {
+		AFR_STACK_UNWIND (access, frame, local->op_ret,
+				  local->op_errno, 0);
+		return 0;
+	}
+
+        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);
-        op_ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.access.last_index);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                op_ret = -1;
-                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);
+	afr_read_txn (frame, this, loc->inode, afr_access_wind,
+		      AFR_METADATA_TRANSACTION);
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (access, frame, op_ret, op_errno);
-        }
+	AFR_STACK_UNWIND (access, frame, -1, op_errno, NULL);
+
         return 0;
 }
 
-
 /* }}} */
 
 /* {{{ stat */
 
-int32_t
+int
 afr_stat_cbk (call_frame_t *frame, void *cookie,
               xlator_t *this, int32_t op_ret, int32_t op_errno,
-              struct iatt *buf)
+              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;
-
-        read_child = (long) cookie;
+        afr_local_t     *local = NULL;
 
         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;
+	if (op_ret < 0) {
+		local->op_ret = op_ret;
+		local->op_errno = op_errno;
 
-                unwind = 0;
+		afr_read_txn_continue (frame, this, (long) cookie);
+		return 0;
+	}
 
-                STACK_WIND_COOKIE (frame, afr_stat_cbk,
-                                   (void *) (long) read_child,
-                                   children[next_call_child],
-                                   children[next_call_child]->fops->stat,
-                                   &local->loc);
-        }
-
-out:
-        if (unwind) {
-                AFR_STACK_UNWIND (stat, frame, op_ret, op_errno, buf);
-        }
+	AFR_STACK_UNWIND (stat, frame, op_ret, op_errno, buf, xdata);
 
         return 0;
 }
 
 
-int32_t
-afr_stat (call_frame_t *frame, xlator_t *this, loc_t *loc)
+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_ret     = -1;
-        int32_t         op_errno   = 0;
-        int32_t         read_child = -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);
-
-        children = priv->children;
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        frame->local = local;
-        op_ret = AFR_LOCAL_INIT (local, priv);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                goto out;
-        }
+	afr_private_t *priv = NULL;
+	afr_local_t *local = NULL;
+
+	priv = this->private;
+	local = frame->local;
+
+	if (subvol == -1) {
+		AFR_STACK_UNWIND (stat, frame, local->op_ret, local->op_errno,
+				  0, 0);
+		return 0;
+	}
+
+        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;
+}
 
-        local->fresh_children = afr_fresh_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                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;
 
-        read_child = afr_inode_get_read_ctx (this, loc->inode,
-                                             local->fresh_children);
-        op_ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.stat.last_index);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                op_ret = -1;
-                goto out;
-        }
-        loc_copy (&local->loc, loc);
+        local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
-        local->cont.stat.ino = loc->inode->ino;
+	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);
+	afr_read_txn (frame, this, loc->inode, afr_stat_wind,
+		      AFR_DATA_TRANSACTION);
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (stat, frame, op_ret, op_errno, NULL);
-        }
+	AFR_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
 
         return 0;
 }
@@ -294,121 +191,76 @@ out:
 
 /* {{{ fstat */
 
-int32_t
+int
 afr_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int32_t op_ret, int32_t op_errno, struct iatt *buf)
+               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;
 
-        read_child = (long) cookie;
+	if (op_ret < 0) {
+		local->op_ret = op_ret;
+		local->op_errno = op_errno;
 
-        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;
+		afr_read_txn_continue (frame, this, (long) cookie);
+		return 0;
+	}
 
-                unwind = 0;
+	AFR_STACK_UNWIND (fstat, frame, op_ret, op_errno, buf, xdata);
 
-                STACK_WIND_COOKIE (frame, afr_fstat_cbk,
-                                   (void *) (long) read_child,
-                                   children[next_call_child],
-                                   children[next_call_child]->fops->fstat,
-                                   local->fd);
-        }
+        return 0;
+}
 
-out:
-        if (unwind) {
-                AFR_STACK_UNWIND (fstat, frame, op_ret, op_errno, buf);
-        }
 
-        return 0;
+int
+afr_fstat_wind (call_frame_t *frame, xlator_t *this, int subvol)
+{
+	afr_private_t *priv = NULL;
+	afr_local_t *local = NULL;
+
+	priv = this->private;
+	local = frame->local;
+
+	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)
+           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_ret     = -1;
-        int32_t         op_errno   = 0;
-        int32_t         read_child = 0;
+        afr_local_t *local = NULL;
+	int op_errno = 0;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (fd, out);
-        VALIDATE_OR_GOTO (this->private, 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_FSTAT;
+	local->fd = fd_ref (fd);
+	if (xdata)
+		local->xdata_req = dict_ref (xdata);
 
-        children = priv->children;
-
-        VALIDATE_OR_GOTO (fd->inode, out);
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        frame->local = local;
-
-        op_ret = AFR_LOCAL_INIT (local, priv);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                goto out;
-        }
-
-        local->fresh_children = afr_fresh_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.fstat.last_index);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                op_ret = -1;
-                goto out;
-        }
-
-        local->cont.fstat.ino = fd->inode->ino;
-        local->fd = fd_ref (fd);
+        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);
+	afr_read_txn (frame, this, fd->inode, afr_fstat_wind,
+		      AFR_DATA_TRANSACTION);
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (fstat, frame, op_ret, op_errno, NULL);
-        }
+	AFR_STACK_UNWIND (fstat, frame, -1, op_errno, NULL, NULL);
 
         return 0;
 }
@@ -417,118 +269,77 @@ out:
 
 /* {{{ readlink */
 
-int32_t
+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)
+                  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;
+	afr_local_t *local = NULL;
 
-        priv     = this->private;
-        children = priv->children;
+	local = frame->local;
 
-        local = frame->local;
+	if (op_ret < 0) {
+		local->op_ret = -1;
+		local->op_errno = op_errno;
 
-        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);
-        }
+		afr_read_txn_continue (frame, this, (long) cookie);
+		return 0;
+	}
 
-out:
-        if (unwind) {
-                AFR_STACK_UNWIND (readlink, frame, op_ret, op_errno, buf, sbuf);
-        }
+	AFR_STACK_UNWIND (readlink, frame, op_ret, op_errno,
+			  buf, sbuf, xdata);
+	return 0;
+}
 
-        return 0;
+int
+afr_readlink_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 (readlink, frame, local->op_ret,
+				  local->op_errno, 0, 0, 0);
+		return 0;
+	}
+
+	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;
 }
 
 
-int32_t
+int
 afr_readlink (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, size_t size)
+              loc_t *loc, size_t size, 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;
+        afr_local_t   * local      = NULL;
         int32_t         op_errno   = 0;
-        int32_t         read_child = -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);
-
-        children = priv->children;
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        frame->local = local;
-        op_ret = AFR_LOCAL_INIT (local, priv);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                goto out;
-        }
-
-        local->fresh_children = afr_fresh_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);
-        op_ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.readlink.last_index);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                op_ret = -1;
-                goto out;
-        }
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
+	local->op = GF_FOP_READLINK;
         loc_copy (&local->loc, loc);
+        local->cont.readlink.size = size;
+	if (xdata)
+		local->xdata_req = dict_ref (xdata);
 
-        local->cont.readlink.size       = size;
-        local->cont.readlink.ino        = loc->inode->ino;
-
-        STACK_WIND_COOKIE (frame, afr_readlink_cbk,
-                           (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->readlink,
-                           loc, size);
+	afr_read_txn (frame, this, loc->inode, afr_readlink_wind,
+		      AFR_DATA_TRANSACTION);
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (readlink, frame, op_ret, op_errno, NULL, NULL);
-        }
         return 0;
+out:
+	AFR_STACK_UNWIND(readlink, frame, -1, op_errno, 0, 0, 0);
+
+	return 0;
 }
 
 
@@ -542,7 +353,7 @@ struct _xattr_key {
 };
 
 
-void
+int
 __gather_xattr_keys (dict_t *dict, char *key, data_t *value,
                      void *data)
 {
@@ -554,18 +365,19 @@ __gather_xattr_keys (dict_t *dict, char *key, data_t *value,
 
                 xkey = GF_CALLOC (1, sizeof (*xkey), gf_afr_mt_xattr_key);
                 if (!xkey)
-                        return;
+                        return -1;
 
                 xkey->key = key;
                 INIT_LIST_HEAD (&xkey->list);
 
                 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;
@@ -586,85 +398,669 @@ __filter_xattrs (dict_t *dict)
 }
 
 
-
-int32_t
+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 *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;
+        afr_local_t *local = NULL;
+
+        local = frame->local;
+
+	if (op_ret < 0) {
+		local->op_ret = op_ret;
+		local->op_errno = op_errno;
+
+		afr_read_txn_continue (frame, this, (long) cookie);
+		return 0;
+	}
+
+	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_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_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;
 
-        read_child = (long) cookie;
+        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_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 (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;
+        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'
+                op_errno = afr_final_errno (local, priv);
+
+                AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, xattr,
+                                  xdata);
+                if (xattr)
+                        dict_unref (xattr);
+        }
+
+        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;
+
+        priv     = this->private;
+        children = priv->children;
+
+        local = frame->local;
+        cky = (long) cookie;
+
+        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_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;
+                }
+
+        unwind:
+                // Updating child_errno with more recent 'events'
+                op_errno = afr_final_errno (local, priv);
+
+                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_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_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;
+
+        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_cbk,
-                                   (void *) (long) read_child,
-                                   children[next_call_child],
-                                   children[next_call_child]->fops->getxattr,
+                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->cont.getxattr.name,
+                                   NULL);
         }
 
-out:
-        if (unwind) {
-                if (op_ret >= 0 && dict)
-                        __filter_xattrs (dict);
+ unwind:
+        if (unwind)
+                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict,
+                                  NULL);
 
-                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
+        return 0;
+}
+
+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;
+
+        LOCK (&frame->lock);
+        {
+                local = frame->local;
+
+                call_cnt = --local->call_count;
+
+                if ((op_ret < 0) || (!dict && !xdata)) {
+                        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;
+                                }
+                        }
+                }
+
+                if (!dict) {
+                        goto unlock;
+                }
+
+                op_ret = dict_get_ptr_and_len (dict, GF_XATTR_LOCKINFO_KEY,
+                                               (void **)&lockinfo_buf, &len);
+
+                if (!lockinfo_buf) {
+                        goto unlock;
+                }
+
+                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 (xdata && local->xdata_rsp) {
+                dict_copy (xdata, local->xdata_rsp);
+        }
+
+        if (!call_cnt) {
+                newdict = dict_new ();
+                if (!newdict) {
+                        local->op_ret = -1;
+                        local->op_errno = ENOMEM;
+                        goto unwind;
+                }
+
+                len = dict_serialized_length (local->dict);
+                if (len == 0) {
+                        goto unwind;
+                }
+
+                lockinfo_buf = GF_CALLOC (1, len, gf_common_mt_char);
+                if (!lockinfo_buf) {
+                        local->op_ret = -1;
+                        local->op_errno = ENOMEM;
+                        goto unwind;
+                }
+
+                op_ret = dict_serialize (local->dict, lockinfo_buf);
+                if (op_ret < 0) {
+                        local->op_ret = -1;
+                        local->op_errno = -op_ret;
+                }
+
+                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;
+                }
+
+        unwind:
+                AFR_STACK_UNWIND (getxattr, frame, op_ret,
+                                  op_errno, newdict,
+                                  local->xdata_rsp);
+        }
+
+        dict_unref (lockinfo);
 
         return 0;
 }
 
 int32_t
-afr_getxattr_unwind (call_frame_t *frame,
-                     int op_ret, int op_errno, dict_t *dict)
-
+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_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
+        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;
+
+                if ((op_ret < 0) || (!dict && !xdata)) {
+                        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;
+                                }
+                        }
+                }
+
+                if (!dict) {
+                        goto unlock;
+                }
+
+                op_ret = dict_get_ptr_and_len (dict, GF_XATTR_LOCKINFO_KEY,
+                                               (void **)&lockinfo_buf, &len);
+
+                if (!lockinfo_buf) {
+                        goto unlock;
+                }
+
+                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 (xdata && local->xdata_rsp) {
+                dict_copy (xdata, local->xdata_rsp);
+        }
+
+        if (!call_cnt) {
+                newdict = dict_new ();
+                if (!newdict) {
+                        local->op_ret = -1;
+                        local->op_errno = ENOMEM;
+                        goto unwind;
+                }
+
+                len = dict_serialized_length (local->dict);
+                if (len <= 0) {
+                        goto unwind;
+                }
+
+                lockinfo_buf = GF_CALLOC (1, len, gf_common_mt_char);
+                if (!lockinfo_buf) {
+                        local->op_ret = -1;
+                        local->op_errno = ENOMEM;
+                        goto unwind;
+                }
+
+                op_ret = dict_serialize (local->dict, lockinfo_buf);
+                if (op_ret < 0) {
+                        local->op_ret = -1;
+                        local->op_errno = -op_ret;
+                }
+
+                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;
+                }
+
+        unwind:
+                AFR_STACK_UNWIND (fgetxattr, frame, op_ret,
+                                  op_errno, newdict,
+                                  local->xdata_rsp);
+        }
+
+        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_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;
+        }
+
+        local = frame->local;
+        cky = (long) cookie;
+
+        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);
+                }
+
+                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;
+                }
+
+                /* closing part */
+                *(xattr_serz + padding + tlen) = ')';
+                *(xattr_serz + padding + tlen + 1) = '\0';
+
+                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");
+
+        unwind:
+                AFR_STACK_UNWIND (fgetxattr, frame, local->op_ret,
+                                  local->op_errno, nxattr, local->xdata_rsp);
+
+                if (nxattr)
+                        dict_unref (nxattr);
+        }
+
+        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 *dict, dict_t *xdata)
 {
-        afr_local_t *local             = NULL;
-        int32_t      callcnt           = 0;
-        int          ret               = 0;
-        char        *pathinfo          = NULL;
-        char        *pathinfo_serz     = NULL;
-        char        pathinfo_cky[1024] = {0,};
-        dict_t      *xattr             = NULL;
-        long         cky               = 0;
-        int32_t      padding           = 0;
-        int32_t      tlen              = 0;
+        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 (this->name, GF_LOG_ERROR, "possible NULL deref");
+                gf_log ("", GF_LOG_ERROR, "possible NULL deref");
                 goto out;
         }
 
@@ -675,6 +1071,14 @@ afr_getxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
                 {
                         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);
+                        }
+
                         if (!dict || (op_ret < 0))
                                 goto out;
 
@@ -682,142 +1086,341 @@ afr_getxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
                                 local->dict = dict_new ();
 
                         if (local->dict) {
-                                ret = dict_get_str (dict, GF_XATTR_PATHINFO_KEY, &pathinfo);
+                                ret = dict_get_str (dict,
+                                                    local->cont.getxattr.name,
+                                                    &xattr);
                                 if (ret)
                                         goto out;
 
-                                pathinfo = gf_strdup (pathinfo);
+                                xattr = gf_strdup (xattr);
 
-                                snprintf (pathinfo_cky, 1024, "%s-%ld", GF_XATTR_PATHINFO_KEY, cky);
-                                ret = dict_set_dynstr (local->dict, pathinfo_cky, pathinfo);
+                                (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 pathinfo cookie key");
+                                        gf_log (this->name, GF_LOG_ERROR,
+                                                "Cannot set xattr cookie key");
                                         goto out;
                                 }
 
-                                local->cont.getxattr.pathinfo_len += strlen (pathinfo) + 1;
+                                local->cont.getxattr.xattr_len += strlen (xattr) + 1;
                         }
                 }
  out:
         UNLOCK (&frame->lock);
 
         if (!callcnt) {
-                if (!local->cont.getxattr.pathinfo_len)
+                if (!local->cont.getxattr.xattr_len)
                         goto unwind;
 
-                xattr = dict_new ();
-                if (!xattr)
+                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.pathinfo_len += (padding + 2);
+                padding += strlen (this->name) + strlen (AFR_PATHINFO_HEADER) + 4;
+                local->cont.getxattr.xattr_len += (padding + 2);
 
-                pathinfo_serz = GF_CALLOC (local->cont.getxattr.pathinfo_len, sizeof (char),
-                                           gf_common_mt_char);
+                xattr_serz = GF_CALLOC (local->cont.getxattr.xattr_len,
+                                        sizeof (char), gf_common_mt_char);
 
-                if (!pathinfo_serz)
+                if (!xattr_serz)
                         goto unwind;
 
                 /* the xlator info */
-                sprintf (pathinfo_serz, "(<"AFR_PATHINFO_HEADER"%s> ", this->name);
+                (void) sprintf (xattr_serz, "(<"AFR_PATHINFO_HEADER"%s> ",
+                                this->name);
 
                 /* actual series of pathinfo */
-                ret = dict_serialize_value_with_delim (local->dict, pathinfo_serz + strlen (pathinfo_serz),
+                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");
+                        gf_log (this->name, GF_LOG_ERROR, "Error serializing"
+                                " dictionary");
                         goto unwind;
                 }
 
                 /* closing part */
-                *(pathinfo_serz + padding + tlen) = ')';
-                *(pathinfo_serz + padding + tlen + 1) = '\0';
+                *(xattr_serz + padding + tlen) = ')';
+                *(xattr_serz + padding + tlen + 1) = '\0';
 
-                ret = dict_set_dynstr (xattr, GF_XATTR_PATHINFO_KEY, pathinfo_serz);
+                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");
+                        gf_log (this->name, GF_LOG_ERROR, "Cannot set pathinfo"
+                                " key in dict");
 
         unwind:
-                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr);
+                AFR_STACK_UNWIND (getxattr, frame, local->op_ret,
+                                  local->op_errno, nxattr, local->xdata_rsp);
 
-                if (local->dict)
-                        dict_unref (local->dict);
-
-                if (xattr)
-                        dict_unref (xattr);
+                if (nxattr)
+                        dict_unref (nxattr);
         }
 
         return ret;
 }
 
+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)
+{
+        afr_local_t *local          = NULL;
+        int32_t      callcnt        = 0;
+
+        if (!frame || !frame->local || !this) {
+                gf_log ("", GF_LOG_ERROR, "possible NULL deref");
+                goto out;
+        }
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                callcnt = --local->call_count;
+
+                if (!dict || (op_ret < 0)) {
+                        local->op_errno = op_errno;
+                        goto cleanup;
+                }
+
+                if (!local->dict)
+                        local->dict = dict_copy_with_ref (dict, NULL);
+                else
+                        dict_foreach (dict, afr_aggregate_stime_xattr,
+                                      local->dict);
+                local->op_ret = 0;
+        }
+
+cleanup:
+        UNLOCK (&frame->lock);
+
+        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 {
+                        *cbk = afr_getxattr_pathinfo_cbk;
+                }
+        } else if (!strncmp (name, GF_XATTR_CLRLK_CMD,
+                             strlen (GF_XATTR_CLRLK_CMD))) {
+                if (is_fgetxattr) {
+                        *cbk = afr_fgetxattr_clrlk_cbk;
+                } else {
+                        *cbk = afr_getxattr_clrlk_cbk;
+                }
+        } else if (!strncmp (name, GF_XATTR_LOCKINFO_KEY,
+                             strlen (GF_XATTR_LOCKINFO_KEY))) {
+                if (is_fgetxattr) {
+                        *cbk = afr_fgetxattr_lockinfo_cbk;
+                } else {
+                        *cbk = afr_getxattr_lockinfo_cbk;
+                }
+        } else if (fnmatch (GF_XATTR_STIME_PATTERN, name, FNM_NOESCAPE) == 0) {
+                *cbk = afr_common_getxattr_stime_cbk;
+        } else {
+                is_spl = _gf_false;
+        }
+
+out:
+        return is_spl;
+}
+
+static void
+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;
+        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 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, name, NULL);
+                        if (!--call_count)
+                                break;
+                }
+        }
+        return;
+}
+
 int32_t
 afr_getxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, const char *name)
+              loc_t *loc, const char *name, dict_t *xdata)
 {
-        afr_private_t   *priv         = NULL;
-        xlator_t        **children    = NULL;
-        int             call_child    = 0;
-        afr_local_t     *local        = NULL;
-        xlator_list_t   *trav         = NULL;
-        xlator_t        **sub_volumes = NULL;
-        int             i             = 0;
-        int32_t         op_ret        = -1;
-        int32_t         op_errno      = 0;
-        int32_t         read_child    = -1;
-
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        afr_private_t           *priv         = NULL;
+        xlator_t                **children    = NULL;
+        afr_local_t             *local        = NULL;
+        xlator_list_t           *trav         = NULL;
+        xlator_t                **sub_volumes = NULL;
+        int                     i             = 0;
+        int32_t                 op_errno      = 0;
+        int                     ret           = -1;
+        fop_getxattr_cbk_t      cbk           = NULL;
+        int                     afr_xtime_gauge[MCNT_MAX] = {0,};
+
+
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
         priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
 
         children = priv->children;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        frame->local = local;
+        loc_copy (&local->loc, loc);
+
+	local->op = GF_FOP_GETXATTR;
+
+	if (xdata)
+		local->xdata_req = dict_ref (xdata);
+
+        if (!name)
+                goto no_name;
+
+        local->cont.getxattr.name = gf_strdup (name);
 
-        op_ret = AFR_LOCAL_INIT (local, priv);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
+	if (!local->cont.getxattr.name) {
+		op_errno = ENOMEM;
+		goto out;
+	}
+
+        if (!strncmp (name, AFR_XATTR_PREFIX,
+                      strlen (AFR_XATTR_PREFIX))) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "%s: no data present for key %s",
+                        loc->path, name);
+                op_errno = ENODATA;
                 goto out;
         }
+        if ((strcmp (GF_XATTR_MARKER_KEY, name) == 0)
+            && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
 
-        loc_copy (&local->loc, loc);
-        if (name)
-                local->cont.getxattr.name = gf_strdup (name);
+                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_UUID_TYPE,
+                                           marker_uuid_default_gauge,
+                                           priv->vol_uuid)) {
 
-        if (name) {
-                if (!strncmp (name, AFR_XATTR_PREFIX,
-                              strlen (AFR_XATTR_PREFIX))) {
                         gf_log (this->name, GF_LOG_INFO,
-                                "%s: no data present for key %s",
+                                "%s: failed to get marker attr (%s)",
                                 loc->path, name);
-                        op_errno = ENODATA;
+                        op_errno = EINVAL;
                         goto out;
                 }
-                if ((strcmp (GF_XATTR_MARKER_KEY, name) == 0)
-                    && (-1 == frame->root->pid)) {
 
+                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_all_subvols (this, frame, name, loc, cbk);
+                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;
+        }
+
+        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 *));
+                        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,
+                        /* don't err out on getting ENOTCONN (brick down)
+                         * from a subset of the bricks
+                         */
+                        memcpy (afr_xtime_gauge, marker_xtime_default_gauge,
+                                sizeof (afr_xtime_gauge));
+                        afr_xtime_gauge[MCNT_NOTFOUND] = 0;
+                        afr_xtime_gauge[MCNT_ENOTCONN] = 0;
+                        if (cluster_getmarkerattr (frame, this, loc,
+                                                   name, local,
+                                                   afr_getxattr_unwind,
                                                    sub_volumes,
                                                    priv->child_count,
-                                                   MARKER_UUID_TYPE,
+                                                   MARKER_XTIME_TYPE,
+                                                   afr_xtime_gauge,
                                                    priv->vol_uuid)) {
-
                                 gf_log (this->name, GF_LOG_INFO,
                                         "%s: failed to get marker attr (%s)",
                                         loc->path, name);
@@ -827,86 +1430,150 @@ afr_getxattr (call_frame_t *frame, xlator_t *this,
 
                         return 0;
                 }
+        }
 
-                /*
-                 * if we are doing getxattr with pathinfo as the key then we
-                 * collect information from all childs
-                 */
-                if (strncmp (name, GF_XATTR_PATHINFO_KEY,
-                             strlen (GF_XATTR_PATHINFO_KEY)) == 0) {
-
-                        local->call_count = priv->child_count;
-                        for (i = 0; i < priv->child_count; i++) {
-                                STACK_WIND_COOKIE (frame, afr_getxattr_pathinfo_cbk,
-                                                   (void *) (long) i,
-                                                   children[i], children[i]->fops->getxattr,
-                                                   loc, name);
-                        }
+no_name:
 
-                        return 0;
-                }
+	afr_read_txn (frame, this, local->loc.inode, afr_getxattr_wind,
+		      AFR_METADATA_TRANSACTION);
 
-                if (*priv->vol_uuid) {
-                        if ((match_uuid_local (name, priv->vol_uuid) == 0)
-                            && (-1 == frame->root->pid)) {
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
 
-                                local->marker.call_count = priv->child_count;
+/* {{{ fgetxattr */
 
-                                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;
+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_local_t *local = NULL;
 
-                                }
+	local = frame->local;
 
-                                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;
-                                }
+	if (op_ret < 0) {
+		local->op_ret = -1;
+		local->op_errno = op_errno;
 
-                                return 0;
-                        }
+		afr_read_txn_continue (frame, this, (long) cookie);
+		return 0;
+	}
+
+	if (dict)
+		afr_filter_xattrs (dict);
+
+	AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, dict, xdata);
+
+	return 0;
+}
+
+int
+afr_fgetxattr_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 (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_all_subvols (xlator_t *this, call_frame_t *frame,
+			   fop_fgetxattr_cbk_t cbk)
+{
+        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 up-children count is 0, afr_local_init would have failed already
+        //and the call would have unwound so not handling it here.
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND_COOKIE (frame, cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->fgetxattr,
+                                           local->fd, local->cont.getxattr.name,
+					   NULL);
+                        if (!--call_count)
+                                break;
                 }
         }
 
-        local->fresh_children = afr_fresh_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+        return;
+}
 
-        read_child = afr_inode_get_read_ctx (this, loc->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;
+
+int
+afr_fgetxattr (call_frame_t *frame, xlator_t *this,
+               fd_t *fd, const char *name, dict_t *xdata)
+{
+        afr_local_t          *local      = NULL;
+        int32_t               op_errno   = 0;
+        fop_fgetxattr_cbk_t   cbk        = NULL;
+
+        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;
         }
 
-        STACK_WIND_COOKIE (frame, afr_getxattr_cbk,
-                           (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->getxattr,
-                           loc, name);
+        afr_fix_open (fd, this);
 
-        op_ret = 0;
+	afr_read_txn (frame, this, fd->inode, afr_fgetxattr_wind,
+		      AFR_METADATA_TRANSACTION);
+
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, NULL);
-        }
+	AFR_STACK_UNWIND (fgetxattr, frame, -1, op_errno, NULL, NULL);
+
         return 0;
 }
 
@@ -915,140 +1582,84 @@ out:
 
 /* {{{ readv */
 
-/**
- * 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
- */
-
-int32_t
+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)
+               struct iobref *iobref, 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;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+	local = frame->local;
 
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
-
-        children = priv->children;
-
-        local = frame->local;
+	if (op_ret < 0) {
+		local->op_ret = -1;
+		local->op_errno = op_errno;
 
-        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;
+		afr_read_txn_continue (frame, this, (long) cookie);
+		return 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);
-        }
+	AFR_STACK_UNWIND (readv, frame, op_ret, op_errno,
+			  vector, count, buf, iobref, xdata);
+	return 0;
+}
 
-out:
-        if (unwind) {
-                AFR_STACK_UNWIND (readv, frame, op_ret, op_errno,
-                                  vector, count, buf, iobref);
-        }
 
-        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;
+
+	if (subvol == -1) {
+		AFR_STACK_UNWIND (readv, frame, local->op_ret, local->op_errno,
+				  0, 0, 0, 0, 0);
+		return 0;
+	}
+
+	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 (call_frame_t *frame, xlator_t *this,
-           fd_t *fd, size_t size, off_t offset)
+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             call_child = 0;
-        int32_t         op_ret     = -1;
         int32_t         op_errno   = 0;
-        int32_t         read_child = -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;
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        frame->local = local;
-        op_ret = AFR_LOCAL_INIT (local, priv);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                goto out;
-        }
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
-        local->fresh_children = afr_fresh_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.readv.last_index);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                op_ret = -1;
-                goto out;
-        }
-
-        local->fd                    = fd_ref (fd);
+	local->op = GF_FOP_READ;
+        local->fd = fd_ref (fd);
+        local->cont.readv.size = size;
+        local->cont.readv.offset = offset;
+        local->cont.readv.flags = flags;
+	if (xdata)
+		local->xdata_req = dict_ref (xdata);
 
-        local->cont.readv.ino        = fd->inode->ino;
-        local->cont.readv.size       = size;
-        local->cont.readv.offset     = offset;
+        afr_fix_open (fd, this);
 
-        STACK_WIND_COOKIE (frame, afr_readv_cbk,
-                           (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->readv,
-                           fd, size, offset);
+	afr_read_txn (frame, this, fd->inode, afr_readv_wind,
+		      AFR_DATA_TRANSACTION);
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (readv, frame, op_ret, op_errno, NULL, 0, NULL,
-                                  NULL);
-        }
         return 0;
+out:
+	AFR_STACK_UNWIND(readv, frame, -1, op_errno, 0, 0, 0, 0, 0);
+
+	return 0;
 }
 
 /* }}} */
diff --git a/xlators/cluster/afr/src/afr-inode-read.h b/xlators/cluster/afr/src/afr-inode-read.h
index acc814fb7..e4091a793 100644
--- a/xlators/cluster/afr/src/afr-inode-read.h
+++ b/xlators/cluster/afr/src/afr-inode-read.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  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 __INODE_READ_H__
@@ -22,26 +13,30 @@
 
 int32_t
 afr_access (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t mask);
+	    loc_t *loc, int32_t mask, dict_t *xdata);
 
 int32_t
 afr_stat (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc);
+	  loc_t *loc, dict_t *xdata);
 
 int32_t
 afr_fstat (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd);
+	   fd_t *fd, dict_t *xdata);
 
 int32_t
 afr_readlink (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, size_t size);
+	      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);
+	   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);
+	      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);
 
 #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 564bb953a..00e0d2676 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -47,140 +38,301 @@
 #include "afr.h"
 #include "afr-transaction.h"
 
-/* {{{ writev */
 
-int
-afr_writev_unwind (call_frame_t *frame, xlator_t *this)
+static void
+__afr_inode_write_finalize (call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	int read_subvol = 0;
+	int i = 0;
+
+	local = frame->local;
+	priv = this->private;
+
+	if (local->inode) {
+		if (local->transaction.type == AFR_METADATA_TRANSACTION)
+			read_subvol = afr_metadata_subvol_get (local->inode, this,
+							       NULL, NULL);
+		else
+			read_subvol = afr_data_subvol_get (local->inode, this,
+							   NULL, NULL);
+	}
+
+	local->op_ret = -1;
+	local->op_errno = afr_final_errno (local, priv);
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!local->replies[i].valid)
+			continue;
+		if (local->replies[i].op_ret < 0) {
+			afr_inode_read_subvol_reset (local->inode, this);
+			continue;
+		}
+
+		/* Order of checks in the compound conditional
+		   below is important.
+
+		   - Highest precedence: largest op_ret
+		   - Next precendence: 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);
+			}
+		}
+	}
+}
 
-        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);
+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 *xdata)
+{
+        afr_local_t *local = NULL;
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (writev, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  &local->cont.writev.prebuf,
-                                  &local->cont.writev.postbuf);
-        }
-        return 0;
+        local = frame->local;
+
+	local->replies[child_index].valid = 1;
+	local->replies[child_index].op_ret = op_ret;
+	local->replies[child_index].op_errno = op_errno;
+
+	if (op_ret >= 0) {
+		if (prebuf)
+			local->replies[child_index].prestat = *prebuf;
+		if (postbuf)
+			local->replies[child_index].poststat = *postbuf;
+		if (xdata)
+			local->replies[child_index].xdata = dict_ref (xdata);
+	} else {
+		afr_transaction_fop_failed (frame, this, child_index);
+	}
+
+        return;
 }
 
 
-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)
+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 *xdata)
 {
-        afr_local_t *   local = NULL;
+        afr_local_t *local = NULL;
         int child_index = (long) cookie;
-        int call_count  = -1;
-        int read_child  = 0;
+        int call_count = -1;
 
         local = frame->local;
 
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
-
         LOCK (&frame->lock);
         {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
-
-                if (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.writev.prebuf  = *prebuf;
-                                local->cont.writev.postbuf = *postbuf;
-                        }
-
-                        if (child_index == read_child) {
-                                local->cont.writev.prebuf  = *prebuf;
-                                local->cont.writev.postbuf = *postbuf;
-                        }
-                }
-
-                local->op_errno = op_errno;
+                __afr_inode_write_fill (frame, this, child_index, op_ret,
+					op_errno, prebuf, postbuf, xdata);
         }
         UNLOCK (&frame->lock);
 
         call_count = afr_frame_return (frame);
 
         if (call_count == 0) {
-                local->transaction.unwind (frame, this);
+		__afr_inode_write_finalize (frame, this);
+
+		if (afr_txn_nothing_failed (frame, this))
+			local->transaction.unwind (frame, this);
 
                 local->transaction.resume (frame, this);
         }
+
         return 0;
 }
 
+/* {{{ writev */
 
-int
-afr_writev_wind (call_frame_t *frame, xlator_t *this)
+void
+afr_writev_copy_outvars (call_frame_t *src_frame, call_frame_t *dst_frame)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int i = 0;
-        int call_count = -1;
+        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);
+}
 
+void
+afr_writev_unwind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *   local = NULL;
         local = frame->local;
-        priv = this->private;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+        AFR_STACK_UNWIND (writev, frame,
+                          local->op_ret, local->op_errno,
+                          &local->cont.inode_wfop.prebuf,
+                          &local->cont.inode_wfop.postbuf,
+                          local->xdata_rsp);
+}
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
+
+int
+afr_transaction_writev_unwind (call_frame_t *frame, xlator_t *this)
+{
+        call_frame_t *fop_frame = NULL;
+
+        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;
+}
 
-        local->call_count = call_count;
+static void
+afr_writev_handle_short_writes (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local = NULL;
+        afr_private_t *priv  = NULL;
+        int           i      = 0;
 
+        local = frame->local;
+        priv = this->private;
+        /*
+         * We already have the best case result of the writev calls staged
+         * as the return value. Any writev that returns some value less
+         * than the best case is now out of sync, so mark the fop as
+         * failed. Note that fops that have returned with errors have
+         * already been marked as failed.
+         */
         for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[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.iobref);
-
-                        if (!--call_count)
-                                break;
-                }
+                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,
+                     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 ret = 0;
+        uint32_t open_fd_count = 0;
+        uint32_t write_is_append = 0;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                __afr_inode_write_fill (frame, this, child_index, op_ret,
+					op_errno, prebuf, postbuf, 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_OPEN_FD_COUNT,
+				       &open_fd_count);
+		if (ret == -1)
+			goto unlock;
+		if ((open_fd_count > local->open_fd_count)) {
+			local->open_fd_count = open_fd_count;
+			local->update_open_fd_count = _gf_true;
+		}
+        }
+unlock:
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
 
+        if (call_count == 0) {
+		if (!local->stable_write && !local->append_write)
+			/* An appended write removes the necessity to
+			   fsync() the file. This is because self-heal
+			   has the logic to check for larger file when
+			   the xattrs are not reliably pointing at
+			   a stale file.
+			*/
+			afr_fd_report_unstable_write (this, local->fd);
+
+		__afr_inode_write_finalize (frame, this);
+
+                afr_writev_handle_short_writes (frame, this);
+
+                if (local->update_open_fd_count)
+                        afr_handle_open_fd_count (frame, this);
+
+                if (!afr_txn_nothing_failed (frame, this)) {
+                        //Don't unwind until post-op is complete
+                        local->transaction.resume (frame, this);
+                } else {
+                /*
+                 * Generally inode-write fops do transaction.unwind then
+                 * transaction.resume, but writev needs to make sure that
+                 * delayed post-op frame is placed in fdctx before unwind
+                 * happens. This prevents the race of flush doing the
+                 * changelog wakeup first in fuse thread and then this
+                 * writev placing its delayed post-op frame in fdctx.
+                 * This helps flush make sure all the delayed post-ops are
+                 * completed.
+                 */
+
+                        fop_frame = afr_transaction_detach_fop_frame (frame);
+                        afr_writev_copy_outvars (frame, fop_frame);
+                        local->transaction.resume (frame, this);
+                        afr_writev_unwind (fop_frame, this);
+                }
+        }
         return 0;
 }
 
 
 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;
+        afr_private_t *priv = NULL;
 
         local = frame->local;
+        priv = this->private;
 
-        iobref_unref (local->cont.writev.iobref);
-        local->cont.writev.iobref = NULL;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
-
+	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;
 }
 
@@ -188,32 +340,39 @@ afr_writev_done (call_frame_t *frame, xlator_t *this)
 int
 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;
-
-        local = frame->local;
+        call_frame_t    *transaction_frame = NULL;
+        afr_local_t     *local             = NULL;
+        int             ret   = -1;
+        int             op_errno = ENOMEM;
 
         transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
+        if (!transaction_frame)
                 goto out;
-        }
 
+        local = frame->local;
         transaction_frame->local = local;
-        frame->local = NULL;
+	frame->local = NULL;
 
-        local->op = GF_FOP_WRITE;
+	if (!AFR_FRAME_INIT (frame, op_errno))
+		goto out;
 
-        local->success_count      = 0;
+        local->op = GF_FOP_WRITE;
 
-        local->transaction.fop    = afr_writev_wind;
-        local->transaction.done   = afr_writev_done;
-        local->transaction.unwind = afr_writev_unwind;
+        local->transaction.wind   = afr_writev_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_transaction_writev_unwind;
 
         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 {
@@ -222,16 +381,18 @@ afr_do_writev (call_frame_t *frame, xlator_t *this)
                                                          local->cont.writev.count);
         }
 
-        afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (writev, frame, op_ret, op_errno, NULL, NULL);
-        }
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
 
+	AFR_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
@@ -239,59 +400,67 @@ out:
 int
 afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
             struct iovec *vector, int32_t count, off_t offset,
-            struct iobref *iobref)
+            uint32_t flags, struct iobref *iobref, dict_t *xdata)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        int ret = -1;
-        int op_ret   = -1;
-        int op_errno = 0;
-        uint64_t ctx = 0;
-        afr_fd_ctx_t *fd_ctx = NULL;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        afr_private_t *priv = NULL;
+        afr_local_t *local = NULL;
+        int op_errno = ENOMEM;
 
         priv = this->private;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+        QUORUM_CHECK(writev,out);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
-
-        frame->local = local;
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
-        local->cont.writev.vector     = iov_dup (vector, count);
+        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.ino        = fd->inode->ino;
+        local->cont.writev.flags      = flags;
         local->cont.writev.iobref     = iobref_ref (iobref);
 
-        local->fd                = fd_ref (fd);
+	if (xdata)
+		local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        ret = fd_ctx_get (fd, this, &ctx);
-        if (ret < 0) {
-                goto out;
-        }
+	if (!local->xdata_req)
+		goto out;
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+        local->fd = fd_ref (fd);
+	local->inode = inode_ref (fd->inode);
 
-        if (fd_ctx->up_count < priv->up_count) {
-                local->openfd_flush_cbk = afr_do_writev;
-                afr_openfd_flush (frame, this, fd);
-        } else {
-                afr_do_writev (frame, this);
-        }
+	if (dict_set_uint32 (local->xdata_req, GLUSTERFS_OPEN_FD_COUNT, 4)) {
+		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));
 
-        op_ret = 0;
+        afr_fix_open (fd, this);
+
+        afr_do_writev (frame, this);
+
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (writev, frame, op_ret, op_errno, NULL, NULL);
-        }
+	AFR_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL, NULL);
 
         return 0;
 }
@@ -309,21 +478,13 @@ afr_truncate_unwind (call_frame_t *frame, xlator_t *this)
 
         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);
-        }
+	main_frame = afr_transaction_detach_fop_frame (frame);
+	if (!main_frame)
+		return 0;
 
+	AFR_STACK_UNWIND (truncate, main_frame, local->op_ret, local->op_errno,
+			  &local->cont.inode_wfop.prebuf,
+			  &local->cont.inode_wfop.postbuf, local->xdata_rsp);
         return 0;
 }
 
@@ -331,175 +492,99 @@ afr_truncate_unwind (call_frame_t *frame, xlator_t *this)
 int
 afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                       struct iatt *postbuf)
+                       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;
-                        }
+	afr_local_t *local = NULL;
 
-                        local->success_count++;
+	local = frame->local;
 
-                        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 (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, xdata);
 }
 
 
-int32_t
-afr_truncate_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_truncate_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_up_children_count (priv->child_count, local->child_up);
-
-        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_truncate_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->truncate,
-                                           &local->loc,
-                                           local->cont.truncate.offset);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
-}
-
-
-int
-afr_truncate_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
-
+	STACK_WIND_COOKIE (frame, afr_truncate_wind_cbk, (void *) (long) subvol,
+			   priv->children[subvol],
+			   priv->children[subvol]->fops->truncate,
+			   &local->loc, local->cont.truncate.offset,
+			   local->xdata_req);
         return 0;
 }
 
 
 int
 afr_truncate (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, off_t offset)
+              loc_t *loc, off_t offset, dict_t *xdata)
 {
         afr_private_t * priv  = NULL;
         afr_local_t   * local = NULL;
         call_frame_t   *transaction_frame = NULL;
-        int ret = -1;
-        int op_ret   = -1;
-        int op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+	int ret = -1;
+        int op_errno = ENOMEM;
 
         priv = this->private;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+        QUORUM_CHECK(truncate,out);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame)
                 goto out;
-        }
-
-        transaction_frame->local = local;
 
-        local->op_ret = -1;
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		goto out;
 
         local->cont.truncate.offset  = offset;
-        local->cont.truncate.ino     = loc->inode->ino;
+	if (xdata)
+		local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        local->transaction.fop    = afr_truncate_wind;
-        local->transaction.done   = afr_truncate_done;
+	if (!local->xdata_req)
+		goto out;
+
+	local->transaction.wind   = afr_truncate_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
         local->transaction.unwind = afr_truncate_unwind;
 
         loc_copy (&local->loc, loc);
+	local->inode = inode_ref (loc->inode);
+
+        local->op = GF_FOP_TRUNCATE;
 
         local->transaction.main_frame = frame;
         local->transaction.start   = offset;
         local->transaction.len     = 0;
 
-        afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+	/* Set it true speculatively, will get reset in afr_truncate_wind_cbk
+	   if truncate was not a NOP */
+	local->stable_write = _gf_true;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (truncate, frame, op_ret, op_errno, NULL, NULL);
+        ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
         }
 
+	return 0;
+out:
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
+
+	AFR_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
@@ -517,20 +602,13 @@ afr_ftruncate_unwind (call_frame_t *frame, xlator_t *this)
 
         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);
+	main_frame = afr_transaction_detach_fop_frame (frame);
+	if (!main_frame)
+		return 0;
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (ftruncate, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.ftruncate.prebuf,
-                                  &local->cont.ftruncate.postbuf);
-        }
+	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;
 }
 
@@ -538,994 +616,1136 @@ afr_ftruncate_unwind (call_frame_t *frame, xlator_t *this)
 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)
+                        struct iatt *postbuf, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int child_index = (long) cookie;
-        int call_count  = -1;
-        int need_unwind = 0;
-        int read_child  = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
-
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
+	afr_local_t *local = NULL;
 
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.ftruncate.prebuf  = *prebuf;
-                                local->cont.ftruncate.postbuf = *postbuf;
-                        }
+	local = frame->local;
 
-                        if (child_index == read_child) {
-                                local->cont.ftruncate.prebuf  = *prebuf;
-                                local->cont.ftruncate.postbuf = *postbuf;
-                        }
+	if (op_ret == 0 && prebuf->ia_size != postbuf->ia_size)
+		local->stable_write = _gf_false;
 
-                        local->success_count++;
+	return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+				      prebuf, postbuf, xdata);
+}
 
-                        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);
+int
+afr_ftruncate_wind (call_frame_t *frame, xlator_t *this, int subvol)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
 
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+	local = frame->local;
+	priv = this->private;
 
+	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_ftruncate_wind (call_frame_t *frame, xlator_t *this)
+afr_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+	       dict_t *xdata)
 {
-        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;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+        QUORUM_CHECK(ftruncate,out);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+	transaction_frame = copy_frame (frame);
+	if (!frame)
+		goto out;
 
-        local->call_count = call_count;
+        local = AFR_FRAME_INIT (transaction_frame, op_errno);
+        if (!local)
+		goto out;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[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);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+        local->cont.ftruncate.offset  = offset;
+	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->fd = fd_ref (fd);
+	local->inode = inode_ref (fd->inode);
 
-int
-afr_ftruncate_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
+        local->op = GF_FOP_FTRUNCATE;
 
-        local = frame->local;
+	local->transaction.wind   = afr_ftruncate_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_ftruncate_unwind;
 
-        local->transaction.unwind (frame, this);
+        local->transaction.main_frame = frame;
 
-        AFR_STACK_DESTROY (frame);
+        local->transaction.start   = local->cont.ftruncate.offset;
+        local->transaction.len     = 0;
+
+        afr_fix_open (fd, this);
+
+	/* Set it true speculatively, will get reset in afr_ftruncate_wind_cbk
+	   if truncate was not a NOP */
+	local->stable_write = _gf_true;
+
+        ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
+        }
+
+	return 0;
+out:
+	AFR_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
 
         return 0;
 }
 
+/* }}} */
+
+/* {{{ setattr */
 
 int
-afr_do_ftruncate (call_frame_t *frame, xlator_t *this)
+afr_setattr_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;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
-
-        transaction_frame->local = local;
-        frame->local = NULL;
+	main_frame = afr_transaction_detach_fop_frame (frame);
+	if (!main_frame)
+		return 0;
 
-        local->op = GF_FOP_FTRUNCATE;
+	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;
+}
 
-        local->transaction.fop    = afr_ftruncate_wind;
-        local->transaction.done   = afr_ftruncate_done;
-        local->transaction.unwind = afr_ftruncate_unwind;
 
-        local->transaction.main_frame = frame;
+int
+afr_setattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno,
+                      struct iatt *preop, struct iatt *postop, dict_t *xdata)
+{
+	return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+				      preop, postop, xdata);
+}
 
-        local->transaction.start   = local->cont.ftruncate.offset;
-        local->transaction.len     = 0;
 
-        afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+int
+afr_setattr_wind (call_frame_t *frame, xlator_t *this, int subvol)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (ftruncate, frame, op_ret, op_errno, NULL, 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 (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, off_t offset)
+afr_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *buf,
+	     int32_t valid, dict_t *xdata)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
+        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;
-        uint64_t ctx = 0;
-        afr_fd_ctx_t *fd_ctx = NULL;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        int op_errno = ENOMEM;
 
         priv = this->private;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        ret = AFR_LOCAL_INIT (local, priv);
+        QUORUM_CHECK(setattr,out);
 
-        if (ret < 0) {
-                op_errno = -ret;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame)
                 goto out;
-        }
 
-        frame->local = local;
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		goto out;
 
-        local->cont.ftruncate.offset  = offset;
-        local->cont.ftruncate.ino     = fd->inode->ino;
+        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 ();
 
-        local->fd = fd_ref (fd);
+	if (!local->xdata_req)
+		goto out;
 
-        ret = fd_ctx_get (fd, this, &ctx);
-        if (ret < 0) {
-                goto out;
-        }
+        local->transaction.wind   = afr_setattr_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_setattr_unwind;
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+        loc_copy (&local->loc, loc);
+	local->inode = inode_ref (loc->inode);
 
-        if (fd_ctx->up_count < priv->up_count) {
-                local->openfd_flush_cbk = afr_do_ftruncate;
-                afr_openfd_flush (frame, this, fd);
-        } else {
-                afr_do_ftruncate (frame, this);
+	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;
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (ftruncate, frame, op_ret, op_errno, NULL, NULL);
-        }
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
 
+	AFR_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
-/* }}} */
-
-/* {{{ setattr */
+/* {{{ fsetattr */
 
 int
-afr_setattr_unwind (call_frame_t *frame, xlator_t *this)
+afr_fsetattr_unwind (call_frame_t *frame, xlator_t *this)
 {
         afr_local_t *   local = NULL;
         call_frame_t   *main_frame = NULL;
 
         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);
-        }
+	main_frame = afr_transaction_detach_fop_frame (frame);
+        if (!main_frame)
+		return 0;
 
+	AFR_STACK_UNWIND (fsetattr, main_frame, local->op_ret, local->op_errno,
+			  &local->cont.inode_wfop.prebuf,
+			  &local->cont.inode_wfop.postbuf, local->xdata_rsp);
         return 0;
 }
 
 
 int
-afr_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)
+afr_fsetattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno,
+                       struct iatt *preop, struct iatt *postop, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int child_index = (long) cookie;
-        int read_child  = 0;
-        int call_count  = -1;
-        int need_unwind = 0;
+	return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+				      preop, postop, 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;
+        priv = this->private;
 
-        read_child = afr_inode_get_read_ctx (this, local->loc.inode, NULL);
+	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;
+}
 
-        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);
+int
+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 = ENOMEM;
 
-                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;
-                        }
+        priv = this->private;
 
-                        if (child_index == read_child) {
-                                local->cont.setattr.preop_buf  = *preop;
-                                local->cont.setattr.postop_buf = *postop;
-                        }
+        QUORUM_CHECK(fsetattr,out);
 
-                        local->success_count++;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame)
+                goto out;
 
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                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.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 ();
 
-        call_count = afr_frame_return (frame);
+	if (!local->xdata_req)
+		goto out;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
+        local->transaction.wind   = afr_fsetattr_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_fsetattr_unwind;
+
+        local->fd                 = fd_ref (fd);
+	local->inode = inode_ref (fd->inode);
+
+	local->op = GF_FOP_FSETATTR;
+
+        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_METADATA_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
         }
 
+	return 0;
+out:
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
+
+	AFR_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
 
-int32_t
-afr_setattr_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;
+/* {{{ setxattr */
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+int
+afr_setxattr_unwind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
 
-        local->call_count = call_count;
+        local = frame->local;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[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);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+	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_setattr_done (call_frame_t *frame, xlator_t *this)
+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;
+        return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+				      NULL, NULL, xdata);
+}
 
-        local = frame->local;
 
-        local->transaction.unwind (frame, this);
+int
+afr_setxattr_wind (call_frame_t *frame, xlator_t *this, int subvol)
+{
+        afr_local_t      *local         = NULL;
+        afr_private_t    *priv          = NULL;
 
-        AFR_STACK_DESTROY (frame);
+        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;
 }
 
 
 int
-afr_setattr (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, struct iatt *buf, int32_t valid)
+afr_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+	      int32_t flags, dict_t *xdata)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
+        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;
+        int             ret               = -1;
+        int             op_errno          = EINVAL;
+
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.afr.*", dict,
+                                   op_errno, out);
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.afr.*", dict,
+                                   op_errno, out);
 
         priv = this->private;
 
+        QUORUM_CHECK(setxattr,out);
+
         transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
+        if (!transaction_frame)
                 goto out;
-        }
-
-        ALLOC_OR_GOTO (local, afr_local_t, out);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
                 goto out;
-        }
-
-        transaction_frame->local = local;
-
-        local->op_ret = -1;
 
-        local->cont.setattr.ino     = loc->inode->ino;
+        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 ();
 
-        local->cont.setattr.in_buf = *buf;
-        local->cont.setattr.valid  = valid;
+	if (!local->xdata_req)
+		goto out;
 
-        local->transaction.fop    = afr_setattr_wind;
-        local->transaction.done   = afr_setattr_done;
-        local->transaction.unwind = afr_setattr_unwind;
+        local->transaction.wind   = afr_setxattr_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_setxattr_unwind;
 
         loc_copy (&local->loc, loc);
+	local->inode = inode_ref (loc->inode);
 
         local->transaction.main_frame = frame;
         local->transaction.start   = LLONG_MAX - 1;
         local->transaction.len     = 0;
 
-        afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+	local->op = GF_FOP_SETXATTR;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (setattr, frame, op_ret, op_errno, NULL, NULL);
+        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);
+
+	AFR_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL);
+
         return 0;
 }
 
-/* {{{ fsetattr */
+/* {{{ fsetxattr */
+
 
 int
-afr_fsetattr_unwind (call_frame_t *frame, xlator_t *this)
+afr_fsetxattr_unwind (call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+        afr_local_t    *local         = NULL;
+        call_frame_t   *main_frame    = NULL;
 
         local = frame->local;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (fsetattr, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.fsetattr.preop_buf,
-                                  &local->cont.fsetattr.postop_buf);
-        }
+	main_frame = afr_transaction_detach_fop_frame (frame);
+        if (!main_frame)
+		return 0;
 
+	AFR_STACK_UNWIND (fsetxattr, main_frame, local->op_ret, local->op_errno,
+			  local->xdata_rsp);
         return 0;
 }
 
 
 int
-afr_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)
+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_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;
+        return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+				      NULL, NULL, xdata);
+}
+
+
+int
+afr_fsetxattr_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;
+        priv = this->private;
 
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+	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;
+}
 
-        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);
+int
+afr_fsetxattr (call_frame_t *frame, xlator_t *this,
+               fd_t *fd, 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          = ENOMEM;
 
-                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;
-                        }
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.afr.*", dict,
+                                   op_errno, out);
 
-                        if (child_index == read_child) {
-                                local->cont.fsetattr.preop_buf  = *preop;
-                                local->cont.fsetattr.postop_buf = *postop;
-                        }
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.afr.*", dict,
+                                   op_errno, out);
 
-                        local->success_count++;
+        priv = this->private;
 
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+        QUORUM_CHECK(fsetxattr,out);
 
-        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.fsetxattr.dict  = dict_ref (dict);
+        local->cont.fsetxattr.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->transaction.wind   = afr_fsetxattr_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_fsetxattr_unwind;
+
+        local->fd                 = fd_ref (fd);
+	local->inode = inode_ref (fd->inode);
+
+	local->op = GF_FOP_FSETXATTR;
+
+        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);
+
+	AFR_STACK_UNWIND (fsetxattr, frame, -1, op_errno, NULL);
         return 0;
 }
 
+/* }}} */
 
-int32_t
-afr_fsetattr_wind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
 
-        local = frame->local;
-        priv = this->private;
+/* {{{ removexattr */
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+int
+afr_removexattr_unwind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
 
-        local->call_count = call_count;
+        local = frame->local;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[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);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+	main_frame = afr_transaction_detach_fop_frame (frame);
+	if (!main_frame)
+		return 0;
 
+	AFR_STACK_UNWIND (removexattr, main_frame, local->op_ret, local->op_errno,
+			  local->xdata_rsp);
         return 0;
 }
 
 
 int
-afr_fsetattr_done (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;
+        return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+				      NULL, NULL, xdata);
+}
 
-        local = frame->local;
 
-        local->transaction.unwind (frame, this);
+int
+afr_removexattr_wind (call_frame_t *frame, xlator_t *this, int subvol)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
 
-        AFR_STACK_DESTROY (frame);
+	local = frame->local;
+	priv = this->private;
 
+	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_fsetattr (call_frame_t *frame, xlator_t *this,
-              fd_t *fd, struct iatt *buf, int32_t valid)
+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_ret   = -1;
-        int op_errno = 0;
+        afr_private_t   *priv              = NULL;
+        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);
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.afr.*",
+                                 name, op_errno, out);
+
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.afr.*",
+                                 name, op_errno, out);
 
         priv = this->private;
 
+        QUORUM_CHECK(removexattr,out);
+
         transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
+        if (!transaction_frame)
                 goto out;
-        }
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		goto out;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+        local->cont.removexattr.name = gf_strdup (name);
 
-        transaction_frame->local = local;
+	if (xdata)
+		local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        local->op_ret = -1;
+	if (!local->xdata_req)
+		goto out;
 
-        local->cont.fsetattr.ino     = fd->inode->ino;
+        local->transaction.wind   = afr_removexattr_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_removexattr_unwind;
 
-        local->cont.fsetattr.in_buf = *buf;
-        local->cont.fsetattr.valid  = valid;
+        loc_copy (&local->loc, loc);
+	local->inode = inode_ref (loc->inode);
 
-        local->transaction.fop    = afr_fsetattr_wind;
-        local->transaction.done   = afr_fsetattr_done;
-        local->transaction.unwind = afr_fsetattr_unwind;
-
-        local->fd                 = fd_ref (fd);
+	local->op = GF_FOP_REMOVEXATTR;
 
         local->transaction.main_frame = frame;
         local->transaction.start   = LLONG_MAX - 1;
         local->transaction.len     = 0;
 
-        afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
+        }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (fsetattr, frame, op_ret, op_errno, NULL, NULL);
-        }
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
 
+	AFR_STACK_UNWIND (removexattr, frame, -1, op_errno, NULL);
         return 0;
 }
 
-
-/* {{{ setxattr */
-
-
+/* ffremovexattr */
 int
-afr_setxattr_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;
 
         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);
+	main_frame = afr_transaction_detach_fop_frame (frame);
+        if (!main_frame)
+		return 0;
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (setxattr, main_frame,
-                                  local->op_ret, local->op_errno)
-                        }
+	AFR_STACK_UNWIND (fremovexattr, 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)
+afr_fremovexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int call_count  = -1;
-        int need_unwind = 0;
-
-        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);
+	return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+				      NULL, NULL, xdata);
+}
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
 
-        call_count = afr_frame_return (frame);
+int
+afr_fremovexattr_wind (call_frame_t *frame, xlator_t *this, int subvol)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+        local = frame->local;
+        priv = this->private;
 
+	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;
 }
 
 
 int
-afr_setxattr_wind (call_frame_t *frame, xlator_t *this)
+afr_fremovexattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
+		  const char *name, dict_t *xdata)
 {
+	afr_private_t *priv = NULL;
         afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
+        call_frame_t *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = ENOMEM;
 
-        local = frame->local;
-        priv = this->private;
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.afr.*",
+                                 name, op_errno, out);
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.afr.*",
+                                 name, op_errno, out);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+	priv = this->private;
 
-        local->call_count = call_count;
+        QUORUM_CHECK(fremovexattr, out);
 
-        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);
-
-                        if (!--call_count)
-                                break;
-                }
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame)
+                goto out;
+
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+                goto out;
+
+        local->cont.removexattr.name = gf_strdup (name);
+	if (xdata)
+		local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
+
+	if (!local->xdata_req)
+		goto out;
+
+        local->transaction.wind   = afr_fremovexattr_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_fremovexattr_unwind;
+
+        local->fd = fd_ref (fd);
+	local->inode = inode_ref (fd->inode);
+
+	local->op = GF_FOP_FREMOVEXATTR;
+
+        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);
+
+	AFR_STACK_UNWIND (fremovexattr, frame, -1, op_errno, NULL);
+
         return 0;
 }
 
 
 int
-afr_setxattr_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);
+        local = frame->local;
 
-        AFR_STACK_DESTROY (frame);
+	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_setxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, dict_t *dict, int32_t flags)
+afr_fallocate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                        struct iatt *postbuf, dict_t *xdata)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
-        int ret = -1;
-        int op_ret   = -1;
-        int op_errno = 0;
+        return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+				      prebuf, postbuf, xdata);
+}
+
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, 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;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+	STACK_WIND_COOKIE (frame, afr_fallocate_wind_cbk, (void *) (long) subvol,
+			   priv->children[subvol],
+			   priv->children[subvol]->fops->fallocate,
+			   local->fd, local->cont.fallocate.mode,
+			   local->cont.fallocate.offset,
+			   local->cont.fallocate.len, local->xdata_req);
+        return 0;
+}
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto 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)
+{
+	afr_private_t *priv = NULL;
+        call_frame_t *transaction_frame = NULL;
+        afr_local_t *local = NULL;
+        int ret = -1;
+        int op_errno = ENOMEM;
+
+	priv = this->private;
+
+        QUORUM_CHECK(fallocate,out);
 
         transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
+        if (!transaction_frame)
                 goto out;
-        }
 
-        transaction_frame->local = local;
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		goto out;
 
-        local->op_ret = -1;
+        local->cont.fallocate.mode = mode;
+        local->cont.fallocate.offset  = offset;
+        local->cont.fallocate.len = len;
 
-        local->cont.setxattr.dict  = dict_ref (dict);
-        local->cont.setxattr.flags = flags;
+        local->fd = fd_ref (fd);
+	local->inode = inode_ref (fd->inode);
 
-        local->transaction.fop    = afr_setxattr_wind;
-        local->transaction.done   = afr_setxattr_done;
-        local->transaction.unwind = afr_setxattr_unwind;
+	if (xdata)
+		local->xdata_req = dict_copy_with_ref (xdata, NULL);
+	else
+		local->xdata_req = dict_new ();
 
-        loc_copy (&local->loc, loc);
+	if (!local->xdata_req)
+		goto out;
+
+        local->op = GF_FOP_FALLOCATE;
+
+        local->transaction.wind   = afr_fallocate_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_fallocate_unwind;
 
         local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
+
+        local->transaction.start   = local->cont.fallocate.offset;
         local->transaction.len     = 0;
 
-        afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        afr_fix_open (fd, this);
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (setxattr, frame, op_ret, op_errno);
+        ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
         }
 
+	return 0;
+out:
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
+
+	AFR_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;
 
         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);
+	main_frame = afr_transaction_detach_fop_frame (frame);
+        if (!main_frame)
+		return 0;
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (removexattr, main_frame,
-                                  local->op_ret, local->op_errno)
-                        }
+	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)
+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;
-
-        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);
+	return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+				      prebuf, postbuf, xdata);
+}
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
 
-        call_count = afr_frame_return (frame);
+int
+afr_discard_wind (call_frame_t *frame, xlator_t *this, int subvol)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+        local = frame->local;
+        priv = this->private;
 
+	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;
 }
 
 
-int32_t
-afr_removexattr_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_discard (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             size_t len, dict_t *xdata)
 {
+        afr_private_t *priv  = NULL;
         afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
+        call_frame_t *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = ENOMEM;
 
-        local = frame->local;
         priv = this->private;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+        QUORUM_CHECK(discard, out);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+	transaction_frame = copy_frame (frame);
+	if (!transaction_frame)
+		goto out;
 
-        local->call_count = call_count;
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		goto out;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[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);
-
-                        if (!--call_count)
-                                break;
-                }
+        local->cont.discard.offset  = offset;
+        local->cont.discard.len = len;
+
+        local->fd = fd_ref (fd);
+	local->inode = inode_ref (fd->inode);
+
+	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_DISCARD;
+
+        local->transaction.wind   = afr_discard_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_discard_unwind;
+
+        local->transaction.main_frame = frame;
+
+        local->transaction.start   = local->cont.discard.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;
         }
 
+	return 0;
+out:
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
+
+	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);
+        local = frame->local;
 
-        AFR_STACK_DESTROY (frame);
+	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)
+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_ret   = -1;
-        int op_errno = 0;
+	return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+				      prebuf, postbuf, xdata);
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (loc, 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;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
+	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;
+}
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+int
+afr_zerofill (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             size_t len, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+        afr_local_t *local = NULL;
+        call_frame_t *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = ENOMEM;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+        priv = this->private;
 
-        transaction_frame->local = local;
+        QUORUM_CHECK(discard, out);
 
-        local->op_ret = -1;
+	transaction_frame = copy_frame (frame);
+	if (!transaction_frame)
+		goto out;
 
-        local->cont.removexattr.name = gf_strdup (name);
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		goto out;
 
-        local->transaction.fop    = afr_removexattr_wind;
-        local->transaction.done   = afr_removexattr_done;
-        local->transaction.unwind = afr_removexattr_unwind;
+        local->cont.zerofill.offset  = offset;
+        local->cont.zerofill.len = len;
 
-        loc_copy (&local->loc, loc);
+        local->fd = fd_ref (fd);
+	local->inode = inode_ref (fd->inode);
+
+	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_ZEROFILL;
+
+        local->transaction.wind   = afr_zerofill_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_zerofill_unwind;
 
         local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
 
-        afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        local->transaction.start   = local->cont.discard.offset;
+        local->transaction.len     = len;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (removexattr, frame, op_ret, op_errno);
+        afr_fix_open (fd, this);
+
+        ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (ret < 0) {
+		op_errno = -ret;
+		goto out;
         }
 
+	return 0;
+out:
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
+
+	AFR_STACK_UNWIND (zerofill, 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 475898722..7b1fc5528 100644
--- a/xlators/cluster/afr/src/afr-inode-write.h
+++ b/xlators/cluster/afr/src/afr-inode-write.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  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 __INODE_WRITE_H__
@@ -22,51 +13,70 @@
 
 int32_t
 afr_chmod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode);
+	   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);
+	   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);
+	    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);
+	    fd_t *fd, mode_t mode, dict_t *xdata);
 
 int32_t
-afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, 
+afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
 	    struct iovec *vector, int32_t count, off_t offset,
-            struct iobref *iobref);
+            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);
+	      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);
+	       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]);
+	     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);
+             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);
+              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);
+              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);
 
 int32_t
 afr_removexattr (call_frame_t *frame, xlator_t *this,
-		 loc_t *loc, const char *name);
+		 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);
 
+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);
 #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 fc54682d1..a2a758f35 100644
--- a/xlators/cluster/afr/src/afr-lk-common.c
+++ b/xlators/cluster/afr/src/afr-lk-common.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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 "dict.h"
@@ -31,8 +22,69 @@
 #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_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index);
+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;
 
@@ -57,12 +109,13 @@ afr_set_lock_number (call_frame_t *frame, xlator_t *this)
 }
 
 void
-afr_set_lk_owner (call_frame_t *frame, xlator_t *this)
+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)frame->root);
-        frame->root->lk_owner = (uint64_t) (unsigned long)frame->root;
+                (unsigned long long) (unsigned long)lk_owner);
+
+        set_lk_owner_from_ptr (&frame->root->lk_owner, lk_owner);
 }
 
 static int
@@ -88,8 +141,7 @@ is_afr_lock_selfheal (afr_local_t *local)
 }
 
 int32_t
-internal_lock_count (call_frame_t *frame, xlator_t *this,
-                     afr_fd_ctx_t *fd_ctx)
+internal_lock_count (call_frame_t *frame, xlator_t *this)
 {
         afr_local_t   *local = NULL;
         afr_private_t *priv  = NULL;
@@ -99,17 +151,9 @@ internal_lock_count (call_frame_t *frame, xlator_t *this,
         local = frame->local;
         priv  = this->private;
 
-        if (fd_ctx) {
-                GF_ASSERT (local->fd);
-                for (i = 0; i < priv->child_count; i++) {
-                        if (local->child_up[i] && fd_ctx->opened_on[i])
-                                ++call_count;
-                }
-        } else {
-                for (i = 0; i < priv->child_count; i++) {
-                        if (local->child_up[i])
-                                ++call_count;
-                }
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i])
+                        ++call_count;
         }
 
         return call_count;
@@ -117,7 +161,7 @@ internal_lock_count (call_frame_t *frame, xlator_t *this,
 
 static void
 afr_print_inodelk (char *str, int size, int cmd,
-                   struct gf_flock *flock, uint64_t owner)
+                   struct gf_flock *flock, gf_lkowner_t *owner)
 {
         char *cmd_str = NULL;
         char *type_str = NULL;
@@ -165,11 +209,11 @@ afr_print_inodelk (char *str, int size, int cmd,
         }
 
         snprintf (str, size, "lock=INODELK, cmd=%s, type=%s, "
-                  "start=%llu, len=%llu, pid=%llu, lk-owner=%llu",
+                  "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,
-                  (unsigned long long) owner);
+                  lkowner_utoa (owner));
 
 }
 
@@ -185,11 +229,11 @@ afr_print_lockee (char *str, int size, loc_t *loc, fd_t *fd,
 
 void
 afr_print_entrylk (char *str, int size, const char *basename,
-                   uint64_t owner)
+                   gf_lkowner_t *owner)
 {
-        snprintf (str, size, "Basename=%s, lk-owner=%llu",
+        snprintf (str, size, "Basename=%s, lk-owner=%s",
                   basename ? basename : "<nul>",
-                  (unsigned long long)owner);
+                  lkowner_utoa (owner));
 }
 
 static void
@@ -243,27 +287,20 @@ afr_set_lock_call_type (afr_lock_call_type_t lock_call_type,
 }
 
 static void
-afr_trace_inodelk_out (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
+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)
 {
-        xlator_t            *this     = NULL;
         afr_internal_lock_t *int_lock = NULL;
         afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
 
         char lockee[256];
         char lock_call_type_str[256];
         char verdict[16];
 
-        this     = THIS;
         local    = frame->local;
         int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        if (!priv->inodelk_trace) {
-                return;
-        }
 
         afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index);
 
@@ -272,39 +309,31 @@ afr_trace_inodelk_out (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
         afr_print_verdict (op_ret, op_errno, verdict);
 
         gf_log (this->name, GF_LOG_INFO,
-                "[%s %s] [%s] Lockee={%s} Number={%llu}",
+                "[%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,
-                lockee,
+                verdict, lkowner_utoa (&frame->root->lk_owner), lockee,
                 (unsigned long long) int_lock->lock_number);
 
 }
 
 static void
-afr_trace_inodelk_in (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
+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)
 {
-        xlator_t            *this     = NULL;
         afr_local_t         *local    = NULL;
         afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
 
         char lock[256];
         char lockee[256];
         char lock_call_type_str[256];
 
-        this     = THIS;
         local    = frame->local;
         int_lock = &local->internal_lock;
-        priv     = this->private;
 
-        if (!priv->inodelk_trace) {
-                return;
-        }
-
-        afr_print_inodelk (lock, 256, cmd, flock, frame->root->lk_owner);
+        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);
@@ -319,20 +348,21 @@ afr_trace_inodelk_in (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
 }
 
 static void
-afr_trace_entrylk_in (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
+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 child_index)
+                      int32_t cookie)
 {
-        xlator_t            *this     = NULL;
         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];
 
-        this     = THIS;
         local    = frame->local;
         int_lock = &local->internal_lock;
         priv     = this->private;
@@ -340,36 +370,41 @@ afr_trace_entrylk_in (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
         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, &local->loc, local->fd, child_index);
+        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}",
+                "[%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);
+                (unsigned long long) int_lock->lock_number,
+                cookie);
 }
 
 static void
-afr_trace_entrylk_out (call_frame_t *frame, 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 child_index)
+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)
 {
-        xlator_t            *this     = NULL;
         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];
 
-        this     = THIS;
         local    = frame->local;
         int_lock = &local->internal_lock;
         priv     = this->private;
@@ -377,20 +412,25 @@ afr_trace_entrylk_out (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
         if (!priv->entrylk_trace) {
                 return;
         }
+        lockee_no = cookie / priv->child_count;
+        child_index = cookie % priv->child_count;
 
-        afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index);
+        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}",
+                "[%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);
+                (unsigned long long) int_lock->lock_number,
+                cookie);
 
 }
 
@@ -443,6 +483,47 @@ is_afr_lock_transaction (afr_local_t *local)
         return ret;
 }
 
+int
+afr_init_entry_lockee (afr_entry_lockee_t *lockee, afr_local_t *local,
+                       loc_t *loc, char *basename, int child_count)
+{
+        int     ret     = -1;
+
+        loc_copy (&lockee->loc, loc);
+        lockee->basename        = (basename)? gf_strdup (basename): NULL;
+        if (basename && !lockee->basename)
+                goto out;
+
+        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;
+
+        ret = 0;
+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;
+}
+
 static int
 initialize_entrylk_variables (call_frame_t *frame, xlator_t *this)
 {
@@ -460,8 +541,13 @@ initialize_entrylk_variables (call_frame_t *frame, xlator_t *this)
         int_lock->lock_op_ret        = -1;
         int_lock->lock_op_errno      = 0;
 
-        for (i = 0; i < priv->child_count; i++) {
-                int_lock->entry_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;
@@ -473,37 +559,37 @@ 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;
+        afr_inodelk_t       *inodelk  = NULL;
 
         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;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
 
-        for (i = 0; i < priv->child_count; i++) {
-                int_lock->inode_locked_nodes[i] = 0;
-        }
+        inodelk->lock_count    = 0;
+        int_lock->lk_attempted_count = 0;
+        int_lock->lock_op_ret   = -1;
+        int_lock->lock_op_errno = 0;
+
+        memset (inodelk->locked_nodes, 0,
+                sizeof (*inodelk->locked_nodes) * priv->child_count);
+        memset (int_lock->locked_nodes, 0,
+                sizeof (*int_lock->locked_nodes) * priv->child_count);
 
         return 0;
 }
 
-loc_t *
-lower_path (loc_t *l1, const char *b1, loc_t *l2, const char *b2)
+int
+afr_lockee_locked_nodes_count (afr_internal_lock_t *int_lock)
 {
-        int ret = 0;
-
-        ret = strcmp (l1->path, l2->path);
+        int call_count  = 0;
+        int i           = 0;
 
-        if (ret == 0)
-                ret = strcmp (b1, b2);
+        for (i = 0; i < int_lock->lockee_count; i++)
+                call_count += int_lock->lockee[i].locked_count;
 
-        if (ret <= 0)
-                return l1;
-        else
-                return l2;
+        return call_count;
 }
 
 int
@@ -524,7 +610,7 @@ afr_locked_nodes_count (unsigned char *locked_nodes, int child_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)
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
         afr_local_t         *local    = NULL;
         afr_internal_lock_t *int_lock = NULL;
@@ -550,23 +636,37 @@ afr_unlock_common_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 static int32_t
 afr_unlock_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno)
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
+        afr_local_t         *local = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk = NULL;
+        int32_t             child_index = (long)cookie;
+        afr_private_t       *priv = NULL;
 
         local = frame->local;
+        int_lock = &local->internal_lock;
 
-        afr_trace_inodelk_out (frame, AFR_INODELK_TRANSACTION,
+        AFR_TRACE_INODELK_OUT (frame, this, AFR_INODELK_TRANSACTION,
                                AFR_UNLOCK_OP, NULL, op_ret,
-                               op_errno, (long) cookie);
+                               op_errno, child_index);
+
+        priv = this->private;
 
         if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: unlock failed %s",
-                        local->loc.path, strerror (op_errno));
+                gf_log (this->name, GF_LOG_INFO, "%s: unlock failed on subvolume %s "
+                        "with lock owner %s", local->loc.path,
+                        priv->children[child_index]->name,
+                        lkowner_utoa (&frame->root->lk_owner));
         }
 
-        afr_unlock_common_cbk (frame, cookie, this, op_ret, op_errno);
+
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+        inodelk->locked_nodes[child_index] &= LOCKED_NO;
+        if (local->transaction.eager_lock)
+                local->transaction.eager_lock[child_index] = 0;
+
+        afr_unlock_common_cbk (frame, cookie, this, op_ret, op_errno, xdata);
 
         return 0;
 
@@ -576,21 +676,30 @@ static int
 afr_unlock_inodelk (call_frame_t *frame, xlator_t *this)
 {
         afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
         afr_local_t         *local    = NULL;
         afr_private_t       *priv     = NULL;
         struct gf_flock flock = {0,};
+        struct gf_flock full_flock = {0,};
+        struct gf_flock *flock_use = NULL;
         int call_count = 0;
         int i = 0;
+        int piggyback = 0;
+        afr_fd_ctx_t        *fd_ctx      = NULL;
+
 
         local    = frame->local;
         int_lock = &local->internal_lock;
         priv     = this->private;
 
-        flock.l_start = int_lock->lk_flock.l_start;
-        flock.l_len   = int_lock->lk_flock.l_len;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+
+        flock.l_start = inodelk->flock.l_start;
+        flock.l_len   = inodelk->flock.l_len;
         flock.l_type  = F_UNLCK;
 
-        call_count = afr_locked_nodes_count (int_lock->inode_locked_nodes,
+        full_flock.l_type = F_UNLCK;
+        call_count = afr_locked_nodes_count (inodelk->locked_nodes,
                                              priv->child_count);
 
         int_lock->lk_call_count = call_count;
@@ -602,55 +711,107 @@ afr_unlock_inodelk (call_frame_t *frame, xlator_t *this)
                 goto out;
         }
 
+        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) {
-                        if (local->fd) {
-                                afr_trace_inodelk_in (frame, 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->finodelk,
-                                                   this->name, local->fd,
-                                                   F_SETLK, &flock);
+                if ((inodelk->locked_nodes[i] & LOCKED_YES) != LOCKED_YES)
+                        continue;
 
-                                if (!--call_count)
-                                        break;
+                if (local->fd) {
+                        flock_use = &flock;
+                        if (!local->transaction.eager_lock[i]) {
+                                goto wind;
+                        }
 
-                        } else {
-                                afr_trace_inodelk_in (frame, AFR_INODELK_TRANSACTION,
-                                                      AFR_UNLOCK_OP, &flock, F_SETLK, i);
+                        piggyback = 0;
 
-                                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);
+                        LOCK (&local->fd->lock);
+                        {
+                                if (fd_ctx->lock_piggyback[i]) {
+                                        fd_ctx->lock_piggyback[i]--;
+                                        piggyback = 1;
+                                } else {
+                                        fd_ctx->lock_acquired[i]--;
+                                }
+                        }
+                        UNLOCK (&local->fd->lock);
 
+                        if (piggyback) {
+                                afr_unlock_inodelk_cbk (frame, (void *) (long) i,
+                                                        this, 1, 0, NULL);
                                 if (!--call_count)
                                         break;
-
+                                continue;
                         }
 
-                }
+                        flock_use = &full_flock;
+                wind:
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_TRANSACTION,
+                                              AFR_UNLOCK_OP, flock_use, F_SETLK,
+                                              i);
 
-        }
+                        STACK_WIND_COOKIE (frame, afr_unlock_inodelk_cbk,
+                                           (void *) (long)i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->finodelk,
+                                           int_lock->domain, local->fd,
+                                           F_SETLK, flock_use, NULL);
 
+                        if (!--call_count)
+                                break;
+
+                } else {
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_TRANSACTION,
+                                              AFR_UNLOCK_OP, &flock, F_SETLK, i);
+
+                        STACK_WIND_COOKIE (frame, afr_unlock_inodelk_cbk,
+                                           (void *) (long)i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->inodelk,
+                                           int_lock->domain, &local->loc,
+                                           F_SETLK, &flock, NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
 out:
         return 0;
 }
 
 static int32_t
 afr_unlock_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno)
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_trace_entrylk_out (frame, AFR_ENTRYLK_TRANSACTION,
-                               AFR_UNLOCK_OP, NULL, op_ret,
-                               op_errno, (long) cookie);
+        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;
 
-        afr_unlock_common_cbk (frame, cookie, this, op_ret, op_errno);
+        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);
 
         return 0;
 }
@@ -658,24 +819,22 @@ afr_unlock_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 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;
-        const char          *basename = NULL;
-        loc_t               *loc      = NULL;
-        int call_count = 0;
-        int i = -1;
+        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;
 
-        basename = int_lock->lk_basename;
-        if (int_lock->lk_loc)
-                loc = int_lock->lk_loc;
+        call_count = afr_lockee_locked_nodes_count (int_lock);
 
-        call_count = afr_locked_nodes_count (int_lock->entry_locked_nodes,
-                                             priv->child_count);
         int_lock->lk_call_count = call_count;
 
         if (!call_count){
@@ -685,18 +844,23 @@ afr_unlock_entrylk (call_frame_t *frame, xlator_t *this)
                 goto out;
         }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (int_lock->entry_locked_nodes[i] & LOCKED_YES) {
-                        afr_trace_entrylk_in (frame, AFR_ENTRYLK_NB_TRANSACTION,
-                                              AFR_UNLOCK_OP, basename, i);
+        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[i],
-                                           priv->children[i]->fops->entrylk,
-                                           this->name,
-                                           loc, basename,
-                                           ENTRYLK_UNLOCK, ENTRYLK_WRLCK);
+                                           priv->children[index],
+                                           priv->children[index]->fops->entrylk,
+                                           int_lock->domain,
+                                           &int_lock->lockee[lockee_no].loc,
+                                           int_lock->lockee[lockee_no].basename,
+                                           ENTRYLK_UNLOCK, ENTRYLK_WRLCK, NULL);
 
                         if (!--call_count)
                                 break;
@@ -710,17 +874,21 @@ out:
 
 static int32_t
 afr_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-              int32_t op_ret, int32_t op_errno)
+              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 done = 0;
-        int child_index = (long) cookie;
+        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;
-        priv     = this->private;
+
+        child_index = ((int)cky) % priv->child_count;
+        lockee_no   = ((int)cky) / priv->child_count;
 
         LOCK (&frame->lock);
         {
@@ -732,13 +900,13 @@ afr_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                         "please load features/locks xlator on server");
                                 local->op_ret = op_ret;
                                 int_lock->lock_op_ret = op_ret;
-                                done = 1;
                         }
 
-                        local->child_up[child_index] = 0;
                         local->op_errno              = op_errno;
                         int_lock->lock_op_errno      = op_errno;
                 }
+
+		int_lock->lk_attempted_count++;
         }
         UNLOCK (&frame->lock);
 
@@ -747,11 +915,17 @@ afr_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 afr_unlock (frame, this);
         } else {
                 if (op_ret == 0) {
-                        int_lock->locked_nodes[child_index]
-                                |= LOCKED_YES;
-                        int_lock->lock_count++;
+                        if (local->transaction.type == AFR_ENTRY_TRANSACTION ||
+                            local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
+                                int_lock->lockee[lockee_no].locked_nodes[child_index] |= LOCKED_YES;
+                                int_lock->lockee[lockee_no].locked_count++;
+                                int_lock->entrylk_lock_count++;
+                        } else {
+                                int_lock->locked_nodes[child_index] |= LOCKED_YES;
+                                int_lock->lock_count++;
+                        }
                 }
-                afr_lock_blocking (frame, this, child_index + 1);
+                afr_lock_blocking (frame, this, cky + 1);
         }
 
         return 0;
@@ -759,104 +933,26 @@ afr_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 static int32_t
 afr_blocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno)
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_trace_inodelk_out (frame, AFR_INODELK_TRANSACTION,
+        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);
+        afr_lock_cbk (frame, cookie, this, op_ret, op_errno, xdata);
         return 0;
 
 }
 
 static int32_t
-afr_lock_lower_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno)
-{
-        afr_internal_lock_t *int_lock    = NULL;
-        afr_private_t       *priv        = NULL;
-        afr_local_t         *local       = NULL;
-        loc_t               *lower       = NULL;
-        loc_t               *higher      = NULL;
-        const char          *lower_name  = NULL;
-        const char          *higher_name = NULL;
-        int child_index = (long) cookie;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        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;
-                        }
-
-                        local->child_up[child_index] = 0;
-                        local->op_errno = op_errno;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (op_ret != 0) {
-                afr_unlock (frame, this);
-                goto out;
-        } else {
-                int_lock->lower_locked_nodes[child_index] |= LOCKED_LOWER;
-                int_lock->lock_count++;
-        }
-
-        /* The lower path has been locked. Now lock the higher path */
-
-        lower = lower_path (&local->transaction.parent_loc,
-                            local->transaction.basename,
-                            &local->transaction.new_parent_loc,
-                            local->transaction.new_basename);
-
-        lower_name = (lower == &local->transaction.parent_loc ?
-                      local->transaction.basename :
-                      local->transaction.new_basename);
-
-        higher = (lower == &local->transaction.parent_loc ?
-                  &local->transaction.new_parent_loc :
-                  &local->transaction.parent_loc);
-
-        higher_name = (higher == &local->transaction.parent_loc ?
-                       local->transaction.basename :
-                       local->transaction.new_basename);
-
-        afr_trace_entrylk_in (frame, AFR_ENTRYLK_TRANSACTION,
-                              AFR_LOCK_OP, higher_name, child_index);
-
-
-        STACK_WIND_COOKIE (frame, afr_lock_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->entrylk,
-                           this->name, higher, higher_name,
-                           ENTRYLK_LOCK, ENTRYLK_WRLCK);
-
-out:
-        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)
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_trace_entrylk_out (frame, AFR_ENTRYLK_TRANSACTION,
+        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);
+        afr_lock_cbk (frame, cookie, this, op_ret, op_errno, xdata);
         return 0;
 }
 
@@ -864,6 +960,7 @@ static int
 afr_copy_locked_nodes (call_frame_t *frame, xlator_t *this)
 {
         afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
         afr_local_t         *local    = NULL;
         afr_private_t       *priv     = NULL;
 
@@ -874,18 +971,16 @@ afr_copy_locked_nodes (call_frame_t *frame, xlator_t *this)
         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;
+                inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+                memcpy (inodelk->locked_nodes, int_lock->locked_nodes,
+                        sizeof (*inodelk->locked_nodes) * priv->child_count);
+                inodelk->lock_count = int_lock->lock_count;
                 break;
 
         case AFR_ENTRY_RENAME_TRANSACTION:
         case AFR_ENTRY_TRANSACTION:
-                memcpy (int_lock->entry_locked_nodes,
-                        int_lock->locked_nodes,
-                        priv->child_count);
-                int_lock->entrylk_lock_count = int_lock->lock_count;
+                /*entrylk_count is being used in both non-blocking and blocking
+                 * modes */
                 break;
         }
 
@@ -893,28 +988,67 @@ afr_copy_locked_nodes (call_frame_t *frame, xlator_t *this)
 
 }
 
+static inline gf_boolean_t
+afr_is_entrylk (afr_internal_lock_t *int_lock,
+                afr_transaction_type trans_type)
+{
+        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)) {
+
+                is_entrylk = _gf_true;
+
+        } else {
+                is_entrylk = _gf_false;
+        }
+
+        return is_entrylk;
+}
+
+static gf_boolean_t
+_is_lock_wind_needed (afr_local_t *local, int child_index)
+{
+        if (!local->child_up[child_index])
+                return _gf_false;
+
+        return _gf_true;
+}
+
 int
-afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index)
+afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
 {
         afr_internal_lock_t *int_lock    = NULL;
+        afr_inodelk_t       *inodelk     = NULL;
         afr_local_t         *local       = NULL;
         afr_private_t       *priv        = NULL;
-        afr_fd_ctx_t        *fd_ctx      = NULL;
-        loc_t               *lower       = NULL;
-        loc_t               *higher      = NULL;
-        const char          *lower_name  = NULL;
-        const char          *higher_name = 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;
+        local         = frame->local;
+        int_lock      = &local->internal_lock;
+        priv          = this->private;
+        child_index   = cookie % priv->child_count;
+        lockee_no     = cookie / priv->child_count;
+        is_entrylk    = afr_is_entrylk (int_lock, local->transaction.type);
 
-        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 (!is_entrylk) {
+                inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+                flock.l_start = inodelk->flock.l_start;
+                flock.l_len   = inodelk->flock.l_len;
+                flock.l_type  = inodelk->flock.l_type;
+        }
 
         if (local->fd) {
                 ret = fd_ctx_get (local->fd, this, &ctx);
@@ -933,46 +1067,26 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index)
 
                         return 0;
                 }
-
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                /* skip over children that or down
-                   or don't have the fd open */
-
-                while ((child_index < priv->child_count)
-                       && (!local->child_up[child_index]
-                           || !fd_ctx->opened_on[child_index]))
-
-                        child_index++;
-        } else {
-                /* skip over children that are down */
-                while ((child_index < priv->child_count)
-                       && !local->child_up[child_index])
-                        child_index++;
         }
 
-        if ((child_index == priv->child_count) &&
-            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;
+        if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
+                if ((is_entrylk && int_lock->entrylk_lock_count == 0) ||
+                    (!is_entrylk && int_lock->lock_count == 0)) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "unable to lock on even one child");
 
-                afr_copy_locked_nodes (frame, this);
+                        local->op_ret           = -1;
+                        int_lock->lock_op_ret   = -1;
 
-                afr_unlock(frame, this);
+                        afr_copy_locked_nodes (frame, this);
 
-                return 0;
+                        afr_unlock(frame, this);
 
+                        return 0;
+                }
         }
 
-        if ((child_index == priv->child_count)
-            || (int_lock->lock_count ==
-                afr_up_children_count (priv->child_count,
-                                       local->child_up))) {
-
+        if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
                 /* we're done locking */
 
                 gf_log (this->name, GF_LOG_DEBUG,
@@ -985,12 +1099,18 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index)
                 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, AFR_INODELK_TRANSACTION,
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_TRANSACTION,
                                               AFR_LOCK_OP, &flock, F_SETLKW,
                                               child_index);
 
@@ -998,11 +1118,12 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index)
                                            (void *) (long) child_index,
                                            priv->children[child_index],
                                            priv->children[child_index]->fops->finodelk,
-                                           this->name, local->fd,
-                                           F_SETLKW, &flock);
+                                           int_lock->domain, local->fd,
+                                           F_SETLKW, &flock, NULL);
 
                 } else {
-                        afr_trace_inodelk_in (frame, AFR_INODELK_TRANSACTION,
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_TRANSACTION,
                                               AFR_LOCK_OP, &flock, F_SETLKW,
                                               child_index);
 
@@ -1010,79 +1131,50 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index)
                                            (void *) (long) child_index,
                                            priv->children[child_index],
                                            priv->children[child_index]->fops->inodelk,
-                                           this->name, &local->loc,
-                                           F_SETLKW, &flock);
+                                           int_lock->domain, &local->loc,
+                                           F_SETLKW, &flock, NULL);
                 }
 
                 break;
 
         case AFR_ENTRY_RENAME_TRANSACTION:
-        {
-                lower = lower_path (&local->transaction.parent_loc,
-                                    local->transaction.basename,
-                                    &local->transaction.new_parent_loc,
-                                    local->transaction.new_basename);
-
-                lower_name = (lower == &local->transaction.parent_loc ?
-                              local->transaction.basename :
-                              local->transaction.new_basename);
-
-                higher = (lower == &local->transaction.parent_loc ?
-                          &local->transaction.new_parent_loc :
-                          &local->transaction.parent_loc);
-
-                higher_name = (higher == &local->transaction.parent_loc ?
-                               local->transaction.basename :
-                               local->transaction.new_basename);
-
-                afr_trace_entrylk_in (frame, AFR_ENTRYLK_TRANSACTION,
-                                      AFR_LOCK_OP, lower_name, child_index);
-
-
-                STACK_WIND_COOKIE (frame, afr_lock_lower_cbk,
-                                   (void *) (long) child_index,
-                                   priv->children[child_index],
-                                   priv->children[child_index]->fops->entrylk,
-                                   this->name, lower, lower_name,
-                                   ENTRYLK_LOCK, ENTRYLK_WRLCK);
-
-                break;
-        }
-
         case AFR_ENTRY_TRANSACTION:
+                /*Accounting for child_index increments on 'down'
+                 *and 'fd-less' children */
+
                 if (local->fd) {
-                        afr_trace_entrylk_in (frame, AFR_ENTRYLK_TRANSACTION,
-                                              AFR_LOCK_OP, local->transaction.basename,
-                                              child_index);
+                        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) child_index,
+                                           (void *) (long) cookie,
                                            priv->children[child_index],
                                            priv->children[child_index]->fops->fentrylk,
-                                           this->name, local->fd,
-                                           local->transaction.basename,
-                                           ENTRYLK_LOCK, ENTRYLK_WRLCK);
+                                           int_lock->domain, local->fd,
+                                           int_lock->lockee[lockee_no].basename,
+                                           ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
                 } else {
-                        afr_trace_entrylk_in (frame, AFR_ENTRYLK_TRANSACTION,
+                        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) child_index,
+                                           (void *) (long) cookie,
                                            priv->children[child_index],
                                            priv->children[child_index]->fops->entrylk,
-                                           this->name,
-                                           &local->transaction.parent_loc,
-                                           local->transaction.basename,
-                                           ENTRYLK_LOCK, ENTRYLK_WRLCK);
+                                           int_lock->domain,
+                                           &int_lock->lockee[lockee_no].loc,
+                                           int_lock->lockee[lockee_no].basename,
+                                           ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
                 }
 
                 break;
         }
 
         return 0;
-
-
 }
 
 int32_t
@@ -1091,6 +1183,7 @@ 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;
@@ -1104,6 +1197,10 @@ afr_blocking_lock (call_frame_t *frame, xlator_t *this)
 
         case AFR_ENTRY_RENAME_TRANSACTION:
         case AFR_ENTRY_TRANSACTION:
+                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_entrylk_variables (frame, this);
                 break;
         }
@@ -1115,59 +1212,68 @@ afr_blocking_lock (call_frame_t *frame, xlator_t *this)
 
 static int32_t
 afr_nonblocking_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno)
+                             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 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;
-        priv     = this->private;
 
-        afr_trace_entrylk_out (frame, AFR_ENTRYLK_TRANSACTION,
-                               AFR_LOCK_OP, NULL, op_ret,
+        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);
-        {
-                call_count = --int_lock->lk_call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (op_ret < 0 ) {
-                if (op_errno == ENOSYS) {
+	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;
+				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++;
+		}
 
-                        local->child_up[child_index] = 0;
-                        int_lock->lock_op_errno      = op_errno;
-                        local->op_errno              = op_errno;
-                }
-        } else if (op_ret == 0) {
-                int_lock->entry_locked_nodes[child_index]
-                        |= LOCKED_YES;
-                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 successfull. Proceed to call FOP */
+                /* all locks successful. Proceed to call FOP */
                 if (int_lock->entrylk_lock_count ==
-                    afr_up_children_count (priv->child_count, local->child_up)) {
+                                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 successfull. Unlock and try locking
+                /* Not all locks were successful. Unlock and try locking
                    again, this time with serially blocking locks */
                 else {
                         gf_log (this->name, GF_LOG_TRACE,
@@ -1184,31 +1290,26 @@ afr_nonblocking_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 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;
-        const char          *basename = NULL;
-        loc_t               *loc      = NULL;
-        int32_t call_count = 0;
+        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;
-        uint64_t  ctx = 0;
-        int ret = 0;
 
         local    = frame->local;
         int_lock = &local->internal_lock;
         priv     = this->private;
 
+        copies = priv->child_count;
         initialize_entrylk_variables (frame, this);
 
-        basename = int_lock->lk_basename;
-        if (int_lock->lk_loc)
-                loc = int_lock->lk_loc;
-
         if (local->fd) {
-                ret = fd_ctx_get (local->fd, this, &ctx);
-
-                if (ret < 0) {
+                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);
@@ -1218,13 +1319,13 @@ afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this)
                         local->op_errno         = EINVAL;
                         int_lock->lock_op_errno = EINVAL;
 
+			afr_unlock (frame, this);
                         return -1;
                 }
 
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                call_count = internal_lock_count (frame, this, fd_ctx);
+                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,
@@ -1235,41 +1336,52 @@ afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this)
 
                 /* Send non-blocking entrylk 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] && fd_ctx->opened_on[i]) {
-                                afr_trace_entrylk_in (frame, AFR_ENTRYLK_NB_TRANSACTION,
-                                                      AFR_LOCK_OP, basename, i);
+                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[i],
-                                                   priv->children[i]->fops->fentrylk,
+                                                   priv->children[index],
+                                                   priv->children[index]->fops->fentrylk,
                                                    this->name, local->fd,
-                                                   basename,
-                                                   ENTRYLK_LOCK_NB, ENTRYLK_WRLCK);
+                                                   int_lock->lockee[lockee_no].basename,
+                                                   ENTRYLK_LOCK_NB, ENTRYLK_WRLCK,
+                                                   NULL);
+                                if (!--call_count)
+                                        break;
                         }
                 }
         } else {
-                GF_ASSERT (loc);
-
-                call_count = internal_lock_count (frame, this, NULL);
+                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 < priv->child_count; i++) {
-                        if (local->child_up[i]) {
-                                afr_trace_entrylk_in (frame, AFR_ENTRYLK_NB_TRANSACTION,
-                                                      AFR_LOCK_OP, basename, i);
+                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[i],
-                                                   priv->children[i]->fops->entrylk,
-                                                   this->name, loc, basename,
-                                                   ENTRYLK_LOCK_NB, ENTRYLK_WRLCK);
+                                                   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;
-
                         }
                 }
         }
@@ -1279,58 +1391,75 @@ out:
 
 int32_t
 afr_nonblocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno)
+                             int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
         afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
         afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = 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;
-        priv     = this->private;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
 
-        afr_trace_inodelk_out (frame, AFR_INODELK_NB_TRANSACTION,
+        AFR_TRACE_INODELK_OUT (frame, this, AFR_INODELK_NB_TRANSACTION,
                                AFR_LOCK_OP, NULL, op_ret,
                                op_errno, (long) cookie);
 
+	if (local->fd)
+		fd_ctx = afr_fd_ctx_get (local->fd, this);
+
         LOCK (&frame->lock);
         {
+		if (op_ret < 0) {
+			if (op_errno == ENOSYS) {
+				/* return ENOTSUP */
+				gf_log (this->name, GF_LOG_ERROR,
+					"subvolume does not support locking. "
+					"please load features/locks xlator on "
+					"server");
+				local->op_ret                = op_ret;
+				int_lock->lock_op_ret        = op_ret;
+				int_lock->lock_op_errno      = op_errno;
+				local->op_errno              = op_errno;
+			}
+			if (local->transaction.eager_lock)
+				local->transaction.eager_lock[child_index] = 0;
+		} else {
+			inodelk->locked_nodes[child_index] |= LOCKED_YES;
+			inodelk->lock_count++;
+
+			if (local->transaction.eager_lock &&
+			    local->transaction.eager_lock[child_index] &&
+			    local->fd) {
+				/* piggybacked */
+				if (op_ret == 1) {
+					/* piggybacked */
+				} else if (op_ret == 0) {
+					/* lock acquired from server */
+                                        fd_ctx->lock_acquired[child_index]++;
+				}
+			}
+		}
+
                 call_count = --int_lock->lk_call_count;
         }
         UNLOCK (&frame->lock);
 
-        if (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;
-                        local->child_up[child_index] = 0;
-                        int_lock->lock_op_errno      = op_errno;
-                        local->op_errno              = op_errno;
-                }
-        } else if (op_ret == 0) {
-                int_lock->inode_locked_nodes[child_index]
-                        |= LOCKED_YES;
-                int_lock->inodelk_lock_count++;
-        }
-
         if (call_count == 0) {
                 gf_log (this->name, GF_LOG_TRACE,
                         "Last inode locking reply received");
-                /* all locks successfull. Proceed to call FOP */
-                if (int_lock->inodelk_lock_count ==
-                    afr_up_children_count (priv->child_count, local->child_up)) {
+                /* all locks successful. Proceed to call FOP */
+                if (inodelk->lock_count == int_lock->lk_expected_count) {
                         gf_log (this->name, GF_LOG_TRACE,
                                 "All servers locked. Calling the cbk");
                         int_lock->lock_op_ret = 0;
                         int_lock->lock_cbk (frame, this);
                 }
-                /* Not all locks were successfull. Unlock and try locking
+                /* Not all locks were successful. Unlock and try locking
                    again, this time with serially blocking locks */
                 else {
                         gf_log (this->name, GF_LOG_TRACE,
@@ -1348,29 +1477,35 @@ int
 afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this)
 {
         afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
         afr_local_t         *local    = NULL;
         afr_private_t       *priv     = NULL;
         afr_fd_ctx_t        *fd_ctx   = NULL;
-        int32_t  call_count = 0;
-        uint64_t ctx        = 0;
-        int      i          = 0;
-        int      ret        = 0;
-        struct gf_flock flock = {0,};
+        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;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+
+        flock.l_start = inodelk->flock.l_start;
+        flock.l_len   = inodelk->flock.l_len;
+        flock.l_type  = inodelk->flock.l_type;
+
+        full_flock.l_type = inodelk->flock.l_type;
 
         initialize_inodelk_variables (frame, this);
 
         if (local->fd) {
-                ret = fd_ctx_get (local->fd, this, &ctx);
-
-                if (ret < 0) {
+                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);
@@ -1380,14 +1515,14 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this)
                         local->op_errno         = EINVAL;
                         int_lock->lock_op_errno = EINVAL;
 
+			afr_unlock (frame, this);
                         ret = -1;
                         goto out;
                 }
 
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                call_count = internal_lock_count (frame, this, fd_ctx);
+                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,
@@ -1399,258 +1534,77 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this)
                 /* 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] && fd_ctx->opened_on[i]) {
-                                afr_trace_inodelk_in (frame, 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->finodelk,
-                                                   this->name, local->fd,
-                                                   F_SETLK, &flock);
-
-                                if (!--call_count)
-                                        break;
+                        if (!local->child_up[i])
+                                continue;
 
+                        flock_use = &flock;
+                        if (!local->transaction.eager_lock_on) {
+                                goto wind;
                         }
 
-                }
-        } else {
-                call_count = internal_lock_count (frame, this, NULL);
-                int_lock->lk_call_count = call_count;
+                        piggyback = 0;
+                        local->transaction.eager_lock[i] = 1;
 
-                for (i = 0; i < priv->child_count; i++) {
-                        if (local->child_up[i]) {
-                                afr_trace_inodelk_in (frame, AFR_INODELK_NB_TRANSACTION,
-                                                      AFR_LOCK_OP, &flock, F_SETLK, i);
+			afr_set_delayed_post_op (frame, this);
 
-                                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);
+                        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;
                         }
-                }
-        }
-
-out:
-        return ret;
-}
-
-static int
-__is_lower_locked (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-        afr_local_t         *local    = NULL;
-        int count = 0;
-        int i     = 0;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (int_lock->lower_locked_nodes[i] & LOCKED_LOWER)
-                        count++;
-        }
-
-        return count;
-
-}
-
-static int
-__is_higher_locked (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-        afr_local_t         *local    = NULL;
-        int count = 0;
-        int i = 0;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (int_lock->locked_nodes[i] & LOCKED_YES)
-                        count++;
-        }
-
-        return count;
-
-}
-
-static int
-afr_unlock_lower_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;
-        const char          *basename = NULL;
-        loc_t               *loc      = NULL;
-        int call_count = 0;
-        int i = -1;
+                        flock_use = &full_flock;
+                wind:
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_NB_TRANSACTION,
+                                              AFR_LOCK_OP, flock_use, F_SETLK, i);
 
-        local = frame->local;
-        int_lock = &local->internal_lock;
-        priv  = this->private;
-
-        basename = int_lock->lk_basename;
-        if (int_lock->lk_loc)
-                loc = int_lock->lk_loc;
-
-        call_count = __is_lower_locked (frame, this);
-        int_lock->lk_call_count = call_count;
+                        STACK_WIND_COOKIE (frame, afr_nonblocking_inodelk_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->finodelk,
+                                           int_lock->domain, local->fd,
+                                           F_SETLK, flock_use, NULL);
 
-        if (!call_count){
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No internal locks unlocked");
-                int_lock->lock_cbk (frame, this);
-                goto out;
-        }
+                        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 (int_lock->lower_locked_nodes[i] & LOCKED_LOWER) {
-                        afr_trace_entrylk_in (frame, AFR_ENTRYLK_NB_TRANSACTION,
-                                              AFR_UNLOCK_OP, basename, i);
+                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_unlock_entrylk_cbk,
+                        STACK_WIND_COOKIE (frame, afr_nonblocking_inodelk_cbk,
                                            (void *) (long) i,
                                            priv->children[i],
-                                           priv->children[i]->fops->entrylk,
-                                           this->name,
-                                           loc, basename,
-                                           ENTRYLK_UNLOCK, ENTRYLK_WRLCK);
+                                           priv->children[i]->fops->inodelk,
+                                           int_lock->domain, &local->loc,
+                                           F_SETLK, &flock, NULL);
 
                         if (!--call_count)
                                 break;
-
                 }
         }
-
 out:
-        return 0;
-
-}
-
-
-static int
-afr_post_unlock_higher_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        local->transaction.done (frame, this);
-        return 0;
-}
-
-static int
-afr_post_unlock_lower_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock    = NULL;
-        afr_local_t         *local       = NULL;
-        loc_t               *lower       = NULL;
-        loc_t               *higher      = NULL;
-        const char          *lower_name  = NULL;
-        const char          *higher_name = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        lower = lower_path (&local->transaction.parent_loc,
-                            local->transaction.basename,
-                            &local->transaction.new_parent_loc,
-                            local->transaction.new_basename);
-
-        lower_name = (lower == &local->transaction.parent_loc ?
-                      local->transaction.basename :
-                      local->transaction.new_basename);
-
-        higher = (lower == &local->transaction.parent_loc ?
-                  &local->transaction.new_parent_loc :
-                  &local->transaction.parent_loc);
-
-        higher_name = (higher == &local->transaction.parent_loc ?
-                       local->transaction.basename :
-                       local->transaction.new_basename);
-
-        if (__is_higher_locked (frame, this)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "unlocking higher");
-                int_lock->lk_basename = higher_name;
-                int_lock->lk_loc      = higher;
-                int_lock->lock_cbk    = afr_post_unlock_higher_cbk;
-
-                afr_unlock_entrylk (frame, this);
-        } else
-                local->transaction.done (frame, this);
-
-        return 0;
-}
-
-static int
-afr_rename_unlock (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock    = NULL;
-        afr_local_t         *local       = NULL;
-        loc_t               *lower       = NULL;
-        loc_t               *higher      = NULL;
-        const char          *lower_name  = NULL;
-        const char          *higher_name = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        lower = lower_path (&local->transaction.parent_loc,
-                            local->transaction.basename,
-                            &local->transaction.new_parent_loc,
-                            local->transaction.new_basename);
-
-        lower_name = (lower == &local->transaction.parent_loc ?
-                      local->transaction.basename :
-                      local->transaction.new_basename);
-
-        higher = (lower == &local->transaction.parent_loc ?
-                  &local->transaction.new_parent_loc :
-                  &local->transaction.parent_loc);
-
-        higher_name = (higher == &local->transaction.parent_loc ?
-                       local->transaction.basename :
-                       local->transaction.new_basename);
-
-
-        if (__is_lower_locked (frame, this)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "unlocking lower");
-                int_lock->lk_basename = lower_name;
-                int_lock->lk_loc      = lower;
-                int_lock->lock_cbk    = afr_post_unlock_lower_cbk;
-
-                afr_unlock_lower_entrylk (frame, this);
-        } else
-                afr_post_unlock_lower_cbk (frame, this);
-
-        return 0;
-}
-
-static int
-afr_rename_transaction (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        return (local->transaction.type ==
-                AFR_ENTRY_RENAME_TRANSACTION);
-
+        return ret;
 }
 
 int32_t
@@ -1664,10 +1618,8 @@ afr_unlock (call_frame_t *frame, xlator_t *this)
                 if (is_afr_lock_transaction (local))
                         afr_unlock_inodelk (frame, this);
                 else
-                        if (!afr_rename_transaction (frame, this))
-                                afr_unlock_entrylk (frame, this);
-                        else
-                                afr_rename_unlock (frame, this);
+                        afr_unlock_entrylk (frame, this);
+
         } else {
                 if (is_afr_lock_selfheal (local))
                         afr_unlock_inodelk (frame, this);
@@ -1679,485 +1631,37 @@ afr_unlock (call_frame_t *frame, xlator_t *this)
 }
 
 int
-afr_mark_locked_nodes (xlator_t *this, fd_t *fd,
-                       unsigned char *locked_nodes)
+afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src, char *dom,
+                          unsigned int child_count)
 {
-        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");
+        afr_local_t         *dst_local   = NULL;
+        afr_local_t         *src_local   = NULL;
+        afr_internal_lock_t *dst_lock    = NULL;
+        afr_internal_lock_t *src_lock    = NULL;
+        afr_inodelk_t       *dst_inodelk = NULL;
+        afr_inodelk_t       *src_inodelk = NULL;
+        int                 ret = -1;
+
+        src_local = src->local;
+        src_lock  = &src_local->internal_lock;
+        src_inodelk = afr_get_inodelk (src_lock, dom);
+        dst_local = dst->local;
+        dst_lock  = &dst_local->internal_lock;
+        dst_inodelk = afr_get_inodelk (dst_lock, dom);
+        if (!dst_inodelk || !src_inodelk)
                 goto out;
-        }
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
-
-        GF_ASSERT (fdctx->locked_on);
-
-        memcpy (fdctx->locked_on, locked_nodes,
-                priv->child_count);
-
-out:
-        return ret;
-}
-
-static int
-__is_fd_saved (xlator_t *this, fd_t *fd)
-{
-        afr_locked_fd_t *locked_fd = NULL;
-        afr_private_t   *priv      = NULL;
-        int              found     = 0;
-
-        priv = this->private;
-
-        list_for_each_entry (locked_fd, &priv->saved_fds, list) {
-                if (locked_fd->fd == fd) {
-                        found = 1;
-                        break;
-                }
-        }
-
-        return found;
-}
-
-static int
-__afr_save_locked_fd (xlator_t *this, fd_t *fd)
-{
-        afr_private_t   *priv      = NULL;
-        afr_locked_fd_t *locked_fd = NULL;
-        int              ret       = 0;
-
-        priv = this->private;
-
-        locked_fd = GF_CALLOC (1, sizeof (*locked_fd),
-                               gf_afr_mt_locked_fd);
-        if (!locked_fd) {
-                ret = -1;
-                goto out;
-        }
-
-        locked_fd->fd = fd;
-        INIT_LIST_HEAD (&locked_fd->list);
-
-        list_add_tail (&locked_fd->list, &priv->saved_fds);
-
-out:
-        return ret;
-}
-
-int
-afr_save_locked_fd (xlator_t *this, fd_t *fd)
-{
-        afr_private_t   *priv      = NULL;
-        int              ret       = 0;
-
-        priv = this->private;
-
-        pthread_mutex_lock (&priv->mutex);
-        {
-                if (__is_fd_saved (this, fd)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "fd=%p already saved", fd);
-                        goto unlock;
-                }
-
-                ret = __afr_save_locked_fd (this, fd);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "fd=%p could not be saved", fd);
-                        goto unlock;
-                }
-        }
-unlock:
-        pthread_mutex_unlock (&priv->mutex);
-
-        return ret;
-}
-
-static int
-afr_lock_recovery_cleanup (call_frame_t *frame, xlator_t *this)
-{
-        afr_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;
-                }
-        }
-
-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);
-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)
-{
-        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);
-
-        return 0;
-
-cleanup:
-        afr_lock_recovery_cleanup (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);
-
-        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)
-{
-        if (op_ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Failed to get locks on fd");
-                goto cleanup;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Got a lock on fd");
-
-        if (is_afr_lock_eol (lock)) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Reached EOL on locks on fd");
-                goto cleanup;
-        }
-
-        afr_recover_lock (frame, this, lock);
-
-        return 0;
-
-cleanup:
-        afr_lock_recovery_cleanup (frame, this);
-
-        return 0;
-}
-
-static int
-afr_lock_recovery (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local        = NULL;
-        afr_private_t *priv         = NULL;
-        fd_t          *fd           = NULL;
-        int            ret          = 0;
-        int32_t        source_child = 0;
-        struct gf_flock   flock        = {0,};
-
-        priv  = this->private;
-        local = frame->local;
-
-        fd = local->fd;
-
-        source_child = afr_get_source_lock_recovery (this, fd);
-        if (source_child < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not recover locks due to lock "
-                        "split brain");
-                ret = -1;
-                goto out;
-        }
-
-        local->source_child = source_child;
-
-        /* the flock can be zero filled as we're querying incrementally
-           the locks held on the fd.
-        */
-        STACK_WIND_COOKIE (frame, afr_get_locks_fd_cbk,
-                           (void *) (long) source_child,
-                           priv->children[source_child],
-                           priv->children[source_child]->fops->lk,
-                           local->fd, F_GETLK_FD, &flock);
-
-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] = 1;
-
-out:
-        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)
-{
-        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);
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
-        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,
-                           fdctx->wbflags);
-
-        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])
-                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       = 0;
-        struct list_head locks_list = {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;
-        }
-
-        local = GF_CALLOC (1, sizeof (*local),
-                           gf_afr_mt_afr_local_t);
-        if (!local) {
-                ret = -1;
-                goto out;
-        }
-
-        AFR_LOCAL_INIT (local, priv);
-        if (!local) {
-                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);
-                }
-        }
-
+        if (src_inodelk->locked_nodes) {
+                memcpy (dst_inodelk->locked_nodes, src_inodelk->locked_nodes,
+                        sizeof (*dst_inodelk->locked_nodes) * child_count);
+                memset (src_inodelk->locked_nodes, 0,
+                        sizeof (*src_inodelk->locked_nodes) * child_count);
+        }
+
+        dst_lock->transaction_lk_type = src_lock->transaction_lk_type;
+        dst_lock->selfheal_lk_type    = src_lock->selfheal_lk_type;
+        dst_inodelk->lock_count = src_inodelk->lock_count;
+        src_inodelk->lock_count = 0;
+        ret = 0;
 out:
         return ret;
 }
diff --git a/xlators/cluster/afr/src/afr-mem-types.h b/xlators/cluster/afr/src/afr-mem-types.h
index de2049589..05df90cc0 100644
--- a/xlators/cluster/afr/src/afr-mem-types.h
+++ b/xlators/cluster/afr/src/afr-mem-types.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -26,7 +17,6 @@
 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_local_t,
         gf_afr_mt_afr_private_t,
         gf_afr_mt_int32_t,
         gf_afr_mt_char,
@@ -43,6 +33,16 @@ enum gf_afr_mem_types_ {
         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_subvol_healer_t,
         gf_afr_mt_end
 };
 #endif
diff --git a/xlators/cluster/afr/src/afr-open.c b/xlators/cluster/afr/src/afr-open.c
index ab636a5b4..f86aa7fd8 100644
--- a/xlators/cluster/afr/src/afr-open.c
+++ b/xlators/cluster/afr/src/afr-open.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include <libgen.h>
@@ -52,19 +43,31 @@
 #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"
+
+
+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;
+
+        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)
+                        struct iatt *postbuf, dict_t *xdata)
 {
         afr_local_t * local = frame->local;
 
         AFR_STACK_UNWIND (open, frame, local->op_ret, local->op_errno,
-                          local->fd);
+                          local->fd, xdata);
         return 0;
 }
 
@@ -72,583 +75,263 @@ afr_open_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 int
 afr_open_cbk (call_frame_t *frame, void *cookie,
               xlator_t *this, int32_t op_ret, int32_t op_errno,
-              fd_t *fd)
+              fd_t *fd, 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 = (long) cookie;
+	afr_fd_ctx_t  *fd_ctx = NULL;
 
         local = frame->local;
+	fd_ctx = local->fd_ctx;
 
         LOCK (&frame->lock);
         {
                 if (op_ret == -1) {
                         local->op_errno = op_errno;
-                }
-
-                if (op_ret >= 0) {
+			fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
+                } else {
                         local->op_ret = op_ret;
-                        local->success_count++;
-
-                        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);
-
-                                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] = 1;
-                        fd_ctx->flags                  = local->cont.open.flags;
-                        fd_ctx->wbflags                = local->cont.open.wbflags;
+			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)) {
+                if ((fd_ctx->flags & O_TRUNC) && (local->op_ret >= 0)) {
                         STACK_WIND (frame, afr_open_ftruncate_cbk,
                                     this, this->fops->ftruncate,
-                                    fd, 0);
+                                    fd, 0, NULL);
                 } else {
                         AFR_STACK_UNWIND (open, frame, local->op_ret,
-                                          local->op_errno, local->fd);
+                                          local->op_errno, local->fd,
+					  local->xdata_rsp);
                 }
         }
 
         return 0;
 }
 
-
 int
 afr_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-          fd_t *fd, int32_t wbflags)
+          fd_t *fd, dict_t *xdata)
 {
         afr_private_t * priv       = NULL;
         afr_local_t *   local      = NULL;
         int             i          = 0;
-        int             ret        = -1;
         int32_t         call_count = 0;
-        int32_t         op_ret     = -1;
         int32_t         op_errno   = 0;
-        int32_t         wind_flags = flags & (~O_TRUNC);
+	afr_fd_ctx_t   *fd_ctx = NULL;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (loc, out);
+        //We can't let truncation to happen outside transaction.
 
         priv = this->private;
 
-        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;
+        if (flags & (O_CREAT|O_TRUNC)) {
+                QUORUM_CHECK(open,out);
         }
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+	fd_ctx = afr_fd_ctx_get (fd, this);
+	if (!fd_ctx) {
+		op_errno = ENOMEM;
+		goto out;
+	}
 
-        frame->local = local;
-        call_count   = local->call_count;
-
-        loc_copy (&local->loc, loc);
+        local->fd = fd_ref (fd);
+	local->fd_ctx = fd_ctx;
+	fd_ctx->flags = flags;
 
-        local->cont.open.flags   = flags;
-        local->cont.open.wbflags = wbflags;
+        call_count = local->call_count;
 
-        local->fd = fd_ref (fd);
+        local->cont.open.flags = flags;
 
         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, wbflags);
-
+                                           loc, (flags & ~O_TRUNC), fd, xdata);
                         if (!--call_count)
                                 break;
                 }
         }
 
-        op_ret = 0;
+	return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (open, frame, op_ret, op_errno, fd);
-        }
+	AFR_STACK_UNWIND (open, frame, -1, op_errno, fd, NULL);
 
         return 0;
 }
 
-
 int
-afr_openfd_sh_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-        afr_internal_lock_t *int_lock    = NULL;
-        afr_local_t         *local       = NULL;
-        afr_private_t       *priv        = NULL;
-        afr_fd_ctx_t        *fd_ctx      = NULL;
-        uint64_t             ctx         = 0;
-        int                  ret         = 0;
-        int                  call_count  = 0;
-        int                  child_index = (long) cookie;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret >= 0) {
-                        ret = fd_ctx_get (fd, this, &ctx);
-
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "failed to get fd context, %p", fd);
-                                goto out;
-                        }
-
-                        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                        fd_ctx->opened_on[child_index] = 1;
-
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "fd for %s opened successfully on subvolume %s",
-                                local->loc.path, priv->children[child_index]->name);
-                }
-        }
-out:
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                int_lock->lock_cbk = local->transaction.done;
-                local->transaction.resume (frame, this);
-        }
-
-        return 0;
-}
-
-
-static int
-__unopened_count (int child_count, unsigned int *opened_on, unsigned char *child_up)
-{
-        int i = 0;
-        int count = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (!opened_on[i] && child_up[i])
-                        count++;
-        }
-
-        return count;
-}
-
-
-int
-afr_openfd_sh_unwind (call_frame_t *frame, xlator_t *this, int32_t op_ret,
-                      int32_t op_errno)
+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;
-        uint64_t       ctx        = 0;
         afr_fd_ctx_t  *fd_ctx     = NULL;
-        int            abandon    = 0;
-        int            ret        = 0;
-        int            i          = 0;
-        int            call_count = 0;
+        int           call_count  = 0;
+        int           child_index = (long) cookie;
 
-        priv  = this->private;
-        local = frame->local;
+        priv     = this->private;
+        local    = frame->local;
 
-        /*
-         * Some subvolumes might have come up on which we never
-         * opened this fd in the first place. Re-open fd's on those
-         * subvolumes now.
-         */
-
-        ret = fd_ctx_get (local->fd, this, &ctx);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "failed to get fd context %p (%s)",
-                        local->fd, local->loc.path);
-                abandon = 1;
-                goto out;
+        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);
+        } 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 = (afr_fd_ctx_t *)(long) ctx;
+        fd_ctx = local->fd_ctx;
 
         LOCK (&local->fd->lock);
         {
-                call_count = __unopened_count (priv->child_count,
-                                               fd_ctx->opened_on,
-                                               local->child_up);
-                for (i = 0; i < priv->child_count; i++) {
-                        fd_ctx->pre_op_done[i] = 0;
-                        fd_ctx->pre_op_piggyback[i] = 0;
+                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);
 
-        if (call_count == 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "fd not open on any subvolume %p (%s)",
-                        local->fd, local->loc.path);
-                abandon = 1;
-                goto out;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (!fd_ctx->opened_on[i] && local->child_up[i]) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "opening fd for %s on subvolume %s",
-                                local->loc.path, priv->children[i]->name);
-
-                        STACK_WIND_COOKIE (frame, afr_openfd_sh_open_cbk,
-                                           (void *)(long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->open,
-                                           &local->loc, fd_ctx->flags, local->fd,
-                                           fd_ctx->wbflags);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-out:
-        if (abandon)
-                local->transaction.resume (frame, this);
+        call_count = afr_frame_return (frame);
+        if (call_count == 0)
+                AFR_STACK_DESTROY (frame);
 
         return 0;
 }
 
 
 static int
-afr_prepare_loc (call_frame_t *frame, fd_t *fd)
-{
-        afr_local_t    *local = NULL;
-        char           *name = NULL;
-        char           *path = NULL;
-        int             ret = 0;
-
-        if ((!fd) || (!fd->inode))
-                return -1;
-
-        local = frame->local;
-        ret = inode_path (fd->inode, NULL, (char **)&path);
-        if (ret <= 0) {
-                gf_log (frame->this->name, GF_LOG_DEBUG,
-                        "Unable to get path for gfid: %s",
-                        uuid_utoa (fd->inode->gfid));
-                return -1;
-        }
-
-        if (local->loc.path) {
-                if (strcmp (path, local->loc.path))
-                        gf_log (frame->this->name, GF_LOG_DEBUG,
-                                "overwriting old loc->path %s with %s",
-                                local->loc.path, path);
-                GF_FREE ((char *)local->loc.path);
-        }
-        local->loc.path = path;
-
-        name = strrchr (local->loc.path, '/');
-        if (name)
-                name++;
-        local->loc.name = name;
-
-        if (local->loc.inode) {
-                inode_unref (local->loc.inode);
-        }
-        local->loc.inode = inode_ref (fd->inode);
-
-        if (local->loc.parent) {
-                inode_unref (local->loc.parent);
-        }
-
-        local->loc.parent = inode_parent (local->loc.inode, 0, NULL);
-
-        return 0;
-}
-
-
-int
-afr_openfd_sh (call_frame_t *frame, xlator_t *this)
+afr_fd_ctx_need_open (fd_t *fd, xlator_t *this, unsigned char *need_open)
 {
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh = NULL;
-        char            sh_type_str[256] = {0,};
-
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        GF_ASSERT (local->loc.path);
-        /* forcibly trigger missing-entries self-heal */
-
-        sh->need_missing_entry_self_heal = _gf_true;
-        sh->need_gfid_self_heal = _gf_true;
-        sh->data_lock_held      = _gf_true;
-        sh->need_data_self_heal = _gf_true;
-        sh->type                = local->fd->inode->ia_type;
-        sh->background          = _gf_false;
-        sh->unwind              = afr_openfd_sh_unwind;
-
-        afr_self_heal_type_str_get(&local->self_heal,
-                                   sh_type_str,
-                                   sizeof(sh_type_str));
-        gf_log (this->name, GF_LOG_INFO, "%s self-heal triggered. "
-                "path: %s, reason: Replicate up down flush, data lock is held",
-                sh_type_str, local->loc.path);
-
-        afr_self_heal (frame, this, local->fd->inode);
-
-        return 0;
+	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;
 }
 
 
-int
-afr_openfd_flush_done (call_frame_t *frame, xlator_t *this)
+void
+afr_fix_open (fd_t *fd, xlator_t *this)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
-
-        uint64_t       ctx;
-        afr_fd_ctx_t * fd_ctx = NULL;
-
-        int _ret = -1;
+        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;
 
         priv  = this->private;
-        local = frame->local;
-
-        LOCK (&local->fd->lock);
-        {
-                _ret = __fd_ctx_get (local->fd, this, &ctx);
-                if (_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "failed to get fd context %p (%s)",
-                                local->fd, local->loc.path);
-                        goto out;
-                }
-
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                fd_ctx->down_count = priv->down_count;
-                fd_ctx->up_count   = priv->up_count;
-        }
-out:
-        UNLOCK (&local->fd->lock);
-
-        afr_local_transaction_cleanup (local, this);
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "The up/down flush is over");
-
-        fd_unref (local->fd);
-        local->openfd_flush_cbk (frame, this);
-
-        return 0;
-}
-
-
-
-int
-afr_openfd_xaction (call_frame_t *frame, xlator_t *this, fd_t *fd)
-{
-        afr_local_t   * local = NULL;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        local = frame->local;
-
-        local->op = GF_FOP_FLUSH;
-
-        local->transaction.fop    = afr_openfd_sh;
-        local->transaction.done   = afr_openfd_flush_done;
-
-        local->transaction.start  = 0;
-        local->transaction.len    = 0;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "doing up/down flush on fd=%p", fd);
-
-        afr_transaction (frame, this, AFR_DATA_TRANSACTION);
-
-out:
-        return 0;
-}
-
-
-
-int
-afr_openfd_xaction_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-        afr_internal_lock_t *int_lock    = NULL;
-        afr_local_t         *local       = NULL;
-        afr_private_t       *priv        = NULL;
-        int                  ret         = 0;
-        uint64_t             ctx         = 0;
-        afr_fd_ctx_t        *fd_ctx      = NULL;
-        int                  call_count  = 0;
-        int                  child_index = (long) cookie;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret >= 0) {
-                        ret = fd_ctx_get (fd, this, &ctx);
-
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "failed to get fd context %p (%s)",
-                                        fd, local->loc.path);
-                                goto out;
-                        }
-
-                        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                        fd_ctx->opened_on[child_index] = 1;
-
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "fd for %s opened successfully on subvolume %s",
-                                local->loc.path, priv->children[child_index]->name);
-                }
-        }
-out:
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
+        if (!afr_is_fd_fixable (fd))
+                goto out;
 
-        if (call_count == 0) {
-                afr_openfd_xaction (frame, this, local->fd);
-        }
+        fd_ctx = afr_fd_ctx_get (fd, this);
+        if (!fd_ctx)
+                goto out;
 
-        return 0;
-}
+	need_open = alloca0 (priv->child_count);
 
+	call_count = afr_fd_ctx_need_open (fd, this, need_open);
+	if (!call_count)
+		goto out;
 
-int
-afr_openfd_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
-{
-        afr_local_t   *local      = NULL;
-        afr_private_t *priv       = NULL;
-        uint64_t       ctx        = 0;
-        afr_fd_ctx_t  *fd_ctx     = NULL;
-        int            no_open    = 0;
-        int            ret        = 0;
-        int            i          = 0;
-        int            call_count = 0;
+        frame = create_frame (this, this->ctx->pool);
+        if (!frame)
+                goto out;
 
-        priv  = this->private;
-        local = frame->local;
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local)
+		goto out;
 
-        /*
-         * If the file is already deleted while the fd is open, no need to
-         * perform the openfd flush, call the flush_cbk and get out.
-         */
-        ret = afr_prepare_loc (frame, fd);
-        if (ret < 0) {
-                local->openfd_flush_cbk (frame, this);
+        local->loc.inode = inode_ref (fd->inode);
+        ret = loc_path (&local->loc, NULL);
+        if (ret < 0)
                 goto out;
-        }
-
-        /*
-         * Some subvolumes might have come up on which we never
-         * opened this fd in the first place. Re-open fd's on those
-         * subvolumes now.
-         */
 
         local->fd = fd_ref (fd);
+	local->fd_ctx = fd_ctx;
 
-        ret = fd_ctx_get (fd, this, &ctx);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "failed to get fd context %p (%s)",
-                        fd, local->loc.path);
-                no_open = 1;
-                goto out;
-        }
+        local->call_count = call_count;
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+        gf_log (this->name, GF_LOG_DEBUG, "need open count: %d",
+                call_count);
 
-        LOCK (&local->fd->lock);
-        {
-                call_count = __unopened_count (priv->child_count,
-                                               fd_ctx->opened_on,
-                                               local->child_up);
-        }
-        UNLOCK (&local->fd->lock);
-
-        if (call_count == 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "fd not open on any subvolume %p (%s)",
-                        fd, local->loc.path);
-                no_open = 1;
-                goto out;
-        }
+        for (i = 0; i < priv->child_count; i++) {
+                if (!need_open[i])
+                        continue;
 
-        local->call_count = call_count;
+                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);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (!fd_ctx->opened_on[i] && local->child_up[i]) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "opening fd for %s on subvolume %s",
+                        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_xaction_open_cbk,
+                        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, fd,
-                                           fd_ctx->wbflags);
-
-                        if (!--call_count)
-                                break;
+                                           &local->loc,
+                                           fd_ctx->flags & (~O_TRUNC),
+                                           local->fd, NULL);
                 }
+
+		if (!--call_count)
+			break;
         }
 
+	return;
 out:
-        if (no_open)
-                afr_openfd_xaction (frame, this, fd);
-
-        return 0;
+        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 000000000..186f68c33
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-read-txn.c
@@ -0,0 +1,239 @@
+/*
+  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"
+
+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;
+	local->readfn (frame, this, subvol);
+
+	return 0;
+}
+
+
+int
+afr_read_txn_refresh_done (call_frame_t *frame, xlator_t *this, int err)
+{
+	afr_local_t *local = NULL;
+	int read_subvol = 0;
+	int event_generation = 0;
+	inode_t *inode = NULL;
+	int ret = -1;
+
+	local = frame->local;
+	inode = local->inode;
+
+	if (err) {
+		local->op_errno = -err;
+		local->op_ret = -1;
+		read_subvol = -1;
+		goto readfn;
+	}
+
+	ret = afr_inode_read_subvol_type_get (inode, this, local->readable,
+					      &event_generation,
+					      local->transaction.type);
+
+	if (ret == -1 || !event_generation) {
+		/* Even after refresh, we don't have a good
+		   read subvolume. Time to bail */
+		local->op_ret = -1;
+		local->op_errno = EIO;
+		read_subvol = -1;
+		goto readfn;
+	}
+
+	read_subvol = afr_read_subvol_select_by_policy (inode, this,
+							local->readable);
+
+	if (read_subvol == -1) {
+		local->op_ret = -1;
+		local->op_errno = 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:
+	local->readfn (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,
+				   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;
+	int read_subvol = -1;
+	int event_generation = 0;
+	int ret = -1;
+
+	priv = this->private;
+	local = frame->local;
+
+	afr_read_txn_wipe (frame, this);
+
+	local->readfn = readfn;
+	local->inode = inode_ref (inode);
+
+	local->transaction.type = type;
+	ret = afr_inode_read_subvol_type_get (inode, this, local->readable,
+					      &event_generation, type);
+	if (ret == -1)
+		/* very first transaction on this inode */
+		goto refresh;
+
+	if (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);
+
+	if (read_subvol < 0 || read_subvol > priv->child_count) {
+		gf_log (this->name, GF_LOG_WARNING, "Unreadable subvolume %d "
+			"found with event generation %d", read_subvol,
+			event_generation);
+		goto refresh;
+	}
+
+	if (!local->child_up[read_subvol]) {
+		/* should never happen, just in case */
+		gf_log (this->name, GF_LOG_WARNING, "subvolume %d is the "
+			"read subvolume in this generation, but is not up",
+			read_subvol);
+		goto refresh;
+	}
+
+	local->read_attempted[read_subvol] = 1;
+
+	local->readfn (frame, this, read_subvol);
+
+	return 0;
+
+refresh:
+	afr_inode_refresh (frame, this, inode, 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 412d3c6db..000000000
--- a/xlators/cluster/afr/src/afr-self-heal-algorithm.c
+++ /dev/null
@@ -1,1090 +0,0 @@
-/*
-  Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-
-#include "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 "md5.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
-*/
-
-
-/*
-  The "full" algorithm. Copies the entire file from
-  source to sinks.
-*/
-
-
-static void
-sh_full_private_cleanup (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *               local   = NULL;
-        afr_self_heal_t *           sh      = NULL;
-        afr_sh_algo_full_private_t *sh_priv = NULL;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        sh_priv = sh->private;
-
-        if (sh_priv)
-                GF_FREE (sh_priv);
-}
-
-
-static int
-sh_full_loop_driver (call_frame_t *frame, xlator_t *this, gf_boolean_t is_first_call);
-
-static int
-sh_full_loop_driver_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t * priv = NULL;
-        afr_local_t * local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        afr_sh_algo_full_private_t *sh_priv = NULL;
-
-        priv    = this->private;
-        local   = frame->local;
-        sh      = &local->self_heal;
-        sh_priv = sh->private;
-
-        sh_full_private_cleanup (frame, this);
-        if (sh->op_failed) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "full self-heal aborting on %s",
-                        local->loc.path);
-
-                local->self_heal.algo_abort_cbk (frame, this);
-        } else {
-                gf_log (this->name, GF_LOG_INFO,
-                        "full self-heal completed on %s",
-                        local->loc.path);
-
-                local->self_heal.algo_completion_cbk (frame, this);
-        }
-        return 0;
-}
-
-static int
-sh_full_loop_return (call_frame_t *rw_frame, xlator_t *this, off_t offset)
-{
-        afr_local_t *               rw_local = NULL;
-        afr_self_heal_t *           rw_sh    = NULL;
-        call_frame_t               *sh_frame = NULL;
-        afr_local_t *               sh_local = NULL;
-        afr_self_heal_t            *sh       = NULL;
-        afr_sh_algo_full_private_t *sh_priv  = NULL;
-
-        rw_local = rw_frame->local;
-        rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        sh_priv  = sh->private;
-
-        AFR_STACK_DESTROY (rw_frame);
-
-        sh_full_loop_driver (sh_frame, this, _gf_false);
-
-        return 0;
-}
-
-
-static int
-sh_full_write_cbk (call_frame_t *rw_frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                   struct iatt *postbuf)
-{
-        afr_private_t *  priv        = NULL;
-        afr_local_t *    rw_local    = NULL;
-        afr_self_heal_t *rw_sh       = NULL;
-        call_frame_t    *sh_frame    = NULL;
-        afr_local_t *    sh_local    = NULL;
-        afr_self_heal_t *sh          = NULL;
-        int              child_index = (long) cookie;
-        int              call_count  = 0;
-
-        priv = this->private;
-
-        rw_local = rw_frame->local;
-        rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-
-        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,
-                rw_sh->offset - op_ret);
-
-        LOCK (&sh_frame->lock);
-        {
-                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->op_failed = 1;
-                }
-        }
-        UNLOCK (&sh_frame->lock);
-
-        call_count = afr_frame_return (rw_frame);
-
-        if (call_count == 0) {
-                sh_full_loop_return (rw_frame, this, rw_sh->offset - op_ret);
-        }
-
-        return 0;
-}
-
-
-static int
-sh_full_read_cbk (call_frame_t *rw_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)
-{
-        afr_private_t *  priv       = NULL;
-        afr_local_t *    rw_local   = NULL;
-        afr_self_heal_t *rw_sh      = NULL;
-        call_frame_t    *sh_frame   = NULL;
-        afr_local_t *    sh_local   = NULL;
-        afr_self_heal_t *sh         = NULL;
-        int              i          = 0;
-        int              call_count = 0;
-        off_t            offset     = (long) cookie;
-
-        priv = this->private;
-        rw_local = rw_frame->local;
-        rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-
-        call_count = sh->active_sinks;
-
-        rw_local->call_count = call_count;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "read %d bytes of data from %s, offset %"PRId64"",
-                op_ret, sh_local->loc.path, offset);
-
-        if (op_ret <= 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "read from %s failed on subvolume %s (%s)",
-                        sh_local->loc.path,
-                        priv->children[sh->source]->name,
-                        strerror (op_errno));
-                sh->op_failed = 1;
-                sh_full_loop_return (rw_frame, this, offset);
-                return 0;
-        }
-
-        rw_sh->offset += op_ret;
-
-        if (sh->file_has_holes) {
-                if (iov_0filled (vector, count) == 0) {
-                        /* the iter function depends on the
-                           sh->offset already being updated
-                           above
-                        */
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "block has all 0 filled");
-                        sh_full_loop_return (rw_frame, this, offset);
-                        goto out;
-                }
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->sources[i] || !sh_local->child_up[i])
-                        continue;
-
-                /* this is a sink, so write to it */
-
-                STACK_WIND_COOKIE (rw_frame, sh_full_write_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->writev,
-                                   sh->healing_fd, vector, count, offset,
-                                   iobref);
-
-                if (!--call_count)
-                        break;
-        }
-
-out:
-        return 0;
-}
-
-
-static int
-sh_full_read_write (call_frame_t *frame, xlator_t *this, off_t offset)
-{
-        afr_private_t *  priv     = NULL;
-        afr_local_t *    local    = NULL;
-        afr_local_t *    rw_local = NULL;
-        afr_self_heal_t *rw_sh    = NULL;
-        afr_self_heal_t *sh       = NULL;
-        call_frame_t    *rw_frame = NULL;
-        int32_t          op_errno = 0;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        rw_frame = copy_frame (frame);
-        if (!rw_frame)
-                goto out;
-
-        ALLOC_OR_GOTO (rw_local, afr_local_t, out);
-
-        rw_frame->local = rw_local;
-        rw_sh           = &rw_local->self_heal;
-
-        rw_sh->offset       = offset;
-        rw_sh->sh_frame     = frame;
-
-        STACK_WIND_COOKIE (rw_frame, sh_full_read_cbk,
-                           (void *) (long) offset,
-                           priv->children[sh->source],
-                           priv->children[sh->source]->fops->readv,
-                           sh->healing_fd, sh->block_size,
-                           offset);
-        return 0;
-
-out:
-        sh->op_failed = 1;
-
-        sh_full_loop_driver (frame, this, _gf_false);
-
-        return 0;
-}
-
-
-static int
-sh_full_loop_driver (call_frame_t *frame, xlator_t *this, gf_boolean_t is_first_call)
-{
-        afr_private_t *             priv           = NULL;
-        afr_local_t *               local          = NULL;
-        afr_self_heal_t            *sh             = NULL;
-        afr_sh_algo_full_private_t *sh_priv        = NULL;
-        gf_boolean_t                is_driver_done = _gf_false;
-        blksize_t                   block_size     = 0;
-        off_t                       offset         = 0;
-        int                         loop           = 0;
-
-        priv    = this->private;
-        local   = frame->local;
-        sh      = &local->self_heal;
-        sh_priv = sh->private;
-
-        LOCK (&sh_priv->lock);
-        {
-                if (_gf_false == is_first_call)
-                        sh_priv->loops_running--;
-                offset           = sh_priv->offset;
-                block_size       = sh->block_size;
-                while ((sh->op_failed == 0) &&
-                       (sh_priv->loops_running < priv->data_self_heal_window_size)
-                       && (sh_priv->offset < sh->file_size)) {
-
-                        loop++;
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "spawning a loop for offset %"PRId64,
-                                sh_priv->offset);
-
-                        sh_priv->offset += sh->block_size;
-                        sh_priv->loops_running++;
-
-                        if (_gf_false == is_first_call)
-                                break;
-
-                }
-                if (0 == sh_priv->loops_running) {
-                        is_driver_done = _gf_true;
-                }
-        }
-        UNLOCK (&sh_priv->lock);
-
-        while (loop--) {
-                if (sh->op_failed) {
-                        // op failed in other loop, stop spawning more loops
-                        sh_full_loop_driver (frame, this, _gf_false);
-                } else {
-                        sh_full_read_write (frame, this, offset);
-                        offset += block_size;
-                }
-        }
-
-        if (is_driver_done) {
-                sh_full_loop_driver_done (frame, this);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_algo_full (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *               local   = NULL;
-        afr_self_heal_t *           sh      = NULL;
-        afr_sh_algo_full_private_t *sh_priv = NULL;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        sh_priv = GF_CALLOC (1, sizeof (*sh_priv),
-                             gf_afr_mt_afr_private_t);
-        if (!sh_priv)
-                goto out;
-
-        LOCK_INIT (&sh_priv->lock);
-
-        sh->private = sh_priv;
-
-        local->call_count = 0;
-
-        sh_full_loop_driver (frame, this, _gf_true);
-out:
-        return 0;
-}
-
-
-/*
- * The "diff" algorithm. Copies only those blocks whose checksums
- * don't match with those of source.
- */
-
-
-static void
-sh_diff_private_cleanup (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *             priv    = NULL;
-        afr_local_t *               local   = NULL;
-        afr_self_heal_t *           sh      = NULL;
-        afr_sh_algo_diff_private_t *sh_priv = NULL;
-        int                         i       = 0;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        sh_priv = sh->private;
-
-        for (i = 0; i < priv->data_self_heal_window_size; i++) {
-                if (sh_priv->loops[i]) {
-                        if (sh_priv->loops[i]->write_needed)
-                                GF_FREE (sh_priv->loops[i]->write_needed);
-
-                        if (sh_priv->loops[i]->checksum)
-                                GF_FREE (sh_priv->loops[i]->checksum);
-
-                        GF_FREE (sh_priv->loops[i]);
-                }
-        }
-
-        if (sh_priv) {
-                if (sh_priv->loops)
-                        GF_FREE (sh_priv->loops);
-
-                GF_FREE (sh_priv);
-        }
-
-
-}
-
-
-static uint32_t
-__make_cookie (int loop_index, int child_index)
-{
-        uint32_t ret = ((loop_index << 16) | child_index);
-        return ret;
-}
-
-
-static int
-__loop_index (uint32_t cookie)
-{
-        return ((cookie & 0xFFFF0000) >> 16);
-}
-
-
-static int
-__child_index (uint32_t cookie)
-{
-        return (cookie & 0x0000FFFF);
-}
-
-
-static void
-sh_diff_loop_state_reset (struct sh_diff_loop_state *loop_state, int child_count)
-{
-        loop_state->active = _gf_false;
-//        loop_state->offset = 0;
-
-        memset (loop_state->write_needed,
-                0, sizeof (*loop_state->write_needed) * child_count);
-
-        memset (loop_state->checksum,
-                0, MD5_DIGEST_LEN * child_count);
-}
-
-
-static int
-sh_diff_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_diff_loop_driver_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *             priv         = NULL;
-        afr_local_t *               local        = NULL;
-        afr_self_heal_t *           sh           = NULL;
-        afr_sh_algo_diff_private_t *sh_priv      = NULL;
-        int32_t                     total_blocks = 0;
-        int32_t                     diff_blocks  = 0;
-
-        priv         = this->private;
-        local        = frame->local;
-        sh           = &local->self_heal;
-        sh_priv      = sh->private;
-        total_blocks = sh_priv->total_blocks;
-        diff_blocks  = sh_priv->diff_blocks;
-
-        sh_diff_private_cleanup (frame, this);
-        if (sh->op_failed) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "diff self-heal aborting on %s",
-                        local->loc.path);
-
-                local->self_heal.algo_abort_cbk (frame, this);
-        } else {
-                gf_log (this->name, GF_LOG_INFO,
-                        "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);
-
-                local->self_heal.algo_completion_cbk (frame, this);
-        }
-
-        return 0;
-}
-
-static int
-sh_diff_loop_driver (call_frame_t *frame, xlator_t *this,
-                     gf_boolean_t is_first_call,
-                     struct sh_diff_loop_state *loop_state);
-
-static int
-sh_diff_loop_return (call_frame_t *rw_frame, xlator_t *this,
-                     struct sh_diff_loop_state *loop_state)
-{
-        afr_private_t *             priv     = NULL;
-        afr_local_t *               rw_local = NULL;
-        afr_self_heal_t *           rw_sh    = NULL;
-        call_frame_t               *sh_frame = NULL;
-        afr_local_t *               sh_local = NULL;
-        afr_self_heal_t            *sh       = NULL;
-        afr_sh_algo_diff_private_t *sh_priv  = NULL;
-
-        priv  = this->private;
-
-        rw_local = rw_frame->local;
-        rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        sh_priv  = sh->private;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "loop for offset %"PRId64" returned", loop_state->offset);
-
-        AFR_STACK_DESTROY (rw_frame);
-
-        sh_diff_loop_driver (sh_frame, this, _gf_false, loop_state);
-
-        return 0;
-}
-
-
-static int
-sh_diff_write_cbk (call_frame_t *rw_frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                   struct iatt *postbuf)
-{
-        afr_private_t *             priv        = NULL;
-        afr_local_t *               rw_local    = NULL;
-        afr_self_heal_t *           rw_sh       = NULL;
-        call_frame_t               *sh_frame    = NULL;
-        afr_local_t *               sh_local    = NULL;
-        afr_self_heal_t            *sh          = NULL;
-        afr_sh_algo_diff_private_t *sh_priv     = NULL;
-        struct sh_diff_loop_state  *loop_state  = NULL;
-        int                         call_count  = 0;
-        int                         child_index = 0;
-        int                         loop_index  = 0;
-
-        priv     = this->private;
-        rw_local = rw_frame->local;
-        rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        sh_priv  = sh->private;
-
-        child_index = __child_index ((uint32_t) (long) cookie);
-        loop_index  = __loop_index ((uint32_t) (long) cookie);
-        loop_state  = sh_priv->loops[loop_index];
-
-        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_state->offset);
-
-        LOCK (&sh_frame->lock);
-        {
-                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->op_failed = 1;
-                }
-        }
-        UNLOCK (&sh_frame->lock);
-
-        call_count = afr_frame_return (rw_frame);
-
-        if (call_count == 0) {
-                sh_diff_loop_return (rw_frame, this, loop_state);
-        }
-
-        return 0;
-}
-
-
-static int
-sh_diff_read_cbk (call_frame_t *rw_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)
-{
-        afr_private_t *               priv       = NULL;
-        afr_local_t *                 rw_local   = NULL;
-        afr_self_heal_t *             rw_sh      = NULL;
-        afr_sh_algo_diff_private_t *  sh_priv    = NULL;
-        call_frame_t                 *sh_frame   = NULL;
-        afr_local_t *                 sh_local   = NULL;
-        afr_self_heal_t              *sh         = NULL;
-        int                           loop_index = 0;
-        struct sh_diff_loop_state    *loop_state = NULL;
-        uint32_t                      wcookie    = 0;
-        int                           i          = 0;
-        int                           call_count = 0;
-
-        priv     = this->private;
-        rw_local = rw_frame->local;
-        rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        sh_priv  = sh->private;
-
-        loop_index = __loop_index ((uint32_t) (long) cookie);
-        loop_state = sh_priv->loops[loop_index];
-
-        call_count = sh_diff_number_of_writes_needed (loop_state->write_needed,
-                                                      priv->child_count);
-
-        rw_local->call_count = call_count;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "read %d bytes of data from %s, offset %"PRId64"",
-                op_ret, sh_local->loc.path, loop_state->offset);
-
-        if ((op_ret <= 0) ||
-            (call_count == 0)) {
-                sh_diff_loop_return (rw_frame, this, loop_state);
-
-                return 0;
-        }
-
-        if (sh->file_has_holes) {
-                if (iov_0filled (vector, count) == 0) {
-                        gf_log (this->name, GF_LOG_DEBUG, "0 filled block");
-                        sh_diff_loop_return (rw_frame, this, loop_state);
-                        goto out;
-                }
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (loop_state->write_needed[i]) {
-                        wcookie = __make_cookie (loop_index, i);
-
-                        STACK_WIND_COOKIE (rw_frame, sh_diff_write_cbk,
-                                           (void *) (long) wcookie,
-                                           priv->children[i],
-                                           priv->children[i]->fops->writev,
-                                           sh->healing_fd, vector, count,
-                                           loop_state->offset, iobref);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-out:
-        return 0;
-}
-
-
-static int
-sh_diff_read (call_frame_t *rw_frame, xlator_t *this,
-              int loop_index)
-{
-        afr_private_t *               priv       = NULL;
-        afr_local_t *                 rw_local   = NULL;
-        afr_self_heal_t *             rw_sh      = NULL;
-        afr_sh_algo_diff_private_t *  sh_priv    = NULL;
-        struct sh_diff_loop_state    *loop_state = NULL;
-        call_frame_t                 *sh_frame   = NULL;
-        afr_local_t *                 sh_local   = NULL;
-        afr_self_heal_t              *sh         = NULL;
-        uint32_t                      cookie     = 0;
-
-        priv     = this->private;
-        rw_local = rw_frame->local;
-        rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        sh_priv  = sh->private;
-
-        loop_state = sh_priv->loops[loop_index];
-
-        cookie = __make_cookie (loop_index, sh->source);
-
-        STACK_WIND_COOKIE (rw_frame, sh_diff_read_cbk,
-                           (void *) (long) cookie,
-                           priv->children[sh->source],
-                           priv->children[sh->source]->fops->readv,
-                           sh->healing_fd, sh_priv->block_size,
-                           loop_state->offset);
-
-        return 0;
-}
-
-
-static int
-sh_diff_checksum_cbk (call_frame_t *rw_frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      uint32_t weak_checksum, uint8_t *strong_checksum)
-{
-        afr_private_t *               priv         = NULL;
-        afr_local_t *                 rw_local     = NULL;
-        afr_self_heal_t              *rw_sh        = NULL;
-        call_frame_t                 *sh_frame     = NULL;
-        afr_local_t *                 sh_local     = NULL;
-        afr_self_heal_t              *sh           = NULL;
-        afr_sh_algo_diff_private_t *  sh_priv      = NULL;
-        int                           loop_index   = 0;
-        int                           child_index  = 0;
-        struct sh_diff_loop_state    *loop_state   = NULL;
-        int                           call_count   = 0;
-        int                           i            = 0;
-        int                           write_needed = 0;
-
-        priv  = this->private;
-
-        rw_local = rw_frame->local;
-        rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-
-        sh_priv = sh->private;
-
-        child_index = __child_index ((uint32_t) (long) cookie);
-        loop_index  = __loop_index ((uint32_t) (long) cookie);
-
-        loop_state  = sh_priv->loops[loop_index];
-
-        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->op_failed = 1;
-        } else {
-                memcpy (loop_state->checksum + child_index * MD5_DIGEST_LEN,
-                        strong_checksum,
-                        MD5_DIGEST_LEN);
-        }
-
-        call_count = afr_frame_return (rw_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_state->checksum + (i * MD5_DIGEST_LEN),
-                                    loop_state->checksum + (sh->source * MD5_DIGEST_LEN),
-                                    MD5_DIGEST_LEN)) {
-                                /*
-                                  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_state->offset);
-
-                                write_needed = loop_state->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 && !sh->op_failed) {
-                        sh_diff_read (rw_frame, this, loop_index);
-                } else {
-                        sh->offset += sh_priv->block_size;
-
-                        sh_diff_loop_return (rw_frame, this, loop_state);
-                }
-        }
-
-        return 0;
-}
-
-
-static int
-sh_diff_find_unused_loop (afr_sh_algo_diff_private_t *sh_priv, int max)
-{
-        int i = 0;
-
-        LOCK (&sh_priv->lock);
-        {
-                for (i = 0; i < max; i++) {
-                        if (sh_priv->loops[i]->active == _gf_false) {
-                                sh_priv->loops[i]->active = _gf_true;
-                                break;
-                        }
-                }
-        }
-        UNLOCK (&sh_priv->lock);
-
-        if (i == max) {
-                gf_log ("[sh-diff]", GF_LOG_ERROR,
-                        "no free loops found! This shouldn't happen. Please"
-                        " report this to gluster-devel@nongnu.org");
-        }
-
-        return i;
-}
-
-
-static int
-sh_diff_checksum (call_frame_t *frame, xlator_t *this, off_t offset)
-{
-        afr_private_t *               priv       = NULL;
-        afr_local_t *                 local      = NULL;
-        afr_local_t *                 rw_local   = NULL;
-        afr_self_heal_t *             sh         = NULL;
-        afr_self_heal_t *             rw_sh      = NULL;
-        afr_sh_algo_diff_private_t *  sh_priv    = NULL;
-        call_frame_t                 *rw_frame   = NULL;
-        uint32_t                      cookie     = 0;
-        int                           loop_index = 0;
-        struct sh_diff_loop_state    *loop_state = NULL;
-        int32_t                       op_errno   = 0;
-        int                           call_count = 0;
-        int                           i          = 0;
-
-        priv    = this->private;
-        local   = frame->local;
-        sh      = &local->self_heal;
-
-        sh_priv = sh->private;
-
-        rw_frame = copy_frame (frame);
-        if (!rw_frame)
-                goto out;
-
-        ALLOC_OR_GOTO (rw_local, afr_local_t, out);
-
-        rw_frame->local = rw_local;
-        rw_sh           = &rw_local->self_heal;
-
-        rw_sh->offset       = sh->offset;
-        rw_sh->sh_frame     = frame;
-
-        call_count = sh->active_sinks + 1;  /* sinks and source */
-
-        rw_local->call_count = call_count;
-
-        loop_index = sh_diff_find_unused_loop (sh_priv, priv->data_self_heal_window_size);
-
-        loop_state = sh_priv->loops[loop_index];
-        loop_state->offset       = offset;
-
-        /* we need to send both the loop index and child index,
-           so squeeze them both into a 32-bit number */
-
-        cookie = __make_cookie (loop_index, sh->source);
-
-        STACK_WIND_COOKIE (rw_frame, sh_diff_checksum_cbk,
-                           (void *) (long) cookie,
-                           priv->children[sh->source],
-                           priv->children[sh->source]->fops->rchecksum,
-                           sh->healing_fd,
-                           offset, sh_priv->block_size);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->sources[i] || !local->child_up[i])
-                        continue;
-
-                cookie = __make_cookie (loop_index, i);
-
-                STACK_WIND_COOKIE (rw_frame, sh_diff_checksum_cbk,
-                                   (void *) (long) cookie,
-                                   priv->children[i],
-                                   priv->children[i]->fops->rchecksum,
-                                   sh->healing_fd,
-                                   offset, sh_priv->block_size);
-
-                if (!--call_count)
-                        break;
-        }
-
-        return 0;
-
-out:
-        sh->op_failed = 1;
-
-        sh_diff_loop_driver (frame, this, _gf_false, loop_state);
-
-        return 0;
-}
-
-
-static int
-sh_diff_loop_driver (call_frame_t *frame, xlator_t *this,
-                     gf_boolean_t is_first_call,
-                     struct sh_diff_loop_state *loop_state)
-{
-        afr_private_t *             priv           = NULL;
-        afr_local_t *               local          = NULL;
-        afr_self_heal_t *           sh             = NULL;
-        afr_sh_algo_diff_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;
-        char                        sh_type_str[256] = {0,};
-
-        priv    = this->private;
-        local   = frame->local;
-        sh      = &local->self_heal;
-        sh_priv = sh->private;
-
-        afr_self_heal_type_str_get(sh, sh_type_str, sizeof(sh_type_str));
-
-        LOCK (&sh_priv->lock);
-        {
-                if (loop_state)
-                        sh_diff_loop_state_reset (loop_state, priv->child_count);
-                if (_gf_false == is_first_call)
-                        sh_priv->loops_running--;
-                offset = sh_priv->offset;
-                block_size = sh_priv->block_size;
-                while ((0 == sh->op_failed) &&
-                       (sh_priv->loops_running < priv->data_self_heal_window_size)
-                       && (sh_priv->offset < sh->file_size)) {
-
-                        loop++;
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "spawning a loop for offset %"PRId64,
-                                sh_priv->offset);
-
-                        sh_priv->offset += sh_priv->block_size;
-                        sh_priv->loops_running++;
-
-                        if (_gf_false == is_first_call)
-                                break;
-
-                }
-                if (0 == sh_priv->loops_running) {
-                        is_driver_done = _gf_true;
-                }
-        }
-        UNLOCK (&sh_priv->lock);
-
-        while (loop--) {
-                if (sh->op_failed) {
-                        // op failed in other loop, stop spawning more loops
-                        sh_diff_loop_driver (frame, this, _gf_false, NULL);
-                } else {
-                        sh_diff_checksum (frame, this, offset);
-                        offset += block_size;
-                }
-        }
-
-        if (is_driver_done) {
-                sh_diff_loop_driver_done (frame, this);
-        }
-        return 0;
-}
-
-
-int
-afr_sh_algo_diff (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *             priv    = NULL;
-        afr_local_t *               local   = NULL;
-        afr_self_heal_t *           sh      = NULL;
-        afr_sh_algo_diff_private_t *sh_priv = NULL;
-        int                         i       = 0;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        sh_priv = GF_CALLOC (1, sizeof (*sh_priv),
-                             gf_afr_mt_afr_private_t);
-        if (!sh_priv)
-                goto err;
-
-        sh_priv->block_size = this->ctx->page_size;
-
-        sh->private = sh_priv;
-
-        LOCK_INIT (&sh_priv->lock);
-
-        local->call_count = 0;
-
-        sh_priv->loops = GF_CALLOC (priv->data_self_heal_window_size,
-                                    sizeof (*sh_priv->loops),
-                                    gf_afr_mt_sh_diff_loop_state);
-        if (!sh_priv->loops)
-                goto err;
-
-        for (i = 0; i < priv->data_self_heal_window_size; i++) {
-                sh_priv->loops[i]               = GF_CALLOC (1, sizeof (*sh_priv->loops[i]),
-                                                             gf_afr_mt_sh_diff_loop_state);
-                if (!sh_priv->loops[i])
-                        goto err;
-
-                sh_priv->loops[i]->checksum     = GF_CALLOC (priv->child_count,
-                                                             MD5_DIGEST_LEN, gf_afr_mt_uint8_t);
-                if (!sh_priv->loops[i]->checksum)
-                        goto err;
-
-                sh_priv->loops[i]->write_needed = GF_CALLOC (priv->child_count,
-                                                             sizeof (*sh_priv->loops[i]->write_needed),
-                                                             gf_afr_mt_char);
-                if (!sh_priv->loops[i]->write_needed)
-                        goto err;
-
-        }
-
-        sh_diff_loop_driver (frame, this, _gf_true, NULL);
-
-        return 0;
-err:
-        if (sh_priv) {
-                if (sh_priv->loops) {
-                        for (i = 0; i < priv->data_self_heal_window_size; i++) {
-                                if (sh_priv->loops[i]->write_needed)
-                                        GF_FREE (sh_priv->loops[i]->write_needed);
-                                if (sh_priv->loops[i]->checksum)
-                                        GF_FREE (sh_priv->loops[i]->checksum);
-                                if (sh_priv->loops[i])
-                                        GF_FREE (sh_priv->loops[i]);
-                        }
-
-                        GF_FREE (sh_priv->loops);
-                }
-
-                GF_FREE (sh_priv);
-        }
-        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 e45621b0e..000000000
--- a/xlators/cluster/afr/src/afr-self-heal-algorithm.h
+++ /dev/null
@@ -1,60 +0,0 @@
-/*
-   Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#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;
-} afr_sh_algo_full_private_t;
-
-struct sh_diff_loop_state {
-        off_t   offset;
-        unsigned char *write_needed;
-        uint8_t *checksum;
-        gf_boolean_t active;
-};
-
-typedef struct {
-        size_t block_size;
-
-        gf_lock_t lock;
-        unsigned int loops_running;
-        off_t offset;
-
-        int32_t total_blocks;
-        int32_t diff_blocks;
-
-        struct sh_diff_loop_state **loops;
-} afr_sh_algo_diff_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 21134d9f7..4dac83113 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.c
+++ b/xlators/cluster/afr/src/afr-self-heal-common.c
@@ -1,2168 +1,1009 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.h"
-#include "xlator.h"
-#include "byte-order.h"
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
 
 #include "afr.h"
-#include "afr-transaction.h"
-#include "afr-self-heal-common.h"
 #include "afr-self-heal.h"
-#include "pump.h"
+#include "byte-order.h"
 
-//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
+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;
-
-        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);
-        }
+	afr_local_t *local = NULL;
+
+	local = frame->local;
+
+	syncbarrier_wake (&local->barrier);
+
+	return 0;
 }
 
-/**
- * select_source - select a source and return it
- */
 
 int
-afr_sh_select_source (int sources[], int child_count)
+afr_selfheal_post_op (call_frame_t *frame, xlator_t *this, inode_t *inode,
+		      int subvol, dict_t *xattr)
 {
-        int i = 0;
-        for (i = 0; i < child_count; i++)
-                if (sources[i])
-                        return i;
+	afr_private_t *priv = NULL;
+	afr_local_t *local = NULL;
+	loc_t loc = {0, };
 
-        return -1;
-}
+	priv = this->private;
+	local = frame->local;
 
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.gfid, inode->gfid);
 
-/**
- * sink_count - return number of sinks in sources array
- */
+	STACK_WIND (frame, afr_selfheal_post_op_cbk, priv->children[subvol],
+		    priv->children[subvol]->fops->xattrop, &loc,
+		    GF_XATTROP_ADD_ARRAY, xattr, NULL);
 
-int
-afr_sh_sink_count (int sources[], int child_count)
-{
-        int i = 0;
-        int sinks = 0;
-        for (i = 0; i < child_count; i++)
-                if (!sources[i])
-                        sinks++;
-        return sinks;
+	syncbarrier_wait (&local->barrier, 1);
+
+	return 0;
 }
 
-int
-afr_sh_source_count (int sources[], int child_count)
+
+dict_t *
+afr_selfheal_output_xattr (xlator_t *this, afr_transaction_type type,
+			   int *output_dirty, int **output_matrix, int subvol)
 {
-        int i = 0;
-        int nsource = 0;
+	dict_t *xattr = NULL;
+	afr_private_t *priv = NULL;
+	int j = 0;
+	int idx = 0;
+	int ret = 0;
+	int *raw = 0;
 
-        for (i = 0; i < child_count; i++)
-                if (sources[i])
-                        nsource++;
-        return nsource;
-}
+	priv = this->private;
+	idx = afr_index_for_transaction_type (type);
 
-void
-afr_sh_set_error (afr_self_heal_t *sh, int32_t op_errno)
-{
-        sh->op_ret = -1;
-        if (afr_error_more_important (sh->op_errno, op_errno))
-                sh->op_errno = op_errno;
-}
+	xattr = dict_new ();
+	if (!xattr)
+		return NULL;
 
-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]);
-                }
-                sprintf (ptr, "]");
-                gf_log (this->name, GF_LOG_TRACE,
-                        "pending_matrix: %s", buf);
-        }
-
-        GF_FREE (buf);
-}
+	if (output_dirty[subvol]) {
+		/* clear dirty */
+		raw = GF_CALLOC (sizeof(int), AFR_NUM_CHANGE_LOGS, gf_afr_mt_int32_t);
+		if (!raw)
+			goto err;
 
-void
-afr_init_pending_matrix (int32_t **pending_matrix, size_t child_count)
-{
-        int             i   = 0;
-        int             j   = 0;
+		raw[idx] = hton32 (output_dirty[subvol]);
+		ret = dict_set_bin (xattr, AFR_DIRTY, raw,
+				    sizeof(int) * AFR_NUM_CHANGE_LOGS);
+		if (ret)
+			goto err;
+	}
 
-        GF_ASSERT (pending_matrix);
+	/* clear/set pending */
+	for (j = 0; j < priv->child_count; j++) {
+		if (!output_matrix[subvol][j])
+			continue;
 
-        for (i = 0; i < child_count; i++) {
-                for (j = 0; j < child_count; j++) {
-                        pending_matrix[i][j] = 0;
-                }
-        }
-}
+		raw = GF_CALLOC (sizeof(int), AFR_NUM_CHANGE_LOGS,
+				 gf_afr_mt_int32_t);
+		if (!raw)
+			goto err;
 
-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;
-                        }
-                }
-        }
+		raw[idx] = hton32 (output_matrix[subvol][j]);
+
+		ret = dict_set_bin (xattr, priv->pending_key[j],
+				    raw, sizeof(int) * AFR_NUM_CHANGE_LOGS);
+		if (ret)
+			goto err;
+	}
+
+	return xattr;
+err:
+	if (xattr)
+		dict_unref (xattr);
+	return NULL;
 }
 
+
 int
-afr_build_pending_matrix (char **pending_key, int32_t **pending_matrix,
-                          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;
-        unsigned char *ignorant_subvols = NULL;
-
-        ignorant_subvols = GF_CALLOC (sizeof (*ignorant_subvols), child_count,
-                                      gf_afr_mt_char);
-        if (NULL == ignorant_subvols)
-                goto out;
-
-        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.
-                                 */
-
-                                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_mark_ignorant_subvols_as_pending (pending_matrix,
-                                              ignorant_subvols,
-                                              child_count);
-        GF_FREE (ignorant_subvols);
-out:
-        return ret;
+afr_selfheal_undo_pending (call_frame_t *frame, xlator_t *this, inode_t *inode,
+			   unsigned char *sources, unsigned char *sinks,
+			   unsigned char *healed_sinks, afr_transaction_type type,
+			   struct afr_reply *replies, unsigned char *locked_on)
+{
+	afr_private_t *priv = NULL;
+	int i = 0;
+	int j = 0;
+	unsigned char *pending = NULL;
+	int *input_dirty = NULL;
+	int **input_matrix = NULL;
+	int *output_dirty = NULL;
+	int **output_matrix = NULL;
+	dict_t *xattr = NULL;
+
+	priv = this->private;
+
+	pending = alloca0 (priv->child_count);
+
+	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);
+
+	afr_selfheal_extract_xattr (this, replies, type, input_dirty,
+				    input_matrix);
+
+	for (i = 0; i < priv->child_count; i++)
+		if (sinks[i] && !healed_sinks[i])
+			pending[i] = 1;
+
+	for (i = 0; i < priv->child_count; i++) {
+		for (j = 0; j < priv->child_count; j++) {
+			if (pending[j])
+				output_matrix[i][j] = 1;
+			else
+				output_matrix[i][j] = -input_matrix[i][j];
+		}
+	}
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!pending[i])
+			output_dirty[i] = -input_dirty[i];
+	}
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!locked_on[i])
+			/* perform post-op only on subvols we had locked
+			   and inspected on.
+			*/
+			continue;
+
+		xattr = afr_selfheal_output_xattr (this, type, output_dirty,
+						   output_matrix, i);
+		if (!xattr) {
+			gf_log (this->name, GF_LOG_ERROR,
+				"unable to allocate xdata for subvol %d", i);
+			continue;
+		}
+
+		afr_selfheal_post_op (frame, this, inode, i, xattr);
+
+		dict_unref (xattr);
+	}
+
+	return 0;
 }
 
-typedef enum {
-        AFR_NODE_INNOCENT,
-        AFR_NODE_FOOL,
-        AFR_NODE_WISE,
-        AFR_NODE_INVALID = -1,
-} afr_node_type;
 
-typedef struct {
-        afr_node_type type;
-        int           wisdom;
-} afr_node_character;
+void
+afr_replies_copy (struct afr_reply *dst, struct afr_reply *src, int count)
+{
+	int i = 0;
+	dict_t *xdata = NULL;
+
+	if (dst == src)
+		return;
+
+	for (i = 0; i < count; i++) {
+		dst[i].valid = src[i].valid;
+		dst[i].op_ret = src[i].op_ret;
+		dst[i].op_errno = src[i].op_errno;
+		dst[i].prestat = src[i].prestat;
+		dst[i].poststat = src[i].poststat;
+		dst[i].preparent = src[i].preparent;
+		dst[i].postparent = src[i].postparent;
+		dst[i].preparent2 = src[i].preparent2;
+		dst[i].postparent2 = src[i].postparent2;
+		if (src[i].xdata)
+			xdata = dict_ref (src[i].xdata);
+		else
+			xdata = NULL;
+		if (dst[i].xdata)
+			dict_unref (dst[i].xdata);
+		dst[i].xdata = xdata;
+		memcpy (dst[i].checksum, src[i].checksum,
+			MD5_DIGEST_LENGTH);
+	}
+}
 
 
-static int
-afr_sh_is_innocent (int32_t *array, int child_count)
+int
+afr_selfheal_fill_dirty (xlator_t *this, int *dirty, int subvol,
+			 int idx, dict_t *xdata)
 {
-        int i   = 0;
-        int ret = 1;   /* innocent until proven guilty */
+	void *pending_raw = NULL;
+	int pending[3] = {0, };
 
-        for (i = 0; i < child_count; i++) {
-                if (array[i]) {
-                        ret = 0;
-                        break;
-                }
-        }
+	if (dict_get_ptr (xdata, AFR_DIRTY, &pending_raw))
+		return -1;
 
-        return ret;
-}
+	if (!pending_raw)
+		return -1;
 
+	memcpy (pending, pending_raw, sizeof(pending));
 
-static int
-afr_sh_is_fool (int32_t *array, int i, int child_count)
-{
-        return array[i];   /* fool if accuses itself */
+	dirty[subvol] = ntoh32 (pending[idx]);
+
+	return 0;
 }
 
 
-static int
-afr_sh_is_wise (int32_t *array, int i, int child_count)
+int
+afr_selfheal_fill_matrix (xlator_t *this, int **matrix, int subvol,
+			  int idx, dict_t *xdata)
 {
-        return !array[i];  /* wise if does not accuse itself */
-}
+	int i = 0;
+	void *pending_raw = NULL;
+	int pending[3] = {0, };
+	afr_private_t *priv = NULL;
 
+	priv = this->private;
 
-static int
-afr_sh_all_nodes_innocent (afr_node_character *characters,
-                           int child_count)
-{
-        int i   = 0;
-        int ret = 1;
+	for (i = 0; i < priv->child_count; i++) {
+		if (dict_get_ptr (xdata, priv->pending_key[i], &pending_raw))
+			continue;
+
+		if (!pending_raw)
+			continue;
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_INNOCENT) {
-                        ret = 0;
-                        break;
-                }
-        }
+		memcpy (pending, pending_raw, sizeof(pending));
 
-        return ret;
+		matrix[subvol][i] = ntoh32 (pending[idx]);
+	}
+
+	return 0;
 }
 
 
-static int
-afr_sh_wise_nodes_exist (afr_node_character *characters, int child_count)
+int
+afr_selfheal_extract_xattr (xlator_t *this, struct afr_reply *replies,
+			    afr_transaction_type type, int *dirty, int **matrix)
 {
-        int i   = 0;
-        int ret = 0;
+	afr_private_t *priv = NULL;
+	int i = 0;
+	dict_t *xdata = NULL;
+	int idx = -1;
+
+	idx = afr_index_for_transaction_type (type);
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type == AFR_NODE_WISE) {
-                        ret = 1;
-                        break;
-                }
-        }
+	priv = this->private;
 
-        return ret;
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].xdata)
+			continue;
+
+		xdata = replies[i].xdata;
+
+		afr_selfheal_fill_dirty (this, dirty, i, idx, xdata);
+		afr_selfheal_fill_matrix (this, matrix, i, idx, xdata);
+	}
+
+	return 0;
 }
 
 
+
 /*
- * 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.
+ * 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.
  *
- * If no nodes with wisdom 1 exist, a split-brain has occured.
  */
 
-static void
-afr_sh_compute_wisdom (int32_t *pending_matrix[],
-                       afr_node_character characters[], int child_count)
-{
-        int i = 0;
-        int j = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type == AFR_NODE_WISE) {
-                        characters[i].wisdom = 1;
-
-                        for (j = 0; j < child_count; j++) {
-                                if ((characters[j].type == AFR_NODE_WISE)
-                                    && pending_matrix[j][i]) {
-
-                                        characters[i].wisdom = 0;
-                                }
-                        }
-                }
-        }
+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)
+{
+	afr_private_t *priv = NULL;
+	int i = 0;
+	int j = 0;
+	int *dirty = NULL;
+	int **matrix = NULL;
+	char *accused = NULL;
+
+	priv = this->private;
+
+	dirty = alloca0 (priv->child_count * sizeof (int));
+	accused = alloca0 (priv->child_count);
+	matrix = ALLOC_MATRIX(priv->child_count, int);
+
+	/* First construct the pending matrix for further analysis */
+	afr_selfheal_extract_xattr (this, replies, type, dirty, matrix);
+
+	/* Next short list all accused to exclude them from being sources */
+	for (i = 0; i < priv->child_count; i++) {
+		for (j = 0; j < priv->child_count; j++) {
+			if (matrix[i][j])
+				accused[j] = 1;
+		}
+	}
+
+	/* Short list all non-accused as sources */
+	memset (sources, 0, priv->child_count);
+	for (i = 0; i < priv->child_count; i++) {
+		if (!accused[i] && locked_on[i])
+			sources[i] = 1;
+	}
+
+	/* Everyone accused by sources are sinks */
+	memset (sinks, 0, priv->child_count);
+	for (i = 0; i < priv->child_count; i++) {
+		if (!sources[i])
+			continue;
+		for (j = 0; j < priv->child_count; j++) {
+			if (matrix[i][j])
+				sinks[j] = 1;
+		}
+	}
+
+	/* If any source has 'dirty' bit, pick first
+	   'dirty' source and make everybody else sinks */
+	for (i = 0; i < priv->child_count; i++) {
+		if (sources[i] && dirty[i]) {
+			for (j = 0; j < priv->child_count; j++) {
+				if (j != i) {
+					sources[j] = 0;
+					sinks[j] = 1;
+				}
+			}
+			break;
+		}
+	}
+
+	/* 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;
+		}
+	}
+
+	return 0;
 }
 
 
-static int
-afr_sh_wise_nodes_conflict (afr_node_character *characters,
-                            int child_count)
+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)
 {
-        int i   = 0;
-        int ret = 1;
+	afr_local_t *local = NULL;
+	int i = -1;
 
-        for (i = 0; i < child_count; i++) {
-                if ((characters[i].type == AFR_NODE_WISE)
-                    && characters[i].wisdom == 1) {
+	local = frame->local;
+	i = (long) cookie;
 
-                        /* There is atleast one bona-fide wise node */
-                        ret = 0;
-                        break;
-                }
-        }
+	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);
 
-        return ret;
+	syncbarrier_wake (&local->barrier);
+
+	return 0;
 }
 
 
-static int
-afr_sh_mark_wisest_as_sources (int sources[],
-                               afr_node_character *characters,
-                               int child_count)
+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)
 {
-        int nsources = 0;
-        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;
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].wisdom == 1) {
-                        sources[i] = 1;
-                        nsources++;
-                }
-        }
+	local = frame->local;
+	priv = frame->this->private;
 
-        return nsources;
-}
+	xattr_req = dict_new ();
+	if (!xattr_req)
+		return NULL;
 
-static void
-afr_compute_witness_of_fools (int32_t *witnesses, int32_t **pending_matrix,
-                              afr_node_character *characters,
-                              int32_t child_count)
-{
-        int i       = 0;
-        int j       = 0;
-        int witness = 0;
-
-        GF_ASSERT (witnesses);
-        GF_ASSERT (pending_matrix);
-        GF_ASSERT (characters);
-        GF_ASSERT (child_count > 0);
-
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_FOOL)
-                        continue;
-
-                witness = 0;
-                for (j = 0; j < child_count; j++) {
-                        if (i == j)
-                                continue;
-                        witness += pending_matrix[i][j];
-                }
-                witnesses[i] = witness;
-        }
-}
+	if (afr_xattr_req_prepare (frame->this, xattr_req) != 0) {
+		dict_destroy (xattr_req);
+		return NULL;
+	}
 
-static int32_t
-afr_find_biggest_witness_among_fools (int32_t *witnesses,
-                                      afr_node_character *characters,
-                                      int32_t child_count)
-{
-        int i               = 0;
-        int biggest_witness = -1;
+	inode = inode_new (parent->table);
+	if (!inode) {
+		dict_destroy (xattr_req);
+		return NULL;
+	}
 
-        GF_ASSERT (witnesses);
-        GF_ASSERT (characters);
-        GF_ASSERT (child_count > 0);
+	loc.parent = inode_ref (parent);
+	uuid_copy (loc.pargfid, parent->gfid);
+	loc.name = name;
+	loc.inode = inode_ref (inode);
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_FOOL)
-                        continue;
+	AFR_ONLIST (lookup_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+		    xattr_req);
 
-                if (biggest_witness < witnesses[i])
-                        biggest_witness = witnesses[i];
-        }
-        return biggest_witness;
-}
+	afr_replies_copy (replies, local->replies, priv->child_count);
 
-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;
-}
+	loc_wipe (&loc);
+	dict_unref (xattr_req);
 
-static int
-afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix,
-                                     afr_node_character *characters,
-                                     int child_count)
-{
-        int32_t       biggest_witness = 0;
-        int           nsources        = 0;
-        int32_t       *witnesses      = NULL;
-
-        GF_ASSERT (child_count > 0);
-
-        witnesses = GF_CALLOC (child_count, sizeof (*witnesses),
-                               gf_afr_mt_int32_t);
-        if (NULL == witnesses) {
-                nsources = -1;
-                goto out;
-        }
-
-        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:
-        if (witnesses)
-                GF_FREE (witnesses);
-        return nsources;
+	return inode;
 }
 
-int
-afr_mark_child_as_source_by_uid (int32_t *sources, struct iatt *bufs,
-                                 int32_t *valid_children, int child_count,
-                                 uint32_t uid)
-{
-        int     i        = 0;
-        int     nsources = 0;
-        int     child    = 0;
-
-        GF_ASSERT (bufs);
-        GF_ASSERT (valid_children);
-        GF_ASSERT (sources);
-        GF_ASSERT (child_count > 0);
-
-        for (i = 0; i < child_count; i++) {
-                if (-1 == valid_children[i])
-                        continue;
-
-                child = valid_children[i];
-                if (uid == bufs[child].ia_uid) {
-                        sources[child] = 1;
-                        nsources++;
-                }
-        }
-        return nsources;
-}
 
 int
-afr_get_child_with_lowest_uid (struct iatt *bufs, int32_t *valid_children,
-                               int child_count)
+afr_selfheal_unlocked_discover_on (call_frame_t *frame, inode_t *inode,
+				   uuid_t gfid, struct afr_reply *replies,
+				   unsigned char *discover_on)
 {
-        int     i        = 0;
-        int     smallest = -1;
-        int     child    = 0;
-
-        GF_ASSERT (bufs);
-        GF_ASSERT (valid_children);
-        GF_ASSERT (child_count > 0);
-
-        for (i = 0; i < child_count; i++) {
-                if (-1 == valid_children[i])
-                        continue;
-                child = valid_children[i];
-                if ((smallest == -1) ||
-                    (bufs[child].ia_uid < bufs[smallest].ia_uid)) {
-                        smallest = child;
-                }
-        }
-        return smallest;
-}
+	loc_t loc = {0, };
+	dict_t *xattr_req = NULL;
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
 
-static int
-afr_sh_mark_lowest_uid_as_source (struct iatt *bufs, int32_t *valid_children,
-                                  int child_count, int32_t *sources)
-{
-        int   nsources              = 0;
-        int   smallest              = 0;
-
-        smallest = afr_get_child_with_lowest_uid (bufs, valid_children,
-                                                  child_count);
-        if (smallest < 0) {
-                nsources = -1;
-                goto out;
-        }
-        nsources = afr_mark_child_as_source_by_uid (sources, bufs,
-                                                    valid_children, child_count,
-                                                    bufs[smallest].ia_uid);
-out:
-        return nsources;
-}
+	local = frame->local;
+	priv = frame->this->private;
 
-char *
-afr_get_character_str (afr_node_type type)
-{
-        char *character = NULL;
-
-        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;
-}
+	xattr_req = dict_new ();
+	if (!xattr_req)
+		return -ENOMEM;
 
-afr_node_type
-afr_find_child_character_type (int32_t *pending_row, int32_t child,
-                               int32_t child_count, const char *xlator_name)
-{
-        afr_node_type type = AFR_NODE_INVALID;
-
-        GF_ASSERT (pending_row);
-        GF_ASSERT (child_count > 0);
-        GF_ASSERT ((child >= 0) && (child < child_count));
-
-        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;
-        else
-                GF_ASSERT (0);
-
-        gf_log (xlator_name, GF_LOG_DEBUG, "child %d character %s",
-                child, afr_get_character_str (type));
-        return type;
-}
+	if (afr_xattr_req_prepare (frame->this, xattr_req) != 0) {
+		dict_destroy (xattr_req);
+		return -ENOMEM;
+	}
 
-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)
-{
-        afr_private_t           *priv = NULL;
-        afr_self_heal_type      sh_type    = AFR_SELF_HEAL_INVALID;
-        int                     nsources   = -1;
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.gfid, gfid);
 
-        priv = this->private;
+	AFR_ONLIST (discover_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+		    xattr_req);
 
-        if (afr_get_children_count (success_children, priv->child_count) == 0)
-                goto out;
+	afr_replies_copy (replies, local->replies, priv->child_count);
 
-        afr_build_pending_matrix (priv->pending_key, pending_matrix,
-                                  xattr, type, priv->child_count);
+	loc_wipe (&loc);
+	dict_unref (xattr_req);
 
-        sh_type = afr_self_heal_type_for_transaction (type);
-        if (AFR_SELF_HEAL_INVALID == sh_type)
-                goto out;
+	return 0;
+}
 
-        afr_sh_print_pending_matrix (pending_matrix, this);
+int
+afr_selfheal_unlocked_discover (call_frame_t *frame, inode_t *inode,
+				uuid_t gfid, struct afr_reply *replies)
+{
+	afr_private_t *priv = NULL;
 
-        nsources = afr_mark_sources (sources, pending_matrix, bufs,
-                                     priv->child_count, sh_type,
-                                     success_children, this->name);
-out:
-        return nsources;
+	priv = frame->this->private;
+
+	return afr_selfheal_unlocked_discover_on (frame, inode, gfid, replies,
+						  priv->child_up);
 }
 
-/**
- * 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.
- *
- * All fools are sinks, unless there are no 'wise' nodes. In that case,
- * one of the fools is made a source.
- */
 
 int
-afr_mark_sources (int32_t *sources, int32_t **pending_matrix, struct iatt *bufs,
-                  int32_t child_count, afr_self_heal_type type,
-                  int32_t *valid_children, const char *xlator_name)
+afr_selfheal_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+		       int op_ret, int op_errno, dict_t *xdata)
 {
-        /* stores the 'characters' (innocent, fool, wise) of the nodes */
-
-        afr_node_character *characters =  NULL;
-        int            i              = 0;
-        int            nsources       = -1;
-        xlator_t      *this           = NULL;
-
-        characters = GF_CALLOC (sizeof (afr_node_character),
-                                child_count, gf_afr_mt_afr_node_character);
-        if (!characters)
-                goto out;
-
-        this = THIS;
-
-        /* start clean */
-        for (i = 0; i < child_count; i++) {
-                sources[i] = 0;
-        }
-
-        nsources = 0;
-        for (i = 0; i < child_count; i++) {
-                characters[i].type =
-                        afr_find_child_character_type (pending_matrix[i], i,
-                                                       child_count,
-                                                       xlator_name);
-                if (AFR_NODE_INVALID == characters[i].type)
-                        gf_log (xlator_name, GF_LOG_WARNING,
-                                "child %d had invalid xattrs", i);
-        }
-
-        if ((type == AFR_SELF_HEAL_METADATA)
-            && afr_sh_all_nodes_innocent (characters, child_count)) {
-
-                nsources = afr_sh_mark_lowest_uid_as_source (bufs,
-                                                             valid_children,
-                                                             child_count,
-                                                             sources);
-                goto out;
-        }
-
-        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)) {
-                        /* split-brain */
-                        gf_log (this->name, GF_LOG_INFO,
-                                "split-brain possible, no source detected");
-                        nsources = -1;
-
-                } else {
-                        nsources = afr_sh_mark_wisest_as_sources (sources,
-                                                                  characters,
-                                                                  child_count);
-                }
-        } else {
-                nsources = afr_mark_biggest_of_fools_as_source (sources,
-                                                                pending_matrix,
-                                                                characters,
-                                                                child_count);
-        }
+	afr_local_t *local = NULL;
+	int i = 0;
 
-out:
-        if (nsources == 0) {
-                for (i = 0; i < child_count; i++) {
-                        if (valid_children[i] != -1)
-                                sources[valid_children[i]] = 1;
-                }
-        }
-        if (characters)
-                GF_FREE (characters);
-
-        gf_log (this->name, GF_LOG_DEBUG, "Number of sources: %d", nsources);
-        return nsources;
-}
+	local = frame->local;
+	i = (long) cookie;
 
-void
-afr_sh_pending_to_delta (afr_private_t *priv, dict_t **xattr,
-                         int32_t *delta_matrix[], int success[],
-                         int child_count, afr_transaction_type type)
-{
-        /* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
-        int32_t  pending[3]  = {0,};
-        void    *pending_raw = NULL;
-        int      ret         = 0;
-        int      i           = 0;
-        int      j           = 0;
-        int      k           = 0;
-
-        /* start clean */
-        for (i = 0; i < child_count; i++) {
-                for (j = 0; j < child_count; j++) {
-                        delta_matrix[i][j] = 0;
-                }
-        }
-
-        for (i = 0; i < child_count; i++) {
-                if (pending_raw)
-                        pending_raw = NULL;
-
-                for (j = 0; j < child_count; j++) {
-                        ret = dict_get_ptr (xattr[i], priv->pending_key[j],
-                                            &pending_raw);
-                        if (ret < 0)
-                                gf_log (THIS->name, GF_LOG_DEBUG,
-                                        "Unable to get dict value.");
-                        if (!success[j])
-                                continue;
-
-                        k = afr_index_for_transaction_type (type);
-
-                        if (pending_raw != NULL) {
-                                memcpy (pending, pending_raw, sizeof(pending));
-                                delta_matrix[i][j] = -(ntoh32 (pending[k]));
-                        } else {
-                                delta_matrix[i][j]  = 0;
-                        }
-
-                }
-        }
+	local->replies[i].valid = 1;
+	local->replies[i].op_ret = op_ret;
+	local->replies[i].op_errno = op_errno;
+
+	syncbarrier_wake (&local->barrier);
+
+	return 0;
 }
 
 
 int
-afr_sh_delta_to_xattr (afr_private_t *priv,
-                       int32_t *delta_matrix[], dict_t *xattr[],
-                       int child_count, afr_transaction_type type)
+afr_selfheal_locked_fill (call_frame_t *frame, xlator_t *this,
+			  unsigned char *locked_on)
 {
-        int      i       = 0;
-        int      j       = 0;
-        int      k       = 0;
-        int      ret     = 0;
-        int32_t *pending = NULL;
-
-        for (i = 0; i < child_count; i++) {
-                if (!xattr[i])
-                        continue;
-
-                for (j = 0; j < child_count; j++) {
-                        pending = GF_CALLOC (sizeof (int32_t), 3,
-                                             gf_afr_mt_int32_t);
-
-                        if (!pending)
-                                continue;
-                        /* 3 = data+metadata+entry */
-
-                        k = afr_index_for_transaction_type (type);
-
-                        pending[k] = hton32 (delta_matrix[i][j]);
-
-                        ret = dict_set_bin (xattr[i], priv->pending_key[j],
-                                            pending,
-                                            3 * sizeof (int32_t));
-                        if (ret < 0)
-                                gf_log (THIS->name, GF_LOG_WARNING,
-                                        "Unable to set dict value.");
-                }
-        }
-        return 0;
+	int i = 0;
+	afr_private_t *priv = NULL;
+	afr_local_t *local = NULL;
+	int count = 0;
+
+	local = frame->local;
+	priv = this->private;
+
+	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 {
+			locked_on[i] = 0;
+		}
+	}
+
+	return count;
 }
 
 
 int
-afr_sh_has_metadata_pending (dict_t *xattr, xlator_t *this)
+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)
 {
-        /* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
-        int32_t        pending[3]  = {0,};
-        void          *pending_raw = NULL;
-        afr_private_t *priv        = NULL;
-        int            ret         = -1;
-        int            i           = 0;
-        int            j           = 0;
+	loc_t loc = {0,};
+	struct gf_flock flock = {0, };
 
-        priv = this->private;
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.gfid, inode->gfid);
 
-        for (i = 0; i < priv->child_count; i++) {
-                ret = dict_get_ptr (xattr, priv->pending_key[i],
-                                    &pending_raw);
+	flock.l_type = F_WRLCK;
+	flock.l_start = off;
+	flock.l_len = size;
 
-                if (ret != 0)
-                        return 0;
+	AFR_ONALL (frame, afr_selfheal_lock_cbk, inodelk, dom,
+		    &loc, F_SETLK, &flock, NULL);
 
-                memcpy (pending, pending_raw, sizeof(pending));
-                j = afr_index_for_transaction_type (AFR_METADATA_TRANSACTION);
+	loc_wipe (&loc);
 
-                if (pending[j])
-                        return 1;
-        }
-
-        return 0;
+	return afr_selfheal_locked_fill (frame, this, locked_on);
 }
 
 
 int
-afr_sh_has_data_pending (dict_t *xattr, xlator_t *this)
+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)
 {
-        /* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
-        int32_t        pending[3]  = {0,};
-        void          *pending_raw = NULL;
-        afr_private_t *priv        = NULL;
-        int            ret         = -1;
-        int            i           = 0;
-        int            j           = 0;
+	loc_t loc = {0,};
+	struct gf_flock flock = {0, };
+	afr_local_t *local = NULL;
+	int i = 0;
+	afr_private_t *priv = NULL;
+
+	priv = this->private;
+	local = frame->local;
+
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.gfid, inode->gfid);
 
-        priv = this->private;
+	flock.l_type = F_WRLCK;
+	flock.l_start = off;
+	flock.l_len = size;
 
-        for (i = 0; i < priv->child_count; i++) {
-                ret = dict_get_ptr (xattr, priv->pending_key[i],
-                                    &pending_raw);
+	AFR_ONALL (frame, afr_selfheal_lock_cbk, inodelk, dom,
+		    &loc, F_SETLK, &flock, NULL);
 
-                if (ret != 0)
-                        return 0;
+	for (i = 0; i < priv->child_count; i++) {
+		if (local->replies[i].op_ret == -1 &&
+		    local->replies[i].op_errno == EAGAIN) {
+			afr_selfheal_locked_fill (frame, this, locked_on);
+			afr_selfheal_uninodelk (frame, this, inode, dom, off,
+						size, locked_on);
 
-                memcpy (pending, pending_raw, sizeof(pending));
-                j = afr_index_for_transaction_type (AFR_DATA_TRANSACTION);
+			AFR_SEQ (frame, afr_selfheal_lock_cbk, inodelk, dom,
+				 &loc, F_SETLKW, &flock, NULL);
+			break;
+		}
+	}
 
-                if (pending[j])
-                        return 1;
-        }
+	loc_wipe (&loc);
 
-        return 0;
+	return afr_selfheal_locked_fill (frame, this, locked_on);
 }
 
 
 int
-afr_sh_has_entry_pending (dict_t *xattr, xlator_t *this)
+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)
 {
-        /* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
-        int32_t        pending[3]  = {0,};
-        void          *pending_raw = NULL;
-        afr_private_t *priv        = NULL;
-        int            ret         = -1;
-        int            i           = 0;
-        int            j           = 0;
+	loc_t loc = {0,};
+	struct gf_flock flock = {0, };
 
-        priv = this->private;
 
-        for (i = 0; i < priv->child_count; i++) {
-                ret = dict_get_ptr (xattr, priv->pending_key[i],
-                                    &pending_raw);
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.gfid, inode->gfid);
 
-                if (ret != 0)
-                        return 0;
+	flock.l_type = F_UNLCK;
+	flock.l_start = off;
+	flock.l_len = size;
 
-                memcpy (pending, pending_raw, sizeof(pending));
-                j = afr_index_for_transaction_type (AFR_ENTRY_TRANSACTION);
+	AFR_ONLIST (locked_on, frame, afr_selfheal_lock_cbk, inodelk,
+		    dom, &loc, F_SETLK, &flock, NULL);
 
-                if (pending[j])
-                        return 1;
-        }
+	loc_wipe (&loc);
 
-        return 0;
+	return 0;
 }
 
 
-/**
- * is_matrix_zero - return true if pending matrix is all zeroes
- */
-
 int
-afr_sh_is_matrix_zero (int32_t *pending_matrix[], int child_count)
+afr_selfheal_tryentrylk (call_frame_t *frame, xlator_t *this, inode_t *inode,
+			 char *dom, const char *name, unsigned char *locked_on)
 {
-        int i = 0;
-        int j = 0;
-
-        for (i = 0; i < child_count; i++)
-                for (j = 0; j < child_count; j++)
-                        if (pending_matrix[i][j])
-                                return 0;
-        return 1;
+	loc_t loc = {0,};
+
+	loc.inode = inode_ref (inode);
+	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_selfheal_locked_fill (frame, this, locked_on);
 }
 
 
 int
-afr_sh_missing_entries_done (call_frame_t *frame, xlator_t *this)
+afr_selfheal_entrylk (call_frame_t *frame, xlator_t *this, inode_t *inode,
+		      char *dom, const char *name, unsigned char *locked_on)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              i = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-//      memset (sh->child_errno, 0, sizeof (int) * priv->child_count);
-        memset (sh->buf, 0, sizeof (struct iatt) * priv->child_count);
-
-        for (i = 0; i < priv->child_count; i++) {
-                sh->locked_nodes[i] = 0;
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->xattr[i])
-                        dict_unref (sh->xattr[i]);
-                sh->xattr[i] = NULL;
-        }
-
-        if (local->govinda_gOvinda || sh->op_failed) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "split brain found, aborting selfheal of %s",
-                        local->loc.path);
-                sh->op_failed = 1;
-                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);
-        }
-
-        return 0;
-}
+	loc_t loc = {0,};
+	afr_local_t *local = NULL;
+	int i = 0;
+	afr_private_t *priv = NULL;
 
+	priv = this->private;
+	local = frame->local;
 
-static int
-afr_sh_missing_entries_finish (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.gfid, inode->gfid);
 
-        local = frame->local;
-        int_lock = &local->internal_lock;
+	AFR_ONALL (frame, afr_selfheal_lock_cbk, entrylk, dom, &loc,
+		   name, ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL);
 
-        int_lock->lock_cbk = afr_sh_missing_entries_done;
-        afr_unlock (frame, this);
+	for (i = 0; i < priv->child_count; i++) {
+		if (local->replies[i].op_ret == -1 &&
+		    local->replies[i].op_errno == EAGAIN) {
+			afr_selfheal_locked_fill (frame, this, locked_on);
+			afr_selfheal_unentrylk (frame, this, inode, dom, name,
+						locked_on);
 
-        return 0;
-}
+			AFR_SEQ (frame, afr_selfheal_lock_cbk, entrylk, dom,
+				 &loc, name, ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
+			break;
+		}
+	}
 
-static 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)
-{
-        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->parentbuf        = *postparent;
-                        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_ERROR,
-                                "path %s on subvolume %s => -1 (%s)",
-                                local->loc.path,
-                                priv->children[child_index]->name,
-                                strerror (op_errno));
-                        local->self_heal.child_errno[child_index] = op_errno;
-                }
-        }
-        UNLOCK (&frame->lock);
-        return;
-}
+	loc_wipe (&loc);
 
-gf_boolean_t
-afr_valid_ia_type (ia_type_t ia_type)
-{
-        switch (ia_type) {
-        case IA_IFSOCK:
-        case IA_IFREG:
-        case IA_IFBLK:
-        case IA_IFCHR:
-        case IA_IFIFO:
-        case IA_IFLNK:
-        case IA_IFDIR:
-                return _gf_true;
-        default:
-                return _gf_false;
-        }
-        return _gf_false;
+	return afr_selfheal_locked_fill (frame, this, locked_on);
 }
 
-void
-afr_sh_call_entry_impunge_recreate (call_frame_t *frame, xlator_t *this,
-                                    int child_index, 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;
-        int32_t         op_errno = 0;
-
-        impunge_frame = copy_frame (frame);
-        if (!impunge_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        ALLOC_OR_GOTO (impunge_local, afr_local_t, out);
-
-        local = frame->local;
-        sh = &local->self_heal;
-        impunge_frame->local = impunge_local;
-        impunge_sh = &impunge_local->self_heal;
-        impunge_sh->sh_frame = frame;
-        impunge_sh->active_source = sh->source;
-        impunge_sh->impunging_entry_mode = st_mode_from_ia (buf->ia_prot,
-                                                            buf->ia_type);
-        impunge_sh->impunge_ret_child = child_index;
-        loc_copy (&impunge_local->loc, &local->loc);
-        sh->impunge_done = impunge_done;
-        impunge_local->call_count = 1;
-        afr_sh_entry_impunge_create (impunge_frame, this, child_index, buf,
-                                     postparent);
-        return;
-out:
-        gf_log (this->name, GF_LOG_ERROR, "impunge of %s failed, reason: %s",
-                local->loc.path, strerror (op_errno));
-        impunge_done (frame, this, child_index, -1, op_errno);
-}
 
 int
-afr_sh_create_entry_cbk (call_frame_t *frame, xlator_t *this, int child,
-                         int32_t op_ret, int32_t op_errno)
+afr_selfheal_unentrylk (call_frame_t *frame, xlator_t *this, inode_t *inode,
+			char *dom, const char *name, unsigned char *locked_on)
 {
-        int             call_count = 0;
-        afr_local_t     *local = NULL;
-
-        local = frame->local;
-
-        if (op_ret == -1)
-                gf_log (this->name, GF_LOG_ERROR,
-                        "create entry %s failed, on child %d reason, %s",
-                        local->loc.path, child, strerror (op_errno));
-        call_count = afr_frame_return (frame);
-        if (call_count == 0)
-                afr_sh_missing_entries_finish (frame, this);
-        return 0;
-}
+	loc_t loc = {0,};
 
-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;
-        afr_private_t   *priv = NULL;
-        int             enoent_count = 0;
-        int             i = 0;
-        struct iatt     *buf = NULL;
-        struct iatt     *postparent = NULL;
-
-        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 == 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "no missing files - %s. proceeding to metadata check",
-                        local->loc.path);
-                /* proceed to next step - metadata self-heal */
-                afr_sh_missing_entries_finish (frame, this);
-                return 0;
-        }
-
-        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);
-                local->govinda_gOvinda = 1;
-                afr_sh_missing_entries_finish (frame, this);
-                goto out;
-        }
-
-        local->call_count = enoent_count;
-        for (i = 0; i < priv->child_count; i++) {
-                //If !child_up errno will be zero
-                if (sh->child_errno[i] != ENOENT)
-                        continue;
-                afr_sh_call_entry_impunge_recreate (frame, this, i,
-                                                    buf, postparent,
-                                                    afr_sh_create_entry_cbk);
-                enoent_count--;
-        }
-        GF_ASSERT (enoent_count == 0);
-out:
-        return 0;
-}
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.gfid, inode->gfid);
 
-void
-afr_sh_missing_entries_lookup_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int32_t         op_errno = 0;
-        ia_type_t       ia_type = IA_INVAL;
-        int32_t         nsources = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        if (afr_get_children_count (sh->success_children,
-                                    priv->child_count) == 0) {
-                op_errno = afr_resultant_errno_get (NULL, sh->child_errno,
-                                                    priv->child_count);
-                goto out;
-        }
-
-        if (afr_gfid_missing_count (this->name, sh->success_children,
-                                    sh->buf, priv->child_count,
-                                    local->loc.path) ||
-            afr_conflicting_iattrs (sh->buf, sh->success_children,
-                                    priv->child_count, local->loc.path,
-                                    this->name)) {
-                //this can happen if finding the fresh parent dir failed
-                local->govinda_gOvinda = 1;
-                sh->op_failed = 1;
-                op_errno = EIO;
-                goto out;
-        }
-
-        //now No chance for the ia_type to conflict
-        ia_type = sh->buf[sh->success_children[0]].ia_type;
-        nsources = afr_build_sources (this, sh->xattr, sh->buf,
-                                      sh->pending_matrix, sh->sources,
-                                      sh->success_children,
-                                      afr_transaction_type_get (ia_type));
-        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);
-                op_errno = EIO;
-                goto out;
-        }
-
-        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;
-        }
-
-        if (sh->gfid_sh_success_cbk)
-                sh->gfid_sh_success_cbk (frame, this);
-        sh_missing_entries_create (frame, this);
-        return;
-out:
-        afr_sh_set_error (sh, op_errno);
-        afr_sh_missing_entries_finish (frame, this);
-        return;
+	AFR_ONLIST (locked_on, frame, afr_selfheal_lock_cbk, entrylk,
+		    dom, &loc, name, ENTRYLK_UNLOCK, ENTRYLK_WRLCK, NULL);
+
+	loc_wipe (&loc);
+
+	return 0;
 }
 
-static int
-afr_sh_missing_entries_lookup_cbk (call_frame_t *frame, void *cookie,
-                                   xlator_t *this, int32_t op_ret,
-                                   int32_t op_errno, inode_t *inode,
-                                   struct iatt *buf, dict_t *xattr,
-                                   struct iatt *postparent)
+
+gf_boolean_t
+afr_is_pending_set (xlator_t *this, dict_t *xdata, int type)
 {
-        int                     call_count = 0;
-        afr_local_t             *local = NULL;
-        afr_self_heal_t         *sh = NULL;
-        afr_private_t           *priv = NULL;
+	int idx = -1;
+	afr_private_t *priv = NULL;
+	void *pending_raw = NULL;
+	int *pending_int = NULL;
+	int i = 0;
 
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
+	priv = this->private;
+	idx = afr_index_for_transaction_type (type);
 
-        afr_sh_common_lookup_resp_handler (frame, cookie, this, op_ret,
-                                           op_errno, inode, buf, xattr,
-                                           postparent);
-        call_count = afr_frame_return (frame);
+	if (dict_get_ptr (xdata, AFR_DIRTY, &pending_raw) == 0) {
+		if (pending_raw) {
+			pending_int = pending_raw;
 
-        if (call_count == 0)
-                afr_sh_missing_entries_lookup_done (frame, this);
+			if (ntoh32 (pending_int[idx]))
+				return _gf_true;
+		}
+	}
 
-        return 0;
-}
+	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;
 
-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->op_failed = 1;
-                }
-                UNLOCK (&frame->lock);
-        }
-        call_count = afr_frame_return (frame);
-        if (call_count == 0)
-                sh->post_remove_call (frame, this);
-        return 0;
-}
+		if (ntoh32 (pending_int[idx]))
+			return _gf_true;
+	}
 
-void
-afr_sh_call_entry_expunge_remove (call_frame_t *frame, xlator_t *this,
-                                  int child_index, struct iatt *buf,
-                                  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;
-
-        expunge_frame = copy_frame (frame);
-        if (!expunge_frame) {
-                goto out;
-        }
-
-        ALLOC_OR_GOTO (expunge_local, afr_local_t, out);
-
-        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);
-        sh->expunge_done = expunge_done;
-        afr_sh_entry_expunge_remove (expunge_frame, this, child_index, buf);
-        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);
+	return _gf_false;
 }
 
-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;
-                }
-        }
-}
 
-int
-afr_sh_purge_stale_entries_done (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_is_data_set (xlator_t *this, dict_t *xdata)
 {
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-        afr_private_t       *priv     = NULL;
-
-        local    = frame->local;
-        sh       = &local->self_heal;
-        priv     = this->private;
-
-        if (sh->op_failed) {
-                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_cbk,
-                                              _gf_true);
-                } 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);
-                }
-        }
-        return 0;
+	return afr_is_pending_set (this, xdata, AFR_DATA_TRANSACTION);
 }
 
 gf_boolean_t
-afr_sh_purge_entry_condition (afr_local_t *local, afr_private_t *priv,
-                              int child)
+afr_is_metadata_set (xlator_t *this, dict_t *xdata)
 {
-        afr_self_heal_t *sh = NULL;
-
-        sh = &local->self_heal;
-
-        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;
-
-        return _gf_false;
+	return afr_is_pending_set (this, xdata, AFR_METADATA_TRANSACTION);
 }
 
 gf_boolean_t
-afr_sh_purge_stale_entry_condition (afr_local_t *local, afr_private_t *priv,
-                                    int child)
+afr_is_entry_set (xlator_t *this, dict_t *xdata)
 {
-        afr_self_heal_t *sh = NULL;
-
-        sh = &local->self_heal;
-
-        if (local->child_up[child] &&
-            (!afr_is_child_present (sh->fresh_children, priv->child_count,
-                                    child))
-             && (sh->child_errno[child] != ENOENT))
-                return _gf_true;
-
-        return _gf_false;
+	return afr_is_pending_set (this, xdata, AFR_ENTRY_TRANSACTION);
 }
 
-void
-afr_sh_purge_entry_common (call_frame_t *frame, xlator_t *this,
-                           gf_boolean_t purge_condition (afr_local_t *local,
-                                                         afr_private_t *priv,
-                                                         int child))
-{
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv = NULL;
-        afr_self_heal_t *sh = NULL;
-        int             i = 0;
-        int             call_count = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (purge_condition (local, priv, i))
-                        call_count++;
-        }
-
-        if (call_count == 0) {
-                sh->post_remove_call (frame, this);
-                goto out;
-        }
-
-        local->call_count = call_count;
-        for (i = 0; i < priv->child_count; i++) {
-                if (!purge_condition (local, priv, i))
-                        continue;
-                afr_sh_call_entry_expunge_remove (frame, this,
-                                                  (long) i, &sh->buf[i],
-                                                  afr_sh_remove_entry_cbk);
-        }
-out:
-        return;
-}
 
 void
-afr_sh_purge_entry (call_frame_t *frame, xlator_t *this)
+afr_inode_link (inode_t *inode, struct iatt *iatt)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
+	inode_t *linked_inode = NULL;
+
+	linked_inode = inode_link (inode, NULL, NULL, iatt);
 
-        local = frame->local;
-        sh = &local->self_heal;
-        sh->post_remove_call = afr_sh_missing_entries_finish;
+	uuid_copy (inode->gfid, iatt->ia_gfid);
+	inode->ia_type = iatt->ia_type;
 
-        afr_sh_purge_entry_common (frame, this, afr_sh_purge_entry_condition);
+	if (linked_inode) {
+		inode_lookup (linked_inode);
+		inode_unref (linked_inode);
+	}
 }
 
-void
-afr_sh_purge_stale_entry (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int             i = 0;
 
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
+/*
+ * 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.
+ */
 
-        sh->post_remove_call = afr_sh_purge_stale_entries_done;
+int
+afr_selfheal_unlocked_inspect (call_frame_t *frame, xlator_t *this,
+			       inode_t *inode, uuid_t gfid,
+			       gf_boolean_t *data_selfheal,
+			       gf_boolean_t *metadata_selfheal,
+			       gf_boolean_t *entry_selfheal)
+{
+	afr_private_t *priv = NULL;
+	int i = 0;
+	int valid_cnt = 0;
+	struct iatt first = {0, };
+	struct afr_reply *replies = NULL;
+	int ret = -1;
+
+	priv = this->private;
+
+	replies = alloca0 (sizeof (*replies) * priv->child_count);
+
+	ret = afr_selfheal_unlocked_discover (frame, inode, gfid, replies);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].valid)
+			continue;
+		if (replies[i].op_ret == -1)
+			continue;
+
+		if (afr_is_data_set (this, replies[i].xdata))
+			*data_selfheal = _gf_true;
+
+		if (afr_is_metadata_set (this, replies[i].xdata))
+			*metadata_selfheal = _gf_true;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (afr_is_child_present (sh->fresh_children,
-                                          priv->child_count, i))
-                        continue;
+		if (afr_is_entry_set (this, replies[i].xdata))
+			*entry_selfheal = _gf_true;
 
-                if ((!local->child_up[i]) || sh->child_errno[i] != 0)
-                        continue;
+		valid_cnt ++;
+		if (valid_cnt == 1) {
+			first = replies[i].poststat;
+			continue;
+		}
 
-                GF_ASSERT (!uuid_is_null (sh->entrybuf.ia_gfid) ||
-                           uuid_is_null (sh->buf[i].ia_gfid));
+		if (!IA_EQUAL (first, replies[i].poststat, type)) {
+			gf_log (this->name, GF_LOG_ERROR,
+				"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));
+			return -EIO;
+		}
 
-                if ((sh->entrybuf.ia_type != sh->buf[i].ia_type) ||
-                    (uuid_compare (sh->buf[i].ia_gfid,
-                                   sh->entrybuf.ia_gfid)))
-                        continue;
+		if (!IA_EQUAL (first, replies[i].poststat, uid)) {
+			gf_log (this->name, GF_LOG_DEBUG,
+				"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));
 
-                afr_fresh_children_add_child (sh->fresh_children,
-                                              i, priv->child_count);
+			*metadata_selfheal = _gf_true;
+		}
 
-        }
-        afr_sh_purge_entry_common (frame, this,
-                                   afr_sh_purge_stale_entry_condition);
-}
+		if (!IA_EQUAL (first, replies[i].poststat, gid)) {
+			gf_log (this->name, GF_LOG_DEBUG,
+				"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));
 
-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);
-}
+			*metadata_selfheal = _gf_true;
+		}
 
-void
-afr_sh_children_lookup_done (call_frame_t *frame, xlator_t *this)
-{
-        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;
-        int32_t          op_errno = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        if (afr_get_children_count (sh->success_children,
-                                    priv->child_count) == 0) {
-                op_errno = afr_resultant_errno_get (NULL, sh->child_errno,
-                                                    priv->child_count);
-                goto fail;
-        }
-
-        //make intersection of (success_children & fresh_parent_dirs) fresh_children
-        //the other success_children will be added to it if they are not stale
-        afr_children_intersection_get (sh->success_children,
-                                       sh->fresh_parent_dirs,
-                                       sh->sources, priv->child_count);
-        afr_get_fresh_children (sh->success_children, sh->sources,
-                                sh->fresh_children, priv->child_count);
-        memset (sh->sources, 0, sizeof (*sh->sources) * 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->op_failed = 1;
-                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;
-                local->govinda_gOvinda = 1;
-                goto fail;
-        }
-
-        return;
-
-fail:
-        afr_sh_set_error (sh, op_errno);
-        afr_sh_missing_entries_finish (frame, this);
-        return;
-}
+		if (!IA_EQUAL (first, replies[i].poststat, prot)) {
+			gf_log (this->name, GF_LOG_DEBUG,
+				"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));
 
-static int
-afr_sh_children_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;
+			*metadata_selfheal = _gf_true;
+		}
 
-        afr_sh_common_lookup_resp_handler (frame, cookie, this, op_ret,
-                                           op_errno, inode, buf, xattr,
-                                           postparent);
-        call_count = afr_frame_return (frame);
+		if (IA_ISREG(first.ia_type) &&
+		    !IA_EQUAL (first, replies[i].poststat, size)) {
+			gf_log (this->name, GF_LOG_DEBUG,
+				"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 (call_count == 0)
-                afr_sh_children_lookup_done (frame, this);
+			*data_selfheal = _gf_true;
+		}
+	}
 
-        return 0;
-}
+	if (valid_cnt > 0)
+		afr_inode_link (inode, &first);
 
-static int
-afr_sh_find_fresh_parents (call_frame_t *frame, xlator_t *this)
-{
-        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;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        /* If We can't find a fresh parent directory here,
-         * we wont know which subvol is correct without finding a parent dir
-         * upwards which has correct xattrs, for that we may have to
-         * do lookups till root, we dont wanna do that,
-         * instead make sure that if there are conflicting gfid
-         * parent dirs, self-heal thus lookup is failed with EIO.
-         * if there are missing entries we dont know whether to delete or
-         * create so fail with EIO,
-         * If there are conflicting xattr fail with EIO.
-         */
-        if (afr_get_children_count (sh->success_children,
-                                    priv->child_count) == 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Parent dir lookup failed "
-                        "for %s, in missing entry self-heal, continuing with "
-                        "the rest of the self-heals", local->loc.path);
-                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, continuing with the rest"
-                        " of the self-heals", local->loc.path);
-                goto out;
-        }
-
-        if (afr_conflicting_iattrs (sh->buf, sh->success_children,
-                                    priv->child_count, sh->parent_loc.path,
-                                    this->name)) {
-                gf_log (this->name, GF_LOG_INFO, "conflicting stat info for "
-                        "parent dirs of %s", local->loc.path);
-                goto out;
-        }
-
-        nsources = afr_build_sources (this, sh->xattr, sh->buf,
-                                      sh->pending_matrix, sh->sources,
-                                      sh->success_children,
-                                      AFR_ENTRY_TRANSACTION);
-        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);
-                goto out;
-        }
-
-        source = afr_sh_select_source (sh->sources, priv->child_count);
-        if (source == -1) {
-                GF_ASSERT (0);
-                gf_log (this->name, GF_LOG_DEBUG, "No active sources found.");
-                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_cbk, _gf_false);
-        return 0;
+	if (valid_cnt < 2)
+		return -ENOTCONN;
 
-out:
-        afr_sh_set_error (sh, EIO);
-        sh->op_failed = 1;
-        afr_sh_missing_entries_finish (frame, this);
-        return 0;
+	return 0;
 }
 
-int
-afr_sh_conflicting_entry_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_sh_common_lookup_resp_handler (frame, cookie, this, op_ret,
-                                           op_errno, inode, buf, xattr,
-                                           postparent);
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0)
-                afr_sh_find_fresh_parents (frame, this);
-
-        return 0;
-}
-
-void
-afr_sh_common_reset (afr_self_heal_t *sh, unsigned int child_count)
+inode_t *
+afr_inode_find (xlator_t *this, uuid_t gfid)
 {
-        int             i = 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);
-}
+	inode_table_t *table = NULL;
+	inode_t *inode = NULL;
 
-/* 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_cbk_t lookup_cbk, gf_boolean_t set_gfid)
-{
-        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 (priv->child_count,
-                                            local->child_up);
-
-        local->call_count = call_count;
-
-        xattr_req = dict_new();
-
-        if (xattr_req) {
-                afr_xattr_req_prepare (this, xattr_req, loc->path);
-                if (set_gfid) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "looking up %s with gfid: %s",
-                                local->loc.path, uuid_utoa (sh->sh_gfid_req));
-                        GF_ASSERT (!uuid_is_null (sh->sh_gfid_req));
-                        afr_set_dict_gfid (xattr_req, sh->sh_gfid_req);
-                }
-        }
-
-        afr_sh_common_reset (sh, priv->child_count);
-        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",
-                                local->loc.path, priv->children[i]->name);
-
-                        STACK_WIND_COOKIE (frame,
-                                           lookup_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->lookup,
-                                           loc, xattr_req);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        if (xattr_req)
-                dict_unref (xattr_req);
-
-        return 0;
-}
+	table = this->itable;
+	if (!table)
+		return NULL;
 
+	inode = inode_find (table, gfid);
+	if (inode)
+		return inode;
 
+	inode = inode_new (table);
+	if (!inode)
+		return NULL;
 
-int
-afr_sh_post_nb_entrylk_conflicting_sh_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
-
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Non blocking entrylks failed.");
-                afr_sh_missing_entries_done (frame, this);
-        } 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_conflicting_entry_lookup_cbk,
-                                      _gf_false);
-        }
-
-        return 0;
-}
+	uuid_copy (inode->gfid, gfid);
 
-int
-afr_sh_post_nb_entrylk_gfid_sh_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Non blocking entrylks failed.");
-                afr_sh_missing_entries_done (frame, this);
-        } else {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking entrylks done. Proceeding to FOP");
-                afr_sh_common_lookup (frame, this, &local->loc,
-                                      afr_sh_missing_entries_lookup_cbk,
-                                      _gf_true);
-        }
-
-        return 0;
+	return inode;
 }
 
-int
-afr_sh_entrylk (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                char *base_name, afr_lock_cbk_t lock_cbk)
-{
-        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;
-
-        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
-        int_lock->selfheal_lk_type    = AFR_ENTRY_SELF_HEAL_LK;
-
-        afr_set_lock_number (frame, this);
+call_frame_t *
+afr_frame_create (xlator_t *this)
+{
+	call_frame_t *frame = NULL;
+	afr_local_t *local = NULL;
+	int op_errno = 0;
+	pid_t pid = -1;
 
-        int_lock->lk_basename = base_name;
-        int_lock->lk_loc      = loc;
-        int_lock->lock_cbk    = lock_cbk;
+	frame = create_frame (this, this->ctx->pool);
+	if (!frame)
+		return NULL;
 
-        afr_nonblocking_entrylk (frame, this);
+	local = AFR_FRAME_INIT (frame, op_errno);
+	if (!local) {
+		STACK_DESTROY (frame->root);
+		return NULL;
+	}
 
-        return 0;
-}
+	syncopctx_setfspid (&pid);
 
-static int
-afr_self_heal_parent_entrylk (call_frame_t *frame, xlator_t *this,
-                              afr_lock_cbk_t lock_cbk)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-        afr_private_t       *priv     = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
-        priv     = this->private;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "attempting to recreate missing entries for path=%s",
-                local->loc.path);
-
-        GF_ASSERT (local->loc.parent);
-        afr_build_parent_loc (&sh->parent_loc, &local->loc);
-
-        afr_sh_entrylk (frame, this, &sh->parent_loc, NULL,
-                        lock_cbk);
-        return 0;
-}
+	frame->root->pid = pid;
 
-static int
-afr_self_heal_conflicting_entries (call_frame_t *frame, xlator_t *this)
-{
-        afr_self_heal_parent_entrylk (frame, this,
-                                      afr_sh_post_nb_entrylk_conflicting_sh_cbk);
-        return 0;
-}
+	afr_set_lk_owner (frame, this, frame->root);
 
-static int
-afr_self_heal_gfids (call_frame_t *frame, xlator_t *this)
-{
-        afr_self_heal_parent_entrylk (frame, this,
-                                      afr_sh_post_nb_entrylk_gfid_sh_cbk);
-        return 0;
+	return frame;
 }
 
-afr_local_t *afr_local_copy (afr_local_t *l, xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        afr_local_t   *lc     = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_self_heal_t *shc = NULL;
-
-        priv = this->private;
-
-        sh = &l->self_heal;
-
-        lc = GF_CALLOC (1, sizeof (afr_local_t),
-                        gf_afr_mt_afr_local_t);
-        if (!lc)
-                goto out;
-
-        shc = &lc->self_heal;
-
-        shc->unwind = sh->unwind;
-        shc->gfid_sh_success_cbk = sh->gfid_sh_success_cbk;
-        shc->need_missing_entry_self_heal = sh->need_missing_entry_self_heal;
-        shc->need_gfid_self_heal = sh->need_gfid_self_heal;
-        shc->need_data_self_heal = sh->need_data_self_heal;
-        shc->need_metadata_self_heal = sh->need_metadata_self_heal;
-        shc->need_entry_self_heal = sh->need_entry_self_heal;
-        shc->forced_merge = sh->forced_merge;
-        shc->healing_fd_opened = sh->healing_fd_opened;
-        shc->data_lock_held = sh->data_lock_held;
-        if (sh->healing_fd && !sh->healing_fd_opened)
-                shc->healing_fd = fd_ref (sh->healing_fd);
-        else
-                shc->healing_fd = sh->healing_fd;
-        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, 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,
-                                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.entry_locked_nodes)
-                lc->internal_lock.entry_locked_nodes =
-                        memdup (l->internal_lock.entry_locked_nodes,
-                                priv->child_count);
-        else
-                lc->internal_lock.entry_locked_nodes =
-                        GF_CALLOC (sizeof (*l->internal_lock.entry_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,
-                                priv->child_count);
-        else
-                lc->internal_lock.locked_nodes =
-                        GF_CALLOC (sizeof (*l->internal_lock.locked_nodes),
-                                   priv->child_count,
-                                   gf_afr_mt_char);
-
-        lc->internal_lock.inodelk_lock_count =
-                l->internal_lock.inodelk_lock_count;
-        lc->internal_lock.entrylk_lock_count =
-                l->internal_lock.entrylk_lock_count;
 
-out:
-        return lc;
-}
+/*
+ * This is the entry point for healing a given GFID
+ */
 
 int
-afr_self_heal_completion_cbk (call_frame_t *bgsh_frame, xlator_t *this)
+afr_selfheal (xlator_t *this, uuid_t gfid)
 {
-        afr_private_t *   priv  = NULL;
-        afr_local_t *     local = NULL;
-        afr_self_heal_t * sh    = NULL;
-        char              sh_type_str[256] = {0,};
-        gf_boolean_t      split_brain = _gf_false;
-
-        priv  = this->private;
-        local = bgsh_frame->local;
-        sh    = &local->self_heal;
-
-        if (local->govinda_gOvinda)
-                split_brain = _gf_true;
-
-        afr_set_split_brain (this, sh->inode, split_brain);
-
-        afr_self_heal_type_str_get (sh, sh_type_str,
-                                    sizeof(sh_type_str));
-        if (sh->op_failed) {
-                gf_log (this->name, GF_LOG_ERROR, "background %s self-heal "
-                        "failed on %s", sh_type_str, local->loc.path);
-        } else {
-                gf_log (this->name, GF_LOG_INFO, "background %s self-heal "
-                        "completed on %s", sh_type_str, local->loc.path);
-        }
-
-        FRAME_SU_UNDO (bgsh_frame, afr_local_t);
-
-        if (!sh->unwound) {
-                sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno);
-        }
-
-        if (sh->background) {
-                LOCK (&priv->lock);
-                {
-                        priv->background_self_heals_started--;
-                }
-                UNLOCK (&priv->lock);
-        }
-
-        AFR_STACK_DESTROY (bgsh_frame);
-
-        return 0;
-}
+	inode_t *inode = NULL;
+	call_frame_t *frame = NULL;
+	int ret = -1;
+	gf_boolean_t data_selfheal = _gf_false;
+	gf_boolean_t metadata_selfheal = _gf_false;
+	gf_boolean_t entry_selfheal = _gf_false;
 
-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;
-        int              i = 0;
-
-        call_frame_t *sh_frame = NULL;
-        afr_local_t  *sh_local = NULL;
-
-        local = frame->local;
-        priv  = this->private;
-
-        GF_ASSERT (local->loc.path);
-
-        if (local->self_heal.background) {
-                LOCK (&priv->lock);
-                {
-                        if (priv->background_self_heals_started
-                            < priv->background_self_heal_count) {
-                                priv->background_self_heals_started++;
-
-
-                        } else {
-                                local->self_heal.background = _gf_false;
-                        }
-                }
-                UNLOCK (&priv->lock);
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "performing self heal on %s (metadata=%d data=%d entry=%d)",
-                local->loc.path,
-                local->self_heal.need_metadata_self_heal,
-                local->self_heal.need_data_self_heal,
-                local->self_heal.need_entry_self_heal);
-
-        sh_frame        = copy_frame (frame);
-        afr_set_lk_owner (sh_frame, this);
-
-        sh_local        = afr_local_copy (local, this);
-        sh_frame->local = sh_local;
-        sh              = &sh_local->self_heal;
-
-        sh->inode       = inode_ref (inode);
-
-        sh->orig_frame  = frame;
-
-        sh->completion_cbk = afr_self_heal_completion_cbk;
-
-        sh->buf = GF_CALLOC (priv->child_count, sizeof (struct iatt),
-                             gf_afr_mt_iatt);
-        sh->parentbufs = GF_CALLOC (priv->child_count, sizeof (struct iatt),
-                                    gf_afr_mt_iatt);
-        sh->child_errno = GF_CALLOC (priv->child_count, sizeof (int),
-                                     gf_afr_mt_int);
-        sh->success = GF_CALLOC (priv->child_count, sizeof (int),
-                                 gf_afr_mt_int);
-        sh->xattr = GF_CALLOC (priv->child_count, sizeof (dict_t *),
-                               gf_afr_mt_dict_t);
-        sh->sources = GF_CALLOC (sizeof (*sh->sources), priv->child_count,
-                                 gf_afr_mt_int);
-        sh->locked_nodes = GF_CALLOC (sizeof (*sh->locked_nodes),
-                                      priv->child_count,
-                                      gf_afr_mt_int);
-
-        sh->pending_matrix = GF_CALLOC (sizeof (int32_t *), priv->child_count,
-                                        gf_afr_mt_int32_t);
-
-        for (i = 0; i < priv->child_count; i++) {
-                sh->pending_matrix[i] = GF_CALLOC (sizeof (int32_t),
-                                                   priv->child_count,
-                                                   gf_afr_mt_int32_t);
-        }
-
-        sh->delta_matrix = GF_CALLOC (sizeof (int32_t *), priv->child_count,
-                                      gf_afr_mt_int32_t);
-        for (i = 0; i < priv->child_count; i++) {
-                sh->delta_matrix[i] = GF_CALLOC (sizeof (int32_t),
-                                                 priv->child_count,
-                                                 gf_afr_mt_int32_t);
-        }
-        sh->success_children = afr_fresh_children_create (priv->child_count);
-        sh->fresh_children = afr_fresh_children_create (priv->child_count);
-        sh->fresh_parent_dirs = afr_fresh_children_create (priv->child_count);
-
-
-        FRAME_SU_DO (sh_frame, afr_local_t);
-        if (sh->need_missing_entry_self_heal) {
-                afr_self_heal_conflicting_entries (sh_frame, this);
-        } else if (sh->need_gfid_self_heal) {
-                GF_ASSERT (!uuid_is_null (sh->sh_gfid_req));
-                afr_self_heal_gfids (sh_frame, this);
-        } else {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "proceeding to metadata check on %s",
-                        local->loc.path);
-
-                afr_sh_missing_entries_done (sh_frame, this);
-        }
-
-        return 0;
-}
+	inode = afr_inode_find (this, gfid);
+	if (!inode)
+		goto out;
 
-void
-afr_self_heal_type_str_get (afr_self_heal_t *self_heal_p, char *str,
-                            size_t size)
-{
-        GF_ASSERT (str && (size > strlen (" missing-entry gfid "
-                                          "meta-data data entry")));
+	frame = afr_frame_create (this);
+	if (!frame)
+		goto out;
 
-        if (self_heal_p->need_metadata_self_heal) {
-                snprintf (str, size, " meta-data");
-        }
+	ret = afr_selfheal_unlocked_inspect (frame, this, inode, gfid,
+					     &data_selfheal,
+					     &metadata_selfheal,
+					     &entry_selfheal);
+	if (ret)
+		goto out;
 
-        if (self_heal_p->need_data_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str), " data");
-        }
+	if (data_selfheal)
+		afr_selfheal_data (frame, this, inode);
 
-        if (self_heal_p->need_entry_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str), " entry");
-        }
+	if (metadata_selfheal)
+		afr_selfheal_metadata (frame, this, inode);
 
-        if (self_heal_p->need_missing_entry_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str),
-                         " missing-entry");
-        }
+	if (entry_selfheal)
+		afr_selfheal_entry (frame, this, inode);
 
-        if (self_heal_p->need_gfid_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str), " gfid");
-        }
-}
+	inode_forget (inode, 1);
+out:
+	if (inode)
+		inode_unref (inode);
+	if (frame)
+		AFR_STACK_DESTROY (frame);
 
-afr_self_heal_type
-afr_self_heal_type_for_transaction (afr_transaction_type type)
-{
-        afr_self_heal_type sh_type = AFR_SELF_HEAL_INVALID;
-
-        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;
+	return ret;
 }
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 c600db825..000000000
--- a/xlators/cluster/afr/src/afr-self-heal-common.h
+++ /dev/null
@@ -1,98 +0,0 @@
-/*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __AFR_SELF_HEAL_COMMON_H__
-#define __AFR_SELF_HEAL_COMMON_H__
-
-#define FILE_HAS_HOLES(buf) (((buf)->ia_size) > ((buf)->ia_blocks * 512))
-
-typedef enum {
-        AFR_SELF_HEAL_ENTRY,
-        AFR_SELF_HEAL_METADATA,
-        AFR_SELF_HEAL_DATA,
-        AFR_SELF_HEAL_INVALID = -1,
-} afr_self_heal_type;
-
-typedef int
-(*afr_lookup_cbk_t) (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
-afr_sh_select_source (int sources[], int child_count);
-
-int
-afr_sh_sink_count (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);
-
-int
-afr_build_pending_matrix (char **pending_key, int32_t **pending_matrix,
-                          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[], int success[],
-                         int child_count, afr_transaction_type type);
-
-int
-afr_mark_sources (int32_t *sources, int32_t **pending_matrix, struct iatt *bufs,
-                  int32_t child_count, afr_self_heal_type type,
-                  int32_t *valid_children, const char *xlator_name);
-
-int
-afr_sh_delta_to_xattr (afr_private_t *priv,
-                       int32_t *delta_matrix[], dict_t *xattr[],
-		       int child_count, afr_transaction_type type);
-
-int
-afr_sh_is_matrix_zero (int32_t *pending_matrix[], int child_count);
-
-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 *xlator, dict_t **xattr, struct iatt *bufs,
-                   int32_t **pending_matrix, int32_t *sources,
-                   int32_t *success_children, afr_transaction_type type);
-void
-afr_sh_common_reset (afr_self_heal_t *sh, unsigned int child_count);
-int
-afr_sh_common_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                      afr_lookup_cbk_t lookup_cbk, gf_boolean_t set_gfid);
-int
-afr_sh_entry_expunge_remove (call_frame_t *expunge_frame, xlator_t *this,
-                             int active_src, struct iatt *buf);
-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, struct iatt *buf,
-                             struct iatt *postparent);
-#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 6d16b170f..c0548d995 100644
--- a/xlators/cluster/afr/src/afr-self-heal-data.c
+++ b/xlators/cluster/afr/src/afr-self-heal-data.c
@@ -1,1198 +1,635 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#include <libgen.h>
-#include <unistd.h>
-#include <fnmatch.h>
-#include <sys/time.h>
-#include <stdlib.h>
-#include <signal.h>
 
 #ifndef _CONFIG_H
 #define _CONFIG_H
 #include "config.h"
 #endif
 
-#include "glusterfs.h"
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-
-#include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-#include "afr-self-heal-algorithm.h"
-
-
-int
-afr_sh_data_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        /*
-          TODO: cleanup sh->*
-        */
-
-        if (sh->healing_fd && !sh->healing_fd_opened) {
-                /* unref only if we created the fd ourselves */
-
-                fd_unref (sh->healing_fd);
-                sh->healing_fd = NULL;
-        }
-
-        /* for (i = 0; i < priv->child_count; i++) */
-        /*        sh->locked_nodes[i] = 0; */
-
-        sh->completion_cbk (frame, this);
-
-        return 0;
-}
-
-
-int
-afr_sh_data_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno)
-{
-        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,
-                                "flush or 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_done (frame, this);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_data_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *statpre,
-                         struct iatt *statpost)
-{
-        afr_sh_data_flush_cbk (frame, cookie, this, op_ret, op_errno);
-
-        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;
-        int              source     = 0;
-        int32_t          valid      = 0;
-        struct iatt      stbuf      = {0,};
-
-        local = frame->local;
-        sh    = &local->self_heal;
-        priv  = this->private;
-
-        source = sh->source;
-
-        valid |= (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME);
-
-        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;
-
-        if (sh->healing_fd_opened) {
-                /* not our job to close the fd */
-
-                afr_sh_data_done (frame, this);
-                return 0;
-        }
-
-        if (!sh->healing_fd) {
-                afr_sh_data_done (frame, this);
-                return 0;
-        }
-
-        call_count        = (sh->active_sinks + 1) * 2;
-        local->call_count = call_count;
-
-        /* closed source */
-        gf_log (this->name, GF_LOG_TRACE,
-                "closing fd of %s on %s",
-                local->loc.path, priv->children[sh->source]->name);
-
-        STACK_WIND_COOKIE (frame, afr_sh_data_flush_cbk,
-                           (void *) (long) sh->source,
-                           priv->children[sh->source],
-                           priv->children[sh->source]->fops->flush,
-                           sh->healing_fd);
-        call_count--;
-
-        STACK_WIND_COOKIE (frame, afr_sh_data_setattr_cbk,
-                           (void *) (long) sh->source,
-                           priv->children[sh->source],
-                           priv->children[sh->source]->fops->setattr,
-                           &local->loc, &stbuf, valid);
-
-        call_count--;
-
-        if (call_count == 0)
-                return 0;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->sources[i] || !local->child_up[i])
-                        continue;
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "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);
-
-                call_count--;
-
-                STACK_WIND_COOKIE (frame, afr_sh_data_setattr_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->setattr,
-                                   &local->loc, &stbuf, valid);
-
-                if (!--call_count)
-                        break;
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_data_unlck_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno)
-{
-        afr_local_t * local = NULL;
-        int           call_count = 0;
-        int           child_index = (long) cookie;
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "locking inode of %s on child %d failed: %s",
-                                local->loc.path, child_index,
-                                strerror (op_errno));
-                } else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "inode of %s on child %d locked",
-                                local->loc.path, child_index);
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                afr_sh_data_close (frame, this);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_data_unlock (call_frame_t *frame, xlator_t *this)
-{
-        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);
-
-        int_lock->lock_cbk = afr_sh_data_close;
-        afr_unlock (frame, this);
-
-        return 0;
-}
-
-
-int
-afr_sh_data_finish (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "finishing data selfheal of %s", local->loc.path);
-
-        if (!sh->data_lock_held)
-                afr_sh_data_unlock (frame, this);
-        else
-                afr_sh_data_close (frame, this);
-
-        return 0;
-}
-
-
-int
-afr_sh_data_erase_pending_cbk (call_frame_t *frame, void *cookie,
-                               xlator_t *this, int32_t op_ret,
-                               int32_t op_errno, dict_t *xattr)
-{
-        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;
-
-        afr_fresh_children_add_child (sh->fresh_children, i, priv->child_count);
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                afr_inode_set_read_ctx (this, sh->inode, sh->source,
-                                        sh->fresh_children);
-                afr_sh_data_finish (frame, this);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_data_erase_pending (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;
-
-        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, AFR_DATA_TRANSACTION);
-
-        erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count,
-                                 gf_afr_mt_dict_t);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->xattr[i]) {
-                        call_count++;
-
-                        erase_xattr[i] = get_new_dict();
-                        dict_ref (erase_xattr[i]);
-                }
-        }
-
-        afr_sh_delta_to_xattr (priv, sh->delta_matrix, erase_xattr,
-                               priv->child_count, AFR_DATA_TRANSACTION);
-
-        local->call_count = call_count;
-        for (i = 0; i < priv->child_count; i++) {
-                if (!erase_xattr[i])
-                        continue;
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "erasing pending flags from %s on %s",
-                        local->loc.path, priv->children[i]->name);
-
-                STACK_WIND_COOKIE (frame, afr_sh_data_erase_pending_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->fxattrop,
-                                   sh->healing_fd,
-                                   GF_XATTROP_ADD_ARRAY, erase_xattr[i]);
-                if (!--call_count)
-                        break;
-        }
+#include "byte-order.h"
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (erase_xattr[i]) {
-                        dict_unref (erase_xattr[i]);
-                }
-        }
-        GF_FREE (erase_xattr);
+enum {
+	AFR_SELFHEAL_DATA_FULL = 0,
+	AFR_SELFHEAL_DATA_DIFF,
+};
 
-        return 0;
-}
 
-
-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)
+#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_private_t * priv = NULL;
-        afr_local_t * local  = NULL;
-        int              call_count = 0;
-        int              child_index = 0;
-
-        priv = this->private;
-        local = frame->local;
-
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1)
-                        gf_log (this->name, GF_LOG_INFO,
-                                "ftruncate of %s on subvolume %s failed (%s)",
-                                local->loc.path,
-                                priv->children[child_index]->name,
-                                strerror (op_errno));
-                else
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "ftruncate of %s on subvolume %s completed",
-                                local->loc.path,
-                                priv->children[child_index]->name);
-        }
-        UNLOCK (&frame->lock);
+	afr_local_t *local = NULL;
+	int i = (long) cookie;
 
-        call_count = afr_frame_return (frame);
+	local = frame->local;
 
-        if (call_count == 0) {
-                afr_sh_data_erase_pending (frame, this);
-        }
+	local->replies[i].valid = 1;
+	local->replies[i].op_ret = op_ret;
+	local->replies[i].op_errno = op_errno;
+	if (strong)
+		memcpy (local->replies[i].checksum, strong, MD5_DIGEST_LENGTH);
 
-        return 0;
+	syncbarrier_wake (&local->barrier);
+	return 0;
 }
 
 
-int
-afr_sh_data_trim_sinks (call_frame_t *frame, xlator_t *this)
+static int
+attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+	  int op_ret, int op_errno, struct iatt *pre, struct iatt *post,
+	  dict_t *xdata)
 {
-        afr_private_t * priv = NULL;
-        afr_local_t * local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        int             *sources = NULL;
-        int              call_count = 0;
-        int              i = 0;
-
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
+	int i = (long) cookie;
+	afr_local_t *local = NULL;
 
-        sources = sh->sources;
-        call_count = sh->active_sinks;
+	local = frame->local;
 
-        local->call_count = call_count;
+	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 < priv->child_count; i++) {
-                if (sources[i] || !local->child_up[i])
-                        continue;
+	syncbarrier_wake (&local->barrier);
 
-                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);
-
-                if (!--call_count)
-                        break;
-        }
-
-        return 0;
+	return 0;
 }
 
 
-static struct afr_sh_algorithm *
-sh_algo_from_name (xlator_t *this, char *name)
+static gf_boolean_t
+__afr_selfheal_data_checksums_match (call_frame_t *frame, xlator_t *this,
+				     fd_t *fd, int source,
+				     unsigned char *healed_sinks,
+				     off_t offset, size_t size)
 {
-        int i = 0;
-
-        while (afr_self_heal_algorithms[i].name) {
-                if (!strcmp (name, afr_self_heal_algorithms[i].name)) {
-                        return &afr_self_heal_algorithms[i];
-                }
-
-                i++;
-        }
-
-        return NULL;
+	afr_private_t *priv = NULL;
+	afr_local_t *local = NULL;
+	unsigned char *wind_subvols = NULL;
+	int i = 0;
+
+	priv = this->private;
+	local = frame->local;
+
+	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, NULL);
+
+	if (!local->replies[source].valid || local->replies[source].op_ret != 0)
+		return _gf_false;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (i == source)
+			continue;
+		if (memcmp (local->replies[source].checksum,
+			    local->replies[i].checksum,
+			    MD5_DIGEST_LENGTH))
+			return _gf_false;
+	}
+
+	return _gf_true;
 }
 
 
 static int
-sh_zero_byte_files_exist (afr_self_heal_t *sh, int child_count)
+__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 i;
-        int ret = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (sh->buf[i].ia_size == 0) {
-                        ret = 1;
-                        break;
-                }
-        }
-
-        return ret;
+	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);
+	if (ret <= 0)
+		return ret;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!healed_sinks[i])
+			continue;
+
+		/*
+		 * TODO: Use fiemap() and discard() to heal holes
+		 * in the future.
+		 *
+		 * For now,
+		 *
+		 * - if the source had any holes at all,
+		 * AND
+		 * - if we are writing past the original file size
+		 *   of the sink
+		 * AND
+		 * - is NOT the last block of the source file. if
+		 *   the block contains EOF, it has to be written
+		 *   in order to set the file size even if the
+		 *   last block is 0-filled.
+		 * AND
+		 * - if the read buffer is filled with only 0's
+		 *
+		 * then, skip writing to this source. We don't depend
+		 * on the write to happen to update the size as we
+		 * have performed an ftruncate() upfront anyways.
+		 */
+#define is_last_block(o,b,s) ((s >= o) && (s <= (o + b)))
+		if (HAS_HOLES ((&replies[source].poststat)) &&
+		    offset >= replies[i].poststat.ia_size &&
+		    !is_last_block (offset, size,
+				    replies[source].poststat.ia_size) &&
+		    (iov_0filled (iovec, count) == 0))
+			continue;
+
+		ret = syncop_writev (priv->children[i], fd, iovec, count,
+				     offset, iobref, 0);
+		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 (iobref)
+		iobref_unref (iobref);
+
+	return ret;
 }
 
 
-struct afr_sh_algorithm *
-afr_sh_data_pick_algo (call_frame_t *frame, xlator_t *this)
+static int
+afr_selfheal_data_block (call_frame_t *frame, xlator_t *this, fd_t *fd,
+			 int source, unsigned char *healed_sinks, off_t offset,
+			 size_t size, int type, struct afr_reply *replies)
 {
-        afr_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 ((local->enoent_count != 0)
-                    || sh_zero_byte_files_exist (sh, 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");
-                }
-        }
-
-        return algo;
+	int ret = -1;
+	int sink_count = 0;
+	afr_private_t *priv = NULL;
+	unsigned char *data_lock = NULL;
+
+	priv = this->private;
+	sink_count = AFR_COUNT (healed_sinks, priv->child_count);
+	data_lock = alloca0 (priv->child_count);
+
+	ret = afr_selfheal_inodelk (frame, this, fd->inode, this->name,
+				    offset, size, data_lock);
+	{
+		if (ret < sink_count) {
+			ret = -ENOTCONN;
+			goto unlock;
+		}
+
+		if (type == AFR_SELFHEAL_DATA_DIFF &&
+		    __afr_selfheal_data_checksums_match (frame, this, fd, source,
+							 healed_sinks, offset, size)) {
+			ret = 0;
+			goto unlock;
+		}
+
+		ret = __afr_selfheal_data_read_write (frame, this, fd, source,
+						      healed_sinks, offset, size,
+						      replies);
+	}
+unlock:
+	afr_selfheal_uninodelk (frame, this, fd->inode, this->name,
+				offset, size, data_lock);
+	return ret;
 }
 
 
-int
-afr_sh_data_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              active_sinks = 0;
-        int              source = 0;
-        int              i = 0;
-        struct afr_sh_algorithm *sh_algo = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        source = sh->source;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->sources[i] == 0 && local->child_up[i] == 1) {
-                        active_sinks++;
-                        sh->success[i] = 1;
-                }
-        }
-        sh->success[source] = 1;
-
-        if (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 0;
-        }
-        sh->active_sinks = active_sinks;
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "self-healing file %s from subvolume %s to %d other",
-                local->loc.path, priv->children[source]->name, active_sinks);
-
-        sh->algo_completion_cbk = afr_sh_data_trim_sinks;
-        sh->algo_abort_cbk      = afr_sh_data_finish;
-
-        sh_algo = afr_sh_data_pick_algo (frame, this);
-
-        sh_algo->fn (frame, this);
-
-        return 0;
-}
-
-
-int
-afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local      = NULL;
-        afr_local_t *    orig_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;
-
-        nsources = afr_build_sources (this, sh->xattr, sh->buf, sh->pending_matrix,
-                                      sh->sources, sh->success_children,
-                                      AFR_DATA_TRANSACTION);
-        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 ((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;
-
-                nsources = afr_sh_source_count (sh->sources,
-                                                priv->child_count);
-        }
-
-        if (nsources == -1) {
-                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.", local->loc.path);
-
-                local->govinda_gOvinda = 1;
-
-                afr_sh_data_finish (frame, this);
-                return 0;
-        }
-
-        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_data_finish (frame, this);
-                return 0;
-        }
-
-        sh->source     = source;
-        sh->block_size = 65536; /* TODO: make it configurable or use macro */
-        sh->file_size  = sh->buf[source].ia_size;
-
-        if (FILE_HAS_HOLES (&sh->buf[source]))
-                sh->file_has_holes = 1;
-
-        orig_local = sh->orig_frame->local;
-        orig_local->cont.lookup.buf.ia_size = sh->buf[source].ia_size;
-
-        /* detect changes not visible through pending flags -- JIC */
-        for (i = 0; i < priv->child_count; i++) {
-                if (i == source || sh->child_errno[i])
-                        continue;
-
-                if (SIZE_DIFFERS (&sh->buf[i], &sh->buf[source]))
-                        sh->sources[i] = 0;
-        }
-
-        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);
-
-        /*
-          quick-read might have read the file, so send xattr from
-          the source subvolume (http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=815)
-        */
-
-        dict_unref (orig_local->cont.lookup.xattr);
-        if (orig_local->cont.lookup.xattrs)
-                orig_local->cont.lookup.xattr = dict_ref (orig_local->cont.lookup.xattrs[sh->source]);
-
-        if (sh->background) {
-                sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno);
-                sh->unwound = _gf_true;
-        }
-
-        afr_sh_data_sync_prepare (frame, this);
-
-        return 0;
-}
-
-static void
-afr_destroy_pending_matrix (int32_t **pending_matrix, int32_t child_count)
+static int
+afr_selfheal_data_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd,
+			 unsigned char *healed_sinks)
 {
-        int             i = 0;
-        GF_ASSERT (child_count > 0);
-        if (pending_matrix) {
-                for (i = 0; i < child_count; i++) {
-                        if (pending_matrix[i])
-                                GF_FREE (pending_matrix[i]);
-                }
-                GF_FREE (pending_matrix);
-        }
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	int i = 0;
+
+	local = frame->local;
+	priv = this->private;
+
+	AFR_ONLIST (healed_sinks, frame, attr_cbk, fsync, fd, 0, NULL);
+
+	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;
 }
 
-static int32_t**
-afr_create_pending_matrix (int32_t child_count)
-{
-        gf_boolean_t            cleanup = _gf_false;
-        int32_t                 **pending_matrix = NULL;
-        int                     i = 0;
-
-        GF_ASSERT (child_count > 0);
-
-        pending_matrix = GF_CALLOC (sizeof (*pending_matrix), child_count,
-                                    gf_afr_mt_int32_t);
-        if (NULL == pending_matrix)
-                goto out;
-        for (i = 0; i < child_count; i++) {
-                pending_matrix[i] = GF_CALLOC (sizeof (**pending_matrix),
-                                               child_count,
-                                               gf_afr_mt_int32_t);
-                if (NULL == pending_matrix[i]) {
-                        cleanup = _gf_true;
-                        goto out;
-                }
-        }
-out:
-        if (_gf_true == cleanup) {
-                afr_destroy_pending_matrix (pending_matrix, child_count);
-                pending_matrix = NULL;
-        }
-        return pending_matrix;
-}
 
-int
-afr_lookup_select_read_child_by_txn_type (xlator_t *this, afr_local_t *local,
-                                          dict_t **xattr,
-                                          afr_transaction_type txn_type)
+static int
+afr_selfheal_data_restore_time (call_frame_t *frame, xlator_t *this,
+				inode_t *inode, int source,
+				unsigned char *healed_sinks,
+				struct afr_reply *replies)
 {
-        afr_private_t            *priv      = NULL;
-        int                      read_child = -1;
-        int                      ret        = -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;
-        afr_self_heal_t          *sh = NULL;
-
-        priv = this->private;
-        bufs = local->cont.lookup.bufs;
-        success_children = local->cont.lookup.success_children;
-        sh = &local->self_heal;
-
-        pending_matrix = afr_create_pending_matrix (priv->child_count);
-        if (NULL == pending_matrix)
-                goto out;
-
-        sources = GF_CALLOC (sizeof (*sources), priv->child_count,
-                             gf_afr_mt_int32_t);
-        if (NULL == sources)
-                goto out;
-
-        nsources = afr_build_sources (this, xattr, bufs, pending_matrix,
-                                      sources, success_children, txn_type);
-        if (nsources < 0) {
-                ret = -1;
-                goto out;
-        }
+	loc_t loc = {0, };
 
-        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);
-        ret = 0;
-        local->cont.lookup.sources = sources;
-out:
-        afr_destroy_pending_matrix (pending_matrix, priv->child_count);
-        if (-1 == ret) {
-                if (sources)
-                        GF_FREE (sources);
-        }
-        gf_log (this->name, GF_LOG_DEBUG, "returning read_child: %d", read_child);
-        return read_child;
-}
-
-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)
-{
-        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++;
-                }
-        }
-        UNLOCK (&frame->lock);
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.gfid, inode->gfid);
 
-        call_count = afr_frame_return (frame);
+	AFR_ONLIST (healed_sinks, frame, attr_cbk, setattr, &loc,
+		    &replies[source].poststat,
+		    (GF_SET_ATTR_ATIME|GF_SET_ATTR_MTIME), NULL);
 
-        if (call_count == 0) {
-                afr_sh_data_fix (frame, this);
-        }
+	loc_wipe (&loc);
 
-        return 0;
+	return 0;
 }
 
-
-int
-afr_sh_data_fstat (call_frame_t *frame, xlator_t *this)
+static int
+afr_data_self_heal_type_get (afr_private_t *priv, unsigned char *healed_sinks,
+                             int source, struct afr_reply *replies)
 {
-        afr_self_heal_t *sh    = NULL;
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv  = NULL;
-        int call_count = 0;
+        int type = AFR_SELFHEAL_DATA_FULL;
         int i = 0;
 
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        call_count = afr_up_children_count (priv->child_count,
-                                            local->child_up);
-
-        local->call_count = call_count;
-
-        afr_reset_children (sh->success_children, 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_fstat_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fstat,
-                                           sh->healing_fd);
-
-                        if (!--call_count)
+        if (priv->data_self_heal_algorithm == NULL) {
+                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;
+                        }
                 }
+        } else if (strcmp (priv->data_self_heal_algorithm, "full") == 0) {
+                type = AFR_SELFHEAL_DATA_FULL;
+        } else if (strcmp (priv->data_self_heal_algorithm, "diff") == 0) {
+                type = AFR_SELFHEAL_DATA_DIFF;
         }
-
-        return 0;
+        return type;
 }
 
-
-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)
-{
-        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,
-                                "fxattrop of %s on %s succeeded",
-                                local->loc.path,
-                                priv->children[child_index]->name);
-
-                        sh->xattr[child_index] = dict_ref (xattr);
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                afr_sh_data_fstat (frame, this);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_data_fxattrop (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_self_heal_t *sh    = NULL;
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv  = NULL;
-        dict_t          *xattr_req = NULL;
-        int32_t         *zero_pending = NULL;
-        int call_count = 0;
-        int i = 0;
-        int ret = 0;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        call_count = afr_up_children_count (priv->child_count,
-                                            local->child_up);
-
-        local->call_count = call_count;
-
-        xattr_req = dict_new();
-        if (!xattr_req) {
-                ret = -1;
-                goto out;
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                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, priv->pending_key[i],
-                                       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;
-                }
-        }
-
-        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);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+	afr_private_t *priv = NULL;
+	int i = 0;
+	off_t off = 0;
+	size_t block = 128 * 1024;
+	int type = AFR_SELFHEAL_DATA_FULL;
+	int ret = -1;
+	call_frame_t *iter_frame = NULL;
+	char *sinks_str = NULL;
+	char *p = NULL;
+
+	priv = this->private;
+
+	sinks_str = alloca0 (priv->child_count * 8);
+	p = sinks_str;
+	for (i = 0; i < priv->child_count; i++) {
+		if (!healed_sinks[i])
+			continue;
+		p += sprintf (p, "%d ", i);
+	}
+
+	gf_log (this->name, GF_LOG_INFO, "performing data selfheal on %s. "
+		"source=%d sinks=%s",
+		uuid_utoa (fd->inode->gfid), source, sinks_str);
+
+        type = afr_data_self_heal_type_get (priv, healed_sinks, source,
+                                            replies);
+
+	iter_frame = afr_copy_frame (frame);
+	if (!iter_frame)
+		return -ENOMEM;
+
+	for (off = 0; off < replies[source].poststat.ia_size; off += block) {
+		ret = afr_selfheal_data_block (iter_frame, this, fd, source,
+					       healed_sinks, off, block, type,
+					       replies);
+		if (ret < 0)
+			goto out;
+
+		AFR_STACK_RESET (iter_frame);
+	}
+
+	afr_selfheal_data_restore_time (frame, this, fd->inode, source,
+					healed_sinks, replies);
+
+	ret = afr_selfheal_data_fsync (frame, this, fd, healed_sinks);
 
 out:
-        if (xattr_req)
-                dict_unref (xattr_req);
-
-        if (ret) {
-                if (zero_pending)
-                        GF_FREE (zero_pending);
-                sh->op_failed = 1;
-                afr_sh_data_done (frame, this);
-        }
-
-        return 0;
+	if (iter_frame)
+		AFR_STACK_DESTROY (iter_frame);
+	return ret;
 }
 
-int
-afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this);
 
-int
-afr_sh_data_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
+static int
+__afr_selfheal_truncate_sinks (call_frame_t *frame, xlator_t *this,
+			       fd_t *fd, unsigned char *healed_sinks,
+			       struct afr_reply *replies, uint64_t size)
 {
-        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 data inodelks "
-                        "failed for %s.", local->loc.path);
-                sh->op_failed = 1;
-                afr_sh_data_done (frame, this);
-        } else {
-
-                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking data inodelks "
-                        "done for %s. Proceeding to FOP", local->loc.path);
-                afr_sh_data_fxattrop (frame, this);
-        }
-
-        return 0;
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	unsigned char *larger_sinks = 0;
+	int i = 0;
+
+	local = frame->local;
+	priv = this->private;
+
+	larger_sinks = alloca0 (priv->child_count);
+	for (i = 0; i < priv->child_count; i++) {
+		if (healed_sinks[i] && replies[i].poststat.ia_size > size)
+			larger_sinks[i] = 1;
+	}
+
+	AFR_ONLIST (larger_sinks, frame, attr_cbk, ftruncate, fd, size, 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;
 }
 
-int
-afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this)
+/*
+ * If by chance there are multiple sources with differing sizes, select
+ * the largest file as the source.
+ *
+ * This can only happen if data was directly modified in the backend.
+ */
+static int
+__afr_selfheal_data_finalize_source (xlator_t *this, unsigned char *sources,
+				     unsigned char *sinks,
+				     unsigned char *healed_sinks,
+				     unsigned char *locked_on,
+				     struct afr_reply *replies)
 {
-        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;
-
-        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 = 0;
-        int_lock->lk_flock.l_len   = 0;
-        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 i = 0;
+	afr_private_t *priv = NULL;
+	uint64_t size = 0;
+	int source = -1;
+	int locked_count = 0;
+	int sources_count = 0;
+	int healed_sinks_count = 0;
+
+	priv = this->private;
+
+	locked_count = AFR_COUNT (locked_on, priv->child_count);
+	sources_count = AFR_COUNT (sources, priv->child_count);
+	healed_sinks_count = AFR_COUNT (healed_sinks, priv->child_count);
+
+	if (locked_count == healed_sinks_count || !sources_count) {
+		/* split brain */
+		return -EIO;
+	}
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!sources[i])
+			continue;
+		if (size <= replies[i].poststat.ia_size) {
+			size = replies[i].poststat.ia_size;
+			source = i;
+		}
+	}
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!sources[i])
+			continue;
+		if (replies[i].poststat.ia_size < size) {
+			sources[i] = 0;
+			sinks[i] = 1;
+		}
+	}
+
+	return source;
 }
 
-
-int
-afr_sh_data_lock (call_frame_t *frame, xlator_t *this)
+/*
+ * __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.
+ */
+static int
+__afr_selfheal_data_prepare (call_frame_t *frame, xlator_t *this, fd_t *fd,
+			     unsigned char *locked_on, unsigned char *sources,
+			     unsigned char *sinks, unsigned char *healed_sinks,
+			     struct afr_reply *replies)
 {
-        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;
-
-        if (sh->data_lock_held) {
-                /* caller has held the lock already,
-                   so skip locking */
-
-                afr_sh_data_fxattrop (frame, this);
-                return 0;
-        }
-
-        return afr_sh_data_lock_rec (frame, this);
+	int ret = -1;
+	int source = -1;
+	afr_private_t *priv = NULL;
+	int i = 0;
+
+	priv = this->private;
+
+	ret = afr_selfheal_unlocked_discover (frame, fd->inode, fd->inode->gfid,
+					      replies);
+	if (ret)
+		return ret;
+
+	ret = afr_selfheal_find_direction (frame, this, replies,
+					   AFR_DATA_TRANSACTION,
+					   locked_on, sources, sinks);
+	if (ret)
+		return ret;
+
+	source = __afr_selfheal_data_finalize_source (this, sources, sinks,
+						      healed_sinks, locked_on,
+						      replies);
+	if (source < 0)
+		return -EIO;
+
+	for (i = 0; i < priv->child_count; i++)
+		/* 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.
+		*/
+		healed_sinks[i] = sinks[i] && locked_on[i];
+
+	return source;
 }
 
 
-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)
+static int
+__afr_selfheal_data (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              call_count = 0;
-        int              child_index = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        child_index = (long) cookie;
-
-        /* TODO: some of the open's might fail.
-           In that case, modify cleanup fn to send flush on those
-           fd's which are already open */
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "open of %s failed on child %s (%s)",
-                                local->loc.path,
-                                priv->children[child_index]->name,
-                                strerror (op_errno));
-                        sh->op_failed = 1;
-                }
-
-                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 (sh->op_failed) {
-                        afr_sh_data_finish (frame, this);
-                        return 0;
-                }
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "fd for %s opened, commencing sync",
-                        local->loc.path);
-
-                afr_sh_data_lock (frame, this);
-        }
-
-        return 0;
+	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;
+	struct afr_reply *locked_replies = NULL;
+	int source = -1;
+	gf_boolean_t compat = _gf_false;
+	unsigned char *compat_lock = NULL;
+
+	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);
+	compat_lock = 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 < 2) {
+			ret = -ENOTCONN;
+			goto unlock;
+		}
+
+		ret = __afr_selfheal_data_prepare (frame, this, fd, data_lock,
+						   sources, sinks, healed_sinks,
+						   locked_replies);
+		if (ret < 0)
+			goto unlock;
+
+		source = ret;
+
+		ret = __afr_selfheal_truncate_sinks (frame, this, fd, healed_sinks,
+						     locked_replies,
+						     locked_replies[source].poststat.ia_size);
+		if (ret < 0)
+			goto unlock;
+
+		ret = 0;
+
+		/* Locking from (LLONG_MAX - 2) to (LLONG_MAX - 1) is for
+		   compatibility with older self-heal clients which do not
+		   hold a lock in the @priv->sh_domain domain to guard
+		   against concurrent ongoing self-heals
+		*/
+		afr_selfheal_inodelk (frame, this, fd->inode, this->name,
+				      LLONG_MAX - 2, 1, compat_lock);
+		compat = _gf_true;
+	}
+unlock:
+	afr_selfheal_uninodelk (frame, this, fd->inode, this->name, 0, 0,
+				data_lock);
+	if (ret < 0)
+		goto out;
+
+	ret = afr_selfheal_data_do (frame, this, fd, source, healed_sinks,
+				    locked_replies);
+	if (ret)
+		goto out;
+
+	ret = afr_selfheal_undo_pending (frame, this, fd->inode, sources, sinks,
+					 healed_sinks, AFR_DATA_TRANSACTION,
+					 locked_replies, data_lock);
+out:
+	if (compat)
+		afr_selfheal_uninodelk (frame, this, fd->inode, this->name,
+					LLONG_MAX - 2, 1, compat_lock);
+	return ret;
 }
 
 
-int
-afr_sh_data_open (call_frame_t *frame, xlator_t *this)
+static fd_t *
+afr_selfheal_data_open (xlator_t *this, inode_t *inode)
 {
-        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;
-
-        if (sh->healing_fd_opened) {
-                /* caller has opened the fd for us already, so skip open */
-
-                afr_sh_data_lock (frame, this);
-                return 0;
-        }
+	loc_t loc = {0,};
+	int ret = 0;
+	fd_t *fd = NULL;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
-        local->call_count = call_count;
+	fd = fd_create (inode, 0);
+	if (!fd)
+		return NULL;
 
-        fd = fd_create (local->loc.inode, frame->root->pid);
-        sh->healing_fd = fd;
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.gfid, inode->gfid);
 
-        /* open sinks */
-        for (i = 0; i < priv->child_count; i++) {
-                if(!local->child_up[i])
-                        continue;
+	ret = syncop_open (this, &loc, O_RDWR|O_LARGEFILE, fd);
+	if (ret) {
+		fd_unref (fd);
+		fd = NULL;
+	} else {
+		fd_bind (fd);
+	}
 
-                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, 0);
+	loc_wipe (&loc);
 
-                if (!--call_count)
-                        break;
-        }
-
-        return 0;
+	return fd;
 }
 
 
 int
-afr_self_heal_data (call_frame_t *frame, xlator_t *this)
+afr_selfheal_data (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 = this->private;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        if (sh->need_data_self_heal && priv->data_self_heal) {
-                afr_sh_data_open (frame, this);
-        } 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;
+	afr_private_t *priv = NULL;
+	unsigned char *locked_on = NULL;
+	int ret = 0;
+	fd_t *fd = NULL;
+
+	priv = this->private;
+
+	fd = afr_selfheal_data_open (this, inode);
+	if (!fd)
+		return -EIO;
+
+	locked_on = alloca0 (priv->child_count);
+
+	ret = afr_selfheal_tryinodelk (frame, this, inode, priv->sh_domain, 0, 0,
+				       locked_on);
+	{
+		if (ret < 2) {
+			/* 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_data (frame, this, fd, locked_on);
+	}
+unlock:
+	afr_selfheal_uninodelk (frame, this, inode, priv->sh_domain, 0, 0, locked_on);
+
+	if (fd)
+		fd_unref (fd);
+
+	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 4a5e7531d..9e714b026 100644
--- a/xlators/cluster/afr/src/afr-self-heal-entry.c
+++ b/xlators/cluster/afr/src/afr-self-heal-entry.c
@@ -1,2382 +1,629 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#include <libgen.h>
-#include <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 "afr-self-heal.h"
 #include "byte-order.h"
-
 #include "afr-transaction.h"
-#include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-
-int
-afr_sh_entry_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        /*
-          TODO: cleanup sh->*
-        */
-
-        if (sh->healing_fd)
-                fd_unref (sh->healing_fd);
-        sh->healing_fd = NULL;
-
-        /* for (i = 0; i < priv->child_count; i++) { */
-        /*        sh->locked_nodes[i] = 0; */
-        /* } */
-
-        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)
-{
-        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_fresh_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;
-}
-
-
-int
-afr_sh_entry_erase_pending (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              need_unwind = 0;
-
-        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, AFR_ENTRY_TRANSACTION);
-
-        erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count,
-                                 gf_afr_mt_dict_t);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->xattr[i]) {
-                        call_count++;
-
-                        erase_xattr[i] = get_new_dict();
-                        dict_ref (erase_xattr[i]);
-                }
-        }
-
-        if (call_count == 0)
-                need_unwind = 1;
-
-        afr_sh_delta_to_xattr (priv, sh->delta_matrix, erase_xattr,
-                               priv->child_count, AFR_ENTRY_TRANSACTION);
-
-        local->call_count = call_count;
-        for (i = 0; i < priv->child_count; i++) {
-                if (!erase_xattr[i])
-                        continue;
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "erasing pending flags from %s on %s",
-                        local->loc.path, priv->children[i]->name);
-
-                STACK_WIND_COOKIE (frame, afr_sh_entry_erase_pending_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->xattrop,
-                                   &local->loc,
-                                   GF_XATTROP_ADD_ARRAY, erase_xattr[i]);
-                if (!--call_count)
-                        break;
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (erase_xattr[i]) {
-                        dict_unref (erase_xattr[i]);
-                }
-        }
-        GF_FREE (erase_xattr);
-
-        if (need_unwind)
-                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)) {
-
-                        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
-build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name)
-{
-        int   ret = -1;
-
-        if (!child) {
-                goto out;
-        }
-
-        if (strcmp (parent->path, "/") == 0)
-                ret = gf_asprintf ((char **)&child->path, "/%s", name);
-        else
-                ret = gf_asprintf ((char **)&child->path, "%s/%s", parent->path,
-                                   name);
+afr_selfheal_entry_delete (call_frame_t *frame, xlator_t *this, inode_t *dir,
+			   const char *name, inode_t *inode, int child,
+			   struct afr_reply *replies)
+{
+	afr_private_t *priv = NULL;
+	xlator_t *subvol = NULL;
+	int ret = 0;
+	loc_t loc = {0, };
+	char g[64];
+
+	priv = this->private;
+
+	subvol = priv->children[child];
+
+	loc.parent = inode_ref (dir);
+	uuid_copy (loc.pargfid, dir->gfid);
+	loc.name = name;
+	loc.inode = inode_ref (inode);
+
+	if (replies[child].valid && replies[child].op_ret == 0) {
+		switch (replies[child].poststat.ia_type) {
+		case IA_IFDIR:
+			gf_log (this->name, GF_LOG_WARNING,
+				"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);
+			break;
+		default:
+			gf_log (this->name, GF_LOG_WARNING,
+				"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);
+			break;
+		}
+	}
+
+	loc_wipe (&loc);
+
+	return ret;
+}
+
+
+int
+afr_selfheal_recreate_entry (call_frame_t *frame, xlator_t *this, int dst,
+			     int source, inode_t *dir, const char *name,
+			     inode_t *inode, struct afr_reply *replies)
+{
+	int ret = 0;
+	loc_t loc = {0,};
+	loc_t srcloc = {0,};
+	afr_private_t *priv = NULL;
+	dict_t *xdata = NULL;
+	struct iatt *iatt = NULL;
+	char *linkname = NULL;
+	mode_t mode = 0;
+	struct iatt newent = {0,};
+
+	priv = this->private;
+
+	xdata = dict_new();
+	if (!xdata)
+		return -ENOMEM;
+
+	loc.parent = inode_ref (dir);
+	uuid_copy (loc.pargfid, dir->gfid);
+	loc.name = name;
+	loc.inode = inode_ref (inode);
+
+	ret = afr_selfheal_entry_delete (frame, this, dir, name, inode, dst,
+					 replies);
+	if (ret)
+		goto out;
+
+	ret = dict_set_static_bin (xdata, "gfid-req",
+				   replies[source].poststat.ia_gfid, 16);
+	if (ret)
+		goto out;
+
+	iatt = &replies[source].poststat;
+
+	srcloc.inode = inode_ref (inode);
+	uuid_copy (srcloc.gfid, iatt->ia_gfid);
+
+	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, xdata, 0);
+		break;
+	case IA_IFLNK:
+		ret = syncop_lookup (priv->children[dst], &srcloc, 0, 0, 0, 0);
+		if (ret == 0) {
+			ret = syncop_link (priv->children[dst], &srcloc, &loc);
+		} else {
+			ret = syncop_readlink (priv->children[source], &srcloc,
+					       &linkname, 4096);
+			if (ret <= 0)
+				goto out;
+			ret = syncop_symlink (priv->children[dst], &loc, linkname,
+					      xdata, NULL);
+		}
+		break;
+	default:
+		ret = dict_set_int32 (xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1);
+		if (ret)
+			goto out;
+		ret = syncop_mknod (priv->children[dst], &loc, mode,
+				    iatt->ia_rdev, xdata, &newent);
+		if (ret == 0 && iatt->ia_size && !newent.ia_size) {
+			/* New entry created. Mark @dst pending on all sources */
+			ret = 1;
+		}
+		break;
+	}
 
-        if (-1 == ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "asprintf failed while setting child path");
-        }
-
-        if (!child->path) {
-                goto out;
-        }
-
-        child->name = strrchr (child->path, '/');
-        if (child->name)
-                child->name++;
-
-        child->parent = inode_ref (parent->inode);
-        child->inode = inode_new (parent->inode->table);
-
-        if (!child->inode) {
-                goto out;
-        }
-
-        ret = 0;
 out:
-        if (ret == -1)
-                loc_wipe (child);
-
-        return ret;
-}
-
-
-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 active_src);
-
-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)
-{
-        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)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *expunge_local = NULL;
-        afr_self_heal_t *expunge_sh = NULL;
-        int              active_src = 0;
-        call_frame_t    *frame = NULL;
-        int32_t          valid = 0;
-
-        priv = this->private;
-        expunge_local = expunge_frame->local;
-        expunge_sh = &expunge_local->self_heal;
-        frame = expunge_sh->sh_frame;
-
-        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;
-        afr_build_parent_loc (&expunge_sh->parent_loc, &expunge_local->loc);
-
-        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);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_expunge_unlink (call_frame_t *expunge_frame, xlator_t *this,
-                             int active_src)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *expunge_local = NULL;
-
-        priv          = this->private;
-        expunge_local = expunge_frame->local;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "expunging file %s on %s",
-                expunge_local->loc.path, priv->children[active_src]->name);
-
-        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_remove_cbk,
-                           (void *) (long) active_src,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->unlink,
-                           &expunge_local->loc);
-
-        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);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_expunge_remove (call_frame_t *expunge_frame, xlator_t *this,
-                             int active_src, struct iatt *buf)
-{
-        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;
-
-        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;
-
-        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_expunge_unlink (expunge_frame, this, active_src);
-                break;
-        case IA_IFDIR:
-                afr_sh_entry_expunge_rmdir (expunge_frame, this, active_src);
-                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;
-        }
-
-        return 0;
-out:
-        AFR_STACK_DESTROY (expunge_frame);
-        sh->expunge_done (frame, this, active_src, -1, EINVAL);
-
-        return 0;
-}
-
-
-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_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));
-                goto out;
-        }
-
-        afr_sh_entry_expunge_remove (expunge_frame, this, active_src, buf);
-
-        return 0;
-out:
-        AFR_STACK_DESTROY (expunge_frame);
-        sh->expunge_done (frame, this, active_src, op_ret, op_errno);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_expunge_purge (call_frame_t *expunge_frame, xlator_t *this,
-                            int active_src)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *expunge_local = NULL;
-
-        priv = this->private;
-        expunge_local = expunge_frame->local;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "looking up %s on %s",
-                expunge_local->loc.path, priv->children[active_src]->name);
-
-        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_lookup_cbk,
-                           (void *) (long) active_src,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->lookup,
-                           &expunge_local->loc, 0);
-
-        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 (need_expunge) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "missing entry %s on %s",
-                        expunge_local->loc.path,
-                        priv->children[source]->name);
-
-                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);
-        } 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)
-            || ((strcmp (local->loc.path, "/") == 0)
-                && (strcmp (name, GF_REPLICATE_TRASH_DIR) == 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 existance of %s under %s",
-                name, local->loc.path);
-
-        expunge_frame = copy_frame (frame);
-        if (!expunge_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        ALLOC_OR_GOTO (expunge_local, afr_local_t, 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;
-
-        ret = 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, 0);
-
-        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)
-{
-        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);
-
-        return 0;
+	if (xdata)
+		dict_unref (xdata);
+	loc_wipe (&loc);
+	loc_wipe (&srcloc);
+	return ret;
 }
 
 
-int
-afr_sh_entry_expunge_all (call_frame_t *frame, xlator_t *this)
+static 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)
 {
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        int              active_src = -1;
+	int ret = 0;
+	int i = 0;
+	afr_private_t *priv = NULL;
+	dict_t *xattr = NULL;
+	int **changelog = NULL;
+	int idx = 0;
 
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
+	priv = this->private;
 
-        sh->offset = 0;
+	idx = afr_index_for_transaction_type (AFR_DATA_TRANSACTION);
 
-        if (sh->source == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no active sources for %s to expunge entries",
-                        local->loc.path);
-                goto out;
-        }
+	uuid_copy (inode->gfid, replies[source].poststat.ia_gfid);
 
-        active_src = next_active_sink (frame, this, sh->active_source);
-        sh->active_source = active_src;
+	changelog = afr_matrix_create (priv->child_count, AFR_NUM_CHANGE_LOGS);
 
-        if (sh->op_failed) {
-                goto out;
-        }
+	xattr = dict_new();
+	if (!xattr)
+		return -ENOMEM;
 
-        if (active_src == -1) {
-                /* completed creating missing files on all subvolumes */
-                goto out;
-        }
+	for (i = 0; i < priv->child_count; i++) {
+		if (!newentry[i])
+			continue;
+		changelog[i][idx] = hton32(1);
+	}
 
-        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_set_pending_dict (priv, xattr, changelog);
 
-        afr_sh_entry_expunge_subvol (frame, this, active_src);
-
-        return 0;
-out:
-        afr_sh_entry_impunge_all (frame, this);
-        return 0;
+	for (i = 0; i < priv->child_count; i++) {
+		if (!sources[i])
+			continue;
+		afr_selfheal_post_op (frame, this, inode, i, xattr);
+	}
 
+	dict_unref (xattr);
+	return ret;
 }
 
 
-int
-afr_sh_entry_impunge_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_impunge_subvol (frame, this, active_src);
-
-        return 0;
+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)
+{
+	int ret = 0;
+	afr_private_t *priv = NULL;
+	int i = 0;
+	unsigned char *newentry = NULL;
+
+	priv = this->private;
+	newentry = alloca0 (priv->child_count);
+
+	if (!replies[source].valid)
+		return -EIO;
+
+	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 (frame, this, fd->inode,
+							 name, inode, i, replies);
+		} else {
+			if (!uuid_compare (replies[i].poststat.ia_gfid,
+					   replies[source].poststat.ia_gfid))
+				continue;
+
+			ret = afr_selfheal_recreate_entry (frame, this, i, source,
+							   fd->inode, name, inode,
+							   replies);
+			if (ret > 0) {
+				newentry[i] = 1;
+				ret = 0;
+			}
+		}
+		if (ret < 0)
+			break;
+	}
+
+	if (AFR_COUNT (newentry, priv->child_count))
+		afr_selfheal_newentry_mark (frame, this, inode, source, replies,
+					    sources, newentry);
+	return ret;
 }
 
 
-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)
+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;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        call_frame_t    *frame = NULL;
-        int              active_src = 0;
-        int              child_index = 0;
-        int32_t          impunge_ret_child = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        frame = impunge_sh->sh_frame;
-        local = frame->local;
-        sh    = &local->self_heal;
-        active_src = impunge_sh->active_source;
-        child_index = (long) cookie;
-
-        if (op_ret == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "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));
-        }
-
-        LOCK (&impunge_frame->lock);
-        {
-                call_count = --impunge_local->call_count;
-        }
-        UNLOCK (&impunge_frame->lock);
-
-        if (call_count == 0) {
-                impunge_ret_child = impunge_sh->impunge_ret_child;
-                AFR_STACK_DESTROY (impunge_frame);
-                sh->impunge_done (frame, this, impunge_ret_child, op_ret,
-                                  op_errno);
-        }
-
-        return 0;
-}
+	int ret = 0;
+	afr_private_t *priv = NULL;
+	int i = 0;
+	int source = -1;
 
+	priv = this->private;
 
-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)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              child_index = 0;
-        struct iatt      stbuf = {0};
-        int32_t          valid = 0;
-
-        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,
-                        "%s: failed to perform xattrop on %s (%s)",
-                        impunge_local->loc.path,
-                        priv->children[child_index]->name,
-                        strerror (op_errno));
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "setting ownership of %s on %s to %d/%d",
-                impunge_local->loc.path,
-                priv->children[child_index]->name,
-                impunge_local->cont.lookup.buf.ia_uid,
-                impunge_local->cont.lookup.buf.ia_gid);
-
-        stbuf.ia_atime = impunge_local->cont.lookup.buf.ia_atime;
-        stbuf.ia_atime_nsec = impunge_local->cont.lookup.buf.ia_atime_nsec;
-        stbuf.ia_mtime = impunge_local->cont.lookup.buf.ia_mtime;
-        stbuf.ia_mtime_nsec = impunge_local->cont.lookup.buf.ia_mtime_nsec;
-
-        stbuf.ia_uid = impunge_local->cont.lookup.buf.ia_uid;
-        stbuf.ia_gid = impunge_local->cont.lookup.buf.ia_gid;
-
-        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) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->setattr,
-                           &impunge_local->loc,
-                           &stbuf, valid);
-        return 0;
-}
+	for (i = 0; i < priv->child_count; i++) {
+		if (replies[i].valid && replies[i].op_ret == 0) {
+			source = i;
+			break;
+		}
+	}
 
+	if (source == -1) {
+		/* entry got deleted in the mean time? */
+		return 0;
+	}
 
-int
-afr_sh_entry_impunge_parent_setattr_cbk (call_frame_t *setattr_frame,
-                                         void *cookie, xlator_t *this,
-                                         int32_t op_ret, int32_t op_errno,
-                                         struct iatt *preop, struct iatt *postop)
-{
-        loc_t *parent_loc = cookie;
-
-        if (op_ret != 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "setattr on parent directory (%s) failed: %s",
-                        parent_loc->path, strerror (op_errno));
-        }
+	for (i = 0; i < priv->child_count; i++) {
+		if (i == source || !healed_sinks[i])
+			continue;
 
-        loc_wipe (parent_loc);
+		if (replies[i].op_errno != ENOENT)
+			continue;
 
-        GF_FREE (parent_loc);
+		ret = afr_selfheal_recreate_entry (frame, this, i, source,
+						   fd->inode, name, inode,
+						   replies);
+	}
 
-        AFR_STACK_DESTROY (setattr_frame);
-        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)
+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;
-        call_frame_t    *frame            = NULL;
-        int              active_src       = 0;
-        int              child_index      = 0;
-        int32_t         *pending_array    = NULL;
-        dict_t          *xattr            = NULL;
-        int              ret              = 0;
-        int              idx              = 0;
-        afr_local_t     *local            = NULL;
-        afr_self_heal_t *sh               = NULL;
-        call_frame_t    *setattr_frame    = NULL;
-        int32_t          valid            = 0;
-        loc_t           *parent_loc       = NULL;
-        struct iatt      parentbuf        = {0,};
-        int32_t          impunge_ret_child = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        frame = impunge_sh->sh_frame;
-        local = frame->local;
-        sh    = &local->self_heal;
-        active_src = impunge_sh->active_source;
-
-        child_index = (long) cookie;
-
-        if (op_ret == -1) {
-                ret = -1;
-                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));
-                goto out;
-        }
-
-        inode->ia_type = stbuf->ia_type;
-
-        xattr = dict_new ();
-        if (!xattr) {
-                ret = -1;
-                goto out;
-        }
-
-        pending_array = (int32_t*) GF_CALLOC (3, sizeof (*pending_array),
-                                              gf_afr_mt_int32_t);
-
-        if (!pending_array) {
-                ret = -1;
-                goto out;
-        }
-        idx = afr_index_for_transaction_type (AFR_METADATA_TRANSACTION);
-        pending_array[idx] = hton32 (1);
-        if (IA_ISDIR (stbuf->ia_type))
-                idx = afr_index_for_transaction_type (AFR_ENTRY_TRANSACTION);
-        else
-                idx = afr_index_for_transaction_type (AFR_DATA_TRANSACTION);
-        pending_array[idx] = hton32 (1);
-
-        ret = dict_set_dynptr (xattr, priv->pending_key[child_index],
-                               pending_array,
-                               3 * sizeof (*pending_array));
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Unable to set dict value.");
-        } else {
-                pending_array = NULL;
-        }
-
-        valid         = GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-        parentbuf     = impunge_sh->parentbuf;
-        setattr_frame = copy_frame (impunge_frame);
-
-        parent_loc = GF_CALLOC (1, sizeof (*parent_loc),
-                                gf_afr_mt_loc_t);
-        if (!parent_loc) {
-                ret = -1;
-                goto out;
-        }
-        afr_build_parent_loc (parent_loc, &impunge_local->loc);
-
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_xattrop_cbk,
-                           (void *) (long) child_index,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->xattrop,
-                           &impunge_local->loc, GF_XATTROP_ADD_ARRAY, xattr);
-
-        STACK_WIND_COOKIE (setattr_frame, afr_sh_entry_impunge_parent_setattr_cbk,
-                           (void *) (long) parent_loc,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->setattr,
-                           parent_loc, &parentbuf, valid);
+	int ret = -1;
 
-out:
-        if (xattr)
-                dict_unref (xattr);
-
-        if (ret) {
-                if (pending_array)
-                        GF_FREE (pending_array);
-
-                LOCK (&impunge_frame->lock);
-                {
-                        call_count = --impunge_local->call_count;
-                }
-                UNLOCK (&impunge_frame->lock);
-
-                if (call_count == 0) {
-                        impunge_ret_child = impunge_sh->impunge_ret_child;
-                        AFR_STACK_DESTROY (impunge_frame);
-                        sh->impunge_done (frame, this, impunge_ret_child, -1,
-                                          op_errno);
-                }
-        }
-
-        return 0;
+	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_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;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "creating missing file %s on %s",
-                impunge_local->loc.path,
-                priv->children[child_index]->name);
-
-        dict = dict_new ();
-        if (!dict)
-                gf_log (this->name, GF_LOG_ERROR, "Out of memory");
-
-        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);
-
-        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)), dict);
-
-        if (dict)
-                dict_unref (dict);
-
-        return 0;
+static int
+afr_selfheal_entry_dirent (call_frame_t *frame, xlator_t *this, fd_t *fd,
+			   int source, unsigned char *sources,
+			   unsigned char *healed_sinks, char *name)
+{
+	afr_private_t *priv = NULL;
+	int ret = 0;
+	unsigned char *locked_on = NULL;
+	struct afr_reply *replies = NULL;
+	inode_t *inode = NULL;
+
+	priv = this->private;
+
+	locked_on = alloca0 (priv->child_count);
+
+	replies = alloca0 (priv->child_count * sizeof(*replies));
+
+	ret = afr_selfheal_entrylk (frame, this, fd->inode, this->name,
+				    name, locked_on);
+	{
+		if (ret < 2) {
+			ret = -ENOTCONN;
+			goto unlock;
+		}
+
+		inode = afr_selfheal_unlocked_lookup_on (frame, fd->inode, name,
+							 replies, locked_on);
+		if (!inode) {
+			ret = -ENOMEM;
+			goto unlock;
+		}
+
+		ret = __afr_selfheal_entry_dirent (frame, this, fd, name, inode,
+						   source, sources, healed_sinks,
+						   locked_on, replies);
+	}
+unlock:
+	afr_selfheal_unentrylk (frame, this, fd->inode, this->name, name,
+				locked_on);
+	if (inode)
+		inode_unref (inode);
+	return ret;
 }
 
 
-
-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_do_subvol (call_frame_t *frame, xlator_t *this, fd_t *fd,
+			      int child, int source, unsigned char *sources,
+			      unsigned char *healed_sinks)
 {
-        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;
-
-        dict = dict_new ();
-        if (!dict) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                return 0;
-        }
-
-        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),
-                           dict);
-
-        if (dict)
-                dict_unref (dict);
-
-        return 0;
-}
+	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;
 
+	priv = this->private;
+	subvol = priv->children[child];
 
-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;
-        call_frame_t    *frame         = NULL;
-        afr_local_t     *local         = NULL;
-        afr_self_heal_t *sh            = NULL;
-        afr_self_heal_t *impunge_sh    = NULL;
-        int32_t          impunge_ret_child = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh    = &impunge_local->self_heal;
-        frame         = impunge_sh->sh_frame;
-        local         = frame->local;
-        sh            = &local->self_heal;
-
-        buf = &impunge_local->cont.symlink.buf;
-
-        dict = dict_new ();
-        if (!dict) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                impunge_ret_child = impunge_sh->impunge_ret_child;
-                AFR_STACK_DESTROY (impunge_frame);
-                sh->impunge_done (impunge_frame, this, impunge_ret_child, -1,
-                                  ENOMEM);
-                goto out;
-        }
-
-        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, dict);
-
-        if (dict)
-                dict_unref (dict);
-out:
-        return 0;
-}
+	INIT_LIST_HEAD (&entries.list);
 
+	iter_frame = afr_copy_frame (frame);
+	if (!iter_frame)
+		return -ENOMEM;
 
-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)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              child_index = -1;
-        call_frame_t    *frame = NULL;
-        int              call_count = -1;
-        int              active_src = -1;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        int32_t          impunge_ret_child = 0;
-
-        priv          = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh    = &impunge_local->self_heal;
-        frame         = impunge_sh->sh_frame;
-        local         = frame->local;
-        sh            = &local->self_heal;
-        active_src    = impunge_sh->active_source;
-
-        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) {
-                impunge_ret_child = impunge_sh->impunge_ret_child;
-                AFR_STACK_DESTROY (impunge_frame);
-                sh->impunge_done (frame, this, impunge_ret_child, op_ret,
-                                  op_errno);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_impunge_symlink_unlink (call_frame_t *impunge_frame, xlator_t *this,
-                                     int child_index)
-{
-        afr_private_t   *priv          = NULL;
-        afr_local_t     *impunge_local = NULL;
+	while ((ret = syncop_readdir (subvol, fd, 131072, offset, &entries))) {
+		if (ret > 0)
+			ret = 0;
+		list_for_each_entry (entry, &entries.list, list) {
+			offset = entry->d_off;
 
-        priv          = this->private;
-        impunge_local = impunge_frame->local;
+			if (!strcmp (entry->d_name, ".") ||
+			    !strcmp (entry->d_name, ".."))
+				continue;
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "unlinking symlink %s with wrong target on %s",
-                impunge_local->loc.path,
-                priv->children[child_index]->name);
+			if (__is_root_gfid (fd->inode->gfid) &&
+			    !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR))
+				continue;
 
-        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);
+			ret = afr_selfheal_entry_dirent (iter_frame, this, fd,
+							 source, sources,
+							 healed_sinks,
+							 entry->d_name);
+			AFR_STACK_RESET (iter_frame);
 
-        return 0;
-}
+			if (ret)
+				break;
+		}
 
+		gf_dirent_free (&entries);
+		if (ret)
+			break;
+	}
 
-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)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              child_index = -1;
-        call_frame_t    *frame = NULL;
-        int              call_count = -1;
-        int              active_src = -1;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        int32_t          impunge_ret_child = 0;
-
-        priv          = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh    = &impunge_local->self_heal;
-        frame         = impunge_sh->sh_frame;
-        local         = frame->local;
-        sh            = &local->self_heal;
-        active_src    = impunge_sh->active_source;
-
-        child_index = (long) cookie;
-
-        if ((op_ret == -1) && (op_errno != ENOENT)) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "readlink of %s on %s failed (%s)",
-                        impunge_local->loc.path,
-                        priv->children[active_src]->name,
-                        strerror (op_errno));
-                goto out;
-        }
-
-        /* symlink doesn't exist on the sink */
-
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                afr_sh_entry_impunge_symlink (impunge_frame, this,
-                                              child_index, impunge_sh->linkname);
-                return 0;
-        }
-
-
-        /* symlink exists on the sink, so check if targets match */
-
-        if (strcmp (linkname, impunge_sh->linkname) == 0) {
-                /* targets match, nothing to do */
-
-                goto out;
-        } else {
-                /*
-                 * Hah! Sneaky wolf in sheep's clothing!
-                 */
-                afr_sh_entry_impunge_symlink_unlink (impunge_frame, this,
-                                                     child_index);
-                return 0;
-        }
-
-out:
-        LOCK (&impunge_frame->lock);
-        {
-                call_count = --impunge_local->call_count;
-        }
-        UNLOCK (&impunge_frame->lock);
-
-        if (call_count == 0) {
-                impunge_ret_child = impunge_sh->impunge_ret_child;
-                AFR_STACK_DESTROY (impunge_frame);
-                sh->impunge_done (frame, this, impunge_ret_child, op_ret,
-                                  op_errno);
-        }
-
-        return 0;
+	AFR_STACK_DESTROY (iter_frame);
+	return ret;
 }
 
-
-int
-afr_sh_entry_impunge_readlink_sink (call_frame_t *impunge_frame, xlator_t *this,
-                                    int child_index)
+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,
+		       struct afr_reply *locked_replies)
 {
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
+	int i = 0;
+	afr_private_t *priv = NULL;
+	int ret = 0;
 
-        priv = this->private;
-        impunge_local = impunge_frame->local;
+	priv = this->private;
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "checking symlink target of %s on %s",
-                impunge_local->loc.path, priv->children[child_index]->name);
+	gf_log (this->name, GF_LOG_INFO, "performing entry selfheal on %s",
+		uuid_utoa (fd->inode->gfid));
 
-        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);
-
-        return 0;
+	for (i = 0; i < priv->child_count; i++) {
+		if (i != source && !healed_sinks[i])
+			continue;
+		ret = afr_selfheal_entry_do_subvol (frame, this, fd, i, source,
+						    sources, healed_sinks);
+		if (ret)
+			break;
+	}
+	return ret;
 }
 
 
-int
-afr_sh_entry_impunge_readlink_cbk (call_frame_t *impunge_frame, void *cookie,
-                                   xlator_t *this,
-                                   int32_t op_ret, int32_t op_errno,
-                                   const char *linkname, struct iatt *sbuf)
+static int
+__afr_selfheal_entry_finalize_source (xlator_t *this, unsigned char *sources,
+				      unsigned char *sinks,
+				      unsigned char *locked_on,
+				      struct afr_reply *replies)
 {
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              child_index = -1;
-        call_frame_t    *frame = NULL;
-        int              call_count = -1;
-        int              active_src = -1;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        int32_t          impunge_ret_child = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        frame = impunge_sh->sh_frame;
-        local         = frame->local;
-        sh            = &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);
-        {
-                call_count = --impunge_local->call_count;
-        }
-        UNLOCK (&impunge_frame->lock);
-
-        if (call_count == 0) {
-                impunge_ret_child = impunge_sh->impunge_ret_child;
-                AFR_STACK_DESTROY (impunge_frame);
-                sh->impunge_done (frame, this, impunge_ret_child, op_ret,
-                                  op_errno);
-        }
-
-        return 0;
-}
+	int i = 0;
+	afr_private_t *priv = NULL;
+	int source = -1;
+	int locked_count = 0;
+	int sources_count = 0;
+	int sinks_count = 0;
 
+	priv = this->private;
 
-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.symlink.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);
-
-        return 0;
-}
+	locked_count = AFR_COUNT (locked_on, priv->child_count);
+	sources_count = AFR_COUNT (sources, priv->child_count);
+	sinks_count = AFR_COUNT (sinks, priv->child_count);
 
-int
-afr_sh_entry_impunge_create (call_frame_t *impunge_frame, xlator_t *this,
-                             int child_index, struct iatt *buf,
-                             struct iatt *postparent)
-{
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        afr_private_t   *priv = NULL;
-        ia_type_t       type = IA_INVAL;
-        int             ret = 0;
-        int             active_src = 0;
-
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        impunge_sh->parentbuf = *postparent;
-        active_src = impunge_sh->active_source;
-        impunge_local->cont.lookup.buf = *buf;
-        afr_update_loc_gfids (&impunge_local->loc, buf, postparent);
-
-        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;
-        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);
-                ret = -1;
-                break;
-        }
-
-        return ret;
-}
+	if (locked_count == sinks_count || !sources_count) {
+		return -1;
+	}
 
-int
-afr_sh_entry_impunge_recreate_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)
-{
-        afr_private_t   *priv = 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;
-        int              active_src = 0;
-        int              child_index = 0;
-        call_frame_t    *frame = NULL;
-        int              call_count = 0;
-        int              ret = 0;
-        int32_t          impunge_ret_child = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        frame = impunge_sh->sh_frame;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        child_index = (long) cookie;
-
-        active_src = impunge_sh->active_source;
-
-        if (op_ret != 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "looking up %s on %s (for %s) failed (%s)",
-                        impunge_local->loc.path,
-                        priv->children[active_src]->name,
-                        priv->children[child_index]->name,
-                        strerror (op_errno));
-                goto out;
-        }
-
-        ret = afr_sh_entry_impunge_create (impunge_frame, this, child_index, buf,
-                                           postparent);
-        if (ret)
-                goto out;
-
-        return 0;
+	for (i = 0; i < priv->child_count; i++) {
+		if (sources[i]) {
+			source = i;
+			break;
+		}
+	}
 
-out:
-        LOCK (&impunge_frame->lock);
-        {
-                call_count = --impunge_local->call_count;
-        }
-        UNLOCK (&impunge_frame->lock);
-
-        if (call_count == 0) {
-                impunge_ret_child = impunge_sh->impunge_ret_child;
-                AFR_STACK_DESTROY (impunge_frame);
-                sh->impunge_done (frame, this, impunge_ret_child, op_ret,
-                                  op_errno);
-        }
-
-        return 0;
+	return source;
 }
 
 
-int
-afr_sh_entry_impunge_recreate (call_frame_t *impunge_frame, xlator_t *this,
-                               int child_index)
+static int
+__afr_selfheal_entry_prepare (call_frame_t *frame, xlator_t *this, fd_t *fd,
+			      unsigned char *locked_on, unsigned char *sources,
+			      unsigned char *sinks, unsigned char *healed_sinks,
+			      struct afr_reply *replies, int *source_p)
 {
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              active_src = 0;
+	int ret = -1;
+	int source = -1;
+	afr_private_t *priv = NULL;
+	int i = 0;
 
+	priv = this->private;
 
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
+	ret = afr_selfheal_unlocked_discover (frame, fd->inode, fd->inode->gfid,
+					      replies);
+	if (ret)
+		return ret;
 
-        active_src = impunge_sh->active_source;
+	ret = afr_selfheal_find_direction (frame, this, replies,
+					   AFR_ENTRY_TRANSACTION,
+					   locked_on, sources, sinks);
+	if (ret)
+		return ret;
 
-        STACK_WIND_COOKIE (impunge_frame,
-                           afr_sh_entry_impunge_recreate_lookup_cbk,
-                           (void *) (long) child_index,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->lookup,
-                           &impunge_local->loc, 0);
+	source = __afr_selfheal_entry_finalize_source (this, sources, sinks,
+						       locked_on, replies);
+	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 0;
-}
+	for (i = 0; i < priv->child_count; i++)
+		/* 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.
+		*/
+		healed_sinks[i] = sinks[i] && locked_on[i];
 
-int
-afr_sh_entry_impunge_entry_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 *x,
-                                struct iatt *postparent)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              call_count = 0;
-        int              child_index = 0;
-        call_frame_t    *frame = NULL;
-        int              active_src = 0;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        int32_t          impunge_ret_child = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        frame = impunge_sh->sh_frame;
-        local         = frame->local;
-        sh            = &local->self_heal;
-        child_index = (long) cookie;
-        active_src = impunge_sh->active_source;
-
-        if ((op_ret == -1 && op_errno == ENOENT)
-            || (IA_ISLNK (impunge_sh->impunging_entry_mode))) {
-
-                /*
-                 * A symlink's target might have changed, so
-                 * always go down the recreate path for them.
-                 */
-
-                /* decrease call_count in recreate-callback */
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "missing entry %s on %s",
-                        impunge_local->loc.path,
-                        priv->children[child_index]->name);
-
-                afr_sh_entry_impunge_recreate (impunge_frame, this,
-                                               child_index);
-                return 0;
-        }
-
-        if (op_ret == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "%s exists under %s",
-                        impunge_local->loc.path,
-                        priv->children[child_index]->name);
-
-                impunge_sh->parentbuf = *postparent;
-        } else {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "looking up %s under %s failed (%s)",
-                        impunge_local->loc.path,
-                        priv->children[child_index]->name,
-                        strerror (op_errno));
-        }
-
-        LOCK (&impunge_frame->lock);
-        {
-                call_count = --impunge_local->call_count;
-        }
-        UNLOCK (&impunge_frame->lock);
-
-        if (call_count == 0) {
-                impunge_ret_child = impunge_sh->impunge_ret_child;
-                AFR_STACK_DESTROY (impunge_frame);
-                sh->impunge_done (frame, this, impunge_ret_child, op_ret,
-                                  op_errno);
-        }
-
-        return 0;
+	return ret;
 }
 
 
-int
-afr_sh_entry_impunge_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    *impunge_frame = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              active_src = 0;
-        int              i = 0;
-        int              call_count = 0;
-        int              op_errno = 0;
-        int              op_ret = -1;
-
-        priv = this->private;
-        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)
-            || ((strcmp (local->loc.path, "/") == 0)
-                && (strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR) == 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 existance of %s under %s",
-                entry->d_name, local->loc.path);
-
-        impunge_frame = copy_frame (frame);
-        if (!impunge_frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        ALLOC_OR_GOTO (impunge_local, afr_local_t, out);
-
-        impunge_frame->local = impunge_local;
-        impunge_sh = &impunge_local->self_heal;
-        impunge_sh->sh_frame = frame;
-        impunge_sh->active_source = active_src;
-        impunge_sh->impunge_ret_child = active_src;
-
-        impunge_sh->impunging_entry_mode =
-                st_mode_from_ia (entry->d_stat.ia_prot, entry->d_stat.ia_type);
-
-        ret = build_child_loc (this, &impunge_local->loc, &local->loc, entry->d_name);
-        if (ret != 0) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (i == active_src)
-                        continue;
-                if (local->child_up[i] == 0)
-                        continue;
-                if (sh->sources[i] == 1)
-                        continue;
-                call_count++;
-        }
-
-        impunge_local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (i == active_src)
-                        continue;
-                if (local->child_up[i] == 0)
-                        continue;
-                if (sh->sources[i] == 1)
-                        continue;
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "looking up %s on %s", impunge_local->loc.path,
-                        priv->children[i]->name);
-
-                STACK_WIND_COOKIE (impunge_frame,
-                                   afr_sh_entry_impunge_entry_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->lookup,
-                                   &impunge_local->loc, 0);
-
-                if (!--call_count)
-                        break;
-        }
-
-        ret = 0;
+static int
+__afr_selfheal_entry (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;
+	struct afr_reply *locked_replies = NULL;
+	int source = -1;
+
+	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);
+
+	locked_replies = alloca0 (sizeof (*locked_replies) * priv->child_count);
+
+	ret = afr_selfheal_entrylk (frame, this, fd->inode, this->name, NULL,
+				    data_lock);
+	{
+		if (ret < 2) {
+			ret = -ENOTCONN;
+			goto unlock;
+		}
+
+		ret = __afr_selfheal_entry_prepare (frame, this, fd, data_lock,
+						    sources, sinks, healed_sinks,
+						    locked_replies, &source);
+	}
+unlock:
+	afr_selfheal_unentrylk (frame, this, fd->inode, this->name, NULL,
+				data_lock);
+	if (ret < 0)
+		goto out;
+
+	ret = afr_selfheal_entry_do (frame, this, fd, source, sources,
+				     healed_sinks, locked_replies);
+	if (ret)
+		goto out;
+
+	ret = afr_selfheal_undo_pending (frame, this, fd->inode, sources, sinks,
+					 healed_sinks, AFR_ENTRY_TRANSACTION,
+					 locked_replies, data_lock);
 out:
-        if (ret == -1)
-                sh->impunge_done (frame, this, active_src, op_ret, op_errno);
-
-        return 0;
+	return ret;
 }
 
 
-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)
+static fd_t *
+afr_selfheal_data_opendir (xlator_t *this, inode_t *inode)
 {
-        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_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_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_impunge_entry (frame, this, entry);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_impunge_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_impunge_readdir_cbk,
-                    priv->children[active_src],
-                    priv->children[active_src]->fops->readdirp,
-                    sh->healing_fd, sh->block_size, sh->offset);
-
-        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 (sh->op_failed) {
-                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, active_src);
-
-        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)
-{
-        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->op_failed = 1;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if (sh->op_failed) {
-                        afr_sh_entry_finish (frame, this);
-                        return 0;
-                }
-                gf_log (this->name, GF_LOG_TRACE,
-                        "fd for %s opened, commencing sync",
-                        local->loc.path);
-
-                sh->active_source = -1;
-                afr_sh_entry_expunge_all (frame, this);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_open (call_frame_t *frame, xlator_t *this)
-{
-        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 = 65536; //131072
-        sh->offset = 0;
-
-        call_count = sh->active_sinks;
-        if (source != -1)
-                call_count++;
+	loc_t loc = {0,};
+	int ret = 0;
+	fd_t *fd = NULL;
 
-        local->call_count = call_count;
+	fd = fd_create (inode, 0);
+	if (!fd)
+		return NULL;
 
-        fd = fd_create (local->loc.inode, frame->root->pid);
-        sh->healing_fd = fd;
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.gfid, inode->gfid);
 
-        if (source != -1) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "opening directory %s on subvolume %s (source)",
-                        local->loc.path, priv->children[source]->name);
+	ret = syncop_opendir (this, &loc, fd);
+	if (ret) {
+		fd_unref (fd);
+		fd = NULL;
+	} else {
+		fd_bind (fd);
+	}
 
-                /* 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);
-                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);
-
-                if (!--call_count)
-                        break;
-        }
-
-        return 0;
-}
-
-
-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              active_sinks = 0;
-        int              source = 0;
-        int              i = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        source = sh->source;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->sources[i] == 0 && local->child_up[i] == 1) {
-                        active_sinks++;
-                        sh->success[i] = 1;
-                }
-        }
-        if (source != -1)
-                sh->success[source] = 1;
-
-        if (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 && 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;
-        }
-        sh->active_sinks = active_sinks;
-
-        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,
-                        active_sinks);
-        else
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no active sources for %s found. "
-                        "merging all entries as a conservative decision",
-                        local->loc.path);
-
-        afr_sh_entry_open (frame, this);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_fix (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;
-
-        int nsources = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        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);
-        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 0;
-        }
-
-        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);
-        afr_inode_set_read_ctx (this, sh->inode, sh->source,
-                                sh->fresh_children);
-
-
-heal:
-        afr_sh_entry_sync_prepare (frame, this);
-
-        return 0;
+	loc_wipe (&loc);
+        return fd;
 }
 
 
-
 int
-afr_sh_entry_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_selfheal_entry (call_frame_t *frame, xlator_t *this, inode_t *inode)
 {
-        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;
+	afr_private_t *priv = NULL;
+	unsigned char *locked_on = NULL;
+	fd_t *fd = NULL;
+	int ret = 0;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret != -1) {
-                        sh->xattr[child_index] = dict_ref (xattr);
-                        sh->buf[child_index] = *buf;
-                        sh->success_children[sh->success_count] = child_index;
-                        sh->success_count++;
-                }
-        }
-        UNLOCK (&frame->lock);
+	priv = this->private;
 
-        call_count = afr_frame_return (frame);
+	fd = afr_selfheal_data_opendir (this, inode);
+	if (!fd)
+		return -EIO;
 
-        if (call_count == 0) {
-                afr_sh_entry_fix (frame, this);
-        }
-
-        return 0;
-}
-
-int
-afr_sh_post_nonblocking_entry_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_ERROR, "Non Blocking entrylks "
-                        "failed for %s.", local->loc.path);
-                sh->op_failed = 1;
-                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_lookup_cbk, _gf_false);
-        }
-
-        return 0;
-}
-
-int
-afr_self_heal_entry (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local = NULL;
-        afr_private_t   *priv = NULL;
+	locked_on = alloca0 (priv->child_count);
 
+	ret = afr_selfheal_tryentrylk (frame, this, inode, priv->sh_domain, NULL,
+				       locked_on);
+	{
+		if (ret < 2) {
+			/* 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;
+		}
 
-        priv = this->private;
-        local = frame->local;
+		ret = __afr_selfheal_entry (frame, this, fd, locked_on);
+	}
+unlock:
+	afr_selfheal_unentrylk (frame, this, inode, priv->sh_domain, NULL, locked_on);
 
-        if (local->self_heal.need_entry_self_heal && priv->entry_self_heal) {
-                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);
-        }
+	if (fd)
+		fd_unref (fd);
 
-        return 0;
+	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 04c5ef4e1..83628297f 100644
--- a/xlators/cluster/afr/src/afr-self-heal-metadata.c
+++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c
@@ -1,697 +1,280 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#include <libgen.h>
-#include <unistd.h>
-#include <fnmatch.h>
-#include <sys/time.h>
-#include <stdlib.h>
-#include <signal.h>
 
 #ifndef _CONFIG_H
 #define _CONFIG_H
 #include "config.h"
 #endif
 
-#include "glusterfs.h"
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-
-#include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-
-
-int
-afr_sh_metadata_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              i = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-//      memset (sh->child_errno, 0, sizeof (int) * priv->child_count);
-        memset (sh->buf, 0, sizeof (struct iatt) * priv->child_count);
-        memset (sh->success, 0, sizeof (int) * priv->child_count);
-
-/*         for (i = 0; i < priv->child_count; i++) { */
-/*                 sh->locked_nodes[i] = 1; */
-/*         } */
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->xattr[i])
-                        dict_unref (sh->xattr[i]);
-                sh->xattr[i] = NULL;
-        }
-
-        if (local->govinda_gOvinda) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "split-brain detected, aborting selfheal of %s",
-                        local->loc.path);
-                sh->op_failed = 1;
-                sh->completion_cbk (frame, this);
-        } else {
-                if (IA_ISREG (sh->type)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "proceeding to data check on %s",
-                                local->loc.path);
-                        afr_self_heal_data (frame, this);
-                        return 0;
-                }
-
-                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);
-                        return 0;
-                }
-                sh->completion_cbk (frame, this);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_metadata_unlck_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno)
-{
-        int               call_count = 0;
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0)
-                afr_sh_metadata_done (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_erase_pending_cbk (call_frame_t *frame, void *cookie,
-                                   xlator_t *this, int32_t op_ret,
-                                   int32_t op_errno, dict_t *xattr)
-{
-        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_fresh_children_add_child (sh->fresh_children, i,
-                                              priv->child_count);
-        }
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if ((!IA_ISREG (sh->buf[sh->source].ia_type)) &&
-                    (!IA_ISDIR (sh->buf[sh->source].ia_type))) {
-                        afr_inode_set_read_ctx (this, sh->inode, sh->source,
-                                                sh->fresh_children);
-                }
-                afr_sh_metadata_finish (frame, this);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_metadata_erase_pending (call_frame_t *frame, xlator_t *this)
-{
-        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;
-
-
-        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,
-                                 AFR_METADATA_TRANSACTION);
-
-        erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count,
-                                 gf_afr_mt_dict_t);
-        if (!erase_xattr)
-                return -ENOMEM;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->xattr[i]) {
-                        call_count++;
-
-                        erase_xattr[i] = get_new_dict();
-                        dict_ref (erase_xattr[i]);
-                }
-        }
-
-        afr_sh_delta_to_xattr (priv, sh->delta_matrix, erase_xattr,
-                               priv->child_count, AFR_METADATA_TRANSACTION);
-
-        local->call_count = call_count;
-
-        if (call_count == 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "metadata of %s not healed on any subvolume",
-                        local->loc.path);
-
-                afr_sh_metadata_finish (frame, this);
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (!erase_xattr[i])
-                        continue;
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "erasing pending flags from %s on %s",
-                        local->loc.path, priv->children[i]->name);
-
-                STACK_WIND_COOKIE (frame, afr_sh_metadata_erase_pending_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->xattrop,
-                                   &local->loc,
-                                   GF_XATTROP_ADD_ARRAY, erase_xattr[i]);
-                if (!--call_count)
-                        break;
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (erase_xattr[i]) {
-                        dict_unref (erase_xattr[i]);
-                }
-        }
-        GF_FREE (erase_xattr);
-
-        return 0;
-}
-
-
-int
-afr_sh_metadata_sync_cbk (call_frame_t *frame, void *cookie, 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              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;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0)
-                afr_sh_metadata_erase_pending (frame, this);
-
-        return 0;
-}
-
-
-int
-afr_sh_metadata_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno,
-                             struct iatt *preop, struct iatt *postop)
-{
-        afr_sh_metadata_sync_cbk (frame, cookie, this, op_ret, op_errno);
-
-        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)
-{
-        afr_sh_metadata_sync_cbk (frame, cookie, this, op_ret, op_errno);
-
-        return 0;
-}
-
-
-int
-afr_sh_metadata_sync (call_frame_t *frame, xlator_t *this, dict_t *xattr)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              source = 0;
-        int              active_sinks = 0;
-        int              call_count = 0;
-        int              i = 0;
-
-        struct iatt      stbuf = {0,};
-        int32_t          valid = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        source = sh->source;
-        active_sinks = sh->active_sinks;
-
-        /*
-         * 2 calls per sink - setattr, setxattr
-         */
-        if (xattr)
-                call_count = active_sinks * 2;
-        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);
-
-                call_count--;
-
-                if (!xattr)
-                        continue;
-
-                STACK_WIND_COOKIE (frame, afr_sh_metadata_xattr_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->setxattr,
-                                   &local->loc, xattr, 0);
-                call_count--;
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_metadata_getxattr_cbk (call_frame_t *frame, void *cookie,
-                              xlator_t *this,
-                              int32_t op_ret, int32_t op_errno, dict_t *xattr)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              source = 0;
-
-        int i;
-
-        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 {
-                for (i = 0; i < priv->child_count; i++) {
-                        dict_del (xattr, priv->pending_key[i]);
-                }
-
-                afr_sh_metadata_sync (frame, this, xattr);
-        }
-
-        return 0;
-}
+#include "byte-order.h"
 
+#define AFR_HEAL_ATTR (GF_SET_ATTR_UID|GF_SET_ATTR_GID|GF_SET_ATTR_MODE)
 
 int
-afr_sh_metadata_sync_prepare (call_frame_t *frame, xlator_t *this)
+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              active_sinks = 0;
-        int              source = 0;
-        int              i = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        source = sh->source;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->sources[i] == 0 && local->child_up[i] == 1) {
-                        active_sinks++;
-                        sh->success[i] = 1;
-                }
-        }
-        sh->success[source] = 1;
-
-        if (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;
-        }
-        sh->active_sinks = active_sinks;
-
-        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, active_sinks);
-
-        STACK_WIND (frame, afr_sh_metadata_getxattr_cbk,
-                    priv->children[source],
-                    priv->children[source]->fops->getxattr,
-                    &local->loc, NULL);
-
-        return 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);
+	uuid_copy (loc.gfid, inode->gfid);
+
+	gf_log (this->name, GF_LOG_INFO, "performing metadata selfheal on %s",
+		uuid_utoa (inode->gfid));
+
+	ret = syncop_getxattr (priv->children[source], &loc, &xattr, NULL);
+	if (ret < 0) {
+		loc_wipe (&loc);
+		return -EIO;
+	}
+
+	afr_filter_xattrs (xattr);
+	dict_del (xattr, GF_SELINUX_XATTR_KEY);
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!healed_sinks[i])
+			continue;
+
+		ret = syncop_setattr (priv->children[i], &loc,
+				      &locked_replies[source].poststat,
+				      AFR_HEAL_ATTR, NULL, NULL);
+		if (ret)
+			healed_sinks[i] = 0;
+
+		old_xattr = NULL;
+		ret = syncop_getxattr (priv->children[i], &loc, &old_xattr, 0);
+		if (old_xattr) {
+			dict_del (old_xattr, GF_SELINUX_XATTR_KEY);
+			afr_filter_xattrs (old_xattr);
+			ret = syncop_removexattr (priv->children[i], &loc, "",
+						  old_xattr);
+		}
+
+		ret = syncop_setxattr (priv->children[i], &loc, xattr, 0);
+		if (ret)
+			healed_sinks[i] = 0;
+	}
+
+	loc_wipe (&loc);
+	if (xattr)
+		dict_unref (xattr);
+
+	return 0;
 }
 
 
-int
-afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this)
+/*
+ * Look for mismatching uid/gid or mode even if xattrs don't say so, and
+ * pick one arbitrarily as winner.
+ */
+
+static int
+__afr_selfheal_metadata_finalize_source (xlator_t *this, unsigned char *sources,
+					 unsigned char *sinks,
+					 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              nsources = 0;
-        int              source = 0;
-        int              i = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        nsources = afr_build_sources (this, sh->xattr, sh->buf,
-                                      sh->pending_matrix, sh->sources,
-                                      sh->success_children,
-                                      AFR_METADATA_TRANSACTION);
-        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);
-                return 0;
-        }
-
-        if ((nsources == -1)
-            && (priv->favorite_child != -1)
-            && (sh->child_errno[priv->favorite_child] == 0)) {
-
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Picking favorite child %s as authentic source to resolve conflicting metadata of %s",
-                        priv->children[priv->favorite_child]->name,
-                        local->loc.path);
-
-                sh->sources[priv->favorite_child] = 1;
-
-                nsources = afr_sh_source_count (sh->sources,
-                                                priv->child_count);
-        }
-
-        if (nsources == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Unable to self-heal permissions/ownership of '%s' "
-                        "(possible split-brain). Please fix the file on "
-                        "all backend volumes", local->loc.path);
-
-                local->govinda_gOvinda = 1;
-
-                afr_sh_metadata_finish (frame, this);
-                return 0;
-        }
-
-        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);
-                return 0;
-        }
-
-        sh->source = source;
-
-        /* detect changes not visible through pending flags -- JIC */
-        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 ((!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);
-        }
-
-        afr_sh_metadata_sync_prepare (frame, this);
-
-        return 0;
+	int i = 0;
+	afr_private_t *priv = NULL;
+	struct iatt first = {0, };
+	int source = -1;
+	int locked_count = 0;
+	int sources_count = 0;
+	int sinks_count = 0;
+
+	priv = this->private;
+
+	locked_count = AFR_COUNT (locked_on, priv->child_count);
+	sources_count = AFR_COUNT (sources, priv->child_count);
+	sinks_count = AFR_COUNT (sinks, priv->child_count);
+
+	if (locked_count == sinks_count || !sources_count) {
+		if (!priv->metadata_splitbrain_forced_heal) {
+			return -EIO;
+		}
+		/* Metadata split brain, select one subvol
+		   arbitrarily */
+		for (i = 0; i < priv->child_count; i++) {
+			if (locked_on[i] && sinks[i]) {
+				sources[i] = 1;
+				sinks[i] = 0;
+				break;
+			}
+		}
+	}
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!sources[i])
+			continue;
+		if (source == -1) {
+			source = i;
+			first = replies[i].poststat;
+		}
+	}
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!sources[i])
+			continue;
+		if (!IA_EQUAL (first, replies[i].poststat, type) ||
+		    !IA_EQUAL (first, replies[i].poststat, uid) ||
+		    !IA_EQUAL (first, replies[i].poststat, gid) ||
+		    !IA_EQUAL (first, replies[i].poststat, prot)) {
+			sources[i] = 0;
+			sinks[i] = 1;
+		}
+	}
+
+	return source;
 }
 
 
-int
-afr_sh_metadata_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)
+static 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,
+				 struct afr_reply *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 == 0) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "path %s on subvolume %s is of mode 0%o",
-                                local->loc.path,
-                                priv->children[child_index]->name,
-                                buf->ia_type);
-
-                        sh->buf[child_index] = *buf;
-                        if (xattr)
-                                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_INFO,
-                                "path %s on subvolume %s => -1 (%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)
-                afr_sh_metadata_fix (frame, this);
-
-        return 0;
+	int ret = -1;
+	int source = -1;
+	afr_private_t *priv = NULL;
+	int i = 0;
+
+	priv = this->private;
+
+	ret = afr_selfheal_unlocked_discover (frame, inode, inode->gfid,
+					      replies);
+	if (ret)
+		return ret;
+
+	ret = afr_selfheal_find_direction (frame, this, replies,
+					   AFR_METADATA_TRANSACTION,
+					   locked_on, sources, sinks);
+	if (ret)
+		return ret;
+
+	source = __afr_selfheal_metadata_finalize_source (this, sources, sinks,
+							  locked_on, replies);
+	if (source < 0)
+		return -EIO;
+
+	for (i = 0; i < priv->child_count; i++)
+		/* 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.
+		*/
+		healed_sinks[i] = sinks[i] && locked_on[i];
+
+	return source;
 }
 
-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_ERROR, "Non Blocking metadata "
-                        "inodelks failed for %s.", local->loc.path);
-                gf_log (this->name, GF_LOG_ERROR, "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_lookup_cbk, _gf_false);
-        }
-
-        return 0;
-}
 
-int
-afr_sh_metadata_lock (call_frame_t *frame, xlator_t *this)
+static int
+__afr_selfheal_metadata (call_frame_t *frame, xlator_t *this, inode_t *inode,
+			 unsigned char *locked_on)
 {
-        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);
-
-        int_lock->lk_flock.l_start = 0;
-        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;
-
-        afr_nonblocking_inodelk (frame, this);
-
-        return 0;
+	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;
+	struct afr_reply *locked_replies = NULL;
+	int source = -1;
+
+	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);
+
+	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 < 2) {
+			ret = -ENOTCONN;
+			goto unlock;
+		}
+
+		ret = __afr_selfheal_metadata_prepare (frame, this, inode, data_lock,
+						       sources, sinks, healed_sinks,
+						       locked_replies);
+		if (ret < 0)
+			goto unlock;
+
+		source = ret;
+		ret = 0;
+	}
+unlock:
+	afr_selfheal_uninodelk (frame, this, inode, this->name,
+				LLONG_MAX -1, 0, data_lock);
+	if (ret < 0)
+		goto out;
+
+	ret = afr_selfheal_metadata_do (frame, this, inode, source, healed_sinks,
+					locked_replies);
+	if (ret)
+		goto out;
+
+	ret = afr_selfheal_undo_pending (frame, this, inode, sources, sinks,
+					 healed_sinks, AFR_METADATA_TRANSACTION,
+					 locked_replies, data_lock);
+out:
+	return ret;
 }
 
 
 int
-afr_self_heal_metadata (call_frame_t *frame, xlator_t *this)
+afr_selfheal_metadata (call_frame_t *frame, xlator_t *this, inode_t *inode)
 {
-        afr_local_t   *local = NULL;
-        afr_private_t *priv = this->private;
-
-
-        local = frame->local;
-
-        if (local->self_heal.need_metadata_self_heal && priv->metadata_self_heal) {
-                afr_sh_metadata_lock (frame, this);
-        } else {
-                afr_sh_metadata_done (frame, this);
-        }
-
-        return 0;
+	afr_private_t *priv = NULL;
+	unsigned char *locked_on = NULL;
+	int ret = 0;
+
+	priv = this->private;
+
+	locked_on = alloca0 (priv->child_count);
+
+	ret = afr_selfheal_tryinodelk (frame, this, inode, priv->sh_domain, 0, 0,
+				       locked_on);
+	{
+		if (ret < 2) {
+			/* 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_metadata (frame, this, inode, locked_on);
+	}
+unlock:
+	afr_selfheal_uninodelk (frame, this, inode, priv->sh_domain, 0, 0, locked_on);
+
+	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 000000000..ce80b8da3
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heal-name.c
@@ -0,0 +1,457 @@
+/*
+  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.
+*/
+
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "afr.h"
+#include "afr-self-heal.h"
+
+
+int
+__afr_selfheal_assign_gfid (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;
+	dict_t *xdata = NULL;
+	int ret = 0;
+	loc_t loc = {0, };
+
+	priv = this->private;
+
+	uuid_copy (parent->gfid, pargfid);
+
+	xdata = dict_new ();
+	if (!xdata) {
+		return -ENOMEM;
+	}
+
+	ret = dict_set_static_bin (xdata, "gfid-req",
+				   replies[gfid_idx].poststat.ia_gfid, 16);
+	if (ret) {
+		dict_destroy (xdata);
+		return -ENOMEM;
+	}
+
+	loc.parent = inode_ref (parent);
+	loc.inode = inode_ref (inode);
+	uuid_copy (loc.pargfid, pargfid);
+	loc.name = bname;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (replies[i].op_ret == 0 || replies[i].op_errno != ENODATA)
+			continue;
+
+		ret = syncop_lookup (priv->children[i], &loc, xdata, 0, 0, 0);
+	}
+
+	loc_wipe (&loc);
+	dict_unref (xdata);
+
+	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;
+
+	priv = this->private;
+
+	uuid_copy (parent->gfid, pargfid);
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].valid)
+			continue;
+
+		if (uuid_compare (replies[i].poststat.ia_gfid,
+				  replies[gfid_idx].poststat.ia_gfid) == 0)
+			continue;
+
+		ret |= afr_selfheal_recreate_entry (frame, this, i, gfid_idx,
+						    parent, bname, inode, replies);
+	}
+
+	return ret;
+}
+
+
+int
+__afr_selfheal_name_expunge (call_frame_t *frame, xlator_t *this, inode_t *parent,
+			     uuid_t pargfid, const char *bname, inode_t *inode,
+			     struct afr_reply *replies)
+{
+	loc_t loc = {0, };
+	int i = 0;
+	afr_private_t *priv = NULL;
+	char g[64];
+	int ret = 0;
+
+	priv = this->private;
+
+	loc.parent = inode_ref (parent);
+	uuid_copy (loc.pargfid, pargfid);
+	loc.name = bname;
+	loc.inode = inode_ref (inode);
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].valid)
+			continue;
+
+		if (replies[i].op_ret)
+			continue;
+
+		switch (replies[i].poststat.ia_type) {
+		case IA_IFDIR:
+			gf_log (this->name, GF_LOG_WARNING,
+				"expunging dir %s/%s (%s) on %s",
+				uuid_utoa (pargfid), bname,
+				uuid_utoa_r (replies[i].poststat.ia_gfid, g),
+				priv->children[i]->name);
+			ret |= syncop_rmdir (priv->children[i], &loc, 1);
+			break;
+		default:
+			gf_log (this->name, GF_LOG_WARNING,
+				"expunging file %s/%s (%s) on %s",
+				uuid_utoa (pargfid), bname,
+				uuid_utoa_r (replies[i].poststat.ia_gfid, g),
+				priv->children[i]->name);
+			ret |= syncop_unlink (priv->children[i], &loc);
+			break;
+		}
+	}
+
+	loc_wipe (&loc);
+
+	return ret;
+
+}
+
+
+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)
+{
+	int i = 0;
+	afr_private_t *priv = NULL;
+	uuid_t gfid = {0, };
+	int gfid_idx = -1;
+	gf_boolean_t source_is_empty = _gf_true;
+	gf_boolean_t need_heal = _gf_false;
+	int first_idx = -1;
+	char g1[64],g2[64];
+
+	priv = this->private;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].valid)
+			continue;
+
+		if (first_idx == -1) {
+			first_idx = i;
+			continue;
+		}
+
+		if (replies[i].op_ret != replies[first_idx].op_ret)
+			need_heal = _gf_true;
+
+		if (uuid_compare (replies[i].poststat.ia_gfid,
+				  replies[first_idx].poststat.ia_gfid))
+			need_heal = _gf_true;
+	}
+
+	if (!need_heal)
+		return 0;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].valid)
+			continue;
+
+		if (!replies[i].op_ret && (source == -1 || sources[i])) {
+			source_is_empty = _gf_false;
+			break;
+		}
+	}
+
+	if (source_is_empty) {
+		return __afr_selfheal_name_expunge (frame, this, parent, pargfid,
+						    bname, inode, replies);
+	}
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].valid)
+			continue;
+
+		if (uuid_is_null (replies[i].poststat.ia_gfid))
+			continue;
+
+		if (uuid_is_null (gfid)) {
+			uuid_copy (gfid, replies[i].poststat.ia_gfid);
+			gfid_idx = i;
+			continue;
+		}
+
+		if (sources[i] || source == -1) {
+			if (gfid_idx != -1 &&
+			    (sources[gfid_idx] || source == -1) &&
+			    uuid_compare (gfid, replies[i].poststat.ia_gfid)) {
+				gf_log (this->name, GF_LOG_WARNING,
+					"GFID mismatch for <gfid:%s>/%s "
+					"%s on %s and %s on %s",
+					uuid_utoa (pargfid), bname,
+					uuid_utoa_r (replies[i].poststat.ia_gfid, g1),
+					priv->children[i]->name,
+					uuid_utoa_r (replies[gfid_idx].poststat.ia_gfid, g2),
+					priv->children[gfid_idx]->name);
+				return -1;
+			}
+
+			uuid_copy (gfid, replies[i].poststat.ia_gfid);
+			gfid_idx = i;
+			continue;
+		}
+	}
+
+	if (gfid_idx == -1)
+		return -1;
+
+	__afr_selfheal_assign_gfid (frame, this, parent, pargfid, bname, inode,
+				    replies, gfid_idx);
+
+	return __afr_selfheal_name_impunge (frame, this, parent, pargfid,
+					    bname, inode, replies, gfid_idx);
+}
+
+
+int
+__afr_selfheal_name_finalize_source (xlator_t *this, unsigned char *sources,
+				     unsigned char *sinks, unsigned char *locked_on,
+				     struct afr_reply *replies)
+{
+	int i = 0;
+	afr_private_t *priv = NULL;
+	int source = -1;
+	int locked_count = 0;
+	int sources_count = 0;
+	int sinks_count = 0;
+
+	priv = this->private;
+
+	locked_count = AFR_COUNT (locked_on, priv->child_count);
+	sources_count = AFR_COUNT (sources, priv->child_count);
+	sinks_count = AFR_COUNT (sinks, priv->child_count);
+
+	if (locked_count == sinks_count || !sources_count) {
+		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, struct afr_reply *replies,
+			     int *source_p)
+{
+	int ret = -1;
+	int source = -1;
+	afr_private_t *priv = NULL;
+	int i = 0;
+
+	priv = this->private;
+
+	ret = afr_selfheal_unlocked_discover (frame, parent, pargfid, replies);
+	if (ret)
+		return ret;
+
+	ret = afr_selfheal_find_direction (frame, this, replies,
+					   AFR_ENTRY_TRANSACTION,
+					   locked_on, sources, sinks);
+	if (ret)
+		return ret;
+
+	source = __afr_selfheal_name_finalize_source (this, sources, sinks,
+						      locked_on, replies);
+	if (source < 0) {
+		/* If source is < 0 (typically split-brain), we perform a
+		   conservative merge of entries rather than erroring out */
+	}
+	*source_p = source;
+
+	for (i = 0; i < priv->child_count; i++)
+		/* 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.
+		*/
+		healed_sinks[i] = sinks[i] && locked_on[i];
+
+	return ret;
+}
+
+
+int
+afr_selfheal_name_do (call_frame_t *frame, xlator_t *this, inode_t *parent,
+		      uuid_t pargfid, const char *bname)
+{
+	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;
+
+	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 < 2) {
+			ret = -ENOTCONN;
+			goto unlock;
+		}
+
+		ret = __afr_selfheal_name_prepare (frame, this, parent, pargfid,
+						   locked_on, sources, sinks,
+						   healed_sinks, replies,
+						   &source);
+		if (ret)
+			goto unlock;
+
+		inode = afr_selfheal_unlocked_lookup_on (frame, parent, bname,
+							 replies, locked_on);
+		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);
+	}
+unlock:
+	afr_selfheal_unentrylk (frame, this, parent, this->name, bname,
+				locked_on);
+	if (inode)
+		inode_unref (inode);
+
+	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;
+
+	priv = this->private;
+
+	replies = alloca0 (sizeof (*replies) * priv->child_count);
+
+	inode = afr_selfheal_unlocked_lookup_on (frame, parent, bname,
+						 replies, priv->child_up);
+	if (!inode)
+		return -ENOMEM;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (!replies[i].valid)
+			continue;
+
+		if (first_idx == -1) {
+			first_idx = i;
+			continue;
+		}
+
+		if (replies[i].op_ret != replies[first_idx].op_ret)
+			*need_heal = _gf_true;
+
+		if (uuid_compare (replies[i].poststat.ia_gfid,
+				  replies[first_idx].poststat.ia_gfid))
+			*need_heal = _gf_true;
+	}
+
+	if (inode)
+		inode_unref (inode);
+	return 0;
+}
+
+int
+afr_selfheal_name (xlator_t *this, uuid_t pargfid, const char *bname)
+{
+	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);
+	if (!frame)
+		goto out;
+
+	ret = afr_selfheal_name_unlocked_inspect (frame, this, parent, pargfid,
+						  bname, &need_heal);
+	if (ret)
+		goto out;
+
+	if (need_heal)
+		afr_selfheal_name_do (frame, this, parent, pargfid, bname);
+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 1056a3662..a1b972ac3 100644
--- a/xlators/cluster/afr/src/afr-self-heal.h
+++ b/xlators/cluster/afr/src/afr-self-heal.h
@@ -1,58 +1,167 @@
 /*
-   Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  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.
 */
 
-#ifndef __AFR_SELF_HEAL_H__
-#define __AFR_SELF_HEAL_H__
 
-#include <sys/stat.h>
+#ifndef _AFR_SELFHEAL_H
+#define _AFR_SELFHEAL_H
+
+
+/* Perform fop on all UP subvolumes and wait for all callbacks to return */
+
+#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;					\
+									\
+	afr_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); \
+		__count++;						\
+	}								\
+	syncbarrier_wait (&__local->barrier, __count);			\
+	} while (0)
+
+
+/* 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, __count = 0;					\
+									\
+	afr_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); \
+		__count++;						\
+	}								\
+	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_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 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 SIZE_GREATER(buf1,buf2) ((buf1)->ia_size > (buf2)->ia_size)
+#define ALLOC_MATRIX(n, type) ({type **__ptr = NULL; \
+	int __i; \
+	__ptr = alloca0 (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)
+
 
 int
-afr_sh_has_metadata_pending (dict_t *xattr, xlator_t *this);
+afr_selfheal (xlator_t *this, uuid_t gfid);
+
 int
-afr_sh_has_entry_pending (dict_t *xattr, xlator_t *this);
+afr_selfheal_name (xlator_t *this, uuid_t gfid, const char *name);
+
 int
-afr_sh_has_data_pending (dict_t *xattr, xlator_t *this);
+afr_selfheal_data (call_frame_t *frame, xlator_t *this, inode_t *inode);
 
 int
-afr_self_heal_entry (call_frame_t *frame, xlator_t *this);
+afr_selfheal_metadata (call_frame_t *frame, xlator_t *this, inode_t *inode);
 
 int
-afr_self_heal_data (call_frame_t *frame, xlator_t *this);
+afr_selfheal_entry (call_frame_t *frame, xlator_t *this, inode_t *inode);
+
+
+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_self_heal_metadata (call_frame_t *frame, xlator_t *this);
+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_self_heal_get_source (xlator_t *this, afr_local_t *local, dict_t **xattr);
+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_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode);
+afr_selfheal_entrylk (call_frame_t *frame, xlator_t *this, inode_t *inode,
+		      char *dom, const char *name, unsigned char *locked_on);
 
 int
-afr_lookup_select_read_child_by_txn_type (xlator_t *this, afr_local_t *local,
-                                          dict_t **xattr,
-                                          afr_transaction_type txn_type);
-#endif /* __AFR_SELF_HEAL_H__ */
+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_unentrylk (call_frame_t *frame, xlator_t *this, inode_t *inode,
+			char *dom, const char *name, unsigned char *locked_on);
+
+int
+afr_selfheal_unlocked_discover (call_frame_t *frame, inode_t *inode,
+				uuid_t gfid, struct afr_reply *replies);
+
+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);
+
+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);
+
+int
+afr_selfheal_extract_xattr (xlator_t *this, struct afr_reply *replies,
+			    afr_transaction_type type, int *dirty, int **matrix);
+
+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, afr_transaction_type type,
+			   struct afr_reply *replies, unsigned char *locked_on);
+
+int
+afr_selfheal_recreate_entry (call_frame_t *frame, xlator_t *this, int dst,
+			     int source, 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);
+
+call_frame_t *
+afr_frame_create (xlator_t *this);
+
+inode_t *
+afr_inode_find (xlator_t *this, uuid_t gfid);
+
+#endif /* !_AFR_SELFHEAL_H */
diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c
new file mode 100644
index 000000000..4bfe909bc
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heald.c
@@ -0,0 +1,1256 @@
+/*
+  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.
+*/
+
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "afr.h"
+#include "afr-self-heal.h"
+#include "afr-self-heald.h"
+#include "protocol-common.h"
+
+#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
+#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])
+
+int afr_shd_gfid_to_path (xlator_t *this, xlator_t *subvol, uuid_t gfid, char **path_p);
+
+char *
+afr_subvol_name (xlator_t *this, int subvol)
+{
+	afr_private_t *priv = NULL;
+
+	priv = this->private;
+	if (subvol < 0 || subvol > priv->child_count)
+		return NULL;
+
+	return priv->children[subvol]->name;
+}
+
+
+void
+afr_destroy_crawl_event_data (void *data)
+{
+        return;
+}
+
+
+void
+afr_destroy_shd_event_data (void *data)
+{
+	shd_event_t *shd_event = data;
+
+	if (!shd_event)
+		return;
+	GF_FREE (shd_event->path);
+
+        return;
+}
+
+
+gf_boolean_t
+afr_shd_is_subvol_local (xlator_t *this, int subvol)
+{
+	char *pathinfo = NULL;
+	afr_private_t *priv = NULL;
+	dict_t *xattr = NULL;
+	int ret = 0;
+	gf_boolean_t is_local = _gf_false;
+	loc_t loc = {0, };
+
+	priv = this->private;
+
+	loc.inode = this->itable->root;
+	uuid_copy (loc.gfid, loc.inode->gfid);
+
+	ret = syncop_getxattr (priv->children[subvol], &loc, &xattr,
+			       GF_XATTR_PATHINFO_KEY);
+	if (ret)
+		return _gf_false;
+	if (!xattr)
+		return _gf_false;
+
+	ret = dict_get_str (xattr, GF_XATTR_PATHINFO_KEY, &pathinfo);
+	if (ret)
+		return _gf_false;
+
+	afr_local_pathinfo (pathinfo, &is_local);
+
+	gf_log (this->name, GF_LOG_DEBUG, "subvol %s is %slocal",
+		priv->children[subvol]->name, is_local? "" : "not ");
+
+	return is_local;
+}
+
+
+int
+__afr_shd_healer_wait (struct subvol_healer *healer)
+{
+	afr_private_t *priv = NULL;
+	struct timespec wait_till = {0, };
+	int ret = 0;
+
+	priv = healer->this->private;
+
+disabled_loop:
+	wait_till.tv_sec = time (NULL) + 60;
+
+	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
+afr_shd_healer_wait (struct subvol_healer *healer)
+{
+	int ret = 0;
+
+	pthread_mutex_lock (&healer->mutex);
+	{
+		ret = __afr_shd_healer_wait (healer);
+	}
+	pthread_mutex_unlock (&healer->mutex);
+
+	return ret;
+}
+
+
+gf_boolean_t
+safe_break (struct subvol_healer *healer)
+{
+	gf_boolean_t ret = _gf_false;
+
+	pthread_mutex_lock (&healer->mutex);
+	{
+		if (healer->rerun)
+			goto unlock;
+
+		healer->running = _gf_false;
+		ret = _gf_true;
+	}
+unlock:
+	pthread_mutex_unlock (&healer->mutex);
+
+	return ret;
+}
+
+
+inode_t *
+afr_shd_inode_find (xlator_t *this, xlator_t *subvol, uuid_t gfid)
+{
+	inode_t *inode = NULL;
+	int ret = 0;
+	loc_t loc = {0, };
+	struct iatt iatt = {0, };
+
+	inode = inode_find (this->itable, gfid);
+	if (inode)
+		goto out;
+
+	loc.inode = inode_new (this->itable);
+	if (!loc.inode)
+		goto out;
+	uuid_copy (loc.gfid, gfid);
+
+	ret = syncop_lookup (subvol, &loc, NULL, &iatt, NULL, NULL);
+	if (ret < 0)
+		goto out;
+
+	inode = inode_link (loc.inode, NULL, NULL, &iatt);
+	if (inode)
+		inode_lookup (inode);
+out:
+	loc_wipe (&loc);
+	return inode;
+}
+
+
+fd_t *
+afr_shd_index_opendir (xlator_t *this, int child)
+{
+	fd_t *fd = NULL;
+	afr_private_t *priv = NULL;
+	xlator_t *subvol = NULL;
+	loc_t rootloc = {0, };
+	inode_t *inode = NULL;
+	int ret = 0;
+	dict_t *xattr = NULL;
+	void *index_gfid = NULL;
+
+	priv = this->private;
+	subvol = priv->children[child];
+
+	rootloc.inode = inode_ref (this->itable->root);
+	uuid_copy (rootloc.gfid, rootloc.inode->gfid);
+
+	ret = syncop_getxattr (subvol, &rootloc, &xattr,
+			       GF_XATTROP_INDEX_GFID);
+	if (ret || !xattr) {
+		errno = -ret;
+		goto out;
+	}
+
+	ret = dict_get_ptr (xattr, GF_XATTROP_INDEX_GFID, &index_gfid);
+	if (ret)
+		goto out;
+
+	gf_log (this->name, GF_LOG_DEBUG, "index-dir gfid for %s: %s",
+		subvol->name, uuid_utoa (index_gfid));
+
+	inode = afr_shd_inode_find (this, subvol, index_gfid);
+	if (!inode)
+		goto out;
+	fd = fd_anonymous (inode);
+out:
+	loc_wipe (&rootloc);
+	if (xattr)
+		dict_unref (xattr);
+	return fd;
+}
+
+
+int
+afr_shd_index_purge (xlator_t *subvol, inode_t *inode, char *name)
+{
+	loc_t loc = {0, };
+	int ret = 0;
+
+	loc.parent = inode_ref (inode);
+	loc.name = name;
+
+	ret = syncop_unlink (subvol, &loc);
+
+	loc_wipe (&loc);
+	return ret;
+}
+
+
+int
+afr_shd_selfheal_name (struct subvol_healer *healer, int child, uuid_t parent,
+		       const char *bname)
+{
+	int ret = -1;
+
+	ret = afr_selfheal_name (THIS, parent, bname);
+
+	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];
+
+	ret = afr_selfheal (this, gfid);
+
+	if (ret == -EIO) {
+		eh = shd->split_brain;
+		crawl_event->split_brain_count++;
+	} else if (ret < 0) {
+		eh = shd->heal_failed;
+		crawl_event->heal_failed_count++;
+	} else if (ret == 0) {
+		eh = shd->healed;
+		crawl_event->healed_count++;
+	}
+
+	afr_shd_gfid_to_path (this, subvol, gfid, &path);
+	if (!path)
+		return ret;
+
+	if (eh) {
+		shd_event = GF_CALLOC (1, sizeof(*shd_event),
+				       gf_afr_mt_shd_event_t);
+		if (!shd_event) {
+			GF_FREE (path);
+			return ret;
+		}
+
+		shd_event->child = child;
+		shd_event->path = path;
+
+		if (eh_save_history (eh, shd_event) < 0) {
+			GF_FREE (shd_event);
+			GF_FREE (path);
+		}
+	}
+	return ret;
+}
+
+
+void
+afr_shd_sweep_prepare (struct subvol_healer *healer)
+{
+	crawl_event_t *event = NULL;
+
+	event = &healer->crawl_event;
+
+	event->healed_count = 0;
+	event->split_brain_count = 0;
+	event->heal_failed_count = 0;
+
+	time (&event->start_time);
+	event->end_time = 0;
+}
+
+
+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;
+
+	event = &healer->crawl_event;
+	shd = &(((afr_private_t *)healer->this->private)->shd);
+
+	time (&event->end_time);
+	history = memdup (event, sizeof (*event));
+	event->start_time = 0;
+
+	if (!history)
+		return;
+
+	if (eh_save_history (shd->statistics[healer->subvol], history) < 0)
+		GF_FREE (history);
+}
+
+
+int
+afr_shd_index_sweep (struct subvol_healer *healer)
+{
+	xlator_t *this = NULL;
+	int child = -1;
+	fd_t *fd = NULL;
+	xlator_t *subvol = NULL;
+	afr_private_t *priv = NULL;
+	off_t offset = 0;
+	gf_dirent_t entries;
+	gf_dirent_t *entry = NULL;
+	uuid_t gfid;
+	int ret = 0;
+	int count = 0;
+
+	this = healer->this;
+	child = healer->subvol;
+	priv = this->private;
+	subvol = priv->children[child];
+
+	fd = afr_shd_index_opendir (this, child);
+	if (!fd) {
+		gf_log (this->name, GF_LOG_WARNING,
+			"unable to opendir index-dir on %s", subvol->name);
+		return -errno;
+	}
+
+	INIT_LIST_HEAD (&entries.list);
+
+	while ((ret = syncop_readdir (subvol, fd, 131072, offset, &entries))) {
+		if (ret > 0)
+			ret = 0;
+		list_for_each_entry (entry, &entries.list, list) {
+			offset = entry->d_off;
+
+			if (!priv->shd.enabled) {
+				ret = -EBUSY;
+				break;
+			}
+
+			if (!strcmp (entry->d_name, ".") ||
+			    !strcmp (entry->d_name, ".."))
+				continue;
+
+			gf_log (this->name, GF_LOG_DEBUG, "got entry: %s",
+				entry->d_name);
+
+			ret = uuid_parse (entry->d_name, gfid);
+			if (ret)
+				continue;
+
+			ret = afr_shd_selfheal (healer, child, gfid);
+			if (ret == 0)
+				count++;
+
+			if (ret == -ENOENT || ret == -ESTALE) {
+				afr_shd_index_purge (subvol, fd->inode,
+						     entry->d_name);
+				ret = 0;
+			}
+		}
+
+		gf_dirent_free (&entries);
+		if (ret)
+			break;
+	}
+
+	if (fd)
+		fd_unref (fd);
+	if (!ret)
+		ret = count;
+	return ret;
+}
+
+
+int
+afr_shd_full_sweep (struct subvol_healer *healer, inode_t *inode)
+{
+	fd_t *fd = NULL;
+	xlator_t *this = NULL;
+	xlator_t *subvol = NULL;
+	afr_private_t *priv = NULL;
+	off_t offset = 0;
+	gf_dirent_t entries;
+	gf_dirent_t *entry = NULL;
+	int ret = 0;
+
+	this = healer->this;
+	priv = this->private;
+	subvol = priv->children[healer->subvol];
+
+	fd = fd_anonymous (inode);
+	if (!fd)
+		return -errno;
+
+	INIT_LIST_HEAD (&entries.list);
+
+	while ((ret = syncop_readdirp (subvol, fd, 131072, offset, 0, &entries))) {
+		if (ret < 0)
+			break;
+
+		ret = gf_link_inodes_from_dirent (this, fd->inode, &entries);
+		if (ret)
+			break;
+
+		list_for_each_entry (entry, &entries.list, list) {
+			offset = entry->d_off;
+
+			if (!priv->shd.enabled) {
+				ret = -EBUSY;
+				break;
+			}
+
+			if (!strcmp (entry->d_name, ".") ||
+			    !strcmp (entry->d_name, ".."))
+				continue;
+
+			afr_shd_selfheal_name (healer, healer->subvol,
+					       inode->gfid, entry->d_name);
+
+			afr_shd_selfheal (healer, healer->subvol,
+					  entry->d_stat.ia_gfid);
+
+			if (entry->d_stat.ia_type == IA_IFDIR) {
+				ret = afr_shd_full_sweep (healer, entry->inode);
+				if (ret)
+					break;
+			}
+		}
+
+		gf_dirent_free (&entries);
+		if (ret)
+			break;
+	}
+
+	if (fd)
+		fd_unref (fd);
+	return ret;
+}
+
+
+void *
+afr_shd_index_healer (void *data)
+{
+	struct subvol_healer *healer = NULL;
+	xlator_t *this = NULL;
+	int ret = 0;
+
+	healer = data;
+	THIS = this = healer->this;
+
+	for (;;) {
+		afr_shd_healer_wait (healer);
+
+		ASSERT_LOCAL(this, healer);
+
+		do {
+			gf_log (this->name, GF_LOG_DEBUG,
+				"starting index sweep on subvol %s",
+				afr_subvol_name (this, healer->subvol));
+
+			afr_shd_sweep_prepare (healer);
+
+			ret = afr_shd_index_sweep (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_log (this->name, GF_LOG_DEBUG,
+				"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);
+	}
+
+	return NULL;
+}
+
+
+void *
+afr_shd_full_healer (void *data)
+{
+	struct subvol_healer *healer = NULL;
+	xlator_t *this = NULL;
+	int run = 0;
+
+	healer = data;
+	THIS = this = healer->this;
+
+	for (;;) {
+		pthread_mutex_lock (&healer->mutex);
+		{
+			run = __afr_shd_healer_wait (healer);
+			if (!run)
+				healer->running = _gf_false;
+		}
+		pthread_mutex_unlock (&healer->mutex);
+
+		if (!run)
+			break;
+
+		ASSERT_LOCAL(this, healer);
+
+		gf_log (this->name, GF_LOG_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_log (this->name, GF_LOG_INFO,
+			"finished full sweep on subvol %s",
+			afr_subvol_name (this, healer->subvol));
+	}
+
+	return NULL;
+}
+
+
+int
+afr_shd_healer_init (xlator_t *this, struct subvol_healer *healer)
+{
+	int ret = 0;
+
+	ret = pthread_mutex_init (&healer->mutex, NULL);
+	if (ret)
+		goto out;
+
+	ret = pthread_cond_init (&healer->cond, NULL);
+	if (ret)
+		goto out;
+
+	healer->this = this;
+	healer->running = _gf_false;
+	healer->rerun = _gf_false;
+	healer->local = _gf_false;
+out:
+	return ret;
+}
+
+
+int
+afr_shd_healer_spawn (xlator_t *this, struct subvol_healer *healer,
+		      void *(threadfn)(void *))
+{
+	int ret = 0;
+
+	pthread_mutex_lock (&healer->mutex);
+	{
+		if (healer->running) {
+			pthread_cond_signal (&healer->cond);
+		} else {
+			ret = gf_thread_create (&healer->thread, NULL,
+						threadfn, healer);
+			if (ret)
+				goto unlock;
+			healer->running = 1;
+		}
+
+		healer->rerun = 1;
+	}
+unlock:
+	pthread_mutex_unlock (&healer->mutex);
+
+	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);
+}
+
+
+int
+afr_shd_index_healer_spawn (xlator_t *this, int subvol)
+{
+	return afr_shd_healer_spawn (this, NTH_INDEX_HEALER (this, subvol),
+				     afr_shd_index_healer);
+}
+
+
+int
+afr_shd_dict_add_crawl_event (xlator_t *this, dict_t *output,
+			      crawl_event_t *crawl_event)
+{
+        int             ret = 0;
+        uint64_t        count = 0;
+        char            key[256] = {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_log (this->name, GF_LOG_ERROR, "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 (key, sizeof (key), "statistics_healed_cnt-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        ret = dict_set_uint64(output, key, healed_count);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+			"Could not add statistics_healed_count to outout");
+                goto out;
+	}
+
+        snprintf (key, sizeof (key), "statistics_sb_cnt-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        ret = dict_set_uint64 (output, key, split_brain_count);
+	if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+			"Could not add statistics_split_brain_count to outout");
+                goto out;
+        }
+
+        snprintf (key, sizeof (key), "statistics_crawl_type-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        ret = dict_set_str (output, key, crawl_type);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+			"Could not add statistics_crawl_type to output");
+                goto out;
+        }
+
+        snprintf (key, sizeof (key), "statistics_heal_failed_cnt-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        ret = dict_set_uint64 (output, key, heal_failed_count);
+	if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+			"Could not add statistics_healed_failed_count to outout");
+                goto out;
+        }
+
+        snprintf (key, sizeof (key), "statistics_strt_time-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        ret = dict_set_dynstr (output, key, start_time_str);
+	if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+			"Could not add statistics_crawl_start_time to outout");
+                goto out;
+        } else {
+		start_time_str = NULL;
+	}
+
+	if (!end_time_str)
+                progress = 1;
+        else
+                progress = 0;
+
+        snprintf (key, sizeof (key), "statistics_end_time-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        if (!end_time_str)
+                end_time_str = gf_strdup ("Could not determine the end time");
+        ret = dict_set_dynstr (output, key, end_time_str);
+	if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+			"Could not add statistics_crawl_end_time to outout");
+                goto out;
+        } else {
+		end_time_str = NULL;
+	}
+
+        snprintf (key, sizeof (key), "statistics_inprogress-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+
+        ret = dict_set_int32 (output, key, progress);
+	if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+			"Could not add statistics_inprogress to outout");
+                goto out;
+        }
+
+	snprintf (key, sizeof (key), "statistics-%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 the counter.");
+                goto out;
+	}
+out:
+	GF_FREE (start_time_str);
+	GF_FREE (end_time_str);
+        return ret;
+}
+
+
+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[256] = {0};
+        int             xl_id = 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;
+        }
+
+        snprintf (key, sizeof (key), "%d-%d-count", xl_id, child);
+        ret = dict_get_uint64 (output, key, &count);
+
+        snprintf (key, sizeof (key), "%d-%d-%"PRIu64, xl_id, child, count);
+	ret = dict_set_dynstr (output, key, path);
+
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "%s: Could not add to output",
+                        path);
+                goto out;
+        }
+
+	if (tv) {
+		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;
+		}
+	}
+
+        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;
+}
+
+
+int
+afr_shd_gfid_to_path (xlator_t *this, xlator_t *subvol, uuid_t gfid, char **path_p)
+{
+	loc_t loc = {0,};
+	char *path = NULL;
+	dict_t *xattr = NULL;
+	int ret = 0;
+
+	uuid_copy (loc.gfid, gfid);
+	loc.inode = inode_new (this->itable);
+
+	ret = syncop_getxattr (subvol, &loc, &xattr, GFID_TO_PATH_KEY);
+	loc_wipe (&loc);
+	if (ret)
+		return ret;
+
+	ret = dict_get_str (xattr, GFID_TO_PATH_KEY, &path);
+	if (ret || !path)
+		return -EINVAL;
+
+	*path_p = gf_strdup (path);
+	if (!*path_p)
+		return -ENOMEM;
+	return 0;
+}
+
+
+int
+afr_shd_gather_index_entries (xlator_t *this, int child, dict_t *output)
+{
+	fd_t *fd = NULL;
+	xlator_t *subvol = NULL;
+	afr_private_t *priv = NULL;
+	off_t offset = 0;
+	gf_dirent_t entries;
+	gf_dirent_t *entry = NULL;
+	uuid_t gfid;
+	int ret = 0;
+	int count = 0;
+	char *path = NULL;
+
+	priv = this->private;
+	subvol = priv->children[child];
+
+	fd = afr_shd_index_opendir (this, child);
+	if (!fd) {
+		gf_log (this->name, GF_LOG_WARNING,
+			"unable to opendir index-dir on %s", subvol->name);
+		return -errno;
+	}
+
+	INIT_LIST_HEAD (&entries.list);
+
+	while ((ret = syncop_readdir (subvol, fd, 131072, offset, &entries))) {
+		if (ret > 0)
+			ret = 0;
+		list_for_each_entry (entry, &entries.list, list) {
+			offset = entry->d_off;
+
+			if (!strcmp (entry->d_name, ".") ||
+			    !strcmp (entry->d_name, ".."))
+				continue;
+
+			gf_log (this->name, GF_LOG_DEBUG, "got entry: %s",
+				entry->d_name);
+
+			ret = uuid_parse (entry->d_name, gfid);
+			if (ret)
+				continue;
+
+			path = NULL;
+			ret = afr_shd_gfid_to_path (this, subvol, gfid, &path);
+
+			if (ret == -ENOENT || ret == -ESTALE) {
+				afr_shd_index_purge (subvol, fd->inode,
+						     entry->d_name);
+				ret = 0;
+				continue;
+			}
+
+			ret = afr_shd_dict_add_path (this, output, child, path,
+						     NULL);
+		}
+
+		gf_dirent_free (&entries);
+		if (ret)
+			break;
+	}
+
+	if (fd)
+		fd_unref (fd);
+	if (!ret)
+		ret = count;
+	return ret;
+}
+
+
+int
+afr_add_shd_event (circular_buffer_t *cb, void *data)
+{
+	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;
+
+	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;
+}
+
+int
+afr_add_crawl_event (circular_buffer_t *cb, void *data)
+{
+	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;
+
+	afr_shd_dict_add_crawl_event (this, output, crawl_event);
+
+	return 0;
+}
+
+
+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;
+
+	this->itable = inode_table_new (SHD_INODE_LRU_LIMIT, this);
+	if (!this->itable)
+		goto out;
+
+	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;
+	}
+
+	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;
+	}
+
+	shd->healed = eh_new (AFR_EH_HEALED_LIMIT, _gf_false,
+			      afr_destroy_shd_event_data);
+        if (!shd->healed)
+		goto out;
+
+	shd->heal_failed = eh_new (AFR_EH_HEAL_FAIL_LIMIT, _gf_false,
+				   afr_destroy_shd_event_data);
+	if (!shd->heal_failed)
+		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:
+	return ret;
+}
+
+
+int
+afr_selfheal_childup (xlator_t *this, int subvol)
+{
+	afr_shd_index_healer_spawn (this, subvol);
+
+	return 0;
+}
+
+
+int64_t
+afr_shd_get_index_count (xlator_t *this, int i)
+{
+	afr_private_t *priv = NULL;
+	xlator_t *subvol = NULL;
+	uint64_t count = 0;
+	loc_t rootloc = {0, };
+	dict_t *xattr = NULL;
+	int ret = -1;
+
+	priv = this->private;
+	subvol = priv->children[i];
+
+	rootloc.inode = inode_ref (this->itable->root);
+	uuid_copy (rootloc.gfid, rootloc.inode->gfid);
+
+	ret = syncop_getxattr (subvol, &rootloc, &xattr,
+			       GF_XATTROP_INDEX_COUNT);
+	loc_wipe (&rootloc);
+
+	if (ret < 0)
+		return -1;
+
+	ret = dict_get_uint64 (xattr, GF_XATTROP_INDEX_COUNT, &count);
+	if (ret)
+		return -1;
+	return count;
+}
+
+
+int
+afr_xl_op (xlator_t *this, dict_t *input, dict_t *output)
+{
+        gf_xl_afr_op_t   op = GF_AFR_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 op_ret = 0;
+	int64_t cnt = 0;
+
+	priv = this->private;
+	shd = &priv->shd;
+
+	for (i = 0; i < priv->child_count; i++)
+		if (priv->child_up[i] == -1)
+			goto out;
+
+        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:
+		op_ret = -1;
+
+		for (i = 0; i < priv->child_count; i++) {
+			healer = &shd->index_healers[i];
+			snprintf (key, 64, "%d-%d-status", xl_id, i);
+
+			if (!priv->child_up[i]) {
+				ret = dict_set_str (output, key,
+						    "Brick is not connected");
+			} else if (AFR_COUNT (priv->child_up,
+					      priv->child_count) < 2) {
+				ret = dict_set_str (output, key,
+						    "< 2 bricks in replica are up");
+			} else if (!afr_shd_is_subvol_local (this, healer->subvol)) {
+				ret = dict_set_str (output, key,
+						    "Brick is remote");
+			} else {
+				ret = dict_set_str (output, key,
+						    "Started self-heal");
+				afr_shd_index_healer_spawn (this, i);
+				op_ret = 0;
+			}
+		}
+                break;
+        case GF_AFR_OP_HEAL_FULL:
+		op_ret = -1;
+
+		for (i = 0; i < priv->child_count; i++) {
+			healer = &shd->full_healers[i];
+			snprintf (key, 64, "%d-%d-status", xl_id, i);
+
+			if (!priv->child_up[i]) {
+				ret = dict_set_str (output, key,
+						    "Brick is not connected");
+			} else if (AFR_COUNT (priv->child_up,
+					      priv->child_count) < 2) {
+				ret = dict_set_str (output, key,
+						    "< 2 bricks in replica are up");
+			} else if (!afr_shd_is_subvol_local (this, healer->subvol)) {
+				ret = dict_set_str (output, key,
+						    "Brick is remote");
+			} else {
+				ret = dict_set_str (output, key,
+						    "Started self-heal");
+				afr_shd_full_healer_spawn (this, i);
+				op_ret = 0;
+			}
+		}
+                break;
+        case GF_AFR_OP_INDEX_SUMMARY:
+		for (i = 0; i < priv->child_count; i++)
+			if (shd->index_healers[i].local)
+				afr_shd_gather_index_entries (this, i, output);
+                break;
+        case GF_AFR_OP_HEALED_FILES:
+		eh_dump (shd->healed, output, afr_add_shd_event);
+                break;
+        case GF_AFR_OP_HEAL_FAILED_FILES:
+		eh_dump (shd->heal_failed, output, afr_add_shd_event);
+                break;
+        case GF_AFR_OP_SPLIT_BRAIN_FILES:
+		eh_dump (shd->split_brain, output, afr_add_shd_event);
+                break;
+        case GF_AFR_OP_STATISTICS:
+		for (i = 0; i < priv->child_count; i++) {
+			eh_dump (shd->statistics[i], output,
+				 afr_add_crawl_event);
+			afr_shd_dict_add_crawl_event (this, output,
+						      &shd->index_healers[i].crawl_event);
+			afr_shd_dict_add_crawl_event (this, output,
+						      &shd->full_healers[i].crawl_event);
+		}
+                break;
+        case GF_AFR_OP_STATISTICS_HEAL_COUNT:
+        case GF_AFR_OP_STATISTICS_HEAL_COUNT_PER_REPLICA:
+		op_ret = -1;
+
+		for (i = 0; i < priv->child_count; i++) {
+			if (!priv->child_up[i]) {
+				snprintf (key, 64, "%d-%d-status", xl_id, i);
+				ret = dict_set_str (output, key,
+						    "Brick is not connected");
+			} else {
+				snprintf (key, 64, "%d-%d-hardlinks", xl_id, i);
+				cnt = afr_shd_get_index_count (this, i);
+				if (cnt >= 0) {
+					ret = dict_set_uint64 (output, key, cnt);
+				}
+				op_ret = 0;
+			}
+		}
+
+//                ret = _do_crawl_op_on_local_subvols (this, INDEX_TO_BE_HEALED,
+//                                                     STATISTICS_TO_BE_HEALED,
+//                                                     output);
+                break;
+
+        default:
+                gf_log (this->name, GF_LOG_ERROR, "Unknown set op %d", op);
+                break;
+        }
+out:
+        dict_del (output, this->name);
+        return op_ret;
+}
diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h
new file mode 100644
index 000000000..10e229ee7
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heald.h
@@ -0,0 +1,72 @@
+/*
+  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.
+*/
+
+
+#ifndef _AFR_SELF_HEALD_H
+#define _AFR_SELF_HEALD_H
+
+#include <pthread.h>
+
+
+typedef struct {
+	int child;
+	char *path;
+} shd_event_t;
+
+typedef struct {
+	int      child;
+	uint64_t healed_count;
+        uint64_t split_brain_count;
+        uint64_t heal_failed_count;
+
+	/* 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;
+} crawl_event_t;
+
+struct subvol_healer {
+	xlator_t        *this;
+	int              subvol;
+	gf_boolean_t     local;
+	gf_boolean_t     running;
+	gf_boolean_t     rerun;
+	crawl_event_t    crawl_event;
+	pthread_mutex_t  mutex;
+	pthread_cond_t   cond;
+	pthread_t        thread;
+};
+
+typedef struct {
+	gf_boolean_t            iamshd;
+	gf_boolean_t            enabled;
+	struct subvol_healer   *index_healers;
+	struct subvol_healer   *full_healers;
+
+	eh_t                    *healed;
+        eh_t                    *heal_failed;
+        eh_t                    *split_brain;
+        eh_t                    **statistics;
+} afr_self_heald_t;
+
+
+int
+afr_selfheal_childup (xlator_t *this, int subvol);
+
+int
+afr_selfheal_daemon_init (xlator_t *this);
+
+int
+afr_xl_op (xlator_t *this, dict_t *input, dict_t *output);
+
+#endif /* !_AFR_SELF_HEALD_H */
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
index 2e2c57265..205ff759e 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -1,186 +1,173 @@
 /*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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 "dict.h"
 #include "byte-order.h"
 #include "common-utils.h"
+#include "timer.h"
 
 #include "afr.h"
 #include "afr-transaction.h"
 
 #include <signal.h>
 
+gf_boolean_t
+afr_changelog_pre_op_uninherit (call_frame_t *frame, xlator_t *this);
 
-#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 */
+gf_boolean_t
+afr_changelog_pre_op_update (call_frame_t *frame, xlator_t *this);
 
+int
+afr_changelog_do (call_frame_t *frame, xlator_t *this, dict_t *xattr,
+		  afr_changelog_resume_t changelog_resume);
 
-afr_fd_ctx_t *
-afr_fd_ctx_get (fd_t *fd, xlator_t *this)
-{
-        uint64_t       ctx = 0;
-        afr_fd_ctx_t  *fd_ctx = NULL;
-        int            ret = 0;
 
-        ret = fd_ctx_get (fd, this, &ctx);
+int
+__afr_txn_write_fop (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
 
-        if (ret < 0)
-                goto out;
+        local = frame->local;
+        priv = this->private;
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+        call_count = AFR_COUNT (local->transaction.pre_op, priv->child_count);
 
-out:
-        return fd_ctx;
-}
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
 
+        local->call_count = call_count;
 
-static void
-afr_pid_save (call_frame_t *frame)
-{
-        afr_local_t * local = NULL;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+			local->transaction.wind (frame, this, i);
 
-        local = frame->local;
+                        if (!--call_count)
+                                break;
+                }
+        }
 
-        local->saved_pid = frame->root->pid;
+        return 0;
 }
 
 
-static void
-afr_pid_restore (call_frame_t *frame)
+int
+__afr_txn_write_done (call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t * local = NULL;
+        afr_local_t *local = NULL;
 
         local = frame->local;
 
-        frame->root->pid = local->saved_pid;
-}
+        local->transaction.unwind (frame, this);
 
+        AFR_STACK_DESTROY (frame);
 
-static void
-__mark_all_pending (int32_t *pending[], int child_count,
-                    afr_transaction_type type)
-{
-        int i = 0;
-        int j = 0;
-
-        for (i = 0; i < child_count; i++) {
-                j = afr_index_for_transaction_type (type);
-                pending[i][j] = hton32 (1);
-        }
+        return 0;
 }
 
 
-static void
-__mark_child_dead (int32_t *pending[], int child_count, int child,
-                   afr_transaction_type type)
+call_frame_t*
+afr_transaction_detach_fop_frame (call_frame_t *frame)
 {
-        int j = 0;
+        afr_local_t *   local = NULL;
+        call_frame_t   *fop_frame = NULL;
 
-        j = afr_index_for_transaction_type (type);
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                fop_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
 
-        pending[child][j] = 0;
+        return fop_frame;
 }
 
 
 static void
-__mark_pre_op_done_on_fd (call_frame_t *frame, xlator_t *this, int child_index)
+afr_save_lk_owner (call_frame_t *frame)
 {
-        afr_local_t   *local = NULL;
-        afr_fd_ctx_t  *fd_ctx = NULL;
+        afr_local_t * local = NULL;
 
         local = frame->local;
 
-        if (!local->fd)
-                return;
-
-        fd_ctx = afr_fd_ctx_get (local->fd, this);
-
-        if (!fd_ctx)
-                goto out;
-
-        LOCK (&local->fd->lock);
-        {
-                if (local->transaction.type == AFR_DATA_TRANSACTION)
-                        fd_ctx->pre_op_done[child_index]++;
-        }
-        UNLOCK (&local->fd->lock);
-out:
-        return;
+        local->saved_lk_owner = frame->root->lk_owner;
 }
 
 
 static void
-__mark_pre_op_undone_on_fd (call_frame_t *frame, xlator_t *this, int child_index)
+afr_restore_lk_owner (call_frame_t *frame)
 {
-        afr_local_t   *local = NULL;
-        afr_fd_ctx_t  *fd_ctx = NULL;
+        afr_local_t * local = NULL;
 
         local = frame->local;
 
-        if (!local->fd)
-                return;
-
-        fd_ctx = afr_fd_ctx_get (local->fd, this);
-
-        if (!fd_ctx)
-                goto out;
-
-        LOCK (&local->fd->lock);
-        {
-                if (local->transaction.type == AFR_DATA_TRANSACTION)
-                        fd_ctx->pre_op_done[child_index]--;
-        }
-        UNLOCK (&local->fd->lock);
-out:
-        return;
+        frame->root->lk_owner = local->saved_lk_owner;
 }
 
-
-static void
-__mark_down_children (int32_t *pending[], int child_count,
-                      unsigned char *child_up, afr_transaction_type type)
+void
+__mark_all_success (call_frame_t *frame, xlator_t *this)
 {
-        int i = 0;
-        int j = 0;
+	afr_private_t *priv = NULL;
+	afr_local_t *local = NULL;
+	int i;
 
-        for (i = 0; i < child_count; i++) {
-                j = afr_index_for_transaction_type (type);
+	local = frame->local;
+	priv = this->private;
 
-                if (!child_up[i])
-                        pending[i][j] = 0;
-        }
+	for (i = 0; i < priv->child_count; i++) {
+		local->transaction.failed_subvols[i] = 0;
+	}
 }
 
 
-static void
-__mark_all_success (int32_t *pending[], int child_count,
-                    afr_transaction_type type)
+int
+afr_transaction_perform_fop (call_frame_t *frame, xlator_t *this)
 {
-        int i;
-        int j;
+        afr_local_t     *local = NULL;
+        fd_t            *fd   = NULL;
 
-        for (i = 0; i < child_count; i++) {
-                j = afr_index_for_transaction_type (type);
-                pending[i][j] = hton32 (-1);
-        }
+        local = frame->local;
+        fd    = local->fd;
+
+        /*  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 (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);
+
+        /* 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);
+
+	return 0;
 }
 
 
@@ -215,39 +202,7 @@ __changelog_enabled (afr_private_t *priv, afr_transaction_type type)
 
 
 static int
-__changelog_needed_pre_op (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-
-        int op_ret   = 0;
-
-        priv  = this->private;
-        local = frame->local;
-
-        if (__changelog_enabled (priv, local->transaction.type)) {
-                switch (local->op) {
-
-                case GF_FOP_WRITE:
-                case GF_FOP_FTRUNCATE:
-                        op_ret = 1;
-                        break;
-
-                case GF_FOP_FLUSH:
-                        op_ret = 0;
-                        break;
-
-                default:
-                        op_ret = 1;
-                }
-        }
-
-        return op_ret;
-}
-
-
-static int
-__changelog_needed_post_op (call_frame_t *frame, xlator_t *this)
+__fop_changelog_needed (call_frame_t *frame, xlator_t *this)
 {
         afr_private_t * priv  = NULL;
         afr_local_t   * local = NULL;
@@ -279,63 +234,30 @@ __changelog_needed_post_op (call_frame_t *frame, xlator_t *this)
 }
 
 
-static int
-afr_set_pending_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending)
+int
+afr_set_pending_dict (afr_private_t *priv, dict_t *xattr, int **pending)
 {
         int i = 0;
         int ret = 0;
+	int pending_zero[AFR_NUM_CHANGE_LOGS] = {0, };
 
         for (i = 0; i < priv->child_count; i++) {
+		if (!memcmp (pending_zero, pending[i], sizeof (pending_zero)))
+			/* don't set xattrs for non-pending servers */
+			continue;
+
                 ret = dict_set_static_bin (xattr, priv->pending_key[i],
-                                           pending[i], 3 * sizeof (int32_t));
+					   pending[i],
+					   AFR_NUM_CHANGE_LOGS * sizeof (int));
                 /* 3 = data+metadata+entry */
 
-                if (ret < 0)
-                        goto out;
-        }
-
-out:
-        return ret;
-}
-
-
-static int
-afr_set_piggyback_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending,
-                        afr_transaction_type type)
-{
-        int i = 0;
-        int ret = 0;
-        int *arr = NULL;
-        int index = 0;
-        size_t pending_xattr_size = 3 * sizeof (int32_t);
-        /* 3 = data+metadata+entry */
-
-        index = afr_index_for_transaction_type (type);
-
-        for (i = 0; i < priv->child_count; i++) {
-                arr = GF_CALLOC (1, pending_xattr_size,
-                                 gf_afr_mt_char);
-                if (!arr) {
-                        ret = -1;
-                        goto out;
-                }
-
-                memcpy (arr, pending[i], pending_xattr_size);
-
-                arr[index]++;
-
-                ret = dict_set_bin (xattr, priv->pending_key[i],
-                                    arr, pending_xattr_size);
-
-                if (ret < 0)
-                        goto out;
+                if (ret)
+                        break;
         }
 
-out:
         return ret;
 }
 
-
 int
 afr_lock_server_count (afr_private_t *priv, afr_transaction_type type)
 {
@@ -361,378 +283,529 @@ afr_lock_server_count (afr_private_t *priv, afr_transaction_type type)
 
 /* {{{ 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)
+
+int
+afr_changelog_post_op_done (call_frame_t *frame, xlator_t *this)
 {
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
         afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-        afr_local_t         *local    = NULL;
-        int                  child_index = 0;
-        int                  call_count = -1;
 
-        priv     = this->private;
-        local    = frame->local;
+	local = frame->local;
+	priv = this->private;
         int_lock = &local->internal_lock;
 
-        child_index = (long) cookie;
-
-        if (op_ret == 1) {
-        }
-
-        if (op_ret == 0) {
-                __mark_pre_op_undone_on_fd (frame, this, child_index);
-        }
-
-        LOCK (&frame->lock);
-        {
-                call_count = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
+	if (local->transaction.resume_stub) {
+		call_resume (local->transaction.resume_stub);
+		local->transaction.resume_stub = NULL;
+	}
 
-        if (call_count == 0) {
-                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 (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);
+	}
 
-        return 0;
+	return 0;
 }
 
 
-void
-afr_update_read_child (call_frame_t *frame, xlator_t *this, inode_t *inode,
-                       afr_transaction_type type)
+afr_inodelk_t*
+afr_get_inodelk (afr_internal_lock_t *int_lock, char *dom)
 {
-        int             curr_read_child = -1;
-        int             new_read_child = -1;
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local = NULL;
-        int             **pending = NULL;
-        int             idx = 0;
-        int32_t         *fresh_children = NULL;
-        size_t          success_count = 0;
-
-        idx = afr_index_for_transaction_type (type);
-
-        priv = this->private;
-        local = frame->local;
-        curr_read_child = afr_inode_get_read_ctx (this, inode, NULL);
-        pending = local->pending;
+        afr_inodelk_t *inodelk = NULL;
+        int           i = 0;
 
-        GF_ASSERT (curr_read_child >= 0);
-
-        if (pending[curr_read_child][idx] != 0)
-                goto out;
-
-        fresh_children = afr_fresh_children_create (priv->child_count);
-        if (!fresh_children)
-                goto out;
-
-        for (new_read_child = 0; new_read_child < priv->child_count;
-             new_read_child++) {
+        for (i = 0; int_lock->inodelk[i].domain; i++) {
+                inodelk = &int_lock->inodelk[i];
+                if (strcmp (dom, inodelk->domain) == 0)
+                        return inodelk;
+        }
+        return NULL;
+}
 
-                if (!priv->child_up[new_read_child])
-                        /* child is down */
-                        continue;
+unsigned char*
+afr_locked_nodes_get (afr_transaction_type type, afr_internal_lock_t *int_lock)
+{
+        unsigned char *locked_nodes = NULL;
+        afr_inodelk_t *inodelk = NULL;
+        switch (type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+                inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+                locked_nodes = inodelk->locked_nodes;
+        break;
 
-                if (pending[new_read_child][idx] == 0)
-                        /* op just failed */
-                        continue;
-                fresh_children[success_count] = new_read_child;
-                success_count++;
+        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;
         }
-
-        afr_inode_set_read_ctx (this, inode, fresh_children[0],
-                                fresh_children);
-out:
-        if (fresh_children)
-                GF_FREE (fresh_children);
-        return;
+        return locked_nodes;
 }
 
 
 int
-afr_changelog_post_op (call_frame_t *frame, xlator_t *this)
+afr_changelog_call_count (afr_transaction_type type,
+			  unsigned char *pre_op_subvols,
+			  unsigned int child_count)
 {
-        afr_private_t * priv = this->private;
-        afr_internal_lock_t *int_lock = NULL;
-        int ret        = 0;
-        int i          = 0;
         int call_count = 0;
 
-        afr_local_t *  local = NULL;
-        afr_fd_ctx_t  *fdctx = NULL;
-        dict_t        **xattr = NULL;
-        int            piggyback = 0;
-        int            index = 0;
-        int            nothing_failed = 1;
+	call_count = AFR_COUNT(pre_op_subvols, child_count);
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+        if (type == AFR_ENTRY_RENAME_TRANSACTION)
+                call_count *= 2;
 
-        __mark_down_children (local->pending, priv->child_count,
-                              local->child_up, local->transaction.type);
+        return call_count;
+}
 
-        if (local->fd)
-                afr_update_read_child (frame, this, local->fd->inode,
-                                       local->transaction.type);
 
-        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] = get_new_dict ();
-                dict_ref (xattr[i]);
-        }
+gf_boolean_t
+afr_txn_nothing_failed (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local = NULL;
+        int i = 0;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+        local = frame->local;
+	priv = this->private;
 
-        if (local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
-                call_count *= 2;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.failed_subvols[i])
+                        return _gf_false;
         }
 
-        local->call_count = call_count;
-
-        if (local->fd)
-                fdctx = afr_fd_ctx_get (local->fd, this);
+        return _gf_true;
+}
 
-        if (call_count == 0) {
-                /* no child is up */
-                for (i = 0; i < priv->child_count; i++) {
-                        dict_unref (xattr[i]);
-                }
 
-                int_lock->lock_cbk = local->transaction.done;
-                afr_unlock (frame, this);
-                return 0;
-        }
-
-        /* check if something has failed, to handle piggybacking */
-        nothing_failed = 1;
-        index = afr_index_for_transaction_type (local->transaction.type);
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->pending[i][index] == 0) {
-                        nothing_failed = 0;
-                        break;
-                }
-        }
+void
+afr_handle_symmetric_errors (call_frame_t *frame, xlator_t *this)
+{
+	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;
+
+	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 on 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 (!op_errno) {
+			op_errno = i_errno;
+		} else if (op_errno != i_errno) {
+			/* Mismatching op_errno's */
+			matching_errors = _gf_false;
+			break;
+		}
+	}
+
+	if (matching_errors)
+		__mark_all_success (frame, this);
+}
 
-        index = afr_index_for_transaction_type (local->transaction.type);
-        if (local->optimistic_change_log &&
-            local->transaction.type != AFR_DATA_TRANSACTION) {
-                /* if nothing_failed, then local->pending[..] == {0 .. 0} */
-                for (i = 0; i < priv->child_count; i++)
-                        local->pending[i][index]++;
-        }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (!local->child_up[i])
-                        continue;
+int
+afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t * priv = this->private;
+        int i          = 0;
+	int ret = 0;
+	int idx = 0;
+        afr_local_t *  local = NULL;
+        dict_t        *xattr = NULL;
+        int            nothing_failed = 1;
+	gf_boolean_t   need_undirty = _gf_false;
 
-                ret = afr_set_pending_dict (priv, xattr[i],
-                                            local->pending);
+        local = frame->local;
+	idx = afr_index_for_transaction_type (local->transaction.type);
 
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_INFO,
-                                "failed to set pending entry");
+        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;
 
-                switch (local->transaction.type) {
-                case AFR_DATA_TRANSACTION:
-                {
-                        if (!fdctx) {
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->xattrop,
-                                            &local->loc,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i]);
-                                break;
-                        }
+	if (nothing_failed && !need_undirty) {
+		afr_changelog_post_op_done (frame, this);
+                goto out;
+	}
+
+	xattr = dict_new ();
+	if (!xattr) {
+		local->op_ret = -1;
+		local->op_errno = ENOMEM;
+		afr_changelog_post_op_done (frame, this);
+		goto out;
+	}
+
+	if (need_undirty) {
+		local->dirty[idx] = hton32(-1);
+
+		ret = dict_set_static_bin (xattr, AFR_DIRTY, local->dirty,
+					   sizeof(int) * AFR_NUM_CHANGE_LOGS);
+		if (ret) {
+			local->op_ret = -1;
+			local->op_errno = ENOMEM;
+			afr_changelog_post_op_done (frame, this);
+			goto out;
+		}
+
+	}
+
+	if (!nothing_failed) {
+		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) {
+			local->op_ret = -1;
+			local->op_errno = ENOMEM;
+			afr_changelog_post_op_done (frame, this);
+			goto out;
+		}
+
+	}
+
+	afr_changelog_do (frame, this, xattr, afr_changelog_post_op_done);
+out:
+	if (xattr)
+                dict_unref (xattr);
 
-                        LOCK (&local->fd->lock);
-                        {
-                                piggyback = 0;
-                                if (fdctx->pre_op_piggyback[i]) {
-                                        fdctx->pre_op_piggyback[i]--;
-                                        piggyback = 1;
-                                }
-                        }
-                        UNLOCK (&local->fd->lock);
+        return 0;
+}
 
-                        if (piggyback && !nothing_failed)
-                                ret = afr_set_piggyback_dict (priv, xattr[i],
-                                                              local->pending,
-                                                              local->transaction.type);
 
-                        if (nothing_failed && piggyback) {
-                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
-                                                           this, 1, 0, xattr[i]);
-                        } 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]);
-                        }
-                }
-                break;
-                case AFR_METADATA_TRANSACTION:
-                {
-                        if (nothing_failed) {
-                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
-                                                           this, 1, 0, xattr[i]);
-                                break;
-                        }
+gf_boolean_t
+afr_changelog_pre_op_uninherit (call_frame_t *frame, xlator_t *this)
+{
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	fd_t *fd = NULL;
+	int i = 0;
+	gf_boolean_t ret = _gf_false;
+	afr_fd_ctx_t *fd_ctx = NULL;
+	int type = 0;
+
+	local = frame->local;
+	priv = this->private;
+	fd = local->fd;
+
+	type = afr_index_for_transaction_type (local->transaction.type);
+	if (type != AFR_DATA_TRANSACTION)
+		return !local->transaction.dirtied;
+
+	if (!fd)
+		return !local->transaction.dirtied;
+
+	fd_ctx = afr_fd_ctx_get (fd, this);
+	if (!fd_ctx)
+		return _gf_false;
+
+	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(&fd->lock);
+	{
+		for (i = 0; i < priv->child_count; i++) {
+			if (local->transaction.pre_op[i] !=
+			    fd_ctx->pre_op_done[type][i]) {
+				ret = !local->transaction.dirtied;
+				goto unlock;
+			}
+		}
+
+		if (fd_ctx->inherited[type]) {
+			ret = _gf_true;
+			fd_ctx->inherited[type]--;
+		} else if (fd_ctx->on_disk[type]) {
+			ret = _gf_false;
+			fd_ctx->on_disk[type]--;
+		} else {
+			/* ASSERT */
+			ret = _gf_false;
+		}
+
+		if (!fd_ctx->inherited[type] && !fd_ctx->on_disk[type]) {
+			for (i = 0; i < priv->child_count; i++)
+				fd_ctx->pre_op_done[type][i] = 0;
+		}
+	}
+unlock:
+	UNLOCK(&fd->lock);
+
+	local->transaction.uninherit_done = _gf_true;
+	local->transaction.uninherit_value = ret;
+
+	return ret;
+}
 
-                        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]);
-                        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]);
-                }
-                break;
 
-                case AFR_ENTRY_RENAME_TRANSACTION:
-                {
-                        if (nothing_failed) {
-                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
-                                                           this, 1, 0, xattr[i]);
-                        } 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]);
-                        }
-                        call_count--;
-                }
+gf_boolean_t
+afr_changelog_pre_op_inherit (call_frame_t *frame, xlator_t *this)
+{
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	fd_t *fd = NULL;
+	int i = 0;
+	gf_boolean_t ret = _gf_false;
+	afr_fd_ctx_t *fd_ctx = NULL;
+	int type = 0;
+
+	local = frame->local;
+	priv = this->private;
+	fd = local->fd;
+
+	if (local->transaction.type != AFR_DATA_TRANSACTION)
+		return _gf_false;
+
+	type = afr_index_for_transaction_type (local->transaction.type);
+
+	if (!fd)
+		return _gf_false;
+
+	fd_ctx = afr_fd_ctx_get (fd, this);
+	if (!fd_ctx)
+		return _gf_false;
+
+	LOCK(&fd->lock);
+	{
+		if (!fd_ctx->on_disk[type]) {
+			/* nothing to inherit yet */
+			ret = _gf_false;
+			goto unlock;
+		}
+
+		for (i = 0; i < priv->child_count; i++) {
+			if (local->transaction.pre_op[i] !=
+			    fd_ctx->pre_op_done[type][i]) {
+				/* either inherit exactly, or don't */
+				ret = _gf_false;
+				goto unlock;
+			}
+		}
+
+		fd_ctx->inherited[type]++;
+
+		ret = _gf_true;
+
+		local->transaction.inherited = _gf_true;
+	}
+unlock:
+	UNLOCK(&fd->lock);
+
+	return ret;
+}
 
-                /*
-                  set it again because previous stack_wind
-                  might have already returned (think of case
-                  where subvolume is posix) and would have
-                  used the dict as placeholder for return
-                  value
-                */
 
-                ret = afr_set_pending_dict (priv, xattr[i],
-                                            local->pending);
+gf_boolean_t
+afr_changelog_pre_op_update (call_frame_t *frame, xlator_t *this)
+{
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	fd_t *fd = NULL;
+	afr_fd_ctx_t *fd_ctx = NULL;
+	int i = 0;
+	gf_boolean_t ret = _gf_false;
+	int type = 0;
+
+	local = frame->local;
+	priv = this->private;
+	fd = local->fd;
+
+	if (!fd)
+		return _gf_false;
+
+	fd_ctx = afr_fd_ctx_get (fd, this);
+	if (!fd_ctx)
+		return _gf_false;
+
+	if (local->transaction.inherited)
+		/* was already inherited in afr_changelog_pre_op */
+		return _gf_false;
+
+	if (!local->transaction.dirtied)
+		return _gf_false;
+
+        if (!afr_txn_nothing_failed (frame, this))
+		return _gf_false;
+
+	type = afr_index_for_transaction_type (local->transaction.type);
+
+	ret = _gf_false;
+
+	LOCK(&fd->lock);
+	{
+		if (!fd_ctx->on_disk[type]) {
+			for (i = 0; i < priv->child_count; i++)
+				fd_ctx->pre_op_done[type][i] =
+					local->transaction.pre_op[i];
+		} else {
+			for (i = 0; i < priv->child_count; i++)
+				if (fd_ctx->pre_op_done[type][i] !=
+				    local->transaction.pre_op[i]) {
+					local->transaction.no_uninherit = 1;
+					goto unlock;
+				}
+		}
+		fd_ctx->on_disk[type]++;
+
+		ret = _gf_true;
+	}
+unlock:
+	UNLOCK(&fd->lock);
+
+	return ret;
+}
 
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_INFO,
-                                "failed to set pending entry");
 
-                /* fall through */
+int
+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_local_t *local = NULL;
+        int call_count = -1;
 
-                case AFR_ENTRY_TRANSACTION:
-                {
-                        if (nothing_failed) {
-                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
-                                                           this, 1, 0, xattr[i]);
-                                break;
-                        }
+        local = frame->local;
 
-                        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]);
-                        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]);
-                }
-                break;
-                }
+	if (op_ret == -1)
+		afr_transaction_fop_failed (frame, this, (long) cookie);
 
-                if (!--call_count)
-                        break;
-        }
+	call_count = afr_frame_return (frame);
 
-        for (i = 0; i < priv->child_count; i++) {
-                dict_unref (xattr[i]);
-        }
+        if (call_count == 0)
+		local->transaction.changelog_resume (frame, this);
 
         return 0;
 }
 
 
-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)
+int
+afr_changelog_do (call_frame_t *frame, xlator_t *this, dict_t *xattr,
+		  afr_changelog_resume_t changelog_resume)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = this->private;
-        loc_t       *   loc   = NULL;
-        int call_count  = -1;
-        int child_index = (long) cookie;
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+	int i = 0;
+	int call_count = 0;
 
-        local = frame->local;
-        loc   = &local->loc;
+	local = frame->local;
+	priv = this->private;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 1) {
-                        /* special op_ret for piggyback */
-                }
+        call_count = afr_changelog_call_count (local->transaction.type,
+					       local->transaction.pre_op,
+					       priv->child_count);
 
-                if (op_ret == 0) {
-                        __mark_pre_op_done_on_fd (frame, this, child_index);
-                }
+	if (call_count == 0) {
+		changelog_resume (frame, this);
+		return 0;
+	}
+
+	local->call_count = call_count;
 
-                if (op_ret == -1) {
-                        local->child_up[child_index] = 0;
+	local->transaction.changelog_resume = changelog_resume;
 
-                        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;
+        for (i = 0; i < priv->child_count; i++) {
+                if (!local->transaction.pre_op[i])
+                        continue;
 
-                        } 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));
+                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,
+						   NULL);
+                        } 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,
+						   NULL);
                         }
-                        local->op_errno = op_errno;
-                }
+			break;
+                case AFR_ENTRY_RENAME_TRANSACTION:
 
-                call_count = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
+			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,
+					   NULL);
+                        call_count--;
 
-        if (call_count == 0) {
-                if ((local->op_ret == -1) &&
-                    (local->op_errno == ENOTSUP)) {
-                        local->transaction.resume (frame, this);
-                } else {
-                        __mark_all_success (local->pending, priv->child_count,
-                                            local->transaction.type);
+                /* fall through */
 
-                        afr_pid_restore (frame);
+                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,
+						   NULL);
+                        else
+                                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,
+						   NULL);
+			break;
+		}
 
-                        local->transaction.fop (frame, this);
-                }
+                if (!--call_count)
+                        break;
         }
 
-        return 0;
+	return 0;
 }
 
 
@@ -743,199 +816,122 @@ afr_changelog_pre_op (call_frame_t *frame, xlator_t *this)
         int i = 0;
         int ret = 0;
         int call_count = 0;
-        dict_t **xattr = NULL;
-        afr_fd_ctx_t *fdctx = NULL;
+	int op_errno = 0;
         afr_local_t *local = NULL;
-        int          piggyback = 0;
+        afr_internal_lock_t *int_lock = NULL;
+        unsigned char       *locked_nodes = NULL;
+	unsigned char       *pending_subvols = 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);
 
-        xattr = alloca (priv->child_count * sizeof (*xattr));
-        memset (xattr, 0, (priv->child_count * sizeof (*xattr)));
+        locked_nodes = afr_locked_nodes_get (local->transaction.type, int_lock);
 
-        for (i = 0; i < priv->child_count; i++) {
-                xattr[i] = get_new_dict ();
-                dict_ref (xattr[i]);
-        }
+	pending_subvols = alloca0 (priv->child_count);
 
-        call_count = afr_up_children_count (priv->child_count,
-                                            local->child_up);
+	for (i = 0; i < priv->child_count; i++) {
+		if (locked_nodes[i]) {
+			local->transaction.pre_op[i] = 1;
+			call_count++;
+		} else {
+			pending_subvols[i] = 1;
+		}
+	}
 
-        if (local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
-                call_count *= 2;
-        }
+	/* TBD: quorum check w/ call_count */
 
         if (call_count == 0) {
-                /* no child is up */
-                for (i = 0; i < priv->child_count; i++) {
-                        dict_unref (xattr[i]);
-                }
-
-                local->internal_lock.lock_cbk =
-                        local->transaction.done;
-                afr_unlock (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        __mark_all_pending (local->pending, priv->child_count,
-                            local->transaction.type);
-
-        if (local->fd)
-                fdctx = afr_fd_ctx_get (local->fd, this);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (!local->child_up[i])
-                        continue;
-                ret = afr_set_pending_dict (priv, xattr[i],
-                                            local->pending);
-
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_INFO,
-                                "failed to set pending entry");
+		op_errno = ENOTCONN;
+		goto err;
+	}
+
+	xdata_req = dict_new();
+	if (!xdata_req) {
+		op_errno = ENOMEM;
+		goto err;
+	}
+
+	pre_nop = _gf_true;
+
+	if (afr_changelog_pre_op_inherit (frame, this))
+		goto next;
+
+	if (call_count < priv->child_count) {
+		/* For subvols we are not performing operation on,
+		   mark them as pending up-front along with the FOP
+		   so that we can safely defer unmarking dirty until
+		   later.
+		*/
+		for (i = 0; i < priv->child_count; i++) {
+			if (pending_subvols[i])
+				local->pending[i][idx] = hton32(1);
+		}
+		ret = afr_set_pending_dict (priv, xdata_req,
+					    local->pending);
+		if (ret < 0) {
+			op_errno = ENOMEM;
+			goto err;
+		}
+		pre_nop = _gf_false;
+	}
+
+	if (call_count > 1 &&
+	    (local->transaction.type == AFR_DATA_TRANSACTION ||
+	     !local->optimistic_change_log)) {
+
+		/* If we are performing change on only one subvol, no
+		   need to mark dirty, because we are setting the pending
+		   counts already anyways
+		*/
+		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;
+		}
+
+		pre_nop = _gf_false;
+		local->transaction.dirtied = 1;
+	}
+
+	if (pre_nop)
+		goto next;
+
+	if (!local->pre_op_compat) {
+		dict_copy (xdata_req, local->xdata_req);
+		goto next;
+	}
+
+	afr_changelog_do (frame, this, xdata_req, afr_transaction_perform_fop);
+
+	if (xdata_req)
+		dict_unref (xdata_req);
+
+	return 0;
+next:
+	afr_transaction_perform_fop (frame, this);
+
+	if (xdata_req)
+		dict_unref (xdata_req);
 
+        return 0;
+err:
+	local->internal_lock.lock_cbk = local->transaction.done;
+	local->op_ret = -1;
+	local->op_errno = op_errno;
 
-                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]);
-                                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);
-
-                        if (piggyback)
-                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
-                                                          this, 1, 0, xattr[i]);
-                        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]);
-                }
-                break;
-                case AFR_METADATA_TRANSACTION:
-                {
-                        if (local->optimistic_change_log) {
-                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
-                                                          this, 1, 0, xattr[i]);
-                                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]);
-                        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]);
-                }
-                break;
-
-                case AFR_ENTRY_RENAME_TRANSACTION:
-                {
-                        if (local->optimistic_change_log) {
-                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
-                                                          this, 1, 0, xattr[i]);
-                        } 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]);
-                        }
-
-                        call_count--;
-                }
-
-
-                /*
-                  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
-                */
-
-                ret = afr_set_pending_dict (priv, xattr[i],
-                                            local->pending);
-
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_INFO,
-                                "failed to set pending entry");
-
-                /* fall through */
-
-                case AFR_ENTRY_TRANSACTION:
-                {
-                        if (local->optimistic_change_log) {
-                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
-                                                          this, 1, 0, xattr[i]);
-                                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]);
-                        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]);
-                }
-                break;
-                }
-
-                if (!--call_count)
-                        break;
-        }
+	afr_unlock (frame, this);
 
-        for (i = 0; i < priv->child_count; i++) {
-                dict_unref (xattr[i]);
-        }
+	if (xdata_req)
+		dict_unref (xdata_req);
 
-        return 0;
+	return 0;
 }
 
 
@@ -1084,12 +1080,14 @@ int
 afr_set_transaction_flock (afr_local_t *local)
 {
         afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
 
         int_lock = &local->internal_lock;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
 
-        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;
+        inodelk->flock.l_len   = local->transaction.len;
+        inodelk->flock.l_start = local->transaction.start;
+        inodelk->flock.l_type  = F_WRLCK;
 
         return 0;
 }
@@ -1104,6 +1102,7 @@ afr_lock_rec (call_frame_t *frame, xlator_t *this)
         int_lock = &local->internal_lock;
 
         int_lock->transaction_lk_type = AFR_TRANSACTION_LK;
+        int_lock->domain = this->name;
 
         switch (local->transaction.type) {
         case AFR_DATA_TRANSACTION:
@@ -1117,8 +1116,8 @@ afr_lock_rec (call_frame_t *frame, xlator_t *this)
 
         case AFR_ENTRY_RENAME_TRANSACTION:
 
-                int_lock->lock_cbk = afr_post_blocking_rename_cbk;
-                afr_blocking_lock (frame, this);
+                int_lock->lock_cbk = afr_post_nonblocking_entrylk_cbk;
+                afr_nonblocking_entrylk (frame, this);
                 break;
 
         case AFR_ENTRY_TRANSACTION:
@@ -1140,12 +1139,6 @@ afr_lock_rec (call_frame_t *frame, xlator_t *this)
 int
 afr_lock (call_frame_t *frame, xlator_t *this)
 {
-        afr_pid_save (frame);
-
-        frame->root->pid = (long) frame->root;
-
-        afr_set_lk_owner (frame, this);
-
         afr_set_lock_number (frame, this);
 
         return afr_lock_rec (frame, this);
@@ -1157,47 +1150,427 @@ afr_lock (call_frame_t *frame, xlator_t *this)
 int
 afr_internal_lock_finish (call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t   *local = NULL;
-        afr_private_t *priv  = NULL;
+        if (__fop_changelog_needed (frame, this)) {
+                afr_changelog_pre_op (frame, this);
+        } else {
+                afr_transaction_perform_fop (frame, this);
+        }
+
+        return 0;
+}
+
+
+void
+afr_set_delayed_post_op (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t    *local = NULL;
+        afr_private_t  *priv = NULL;
+
+        /* call this function from any of the related optimizations
+           which benefit from delaying post op are enabled, namely:
+
+           - changelog piggybacking
+           - eager locking
+        */
+
+        priv = this->private;
+        if (!priv)
+                return;
+
+        if (!priv->post_op_delay_secs)
+                return;
 
-        priv  = this->private;
         local = frame->local;
+        if (!local->transaction.eager_lock_on)
+                return;
 
-        if (__changelog_needed_pre_op (frame, this)) {
-                afr_changelog_pre_op (frame, this);
-        } else {
-                __mark_all_success (local->pending, priv->child_count,
-                                    local->transaction.type);
+        if (!local)
+                return;
+
+        if (!local->fd)
+                return;
+
+        if (local->op == GF_FOP_WRITE)
+                local->delayed_post_op = _gf_true;
+}
+
+gf_boolean_t
+afr_are_multiple_fds_opened (fd_t *fd, xlator_t *this)
+{
+        afr_fd_ctx_t *fd_ctx = NULL;
+
+        if (!fd) {
+                /* If false is returned, it may keep on taking eager-lock
+                 * which may lead to starvation, so return true to avoid that.
+                 */
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid fd");
+                return _gf_true;
+        }
+        /* Lets say mount1 has eager-lock(full-lock) and after the eager-lock
+         * is taken mount2 opened the same file, it won't be able to
+         * perform any data operations until mount1 releases eager-lock.
+         * To avoid such scenario do not enable eager-lock for this transaction
+         * if open-fd-count is > 1
+         */
+
+        fd_ctx = afr_fd_ctx_get (fd, this);
+        if (!fd_ctx)
+                return _gf_true;
+
+        if (fd_ctx->open_fd_count > 1)
+                return _gf_true;
+
+        return _gf_false;
+}
+
+
+gf_boolean_t
+is_afr_delayed_changelog_post_op_needed (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t      *local = NULL;
+        gf_boolean_t      res = _gf_false;
+
+        local = frame->local;
+        if (!local)
+                goto out;
+
+        if (!local->delayed_post_op)
+                goto out;
+
+        //Mark pending changelog ASAP
+        if (!afr_txn_nothing_failed (frame, this))
+                goto out;
+
+        if (local->fd && afr_are_multiple_fds_opened (local->fd, this))
+                goto out;
 
-                afr_pid_restore (frame);
+        res = _gf_true;
+out:
+        return res;
+}
+
+
+void
+afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
+                               call_stub_t *stub);
+
+void
+afr_delayed_changelog_wake_up_cbk (void *data)
+{
+        fd_t           *fd = NULL;
+
+        fd = data;
+
+        afr_delayed_changelog_wake_up (THIS, fd);
+}
 
-                local->transaction.fop (frame, this);
+
+/* SET operation */
+int
+afr_fd_report_unstable_write (xlator_t *this, fd_t *fd)
+{
+        afr_fd_ctx_t *fdctx = NULL;
+
+        fdctx = afr_fd_ctx_get (fd, this);
+
+        LOCK(&fd->lock);
+        {
+                fdctx->witnessed_unstable_write = _gf_true;
         }
+        UNLOCK(&fd->lock);
 
         return 0;
 }
 
+/* TEST and CLEAR operation */
+gf_boolean_t
+afr_fd_has_witnessed_unstable_write (xlator_t *this, fd_t *fd)
+{
+        afr_fd_ctx_t *fdctx = NULL;
+        gf_boolean_t witness = _gf_false;
+
+        fdctx = afr_fd_ctx_get (fd, this);
+        if (!fdctx)
+                return _gf_true;
+
+        LOCK(&fd->lock);
+        {
+                if (fdctx->witnessed_unstable_write) {
+                        witness = _gf_true;
+                        fdctx->witnessed_unstable_write = _gf_false;
+                }
+        }
+        UNLOCK (&fd->lock);
+
+        return witness;
+}
+
+
+int
+afr_changelog_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, struct iatt *pre,
+                         struct iatt *post, dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        int child_index = (long) cookie;
+        int call_count = -1;
+        afr_local_t *local = NULL;
+
+        priv = this->private;
+        local = frame->local;
+
+        if (op_ret != 0) {
+                /* Failure of fsync() is as good as failure of previous
+                   write(). So treat it like one.
+		*/
+                gf_log (this->name, GF_LOG_WARNING,
+                        "fsync(%s) failed on subvolume %s. Transaction was %s",
+                        uuid_utoa (local->fd->inode->gfid),
+                        priv->children[child_index]->name,
+                        gf_fop_list[local->op]);
+
+                afr_transaction_fop_failed (frame, this, child_index);
+	}
+
+	call_count = afr_frame_return (frame);
+
+	if (call_count == 0)
+		afr_changelog_post_op_now (frame, this);
+
+	return 0;
+}
+
+
+int
+afr_changelog_fsync (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+        int i = 0;
+        int call_count = 0;
+        afr_private_t *priv = NULL;
+        dict_t *xdata = NULL;
+        GF_UNUSED int ret = -1;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = AFR_COUNT (local->transaction.pre_op, priv->child_count);
+
+        if (!call_count) {
+                /* will go straight to unlock */
+                afr_changelog_post_op_now (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        xdata = dict_new();
+        if (xdata)
+                ret = dict_set_int32 (xdata, "batch-fsync", 1);
+
+        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;
+        }
+
+        if (xdata)
+                dict_unref (xdata);
+
+        return 0;
+}
+
+
+int
+afr_changelog_post_op_safe (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->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->fd)) {
+                afr_changelog_post_op_now (frame, this);
+                return 0;
+        }
+
+        /* Check whether users want durability and perform fsync/post-op
+         * accordingly.
+         */
+        if (priv->ensure_durability) {
+                /* Time to fsync() */
+                afr_changelog_fsync (frame, this);
+        } else {
+                afr_changelog_post_op_now (frame, this);
+        }
+
+        return 0;
+}
+
+
+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 timespec    delta = {0, };
+	afr_private_t     *priv = NULL;
+	afr_local_t       *local = NULL;
+
+	priv = this->private;
+
+	fd_ctx = afr_fd_ctx_get (fd, this);
+	if (!fd_ctx)
+                goto out;
+
+	delta.tv_sec = priv->post_op_delay_secs;
+	delta.tv_nsec = 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);
+
+out:
+	if (prev_frame) {
+		local = prev_frame->local;
+		local->transaction.resume_stub = stub;
+		afr_changelog_post_op_now (prev_frame, this);
+	} else if (stub) {
+		call_resume (stub);
+	}
+}
+
+
+void
+afr_changelog_post_op (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t  *local = NULL;
+
+        local = frame->local;
+
+        if (is_afr_delayed_changelog_post_op_needed (frame, this))
+                afr_delayed_changelog_post_op (this, frame, local->fd, NULL);
+        else
+                afr_changelog_post_op_safe (frame, this);
+}
+
+
+
+/* Wake up the sleeping/delayed post-op, and also register
+   a stub to have it resumed after this transaction
+   completely finishes.
+
+   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);
+}
+
+
+void
+afr_delayed_changelog_wake_up (xlator_t *this, fd_t *fd)
+{
+        afr_delayed_changelog_post_op (this, NULL, fd, NULL);
+}
+
 
 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 (__changelog_needed_post_op (frame, this)) {
+        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 */
+                afr_remove_eager_lock_stub (local);
+        }
+
+        afr_restore_lk_owner (frame);
+
+        afr_handle_symmetric_errors (frame, this);
+
+	if (!local->pre_op_compat)
+		/* new mode, pre-op was done along
+		   with OP */
+		afr_changelog_pre_op_update (frame, this);
+
+        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_changelog_post_op_done (frame, this);
         }
 
         return 0;
@@ -1209,16 +1582,96 @@ afr_transaction_resume (call_frame_t *frame, xlator_t *this)
  */
 
 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;
 
         local = frame->local;
-        priv  = this->private;
 
-        __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)
+{
+        uint64_t start1 = local1->transaction.start;
+        uint64_t start2 = local2->transaction.start;
+        uint64_t end1 = 0;
+        uint64_t end2 = 0;
+
+        if (local1->transaction.len)
+                end1 = start1 + local1->transaction.len - 1;
+        else
+                end1 = ULLONG_MAX;
+
+        if (local2->transaction.len)
+                end2 = start2 + local2->transaction.len - 1;
+        else
+                end2 = ULLONG_MAX;
+
+        return ((end1 >= start2) && (end2 >= start1));
+}
+
+void
+afr_transaction_eager_lock_init (afr_local_t *local, xlator_t *this)
+{
+        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;
+
+        if (afr_are_multiple_fds_opened (local->fd, this))
+                return;
+        /*
+         * Once full file lock is acquired in eager-lock phase, overlapping
+         * writes do not compete for inode-locks, instead are transferred to the
+         * next writes. Because of this overlapping writes are not ordered.
+         * This can cause inconsistencies in replication.
+         * Example:
+         * Two overlapping writes w1, w2 are sent in parallel on same fd
+         * in two threads t1, t2.
+         * Both threads can execute afr_writev_wind in the following manner.
+         * t1 winds w1 on brick-0
+         * t2 winds w2 on brick-0
+         * t2 winds w2 on brick-1
+         * t1 winds w1 on brick-1
+         *
+         * This check makes sure the locks are not transferred for
+         * overlapping writes.
+         */
+        LOCK (&local->fd->lock);
+        {
+                list_for_each_entry (each, &fdctx->eager_locked,
+                                     transaction.eager_locked) {
+                        if (afr_locals_overlap (each, local)) {
+                                local->transaction.eager_lock_on = _gf_false;
+                                goto unlock;
+                        }
+                }
+
+                local->transaction.eager_lock_on = _gf_true;
+                list_add_tail (&local->transaction.eager_locked,
+                               &fdctx->eager_locked);
+        }
+unlock:
+        UNLOCK (&local->fd->lock);
 }
 
 
@@ -1227,20 +1680,43 @@ afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type)
 {
         afr_local_t *   local = NULL;
         afr_private_t * priv  = NULL;
+        fd_t            *fd   = NULL;
+        int             ret   = -1;
 
         local = frame->local;
         priv  = this->private;
 
-        afr_transaction_local_init (local, priv);
-
         local->transaction.resume = afr_transaction_resume;
         local->transaction.type   = type;
 
+        ret = afr_transaction_local_init (local, this);
+        if (ret < 0)
+            goto out;
+
+        afr_transaction_eager_lock_init (local, this);
+
+        if (local->fd && local->transaction.eager_lock_on)
+                afr_set_lk_owner (frame, this, local->fd);
+        else
+                afr_set_lk_owner (frame, this, frame->root);
+
+        if (!local->transaction.eager_lock_on && local->loc.inode) {
+                fd = fd_lookup (local->loc.inode, frame->root->pid);
+                if (fd == NULL)
+                        fd = fd_lookup_anonymous (local->loc.inode);
+
+                if (fd) {
+                        afr_delayed_changelog_wake_up (this, fd);
+                        fd_unref (fd);
+                }
+        }
+
         if (afr_lock_server_count (priv, local->transaction.type) == 0) {
                 afr_internal_lock_finish (frame, this);
         } else {
                 afr_lock (frame, this);
         }
-
-        return 0;
+        ret = 0;
+out:
+        return ret;
 }
diff --git a/xlators/cluster/afr/src/afr-transaction.h b/xlators/cluster/afr/src/afr-transaction.h
index 84cf31d63..77cc8eed0 100644
--- a/xlators/cluster/afr/src/afr-transaction.h
+++ b/xlators/cluster/afr/src/afr-transaction.h
@@ -1,25 +1,18 @@
 /*
-   Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  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 __TRANSACTION_H__
 #define __TRANSACTION_H__
 
+#include "afr.h"
+
 void
 afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this,
 			    int child_index);
@@ -27,7 +20,34 @@ afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this,
 int
 afr_lock_server_count (afr_private_t *priv, afr_transaction_type type);
 
+afr_inodelk_t*
+afr_get_inodelk (afr_internal_lock_t *int_lock, char *dom);
+
 int32_t
 afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type);
 
+int
+afr_set_pending_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending);
+
+void
+afr_set_delayed_post_op (call_frame_t *frame, xlator_t *this);
+
+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);
+
+int afr_read_txn_continue (call_frame_t *frame, xlator_t *this, int subvol);
+
+int __afr_txn_write_fop (call_frame_t *frame, xlator_t *this);
+int __afr_txn_write_done (call_frame_t *frame, xlator_t *this);
+call_frame_t *afr_transaction_detach_fop_frame (call_frame_t *frame);
+
 #endif /* __TRANSACTION_H__ */
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index c6705fc68..ead08425f 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include <libgen.h>
@@ -37,8 +28,13 @@ notify (xlator_t *this, int32_t event,
         void *data, ...)
 {
         int ret = -1;
+        va_list         ap;
+        void *data2 = NULL;
 
-        ret = afr_notify (this, event, data);
+        va_start (ap, data);
+        data2 = va_arg (ap, dict_t*);
+        va_end (ap);
+        ret = afr_notify (this, event, data, data2);
 
         return ret;
 }
@@ -62,284 +58,147 @@ mem_acct_init (xlator_t *this)
         return ret;
 }
 
+
 int
-validate_options (xlator_t *this, char **op_errstr)
+xlator_subvolume_index (xlator_t *this, xlator_t *subvol)
 {
-        int                 ret = 0;
-        volume_opt_list_t  *vol_opt = NULL;
-        volume_opt_list_t  *tmp;
+        int index = -1;
+        int i = 0;
+        xlator_list_t *list = NULL;
 
-        if (!this) {
-                gf_log (this->name, GF_LOG_DEBUG, "'this' not a valid ptr");
-                ret =-1;
-                goto out;
-        }
+        list = this->children;
 
-        if (list_empty (&this->volume_options))
-                goto out;
-
-        vol_opt = list_entry (this->volume_options.next,
-                              volume_opt_list_t, list);
-        list_for_each_entry_safe (vol_opt, tmp, &this->volume_options, list) {
-                ret = validate_xlator_volume_options_attacherr (this,
-                                                                vol_opt->given_opt,
-                                                                op_errstr);
+        while (list) {
+                if (subvol == list->xlator ||
+                    strcmp (subvol->name, list->xlator->name) == 0) {
+                        index = i;
+                        break;
+                }
+                list = list->next;
+                i++;
         }
 
-out:
-
-        return ret;
+        return index;
 }
 
+void
+fix_quorum_options (xlator_t *this, afr_private_t *priv, char *qtype)
+{
+        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;
+        }
+}
 
 int
 reconfigure (xlator_t *this, dict_t *options)
 {
-
-        gf_boolean_t metadata_self_heal;   /* on/off */
-        gf_boolean_t entry_self_heal;
-        gf_boolean_t data_self_heal;
-        gf_boolean_t data_change_log;       /* on/off */
-        gf_boolean_t metadata_change_log;   /* on/off */
-        gf_boolean_t entry_change_log;      /* on/off */
-        gf_boolean_t strict_readdir;
-
-        afr_private_t * priv        = NULL;
-        xlator_list_t * trav        = NULL;
-
-        char * read_subvol     = NULL;
-        char * self_heal       = NULL;
-        char * change_log      = NULL;
-        char * str_readdir     = NULL;
-        char * self_heal_algo  = NULL;
-
-        int32_t background_count  = 0;
-        int32_t window_size       = 0;
-
-        int    read_ret      = -1;
-        int    dict_ret      = -1;
-        int    flag          = 1;
-        int    ret           = 0;
-        int    temp_ret      = -1;
+        afr_private_t *priv        = NULL;
+        xlator_t      *read_subvol = NULL;
+        int            read_subvol_index = -1;
+        int            ret         = -1;
+        int            index       = -1;
+        char          *qtype       = NULL;
 
         priv = this->private;
 
-        dict_ret = dict_get_int32 (options, "background-self-heal-count",
-                                   &background_count);
-        if (dict_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reconfiguring background self-heal count to %d",
-                        background_count);
+	GF_OPTION_RECONF ("afr-dirty-xattr",
+			  priv->afr_dirty, options, str,
+			  out);
 
-                priv->background_self_heal_count = background_count;
-        }
+	GF_OPTION_RECONF ("metadata-splitbrain-forced-heal",
+			  priv->metadata_splitbrain_forced_heal, options, bool,
+			  out);
 
-        dict_ret = dict_get_str (options, "metadata-self-heal",
-                                 &self_heal);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (self_heal, &metadata_self_heal);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Reconfiguration Invalid 'option metadata"
-                                "-self-heal %s'. Defaulting to old value.",
-                                self_heal);
-                        ret = -1;
-                        goto out;
-                }
+        GF_OPTION_RECONF ("background-self-heal-count",
+                          priv->background_self_heal_count, options, uint32,
+                          out);
 
-                priv->metadata_self_heal = metadata_self_heal;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reconfiguring 'option metadata"
-                        "-self-heal %s'.",
-                        self_heal);
-        }
+        GF_OPTION_RECONF ("metadata-self-heal",
+                          priv->metadata_self_heal, options, bool, out);
 
-        dict_ret = dict_get_str (options, "data-self-heal", &self_heal);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (self_heal, &data_self_heal);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Reconfiguration Invalid 'option data"
-                                "-self-heal %s'. Defaulting to old value.",
-                                self_heal);
-                        ret = -1;
-                        goto out;
-                }
+        GF_OPTION_RECONF ("data-self-heal", priv->data_self_heal, options, str,
+                          out);
 
-                priv->data_self_heal = data_self_heal;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reconfiguring 'option data"
-                        "-self-heal %s'.", self_heal);
-        }
+        GF_OPTION_RECONF ("entry-self-heal", priv->entry_self_heal, options,
+                          bool, out);
 
-        dict_ret = dict_get_str (options, "entry-self-heal",
-                                 &self_heal);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (self_heal, &entry_self_heal);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Reconfiguration Invalid 'option data"
-                                "-self-heal %s'. Defaulting to old value.",
-                                self_heal);
-                        ret = -1;
-                        goto out;
-                }
+        GF_OPTION_RECONF ("data-self-heal-window-size",
+                          priv->data_self_heal_window_size, options,
+                          uint32, out);
 
-                priv->entry_self_heal = entry_self_heal;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reconfiguring 'option entry"
-                        "-self-heal %s'.", self_heal);
-        }
+        GF_OPTION_RECONF ("data-change-log", priv->data_change_log, options,
+                          bool, out);
 
+        GF_OPTION_RECONF ("metadata-change-log",
+                          priv->metadata_change_log, options, bool, out);
 
-        dict_ret = dict_get_str (options, "strict-readdir",
-                                 &str_readdir);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (str_readdir, &strict_readdir);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option strict-readdir %s'. "
-                                "Defaulting to old value.",
-                                str_readdir);
-                        ret = -1;
-                        goto out;
-                }
-
-                priv->strict_readdir = strict_readdir;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reconfiguring 'option strict"
-                        "-readdir %s'.", str_readdir);
-        }
+        GF_OPTION_RECONF ("entry-change-log", priv->entry_change_log, options,
+                          bool, out);
 
-        dict_ret = dict_get_int32 (options, "data-self-heal-window-size",
-                                   &window_size);
-        if (dict_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reconfiguring, Setting data self-heal window size to %d",
-                        window_size);
+        GF_OPTION_RECONF ("data-self-heal-algorithm",
+                          priv->data_self_heal_algorithm, options, str, out);
 
-                priv->data_self_heal_window_size = window_size;
-        }
-        else {
-                priv->data_self_heal_window_size = 16;
-        }
+        GF_OPTION_RECONF ("read-subvolume", read_subvol, options, xlator, out);
 
+        GF_OPTION_RECONF ("read-hash-mode", priv->hash_mode,
+                          options, uint32, out);
 
-        dict_ret = dict_get_str (options, "data-change-log", &change_log);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (change_log, &data_change_log);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Reconfiguration Invalid 'option data-"
-                                "change-log %s'. Defaulting to old value.",
-                                change_log);
-                        ret = -1;
+        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->data_change_log = data_change_log;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reconfiguring 'option data-"
-                        "change-log %s'.", change_log);
+                priv->read_child = index;
         }
 
-        dict_ret = dict_get_str (options, "metadata-change-log",
-                                 &change_log);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (change_log,
-                                              &metadata_change_log);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option metadata-change-log %s'. "
-                                "Defaulting to metadata-change-log as 'off'.",
-                                change_log);
-                        ret = -1;
-                        goto out;
-                }
+        GF_OPTION_RECONF ("read-subvolume-index",read_subvol_index, options,int32,out);
 
-                priv->metadata_change_log = metadata_change_log;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reconfiguring 'option metadata-"
-                        "change-log %s'.", change_log);
-        }
-
-        dict_ret = dict_get_str (options, "entry-change-log",
-                                 &change_log);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (change_log, &entry_change_log);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option entry-change-log %s'. "
-                                "Defaulting to entry-change-log as 'on'.",
-                                change_log);
-                        ret = -1;
+        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->entry_change_log = entry_change_log;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reconfiguring 'option entry-"
-                        "change-log %s'.", change_log);
+                priv->read_child = index;
         }
 
-        dict_ret = dict_get_str (options, "data-self-heal-algorithm",
-                                 &self_heal_algo);
-        if (dict_ret == 0) {
-                /* Handling both strcmp cases - s1 > s2 and s1 < s2 */
-
-                if (!strcmp (self_heal_algo, "full")) {
-                        priv->data_self_heal_algorithm = self_heal_algo;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Reconfiguring 'option data-self"
-                                "heal-algorithm %s'.", self_heal_algo);
-                        goto next;
-                }
+        GF_OPTION_RECONF ("pre-op-compat", priv->pre_op_compat, options, bool, out);
 
-                if (!strcmp (self_heal_algo, "diff")) {
-                        priv->data_self_heal_algorithm = self_heal_algo;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Reconfiguring 'option data-self"
-                                "heal-algorithm %s'.", self_heal_algo);
-                        goto next;
-                }
+        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_log (this->name, GF_LOG_WARNING,
-                        "Invalid self-heal algorithm %s,"
-                        "defaulting back to old value",
-                        self_heal_algo);
-                ret = -1;
-                goto out;
-        }
+	GF_OPTION_RECONF ("post-op-delay-secs", priv->post_op_delay_secs, options,
+			  uint32, out);
 
-        read_ret = dict_get_str (options, "read-subvolume", &read_subvol);
+        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);
 
-        if (read_ret < 0)
-                goto next;// No need to traverse, hence set the next option
+	GF_OPTION_RECONF ("self-heal-daemon", priv->shd.enabled, options,
+			  bool, out);
 
-        trav = this->children;
-        flag = 0;
-        while (trav) {
-                if (!read_ret && !strcmp (read_subvol, trav->xlator->name)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Subvolume '%s' specified as read child.",
-                                trav->xlator->name);
-
-                        flag = 1;
-                        break;
-                }
+	GF_OPTION_RECONF ("iam-self-heal-daemon", priv->shd.iamshd, options,
+			  bool, out);
 
-                trav = trav->next;
-        }
-
-        if (flag == 0 ) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Invalid 'option read-subvolume %s', no such subvolume"
-                        , read_subvol);
-                ret = -1;
-                goto out;
-        }
+        priv->did_discovery = _gf_false;
 
-next:
+        ret = 0;
 out:
         return ret;
 
@@ -354,39 +213,20 @@ static const char *favorite_child_warning_str = "You have specified subvolume '%
         "subvolumes. All versions of the file except that on '%s' "
         "WILL BE LOST.";
 
-static const char *no_lock_servers_warning_str = "You have set lock-server-count = 0. "
-        "This means correctness is NO LONGER GUARANTEED in all cases. If two or more "
-        "applications write to the same region of a file, there is a possibility that "
-        "its copies will be INCONSISTENT. Set it to a value greater than 0 unless you "
-        "are ABSOLUTELY SURE of what you are doing and WILL NOT HOLD GlusterFS "
-        "RESPONSIBLE for inconsistent data. If you are in doubt, set it to a value "
-        "greater than 0.";
 
 int32_t
 init (xlator_t *this)
 {
-        afr_private_t * priv        = NULL;
-        int             child_count = 0;
-        xlator_list_t * trav        = NULL;
-        int             i           = 0;
-        int             ret         = -1;
-        int             op_errno    = 0;
-        char * read_subvol     = NULL;
-        char * fav_child       = NULL;
-        char * self_heal       = NULL;
-        char * algo            = NULL;
-        char * change_log      = NULL;
-        char * strict_readdir  = NULL;
-        char * inodelk_trace   = NULL;
-        char * entrylk_trace   = NULL;
-        char * def_val         = NULL;
-        int32_t background_count  = 0;
-        int32_t lock_server_count = 1;
-        int32_t window_size       = 0;
-        int    fav_ret       = -1;
-        int    read_ret      = -1;
-        int    dict_ret      = -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;
+        xlator_t      *fav_child   = NULL;
+        char          *qtype       = NULL;
 
         if (!this->children) {
                 gf_log (this->name, GF_LOG_ERROR,
@@ -400,288 +240,111 @@ init (xlator_t *this)
                         "Volume is dangling.");
         }
 
-        ALLOC_OR_GOTO (this->private, afr_private_t, 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);
 
-        read_ret = dict_get_str (this->options, "read-subvolume", &read_subvol);
-        priv->read_child = -1;
-
-        fav_ret = dict_get_str (this->options, "favorite-child", &fav_child);
-        priv->favorite_child = -1;
-
-        priv->background_self_heal_count = 16;
+        child_count = xlator_subvolume_count (this);
 
-        dict_ret = dict_get_int32 (this->options, "background-self-heal-count",
-                                   &background_count);
-        if (dict_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Setting background self-heal count to %d",
-                        background_count);
+        priv->child_count = child_count;
 
-                priv->background_self_heal_count = background_count;
-        }
+        priv->read_child = -1;
 
-        /* Default values */
-
-        priv->data_self_heal     = 1;
-        priv->metadata_self_heal = 1;
-        priv->entry_self_heal    = 1;
-
-        dict_ret = dict_get_str (this->options, "data-self-heal", &self_heal);
-        if (dict_ret == 0) {
-                ret = gf_string2boolean (self_heal, &priv->data_self_heal);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option data-self-heal %s'. "
-                                "Defaulting to data-self-heal as 'on'",
-                                self_heal);
-                        priv->data_self_heal = 1;
-                }
-        }
+	GF_OPTION_INIT ("afr-dirty-xattr", priv->afr_dirty, str, out);
 
-        if (xlator_get_volopt_info (&this->volume_options,
-                                    "data-self-heal-algorithm", &def_val, NULL)) {
-                gf_log (this->name, GF_LOG_ERROR, "Default value of "
-                                " data-self-heal-algorithm not found");
-                ret = -1;
-                goto out;
-        } else {
-                priv->data_self_heal_algorithm = def_val;
-        }
-        dict_ret = dict_get_str (this->options, "data-self-heal-algorithm",
-                                 &algo);
-        if (dict_ret == 0) {
-                priv->data_self_heal_algorithm = gf_strdup (algo);
-        }
+	GF_OPTION_INIT ("metadata-splitbrain-forced-heal",
+			priv->metadata_splitbrain_forced_heal, bool, out);
 
-        if (xlator_get_volopt_info (&this->volume_options,
-                                    "data-self-heal-window-size",&def_val,
-                                    NULL)) {
-                gf_log (this->name, GF_LOG_ERROR, "Default value of "
-                          "data-self-heal-window-size not found");
-                ret = -1;
-                goto out;
-        } else {
-                if (gf_string2int32 (def_val,
-                                    (int *)&priv->data_self_heal_window_size)) {
-                        gf_log (this->name, GF_LOG_ERROR, "Default value of "
-                                 "data-self-heal-window-size  corrupt");
-                        ret = -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;
                 }
         }
-
-        dict_ret = dict_get_int32 (this->options, "data-self-heal-window-size",
-                                   &window_size);
-        if (dict_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Setting data self-heal window size to %d",
-                        window_size);
-
-                priv->data_self_heal_window_size = window_size;
-        }
-
-        dict_ret = dict_get_str (this->options, "metadata-self-heal",
-                                 &self_heal);
-        if (dict_ret == 0) {
-                ret = gf_string2boolean (self_heal, &priv->metadata_self_heal);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option metadata-self-heal %s'. "
-                                "Defaulting to metadata-self-heal as 'on'.",
-                                self_heal);
-                        priv->metadata_self_heal = 1;
-                }
-        }
-
-        dict_ret = dict_get_str (this->options, "entry-self-heal", &self_heal);
-        if (dict_ret == 0) {
-                ret = gf_string2boolean (self_heal, &priv->entry_self_heal);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option entry-self-heal %s'. "
-                                "Defaulting to entry-self-heal as 'on'.",
-                                self_heal);
-                        priv->entry_self_heal = 1;
-                }
-        }
-
-        /* Change log options */
-
-        priv->data_change_log     = 1;
-        priv->metadata_change_log = 1;
-        priv->entry_change_log    = 1;
-        priv->optimistic_change_log = 1;
-
-        dict_ret = dict_get_str (this->options, "data-change-log",
-                                 &change_log);
-        if (dict_ret == 0) {
-                ret = gf_string2boolean (change_log, &priv->data_change_log);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option data-change-log %s'. "
-                                "Defaulting to data-change-log as 'on'.",
-                                change_log);
-                        priv->data_change_log = 1;
-                }
-        }
-
-        dict_ret = dict_get_str (this->options, "metadata-change-log",
-                                 &change_log);
-        if (dict_ret == 0) {
-                ret = gf_string2boolean (change_log,
-                                         &priv->metadata_change_log);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option metadata-change-log %s'. "
-                                "Defaulting to metadata-change-log as 'off'.",
-                                change_log);
-                        priv->metadata_change_log = 0;
-                }
-        }
-
-        dict_ret = dict_get_str (this->options, "entry-change-log",
-                                 &change_log);
-        if (dict_ret == 0) {
-                ret = gf_string2boolean (change_log, &priv->entry_change_log);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option entry-change-log %s'. "
-                                "Defaulting to entry-change-log as 'on'.",
-                                change_log);
-                        priv->entry_change_log = 1;
-                }
-        }
-
-        dict_ret = dict_get_str (this->options, "optimistic-change-log",
-                                 &change_log);
-        if (dict_ret == 0) {
-                ret = gf_string2boolean (change_log, &priv->optimistic_change_log);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option optimistic-change-log %s'. "
-                                "Defaulting to optimistic-change-log as 'on'.",
-                                change_log);
-                        priv->optimistic_change_log = 1;
+        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);
 
-        /* Locking options */
-
-        priv->inodelk_trace = 0;
-        priv->entrylk_trace = 0;
-
-        dict_ret = dict_get_str (this->options, "inodelk-trace",
-                                 &inodelk_trace);
-        if (dict_ret == 0) {
-                ret = gf_string2boolean (inodelk_trace, &priv->inodelk_trace);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option inodelk-trace %s' ",
-                                inodelk_trace);
+        GF_OPTION_INIT ("read-hash-mode", priv->hash_mode, uint32, out);
 
-                        priv->inodelk_trace = 0;
+        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);
         }
 
 
-        dict_ret = dict_get_str (this->options, "entrylk-trace",
-                                 &entrylk_trace);
-        if (dict_ret == 0) {
-                ret = gf_string2boolean (entrylk_trace, &priv->entrylk_trace);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option entrylk-trace %s' ",
-                                inodelk_trace);
+        GF_OPTION_INIT ("background-self-heal-count",
+                        priv->background_self_heal_count, uint32, out);
 
-                        priv->entrylk_trace = 0;
-                }
-        }
+        GF_OPTION_INIT ("data-self-heal", priv->data_self_heal, str, out);
 
+        GF_OPTION_INIT ("data-self-heal-algorithm",
+                        priv->data_self_heal_algorithm, str, out);
 
-        priv->data_lock_server_count = 1;
-        priv->metadata_lock_server_count = 0;
-        priv->entry_lock_server_count = 1;
+        GF_OPTION_INIT ("data-self-heal-window-size",
+                        priv->data_self_heal_window_size, uint32, out);
 
-        dict_ret = dict_get_int32 (this->options, "data-lock-server-count",
-                                   &lock_server_count);
-        if (dict_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Setting data lock server count to %d.",
-                        lock_server_count);
+        GF_OPTION_INIT ("metadata-self-heal", priv->metadata_self_heal, bool,
+                        out);
 
-                if (lock_server_count == 0)
-                        gf_log (this->name, GF_LOG_WARNING, "%s",
-                                no_lock_servers_warning_str);
+        GF_OPTION_INIT ("entry-self-heal", priv->entry_self_heal, bool, out);
 
-                priv->data_lock_server_count = lock_server_count;
-        }
+        GF_OPTION_INIT ("data-change-log", priv->data_change_log, bool, out);
 
+        GF_OPTION_INIT ("metadata-change-log", priv->metadata_change_log, bool,
+                        out);
 
-        dict_ret = dict_get_int32 (this->options,
-                                   "metadata-lock-server-count",
-                                   &lock_server_count);
-        if (dict_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Setting metadata lock server count to %d.",
-                        lock_server_count);
-                priv->metadata_lock_server_count = lock_server_count;
-        }
+        GF_OPTION_INIT ("entry-change-log", priv->entry_change_log, bool, out);
 
+        GF_OPTION_INIT ("optimistic-change-log", priv->optimistic_change_log,
+                        bool, out);
 
-        dict_ret = dict_get_int32 (this->options, "entry-lock-server-count",
-                                   &lock_server_count);
-        if (dict_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Setting entry lock server count to %d.",
-                        lock_server_count);
+        GF_OPTION_INIT ("inodelk-trace", priv->inodelk_trace, bool, out);
 
-                priv->entry_lock_server_count = lock_server_count;
-        }
+        GF_OPTION_INIT ("entrylk-trace", priv->entrylk_trace, bool, out);
 
-        priv->strict_readdir = _gf_false;
-
-        dict_ret = dict_get_str (this->options, "strict-readdir",
-                                 &strict_readdir);
-        if (dict_ret == 0) {
-                ret = gf_string2boolean (strict_readdir, &priv->strict_readdir);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Invalid 'option strict-readdir %s'. "
-                                "Defaulting to strict-readdir as 'off'.",
-                                strict_readdir);
-                }
-        }
+        GF_OPTION_INIT ("pre-op-compat", priv->pre_op_compat, bool, out);
 
-        trav = this->children;
-        while (trav) {
-                if (!read_ret && !strcmp (read_subvol, trav->xlator->name)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Subvolume '%s' specified as read child.",
-                                trav->xlator->name);
+        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);
 
-                        priv->read_child = child_count;
-                }
+	GF_OPTION_INIT ("post-op-delay-secs", priv->post_op_delay_secs, uint32, out);
+        GF_OPTION_INIT ("ensure-durability", priv->ensure_durability, bool,
+                        out);
 
-                if (fav_ret == 0 && !strcmp (fav_child, trav->xlator->name)) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                favorite_child_warning_str, trav->xlator->name,
-                                trav->xlator->name, trav->xlator->name);
-                        priv->favorite_child = child_count;
-                }
+	GF_OPTION_INIT ("self-heal-daemon", priv->shd.enabled, bool, out);
 
-                child_count++;
-                trav = trav->next;
-        }
+	GF_OPTION_INIT ("iam-self-heal-daemon", priv->shd.iamshd, bool, out);
 
         priv->wait_count = 1;
 
-        priv->child_count = child_count;
-
-        LOCK_INIT (&priv->lock);
-        LOCK_INIT (&priv->read_child_lock);
-
         priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count,
                                     gf_afr_mt_char);
         if (!priv->child_up) {
@@ -720,8 +383,6 @@ init (xlator_t *this)
                                    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;
                 }
@@ -730,6 +391,13 @@ init (xlator_t *this)
                 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) {
@@ -737,12 +405,22 @@ init (xlator_t *this)
                 goto out;
         }
 
-        LOCK_INIT (&priv->root_inode_lk);
-        priv->first_lookup = 1;
-        priv->root_inode = NULL;
+	ret = afr_selfheal_daemon_init (this);
+	if (ret) {
+		ret = -ENOMEM;
+		goto out;
+	}
 
-        pthread_mutex_init (&priv->mutex, NULL);
-        INIT_LIST_HEAD (&priv->saved_fds);
+        /* 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->root_inode = NULL;
 
         ret = 0;
 out:
@@ -753,6 +431,13 @@ out:
 int
 fini (xlator_t *this)
 {
+        afr_private_t *priv = NULL;
+
+        priv = this->private;
+        this->private = NULL;
+        afr_priv_destroy (priv);
+        //if (this->itable);//I dont see any destroy func
+
         return 0;
 }
 
@@ -771,6 +456,9 @@ struct xlator_fops fops = {
         .finodelk    = afr_finodelk,
         .entrylk     = afr_entrylk,
         .fentrylk    = afr_fentrylk,
+	.fallocate   = afr_fallocate,
+	.discard     = afr_discard,
+        .zerofill    = afr_zerofill,
 
         /* inode read */
         .access      = afr_access,
@@ -778,6 +466,7 @@ struct xlator_fops fops = {
         .fstat       = afr_fstat,
         .readlink    = afr_readlink,
         .getxattr    = afr_getxattr,
+        .fgetxattr   = afr_fgetxattr,
         .readv       = afr_readv,
 
         /* inode write */
@@ -785,9 +474,11 @@ struct xlator_fops fops = {
         .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,
@@ -820,68 +511,239 @@ struct xlator_cbks cbks = {
 
 struct volume_options options[] = {
         { .key  = {"read-subvolume" },
-          .type = GF_OPTION_TYPE_XLATOR
+          .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 = "1",
+          .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 up server, "
+                         "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
+          .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
+          .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_BOOL
+          .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,
-          .default_value = "",
           .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.",
-          .value = { "diff", "full" }
+                           "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 = "16",
+          .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
+          .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
+          .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
+          .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
+          .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
+          .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
+          .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  = {"data-lock-server-count"},
-          .type = GF_OPTION_TYPE_INT,
-          .min  = 0
+        { .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",
+          .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 "
+                         "acquisition 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 \"optimized\" transaction."
+
+        },
+        { .key = {"self-heal-daemon"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "on",
+          .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  = {"metadata-lock-server-count"},
+        { .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  = 0
+          .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  = {"entry-lock-server-count"},
+        { .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
+          .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  = {"strict-readdir"},
+        { .key = {"ensure-durability"},
           .type = GF_OPTION_TYPE_BOOL,
+          .description = "Afr performs fsyncs for transactions if this "
+                         "option is on to make sure the changelogs/data is "
+                         "written to the disk",
+          .default_value = "on",
         },
+	{ .key = {"afr-dirty-xattr"},
+	  .type = GF_OPTION_TYPE_STR,
+	  .default_value = AFR_DIRTY_DEFAULT,
+	},
+	{ .key = {"metadata-splitbrain-forced-heal"},
+	  .type = GF_OPTION_TYPE_BOOL,
+	  .default_value = "off",
+	},
         { .key  = {NULL} },
 };
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 7d5d05479..36042f7b2 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -31,57 +22,47 @@
 #include "afr-mem-types.h"
 
 #include "libxlator.h"
+#include "timer.h"
+#include "syncop.h"
+
+#include "afr-self-heald.h"
 
 #define AFR_XATTR_PREFIX "trusted.afr"
 #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)
 
-struct _pump_private;
+#define AFR_LOCKEE_COUNT_MAX    3
+#define AFR_DOM_COUNT_MAX    3
+#define AFR_NUM_CHANGE_LOGS            3 /*data + metadata + entry*/
 
-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_lock_cbk_t) (call_frame_t *frame, xlator_t *this);
 
-typedef int (*afr_impunge_done_cbk_t) (call_frame_t *frame, xlator_t *this,
-                                       int child, 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_read_txn_wind_t) (call_frame_t *frame, xlator_t *this, int subvol);
 
-typedef int (*afr_lock_cbk_t) (call_frame_t *frame, xlator_t *this);
+typedef int (*afr_inode_refresh_cbk_t) (call_frame_t *frame, xlator_t *this, int err);
 
-typedef struct afr_inode_params_ {
-        uint64_t mask_type;
-        union {
-                gf_boolean_t value;
-                struct {
-                        int32_t read_child;
-                        int32_t *fresh_children;
-                } read_ctx;
-        } u;
-} afr_inode_params_t;
+typedef int (*afr_changelog_resume_t) (call_frame_t *frame, xlator_t *this);
 
-typedef struct afr_inode_ctx_ {
-        uint64_t masks;
-        int32_t  *fresh_children;//increasing order of latency
-} afr_inode_ctx_t;
+#define alloca0(size) ({void *__ptr; __ptr = alloca(size); memset(__ptr, 0, size); __ptr;})
+#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];})
 
 typedef struct _afr_private {
         gf_lock_t lock;               /* to guard access to child_count, etc */
         unsigned int child_count;     /* total number of children   */
 
-        unsigned int read_child_rr;   /* round-robin index of the read_child */
-        gf_lock_t read_child_lock;    /* lock to protect above */
-
         xlator_t **children;
 
-        gf_lock_t root_inode_lk;
-        int first_lookup;
         inode_t *root_inode;
 
         unsigned char *child_up;
 
         char **pending_key;
 
-        gf_boolean_t data_self_heal;              /* on/off */
+        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 */
@@ -95,141 +76,53 @@ typedef struct _afr_private {
         gf_boolean_t metadata_change_log;   /* on/off */
         gf_boolean_t entry_change_log;      /* on/off */
 
+	gf_boolean_t metadata_splitbrain_forced_heal; /* on/off */
         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 */
 
-        unsigned int data_lock_server_count;
-        unsigned int metadata_lock_server_count;
-        unsigned int entry_lock_server_count;
-
         gf_boolean_t inodelk_trace;
         gf_boolean_t entrylk_trace;
 
-        gf_boolean_t strict_readdir;
-
         unsigned int wait_count;      /* # of servers to wait for success */
 
         uint64_t up_count;      /* number of CHILD_UPs we have seen */
         uint64_t down_count;    /* number of CHILD_DOWNs we have seen */
 
-        struct _pump_private *pump_private; /* Set if we are loaded as pump */
-        int                   use_afr_in_pump;
-
-        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      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;
 
         char                   vol_uuid[UUID_SIZE + 1];
         int32_t                *last_event;
-} afr_private_t;
-
-typedef struct {
-        /* External interface: These are variables (some optional) that
-           are set by whoever has triggered self-heal */
-
-        gf_boolean_t need_data_self_heal;
-        gf_boolean_t need_metadata_self_heal;
-        gf_boolean_t need_entry_self_heal;
-        gf_boolean_t need_gfid_self_heal;
-        gf_boolean_t need_missing_entry_self_heal;
-
-        gf_boolean_t forced_merge;        /* Is this a self-heal triggered to
-                                             forcibly merge the directories? */
-
-        gf_boolean_t healing_fd_opened;   /* true if caller has already
-                                             opened fd */
-
-        gf_boolean_t data_lock_held;      /* true if caller has already
-                                             acquired 0-0 lock */
-
-        fd_t *healing_fd;                 /* set if callers has opened fd */
-
-        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);
-
-        /* 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;
-        int32_t impunge_ret_child;
-
-        /* 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;
-
-        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;
-        int *success;
-        unsigned char *locked_nodes;
-        int lock_count;
-
-        mode_t impunging_entry_mode;
-        const char *linkname;
 
-        int   op_failed;
-
-        int   file_has_holes;
-        blksize_t block_size;
-        off_t file_size;
-        off_t offset;
-        afr_post_remove_call_t post_remove_call;
-
-        loc_t parent_loc;
-
-        call_frame_t *orig_frame;
-        gf_boolean_t unwound;
-
-        /* private data for the particular self-heal algorithm */
-        void *private;
-
-        int (*flush_self_heal_cbk) (call_frame_t *frame, xlator_t *this);
-
-        int (*completion_cbk) (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;
-} afr_self_heal_t;
+	/* @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;
+
+        gf_boolean_t           choose_local;
+        gf_boolean_t           did_discovery;
+        uint64_t               sh_readdir_size;
+        gf_boolean_t           ensure_durability;
+        char                   *sh_domain;
+	char                   *afr_dirty;
+
+	afr_self_heald_t       shd;
+
+	/* pump dependencies */
+	void                   *pump_private;
+	gf_boolean_t           use_afr_in_pump;
+} afr_private_t;
 
 
 typedef enum {
@@ -292,11 +185,31 @@ afr_index_for_transaction_type (afr_transaction_type type)
         return -1;  /* make gcc happy */
 }
 
+typedef struct {
+        loc_t                   loc;
+        char                    *basename;
+        unsigned char           *locked_nodes;
+        int                     locked_count;
+
+} afr_entry_lockee_t;
+
+int
+afr_entry_lockee_cmp (const void *l1, const void *l2);
+
+typedef struct {
+        char    *domain; /* Domain on which inodelk is taken */
+        struct gf_flock flock;
+        unsigned char *locked_nodes;
+        int32_t lock_count;
+} afr_inodelk_t;
 
 typedef struct {
         loc_t *lk_loc;
-        struct gf_flock lk_flock;
 
+        int                     lockee_count;
+        afr_entry_lockee_t      lockee[AFR_LOCKEE_COUNT_MAX];
+
+        afr_inodelk_t       inodelk[AFR_DOM_COUNT_MAX];
         const char *lk_basename;
         const char *lower_basename;
         const char *higher_basename;
@@ -305,86 +218,203 @@ typedef struct {
 
         unsigned char *locked_nodes;
         unsigned char *lower_locked_nodes;
-        unsigned char *inode_locked_nodes;
-        unsigned char *entry_locked_nodes;
 
         selfheal_lk_type_t selfheal_lk_type;
         transaction_lk_type_t transaction_lk_type;
 
         int32_t lock_count;
-        int32_t inodelk_lock_count;
         int32_t entrylk_lock_count;
 
         uint64_t lock_number;
         int32_t lk_call_count;
+        int32_t lk_expected_count;
+        int32_t lk_attempted_count;
 
         int32_t lock_op_ret;
         int32_t lock_op_errno;
+        afr_lock_cbk_t lock_cbk;
+        char *domain; /* Domain on which inode/entry lock/unlock in progress.*/
+} afr_internal_lock_t;
 
-        int (*lock_cbk) (call_frame_t *, xlator_t *);
+struct afr_reply {
+	int	valid;
+	int32_t	op_ret;
+	int32_t	op_errno;
+	dict_t *xdata;
+	struct iatt poststat;
+	struct iatt postparent;
+	struct iatt prestat;
+	struct iatt preparent;
+	struct iatt preparent2;
+	struct iatt postparent2;
+	uint8_t checksum[MD5_DIGEST_LENGTH];
+};
 
-} afr_internal_lock_t;
+typedef enum {
+        AFR_FD_NOT_OPENED,
+        AFR_FD_OPENED,
+        AFR_FD_OPENING
+} afr_fd_open_status_t;
+
+typedef struct {
+        unsigned int *pre_op_done[AFR_NUM_CHANGE_LOGS];
+	int inherited[AFR_NUM_CHANGE_LOGS];
+	int on_disk[AFR_NUM_CHANGE_LOGS];
+        afr_fd_open_status_t *opened_on; /* which subvolumes the fd is open on */
+
+        unsigned int *lock_piggyback;
+        unsigned int *lock_acquired;
+
+        int flags;
+
+	/* used for delayed-post-op optimization */
+	pthread_mutex_t    delay_lock;
+	gf_timer_t        *delay_timer;
+	call_frame_t      *delay_frame;
+
+	/* 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;
+
+	/* @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;
+
+
+	/* list of frames currently in progress */
+	struct list_head  eager_locked;
+} afr_fd_ctx_t;
 
-typedef struct _afr_locked_fd {
-        fd_t  *fd;
-        struct list_head list;
-} afr_locked_fd_t;
 
 typedef struct _afr_local {
-        int     uid;
-        int     gid;
+	glusterfs_fop_t  op;
         unsigned int call_count;
-        unsigned int success_count;
-        unsigned int enoent_count;
-
 
-        unsigned int govinda_gOvinda;
+	/* @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;
 
-        unsigned int read_child_index;
-        unsigned char read_child_returned;
-        unsigned int first_up_child;
+        uint32_t     open_fd_count;
+        gf_boolean_t update_open_fd_count;
 
-        pid_t saved_pid;
+	gf_lkowner_t  saved_lk_owner;
 
         int32_t op_ret;
         int32_t op_errno;
 
         int32_t **pending;
 
+	int dirty[AFR_NUM_CHANGE_LOGS];
+
         loc_t loc;
         loc_t newloc;
 
         fd_t *fd;
+	afr_fd_ctx_t *fd_ctx;
 
-        glusterfs_fop_t fop;
-
+	/* @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;
-        int32_t       *fresh_children; //in the order of response
 
-        int32_t *child_errno;
+	/* @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;
 
-        dict_t  *xattr_req;
+	/* @readfn:
 
-        int32_t  inodelk_count;
-        int32_t  entrylk_count;
+	   pointer to function which will perform the read operation on a given
+	   subvolume. Used in read transactions.
+	*/
 
-        afr_internal_lock_t internal_lock;
+	afr_read_txn_wind_t readfn;
+
+	/* @refreshed:
+
+	   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;
+
+	/* @inode:
+
+	   the inode on which the read txn is performed on. ref'ed and copied
+	   from either fd->inode or loc.inode
+	*/
+
+	inode_t *inode;
+
+	/* @parent[2]:
+
+	   parent inode[s] on which directory transactions are performed.
+	*/
 
-        afr_locked_fd_t *locked_fd;
-        int32_t          source_child;
-        int32_t          lock_recovery_child;
+	inode_t *parent;
+	inode_t *parent2;
+
+	/* @readable:
+
+	   array of flags representing servers from which a read can be
+	   performed. This is the output of afr_inode_refresh()
+	*/
+	unsigned char *readable;
+
+	afr_inode_refresh_cbk_t refreshfn;
+
+	/* @refreshinode:
+
+	   Inode currently getting refreshed.
+	*/
+	inode_t *refreshinode;
+
+	/*
+	  @pre_op_compat:
+
+	  compatibility mode of pre-op. send a separate pre-op and
+	  op operations as part of transaction, rather than combining
+	*/
+
+	gf_boolean_t pre_op_compat;
+
+        dict_t  *xattr_req;
+
+        afr_internal_lock_t internal_lock;
 
         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;
 
-        int (*openfd_flush_cbk) (call_frame_t *frame, xlator_t *this);
+	/* 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
         */
 
-        int   op;
         struct {
                 struct {
                         unsigned char buf_set;
@@ -392,25 +422,7 @@ typedef struct _afr_local {
                 } statfs;
 
                 struct {
-                        uuid_t  gfid_req;
-                        inode_t *inode;
-                        struct iatt buf;
-                        struct iatt postparent;
-                        ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
-                        dict_t **xattrs;
-                        dict_t *xattr;
-                        struct iatt *postparents;
-                        struct iatt *bufs;
-                        int32_t read_child;
-                        int32_t *sources;
-                        int32_t *success_children;
-                } lookup;
-
-                struct {
                         int32_t flags;
-                        int32_t wbflags;
                 } open;
 
                 struct {
@@ -429,31 +441,28 @@ typedef struct _afr_local {
 
                 struct {
                         int last_index;
-                        ino_t ino;
                 } stat;
 
                 struct {
                         int last_index;
-                        ino_t ino;
                 } fstat;
 
                 struct {
                         size_t size;
                         int last_index;
-                        ino_t ino;
                 } readlink;
 
                 struct {
                         char *name;
                         int last_index;
-                        long pathinfo_len;
+                        long xattr_len;
                 } getxattr;
 
                 struct {
-                        ino_t ino;
                         size_t size;
                         off_t offset;
                         int last_index;
+                        uint32_t flags;
                 } readv;
 
                 /* dir read */
@@ -471,59 +480,43 @@ typedef struct _afr_local {
                         int32_t op_errno;
                         size_t size;
                         off_t offset;
-
+                        dict_t *dict;
                         gf_boolean_t failed;
                         int last_index;
                 } readdir;
                 /* inode write */
 
                 struct {
-                        ino_t ino;
                         struct iatt prebuf;
                         struct iatt postbuf;
+                } inode_wfop; //common structure for all inode-write-fops
 
+                struct {
                         int32_t op_ret;
 
                         struct iovec *vector;
                         struct iobref *iobref;
                         int32_t count;
                         off_t offset;
+                        uint32_t flags;
                 } writev;
 
                 struct {
-                        ino_t ino;
-                        struct iatt prebuf;
-                        struct iatt postbuf;
-                } fsync;
-
-                struct {
-                        ino_t ino;
                         off_t offset;
-                        struct iatt prebuf;
-                        struct iatt postbuf;
                 } truncate;
 
                 struct {
-                        ino_t ino;
                         off_t offset;
-                        struct iatt prebuf;
-                        struct iatt postbuf;
                 } ftruncate;
 
                 struct {
-                        ino_t ino;
                         struct iatt in_buf;
                         int32_t valid;
-                        struct iatt preop_buf;
-                        struct iatt postop_buf;
                 } setattr;
 
                 struct {
-                        ino_t ino;
                         struct iatt in_buf;
                         int32_t valid;
-                        struct iatt preop_buf;
-                        struct iatt postop_buf;
                 } fsetattr;
 
                 struct {
@@ -532,116 +525,87 @@ typedef struct _afr_local {
                 } 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 {
-                        ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
-                        fd_t *fd;
-                        dict_t *params;
-                        int32_t flags;
-                        mode_t mode;
                         inode_t *inode;
                         struct iatt buf;
                         struct iatt preparent;
                         struct iatt postparent;
-                        struct iatt read_child_buf;
+                        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 {
-                        ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
                         dev_t dev;
                         mode_t mode;
                         dict_t *params;
-                        inode_t *inode;
-                        struct iatt buf;
-                        struct iatt preparent;
-                        struct iatt postparent;
-                        struct iatt read_child_buf;
                 } mknod;
 
                 struct {
-                        ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
                         int32_t mode;
                         dict_t *params;
-                        inode_t *inode;
-                        struct iatt buf;
-                        struct iatt read_child_buf;
-                        struct iatt preparent;
-                        struct iatt postparent;
                 } mkdir;
 
                 struct {
-                        ino_t parent_ino;
-                        int32_t op_ret;
-                        int32_t op_errno;
-                        struct iatt preparent;
-                        struct iatt postparent;
-                } unlink;
-
-                struct {
-                        int   flags;
-                        ino_t parent_ino;
-                        int32_t op_ret;
-                        int32_t op_errno;
-                        struct iatt preparent;
-                        struct iatt postparent;
+                        int flags;
                 } rmdir;
 
                 struct {
-                        ino_t oldparent_ino;
-                        ino_t newparent_ino;
-                        ino_t ino;
-                        struct iatt buf;
-                        struct iatt read_child_buf;
-                        struct iatt preoldparent;
-                        struct iatt prenewparent;
-                        struct iatt postoldparent;
-                        struct iatt postnewparent;
-                } rename;
-
-                struct {
-                        ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
-                        inode_t *inode;
-                        struct iatt buf;
-                        struct iatt read_child_buf;
-                        struct iatt preparent;
-                        struct iatt postparent;
-                } link;
-
-                struct {
-                        ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
-                        inode_t *inode;
                         dict_t *params;
-                        struct iatt buf;
-                        struct iatt read_child_buf;
                         char *linkpath;
-                        struct iatt preparent;
-                        struct iatt postparent;
                 } symlink;
 
+		struct {
+			int32_t mode;
+			off_t offset;
+			size_t len;
+		} fallocate;
+
+		struct {
+			off_t offset;
+			size_t len;
+		} discard;
+
                 struct {
-                        int32_t flags;
-                        dir_entry_t *entries;
-                        int32_t count;
-                } setdents;
+                        off_t offset;
+                        off_t len;
+                        struct iatt prebuf;
+                        struct iatt postbuf;
+                } zerofill;
+
+
         } cont;
 
         struct {
                 off_t start, len;
 
+                gf_boolean_t    eager_lock_on;
+                int *eager_lock;
+
                 char *basename;
                 char *new_basename;
 
@@ -650,15 +614,67 @@ typedef struct _afr_local {
 
                 afr_transaction_type type;
 
-                int success_count;
-                int erase_pending;
-                int failure_count;
+		/* stub to resume on destruction
+		   of the transaction frame */
+		call_stub_t      *resume_stub;
+
+		struct list_head  eager_locked;
+
+                unsigned char   *pre_op;
+
+		/* @fop_subvols: subvolumes on which FOP will be attempted */
+                unsigned char   *fop_subvols;
+
+		/* @failed_subvols: subvolumes on which FOP failed. Always
+		   a subset of @fop_subvols */
+                unsigned char   *failed_subvols;
+
+		/* @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;
 
-                int last_tried;
-                int32_t *child_errno;
+		/* @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;
+
+		/* @changelog_resume: function to be called after changlogging
+		   (either pre-op or post-op) is done
+		*/
+
+		afr_changelog_resume_t changelog_resume;
 
                 call_frame_t *main_frame;
 
+                int (*wind) (call_frame_t *frame, xlator_t *this, int subvol);
+
                 int (*fop) (call_frame_t *frame, xlator_t *this);
 
                 int (*done) (call_frame_t *frame, xlator_t *this);
@@ -670,85 +686,96 @@ typedef struct _afr_local {
                 /* post-op hook */
         } transaction;
 
-        afr_self_heal_t self_heal;
+	syncbarrier_t barrier;
 
         struct marker_str     marker;
-} afr_local_t;
 
+        /* extra data for fops */
+        dict_t         *xdata_req;
+        dict_t         *xdata_rsp;
 
-typedef struct {
-        unsigned int *pre_op_done;
-        unsigned int *opened_on;     /* which subvolumes the fd is open on */
-        unsigned int *pre_op_piggyback;
+        mode_t          umask;
+        int             xflag;
+        gf_boolean_t    do_discovery;
+	struct afr_reply *replies;
+} afr_local_t;
 
-        int flags;
-        int32_t wbflags;
-        uint64_t up_count;   /* number of CHILD_UPs this fd has seen */
-        uint64_t down_count; /* number of CHILD_DOWNs this fd has seen */
 
-        int32_t last_tried;
+/* 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)))
 
-        int  hit, miss;
-        gf_boolean_t failed_over;
-        struct list_head entries; /* needed for readdir failover */
+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_inode_read_subvol_get (inode_t *inode, xlator_t *this,
+			     unsigned char *data_subvols,
+			     unsigned char *metadata_subvols,
+			     int *event_generation);
 
-        unsigned char *locked_on; /* which subvolumes locks have been successful */
-} afr_fd_ctx_t;
+int
+__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_inode_read_subvol_reset (inode_t *inode, xlator_t *this);
 
-/* try alloc and if it fails, goto label */
-#define ALLOC_OR_GOTO(var, type, label) do {                    \
-                var = GF_CALLOC (sizeof (type), 1,              \
-                                 gf_afr_mt_##type);             \
-                if (!var) {                                     \
-                        gf_log (this->name, GF_LOG_ERROR,       \
-                                "out of memory :(");            \
-                        op_errno = ENOMEM;                      \
-                        goto label;                             \
-                }                                               \
-        } while (0);
+int
+afr_read_subvol_select_by_policy (inode_t *inode, xlator_t *this,
+				  unsigned char *readable);
 
+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,
+		     int *event_p, afr_transaction_type type);
 
-/* 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)))
+#define afr_data_subvol_get(i, t, s, e) \
+	afr_read_subvol_get(i, t, s, e, AFR_DATA_TRANSACTION)
 
-#define afr_fop_failed(op_ret, op_errno) ((op_ret) == -1)
+#define afr_metadata_subvol_get(i, t, s, e) \
+	afr_read_subvol_get(i, t, s, e, AFR_METADATA_TRANSACTION)
 
-/* have we tried all children? */
-#define all_tried(i, count)  ((i) == (count) - 1)
+int
+afr_inode_refresh (call_frame_t *frame, xlator_t *this, inode_t *inode,
+		   afr_inode_refresh_cbk_t cbk);
 
 int32_t
-afr_set_dict_gfid (dict_t *dict, uuid_t gfid);
+afr_notify (xlator_t *this, int32_t event, void *data, void *data2);
 
 int
-pump_command_reply (call_frame_t *frame, xlator_t *this);
+afr_init_entry_lockee (afr_entry_lockee_t *lockee, afr_local_t *local,
+                       loc_t *loc, char *basename, int child_count);
 
-int32_t
-afr_notify (xlator_t *this, int32_t event,
-            void *data, ...);
+void
+afr_entry_lockee_cleanup (afr_internal_lock_t *int_lock);
 
 int
 afr_attempt_lock_recovery (xlator_t *this, int32_t child_index);
 
 int
-afr_save_locked_fd (xlator_t *this, fd_t *fd);
-
-int
 afr_mark_locked_nodes (xlator_t *this, fd_t *fd,
                        unsigned char *locked_nodes);
 
 void
-afr_set_lk_owner (call_frame_t *frame, xlator_t *this);
+afr_set_lk_owner (call_frame_t *frame, xlator_t *this, void *lk_owner);
 
 int
 afr_set_lock_number (call_frame_t *frame, xlator_t *this);
 
-
-loc_t *
-lower_path (loc_t *l1, const char *b1, loc_t *l2, const char *b2);
-
 int32_t
 afr_unlock (call_frame_t *frame, xlator_t *this);
 
@@ -764,33 +791,30 @@ afr_blocking_lock (call_frame_t *frame, xlator_t *this);
 int
 afr_internal_lock_finish (call_frame_t *frame, xlator_t *this);
 
+int
+afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src, char *dom,
+                          unsigned int child_count);
 
-int pump_start (call_frame_t *frame, xlator_t *this);
+int
+__afr_fd_ctx_set (xlator_t *this, fd_t *fd);
 
 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);
-
-void
-afr_inode_set_read_ctx (xlator_t *this, inode_t *inode, int32_t read_child,
-                        int32_t *fresh_children);
+afr_fd_ctx_t *
+afr_fd_ctx_get (fd_t *fd, xlator_t *this);
 
-void
-afr_build_parent_loc (loc_t *parent, loc_t *child);
+int
+afr_build_parent_loc (loc_t *parent, loc_t *child, int32_t *op_errno);
 
 int
-afr_up_children_count (int child_count, unsigned char *child_up);
+afr_locked_nodes_count (unsigned char *locked_nodes, int child_count);
 
-gf_boolean_t
-afr_is_fresh_lookup (loc_t *loc, xlator_t *this);
+int
+afr_replies_interpret (call_frame_t *frame, xlator_t *this, inode_t *inode);
 
 void
-afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *postparent);
-
-int
-afr_locked_nodes_count (unsigned char *locked_nodes, int child_count);
+afr_replies_wipe (afr_local_t *local, afr_private_t *priv);
 
 void
 afr_local_cleanup (afr_local_t *local, xlator_t *this);
@@ -798,21 +822,9 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this);
 int
 afr_frame_return (call_frame_t *frame);
 
-uint64_t
-afr_is_split_brain (xlator_t *this, inode_t *inode);
-
-void
-afr_set_split_brain (xlator_t *this, inode_t *inode, gf_boolean_t set);
-
 int
 afr_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-          fd_t *fd, int32_t wbflags);
-
-void
-afr_set_opendir_done (xlator_t *this, inode_t *inode);
-
-gf_boolean_t
-afr_is_opendir_done (xlator_t *this, inode_t *inode);
+          fd_t *fd, dict_t *xdata);
 
 void
 afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this);
@@ -820,9 +832,6 @@ afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this);
 int
 afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd);
 
-int
-afr_openfd_flush (call_frame_t *frame, xlator_t *this, fd_t *fd);
-
 #define AFR_STACK_UNWIND(fop, frame, params ...)                \
         do {                                                    \
                 afr_local_t *__local = NULL;                    \
@@ -833,22 +842,36 @@ afr_openfd_flush (call_frame_t *frame, xlator_t *this, fd_t *fd);
                         frame->local = NULL;                    \
                 }                                               \
                 STACK_UNWIND_STRICT (fop, frame, params);       \
-                afr_local_cleanup (__local, __this);            \
-                GF_FREE (__local);                              \
-        } while (0);
+                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);            \
-                afr_local_cleanup (__local, __this);    \
-                GF_FREE (__local);                      \
+#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, THIS->private, &op_errno)) { \
+		afr_local_cleanup (frame->local, THIS);		       \
+		mem_put (frame->local);				       \
+		frame->local = NULL; };				       \
+	frame->local;})
+
+#define AFR_STACK_RESET(frame) do { int opr; STACK_RESET (frame->root); AFR_FRAME_INIT(frame, opr);} while (0)
+
 /* allocate and return a string that is the basename of argument */
 static inline char *
 AFR_BASENAME (const char *str)
@@ -861,84 +884,93 @@ AFR_BASENAME (const char *str)
         return __basename_str;
 }
 
+call_frame_t *
+afr_copy_frame (call_frame_t *base);
+
 int
-afr_transaction_local_init (afr_local_t *local, afr_private_t *priv);
+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 );
 
-int32_t *
-afr_fresh_children_create (int32_t child_count);
-
 int
-AFR_LOCAL_INIT (afr_local_t *local, afr_private_t *priv);
+afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno);
 
 int
 afr_internal_lock_init (afr_internal_lock_t *lk, size_t child_count,
                         transaction_lk_type_t lk_type);
 
 int
-afr_first_up_child (unsigned char *child_up, size_t child_count);
+afr_higher_errno (int32_t old_errno, int32_t new_errno);
+
+int
+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);
+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);
+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 *fresh_children, unsigned int child_count);
-void
-afr_fresh_children_add_child (int32_t *fresh_children, int32_t child,
-                              int32_t child_count);
 void
-afr_reset_children (int32_t *fresh_children, int32_t child_count);
-gf_boolean_t
-afr_error_more_important (int32_t old_errno, int32_t new_errno);
-int
-afr_errno_count (int32_t *children, int *child_errno,
-                 unsigned int child_count, int32_t op_errno);
+afr_set_low_priority (call_frame_t *frame);
 int
-afr_get_children_count (int32_t *fresh_children, unsigned int child_count);
+afr_child_fd_ctx_set (xlator_t *this, fd_t *fd, int32_t child,
+                      int flags);
+
 gf_boolean_t
-afr_is_child_present (int32_t *success_children, int32_t child_count,
-                      int32_t child);
+afr_have_quorum (char *logname, afr_private_t *priv);
+
 void
-afr_update_gfid_from_iatts (uuid_t uuid, struct iatt *bufs,
-                            int32_t *success_children,
-                            unsigned int child_count);
+afr_matrix_cleanup (int32_t **pending, unsigned int m);
+
+int32_t**
+afr_matrix_create (unsigned int m, unsigned int n);
+
 void
-afr_reset_xattr (dict_t **xattr, unsigned int child_count);
+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
+
+/*
+ * 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);
+
+int
+afr_fd_report_unstable_write (xlator_t *this, fd_t *fd);
+
 gf_boolean_t
-afr_conflicting_iattrs (struct iatt *bufs, int32_t *success_children,
-                        unsigned int child_count, const char *path,
-                        const char *xlator_name);
+afr_fd_has_witnessed_unstable_write (xlator_t *this, fd_t *fd);
+
+void
+afr_delayed_changelog_wake_resume (xlator_t *this, fd_t *fd, call_stub_t *stub);
+
 int
-afr_gfid_missing_count (const char *xlator_name, int32_t *children,
-                        struct iatt *bufs, unsigned int child_count,
-                        const char *path);
+afr_inodelk_init (afr_inodelk_t *lk, char *dom, size_t child_count);
+
 void
-afr_xattr_req_prepare (xlator_t *this, dict_t *xattr_req, const char *path);
+afr_handle_open_fd_count (call_frame_t *frame, xlator_t *this);
+
+int
+afr_local_pathinfo (char *pathinfo, gf_boolean_t *is_local);
+
 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);
-int32_t
-afr_resultant_errno_get (int32_t *children,
-                         int *child_errno, unsigned int child_count);
+afr_remove_eager_lock_stub (afr_local_t *local);
 #endif /* __AFR_H__ */
diff --git a/xlators/cluster/afr/src/pump.c b/xlators/cluster/afr/src/pump.c
index 3adac9502..eed509956 100644
--- a/xlators/cluster/afr/src/pump.c
+++ b/xlators/cluster/afr/src/pump.c
@@ -1,25 +1,17 @@
 /*
-   Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  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
@@ -28,8 +20,130 @@
 
 #include "afr-common.c"
 #include "defaults.c"
+#include "glusterfs.h"
+#include "pump.h"
+
+
+static int
+afr_set_dict_gfid (dict_t *dict, uuid_t gfid)
+{
+        int ret       = 0;
+        uuid_t *pgfid = NULL;
+
+        GF_ASSERT (gfid);
+
+        pgfid = GF_CALLOC (1, sizeof (uuid_t), gf_common_mt_char);
+        if (!pgfid) {
+                ret = -1;
+                goto out;
+        }
+
+        uuid_copy (*pgfid, gfid);
+
+        ret = dict_set_dynptr (dict, "gfid-req", pgfid, sizeof (uuid_t));
+        if (ret)
+                gf_log (THIS->name, GF_LOG_ERROR, "gfid set failed");
+
+out:
+        if (ret && pgfid)
+                GF_FREE (pgfid);
+	return ret;
+}
+
+static int
+afr_set_root_gfid (dict_t *dict)
+{
+        uuid_t gfid;
+        int ret = 0;
+
+        memset (gfid, 0, 16);
+        gfid[15] = 1;
+
+        ret = afr_set_dict_gfid (dict, gfid);
+
+        return ret;
+}
+
+static int
+afr_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name)
+{
+        int   ret = -1;
+        uuid_t pargfid = {0};
+
+        if (!child)
+                goto out;
+
+        if (!uuid_is_null (parent->inode->gfid))
+                uuid_copy (pargfid, parent->inode->gfid);
+        else if (!uuid_is_null (parent->gfid))
+                uuid_copy (pargfid, parent->gfid);
+
+        if (uuid_is_null (pargfid))
+                goto out;
+
+        if (strcmp (parent->path, "/") == 0)
+                ret = gf_asprintf ((char **)&child->path, "/%s", name);
+        else
+                ret = gf_asprintf ((char **)&child->path, "%s/%s", parent->path,
+                                   name);
+
+        if (-1 == ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "asprintf failed while setting child path");
+        }
+
+        child->name = strrchr (child->path, '/');
+        if (child->name)
+                child->name++;
+
+        child->parent = inode_ref (parent->inode);
+        child->inode = inode_new (parent->inode->table);
+        uuid_copy (child->pargfid, pargfid);
+
+        if (!child->inode) {
+                ret = -1;
+                goto out;
+        }
+
+        ret = 0;
+out:
+        if ((ret == -1) && child)
+                loc_wipe (child);
+
+        return ret;
+}
+
+static void
+afr_build_root_loc (xlator_t *this, loc_t *loc)
+{
+        afr_private_t   *priv = NULL;
+
+        priv = this->private;
+        loc->path = gf_strdup ("/");
+        loc->name = "";
+        loc->inode = inode_ref (priv->root_inode);
+        uuid_copy (loc->gfid, loc->inode->gfid);
+}
+
+static void
+afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *postparent)
+{
+        GF_ASSERT (loc);
+        GF_ASSERT (buf);
+
+        uuid_copy (loc->gfid, buf->ia_gfid);
+        if (postparent)
+                uuid_copy (loc->pargfid, postparent->ia_gfid);
+}
 
 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)
 {
@@ -140,85 +254,17 @@ pump_set_resume_path (xlator_t *this, const char *path)
 
         LOCK (&pump_priv->resume_path_lock);
         {
-                pump_priv->resume_path = strdup (path);
-                if (!pump_priv->resume_path)
-                        ret = -1;
+                strncpy (pump_priv->resume_path, path, strlen (path) + 1);
         }
         UNLOCK (&pump_priv->resume_path_lock);
 
         return ret;
 }
 
-static void
-build_child_loc (loc_t *parent, loc_t *child, char *path, char *name)
-{
-        child->path = path;
-        child->name = name;
-
-        child->parent = inode_ref (parent->inode);
-        child->inode = inode_new (parent->inode->table);
-}
-
-static char *
-build_file_path (loc_t *loc, gf_dirent_t *entry)
-{
-        xlator_t *this = NULL;
-        char *file_path = NULL;
-        int pathlen = 0;
-        int total_size = 0;
-
-        this = THIS;
-
-        pathlen = STRLEN_0 (loc->path);
-
-        if (IS_ROOT_PATH (loc->path)) {
-                total_size = pathlen + entry->d_len;
-                file_path = GF_CALLOC (1, total_size, gf_afr_mt_char);
-                if (!file_path) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        return NULL;
-                }
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "constructing file path of size=%d"
-                        "pathlen=%d, d_len=%d",
-                        total_size, pathlen,
-                        entry->d_len);
-
-                snprintf(file_path, total_size, "%s%s", loc->path, entry->d_name);
-
-        } else {
-                total_size = pathlen + entry->d_len + 1; /* for the extra '/' in the path */
-                file_path = GF_CALLOC (1, total_size + 1, gf_afr_mt_char);
-                if (!file_path) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        return NULL;
-                }
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "constructing file path of size=%d"
-                        "pathlen=%d, d_len=%d",
-                        total_size, pathlen,
-                        entry->d_len);
-
-                snprintf(file_path, total_size, "%s/%s", loc->path, entry->d_name);
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "path=%s and d_name=%s", loc->path, entry->d_name);
-        gf_log (this->name, GF_LOG_TRACE,
-                "constructed file_path=%s of size=%d", file_path, total_size);
-
-        return file_path;
-}
-
 static int
 pump_save_path (xlator_t *this, const char *path)
 {
         afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
         pump_state_t state;
         dict_t *dict = NULL;
         loc_t  loc = {0};
@@ -230,29 +276,30 @@ pump_save_path (xlator_t *this, const char *path)
                 return 0;
 
         priv = this->private;
-        pump_priv = priv->pump_private;
 
         GF_ASSERT (priv->root_inode);
 
-        build_root_loc (priv->root_inode, &loc);
+        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_DEBUG,
+                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);
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Saving path for status info");
         }
 
         dict_unref (dict);
 
+        loc_wipe (&loc);
         return 0;
 }
 
@@ -315,15 +362,9 @@ pump_get_resume_path (xlator_t *this)
 static int
 pump_update_resume_state (xlator_t *this, const char *path)
 {
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
         pump_state_t state;
         const char *resume_path = NULL;
 
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
         state = pump_get_state ();
 
         if (state == PUMP_STATE_RESUME) {
@@ -351,16 +392,10 @@ pump_update_resume_state (xlator_t *this, const char *path)
 static gf_boolean_t
 is_pump_traversal_allowed (xlator_t *this, const char *path)
 {
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
         pump_state_t state;
         const char *resume_path = NULL;
         gf_boolean_t ret = _gf_true;
 
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
         state = pump_get_state ();
 
         if (state == PUMP_STATE_RESUME) {
@@ -403,26 +438,22 @@ pump_save_file_stats (xlator_t *this, const char *path)
 static int
 gf_pump_traverse_directory (loc_t *loc)
 {
-        xlator_t *this = NULL;
-        afr_private_t *priv = NULL;
-        fd_t     *fd   = NULL;
-
-        off_t       offset   = 0;
-        loc_t       entry_loc;
-        gf_dirent_t *entry = NULL;
-        gf_dirent_t *tmp = NULL;
-        gf_dirent_t entries;
-
-	struct iatt iatt, parent;
-	dict_t *xattr_rsp;
-
-        char *file_path = NULL;
-        int ret = 0;
-        gf_boolean_t is_directory_empty = _gf_true;
+        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;
-        priv = this->private;
 
         GF_ASSERT (loc->inode);
 
@@ -444,7 +475,8 @@ gf_pump_traverse_directory (loc_t *loc)
                 "pump opendir on %s returned=%d",
                 loc->path, ret);
 
-        while (syncop_readdirp (this, fd, 131072, offset, &entries)) {
+        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,
@@ -456,110 +488,108 @@ gf_pump_traverse_directory (loc_t *loc)
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "found readdir entry=%s", entry->d_name);
 
-                        file_path = build_file_path (loc, entry);
-                        if (!file_path) {
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "file path construction failed");
-                                goto out;
+                        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;
 
-                        build_child_loc (loc, &entry_loc, file_path, entry->d_name);
-
-                        if (!IS_ENTRY_CWD (entry->d_name) &&
-                                           !IS_ENTRY_PARENT (entry->d_name)) {
-
-                                    is_directory_empty = _gf_false;
-                                    ret = syncop_lookup (this, &entry_loc, NULL,
-                                                         &iatt, &xattr_rsp, &parent);
+			if ((strcmp (entry->d_name, ".") == 0) ||
+			    (strcmp (entry->d_name, "..") == 0))
+				continue;
 
-                                    memcpy (entry_loc.inode->gfid, iatt.ia_gfid, 16);
+			is_directory_empty = _gf_false;
+			gf_log (this->name, GF_LOG_DEBUG,
+				"lookup %s => %"PRId64,
+				entry_loc.path,
+				iatt.ia_ino);
 
-                                    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);
 
-                                    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;
+			}
 
+			ret = afr_selfheal_name (this, loc->gfid, entry->d_name);
+			if (ret) {
+				gf_log (this->name, GF_LOG_ERROR,
+					"%s: name self-heal failed (%s/%s)",
+					entry_loc.path, uuid_utoa (loc->gfid),
+					entry->d_name);
+				continue;
+			}
 
-                                    gf_log (this->name, GF_LOG_DEBUG,
-                                            "second lookup ret=%d: %s => %"PRId64,
-                                            ret,
-                                            entry_loc.path,
-                                            iatt.ia_ino);
+			ret = afr_selfheal (this, iatt.ia_gfid);
+			if (ret) {
+				gf_log (this->name, GF_LOG_ERROR,
+					"%s: self-heal failed (%s)",
+					entry_loc.path, uuid_utoa (iatt.ia_gfid));
+				continue;
+			}
 
-                                    pump_update_resume_state (this, entry_loc.path);
+			pump_fill_loc_info (&entry_loc, &iatt, &parent);
 
-                                    pump_save_path (this, entry_loc.path);
-                                    pump_save_file_stats (this, entry_loc.path);
+			pump_update_resume_state (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;
-                                    }
+			pump_save_path (this, entry_loc.path);
+			pump_save_file_stats (this, entry_loc.path);
 
-                                    gf_log (this->name, GF_LOG_TRACE,
-                                            "type of file=%d, IFDIR=%d",
-                                            iatt.ia_type, IA_IFDIR);
+			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);
-                                            }
-                                    }
-                            }
-                        offset = entry->d_off;
-                        loc_wipe (&entry_loc);
+			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);
-                gf_log (this->name, GF_LOG_TRACE,
-                        "offset incremented to %d",
+                free_entries = _gf_false;
+                gf_log (this->name, GF_LOG_TRACE, "offset incremented to %d",
                         (int32_t ) offset);
 
         }
 
-        if (is_directory_empty && IS_ROOT_PATH (loc->path)) {
+        ret = syncop_close (fd);
+        if (ret < 0)
+                gf_log (this->name, GF_LOG_DEBUG, "closing the fd failed");
+
+        if (is_directory_empty && (strcmp (loc->path, "/") == 0)) {
                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;
-
-}
-
-void
-build_root_loc (inode_t *inode, loc_t *loc)
-{
-        loc->path = "/";
-        loc->name = "";
-        loc->inode = inode;
-        loc->ino = 1;
-        loc->inode->ino = 1;
-        memset (loc->inode->gfid, 0, 16);
-        loc->inode->gfid[15] = 1;
-
 }
 
 static int
 pump_update_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_get_resume_path (this);
 
         if (resume_path) {
@@ -580,10 +610,9 @@ pump_update_resume_path (xlator_t *this)
 
 static int32_t
 pump_xattr_cleaner (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno)
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
         afr_private_t  *priv      = NULL;
-        pump_private_t *pump_priv = NULL;
         loc_t           loc       = {0};
         int             i         = 0;
         int             ret       = 0;
@@ -591,24 +620,25 @@ pump_xattr_cleaner (call_frame_t *frame, void *cookie, xlator_t *this,
         int             sink      = 1;
 
         priv      = this->private;
-        pump_priv = priv->pump_private;
 
-        build_root_loc (priv->root_inode, &loc);
+        afr_build_root_loc (this, &loc);
 
         ret = syncop_removexattr (priv->children[source], &loc,
-                                          PUMP_PATH);
+				  PUMP_PATH, 0);
 
         ret = syncop_removexattr (priv->children[sink], &loc,
-                                  PUMP_SINK_COMPLETE);
+                                  PUMP_SINK_COMPLETE, 0);
 
         for (i = 0; i < priv->child_count; i++) {
                 ret = syncop_removexattr (priv->children[i], &loc,
-                                          PUMP_SOURCE_COMPLETE);
-                if (ret)
+                                          PUMP_SOURCE_COMPLETE, 0);
+                if (ret) {
                         gf_log (this->name, GF_LOG_DEBUG, "removexattr "
-                                "failed with %s", strerror (errno));
+                                "failed with %s", strerror (-ret));
+                }
         }
 
+        loc_wipe (&loc);
         return pump_command_reply (frame, this);
 }
 
@@ -628,7 +658,7 @@ pump_complete_migration (xlator_t *this)
 
         GF_ASSERT (priv->root_inode);
 
-        build_root_loc (priv->root_inode, &loc);
+        afr_build_root_loc (this, &loc);
 
         dict = dict_new ();
 
@@ -640,6 +670,10 @@ pump_complete_migration (xlator_t *this)
                 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) {
@@ -647,6 +681,10 @@ pump_complete_migration (xlator_t *this)
                                 "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) {
@@ -662,24 +700,11 @@ pump_complete_migration (xlator_t *this)
                 call_resume (pump_priv->cleaner);
         }
 
+        loc_wipe (&loc);
         return 0;
 }
 
 static int
-pump_set_root_gfid (dict_t *dict)
-{
-        uuid_t gfid;
-        int ret = 0;
-
-        memset (gfid, 0, 16);
-        gfid[15] = 1;
-
-        ret = afr_set_dict_gfid (dict, gfid);
-
-        return ret;
-}
-
-static int
 pump_lookup_sink (loc_t *loc)
 {
         xlator_t *this = NULL;
@@ -692,7 +717,7 @@ pump_lookup_sink (loc_t *loc)
 
         xattr_req = dict_new ();
 
-        ret = pump_set_root_gfid (xattr_req);
+        ret = afr_set_root_gfid (xattr_req);
         if (ret)
                 goto out;
 
@@ -702,6 +727,7 @@ pump_lookup_sink (loc_t *loc)
         if (ret) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "Lookup on sink child failed");
+                ret = -1;
                 goto out;
         }
 
@@ -731,7 +757,7 @@ pump_task (void *data)
 
         GF_ASSERT (priv->root_inode);
 
-        build_root_loc (priv->root_inode, &loc);
+        afr_build_root_loc (this, &loc);
         xattr_req = dict_new ();
         if (!xattr_req) {
                 gf_log (this->name, GF_LOG_DEBUG,
@@ -740,14 +766,13 @@ pump_task (void *data)
                 goto out;
         }
 
-        pump_set_root_gfid (xattr_req);
+        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: ino=%"PRId64", path=%s",
-                loc.ino,
-                loc.path);
+                "lookup: path=%s gfid=%s",
+                loc.path, uuid_utoa (loc.inode->gfid));
 
         ret = pump_check_and_update_status (this);
         if (ret < 0) {
@@ -756,7 +781,7 @@ pump_task (void *data)
 
         pump_update_resume_path (this);
 
-        pump_set_root_gfid (xattr_req);
+        afr_set_root_gfid (xattr_req);
         ret = pump_lookup_sink (&loc);
         if (ret) {
                 pump_update_resume_path (this);
@@ -770,6 +795,7 @@ out:
         if (xattr_req)
                 dict_unref (xattr_req);
 
+        loc_wipe (&loc);
 	return 0;
 }
 
@@ -779,12 +805,10 @@ pump_task_completion (int ret, call_frame_t *sync_frame, void *data)
 {
         xlator_t *this = NULL;
         afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
 
         this = THIS;
 
         priv = this->private;
-        pump_priv = priv->pump_private;
 
         inode_unref (priv->root_inode);
         STACK_DESTROY (sync_frame->root);
@@ -805,7 +829,7 @@ pump_start (call_frame_t *pump_frame, xlator_t *this)
 	priv = this->private;
         pump_priv = priv->pump_private;
 
-	pump_frame->root->lk_owner = (uint64_t) (unsigned long)pump_frame->root;
+        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,
@@ -819,8 +843,8 @@ pump_start (call_frame_t *pump_frame, xlator_t *this)
         }
 
         gf_log (this->name, GF_LOG_DEBUG,
-                "setting pump as started lk_owner: %"PRIu64" %"PRIu64,
-                pump_frame->root->lk_owner, pump_pid);
+                "setting pump as started lk_owner: %s %"PRIu64,
+                lkowner_utoa (&pump_frame->root->lk_owner), pump_pid);
 
         priv->use_afr_in_pump = 1;
 out:
@@ -854,7 +878,7 @@ pump_cmd_start_setxattr_cbk (call_frame_t *frame,
                              void *cookie,
                              xlator_t *this,
                              int32_t op_ret,
-                             int32_t op_errno)
+                             int32_t op_errno, dict_t *xdata)
 
 {
         call_frame_t *prev = NULL;
@@ -906,9 +930,9 @@ pump_initiate_sink_connect (call_frame_t *frame, xlator_t *this)
 
         GF_ASSERT (priv->root_inode);
 
-        build_root_loc (priv->root_inode, &loc);
+        afr_build_root_loc (this, &loc);
 
-        data = data_ref (dict_get (local->dict, PUMP_CMD_START));
+        data = data_ref (dict_get (local->dict, RB_PUMP_CMD_START));
         if (!data) {
                 ret = -1;
                 gf_log (this->name, GF_LOG_ERROR,
@@ -947,7 +971,7 @@ pump_initiate_sink_connect (call_frame_t *frame, xlator_t *this)
 		    PUMP_SINK_CHILD(this)->fops->setxattr,
 		    &loc,
 		    dict,
-		    0);
+		    0, NULL);
 
         ret = 0;
 
@@ -961,6 +985,7 @@ out:
         if (ret && clnt_cmd)
                 GF_FREE (clnt_cmd);
 
+        loc_wipe (&loc);
         return ret;
 }
 
@@ -980,7 +1005,7 @@ pump_cmd_start_getxattr_cbk (call_frame_t *frame,
                              xlator_t *this,
                              int32_t op_ret,
                              int32_t op_errno,
-                             dict_t *dict)
+                             dict_t *dict, dict_t *xdata)
 {
         afr_local_t *local = NULL;
         char *path = NULL;
@@ -1047,6 +1072,7 @@ pump_execute_status (call_frame_t *frame, xlator_t *this)
         uint64_t number_files = 0;
 
         char filename[PATH_MAX];
+        char summary[PATH_MAX+256];
         char *dict_str = NULL;
 
         int32_t op_ret = 0;
@@ -1075,16 +1101,19 @@ pump_execute_status (call_frame_t *frame, xlator_t *this)
         }
 
         if (pump_priv->pump_finished) {
-        snprintf (dict_str, PATH_MAX + 256, "Number of files migrated = %"PRIu64"        Migration complete ",
-                  number_files);
+                snprintf (summary, PATH_MAX+256,
+                          "no_of_files=%"PRIu64, number_files);
         } else {
-        snprintf (dict_str, PATH_MAX + 256, "Number of files migrated = %"PRIu64"       Current file= %s ",
-                  number_files, filename);
+                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, PUMP_CMD_STATUS, dict_str);
+        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");
@@ -1096,13 +1125,12 @@ pump_execute_status (call_frame_t *frame, xlator_t *this)
 
 out:
 
-        AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
+        AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, NULL);
 
         if (dict)
                 dict_unref (dict);
 
-        if (dict_str)
-                GF_FREE (dict_str);
+        GF_FREE (dict_str);
 
         return 0;
 }
@@ -1144,14 +1172,14 @@ pump_execute_start (call_frame_t *frame, xlator_t *this)
 
         GF_ASSERT (priv->root_inode);
 
-        build_root_loc (priv->root_inode, &loc);
+        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);
+		    PUMP_PATH, NULL);
 
         ret = 0;
 
@@ -1161,6 +1189,7 @@ out:
                 pump_command_reply (frame, this);
         }
 
+        loc_wipe (&loc);
 	return 0;
 }
 
@@ -1168,7 +1197,7 @@ static int
 pump_cleanup_helper (void *data) {
         call_frame_t *frame = data;
 
-        pump_xattr_cleaner (frame, 0, frame->this, 0, 0);
+        pump_xattr_cleaner (frame, 0, frame->this, 0, 0, NULL);
 
         return 0;
 }
@@ -1194,14 +1223,6 @@ pump_execute_commit (call_frame_t *frame, xlator_t *this)
         pump_priv = priv->pump_private;
         local     = frame->local;
 
-
-        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) {
                 pump_change_state (this, PUMP_STATE_COMMIT);
@@ -1258,7 +1279,7 @@ pump_execute_abort (call_frame_t *frame, xlator_t *this)
         } else {
                 pump_priv->cleaner = fop_setxattr_cbk_stub (frame,
                                                             pump_xattr_cleaner,
-                                                            0, 0);
+                                                            0, 0, NULL);
         }
 
         return 0;
@@ -1271,7 +1292,7 @@ pump_command_status (xlator_t *this, dict_t *dict)
         int dict_ret = -1;
         int ret = _gf_true;
 
-        dict_ret = dict_get_str (dict, PUMP_CMD_STATUS, &cmd);
+        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");
@@ -1295,7 +1316,7 @@ pump_command_pause (xlator_t *this, dict_t *dict)
         int dict_ret = -1;
         int ret = _gf_true;
 
-        dict_ret = dict_get_str (dict, PUMP_CMD_PAUSE, &cmd);
+        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");
@@ -1319,7 +1340,7 @@ pump_command_commit (xlator_t *this, dict_t *dict)
         int dict_ret = -1;
         int ret = _gf_true;
 
-        dict_ret = dict_get_str (dict, PUMP_CMD_COMMIT, &cmd);
+        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");
@@ -1343,7 +1364,7 @@ pump_command_abort (xlator_t *this, dict_t *dict)
         int dict_ret = -1;
         int ret = _gf_true;
 
-        dict_ret = dict_get_str (dict, PUMP_CMD_ABORT, &cmd);
+        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");
@@ -1367,7 +1388,7 @@ pump_command_start (xlator_t *this, dict_t *dict)
         int dict_ret = -1;
         int ret = _gf_true;
 
-        dict_ret = dict_get_str (dict, PUMP_CMD_START, &cmd);
+        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");
@@ -1384,135 +1405,22 @@ out:
 
 }
 
-struct _xattr_key {
-        char *key;
-        struct list_head list;
-};
-
-static void
-__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;
-
-                xkey->key = key;
-                INIT_LIST_HEAD (&xkey->list);
-
-                list_add_tail (&xkey->list, list);
-        }
-}
-
-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)
-{
-	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);
-	}
-
-out:
-	if (unwind) {
-                if (op_ret >= 0 && dict)
-                        __filter_xattrs (dict);
-
-		AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
-	}
-
-	return 0;
-}
-
-int32_t
-pump_getxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, const char *name)
+int
+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           op_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;
+	afr_private_t *priv = NULL;
+	int op_errno = 0;
+	int ret = 0;
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	frame->local = local;
+	priv = this->private;
 
-        op_ret = AFR_LOCAL_INIT (local, priv);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                goto out;
+        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;
         }
 
         if (name) {
@@ -1523,187 +1431,21 @@ pump_getxattr (call_frame_t *frame, xlator_t *this,
                         goto out;
                 }
 
-                if (!strcmp (name, PUMP_CMD_STATUS)) {
+                if (!strcmp (name, RB_PUMP_CMD_STATUS)) {
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "Hit pump command - status");
                         pump_execute_status (frame, this);
-                        op_ret = 0;
+                        ret = 0;
                         goto out;
                 }
         }
 
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_getxattr_cbk,
-                            FIRST_CHILD (this),
-                            (FIRST_CHILD (this))->fops->getxattr,
-                            loc, name);
-                return 0;
-        }
-
-        local->fresh_children = GF_CALLOC (priv->child_count,
-                                          sizeof (*local->fresh_children),
-                                          gf_afr_mt_int32_t);
-        if (local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+	afr_getxattr (frame, this, loc, name, xdata);
 
-        read_child = afr_inode_get_read_ctx (this, loc->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;
-        }
-	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);
-
-	op_ret = 0;
+	ret = 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, NULL);
-	}
-	return 0;
-}
-
-static int
-afr_setxattr_unwind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = 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 (setxattr, main_frame,
-                                  local->op_ret, local->op_errno)
-	}
-	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)
-{
-	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 (priv->child_count, local->child_up);
-
-	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);
-
-			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)
-{
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno);
+	if (ret < 0)
+		AFR_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
 	return 0;
 }
 
@@ -1721,120 +1463,85 @@ pump_command_reply (call_frame_t *frame, xlator_t *this)
                 gf_log (this->name, GF_LOG_INFO,
                         "Command succeeded");
 
-        dict_unref (local->dict);
-
         AFR_STACK_UNWIND (setxattr,
                           frame,
                           local->op_ret,
-                          local->op_errno);
+                          local->op_errno, NULL);
 
         return 0;
 }
 
 int
-pump_parse_command (call_frame_t *frame, xlator_t *this,
-                    afr_local_t *local, dict_t *dict)
+pump_parse_command (call_frame_t *frame, xlator_t *this, dict_t *dict,
+		    int *op_errno_p)
 {
-
+	afr_local_t *local = NULL;
         int ret = -1;
+	int op_errno = 0;
 
         if (pump_command_start (this, dict)) {
-                frame->local = local;
+                local = AFR_FRAME_INIT (frame, op_errno);
+		if (!local)
+			goto out;
                 local->dict = dict_ref (dict);
                 ret = pump_execute_start (frame, this);
 
         } else if (pump_command_pause (this, dict)) {
-                frame->local = local;
+		local = AFR_FRAME_INIT (frame, op_errno);
+		if (!local)
+			goto out;
                 local->dict = dict_ref (dict);
                 ret = pump_execute_pause (frame, this);
 
         } else if (pump_command_abort (this, dict)) {
-                frame->local = local;
+		local = AFR_FRAME_INIT (frame, op_errno);
+		if (!local)
+			goto out;
                 local->dict = dict_ref (dict);
                 ret = pump_execute_abort (frame, this);
 
         } else if (pump_command_commit (this, dict)) {
-                frame->local = local;
+		local = AFR_FRAME_INIT (frame, op_errno);
+		if (!local)
+			goto out;
                 local->dict = dict_ref (dict);
                 ret = pump_execute_commit (frame, this);
         }
+out:
+	if (op_errno_p)
+		*op_errno_p = op_errno;
         return ret;
 }
 
 int
-pump_setxattr (call_frame_t *frame, xlator_t *this,
-               loc_t *loc, dict_t *dict, int32_t flags)
+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;
-
+	afr_private_t *priv = NULL;
 	int ret = -1;
-
-	int op_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;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-        ret = pump_parse_command (frame, this,
-                                  local, dict);
-        if (ret >= 0) {
-                op_ret = 0;
-                goto out;
-        }
-
         if (!priv->use_afr_in_pump) {
                 STACK_WIND (frame, default_setxattr_cbk,
                             FIRST_CHILD (this),
                             (FIRST_CHILD (this))->fops->setxattr,
-                            loc, dict, flags);
+                            loc, dict, flags, xdata);
                 return 0;
         }
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		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;
+        ret = pump_parse_command (frame, this, dict, &op_errno);
+        if (ret >= 0)
+                goto out;
 
-	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+	afr_setxattr (frame, this, loc, dict, flags, xdata);
 
-	op_ret = 0;
+	ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (setxattr, frame, op_ret, op_errno);
+	if (ret < 0) {
+		AFR_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL);
 	}
 
 	return 0;
@@ -1868,7 +1575,7 @@ static int32_t
 pump_truncate (call_frame_t *frame,
                xlator_t *this,
                loc_t *loc,
-               off_t offset)
+               off_t offset, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -1878,11 +1585,11 @@ pump_truncate (call_frame_t *frame,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->truncate,
                             loc,
-                            offset);
+                            offset, xdata);
                 return 0;
         }
 
-        afr_truncate (frame, this, loc, offset);
+        afr_truncate (frame, this, loc, offset, xdata);
         return 0;
 }
 
@@ -1891,7 +1598,7 @@ static int32_t
 pump_ftruncate (call_frame_t *frame,
                 xlator_t *this,
                 fd_t *fd,
-                off_t offset)
+                off_t offset, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -1901,11 +1608,11 @@ pump_ftruncate (call_frame_t *frame,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->ftruncate,
                             fd,
-                            offset);
+                            offset, xdata);
                 return 0;
         }
 
-        afr_ftruncate (frame, this, fd, offset);
+        afr_ftruncate (frame, this, fd, offset, xdata);
         return 0;
 }
 
@@ -1914,7 +1621,7 @@ pump_ftruncate (call_frame_t *frame,
 
 int
 pump_mknod (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, mode_t mode, dev_t rdev, dict_t *parms)
+            loc_t *loc, mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -1922,10 +1629,10 @@ pump_mknod (call_frame_t *frame, xlator_t *this,
                 STACK_WIND (frame, default_mknod_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->mknod,
-                            loc, mode, rdev, parms);
+                            loc, mode, rdev, umask, xdata);
                 return 0;
         }
-        afr_mknod (frame, this, loc, mode, rdev, parms);
+        afr_mknod (frame, this, loc, mode, rdev, umask, xdata);
         return 0;
 
 }
@@ -1934,7 +1641,7 @@ pump_mknod (call_frame_t *frame, xlator_t *this,
 
 int
 pump_mkdir (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, mode_t mode, dict_t *params)
+            loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -1942,10 +1649,10 @@ pump_mkdir (call_frame_t *frame, xlator_t *this,
                 STACK_WIND (frame, default_mkdir_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->mkdir,
-                            loc, mode, params);
+                            loc, mode, umask, xdata);
                 return 0;
         }
-        afr_mkdir (frame, this, loc, mode, params);
+        afr_mkdir (frame, this, loc, mode, umask, xdata);
         return 0;
 
 }
@@ -1954,7 +1661,7 @@ pump_mkdir (call_frame_t *frame, xlator_t *this,
 static int32_t
 pump_unlink (call_frame_t *frame,
              xlator_t *this,
-             loc_t *loc)
+             loc_t *loc, int xflag, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -1963,10 +1670,10 @@ pump_unlink (call_frame_t *frame,
                             default_unlink_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->unlink,
-                            loc);
+                            loc, xflag, xdata);
                 return 0;
         }
-        afr_unlink (frame, this, loc);
+        afr_unlink (frame, this, loc, xflag, xdata);
         return 0;
 
 }
@@ -1974,7 +1681,7 @@ pump_unlink (call_frame_t *frame,
 
 static int
 pump_rmdir (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, int flags)
+            loc_t *loc, int flags, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 
@@ -1984,11 +1691,11 @@ pump_rmdir (call_frame_t *frame, xlator_t *this,
                 STACK_WIND (frame, default_rmdir_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->rmdir,
-                            loc, flags);
+                            loc, flags, xdata);
                 return 0;
         }
 
-        afr_rmdir (frame, this, loc, flags);
+        afr_rmdir (frame, this, loc, flags, xdata);
         return 0;
 
 }
@@ -1997,7 +1704,7 @@ pump_rmdir (call_frame_t *frame, xlator_t *this,
 
 int
 pump_symlink (call_frame_t *frame, xlator_t *this,
-              const char *linkpath, loc_t *loc, dict_t *params)
+              const char *linkpath, loc_t *loc, mode_t umask, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2005,10 +1712,10 @@ pump_symlink (call_frame_t *frame, xlator_t *this,
                 STACK_WIND (frame, default_symlink_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->symlink,
-                            linkpath, loc, params);
+                            linkpath, loc, umask, xdata);
                 return 0;
         }
-        afr_symlink (frame, this, linkpath, loc, params);
+        afr_symlink (frame, this, linkpath, loc, umask, xdata);
         return 0;
 
 }
@@ -2018,7 +1725,7 @@ static int32_t
 pump_rename (call_frame_t *frame,
              xlator_t *this,
              loc_t *oldloc,
-             loc_t *newloc)
+             loc_t *newloc, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2027,10 +1734,10 @@ pump_rename (call_frame_t *frame,
                             default_rename_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->rename,
-                            oldloc, newloc);
+                            oldloc, newloc, xdata);
                 return 0;
         }
-        afr_rename (frame, this, oldloc, newloc);
+        afr_rename (frame, this, oldloc, newloc, xdata);
         return 0;
 
 }
@@ -2040,7 +1747,7 @@ static int32_t
 pump_link (call_frame_t *frame,
            xlator_t *this,
            loc_t *oldloc,
-           loc_t *newloc)
+           loc_t *newloc, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2049,10 +1756,10 @@ pump_link (call_frame_t *frame,
                             default_link_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->link,
-                            oldloc, newloc);
+                            oldloc, newloc, xdata);
                 return 0;
         }
-        afr_link (frame, this, oldloc, newloc);
+        afr_link (frame, this, oldloc, newloc, xdata);
         return 0;
 
 }
@@ -2061,7 +1768,7 @@ pump_link (call_frame_t *frame,
 static int32_t
 pump_create (call_frame_t *frame, xlator_t *this,
              loc_t *loc, int32_t flags, mode_t mode,
-             fd_t *fd, dict_t *params)
+             mode_t umask, fd_t *fd, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2069,10 +1776,10 @@ pump_create (call_frame_t *frame, xlator_t *this,
                 STACK_WIND (frame, default_create_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->create,
-                            loc, flags, mode, fd, params);
+                            loc, flags, mode, umask, fd, xdata);
                 return 0;
         }
-        afr_create (frame, this, loc, flags, mode, fd, params);
+        afr_create (frame, this, loc, flags, mode, umask, fd, xdata);
         return 0;
 
 }
@@ -2082,8 +1789,7 @@ static int32_t
 pump_open (call_frame_t *frame,
            xlator_t *this,
            loc_t *loc,
-           int32_t flags, fd_t *fd,
-           int32_t wbflags)
+           int32_t flags, fd_t *fd, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2092,10 +1798,10 @@ pump_open (call_frame_t *frame,
                             default_open_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->open,
-                            loc, flags, fd, wbflags);
+                            loc, flags, fd, xdata);
                 return 0;
         }
-        afr_open (frame, this, loc, flags, fd, wbflags);
+        afr_open (frame, this, loc, flags, fd, xdata);
         return 0;
 
 }
@@ -2107,8 +1813,8 @@ pump_writev (call_frame_t *frame,
              fd_t *fd,
              struct iovec *vector,
              int32_t count,
-             off_t off,
-             struct iobref *iobref)
+             off_t off, uint32_t flags,
+             struct iobref *iobref, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2120,20 +1826,20 @@ pump_writev (call_frame_t *frame,
                             fd,
                             vector,
                             count,
-                            off,
-                            iobref);
+                            off, flags,
+                            iobref, xdata);
                 return 0;
         }
-        afr_writev (frame, this, fd, vector, count, off, iobref);
-        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)
+            fd_t *fd, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2142,10 +1848,10 @@ pump_flush (call_frame_t *frame,
                             default_flush_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->flush,
-                            fd);
+                            fd, xdata);
                 return 0;
         }
-        afr_flush (frame, this, fd);
+        afr_flush (frame, this, fd, xdata);
         return 0;
 
 }
@@ -2155,7 +1861,7 @@ static int32_t
 pump_fsync (call_frame_t *frame,
             xlator_t *this,
             fd_t *fd,
-            int32_t flags)
+            int32_t flags, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2165,10 +1871,10 @@ pump_fsync (call_frame_t *frame,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->fsync,
                             fd,
-                            flags);
+                            flags, xdata);
                 return 0;
         }
-        afr_fsync (frame, this, fd, flags);
+        afr_fsync (frame, this, fd, flags, xdata);
         return 0;
 
 }
@@ -2177,7 +1883,7 @@ pump_fsync (call_frame_t *frame,
 static int32_t
 pump_opendir (call_frame_t *frame,
               xlator_t *this,
-              loc_t *loc, fd_t *fd)
+              loc_t *loc, fd_t *fd, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2186,10 +1892,10 @@ pump_opendir (call_frame_t *frame,
                             default_opendir_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->opendir,
-                            loc, fd);
+                            loc, fd, xdata);
                 return 0;
         }
-        afr_opendir (frame, this, loc, fd);
+        afr_opendir (frame, this, loc, fd, xdata);
         return 0;
 
 }
@@ -2199,7 +1905,7 @@ static int32_t
 pump_fsyncdir (call_frame_t *frame,
                xlator_t *this,
                fd_t *fd,
-               int32_t flags)
+               int32_t flags, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2209,10 +1915,10 @@ pump_fsyncdir (call_frame_t *frame,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->fsyncdir,
                             fd,
-                            flags);
+                            flags, xdata);
                 return 0;
         }
-        afr_fsyncdir (frame, this, fd, flags);
+        afr_fsyncdir (frame, this, fd, flags, xdata);
         return 0;
 
 }
@@ -2223,7 +1929,7 @@ pump_xattrop (call_frame_t *frame,
               xlator_t *this,
               loc_t *loc,
               gf_xattrop_flags_t flags,
-              dict_t *dict)
+              dict_t *dict, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2234,10 +1940,10 @@ pump_xattrop (call_frame_t *frame,
                             FIRST_CHILD(this)->fops->xattrop,
                             loc,
                             flags,
-                            dict);
+                            dict, xdata);
                 return 0;
         }
-        afr_xattrop (frame, this, loc, flags, dict);
+        afr_xattrop (frame, this, loc, flags, dict, xdata);
         return 0;
 
 }
@@ -2247,7 +1953,7 @@ pump_fxattrop (call_frame_t *frame,
                xlator_t *this,
                fd_t *fd,
                gf_xattrop_flags_t flags,
-               dict_t *dict)
+               dict_t *dict, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2258,10 +1964,10 @@ pump_fxattrop (call_frame_t *frame,
                             FIRST_CHILD(this)->fops->fxattrop,
                             fd,
                             flags,
-                            dict);
+                            dict, xdata);
                 return 0;
         }
-        afr_fxattrop (frame, this, fd, flags, dict);
+        afr_fxattrop (frame, this, fd, flags, dict, xdata);
         return 0;
 
 }
@@ -2271,9 +1977,17 @@ static int32_t
 pump_removexattr (call_frame_t *frame,
                   xlator_t *this,
                   loc_t *loc,
-                  const char *name)
+                  const char *name, dict_t *xdata)
 {
-        afr_private_t *priv  = NULL;
+        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,
@@ -2281,10 +1995,14 @@ pump_removexattr (call_frame_t *frame,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->removexattr,
                             loc,
-                            name);
+                            name, xdata);
                 return 0;
         }
-        afr_removexattr (frame, this, loc, name);
+        afr_removexattr (frame, this, loc, name, xdata);
+
+ out:
+        if (op_errno)
+                AFR_STACK_UNWIND (removexattr, frame, -1, op_errno, NULL);
         return 0;
 
 }
@@ -2296,7 +2014,7 @@ pump_readdir (call_frame_t *frame,
               xlator_t *this,
               fd_t *fd,
               size_t size,
-              off_t off)
+              off_t off, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2305,21 +2023,18 @@ pump_readdir (call_frame_t *frame,
                             default_readdir_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->readdir,
-                            fd, size, off);
+                            fd, size, off, xdata);
                 return 0;
         }
-        afr_readdir (frame, this, fd, size, off);
+        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)
+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;
@@ -2328,10 +2043,10 @@ pump_readdirp (call_frame_t *frame,
                             default_readdirp_cbk,
                             FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->readdirp,
-                            fd, size, off);
+                            fd, size, off, dict);
                 return 0;
         }
-        afr_readdirp (frame, this, fd, size, off);
+        afr_readdirp (frame, this, fd, size, off, dict);
         return 0;
 
 }
@@ -2362,13 +2077,24 @@ pump_release (xlator_t *this,
 
 }
 
+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)
+              int32_t valid, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2377,10 +2103,10 @@ pump_setattr (call_frame_t *frame,
                             default_setattr_cbk,
                             FIRST_CHILD (this),
                             FIRST_CHILD (this)->fops->setattr,
-                            loc, stbuf, valid);
+                            loc, stbuf, valid, xdata);
                 return 0;
         }
-        afr_setattr (frame, this, loc, stbuf, valid);
+        afr_setattr (frame, this, loc, stbuf, valid, xdata);
         return 0;
 
 }
@@ -2391,7 +2117,7 @@ pump_fsetattr (call_frame_t *frame,
                xlator_t *this,
                fd_t *fd,
                struct iatt *stbuf,
-               int32_t valid)
+               int32_t valid, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
 	priv = this->private;
@@ -2400,10 +2126,10 @@ pump_fsetattr (call_frame_t *frame,
                             default_fsetattr_cbk,
                             FIRST_CHILD (this),
                             FIRST_CHILD (this)->fops->fsetattr,
-                            fd, stbuf, valid);
+                            fd, stbuf, valid, xdata);
                 return 0;
         }
-        afr_fsetattr (frame, this, fd, stbuf, valid);
+        afr_fsetattr (frame, this, fd, stbuf, valid, xdata);
         return 0;
 
 }
@@ -2452,7 +2178,7 @@ notify (xlator_t *this, int32_t event,
 
         child_xl = (xlator_t *) data;
 
-        ret = afr_notify (this, event, data);
+        ret = afr_notify (this, event, data, NULL);
 
 	switch (event) {
 	case GF_EVENT_CHILD_DOWN:
@@ -2482,12 +2208,12 @@ int32_t
 init (xlator_t *this)
 {
 	afr_private_t * priv        = NULL;
-        pump_private_t *pump_priv = NULL;
+        pump_private_t *pump_priv   = NULL;
 	int             child_count = 0;
 	xlator_list_t * trav        = NULL;
 	int             i           = 0;
 	int             ret         = -1;
-	int             op_errno    = 0;
+	GF_UNUSED int   op_errno    = 0;
 
         int source_child = 0;
 
@@ -2503,26 +2229,36 @@ init (xlator_t *this)
 			"Volume is dangling.");
 	}
 
-	ALLOC_OR_GOTO (this->private, afr_private_t, out);
+	priv = GF_CALLOC (1, sizeof (afr_private_t), gf_afr_mt_afr_private_t);
+        if (!priv)
+                goto out;
 
-	priv = this->private;
+        LOCK_INIT (&priv->lock);
+
+        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     = 1;
+	priv->data_self_heal     = "on";
 	priv->metadata_self_heal = 1;
 	priv->entry_self_heal    = 1;
 
-        priv->data_self_heal_algorithm = "";
-
         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 */
 
@@ -2531,31 +2267,6 @@ init (xlator_t *this)
            and the sink.
         */
 
-	priv->data_lock_server_count = 2;
-	priv->metadata_lock_server_count = 2;
-	priv->entry_lock_server_count = 2;
-
-	priv->strict_readdir = _gf_false;
-
-        trav = this->children;
-        while (trav) {
-                child_count++;
-                trav = trav->next;
-        }
-
-	priv->wait_count = 1;
-
-        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;
-
-	LOCK_INIT (&priv->lock);
-        LOCK_INIT (&priv->read_child_lock);
-
 	priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count,
                                  gf_afr_mt_char);
 	if (!priv->child_up) {
@@ -2589,7 +2300,8 @@ init (xlator_t *this)
 	while (i < child_count) {
 		priv->children[i] = trav->xlator;
 
-                ret = gf_asprintf (&priv->pending_key[i], "%s.%s", AFR_XATTR_PREFIX,
+                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,
@@ -2602,7 +2314,12 @@ init (xlator_t *this)
 		i++;
 	}
 
-        priv->first_lookup = 1;
+        ret = gf_asprintf (&priv->sh_domain, "%s-self-heal", this->name);
+        if (-1 == ret) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
         priv->root_inode = NULL;
 
         priv->last_event = GF_CALLOC (child_count, sizeof (*priv->last_event),
@@ -2627,13 +2344,12 @@ init (xlator_t *this)
         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");
+                gf_log (this->name, GF_LOG_ERROR, "Out of memory");
                 ret = -1;
                 goto out;
         }
 
-	pump_priv->env = syncenv_new (0);
+	pump_priv->env = this->ctx->env;
         if (!pump_priv->env) {
                 gf_log (this->name, GF_LOG_ERROR,
                         "Could not create new sync-environment");
@@ -2641,21 +2357,67 @@ init (xlator_t *this)
                 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_priv = NULL;
 
-        pthread_mutex_init (&priv->mutex, NULL);
-        INIT_LIST_HEAD (&priv->saved_fds);
+        this->private = priv;
+        priv = NULL;
 
         pump_change_state (this, PUMP_STATE_ABORT);
 
 	ret = 0;
 out:
+
+        if (pump_priv) {
+                GF_FREE (pump_priv->resume_path);
+                LOCK_DESTROY (&pump_priv->resume_path_lock);
+                LOCK_DESTROY (&pump_priv->pump_state_lock);
+                GF_FREE (pump_priv);
+        }
+
+        if (priv) {
+                GF_FREE (priv->child_up);
+                GF_FREE (priv->children);
+                GF_FREE (priv->pending_key);
+                GF_FREE (priv->last_event);
+                LOCK_DESTROY (&priv->lock);
+                GF_FREE (priv);
+        }
+
 	return ret;
 }
 
 int
 fini (xlator_t *this)
 {
+        afr_private_t * priv        = NULL;
+        pump_private_t *pump_priv = NULL;
+
+        priv      = this->private;
+        this->private = NULL;
+        if (!priv)
+                goto out;
+
+        pump_priv = priv->pump_private;
+        if (!pump_priv)
+                goto afr_priv;
+
+        GF_FREE (pump_priv->resume_path);
+        LOCK_DESTROY (&pump_priv->resume_path_lock);
+        LOCK_DESTROY (&pump_priv->pump_state_lock);
+        GF_FREE (pump_priv);
+afr_priv:
+        afr_priv_destroy (priv);
+out:
 	return 0;
 }
 
@@ -2703,6 +2465,7 @@ struct xlator_dumpops dumpops = {
 struct xlator_cbks cbks = {
 	.release     = pump_release,
 	.releasedir  = pump_releasedir,
+        .forget      = pump_forget,
 };
 
 struct volume_options options[] = {
diff --git a/xlators/cluster/afr/src/pump.h b/xlators/cluster/afr/src/pump.h
index d69db9368..9d0b6db6a 100644
--- a/xlators/cluster/afr/src/pump.h
+++ b/xlators/cluster/afr/src/pump.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  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__
@@ -26,16 +17,6 @@
 #define CLIENT_CMD_CONNECT "trusted.glusterfs.client-connect"
 #define CLIENT_CMD_DISCONNECT "trusted.glusterfs.client-disconnect"
 
-#define IS_ROOT_PATH(path) (!strcmp (path, "/"))
-#define IS_ENTRY_CWD(entry) (!strcmp (entry, "."))
-#define IS_ENTRY_PARENT(entry) (!strcmp (entry, ".."))
-
-#define PUMP_CMD_START  "trusted.glusterfs.pump.start"
-#define PUMP_CMD_COMMIT "trusted.glusterfs.pump.commit"
-#define PUMP_CMD_ABORT  "trusted.glusterfs.pump.abort"
-#define PUMP_CMD_PAUSE  "trusted.glusterfs.pump.pause"
-#define PUMP_CMD_STATUS "trusted.glusterfs.pump.status"
-
 #define PUMP_SOURCE_COMPLETE "trusted.glusterfs.pump-source-complete"
 #define PUMP_SINK_COMPLETE "trusted.glusterfs.pump-sink-complete"
 
@@ -54,7 +35,7 @@ typedef enum {
 
 typedef struct _pump_private {
 	struct syncenv *env;            /* The env pointer to the pump synctask */
-        const char *resume_path;        /* path to resume from the last pause */
+        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 */
@@ -94,4 +75,7 @@ pump_command_status (xlator_t *this, dict_t *dict);
 int
 pump_execute_status (call_frame_t *frame, xlator_t *this);
 
+int
+pump_command_reply (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 cf883a974..3fc29bf81 100644
--- a/xlators/cluster/dht/src/Makefile.am
+++ b/xlators/cluster/dht/src/Makefile.am
@@ -1,32 +1,33 @@
-
 xlator_LTLIBRARIES = dht.la nufa.la switch.la
 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 \
-		$(top_builddir)/xlators/lib/src/libxlator.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_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 -avoidversion
+dht_la_LDFLAGS = -module -avoid-version
 dht_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-nufa_la_LDFLAGS = -module -avoidversion
+nufa_la_LDFLAGS = -module -avoid-version
 nufa_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-switch_la_LDFLAGS = -module -avoidversion
+switch_la_LDFLAGS = -module -avoid-version
 switch_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-noinst_HEADERS = dht-common.h dht-common.c dht-mem-types.h $(top_builddir)/xlators/lib/src/libxlator.h
+noinst_HEADERS = dht-common.h dht-mem-types.h \
+	$(top_builddir)/xlators/lib/src/libxlator.h
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS) \
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
 	-I$(top_srcdir)/xlators/lib/src
 
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
 CLEANFILES =
 
 uninstall-local:
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index c4bc0b4df..3868fc38f 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2009-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -31,17 +22,18 @@
 #include "dht-common.h"
 #include "defaults.h"
 #include "byte-order.h"
+#include "glusterfs-acl.h"
 
 #include <sys/time.h>
 #include <libgen.h>
 
-void
+int
 dht_aggregate (dict_t *this, char *key, data_t *value, void *data)
 {
         dict_t  *dst  = NULL;
         int64_t *ptr  = 0, *size = NULL;
         int32_t  ret  = -1;
-        data_pair_t  *data_pair = NULL;
+        data_t  *dict_data = NULL;
 
         dst = data;
 
@@ -53,32 +45,37 @@ dht_aggregate (dict_t *this, char *key, data_t *value, void *data)
                         if (size == NULL) {
                                 gf_log ("dht", GF_LOG_WARNING,
                                         "memory allocation failed");
-                                return;
+                                return -1;
                         }
                         ret = dict_set_bin (dst, key, size, sizeof (int64_t));
                         if (ret < 0) {
                                 gf_log ("dht", GF_LOG_WARNING,
                                         "dht aggregate dict set failed");
                                 GF_FREE (size);
-                                return;
+                                return -1;
                         }
                 }
 
                 ptr = data_to_bin (value);
                 if (ptr == NULL) {
                         gf_log ("dht", GF_LOG_WARNING, "data to bin failed");
-                        return;
+                        return -1;
                 }
 
                 *size = hton64 (ntoh64 (*size) + ntoh64 (*ptr));
+
+        } else if (fnmatch (GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0) {
+                ret = gf_get_min_stime (THIS, dst, key, value);
+                if (ret < 0)
+                        return ret;
         } else {
                 /* compare user xattrs only */
                 if (!strncmp (key, "user.", strlen ("user."))) {
-                        ret = dict_lookup (dst, key, &data_pair); 
-                        if (!ret && data) {
-                                ret = is_data_equal (data_pair->value, value);
+                        ret = dict_lookup (dst, key, &dict_data);
+                        if (!ret && dict_data && value) {
+                                ret = is_data_equal (dict_data, value);
                                 if (!ret)
-                                        gf_log ("dht", GF_LOG_WARNING,
+                                        gf_log ("dht", GF_LOG_DEBUG,
                                                 "xattr mismatch for %s", key);
                         }
                 }
@@ -87,7 +84,7 @@ dht_aggregate (dict_t *this, char *key, data_t *value, void *data)
                         gf_log ("dht", GF_LOG_WARNING, "xattr dict set failed");
         }
 
-        return;
+        return 0;
 }
 
 
@@ -114,7 +111,7 @@ out:
 int
 dht_lookup_selfheal_cbk (call_frame_t *frame, void *cookie,
                          xlator_t *this,
-                         int op_ret, int op_errno)
+                         int op_ret, int op_errno, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         dht_layout_t *layout = NULL;
@@ -134,7 +131,12 @@ dht_lookup_selfheal_cbk (call_frame_t *frame, void *cookie,
                 ret = dht_layout_set (this, local->inode, layout);
         }
 
-        WIPE (&local->postparent);
+        if (local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           &local->postparent, 1);
+        }
+
+        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
 
         DHT_STACK_UNWIND (lookup, frame, ret, local->op_errno, local->inode,
                           &local->stbuf, local->xattr, &local->postparent);
@@ -145,12 +147,261 @@ out:
 
 
 int
+dht_discover_complete (xlator_t *this, call_frame_t *discover_frame)
+{
+        dht_local_t     *local = NULL;
+        call_frame_t    *main_frame = NULL;
+        int              op_errno = 0;
+        int              ret = -1;
+        dht_layout_t    *layout = NULL;
+        dht_conf_t      *conf = NULL;
+
+        local = discover_frame->local;
+        layout = local->layout;
+        conf = this->private;
+
+        LOCK(&discover_frame->lock);
+        {
+                main_frame = local->main_frame;
+                local->main_frame = NULL;
+        }
+        UNLOCK(&discover_frame->lock);
+
+        if (!main_frame)
+                return 0;
+
+        if (local->file_count && local->dir_count) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "path %s exists as a file on one subvolume "
+                        "and directory on another. "
+                        "Please fix it manually",
+                        local->loc.path);
+                op_errno = EIO;
+                goto out;
+        }
+
+        if (local->cached_subvol) {
+                ret = dht_layout_preset (this, local->cached_subvol,
+                                         local->inode);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "failed to set layout for subvolume %s",
+                                local->cached_subvol ? local->cached_subvol->name : "<nil>");
+                        op_errno = EINVAL;
+                        goto out;
+                }
+        } else {
+                ret = dht_layout_normalize (this, &local->loc, layout);
+                if ((ret < 0) || ((ret > 0) && (local->op_ret != 0))) {
+                        /* either the layout is incorrect or the directory is
+                         * not found even in one subvolume.
+                         */
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "normalizing failed on %s "
+                                "(overlaps/holes present: %s, "
+                                "ENOENT errors: %d)", local->loc.path,
+                                (ret < 0) ? "yes" : "no", (ret > 0) ? ret : 0);
+                        if ((ret > 0) && (ret == conf->subvolume_cnt)) {
+                                op_errno = ESTALE;
+                                goto out;
+                        }
+                }
+
+                if (local->inode)
+                        dht_layout_set (this, local->inode, layout);
+        }
+
+        DHT_STACK_UNWIND (lookup, main_frame, local->op_ret, local->op_errno,
+                          local->inode, &local->stbuf, local->xattr,
+                          &local->postparent);
+        return 0;
+out:
+        DHT_STACK_UNWIND (lookup, main_frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL);
+
+        return ret;
+}
+
+
+int
+dht_discover_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int op_ret, int op_errno,
+                  inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                  struct iatt *postparent)
+{
+        dht_local_t  *local                   = NULL;
+        int           this_call_cnt           = 0;
+        call_frame_t *prev                    = NULL;
+        dht_layout_t *layout                  = NULL;
+        int           ret                     = -1;
+        int           is_dir                  = 0;
+        int           is_linkfile             = 0;
+        int           attempt_unwind          = 0;
+        dht_conf_t   *conf                    = 0;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, out);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev  = cookie;
+        conf = this->private;
+
+        layout = local->layout;
+
+        /* Check if the gfid is different for file from other node */
+        if (!op_ret && uuid_compare (local->gfid, stbuf->ia_gfid)) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: gfid different on %s",
+                        local->loc.path, prev->this->name);
+        }
+
+
+        LOCK (&frame->lock);
+        {
+                /* TODO: assert equal mode on stbuf->st_mode and
+                   local->stbuf->st_mode
+
+                   else mkdir/chmod/chown and fix
+                */
+                ret = dht_layout_merge (this, layout, prev->this,
+                                        op_ret, op_errno, xattr);
+                if (ret)
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: failed to merge layouts", local->loc.path);
+
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "lookup of %s on %s returned error (%s)",
+                                local->loc.path, prev->this->name,
+                                strerror (op_errno));
+
+                        goto unlock;
+                }
+
+                is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+                                                 conf->link_xattr_name);
+                is_dir = check_is_dir (inode, stbuf, xattr);
+
+                if (is_dir) {
+                        local->dir_count ++;
+                } else {
+                        local->file_count ++;
+
+                        if (!is_linkfile) {
+                                /* real file */
+                                local->cached_subvol = prev->this;
+                                attempt_unwind = 1;
+                        } else {
+                                goto unlock;
+                        }
+                }
+
+                local->op_ret = 0;
+
+                if (local->xattr == NULL) {
+                        local->xattr = dict_ref (xattr);
+                } else {
+                        dht_aggregate_xattr (local->xattr, xattr);
+                }
+
+                if (local->inode == NULL)
+                        local->inode = inode_ref (inode);
+
+                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+                dht_iatt_merge (this, &local->postparent, postparent,
+                                prev->this);
+        }
+unlock:
+        UNLOCK (&frame->lock);
+out:
+        this_call_cnt = dht_frame_return (frame);
+
+        if (is_last_call (this_call_cnt) || attempt_unwind) {
+                dht_discover_complete (this, frame);
+        }
+
+        if (is_last_call (this_call_cnt))
+                DHT_STACK_DESTROY (frame);
+
+        return 0;
+}
+
+
+int
+dht_discover (call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+        int          ret;
+        dht_local_t *local = NULL;
+        dht_conf_t  *conf = NULL;
+        int          call_cnt = 0;
+        int          op_errno = EINVAL;
+        int          i = 0;
+        call_frame_t *discover_frame = NULL;
+
+        conf = this->private;
+        local = frame->local;
+
+        ret = dict_set_uint32 (local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to set '%s' key",
+                        loc->path, conf->xattr_name);
+
+        ret = dict_set_uint32 (local->xattr_req, conf->link_xattr_name, 256);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to set '%s' key",
+                        loc->path, conf->link_xattr_name);
+
+        call_cnt        = conf->subvolume_cnt;
+        local->call_cnt = call_cnt;
+
+        local->layout = dht_layout_new (this, conf->subvolume_cnt);
+
+        if (!local->layout) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        uuid_copy (local->gfid, loc->gfid);
+
+        discover_frame = copy_frame (frame);
+        if (!discover_frame) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        discover_frame->local = local;
+        frame->local = NULL;
+        local->main_frame = frame;
+
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (discover_frame, dht_discover_cbk,
+                            conf->subvolumes[i],
+                            conf->subvolumes[i]->fops->lookup,
+                            &local->loc, local->xattr_req);
+        }
+
+        return 0;
+
+err:
+        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL);
+
+        return 0;
+}
+
+
+int
 dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                     int op_ret, int op_errno,
                     inode_t *inode, struct iatt *stbuf, dict_t *xattr,
                     struct iatt *postparent)
 {
-        dht_conf_t   *conf                    = NULL;
         dht_local_t  *local                   = NULL;
         int           this_call_cnt           = 0;
         call_frame_t *prev                    = NULL;
@@ -164,7 +415,6 @@ dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         GF_VALIDATE_OR_GOTO ("dht", this->private, out);
         GF_VALIDATE_OR_GOTO ("dht", cookie, out);
 
-        conf  = this->private;
         local = frame->local;
         prev  = cookie;
 
@@ -191,7 +441,7 @@ dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                         op_ret, op_errno, xattr);
 
                 if (op_ret == -1) {
-                        local->op_errno = ENOENT;
+                        local->op_errno = op_errno;
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "lookup of %s on %s returned error (%s)",
                                 local->loc.path, prev->this->name,
@@ -250,6 +500,12 @@ unlock:
                         dht_layout_set (this, local->inode, layout);
                 }
 
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
+
+                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
                 DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
                                   local->inode, &local->stbuf, local->xattr,
                                   &local->postparent);
@@ -259,6 +515,7 @@ unlock:
 
 selfheal:
         FRAME_SU_DO (frame, dht_local_t);
+        uuid_copy (local->loc.gfid, local->gfid);
         ret = dht_selfheal_directory (frame, dht_lookup_selfheal_cbk,
                                       &local->loc, layout);
 out:
@@ -279,6 +536,8 @@ dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         int           ret  = -1;
         int           is_dir = 0;
         int           is_linkfile = 0;
+        call_frame_t *copy          = NULL;
+        dht_local_t  *copy_local    = NULL;
 
         GF_VALIDATE_OR_GOTO ("dht", frame, err);
         GF_VALIDATE_OR_GOTO ("dht", this, err);
@@ -305,12 +564,20 @@ dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                         strerror (op_errno));
 			}
                         if (op_errno == ESTALE) {
-                                /* propogate the ESTALE to parent.
+                                /* propagate the ESTALE to parent.
                                  * setting local->return_estale would send
                                  * ESTALE to parent. */
                                 local->return_estale = 1;
                         }
 
+                        /* if it is ENOENT, we may have to do a
+                         * 'lookup_everywhere()' to make sure
+                         * the file is not migrated */
+                        if (op_errno == ENOENT) {
+                                if (IA_ISREG (local->loc.inode->ia_type)) {
+                                        local->need_lookup_everywhere = 1;
+                                }
+                        }
                         goto unlock;
                 }
 
@@ -329,7 +596,8 @@ dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 layout = local->layout;
 
                 is_dir = check_is_dir (inode, stbuf, xattr);
-                is_linkfile = check_is_linkfile (inode, stbuf, xattr);
+                is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+                                                 conf->link_xattr_name);
 
                 if (is_linkfile) {
                         gf_log (this->name, GF_LOG_INFO,
@@ -341,6 +609,23 @@ dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 }
 
                 if (is_dir) {
+                        ret = dht_dir_has_layout (xattr, conf->xattr_name);
+                        if (ret >= 0) {
+                                if (is_greater_time(local->stbuf.ia_ctime,
+                                                    local->stbuf.ia_ctime_nsec,
+                                                    stbuf->ia_ctime,
+                                                    stbuf->ia_ctime_nsec)) {
+                                        local->prebuf.ia_gid = stbuf->ia_gid;
+                                        local->prebuf.ia_uid = stbuf->ia_uid;
+                                }
+                        }
+                        if (local->stbuf.ia_type != IA_INVAL)
+                        {
+                                if ((local->stbuf.ia_gid != stbuf->ia_gid) ||
+                                    (local->stbuf.ia_uid != stbuf->ia_uid)) {
+                                        local->need_selfheal = 1;
+                                }
+                        }
                         ret = dht_layout_dir_mismatch (this, layout,
                                                        prev->this, &local->loc,
                                                        xattr);
@@ -379,7 +664,28 @@ out:
                     && (conf && conf->unhashed_sticky_bit)) {
                         local->stbuf.ia_prot.sticky = 1;
                 }
-                if (local->layout_mismatch) {
+                if (local->need_selfheal) {
+                        local->need_selfheal = 0;
+                        uuid_copy (local->gfid, local->stbuf.ia_gfid);
+                        local->stbuf.ia_gid = local->prebuf.ia_gid;
+                        local->stbuf.ia_uid = local->prebuf.ia_uid;
+                        copy = create_frame (this, this->ctx->pool);
+                        if (copy) {
+                                copy_local = dht_local_init (copy, &local->loc,
+                                                             NULL, 0);
+                                if (!copy_local)
+                                        goto cont;
+                                copy_local->stbuf = local->stbuf;
+                                copy->local = copy_local;
+                                FRAME_SU_DO (copy, dht_local_t);
+                                ret = synctask_new (this->ctx->env,
+                                                    dht_dir_attr_heal,
+                                                    dht_dir_attr_heal_done,
+                                                    copy, copy);
+                        }
+                }
+cont:
+		if (local->layout_mismatch) {
                         /* Found layout mismatch in the directory, need to
                            fix this in the inode context */
                         dht_layout_unref (this, local->layout);
@@ -388,13 +694,29 @@ out:
                         return 0;
                 }
 
+                if (local->need_lookup_everywhere) {
+                        /* As the current layout gave ENOENT error, we would
+                           need a new layout */
+                        dht_layout_unref (this, local->layout);
+                        local->layout = NULL;
+
+                        /* We know that current cached subvol is no more
+                           valid, get the new one */
+                        local->cached_subvol = NULL;
+                        dht_lookup_everywhere (frame, this, &local->loc);
+                        return 0;
+                }
                 if (local->return_estale) {
                         local->op_ret = -1;
                         local->op_errno = ESTALE;
                 }
 
-                WIPE (&local->postparent);
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
 
+                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
                 DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
                                   local->inode, &local->stbuf, local->xattr,
                                   &local->postparent);
@@ -410,7 +732,8 @@ dht_lookup_linkfile_create_cbk (call_frame_t *frame, void *cookie,
                                 xlator_t *this,
                                 int32_t op_ret, int32_t op_errno,
                                 inode_t *inode, struct iatt *stbuf,
-                                struct iatt *preparent, struct iatt *postparent)
+                                struct iatt *preparent, struct iatt *postparent,
+                                dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         xlator_t     *cached_subvol = NULL;
@@ -427,7 +750,7 @@ dht_lookup_linkfile_create_cbk (call_frame_t *frame, void *cookie,
         cached_subvol = local->cached_subvol;
         conf = this->private;
 
-        ret = dht_layout_preset (this, local->cached_subvol, inode);
+        ret = dht_layout_preset (this, local->cached_subvol, local->loc.inode);
         if (ret < 0) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "failed to set layout for subvolume %s",
@@ -443,9 +766,16 @@ dht_lookup_linkfile_create_cbk (call_frame_t *frame, void *cookie,
                 local->stbuf.ia_prot.sticky = 1;
         }
 
+        if (local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
+        }
+
 unwind:
-        WIPE (&local->postparent);
+        if (local->linked == _gf_true)
+                dht_linkfile_attr_heal (frame, this);
 
+        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
         DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
                           local->inode, &local->stbuf, local->xattr,
                           &local->postparent);
@@ -461,7 +791,7 @@ dht_lookup_everywhere_done (call_frame_t *frame, xlator_t *this)
         dht_local_t  *local = NULL;
         xlator_t     *hashed_subvol = NULL;
         xlator_t     *cached_subvol = NULL;
-
+        dht_layout_t *layout = NULL;
 
         local = frame->local;
         hashed_subvol = local->hashed_subvol;
@@ -489,6 +819,46 @@ dht_lookup_everywhere_done (call_frame_t *frame, xlator_t *this)
                 return 0;
         }
 
+        if (local->need_lookup_everywhere) {
+                if (uuid_compare (local->gfid, local->inode->gfid)) {
+                        /* GFID different, return error */
+                        DHT_STACK_UNWIND (lookup, frame, -1, ENOENT, NULL,
+                                          NULL, NULL, NULL);
+                        return 0;
+                }
+                local->op_ret = 0;
+                local->op_errno = 0;
+                layout = dht_layout_for_subvol (this, cached_subvol);
+                if (!layout) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "%s: no pre-set layout for subvolume %s",
+                                local->loc.path, (cached_subvol ?
+                                                  cached_subvol->name :
+                                                  "<nil>"));
+                }
+
+                ret = dht_layout_set (this, local->inode, layout);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "%s: failed to set layout for subvol %s",
+                                local->loc.path, (cached_subvol ?
+                                                  cached_subvol->name :
+                                                  "<nil>"));
+                }
+
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
+
+                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
+                DHT_STACK_UNWIND (lookup, frame, local->op_ret,
+                                  local->op_errno, local->inode,
+                                  &local->stbuf, local->xattr,
+                                  &local->postparent);
+                return 0;
+        }
+
         if (!hashed_subvol) {
                 gf_log (this->name, GF_LOG_INFO,
                         "cannot create linkfile file for %s on %s: "
@@ -509,8 +879,12 @@ dht_lookup_everywhere_done (call_frame_t *frame, xlator_t *this)
                         local->op_errno = EINVAL;
                 }
 
-                WIPE (&local->postparent);
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
 
+                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
                 DHT_STACK_UNWIND (lookup, frame, local->op_ret,
                                   local->op_errno, local->inode,
                                   &local->stbuf, local->xattr,
@@ -524,7 +898,7 @@ dht_lookup_everywhere_done (call_frame_t *frame, xlator_t *this)
                 hashed_subvol->name);
 
         ret = dht_linkfile_create (frame,
-                                   dht_lookup_linkfile_create_cbk,
+                                   dht_lookup_linkfile_create_cbk, this,
                                    cached_subvol, hashed_subvol, &local->loc);
 
         return ret;
@@ -534,7 +908,8 @@ dht_lookup_everywhere_done (call_frame_t *frame, xlator_t *this)
 int
 dht_lookup_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                        int op_ret, int op_errno,
-                       struct iatt *preparent, struct iatt *postparent)
+                       struct iatt *preparent, struct iatt *postparent,
+                       dict_t *xdata)
 {
         int  this_call_cnt = 0;
 
@@ -553,7 +928,6 @@ dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                            inode_t *inode, struct iatt *buf, dict_t *xattr,
                            struct iatt *postparent)
 {
-        dht_conf_t   *conf          = NULL;
         dht_local_t  *local         = NULL;
         int           this_call_cnt = 0;
         call_frame_t *prev          = NULL;
@@ -562,7 +936,9 @@ dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         xlator_t     *subvol        = NULL;
         loc_t        *loc           = NULL;
         xlator_t     *link_subvol   = NULL;
-        int           ret = -1;
+        int           ret           = -1;
+        int32_t       fd_count      = 0;
+        dht_conf_t   *conf          = NULL;
 
         GF_VALIDATE_OR_GOTO ("dht", frame, out);
         GF_VALIDATE_OR_GOTO ("dht", this, out);
@@ -570,10 +946,9 @@ dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         GF_VALIDATE_OR_GOTO ("dht", cookie, out);
         GF_VALIDATE_OR_GOTO ("dht", this->private, out);
 
-        conf   = this->private;
-
         local  = frame->local;
         loc    = &local->loc;
+        conf   = this->private;
 
         prev   = cookie;
         subvol = prev->this;
@@ -595,7 +970,8 @@ dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                 loc->path, prev->this->name);
                 }
 
-                is_linkfile = check_is_linkfile (inode, buf, xattr);
+                is_linkfile = check_is_linkfile (inode, buf, xattr,
+                                                 conf->link_xattr_name);
                 is_dir = check_is_dir (inode, buf, xattr);
 
                 if (is_linkfile) {
@@ -636,7 +1012,7 @@ dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                 /* This is where we need 'rename' both entries logic */
                                 gf_log (this->name, GF_LOG_WARNING,
                                         "multiple subvolumes (%s and %s) have "
-                                        "file %s (preferrably rename the file "
+                                        "file %s (preferably rename the file "
                                         "in the backend, and do a fresh lookup)",
                                         local->cached_subvol->name,
                                         subvol->name, local->loc.path);
@@ -647,12 +1023,17 @@ unlock:
         UNLOCK (&frame->lock);
 
         if (is_linkfile) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "deleting stale linkfile %s on %s",
-                        loc->path, subvol->name);
-                STACK_WIND (frame, dht_lookup_unlink_cbk,
-                            subvol, subvol->fops->unlink, loc);
-                return 0;
+                ret = dict_get_int32 (xattr, GLUSTERFS_OPEN_FD_COUNT, &fd_count);
+                /* Delete the linkfile only if there are no open fds on it.
+                   if there is a open-fd, it may be in migration */
+                if (!ret && (fd_count == 0)) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "deleting stale linkfile %s on %s",
+                                loc->path, subvol->name);
+                        STACK_WIND (frame, dht_lookup_unlink_cbk,
+                                    subvol, subvol->fops->unlink, loc, 0, NULL);
+                        return 0;
+                }
         }
 
         this_call_cnt = dht_frame_return (frame);
@@ -732,7 +1113,16 @@ dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
                 gf_log (this->name, GF_LOG_INFO,
                         "lookup of %s on %s (following linkfile) failed (%s)",
                         local->loc.path, subvol->name, strerror (op_errno));
-                goto err;
+
+                /* If cached subvol returned ENOTCONN, do not do
+                lookup_everywhere. We need to make sure linkfile does not get
+                removed, which can take away the namespace, and subvol is
+                anyways down. */
+
+                if (op_errno != ENOTCONN)
+                        goto err;
+                else
+                        goto unwind;
         }
 
         if (check_is_dir (inode, stbuf, xattr)) {
@@ -742,7 +1132,7 @@ dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
                 goto err;
         }
 
-        if (check_is_linkfile (inode, stbuf, xattr)) {
+        if (check_is_linkfile (inode, stbuf, xattr, conf->link_xattr_name)) {
                 gf_log (this->name, GF_LOG_INFO,
                         "lookup of %s on %s (following linkfile) reached link",
                         local->loc.path, subvol->name);
@@ -770,9 +1160,13 @@ dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
                 op_errno = EINVAL;
         }
 
-unwind:
-        WIPE (postparent);
+        if (local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
+        }
 
+unwind:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
         DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
                           postparent);
 
@@ -816,9 +1210,13 @@ dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc)
                 local->xattr = NULL;
         }
 
-        if (!uuid_is_null (local->gfid))
+        if (!uuid_is_null (local->gfid)) {
                 ret = dict_set_static_bin (local->xattr_req, "gfid-req",
                                            local->gfid, 16);
+                if (ret)
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: failed to set gfid", local->loc.path);
+        }
 
         for (i = 0; i < call_cnt; i++) {
                 STACK_WIND (frame, dht_lookup_dir_cbk,
@@ -849,7 +1247,6 @@ dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         loc_t        *loc           = NULL;
         call_frame_t *prev          = NULL;
         int           ret           = 0;
-        uint64_t      tmp_layout    = 0;
         dht_layout_t *parent_layout = NULL;
 
         GF_VALIDATE_OR_GOTO ("dht", frame, err);
@@ -880,8 +1277,10 @@ dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 }
                 if ((conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) &&
                     (loc->parent)) {
-                        ret = inode_ctx_get (loc->parent, this, &tmp_layout);
-                        parent_layout = (dht_layout_t *)(long)tmp_layout;
+                        ret = dht_inode_ctx_layout_get (loc->parent, this,
+                                                        &parent_layout);
+                        if (ret || !parent_layout)
+                                goto out;
                         if (parent_layout->search_unhashed) {
                                 local->op_errno = ENOENT;
                                 dht_lookup_everywhere (frame, this, loc);
@@ -910,7 +1309,8 @@ dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 goto out;
         }
 
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr);
+        is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+                                         conf->link_xattr_name);
 
         if (!is_linkfile) {
                 /* non-directory and not a linkfile */
@@ -950,14 +1350,51 @@ out:
          * from each of the subvolume. See dht_iatt_merge for reference.
          */
 
-        WIPE (postparent);
+        if (!op_ret && local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
+        }
 
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
         DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
                           postparent);
 err:
         return 0;
 }
 
+/* For directories, check if acl xattrs have been requested (by the acl xlator),
+ * if not, request for them. These xattrs are needed for dht dir self-heal to
+ * perform proper self-healing of dirs
+ */
+void
+dht_check_and_set_acl_xattr_req (inode_t *inode, dict_t *xattr_req)
+{
+        int     ret = 0;
+
+        GF_ASSERT (inode);
+        GF_ASSERT (xattr_req);
+
+        if (inode->ia_type != IA_IFDIR)
+                return;
+
+        if (!dict_get (xattr_req, POSIX_ACL_ACCESS_XATTR)) {
+                ret = dict_set_int8 (xattr_req, POSIX_ACL_ACCESS_XATTR, 0);
+                if (ret)
+                        gf_log (THIS->name, GF_LOG_WARNING,
+                                "failed to set key %s",
+                                POSIX_ACL_ACCESS_XATTR);
+        }
+
+        if (!dict_get (xattr_req, POSIX_ACL_DEFAULT_XATTR)) {
+                ret = dict_set_int8 (xattr_req, POSIX_ACL_DEFAULT_XATTR, 0);
+                if (ret)
+                        gf_log (THIS->name, GF_LOG_WARNING,
+                                "failed to set key %s",
+                                POSIX_ACL_DEFAULT_XATTR);
+        }
+
+        return;
+}
 
 int
 dht_lookup (call_frame_t *frame, xlator_t *this,
@@ -965,7 +1402,6 @@ dht_lookup (call_frame_t *frame, xlator_t *this,
 {
         xlator_t     *subvol = NULL;
         xlator_t     *hashed_subvol = NULL;
-        xlator_t     *cached_subvol = NULL;
         dht_local_t  *local  = NULL;
         dht_conf_t   *conf = NULL;
         int           ret    = -1;
@@ -973,26 +1409,33 @@ dht_lookup (call_frame_t *frame, xlator_t *this,
         dht_layout_t *layout = NULL;
         int           i = 0;
         int           call_cnt = 0;
-
+        loc_t         new_loc = {0,};
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
 
         conf = this->private;
         if (!conf)
                 goto err;
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_LOOKUP);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
-        if (!dht_filter_loc_subvol_key (this, loc, &local->loc,
-                                        &hashed_subvol)) {
-                ret = loc_dup (loc, &local->loc);
+
+        ret = dht_filter_loc_subvol_key (this, loc, &new_loc,
+                                         &hashed_subvol);
+        if (ret) {
+                loc_wipe (&local->loc);
+                ret = loc_dup (&new_loc, &local->loc);
+
+                /* we no more need 'new_loc' entries */
+                loc_wipe (&new_loc);
+
+                /* check if loc_dup() is successful */
                 if (ret == -1) {
                         op_errno = errno;
                         gf_log (this->name, GF_LOG_DEBUG,
@@ -1008,16 +1451,19 @@ dht_lookup (call_frame_t *frame, xlator_t *this,
                 local->xattr_req = dict_new ();
         }
 
+        if (uuid_is_null (loc->pargfid) && !uuid_is_null (loc->gfid) &&
+            !__is_root_gfid (loc->inode->gfid)) {
+                local->cached_subvol = NULL;
+                dht_discover (frame, this, loc);
+                return 0;
+        }
+
         if (!hashed_subvol)
                 hashed_subvol = dht_subvol_get_hashed (this, loc);
-        cached_subvol = dht_subvol_get_cached (this, loc->inode);
-
-        local->cached_subvol = cached_subvol;
         local->hashed_subvol = hashed_subvol;
 
         if (is_revalidate (loc)) {
-                local->layout = layout = dht_layout_get (this, loc->inode);
-
+                layout = local->layout;
                 if (!layout) {
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "revalidate without cache. path=%s",
@@ -1033,38 +1479,63 @@ dht_lookup (call_frame_t *frame, xlator_t *this,
 
                         dht_layout_unref (this, local->layout);
                         local->layout = NULL;
+                        local->cached_subvol = NULL;
                         goto do_fresh_lookup;
                 }
 
-                local->inode    = inode_ref (loc->inode);
-
-                local->call_cnt = layout->cnt;
-                call_cnt = local->call_cnt;
+                local->inode = inode_ref (loc->inode);
 
                 /* NOTE: we don't require 'trusted.glusterfs.dht.linkto' attribute,
                  *       revalidates directly go to the cached-subvolume.
                  */
                 ret = dict_set_uint32 (local->xattr_req,
-                                       "trusted.glusterfs.dht", 4 * 4);
+                                       conf->xattr_name, 4 * 4);
 
-		for (i = 0; i < layout->cnt; i++) {
+                if (IA_ISDIR (local->inode->ia_type)) {
+                        local->call_cnt = call_cnt = conf->subvolume_cnt;
+                        for (i = 0; i < call_cnt; i++) {
+                                STACK_WIND (frame, dht_revalidate_cbk,
+                                            conf->subvolumes[i],
+                                            conf->subvolumes[i]->fops->lookup,
+                                            loc, local->xattr_req);
+                        }
+                        return 0;
+                }
+
+                call_cnt = local->call_cnt = layout->cnt;
+
+                /* need it for self-healing linkfiles which is
+                   'in-migration' state */
+                ret = dict_set_uint32 (local->xattr_req,
+                                       GLUSTERFS_OPEN_FD_COUNT, 4);
+
+                /* need it for dir self-heal */
+                dht_check_and_set_acl_xattr_req (loc->inode, local->xattr_req);
+
+		for (i = 0; i < call_cnt; i++) {
 			subvol = layout->list[i].xlator;
 
 			STACK_WIND (frame, dht_revalidate_cbk,
 				    subvol, subvol->fops->lookup,
 				    &local->loc, local->xattr_req);
 
-			if (!--call_cnt)
-				break;
 		}
         } else {
         do_fresh_lookup:
                 /* TODO: remove the hard-coding */
                 ret = dict_set_uint32 (local->xattr_req,
-                                       "trusted.glusterfs.dht", 4 * 4);
+                                       conf->xattr_name, 4 * 4);
+
+                ret = dict_set_uint32 (local->xattr_req,
+                                       conf->link_xattr_name, 256);
 
+                /* need it for self-healing linkfiles which is
+                   'in-migration' state */
                 ret = dict_set_uint32 (local->xattr_req,
-                                       "trusted.glusterfs.dht.linkto", 256);
+                                       GLUSTERFS_OPEN_FD_COUNT, 4);
+
+                /* need it for dir self-heal */
+                dht_check_and_set_acl_xattr_req (loc->inode, local->xattr_req);
 
                 if (!hashed_subvol) {
                         gf_log (this->name, GF_LOG_DEBUG,
@@ -1099,287 +1570,8 @@ dht_lookup (call_frame_t *frame, xlator_t *this,
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int
-dht_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno, struct iatt *prebuf,
-                  struct iatt *postbuf)
-{
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
-
-        local = frame->local;
-        prev = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        local->op_ret = -1;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
-
-                dht_iatt_merge (this, &local->prebuf, prebuf, prev->this);
-                dht_iatt_merge (this, &local->stbuf, postbuf, prev->this);
-
-                local->op_ret = 0;
-        }
-unlock:
-        UNLOCK (&frame->lock);
-out:
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_UNWIND (truncate, frame, local->op_ret, local->op_errno,
-                                  &local->prebuf, &local->stbuf);
-err:
-        return 0;
-}
-
-
-
-int
-dht_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-              int op_ret, int op_errno, struct iatt *stbuf)
-{
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
-
-        local = frame->local;
-        prev = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
-
-                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-
-                local->op_ret = 0;
-        }
-unlock:
-        UNLOCK (&frame->lock);
-out:
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_UNWIND (stat, frame, local->op_ret, local->op_errno,
-                                  &local->stbuf);
-err:
-        return 0;
-}
-
-
-int
-dht_stat (call_frame_t *frame, xlator_t *this,
-          loc_t *loc)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-        dht_layout_t *layout = NULL;
-        int           i = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        local->layout = layout = dht_layout_get (this, loc->inode);
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local->inode = inode_ref (loc->inode);
-        local->call_cnt = layout->cnt;
-
-        for (i = 0; i < layout->cnt; i++) {
-                subvol = layout->list[i].xlator;
-
-                STACK_WIND (frame, dht_attr_cbk,
-                            subvol, subvol->fops->stat,
-                            loc);
-        }
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL);
-
-        return 0;
-}
-
-
-int
-dht_fstat (call_frame_t *frame, xlator_t *this,
-           fd_t *fd)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-        dht_layout_t *layout = NULL;
-        int           i = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        local->layout = layout = dht_layout_get (this, fd->inode);
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local->inode    = inode_ref (fd->inode);
-        local->call_cnt = layout->cnt;;
-
-        for (i = 0; i < layout->cnt; i++) {
-                subvol = layout->list[i].xlator;
-                STACK_WIND (frame, dht_attr_cbk,
-                            subvol, subvol->fops->fstat,
-                            fd);
-        }
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fstat, frame, -1, op_errno, NULL);
-
-        return 0;
-}
-
-
-int
-dht_truncate (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, off_t offset)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        subvol = dht_subvol_get_cached (this, loc->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        local->inode = inode_ref (loc->inode);
-        local->call_cnt = 1;
-
-        STACK_WIND (frame, dht_truncate_cbk,
-                    subvol, subvol->fops->truncate,
-                    loc, offset);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL);
-
-        return 0;
-}
-
-
-int
-dht_ftruncate (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, off_t offset)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        local->inode = inode_ref (fd->inode);
-        local->call_cnt = 1;
-
-        STACK_WIND (frame, dht_truncate_cbk,
-                    subvol, subvol->fops->ftruncate,
-                    fd, offset);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL);
-
+        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL);
         return 0;
 }
 
@@ -1387,7 +1579,7 @@ err:
 int
 dht_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 int op_ret, int op_errno, struct iatt *preparent,
-                struct iatt *postparent)
+                struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         call_frame_t *prev  = NULL;
@@ -1411,14 +1603,18 @@ dht_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 local->postparent = *postparent;
                 local->preparent = *preparent;
 
-                WIPE (&local->postparent);
-                WIPE (&local->preparent);
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->preparent, 0);
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
         }
 unlock:
         UNLOCK (&frame->lock);
 
         DHT_STACK_UNWIND (unlink, frame, local->op_ret, local->op_errno,
-                          &local->preparent, &local->postparent);
+                          &local->preparent, &local->postparent, NULL);
 
         return 0;
 }
@@ -1427,7 +1623,7 @@ unlock:
 int
 dht_unlink_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                          int op_ret, int op_errno, struct iatt *preparent,
-                         struct iatt *postparent)
+                         struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         call_frame_t *prev = NULL;
@@ -1439,7 +1635,8 @@ dht_unlink_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         LOCK (&frame->lock);
         {
-                if (op_ret == -1) {
+                if ((op_ret == -1) && !((op_errno == ENOENT) ||
+                                        (op_errno == ENOTCONN))) {
                         local->op_errno = op_errno;
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "subvolume %s returned -1 (%s)",
@@ -1452,7 +1649,7 @@ dht_unlink_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 unlock:
         UNLOCK (&frame->lock);
 
-        if (op_ret == -1)
+        if (local->op_ret == -1)
                 goto err;
 
         cached_subvol = dht_subvol_get_cached (this, local->loc.inode);
@@ -1466,26 +1663,24 @@ unlock:
 
         STACK_WIND (frame, dht_unlink_cbk,
                     cached_subvol, cached_subvol->fops->unlink,
-                    &local->loc);
+                    &local->loc, local->flags, NULL);
 
         return 0;
 
 err:
         DHT_STACK_UNWIND (unlink, frame, -1, local->op_errno,
-                          NULL, NULL);
+                          NULL, NULL, NULL);
         return 0;
 }
 
-
 int
-dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
-               int op_errno, struct iatt *prebuf, struct iatt *postbuf)
+dht_err_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+             int op_ret, int op_errno, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         int           this_call_cnt = 0;
         call_frame_t *prev = NULL;
 
-
         local = frame->local;
         prev = cookie;
 
@@ -1505,292 +1700,277 @@ unlock:
         UNLOCK (&frame->lock);
 
         this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_UNWIND (fsync, frame, local->op_ret, local->op_errno,
-                                  prebuf, postbuf);
+        if (is_last_call (this_call_cnt)) {
+                DHT_STACK_UNWIND (setxattr, frame, local->op_ret,
+                                  local->op_errno, NULL);
+        }
 
         return 0;
 }
 
-
-
-int
-dht_err_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-             int op_ret, int op_errno)
+static void
+fill_layout_info (dht_layout_t *layout, char *buf)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
-
-
-        local = frame->local;
-        prev = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
-
-                local->op_ret = 0;
-        }
-unlock:
-        UNLOCK (&frame->lock);
+        int i = 0;
+        char tmp_buf[128] = {0,};
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                DHT_STACK_UNWIND (setxattr, frame, local->op_ret, local->op_errno);
+        for (i = 0; i < layout->cnt; i++) {
+                snprintf (tmp_buf, 128, "(%s %u %u)",
+                          layout->list[i].xlator->name,
+                          layout->list[i].start,
+                          layout->list[i].stop);
+                if (i)
+                        strcat (buf, " ");
+                strcat (buf, tmp_buf);
         }
-
-        return 0;
 }
 
-
-int
-dht_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-             int op_ret, int op_errno)
+void
+dht_fill_pathinfo_xattr (xlator_t *this, dht_local_t *local,
+                         char *xattr_buf, int32_t alloc_len,
+                         int flag, char *layout_buf)
 {
-        DHT_STACK_UNWIND (access, frame, op_ret, op_errno);
-        return 0;
+        if (flag && local->xattr_val)
+                snprintf (xattr_buf, alloc_len,
+                          "((<"DHT_PATHINFO_HEADER"%s> %s) (%s-layout %s))",
+                          this->name, local->xattr_val, this->name,
+                          layout_buf);
+        else if (local->xattr_val)
+                snprintf (xattr_buf, alloc_len,
+                          "(<"DHT_PATHINFO_HEADER"%s> %s)",
+                          this->name, local->xattr_val);
+        else if (flag)
+                snprintf (xattr_buf, alloc_len, "(%s-layout %s)",
+                          this->name, layout_buf);
 }
 
-
 int
-dht_access (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, int32_t mask)
+dht_vgetxattr_alloc_and_fill (dht_local_t *local, dict_t *xattr, xlator_t *this,
+                              int op_errno)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        subvol = dht_subvol_get_cached (this, loc->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+        int      ret       = -1;
+        char    *value     = NULL;
+        int32_t  plen      = 0;
 
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
+        ret = dict_get_str (xattr, local->xsel, &value);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Subvolume %s returned -1 (%s)", this->name,
+                        strerror (op_errno));
+                local->op_ret = -1;
+                local->op_errno = op_errno;
+                goto out;
         }
 
-        local->call_cnt = 1;
-
-        STACK_WIND (frame, dht_access_cbk,
-                    subvol, subvol->fops->access,
-                    loc, mask);
-
-        return 0;
+        local->alloc_len += strlen(value);
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (access, frame, -1, op_errno);
-
-        return 0;
-}
-
-
-int
-dht_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno, const char *path, struct iatt *sbuf)
-{
-        dht_local_t *local = NULL;
+        if (!local->xattr_val) {
+                local->alloc_len += (strlen (DHT_PATHINFO_HEADER) + 10);
+                local->xattr_val = GF_CALLOC (local->alloc_len, sizeof (char),
+                                              gf_common_mt_char);
+                if (!local->xattr_val) {
+                        ret = -1;
+                        goto out;
+                }
+        }
 
-        local = frame->local;
-        if (op_ret == -1)
-                goto err;
+        if (local->xattr_val) {
+                plen = strlen (local->xattr_val);
+                if (plen) {
+                        /* extra byte(s) for \0 to be safe */
+                        local->alloc_len += (plen + 2);
+                        local->xattr_val = GF_REALLOC (local->xattr_val,
+                                                       local->alloc_len);
+                        if (!local->xattr_val) {
+                                ret = -1;
+                                goto out;
+                        }
+                }
 
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
+                (void) strcat (local->xattr_val, value);
+                (void) strcat (local->xattr_val, " ");
+                local->op_ret = 0;
         }
 
-err:
-        DHT_STACK_UNWIND (readlink, frame, op_ret, op_errno, path, sbuf);
+        ret = 0;
 
-        return 0;
+ out:
+        return ret;
 }
 
-
 int
-dht_readlink (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, size_t size)
+dht_vgetxattr_fill_and_set (dht_local_t *local, dict_t **dict, xlator_t *this,
+                            gf_boolean_t flag)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        subvol = dht_subvol_get_cached (this, loc->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+        int   ret             = -1;
+        char *xattr_buf       = NULL;
+        char layout_buf[8192] = {0,};
 
-        STACK_WIND (frame, dht_readlink_cbk,
-                    subvol, subvol->fops->readlink,
-                    loc, size);
+        if (flag)
+                fill_layout_info (local->layout, layout_buf);
 
-        return 0;
+        *dict = dict_new ();
+        if (!*dict)
+                goto out;
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (readlink, frame, -1, op_errno, NULL, NULL);
+        local->xattr_val[strlen (local->xattr_val) - 1] = '\0';
+
+        /* we would need max this many bytes to create xattr string
+         * extra 40 bytes is just an estimated amount of additional
+         * space required as we include translator name and some
+         * spaces, brackets etc. when forming the pathinfo string.
+         *
+         * For node-uuid we just don't have all the pretty formatting,
+         * but since this is a generic routine for pathinfo & node-uuid
+         * we dont have conditional space allocation and try to be
+         * generic
+         */
+        local->alloc_len += (2 * strlen (this->name))
+                + strlen (layout_buf)
+                + 40;
+        xattr_buf = GF_CALLOC (local->alloc_len, sizeof (char),
+                               gf_common_mt_char);
+        if (!xattr_buf)
+                goto out;
 
-        return 0;
-}
+        if (XATTR_IS_PATHINFO (local->xsel)) {
+                (void) dht_fill_pathinfo_xattr (this, local, xattr_buf,
+                                                local->alloc_len, flag,
+                                                layout_buf);
+        } else if (XATTR_IS_NODE_UUID (local->xsel)) {
+                (void) snprintf (xattr_buf, local->alloc_len, "%s",
+                                 local->xattr_val);
+        } else {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Unknown local->xsel (%s)", local->xsel);
+                GF_FREE (xattr_buf);
+                goto out;
+        }
 
+        ret = dict_set_dynstr (*dict, local->xsel, xattr_buf);
+        if (ret)
+                GF_FREE (xattr_buf);
+        GF_FREE (local->xattr_val);
 
-static void
-fill_layout_info (dht_layout_t *layout, char *buf)
-{
-        int i = 0;
-        char tmp_buf[128] = {0,};
-
-        for (i = 0; i < layout->cnt; i++) {
-                snprintf (tmp_buf, 128, "(%s %u %u)",
-                          layout->list[i].xlator->name,
-                          layout->list[i].start,
-                          layout->list[i].stop);
-                if (i)
-                        strcat (buf, " ");
-                strcat (buf, tmp_buf);
-        }
+ out:
+        return ret;
 }
 
 int
-dht_pathinfo_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int op_ret, int op_errno, dict_t *xattr)
+dht_vgetxattr_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
 {
-        dht_local_t *local         = NULL;
         int          ret           = 0;
-        int          flag          = 0;
+        dht_local_t *local         = NULL;
         int          this_call_cnt = 0;
-        char        *value_got     = NULL;
-        char  layout_buf[8192]     = {0,};
-        char        *xattr_buf     = NULL;
         dict_t      *dict          = NULL;
-        int32_t      alloc_len     = 0;
-        int32_t      plen          = 0;
 
-        local = frame->local;
-
-        if (op_ret != -1) {
-                ret = dict_get_str (xattr, GF_XATTR_PATHINFO_KEY, &value_got);
-                if (!ret) {
-                        alloc_len = strlen (value_got);
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (frame->local, out);
 
-                        /**
-                         * allocate the buffer:- we allocate 10 bytes extra in case we need to
-                         * append ' Link: ' in the buffer for another STACK_WIND
-                         */
-                        if (!local->pathinfo)
-                                local->pathinfo = GF_CALLOC (alloc_len + strlen (DHT_PATHINFO_HEADER) + 10,
-                                                             sizeof (char), gf_common_mt_char);
-
-                        if (local->pathinfo) {
-                                plen = strlen (local->pathinfo);
-                                if (plen) {
-                                        alloc_len += plen;
-                                        local->pathinfo = GF_REALLOC (local->pathinfo,
-                                                                      alloc_len);
-                                        if (!local->pathinfo)
-                                                goto out;
-                                }
+        local = frame->local;
 
-                                strcat (local->pathinfo, value_got);
+        LOCK (&frame->lock);
+        {
+                this_call_cnt = --local->call_cnt;
+                if (op_ret < 0) {
+                        if (op_errno != ENOTCONN) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "getxattr err (%s) for dir",
+                                        strerror (op_errno));
+				local->op_ret = -1;
+				local->op_errno = op_errno;
                         }
+
+                        goto unlock;
                 }
+
+                ret = dht_vgetxattr_alloc_and_fill (local, xattr, this,
+                                                    op_errno);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "alloc or fill failure");
         }
+ unlock:
+        UNLOCK (&frame->lock);
 
- out:
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                if (local->layout->cnt > 1) {
-                        /* Set it for directory */
-                        fill_layout_info (local->layout, layout_buf);
-                        flag = 1;
-                }
+        if (!is_last_call (this_call_cnt))
+                goto out;
 
-                dict = dict_new ();
+        /* -- last call: do patch ups -- */
 
-                /* we would need max-to-max this many bytes to create pathinfo string */
-                alloc_len += (2 * strlen (this->name)) + strlen (layout_buf) + 40;
-                xattr_buf = GF_CALLOC (alloc_len, sizeof (char), gf_common_mt_char);
+        if (local->op_ret == -1) {
+                goto unwind;
+        }
 
-                if (flag && local->pathinfo)
-                        snprintf (xattr_buf, alloc_len, "((<"DHT_PATHINFO_HEADER"%s> %s) (%s-layout %s))",
-                                  this->name, local->pathinfo, this->name,
-                                  layout_buf);
-                else if (local->pathinfo)
-                        snprintf (xattr_buf, alloc_len, "(<"DHT_PATHINFO_HEADER"%s> %s)",
-                                  this->name, local->pathinfo);
-                else if (flag)
-                        snprintf (xattr_buf, alloc_len, "(%s-layout %s)",
-                                  this->name, layout_buf);
+        ret = dht_vgetxattr_fill_and_set (local, &dict, this, _gf_true);
+        if (ret)
+                goto unwind;
 
-                ret = dict_set_dynstr (dict, GF_XATTR_PATHINFO_KEY,
-                                       xattr_buf);
+        DHT_STACK_UNWIND (getxattr, frame, 0, 0, dict, xdata);
+        goto cleanup;
 
-                if (local->pathinfo)
-                        GF_FREE (local->pathinfo);
+ unwind:
+        DHT_STACK_UNWIND (getxattr, frame, -1, local->op_errno, NULL, NULL);
+ cleanup:
+        if (dict)
+                dict_unref (dict);
+ out:
+        return 0;
+}
 
-                DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
+int
+dht_vgetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+        dht_local_t  *local         = NULL;
+        int           ret           = 0;
+        dict_t       *dict          = NULL;
+        call_frame_t *prev          = NULL;
+        gf_boolean_t  flag          = _gf_true;
 
-                if (dict)
-                        dict_unref (dict);
+        local = frame->local;
+        prev = cookie;
 
-                return 0;
+        if (op_ret < 0) {
+                local->op_ret = -1;
+                local->op_errno = op_errno;
+                gf_log (this->name, GF_LOG_ERROR, "Subvolume %s returned -1 "
+                        "(%s)", prev->this->name, strerror (op_errno));
+                goto unwind;
         }
 
-        if (local->pathinfo)
-                strcat (local->pathinfo, " Link: ");
-        if (local->hashed_subvol) {
-                /* This will happen if there pending */
-                STACK_WIND (frame, dht_pathinfo_getxattr_cbk, local->hashed_subvol,
-                            local->hashed_subvol->fops->getxattr,
-                            &local->loc, local->key);
-
-                return 0;
+        ret = dht_vgetxattr_alloc_and_fill (local, xattr, this,
+                                            op_errno);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "alloc or fill failure");
+                goto unwind;
         }
 
-        gf_log ("this->name", GF_LOG_ERROR, "Unable to find hashed_subvol for path"
-                " %s", local->pathinfo);
+        flag = (local->layout->cnt > 1) ? _gf_true : _gf_false;
+
+        ret = dht_vgetxattr_fill_and_set (local, &dict, this, flag);
+        if (ret)
+                goto unwind;
+
+        DHT_STACK_UNWIND (getxattr, frame, 0, 0, dict, xdata);
+        goto cleanup;
+
+ unwind:
+        DHT_STACK_UNWIND (getxattr, frame, -1, local->op_errno,
+                          NULL, NULL);
+ cleanup:
+        if (dict)
+                dict_unref (dict);
 
-        DHT_STACK_UNWIND (getxattr, frame, -1, op_errno, dict);
         return 0;
 }
 
 int
 dht_linkinfo_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int op_ret, int op_errno, dict_t *xattr)
+                           int op_ret, int op_errno, dict_t *xattr,
+                           dict_t *xdata)
 {
         int   ret   = 0;
         char *value = NULL;
@@ -1805,21 +1985,24 @@ dht_linkinfo_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 }
         }
 
-        DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr);
+        DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr, xdata);
 
         return 0;
 }
 
 int
 dht_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno, dict_t *xattr)
+                  int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
 {
         int             this_call_cnt = 0;
         dht_local_t     *local = NULL;
+        dht_conf_t      *conf = NULL;
 
         VALIDATE_OR_GOTO (frame, out);
         VALIDATE_OR_GOTO (frame->local, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
+        conf = this->private;
         local = frame->local;
 
         this_call_cnt = dht_frame_return (frame);
@@ -1827,39 +2010,106 @@ dht_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         if (!xattr || (op_ret == -1))
                 goto out;
 
-        if (dict_get (xattr, "trusted.glusterfs.dht")) {
-                dict_del (xattr, "trusted.glusterfs.dht");
+        if (dict_get (xattr, conf->xattr_name)) {
+                dict_del (xattr, conf->xattr_name);
+        }
+
+        if (frame->root->pid >= 0 ) {
+                GF_REMOVE_INTERNAL_XATTR("trusted.glusterfs.quota*", xattr);
+                GF_REMOVE_INTERNAL_XATTR("trusted.pgfid*", xattr);
         }
+
         local->op_ret = 0;
 
         if (!local->xattr) {
                 local->xattr = dict_copy_with_ref (xattr, NULL);
         } else {
-                /* first aggregate everything into xattr and then copy into
-                 * local->xattr.
-                 */
-                dht_aggregate_xattr (xattr, local->xattr);
+                dht_aggregate_xattr (local->xattr, xattr);
         }
 out:
         if (is_last_call (this_call_cnt)) {
-                DHT_STACK_UNWIND (getxattr, frame, local->op_ret, op_errno, local->xattr);
+                DHT_STACK_UNWIND (getxattr, frame, local->op_ret, op_errno,
+                                  local->xattr, NULL);
         }
         return 0;
 }
 
 int32_t
 dht_getxattr_unwind (call_frame_t *frame,
-                     int op_ret, int op_errno, dict_t *dict)
+                     int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
 {
-        DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
+        DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, xdata);
         return 0;
 }
 
 
 int
+dht_getxattr_get_real_filename_cbk (call_frame_t *frame, void *cookie,
+				    xlator_t *this, int op_ret, int op_errno,
+				    dict_t *xattr, dict_t *xdata)
+{
+        int             this_call_cnt = 0;
+        dht_local_t     *local = NULL;
+
+
+        local = frame->local;
+
+	if (op_ret != -1) {
+		if (local->xattr)
+			dict_unref (local->xattr);
+		local->xattr = dict_ref (xattr);
+
+		if (local->xattr_req)
+			dict_unref (local->xattr_req);
+		local->xattr_req = dict_ref (xdata);
+	}
+
+	this_call_cnt = dht_frame_return (frame);
+	if (is_last_call (this_call_cnt)) {
+		DHT_STACK_UNWIND (getxattr, frame, local->op_ret, op_errno,
+				  local->xattr, local->xattr_req);
+	}
+
+	return 0;
+}
+
+
+int
+dht_getxattr_get_real_filename (call_frame_t *frame, xlator_t *this,
+				loc_t *loc, const char *key, dict_t *xdata)
+{
+	dht_local_t     *local = NULL;
+	int              i = 0;
+	dht_layout_t    *layout = NULL;
+	int              cnt = 0;
+	xlator_t        *subvol = NULL;
+
+
+	local = frame->local;
+	layout = local->layout;
+
+	cnt = local->call_cnt = layout->cnt;
+
+	local->op_ret = -1;
+	local->op_errno = ENODATA;
+
+	for (i = 0; i < cnt; i++) {
+		subvol = layout->list[i].xlator;
+		STACK_WIND (frame, dht_getxattr_get_real_filename_cbk,
+			    subvol, subvol->fops->getxattr,
+			    loc, key, xdata);
+	}
+
+	return 0;
+}
+
+
+int
 dht_getxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, const char *key)
+              loc_t *loc, const char *key, dict_t *xdata)
+#define DHT_IS_DIR(layout)  (layout->cnt > 1)
 {
+
         xlator_t     *subvol        = NULL;
         xlator_t     *hashed_subvol = NULL;
         xlator_t     *cached_subvol = NULL;
@@ -1868,7 +2118,6 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
         dht_layout_t *layout        = NULL;
         xlator_t     **sub_volumes  = NULL;
         int           op_errno      = -1;
-        int           ret           = 0;
         int           i             = 0;
         int           cnt           = 0;
 
@@ -1876,57 +2125,94 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
         VALIDATE_OR_GOTO (this->private, err);
 
         conf   = this->private;
-        layout = dht_layout_get (this, loc->inode);
-        if (!layout) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "layout is NULL");
-                op_errno = ENOENT;
-                goto err;
-        }
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_GETXATTR);
         if (!local) {
                 op_errno = ENOMEM;
 
                 goto err;
         }
 
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "layout is NULL");
+                op_errno = ENOENT;
                 goto err;
         }
-        local->layout = layout;
-
-        if (key && (strcmp (key, GF_XATTR_PATHINFO_KEY) == 0)) {
-                hashed_subvol = dht_subvol_get_hashed (this, loc);
-                cached_subvol = dht_subvol_get_cached (this, loc->inode);
 
+        if (key) {
                 local->key = gf_strdup (key);
                 if (!local->key) {
                         op_errno = ENOMEM;
-
                         goto err;
                 }
+        }
 
-                local->call_cnt = 1;
-                if (hashed_subvol != cached_subvol) {
-                        local->call_cnt = 2;
-                        local->hashed_subvol = hashed_subvol;
+	if (key &&
+	    (strncmp (key, GF_XATTR_GET_REAL_FILENAME_KEY,
+		      strlen (GF_XATTR_GET_REAL_FILENAME_KEY)) == 0)
+	    && DHT_IS_DIR(layout)) {
+		dht_getxattr_get_real_filename (frame, this, loc, key, xdata);
+		return 0;
+	}
+
+        /* for file use cached subvolume (obviously!): see if {}
+         * below
+         * for directory:
+         *  wind to all subvolumes and exclude subvolumes which
+         *  return ENOTCONN (in callback)
+         *
+         * NOTE: Don't trust inode here, as that may not be valid
+         *       (until inode_link() happens)
+         */
+        if (key && DHT_IS_DIR(layout) &&
+            (XATTR_IS_PATHINFO (key)
+             || (strcmp (key, GF_XATTR_NODE_UUID_KEY) == 0))) {
+                (void) strncpy (local->xsel, key, 256);
+                cnt = local->call_cnt = layout->cnt;
+                for (i = 0; i < cnt; i++) {
+                        subvol = layout->list[i].xlator;
+                        STACK_WIND (frame, dht_vgetxattr_dir_cbk,
+                                    subvol, subvol->fops->getxattr,
+                                    loc, key, NULL);
                 }
+                return 0;
+        }
 
-                STACK_WIND (frame, dht_pathinfo_getxattr_cbk, cached_subvol,
-                            cached_subvol->fops->getxattr, loc, key);
+        /* node-uuid or pathinfo for files */
+        if (key && ((strcmp (key, GF_XATTR_NODE_UUID_KEY) == 0)
+                    || XATTR_IS_PATHINFO (key))) {
+                cached_subvol = local->cached_subvol;
+                (void) strncpy (local->xsel, key, 256);
+
+                local->call_cnt = 1;
+                STACK_WIND (frame, dht_vgetxattr_cbk, cached_subvol,
+                            cached_subvol->fops->getxattr, loc, key, NULL);
 
                 return 0;
         }
+
         if (key && (strcmp (key, GF_XATTR_LINKINFO_KEY) == 0)) {
                 hashed_subvol = dht_subvol_get_hashed (this, loc);
+                if (!hashed_subvol) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to get"
+                                "hashed subvol for %s", loc->path);
+                        op_errno = EINVAL;
+                        goto err;
+                }
+
                 cached_subvol = dht_subvol_get_cached (this, loc->inode);
+                if (!cached_subvol) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to get"
+                                "cached subvol for %s", loc->path);
+                        op_errno = EINVAL;
+                        goto err;
+                }
+
                 if (hashed_subvol == cached_subvol) {
                         op_errno = ENODATA;
                         goto err;
@@ -1934,7 +2220,7 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
                 if (hashed_subvol) {
                         STACK_WIND (frame, dht_linkinfo_getxattr_cbk, hashed_subvol,
                                     hashed_subvol->fops->getxattr, loc,
-                                    GF_XATTR_PATHINFO_KEY);
+                                    GF_XATTR_PATHINFO_KEY, NULL);
                         return 0;
                 }
                 op_errno = ENODATA;
@@ -1942,13 +2228,13 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
         }
 
         if (key && (!strcmp (GF_XATTR_MARKER_KEY, key))
-            && (-1 == frame->root->pid)) {
-
-                if (loc->inode-> ia_type == IA_IFDIR) {
+            && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
+                if (DHT_IS_DIR(layout)) {
                         cnt = layout->cnt;
                 } else {
                         cnt = 1;
                 }
+
                 sub_volumes = alloca ( cnt * sizeof (xlator_t *));
                 for (i = 0; i < cnt; i++)
                         *(sub_volumes + i) = layout->list[i].xlator;
@@ -1956,7 +2242,8 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
                 if (cluster_getmarkerattr (frame, this, loc, key,
                                            local, dht_getxattr_unwind,
                                            sub_volumes, cnt,
-                                           MARKER_UUID_TYPE, conf->vol_uuid)) {
+                                           MARKER_UUID_TYPE, marker_uuid_default_gauge,
+                                           conf->vol_uuid)) {
                         op_errno = EINVAL;
                         goto err;
                 }
@@ -1964,10 +2251,22 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
                 return 0;
         }
 
+        if (key && !strcmp (GF_XATTR_QUOTA_LIMIT_LIST, key)) {
+                /* quota hardlimit and aggregated size of a directory is stored
+                 * in inode contexts of each brick. Hence its good enough that
+                 * we send getxattr for this key to any brick.
+                 */
+                local->call_cnt = 1;
+                subvol = dht_first_up_subvol (this);
+                STACK_WIND (frame, dht_getxattr_cbk, subvol,
+                            subvol->fops->getxattr, loc, key, xdata);
+                return 0;
+        }
+
         if (key && *conf->vol_uuid) {
                 if ((match_uuid_local (key, conf->vol_uuid) == 0) &&
-                    (-1 == frame->root->pid)) {
-                        if (loc->inode-> ia_type == IA_IFDIR) {
+                    (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
+                        if (DHT_IS_DIR(layout)) {
                                 cnt = layout->cnt;
                         } else {
                                 cnt = 1;
@@ -1980,6 +2279,7 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
                                                    local, dht_getxattr_unwind,
                                                    sub_volumes, cnt,
                                                    MARKER_XTIME_TYPE,
+                                                   marker_xtime_default_gauge,
                                                    conf->vol_uuid)) {
                                 op_errno = EINVAL;
                                 goto err;
@@ -1989,6 +2289,60 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
                 }
         }
 
+        if (DHT_IS_DIR(layout)) {
+                cnt = local->call_cnt = layout->cnt;
+        } else {
+                cnt = local->call_cnt  = 1;
+        }
+
+        for (i = 0; i < cnt; i++) {
+                subvol = layout->list[i].xlator;
+                STACK_WIND (frame, dht_getxattr_cbk,
+                            subvol, subvol->fops->getxattr,
+                            loc, key, NULL);
+        }
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+#undef DHT_IS_DIR
+
+int
+dht_fgetxattr (call_frame_t *frame, xlator_t *this,
+               fd_t *fd, const char *key, dict_t *xdata)
+{
+        xlator_t     *subvol        = NULL;
+        dht_local_t  *local         = NULL;
+        dht_layout_t *layout        = NULL;
+        int           op_errno      = -1;
+        int           i             = 0;
+        int           cnt           = 0;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (fd->inode, err);
+        VALIDATE_OR_GOTO (this->private, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FGETXATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+
+                goto err;
+        }
+
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "layout is NULL");
+                op_errno = ENOENT;
+                goto err;
+        }
+
         if (key) {
                 local->key = gf_strdup (key);
                 if (!local->key) {
@@ -1997,7 +2351,10 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
                 }
         }
 
-        if (loc->inode-> ia_type == IA_IFDIR) {
+        if ((fd->inode->ia_type == IA_IFDIR)
+	    && key
+            && (strncmp (key, GF_XATTR_LOCKINFO_KEY,
+                         strlen (GF_XATTR_LOCKINFO_KEY) != 0))) {
                 cnt = local->call_cnt = layout->cnt;
         } else {
                 cnt = local->call_cnt  = 1;
@@ -2006,32 +2363,45 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
         for (i = 0; i < cnt; i++) {
                 subvol = layout->list[i].xlator;
                 STACK_WIND (frame, dht_getxattr_cbk,
-                            subvol, subvol->fops->getxattr,
-                            loc, key);
+                            subvol, subvol->fops->fgetxattr,
+                            fd, key, NULL);
         }
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL);
+        DHT_STACK_UNWIND (fgetxattr, frame, -1, op_errno, NULL, NULL);
 
         return 0;
 }
 
 int
 dht_fsetxattr (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, dict_t *xattr, int flags)
+               fd_t *fd, dict_t *xattr, int flags, dict_t *xdata)
 {
         xlator_t     *subvol   = NULL;
         dht_local_t  *local    = NULL;
         int           op_errno = EINVAL;
+        dht_conf_t   *conf     = NULL;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
         VALIDATE_OR_GOTO (fd->inode, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
-        subvol = dht_subvol_get_cached (this, fd->inode);
+        conf = this->private;
+
+        GF_IF_INTERNAL_XATTR_GOTO (conf->wild_xattr_name, xattr,
+                                   op_errno, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FSETXATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
         if (!subvol) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "no cached subvolume for fd=%p", fd);
@@ -2039,23 +2409,16 @@ dht_fsetxattr (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        local->inode = inode_ref (fd->inode);
         local->call_cnt = 1;
 
         STACK_WIND (frame, dht_err_cbk, subvol, subvol->fops->fsetxattr,
-                    fd, xattr, flags);
+                    fd, xattr, flags, NULL);
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fsetxattr, frame, -1, op_errno);
+        DHT_STACK_UNWIND (fsetxattr, frame, -1, op_errno, NULL);
 
         return 0;
 }
@@ -2063,16 +2426,58 @@ err:
 
 static int
 dht_common_setxattr_cbk (call_frame_t *frame, void *cookie,
-                         xlator_t *this, int32_t op_ret, int32_t op_errno)
+                         xlator_t *this, int32_t op_ret, int32_t op_errno,
+                         dict_t *xdata)
 {
-        DHT_STACK_UNWIND (setxattr, frame, op_ret, op_errno);
+        DHT_STACK_UNWIND (setxattr, frame, op_ret, op_errno, xdata);
 
         return 0;
 }
 
 int
+dht_checking_pathinfo_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                           int op_ret, int op_errno, dict_t *xattr,
+                           dict_t *xdata)
+{
+        int           i     = -1;
+        int           ret   = -1;
+        char         *value = NULL;
+        dht_local_t  *local = NULL;
+        dht_conf_t   *conf  = NULL;
+        call_frame_t *prev  = NULL;
+        int           this_call_cnt = 0;
+
+        local = frame->local;
+        prev = cookie;
+        conf = this->private;
+
+        if (op_ret == -1)
+                goto out;
+
+
+        ret = dict_get_str (xattr, GF_XATTR_PATHINFO_KEY, &value);
+        if (ret)
+                goto out;
+
+        if (!strcmp (value, local->key)) {
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                        if (conf->subvolumes[i] == prev->this)
+                                conf->decommissioned_bricks[i] = prev->this;
+                }
+        }
+
+out:
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+                DHT_STACK_UNWIND (setxattr, frame, local->op_ret, ENOTSUP, NULL);
+        }
+        return 0;
+
+}
+
+int
 dht_setxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, dict_t *xattr, int flags)
+              loc_t *loc, dict_t *xattr, int flags, dict_t *xdata)
 {
         xlator_t     *subvol   = NULL;
         dht_local_t  *local    = NULL;
@@ -2084,17 +2489,26 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
         data_t       *tmp      = NULL;
         uint32_t      dir_spread = 0;
         char          value[4096] = {0,};
-        int           forced_rebalance = 0;
-
+        gf_dht_migrate_data_type_t forced_rebalance = GF_DHT_MIGRATE_DATA;
+        int           call_cnt = 0;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
 
         conf   = this->private;
-        subvol = dht_subvol_get_cached (this, loc->inode);
+
+        GF_IF_INTERNAL_XATTR_GOTO (conf->wild_xattr_name, xattr,
+                                   op_errno, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_SETXATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
         if (!subvol) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "no cached subvolume for path=%s", loc->path);
@@ -2102,13 +2516,7 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        local->layout = layout = dht_layout_get (this, loc->inode);
+        layout = local->layout;
         if (!layout) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "no layout for path=%s", loc->path);
@@ -2116,15 +2524,11 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+        local->call_cnt = call_cnt = layout->cnt;
 
         tmp = dict_get (xattr, "distribute.migrate-data");
         if (tmp) {
-                if (!IA_ISREG (loc->inode->ia_type)) {
+                if (IA_ISDIR (loc->inode->ia_type)) {
                         op_errno = ENOTSUP;
                         goto err;
                 }
@@ -2133,15 +2537,27 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
                    (ie, 'target' subvolume given there, etc) */
                 memcpy (value, tmp->data, tmp->len);
                 if (strcmp (value, "force") == 0)
-                        forced_rebalance = 1;
+                       forced_rebalance =
+                                GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS;
 
-                local->to_subvol   = dht_subvol_get_hashed (this, loc);
-                local->from_subvol = dht_subvol_get_cached (this, loc->inode);
-                if (local->to_subvol == local->from_subvol) {
-                        op_errno = ENOTSUP;
+                if (conf->decommission_in_progress)
+                        forced_rebalance = GF_DHT_MIGRATE_HARDLINK;
+
+                local->rebalance.target_node = dht_subvol_get_hashed (this, loc);
+                if (!local->rebalance.target_node) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to get "
+                                "hashed subvol for %s", loc->path);
+                        op_errno = EINVAL;
                         goto err;
                 }
-                if (local->to_subvol) {
+
+                local->rebalance.from_subvol = local->cached_subvol;
+
+                if (local->rebalance.target_node == local->rebalance.from_subvol) {
+                        op_errno = EEXIST;
+                        goto err;
+                }
+                if (local->rebalance.target_node) {
                         local->flags = forced_rebalance;
 
                         ret = dht_start_rebalance_task (this, frame);
@@ -2152,19 +2568,45 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
                                 "%s: failed to create a new synctask",
                                 loc->path);
                 }
-                op_errno = ENOTSUP;
+                op_errno = EINVAL;
                 goto err;
 
         }
 
+        tmp = dict_get (xattr, "decommission-brick");
+        if (tmp) {
+                /* This operation should happen only on '/' */
+                if (!__is_root_gfid (loc->inode->gfid)) {
+                        op_errno = ENOTSUP;
+                        goto err;
+                }
+
+                memcpy (value, tmp->data, ((tmp->len < 4095) ? tmp->len : 4095));
+                local->key = gf_strdup (value);
+                local->call_cnt = conf->subvolume_cnt;
+
+                for (i = 0 ; i < conf->subvolume_cnt; i++) {
+                        /* Get the pathinfo, and then compare */
+                        STACK_WIND (frame, dht_checking_pathinfo_cbk,
+                                    conf->subvolumes[i],
+                                    conf->subvolumes[i]->fops->getxattr,
+                                    loc, GF_XATTR_PATHINFO_KEY, NULL);
+                }
+                return 0;
+        }
+
         tmp = dict_get (xattr, GF_XATTR_FIX_LAYOUT_KEY);
         if (tmp) {
                 gf_log (this->name, GF_LOG_INFO,
                         "fixing the layout of %s", loc->path);
 
-                dht_fix_directory_layout (frame, dht_common_setxattr_cbk,
-                                          layout);
-                return 0;
+                ret = dht_fix_directory_layout (frame, dht_common_setxattr_cbk,
+                                                layout);
+                if (ret) {
+                        op_errno = ENOTCONN;
+                        goto err;
+                }
+                return ret;
         }
 
         tmp = dict_get (xattr, "distribute.directory-spread-count");
@@ -2176,10 +2618,14 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
                              (dir_spread > 0))) {
                         layout->spread_cnt = dir_spread;
 
-                        dht_fix_directory_layout (frame,
-                                                  dht_common_setxattr_cbk,
-                                                  layout);
-                        return 0;
+                        ret = dht_fix_directory_layout (frame,
+                                                        dht_common_setxattr_cbk,
+                                                        layout);
+                        if (ret) {
+                                op_errno = ENOTCONN;
+                                goto err;
+                        }
+                        return ret;
                 }
                 gf_log (this->name, GF_LOG_ERROR,
                         "wrong 'directory-spread-count' value (%s)", value);
@@ -2187,20 +2633,18 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        local->call_cnt = layout->cnt;
-
-        for (i = 0; i < layout->cnt; i++) {
+        for (i = 0; i < call_cnt; i++) {
                 STACK_WIND (frame, dht_err_cbk,
                             layout->list[i].xlator,
                             layout->list[i].xlator->fops->setxattr,
-                            loc, xattr, flags);
+                            loc, xattr, flags, xdata);
         }
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (setxattr, frame, -1, op_errno);
+        DHT_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL);
 
         return 0;
 }
@@ -2208,7 +2652,7 @@ err:
 
 int
 dht_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int op_ret, int op_errno)
+                     int op_ret, int op_errno, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         int           this_call_cnt = 0;
@@ -2234,7 +2678,8 @@ unlock:
 
         this_call_cnt = dht_frame_return (frame);
         if (is_last_call (this_call_cnt)) {
-                DHT_STACK_UNWIND (removexattr, frame, local->op_ret, local->op_errno);
+                DHT_STACK_UNWIND (removexattr, frame, local->op_ret,
+                                  local->op_errno, NULL);
         }
 
         return 0;
@@ -2243,22 +2688,35 @@ unlock:
 
 int
 dht_removexattr (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, const char *key)
+                 loc_t *loc, const char *key, dict_t *xdata)
 {
         xlator_t     *subvol = NULL;
         int           op_errno = -1;
         dht_local_t  *local = NULL;
         dht_layout_t *layout = NULL;
+        int           call_cnt = 0;
+        dht_conf_t   *conf = NULL;
 
         int i;
 
-        VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (this->private, err);
+
+        conf = this->private;
+
+        GF_IF_NATIVE_XATTR_GOTO (conf->wild_xattr_name, key, op_errno, err);
+
+        VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
 
-        subvol = dht_subvol_get_cached (this, loc->inode);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_REMOVEXATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
         if (!subvol) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "no cached subvolume for path=%s", loc->path);
@@ -2266,13 +2724,7 @@ dht_removexattr (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        local->layout = layout = dht_layout_get (this, loc->inode);
+        layout = local->layout;
         if (!local->layout) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "no layout for path=%s", loc->path);
@@ -2280,368 +2732,120 @@ dht_removexattr (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        local->call_cnt = layout->cnt;
+        local->call_cnt = call_cnt = layout->cnt;
+        local->key = gf_strdup (key);
 
-        for (i = 0; i < layout->cnt; i++) {
+        for (i = 0; i < call_cnt; i++) {
                 STACK_WIND (frame, dht_removexattr_cbk,
                             layout->list[i].xlator,
                             layout->list[i].xlator->fops->removexattr,
-                            loc, key);
+                            loc, key, NULL);
         }
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (removexattr, frame, -1, op_errno);
-
-        return 0;
-}
-
-
-int
-dht_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-            int op_ret, int op_errno, fd_t *fd)
-{
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
-
-
-        local = frame->local;
-        prev = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
-
-                local->op_ret = 0;
-        }
-unlock:
-        UNLOCK (&frame->lock);
-
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_UNWIND (open, frame, local->op_ret, local->op_errno,
-                                  local->fd);
+        DHT_STACK_UNWIND (removexattr, frame, -1, op_errno, NULL);
 
         return 0;
 }
 
-
 int
-dht_open (call_frame_t *frame, xlator_t *this,
-          loc_t *loc, int flags, fd_t *fd, int wbflags)
+dht_fremovexattr (call_frame_t *frame, xlator_t *this,
+                  fd_t *fd, const char *key, dict_t *xdata)
 {
         xlator_t     *subvol = NULL;
-        int           ret = -1;
         int           op_errno = -1;
         dht_local_t  *local = NULL;
+        dht_layout_t *layout = NULL;
+        int           call_cnt = 0;
+        dht_conf_t   *conf = 0;
 
+        int i;
 
-        VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
-
-        local->fd = fd_ref (fd);
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
-
-        local->call_cnt = 1;
-
-        STACK_WIND (frame, dht_fd_cbk,
-                    subvol, subvol->fops->open,
-                    loc, flags, fd, wbflags);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (open, frame, -1, op_errno, NULL);
-
-        return 0;
-}
-
-
-int
-dht_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int op_ret, int op_errno,
-               struct iovec *vector, int count, struct iatt *stbuf,
-               struct iobref *iobref)
-{
-        dht_local_t     *local = frame->local;
-
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
-                goto out;
-        }
-
-out:
-        DHT_STACK_UNWIND (readv, frame, op_ret, op_errno, vector, count, stbuf,
-                          iobref);
+        VALIDATE_OR_GOTO (this->private, err);
 
-        return 0;
-}
+        conf = this->private;
 
-
-int
-dht_readv (call_frame_t *frame, xlator_t *this,
-           fd_t *fd, size_t size, off_t off)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
+        GF_IF_NATIVE_XATTR_GOTO (conf->wild_xattr_name, key, op_errno, err);
 
         VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FREMOVEXATTR);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
 
-        STACK_WIND (frame, dht_readv_cbk,
-                    subvol, subvol->fops->readv,
-                    fd, size, off);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL);
-
-        return 0;
-}
-
-
-int
-dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno, struct iatt *prebuf,
-                struct iatt *postbuf)
-{
-        dht_local_t *local = NULL;
-
-        if (op_ret == -1) {
-                goto out;
-        }
-
-        local = frame->local;
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
-                goto out;
-        }
-
-out:
-        DHT_STACK_UNWIND (writev, frame, op_ret, op_errno, prebuf, postbuf);
-
-        return 0;
-}
-
-
-int
-dht_writev (call_frame_t *frame, xlator_t *this,
-            fd_t *fd, struct iovec *vector, int count, off_t off,
-            struct iobref *iobref)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        subvol = dht_subvol_get_cached (this, fd->inode);
+        subvol = local->cached_subvol;
         if (!subvol) {
                 gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
+                        "no cached subvolume for inode=%s",
+                        uuid_utoa (fd->inode->gfid));
                 op_errno = EINVAL;
                 goto err;
         }
 
-        local = dht_local_init (frame);
-        if (!local) {
-
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        STACK_WIND (frame, dht_writev_cbk,
-                    subvol, subvol->fops->writev,
-                    fd, vector, count, off, iobref);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL);
-
-        return 0;
-}
-
-
-int
-dht_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
+        layout = local->layout;
+        if (!local->layout) {
                 gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
+                        "no layout for inode=%s", uuid_utoa (fd->inode->gfid));
                 op_errno = EINVAL;
                 goto err;
         }
 
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
+        local->call_cnt = call_cnt = layout->cnt;
+        local->key = gf_strdup (key);
 
-                goto err;
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (frame, dht_removexattr_cbk,
+                            layout->list[i].xlator,
+                            layout->list[i].xlator->fops->fremovexattr,
+                            fd, key, NULL);
         }
 
-        local->fd = fd_ref (fd);
-        local->call_cnt = 1;
-
-        STACK_WIND (frame, dht_err_cbk,
-                    subvol, subvol->fops->flush, fd);
-
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (flush, frame, -1, op_errno);
+        DHT_STACK_UNWIND (fremovexattr, frame, -1, op_errno, NULL);
 
         return 0;
 }
 
 
 int
-dht_fsync (call_frame_t *frame, xlator_t *this,
-           fd_t *fd, int datasync)
+dht_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+            int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
         dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
 
+        local = frame->local;
+        prev = cookie;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
-        local->call_cnt = 1;
-
-        STACK_WIND (frame, dht_fsync_cbk,
-                    subvol, subvol->fops->fsync,
-                    fd, datasync);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL);
-
-        return 0;
-}
-
-
-int
-dht_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-            int op_ret, int op_errno, struct gf_flock *flock)
-{
-        DHT_STACK_UNWIND (lk, frame, op_ret, op_errno, flock);
-
-        return 0;
-}
-
-
-int
-dht_lk (call_frame_t *frame, xlator_t *this,
-        fd_t *fd, int cmd, struct gf_flock *flock)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "subvolume %s returned -1 (%s)",
+                                prev->this->name, strerror (op_errno));
+                        goto unlock;
+                }
 
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
+                local->op_ret = 0;
         }
+unlock:
+        UNLOCK (&frame->lock);
 
-        STACK_WIND (frame, dht_lk_cbk,
-                    subvol, subvol->fops->lk,
-                    fd, cmd, flock);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lk, frame, -1, op_errno, NULL);
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt))
+                DHT_STACK_UNWIND (open, frame, local->op_ret, local->op_errno,
+                                  local->fd, NULL);
 
         return 0;
 }
@@ -2671,13 +2875,18 @@ dht_normalize_stats (struct statvfs *buf, unsigned long bsize,
 
 int
 dht_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno, struct statvfs *statvfs)
+                int op_ret, int op_errno, struct statvfs *statvfs, dict_t *xdata)
 {
-        dht_local_t *local         = NULL;
-        int          this_call_cnt = 0;
-        int          bsize         = 0;
-        int          frsize        = 0;
+        dht_local_t *local              = NULL;
+        int          this_call_cnt      = 0;
+        int          bsize              = 0;
+        int          frsize             = 0;
+        int8_t       quota_deem_statfs  = 0;
+        GF_UNUSED int     ret           = 0;
+        unsigned long     new_usage     = 0;
+        unsigned long     cur_usage     = 0;
 
+        ret = dict_get_int8 (xdata, "quota-deem-statfs", &quota_deem_statfs);
 
         local = frame->local;
 
@@ -2687,8 +2896,22 @@ dht_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->op_errno = op_errno;
                         goto unlock;
                 }
+                if (!statvfs) {
+                        op_errno = EINVAL;
+                        local->op_ret = -1;
+                        goto unlock;
+                }
                 local->op_ret = 0;
 
+                if (quota_deem_statfs) {
+                        new_usage = statvfs->f_blocks - statvfs->f_bfree;
+                        cur_usage = local->statvfs.f_blocks - local->statvfs.f_bfree;
+                        /* We take the maximux of the usage from the subvols */
+                        if (new_usage >= cur_usage)
+                                local->statvfs = *statvfs;
+                        goto unlock;
+                }
+
                 if (local->statvfs.f_bsize != 0) {
                         bsize = max(local->statvfs.f_bsize, statvfs->f_bsize);
                         frsize = max(local->statvfs.f_frsize, statvfs->f_frsize);
@@ -2709,6 +2932,7 @@ dht_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 local->statvfs.f_flag     = statvfs->f_flag;
                 local->statvfs.f_namemax  = statvfs->f_namemax;
 
+
         }
 unlock:
         UNLOCK (&frame->lock);
@@ -2717,14 +2941,14 @@ unlock:
         this_call_cnt = dht_frame_return (frame);
         if (is_last_call (this_call_cnt))
                 DHT_STACK_UNWIND (statfs, frame, local->op_ret, local->op_errno,
-                                  &local->statvfs);
+                                  &local->statvfs, xdata);
 
         return 0;
 }
 
 
 int
-dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
+dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
         xlator_t     *subvol = NULL;
         dht_local_t  *local  = NULL;
@@ -2736,12 +2960,11 @@ dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
         VALIDATE_OR_GOTO (this->private, err);
 
         conf = this->private;
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, NULL, NULL, GF_FOP_STATFS);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -2753,7 +2976,8 @@ dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
                 for (i = 0; i < conf->subvolume_cnt; i++) {
                         STACK_WIND (frame, dht_statfs_cbk,
                                     conf->subvolumes[i],
-                                    conf->subvolumes[i]->fops->statfs, loc);
+                                    conf->subvolumes[i]->fops->statfs, loc,
+                                    xdata);
                 }
                 return 0;
         }
@@ -2769,28 +2993,27 @@ dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
         local->call_cnt = 1;
 
         STACK_WIND (frame, dht_statfs_cbk,
-                    subvol, subvol->fops->statfs, loc);
+                    subvol, subvol->fops->statfs, loc, xdata);
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (statfs, frame, -1, op_errno, NULL);
+        DHT_STACK_UNWIND (statfs, frame, -1, op_errno, NULL, NULL);
 
         return 0;
 }
 
 
 int
-dht_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
+dht_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+             dict_t *xdata)
 {
         dht_local_t  *local  = NULL;
         dht_conf_t   *conf = NULL;
-        int           ret = -1;
         int           op_errno = -1;
         int           i = -1;
 
-
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
@@ -2798,35 +3021,27 @@ dht_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
 
         conf = this->private;
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, fd, GF_FOP_OPENDIR);
         if (!local) {
                 op_errno = ENOMEM;
 
                 goto err;
         }
 
-        local->fd = fd_ref (fd);
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
-
         local->call_cnt = conf->subvolume_cnt;
 
         for (i = 0; i < conf->subvolume_cnt; i++) {
                 STACK_WIND (frame, dht_fd_cbk,
                             conf->subvolumes[i],
                             conf->subvolumes[i]->fops->opendir,
-                            loc, fd);
+                            loc, fd, xdata);
         }
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (opendir, frame, -1, op_errno, NULL);
+        DHT_STACK_UNWIND (opendir, frame, -1, op_errno, NULL, NULL);
 
         return 0;
 }
@@ -2834,7 +3049,7 @@ err:
 
 int
 dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
-                  int op_errno, gf_dirent_t *orig_entries)
+                  int op_errno, gf_dirent_t *orig_entries, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         gf_dirent_t   entries;
@@ -2847,6 +3062,7 @@ dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
         dht_layout_t *layout = 0;
         dht_conf_t   *conf   = NULL;
         xlator_t     *subvol = 0;
+        int           ret    = 0;
 
         INIT_LIST_HEAD (&entries.list);
         prev = cookie;
@@ -2863,10 +3079,13 @@ dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
 
         list_for_each_entry (orig_entry, (&orig_entries->list), list) {
                 next_offset = orig_entry->d_off;
-
-                if (check_is_linkfile (NULL, (&orig_entry->d_stat), NULL)
-                    || (check_is_dir (NULL, (&orig_entry->d_stat), NULL)
-                        && (prev->this != dht_first_up_subvol (this)))) {
+                if (check_is_dir (NULL, (&orig_entry->d_stat), NULL) &&
+                    (prev->this != local->first_up_subvol)) {
+                        continue;
+                }
+                if (check_is_linkfile (NULL, (&orig_entry->d_stat),
+                                       orig_entry->dict,
+                                       conf->link_xattr_name)) {
                         continue;
                 }
 
@@ -2882,7 +3101,7 @@ dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
                                                     orig_entry->d_name);
                         if (!subvol || (subvol != prev->this)) {
                                 /* TODO: Count the number of entries which need
-                                   linkfile to prove its existance in fs */
+                                   linkfile to prove its existence in fs */
                                 layout->search_unhashed++;
                         }
                 }
@@ -2895,6 +3114,24 @@ dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
                 entry->d_type = orig_entry->d_type;
                 entry->d_len  = orig_entry->d_len;
 
+                if (orig_entry->dict)
+                        entry->dict = dict_ref (orig_entry->dict);
+
+                /* making sure we set the inode ctx right with layout,
+                   currently possible only for non-directories, so for
+                   directories don't set entry inodes */
+                if (!IA_ISDIR(entry->d_stat.ia_type) && orig_entry->inode) {
+                        ret = dht_layout_preset (this, prev->this,
+                                                 orig_entry->inode);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "failed to link the layout in inode");
+                        entry->inode = inode_ref (orig_entry->inode);
+                } else if (orig_entry->inode) {
+                        dht_inode_ctx_time_update (orig_entry->inode, this,
+                                                   &entry->d_stat, 1);
+                }
+
                 list_add_tail (&entry->list, &entries.list);
                 count++;
         }
@@ -2924,9 +3161,23 @@ done:
                         goto unwind;
                 }
 
+		if (conf->readdir_optimize == _gf_true) {
+                        if (next_subvol != local->first_up_subvol) {
+                                ret = dict_set_int32 (local->xattr,
+                                                      GF_READDIR_SKIP_DIRS, 1);
+                                if (ret)
+                                        gf_log (this->name, GF_LOG_ERROR,
+					         "dict set failed");
+		        } else {
+                                 dict_del (local->xattr,
+                                           GF_READDIR_SKIP_DIRS);
+                        }
+                }
+
                 STACK_WIND (frame, dht_readdirp_cbk,
                             next_subvol, next_subvol->fops->readdirp,
-                            local->fd, local->size, next_offset);
+                            local->fd, local->size, next_offset,
+                            local->xattr);
                 return 0;
         }
 
@@ -2934,7 +3185,7 @@ unwind:
         if (op_ret < 0)
                 op_ret = 0;
 
-        DHT_STACK_UNWIND (readdirp, frame, op_ret, op_errno, &entries);
+        DHT_STACK_UNWIND (readdirp, frame, op_ret, op_errno, &entries, NULL);
 
         gf_dirent_free (&entries);
 
@@ -2945,7 +3196,8 @@ unwind:
 
 int
 dht_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, gf_dirent_t *orig_entries)
+                 int op_ret, int op_errno, gf_dirent_t *orig_entries,
+                 dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         gf_dirent_t   entries;
@@ -2956,13 +3208,11 @@ dht_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         off_t         next_offset = 0;
         int           count = 0;
         dht_layout_t *layout = 0;
-        dht_conf_t   *conf   = NULL;
         xlator_t     *subvol = 0;
 
         INIT_LIST_HEAD (&entries.list);
         prev = cookie;
         local = frame->local;
-        conf  = this->private;
 
         if (op_ret < 0)
                 goto done;
@@ -3024,7 +3274,7 @@ done:
 
                 STACK_WIND (frame, dht_readdir_cbk,
                             next_subvol, next_subvol->fops->readdir,
-                            local->fd, local->size, next_offset);
+                            local->fd, local->size, next_offset, NULL);
                 return 0;
         }
 
@@ -3032,7 +3282,7 @@ unwind:
         if (op_ret < 0)
                 op_ret = 0;
 
-        DHT_STACK_UNWIND (readdir, frame, op_ret, op_errno, &entries);
+        DHT_STACK_UNWIND (readdir, frame, op_ret, op_errno, &entries, NULL);
 
         gf_dirent_free (&entries);
 
@@ -3042,46 +3292,76 @@ unwind:
 
 int
 dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                off_t yoff, int whichop)
+                off_t yoff, int whichop, dict_t *dict)
 {
         dht_local_t  *local  = NULL;
-        dht_conf_t   *conf = NULL;
         int           op_errno = -1;
         xlator_t     *xvol = NULL;
         off_t         xoff = 0;
-
+        int           ret = 0;
+        dht_conf_t   *conf = NULL;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
         conf = this->private;
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, NULL, NULL, whichop);
         if (!local) {
-
                 op_errno = ENOMEM;
                 goto err;
         }
 
         local->fd = fd_ref (fd);
         local->size = size;
+        local->xattr_req = (dict)? dict_ref (dict) : NULL;
+        local->first_up_subvol = dht_first_up_subvol (this);
 
         dht_deitransform (this, yoff, &xvol, (uint64_t *)&xoff);
 
         /* TODO: do proper readdir */
-        if (whichop == GF_FOP_READDIR)
-                STACK_WIND (frame, dht_readdir_cbk, xvol, xvol->fops->readdir,
-                            fd, size, xoff);
-        else
+        if (whichop == GF_FOP_READDIRP) {
+                if (dict)
+                        local->xattr = dict_ref (dict);
+                else
+                        local->xattr = dict_new ();
+
+                if (local->xattr) {
+                        ret = dict_set_uint32 (local->xattr,
+                                               conf->link_xattr_name, 256);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "failed to set '%s' key",
+                                        conf->link_xattr_name);
+			if (conf->readdir_optimize == _gf_true) {
+                                if (xvol != local->first_up_subvol) {
+				        ret = dict_set_int32 (local->xattr,
+			                               GF_READDIR_SKIP_DIRS, 1);
+				        if (ret)
+					        gf_log (this->name,
+                                                        GF_LOG_ERROR,
+						        "Dict set failed");
+                                } else {
+                                        dict_del (local->xattr,
+                                                  GF_READDIR_SKIP_DIRS);
+                                }
+			}
+                }
+
                 STACK_WIND (frame, dht_readdirp_cbk, xvol, xvol->fops->readdirp,
-                            fd, size, xoff);
+                            fd, size, xoff, local->xattr);
+        } else {
+                STACK_WIND (frame, dht_readdir_cbk, xvol, xvol->fops->readdir,
+                            fd, size, xoff, local->xattr);
+        }
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (readdir, frame, -1, op_errno, NULL);
+        DHT_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
 
         return 0;
 }
@@ -3089,7 +3369,7 @@ err:
 
 int
 dht_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t yoff)
+             off_t yoff, dict_t *xdata)
 {
         int          op = GF_FOP_READDIR;
         dht_conf_t  *conf = NULL;
@@ -3110,15 +3390,15 @@ dht_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
                 op = GF_FOP_READDIRP;
 
 out:
-        dht_do_readdir (frame, this, fd, size, yoff, op);
+        dht_do_readdir (frame, this, fd, size, yoff, op, 0);
         return 0;
 }
 
 int
 dht_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-              off_t yoff)
+              off_t yoff, dict_t *dict)
 {
-        dht_do_readdir (frame, this, fd, size, yoff, GF_FOP_READDIRP);
+        dht_do_readdir (frame, this, fd, size, yoff, GF_FOP_READDIRP, dict);
         return 0;
 }
 
@@ -3126,7 +3406,7 @@ dht_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
 
 int
 dht_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno)
+                  int op_ret, int op_errno, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         int           this_call_cnt = 0;
@@ -3146,21 +3426,22 @@ dht_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         this_call_cnt = dht_frame_return (frame);
         if (is_last_call (this_call_cnt))
-                DHT_STACK_UNWIND (fsyncdir, frame, local->op_ret, local->op_errno);
+                DHT_STACK_UNWIND (fsyncdir, frame, local->op_ret,
+                                  local->op_errno, xdata);
 
         return 0;
 }
 
 
 int
-dht_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync)
+dht_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
+              int datasync, dict_t *xdata)
 {
         dht_local_t  *local  = NULL;
         dht_conf_t   *conf = NULL;
         int           op_errno = -1;
         int           i = -1;
 
-
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
@@ -3168,10 +3449,9 @@ dht_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync)
 
         conf = this->private;
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, NULL, NULL, GF_FOP_FSYNCDIR);
         if (!local) {
                 op_errno = ENOMEM;
-
                 goto err;
         }
 
@@ -3182,14 +3462,14 @@ dht_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync)
                 STACK_WIND (frame, dht_fsyncdir_cbk,
                             conf->subvolumes[i],
                             conf->subvolumes[i]->fops->fsyncdir,
-                            fd, datasync);
+                            fd, datasync, xdata);
         }
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fsyncdir, frame, -1, op_errno);
+        DHT_STACK_UNWIND (fsyncdir, frame, -1, op_errno, NULL);
 
         return 0;
 }
@@ -3199,9 +3479,9 @@ int
 dht_newfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                  int op_ret, int op_errno,
                  inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
-                 struct iatt *postparent)
+                 struct iatt *postparent, dict_t *xdata)
 {
-        call_frame_t *prev = NULL;
+        xlator_t     *prev = NULL;
         int           ret = -1;
         dht_local_t  *local = NULL;
 
@@ -3219,19 +3499,24 @@ dht_newfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         prev = cookie;
 
         if (local->loc.parent) {
-                WIPE (preparent);
-                WIPE (postparent);
+
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           preparent, 0);
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
         }
 
-        ret = dht_layout_preset (this, prev->this, inode);
+        ret = dht_layout_preset (this, prev, inode);
         if (ret < 0) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "could not set pre-set layout for subvolume %s",
-                        prev->this->name);
+                        prev? prev->name: NULL);
                 op_ret   = -1;
                 op_errno = EINVAL;
                 goto out;
         }
+        if (local->linked == _gf_true)
+                dht_linkfile_attr_heal (frame, this);
 out:
         /*
          * FIXME: ia_size and st_blocks of preparent and postparent do not have
@@ -3240,9 +3525,9 @@ out:
          * corresponding values from each of the subvolume.
          * See dht_iatt_merge for reference.
          */
-
-        DHT_STACK_UNWIND (mknod, frame, op_ret, op_errno, inode, stbuf, preparent,
-                          postparent);
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (mknod, frame, op_ret, op_errno, inode, stbuf,
+                          preparent, postparent, xdata);
         return 0;
 }
 
@@ -3251,7 +3536,8 @@ dht_mknod_linkfile_create_cbk (call_frame_t *frame, void *cookie,
                                xlator_t *this,
                                int32_t op_ret, int32_t op_errno,
                                inode_t *inode, struct iatt *stbuf,
-                               struct iatt *preparent, struct iatt *postparent)
+                               struct iatt *preparent, struct iatt *postparent,
+                               dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         xlator_t     *cached_subvol = NULL;
@@ -3260,42 +3546,43 @@ dht_mknod_linkfile_create_cbk (call_frame_t *frame, void *cookie,
                 goto err;
 
         local = frame->local;
+        if (!local || !local->cached_subvol) {
+                op_errno = EINVAL;
+                goto err;
+        }
+
         cached_subvol = local->cached_subvol;
 
-        STACK_WIND (frame, dht_newfile_cbk,
-                    cached_subvol, cached_subvol->fops->mknod,
-                    &local->loc, local->mode, local->rdev,
-                    local->params);
+        STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)cached_subvol,
+                           cached_subvol, cached_subvol->fops->mknod,
+                           &local->loc, local->mode, local->rdev, local->umask,
+                           local->params);
 
         return 0;
 err:
-        DHT_STACK_UNWIND (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+        DHT_STACK_UNWIND (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                          NULL);
         return 0;
 }
 
 int
 dht_mknod (call_frame_t *frame, xlator_t *this,
-           loc_t *loc, mode_t mode, dev_t rdev, dict_t *params)
+           loc_t *loc, mode_t mode, dev_t rdev, mode_t umask, dict_t *params)
 {
         xlator_t    *subvol = NULL;
         int          op_errno = -1;
-        int          ret = -1;
         xlator_t    *avail_subvol = NULL;
-        dht_conf_t  *conf = NULL;
         dht_local_t *local = NULL;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
 
-        conf = this->private;
-
         dht_get_du_info (frame, this, loc);
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_MKNOD);
         if (!local) {
                 op_errno = ENOMEM;
-
                 goto err;
         }
 
@@ -3308,22 +3595,17 @@ dht_mknod (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
-
         if (!dht_is_subvol_filled (this, subvol)) {
                 gf_log (this->name, GF_LOG_TRACE,
                         "creating %s on %s", loc->path, subvol->name);
 
-                STACK_WIND (frame, dht_newfile_cbk,
-                            subvol, subvol->fops->mknod,
-                            loc, mode, rdev, params);
+                STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol,
+                                   subvol, subvol->fops->mknod, loc, mode,
+                                   rdev, umask, params);
         } else {
-                avail_subvol = dht_free_disk_available_subvol (this, subvol);
+
+                avail_subvol = dht_free_disk_available_subvol (this, subvol,
+                                                               local);
                 if (avail_subvol != subvol) {
                         /* Choose the minimum filled volume, and create the
                            files there */
@@ -3332,17 +3614,18 @@ dht_mknod (call_frame_t *frame, xlator_t *this,
                         local->cached_subvol = avail_subvol;
                         local->mode = mode;
                         local->rdev = rdev;
-
+                        local->umask = umask;
                         dht_linkfile_create (frame,
                                              dht_mknod_linkfile_create_cbk,
-                                             avail_subvol, subvol, loc);
+                                             this, avail_subvol, subvol, loc);
                 } else {
                         gf_log (this->name, GF_LOG_TRACE,
                                 "creating %s on %s", loc->path, subvol->name);
 
-                        STACK_WIND (frame, dht_newfile_cbk,
-                                    subvol, subvol->fops->mknod,
-                                    loc, mode, rdev, params);
+                        STACK_WIND_COOKIE (frame, dht_newfile_cbk,
+                                           (void *)subvol, subvol,
+                                           subvol->fops->mknod, loc, mode,
+                                           rdev, umask, params);
                 }
         }
 
@@ -3351,7 +3634,7 @@ dht_mknod (call_frame_t *frame, xlator_t *this,
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
         DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL);
+                          NULL, NULL, NULL, NULL, NULL);
 
         return 0;
 }
@@ -3359,22 +3642,19 @@ err:
 
 int
 dht_symlink (call_frame_t *frame, xlator_t *this,
-             const char *linkname, loc_t *loc, dict_t *params)
+             const char *linkname, loc_t *loc, mode_t umask, dict_t *params)
 {
         xlator_t    *subvol = NULL;
         int          op_errno = -1;
         dht_local_t *local = NULL;
-        int          ret = -1;
-
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_SYMLINK);
         if (!local) {
                 op_errno = ENOMEM;
-
                 goto err;
         }
 
@@ -3387,41 +3667,33 @@ dht_symlink (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        ret = loc_copy (&local->loc, loc);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_TRACE, "Failed to copy loc");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
         gf_log (this->name, GF_LOG_TRACE,
                 "creating %s on %s", loc->path, subvol->name);
 
-        STACK_WIND (frame, dht_newfile_cbk,
-                    subvol, subvol->fops->symlink,
-                    linkname, loc, params);
+        STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol, subvol,
+                           subvol->fops->symlink, linkname, loc, umask,
+                           params);
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
         DHT_STACK_UNWIND (link, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL);
+                          NULL, NULL, NULL, NULL, NULL);
 
         return 0;
 }
 
 
 int
-dht_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
+dht_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+            dict_t *xdata)
 {
         xlator_t    *cached_subvol = NULL;
         xlator_t    *hashed_subvol = NULL;
-        int          ret = -1;
         int          op_errno = -1;
         dht_local_t *local = NULL;
 
-
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
@@ -3433,53 +3705,49 @@ dht_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
                         local->loc.path, cached_subvol->name, loc->path);
                 STACK_WIND (frame, dht_unlink_cbk,
                             cached_subvol, cached_subvol->fops->unlink,
-                            &local->loc);
+                            &local->loc, xflag, xdata);
                 goto done;
         }
 
-        cached_subvol = dht_subvol_get_cached (this, loc->inode);
-        if (!cached_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
+        local = dht_local_init (frame, loc, NULL, GF_FOP_UNLINK);
+        if (!local) {
+                op_errno = ENOMEM;
+
                 goto err;
         }
 
         hashed_subvol = dht_subvol_get_hashed (this, loc);
+        /* Dont fail unlink if hashed_subvol is NULL which can be the result
+         * of layout anomaly */
         if (!hashed_subvol) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "no subvolume in layout for path=%s",
                         loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-
-                goto err;
         }
 
-        ret = loc_copy (&local->loc, loc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-
+        cached_subvol = local->cached_subvol;
+        if (!cached_subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
                 goto err;
         }
 
-        if (hashed_subvol != cached_subvol) {
+        local->flags = xflag;
+        if (hashed_subvol && hashed_subvol != cached_subvol) {
                 STACK_WIND (frame, dht_unlink_linkfile_cbk,
-                            hashed_subvol, hashed_subvol->fops->unlink, loc);
+                            hashed_subvol, hashed_subvol->fops->unlink, loc,
+                            xflag, xdata);
         } else {
                 STACK_WIND (frame, dht_unlink_cbk,
-                            cached_subvol, cached_subvol->fops->unlink, loc);
+                            cached_subvol, cached_subvol->fops->unlink, loc,
+                            xflag, xdata);
         }
 done:
         return 0;
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL);
+        DHT_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL);
 
         return 0;
 }
@@ -3489,13 +3757,14 @@ int
 dht_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
               int op_ret, int op_errno,
               inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
-              struct iatt *postparent)
+              struct iatt *postparent, dict_t *xdata)
 {
         call_frame_t *prev = NULL;
         dht_layout_t *layout = NULL;
         dht_local_t  *local = NULL;
 
         prev = cookie;
+
         local = frame->local;
 
         if (op_ret == -1)
@@ -3511,12 +3780,20 @@ dht_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 goto out;
         }
 
-        WIPE (preparent);
-        WIPE (postparent);
-
+        if (local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           preparent, 0);
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
+        }
+        if (local->linked == _gf_true) {
+                local->stbuf = *stbuf;
+                dht_linkfile_attr_heal (frame, this);
+        }
 out:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
         DHT_STACK_UNWIND (link, frame, op_ret, op_errno, inode, stbuf, preparent,
-                          postparent);
+                          postparent, NULL);
 
         return 0;
 }
@@ -3526,27 +3803,27 @@ int
 dht_link_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                        int op_ret, int op_errno,
                        inode_t *inode, struct iatt *stbuf,
-                       struct iatt *preparent, struct iatt *postparent)
+                       struct iatt *preparent, struct iatt *postparent,
+                       dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         xlator_t     *srcvol = NULL;
 
-
         if (op_ret == -1)
                 goto err;
 
         local = frame->local;
         srcvol = local->linkfile.srcvol;
 
-        STACK_WIND (frame, dht_link_cbk,
-                    srcvol, srcvol->fops->link,
-                    &local->loc, &local->loc2);
+        STACK_WIND (frame, dht_link_cbk, srcvol, srcvol->fops->link,
+                    &local->loc, &local->loc2, xdata);
 
         return 0;
 
 err:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
         DHT_STACK_UNWIND (link, frame, op_ret, op_errno, inode, stbuf, preparent,
-                          postparent);
+                          postparent, NULL);
 
         return 0;
 }
@@ -3554,7 +3831,7 @@ err:
 
 int
 dht_link (call_frame_t *frame, xlator_t *this,
-          loc_t *oldloc, loc_t *newloc)
+          loc_t *oldloc, loc_t *newloc, dict_t *xdata)
 {
         xlator_t    *cached_subvol = NULL;
         xlator_t    *hashed_subvol = NULL;
@@ -3562,13 +3839,19 @@ dht_link (call_frame_t *frame, xlator_t *this,
         int          ret = -1;
         dht_local_t *local = NULL;
 
-
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (oldloc, err);
         VALIDATE_OR_GOTO (newloc, err);
 
-        cached_subvol = dht_subvol_get_cached (this, oldloc->inode);
+        local = dht_local_init (frame, oldloc, NULL, GF_FOP_LINK);
+        if (!local) {
+                op_errno = ENOMEM;
+
+                goto err;
+        }
+
+        cached_subvol = local->cached_subvol;
         if (!cached_subvol) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "no cached subvolume for path=%s", oldloc->path);
@@ -3585,42 +3868,27 @@ dht_link (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
-
-        ret = loc_copy (&local->loc, oldloc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
-
         ret = loc_copy (&local->loc2, newloc);
         if (ret == -1) {
                 op_errno = ENOMEM;
-
                 goto err;
         }
 
         if (hashed_subvol != cached_subvol) {
-                memcpy (local->gfid, oldloc->inode->gfid, 16);
-                dht_linkfile_create (frame, dht_link_linkfile_cbk,
+                uuid_copy (local->gfid, oldloc->inode->gfid);
+                dht_linkfile_create (frame, dht_link_linkfile_cbk, this,
                                      cached_subvol, hashed_subvol, newloc);
         } else {
                 STACK_WIND (frame, dht_link_cbk,
                             cached_subvol, cached_subvol->fops->link,
-                            oldloc, newloc);
+                            oldloc, newloc, xdata);
         }
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (link, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+        DHT_STACK_UNWIND (link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
         return 0;
 }
@@ -3630,7 +3898,7 @@ int
 dht_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 int op_ret, int op_errno,
                 fd_t *fd, inode_t *inode, struct iatt *stbuf,
-                struct iatt *preparent, struct iatt *postparent)
+                struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
         call_frame_t *prev = NULL;
         int           ret = -1;
@@ -3649,8 +3917,11 @@ dht_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         prev = cookie;
 
         if (local->loc.parent) {
-                WIPE (preparent);
-                WIPE (postparent);
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           preparent, 0);
+
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
         }
 
         ret = dht_layout_preset (this, prev->this, inode);
@@ -3662,10 +3933,14 @@ dht_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 op_errno = EINVAL;
                 goto out;
         }
-
+        if (local->linked == _gf_true) {
+                local->stbuf = *stbuf;
+                dht_linkfile_attr_heal (frame, this);
+        }
 out:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
         DHT_STACK_UNWIND (create, frame, op_ret, op_errno, fd, inode, stbuf, preparent,
-                          postparent);
+                          postparent, NULL);
         return 0;
 }
 
@@ -3675,7 +3950,8 @@ dht_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
                                 xlator_t *this,
                                 int32_t op_ret, int32_t op_errno,
                                 inode_t *inode, struct iatt *stbuf,
-                                struct iatt *preparent, struct iatt *postparent)
+                                struct iatt *preparent, struct iatt *postparent,
+                                dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         xlator_t     *cached_subvol = NULL;
@@ -3689,23 +3965,22 @@ dht_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
         STACK_WIND (frame, dht_create_cbk,
                     cached_subvol, cached_subvol->fops->create,
                     &local->loc, local->flags, local->mode,
-                    local->fd, local->params);
+                    local->umask, local->fd, local->params);
 
         return 0;
 err:
-        DHT_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+        DHT_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL, NULL, NULL);
         return 0;
 }
 
 int
 dht_create (call_frame_t *frame, xlator_t *this,
             loc_t *loc, int32_t flags, mode_t mode,
-            fd_t *fd, dict_t *params)
+            mode_t umask, fd_t *fd, dict_t *params)
 {
         int          op_errno = -1;
-        int          ret = -1;
         xlator_t    *subvol = NULL;
-        dht_conf_t  *conf = NULL;
         dht_local_t *local = NULL;
         xlator_t    *avail_subvol = NULL;
 
@@ -3713,13 +3988,10 @@ dht_create (call_frame_t *frame, xlator_t *this,
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
 
-        conf = this->private;
-
         dht_get_du_info (frame, this, loc);
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, fd, GF_FOP_CREATE);
         if (!local) {
-
                 op_errno = ENOMEM;
                 goto err;
         }
@@ -3731,16 +4003,10 @@ dht_create (call_frame_t *frame, xlator_t *this,
                         local->loc.path, subvol->name, loc->path);
                 STACK_WIND (frame, dht_create_cbk,
                             subvol, subvol->fops->create,
-                            &local->loc, flags, mode, fd, params);
+                            &local->loc, flags, mode, umask, fd, params);
                 goto done;
         }
 
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
         subvol = dht_subvol_get_hashed (this, loc);
         if (!subvol) {
                 gf_log (this->name, GF_LOG_DEBUG,
@@ -3755,40 +4021,38 @@ dht_create (call_frame_t *frame, xlator_t *this,
                         "creating %s on %s", loc->path, subvol->name);
                 STACK_WIND (frame, dht_create_cbk,
                             subvol, subvol->fops->create,
-                            loc, flags, mode, fd, params);
+                            loc, flags, mode, umask, fd, params);
                 goto done;
         }
         /* Choose the minimum filled volume, and create the
            files there */
-        /* TODO */
-        avail_subvol = dht_free_disk_available_subvol (this, subvol);
+        avail_subvol = dht_free_disk_available_subvol (this, subvol, local);
         if (avail_subvol != subvol) {
-                local->fd = fd_ref (fd);
                 local->params = dict_ref (params);
                 local->flags = flags;
                 local->mode = mode;
-
+                local->umask = umask;
                 local->cached_subvol = avail_subvol;
                 local->hashed_subvol = subvol;
                 gf_log (this->name, GF_LOG_TRACE,
                         "creating %s on %s (link at %s)", loc->path,
                         avail_subvol->name, subvol->name);
-                dht_linkfile_create (frame,
-                                     dht_create_linkfile_create_cbk,
-                                     avail_subvol, subvol, loc);
+                dht_linkfile_create (frame, dht_create_linkfile_create_cbk,
+                                     this, avail_subvol, subvol, loc);
                 goto done;
         }
         gf_log (this->name, GF_LOG_TRACE,
                 "creating %s on %s", loc->path, subvol->name);
         STACK_WIND (frame, dht_create_cbk,
                     subvol, subvol->fops->create,
-                    loc, flags, mode, fd, params);
+                    loc, flags, mode, umask, fd, params);
 done:
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+        DHT_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL, NULL, NULL);
 
         return 0;
 }
@@ -3797,26 +4061,28 @@ err:
 int
 dht_mkdir_selfheal_cbk (call_frame_t *frame, void *cookie,
                         xlator_t *this,
-                        int32_t op_ret, int32_t op_errno)
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
         dht_local_t   *local = NULL;
         dht_layout_t  *layout = NULL;
 
-
         local = frame->local;
         layout = local->selfheal.layout;
 
         if (op_ret == 0) {
                 dht_layout_set (this, local->inode, layout);
                 if (local->loc.parent) {
-                        WIPE (&local->preparent);
-                        WIPE (&local->postparent);
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->preparent, 0);
+
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
                 }
         }
 
         DHT_STACK_UNWIND (mkdir, frame, op_ret, op_errno,
                           local->inode, &local->stbuf, &local->preparent,
-                          &local->postparent);
+                          &local->postparent, NULL);
 
         return 0;
 }
@@ -3824,17 +4090,15 @@ dht_mkdir_selfheal_cbk (call_frame_t *frame, void *cookie,
 int
 dht_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
-               struct iatt *preparent, struct iatt *postparent)
+               struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         int           this_call_cnt = 0;
         int           ret = -1;
-        int           subvol_filled = 0;
+        gf_boolean_t subvol_filled = _gf_false;
         call_frame_t *prev = NULL;
         dht_layout_t *layout = NULL;
-        dht_conf_t   *conf = NULL;
 
-        conf = this->private;
         local = frame->local;
         prev  = cookie;
         layout = local->layout;
@@ -3847,9 +4111,21 @@ dht_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         ret = dht_layout_merge (this, layout, prev->this,
                                                 -1, ENOSPC, NULL);
                 } else {
+			if (op_ret == -1 && op_errno == EEXIST)
+				/* Very likely just a race between mkdir and
+				   self-heal (from lookup of a concurrent mkdir
+				   attempt).
+				   Ignore error for now. layout setting will
+				   anyways fail if this was a different (old)
+				   pre-existing different directory.
+				*/
+				op_ret = 0;
                         ret = dht_layout_merge (this, layout, prev->this,
                                                 op_ret, op_errno, NULL);
                 }
+                if (ret)
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: failed to merge layouts", local->loc.path);
 
                 if (op_ret == -1) {
                         local->op_errno = op_errno;
@@ -3876,7 +4152,8 @@ int
 dht_mkdir_hashed_cbk (call_frame_t *frame, void *cookie,
                       xlator_t *this, int op_ret, int op_errno,
                       inode_t *inode, struct iatt *stbuf,
-                      struct iatt *preparent, struct iatt *postparent)
+                      struct iatt *preparent, struct iatt *postparent,
+                      dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         int           ret = -1;
@@ -3904,6 +4181,12 @@ dht_mkdir_hashed_cbk (call_frame_t *frame, void *cookie,
                 ret = dht_layout_merge (this, layout, prev->this,
                                         op_ret, op_errno, NULL);
 
+        /* TODO: we may have to return from the function
+           if layout merge fails. For now, lets just log an error */
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to merge layouts", local->loc.path);
+
         if (op_ret == -1) {
                 local->op_errno = op_errno;
                 goto err;
@@ -3916,6 +4199,8 @@ dht_mkdir_hashed_cbk (call_frame_t *frame, void *cookie,
 
         local->call_cnt = conf->subvolume_cnt - 1;
 
+        if (uuid_is_null (local->loc.gfid))
+                uuid_copy (local->loc.gfid, stbuf->ia_gfid);
         if (local->call_cnt == 0) {
                 dht_selfheal_directory (frame, dht_mkdir_selfheal_cbk,
                                         &local->loc, layout);
@@ -3925,24 +4210,24 @@ dht_mkdir_hashed_cbk (call_frame_t *frame, void *cookie,
                         continue;
                 STACK_WIND (frame, dht_mkdir_cbk,
                             conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->mkdir,
-                            &local->loc, local->mode, local->params);
+                            conf->subvolumes[i]->fops->mkdir, &local->loc,
+                            local->mode, local->umask, local->params);
         }
         return 0;
 err:
-        DHT_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+        DHT_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL, NULL);
         return 0;
 }
 
 
-int
+ int
 dht_mkdir (call_frame_t *frame, xlator_t *this,
-           loc_t *loc, mode_t mode, dict_t *params)
+           loc_t *loc, mode_t mode, mode_t umask, dict_t *params)
 {
         dht_local_t  *local  = NULL;
         dht_conf_t   *conf = NULL;
         int           op_errno = -1;
-        int           ret = -1;
         xlator_t     *hashed_subvol = NULL;
 
 
@@ -3957,15 +4242,13 @@ dht_mkdir (call_frame_t *frame, xlator_t *this,
 
         dht_get_du_info (frame, this, loc);
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_MKDIR);
         if (!local) {
-
                 op_errno = ENOMEM;
                 goto err;
         }
 
         hashed_subvol = dht_subvol_get_hashed (this, loc);
-
         if (hashed_subvol == NULL) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "hashed subvol not found for %s",
@@ -3975,21 +4258,13 @@ dht_mkdir (call_frame_t *frame, xlator_t *this,
         }
 
         local->hashed_subvol = hashed_subvol;
-        local->inode = inode_ref (loc->inode);
-        ret = loc_copy (&local->loc, loc);
         local->mode = mode;
-
-        if (ret == -1) {
-
-                op_errno = ENOMEM;
-                goto err;
-        }
-
+        local->umask = umask;
         local->params = dict_ref (params);
+        local->inode  = inode_ref (loc->inode);
 
         local->layout = dht_layout_new (this, conf->subvolume_cnt);
         if (!local->layout) {
-
                 op_errno = ENOMEM;
                 goto err;
         }
@@ -3997,13 +4272,14 @@ dht_mkdir (call_frame_t *frame, xlator_t *this,
         STACK_WIND (frame, dht_mkdir_hashed_cbk,
                     hashed_subvol,
                     hashed_subvol->fops->mkdir,
-                    loc, mode, params);
+                    loc, mode, umask, params);
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+        DHT_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL, NULL);
 
         return 0;
 }
@@ -4011,14 +4287,87 @@ err:
 
 int
 dht_rmdir_selfheal_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int op_ret, int op_errno)
+                        int op_ret, int op_errno, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
 
         local = frame->local;
 
         DHT_STACK_UNWIND (rmdir, frame, local->op_ret, local->op_errno,
-                          &local->preparent, &local->postparent);
+                          &local->preparent, &local->postparent, NULL);
+
+        return 0;
+}
+
+
+int
+dht_rmdir_hashed_subvol_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int op_ret, int op_errno, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
+
+        local = frame->local;
+        prev  = cookie;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        local->op_ret   = -1;
+                        if (op_errno != ENOENT && op_errno != EACCES) {
+                                local->need_selfheal = 1;
+                        }
+
+
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "rmdir on %s for %s failed (%s)",
+                                prev->this->name, local->loc.path,
+                                strerror (op_errno));
+                        goto unlock;
+                }
+
+                dht_iatt_merge (this, &local->preparent, preparent, prev->this);
+                dht_iatt_merge (this, &local->postparent, postparent,
+                                prev->this);
+
+        }
+unlock:
+        UNLOCK (&frame->lock);
+
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+               if (local->need_selfheal) {
+                        local->layout =
+                                dht_layout_get (this, local->loc.inode);
+
+                        /* TODO: neater interface needed below */
+                        local->stbuf.ia_type = local->loc.inode->ia_type;
+
+                        uuid_copy (local->gfid, local->loc.inode->gfid);
+                        dht_selfheal_restore (frame, dht_rmdir_selfheal_cbk,
+                                              &local->loc, local->layout);
+               } else {
+
+                        if (local->loc.parent) {
+                                dht_inode_ctx_time_update (local->loc.parent,
+                                                           this,
+                                                           &local->preparent,
+                                                           0);
+
+                                dht_inode_ctx_time_update (local->loc.parent,
+                                                           this,
+                                                           &local->postparent,
+                                                           1);
+                        }
+
+                        DHT_STACK_UNWIND (rmdir, frame, local->op_ret,
+                                          local->op_errno, &local->preparent,
+                                          &local->postparent, NULL);
+               }
+        }
 
         return 0;
 }
@@ -4027,11 +4376,12 @@ dht_rmdir_selfheal_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 int
 dht_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                int op_ret, int op_errno, struct iatt *preparent,
-               struct iatt *postparent)
+               struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         int           this_call_cnt = 0;
         call_frame_t *prev = NULL;
+        int           done = 0;
 
         local = frame->local;
         prev  = cookie;
@@ -4042,8 +4392,9 @@ dht_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->op_errno = op_errno;
                         local->op_ret   = -1;
 
-                        if (op_errno != ENOENT)
+                        if (op_errno != ENOENT && op_errno != EACCES) {
                                 local->need_selfheal = 1;
+                        }
 
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "rmdir on %s for %s failed (%s)",
@@ -4052,6 +4403,8 @@ dht_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         goto unlock;
                 }
 
+                /* Track if rmdir succeeded on atleast one subvol*/
+                local->fop_succeeded = 1;
                 dht_iatt_merge (this, &local->preparent, preparent, prev->this);
                 dht_iatt_merge (this, &local->postparent, postparent,
                                 prev->this);
@@ -4061,8 +4414,17 @@ unlock:
 
 
         this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                if (local->need_selfheal) {
+
+        /* if local->hashed_subvol, we are yet to wind to hashed_subvol. */
+        if (local->hashed_subvol && (this_call_cnt == 1)) {
+                done = 1;
+        } else if (!local->hashed_subvol && !this_call_cnt) {
+                done = 1;
+        }
+
+
+        if (done) {
+                if (local->need_selfheal && local->fop_succeeded) {
                         local->layout =
                                 dht_layout_get (this, local->loc.inode);
 
@@ -4072,15 +4434,34 @@ unlock:
                         uuid_copy (local->gfid, local->loc.inode->gfid);
                         dht_selfheal_restore (frame, dht_rmdir_selfheal_cbk,
                                               &local->loc, local->layout);
-                } else {
+                } else if (this_call_cnt) {
+                        /* If non-hashed subvol's have responded, proceed */
+
+                        local->need_selfheal = 0;
+                        STACK_WIND (frame, dht_rmdir_hashed_subvol_cbk,
+                                    local->hashed_subvol,
+                                    local->hashed_subvol->fops->rmdir,
+                                    &local->loc, local->flags, NULL);
+                } else if (!this_call_cnt) {
+                        /* All subvol's have responded, proceed */
+
                         if (local->loc.parent) {
-                                WIPE (&local->preparent);
-                                WIPE (&local->postparent);
+
+                                dht_inode_ctx_time_update (local->loc.parent,
+                                                           this,
+                                                           &local->preparent,
+                                                           0);
+
+                                dht_inode_ctx_time_update (local->loc.parent,
+                                                           this,
+                                                           &local->postparent,
+                                                           1);
+
                         }
 
                         DHT_STACK_UNWIND (rmdir, frame, local->op_ret,
                                           local->op_errno, &local->preparent,
-                                          &local->postparent);
+                                          &local->postparent, NULL);
                 }
         }
 
@@ -4094,6 +4475,7 @@ dht_rmdir_do (call_frame_t *frame, xlator_t *this)
         dht_local_t  *local = NULL;
         dht_conf_t   *conf = NULL;
         int           i = 0;
+        xlator_t     *hashed_subvol = NULL;
 
         VALIDATE_OR_GOTO (this->private, err);
 
@@ -4105,18 +4487,41 @@ dht_rmdir_do (call_frame_t *frame, xlator_t *this)
 
         local->call_cnt = conf->subvolume_cnt;
 
+        /* first remove from non-hashed_subvol */
+        hashed_subvol = dht_subvol_get_hashed (this, &local->loc);
+
+        if (!hashed_subvol) {
+                gf_log (this->name, GF_LOG_WARNING, "failed to get hashed "
+                        "subvol for %s",local->loc.path);
+        } else {
+                local->hashed_subvol = hashed_subvol;
+        }
+
+        /* When DHT has only 1 child */
+        if (conf->subvolume_cnt == 1) {
+                STACK_WIND (frame, dht_rmdir_hashed_subvol_cbk,
+                            conf->subvolumes[0],
+                            conf->subvolumes[0]->fops->rmdir,
+                            &local->loc, local->flags, NULL);
+                return 0;
+        }
+
         for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (hashed_subvol &&
+                    (hashed_subvol == conf->subvolumes[i]))
+                        continue;
+
                 STACK_WIND (frame, dht_rmdir_cbk,
                             conf->subvolumes[i],
                             conf->subvolumes[i]->fops->rmdir,
-                            &local->loc, local->flags);
+                            &local->loc, local->flags, NULL);
         }
 
         return 0;
 
 err:
         DHT_STACK_UNWIND (rmdir, frame, local->op_ret, local->op_errno,
-                          &local->preparent, &local->postparent);
+                          &local->preparent, &local->postparent, NULL);
         return 0;
 }
 
@@ -4124,7 +4529,7 @@ err:
 int
 dht_rmdir_linkfile_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                int op_ret, int op_errno, struct iatt *preparent,
-                               struct iatt *postparent)
+                               struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t    *local = NULL;
         call_frame_t   *prev = NULL;
@@ -4172,6 +4577,7 @@ dht_rmdir_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         call_frame_t   *main_frame = NULL;
         dht_local_t    *main_local = NULL;
         int             this_call_cnt = 0;
+        dht_conf_t     *conf = this->private;
 
         local = frame->local;
         prev  = cookie;
@@ -4183,7 +4589,7 @@ dht_rmdir_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         if (op_ret != 0)
                 goto err;
 
-        if (check_is_linkfile (inode, stbuf, xattr) == 0) {
+        if (!check_is_linkfile (inode, stbuf, xattr, conf->link_xattr_name)) {
                 main_local->op_ret  = -1;
                 main_local->op_errno = ENOTEMPTY;
 
@@ -4194,7 +4600,74 @@ dht_rmdir_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         }
 
         STACK_WIND (frame, dht_rmdir_linkfile_unlink_cbk,
-                    src, src->fops->unlink, &local->loc);
+                    src, src->fops->unlink, &local->loc, 0, NULL);
+        return 0;
+err:
+
+        this_call_cnt = dht_frame_return (main_frame);
+        if (is_last_call (this_call_cnt))
+                dht_rmdir_do (main_frame, this);
+
+        DHT_STACK_DESTROY (frame);
+        return 0;
+}
+
+
+int
+dht_rmdir_cached_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                             int op_ret, int op_errno, inode_t *inode,
+                             struct iatt *stbuf, dict_t *xattr,
+                             struct iatt *parent)
+{
+        dht_local_t    *local         = NULL;
+        xlator_t       *src           = NULL;
+        call_frame_t   *main_frame    = NULL;
+        dht_local_t    *main_local    = NULL;
+        int             this_call_cnt = 0;
+        dht_conf_t     *conf          = this->private;
+        dict_t         *xattrs        = NULL;
+        int             ret           = 0;
+
+        local = frame->local;
+        src   = local->hashed_subvol;
+
+        main_frame = local->main_frame;
+        main_local = main_frame->local;
+
+        if (op_ret == 0) {
+                main_local->op_ret  = -1;
+                main_local->op_errno = ENOTEMPTY;
+
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s found on cached subvol %s",
+                        local->loc.path, src->name);
+                goto err;
+        } else if (op_errno != ENOENT) {
+                main_local->op_ret  = -1;
+                main_local->op_errno = op_errno;
+                goto err;
+        }
+
+        xattrs = dict_new ();
+        if (!xattrs) {
+                gf_log (this->name, GF_LOG_ERROR, "dict_new failed");
+                goto err;
+        }
+
+        ret = dict_set_uint32 (xattrs, conf->link_xattr_name, 256);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "failed to set linkto key"
+                        " in dict");
+                if (xattrs)
+                        dict_unref (xattrs);
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_rmdir_lookup_cbk,
+                    src, src->fops->lookup, &local->loc, xattrs);
+        if (xattrs)
+                dict_unref (xattrs);
+
         return 0;
 err:
 
@@ -4211,12 +4684,15 @@ int
 dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
                            gf_dirent_t *entries, xlator_t *src)
 {
-        int                 ret = 0;
-        int                 build_ret = 0;
-        gf_dirent_t        *trav = NULL;
+        int                 ret          = 0;
+        int                 build_ret    = 0;
+        gf_dirent_t        *trav         = NULL;
         call_frame_t       *lookup_frame = NULL;
         dht_local_t        *lookup_local = NULL;
-        dht_local_t        *local = NULL;
+        dht_local_t        *local        = NULL;
+        dict_t             *xattrs       = NULL;
+        dht_conf_t         *conf         = this->private;
+        xlator_t           *subvol       = NULL;
 
         local = frame->local;
 
@@ -4225,7 +4701,8 @@ dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
                         continue;
                 if (strcmp (trav->d_name, "..") == 0)
                         continue;
-                if (check_is_linkfile (NULL, (&trav->d_stat), NULL) == 1) {
+                if (check_is_linkfile (NULL, (&trav->d_stat), trav->dict,
+                                              conf->link_xattr_name)) {
                         ret++;
                         continue;
                 }
@@ -4237,6 +4714,21 @@ dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
                 return 0;
         }
 
+        xattrs = dict_new ();
+        if (!xattrs) {
+                gf_log (this->name, GF_LOG_ERROR, "dict_new failed");
+                return -1;
+        }
+
+        ret = dict_set_uint32 (xattrs, conf->link_xattr_name, 256);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "failed to set linkto key"
+                        " in dict");
+                if (xattrs)
+                        dict_unref (xattrs);
+                return -1;
+        }
+
         list_for_each_entry (trav, &entries->list, list) {
                 if (strcmp (trav->d_name, ".") == 0)
                         continue;
@@ -4253,20 +4745,22 @@ dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
                         goto err;
                 }
 
-                lookup_local = GF_CALLOC (sizeof (*local), 1,
-                                          gf_dht_mt_dht_local_t);
+                lookup_local = mem_get0 (this->local_pool);
                 if (!lookup_local) {
                         goto err;
                 }
 
                 lookup_frame->local = lookup_local;
                 lookup_local->main_frame = frame;
+                lookup_local->hashed_subvol = src;
 
                 build_ret = dht_build_child_loc (this, &lookup_local->loc,
                                                  &local->loc, trav->d_name);
                 if (build_ret != 0)
                         goto err;
 
+                uuid_copy (lookup_local->loc.gfid, trav->d_stat.ia_gfid);
+
                 gf_log (this->name, GF_LOG_TRACE,
                         "looking up %s on %s",
                         lookup_local->loc.path, src->name);
@@ -4277,14 +4771,31 @@ dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
                 }
                 UNLOCK (&frame->lock);
 
-                STACK_WIND (lookup_frame, dht_rmdir_lookup_cbk,
-                            src, src->fops->lookup,
-                            &lookup_local->loc, NULL);
+                subvol = dht_linkfile_subvol (this, NULL, &trav->d_stat,
+                                              trav->dict);
+                if (!subvol) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "linkfile not having link subvolume. path=%s",
+                                lookup_local->loc.path);
+                        STACK_WIND (lookup_frame, dht_rmdir_lookup_cbk,
+                                    src, src->fops->lookup,
+                                    &lookup_local->loc, xattrs);
+                } else {
+                        STACK_WIND (lookup_frame, dht_rmdir_cached_lookup_cbk,
+                                    subvol, subvol->fops->lookup,
+                                    &lookup_local->loc, xattrs);
+                }
                 ret++;
         }
 
+        if (xattrs)
+                dict_unref (xattrs);
+
         return ret;
 err:
+        if (xattrs)
+                dict_unref (xattrs);
+
         DHT_STACK_DESTROY (lookup_frame);
         return 0;
 }
@@ -4292,7 +4803,8 @@ err:
 
 int
 dht_rmdir_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int op_ret, int op_errno, gf_dirent_t *entries)
+                        int op_ret, int op_errno, gf_dirent_t *entries,
+                        dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         int           this_call_cnt = -1;
@@ -4336,35 +4848,65 @@ dht_rmdir_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 int
 dht_rmdir_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int op_ret, int op_errno, fd_t *fd)
+                       int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
+        dht_local_t  *local         = NULL;
         int           this_call_cnt = -1;
-        call_frame_t *prev = NULL;
-
+        call_frame_t *prev          = NULL;
+        dict_t       *dict          = NULL;
+        int           ret           = 0;
+        dht_conf_t   *conf          = this->private;
+        int           i             = 0;
 
         local = frame->local;
         prev  = cookie;
 
+        this_call_cnt = dht_frame_return (frame);
         if (op_ret == -1) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "opendir on %s for %s failed (%s)",
                         prev->this->name, local->loc.path,
                         strerror (op_errno));
+                if (op_errno != ENOENT) {
+                        local->op_ret = -1;
+                        local->op_errno = op_errno;
+                }
+                goto err;
+        }
+
+        if (!is_last_call (this_call_cnt))
+                return 0;
+
+        if (local->op_ret == -1)
+                goto err;
+
+        dict = dict_new ();
+        if (!dict) {
                 local->op_ret = -1;
-                local->op_errno = op_errno;
+                local->op_errno = ENOMEM;
                 goto err;
         }
 
-        STACK_WIND (frame, dht_rmdir_readdirp_cbk,
-                    prev->this, prev->this->fops->readdirp,
-                    local->fd, 4096, 0);
+        ret = dict_set_uint32 (dict, conf->link_xattr_name, 256);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to set '%s' key",
+                        local->loc.path, conf->link_xattr_name);
+
+        local->call_cnt = conf->subvolume_cnt;
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                STACK_WIND (frame, dht_rmdir_readdirp_cbk,
+                            conf->subvolumes[i],
+                            conf->subvolumes[i]->fops->readdirp,
+                            local->fd, 4096, 0, dict);
+        }
+
+        if (dict)
+                dict_unref (dict);
 
         return 0;
 
 err:
-        this_call_cnt = dht_frame_return (frame);
-
         if (is_last_call (this_call_cnt)) {
                 dht_rmdir_do (frame, this);
         }
@@ -4374,14 +4916,13 @@ err:
 
 
 int
-dht_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
+dht_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+           dict_t *xdata)
 {
         dht_local_t  *local  = NULL;
         dht_conf_t   *conf = NULL;
         int           op_errno = -1;
         int           i = -1;
-        int           ret = -1;
-
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -4392,22 +4933,15 @@ dht_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
 
         conf = this->private;
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_RMDIR);
         if (!local) {
-
                 op_errno = ENOMEM;
                 goto err;
         }
 
         local->call_cnt = conf->subvolume_cnt;
         local->op_ret   = 0;
-
-        ret = loc_copy (&local->loc, loc);
-        if (ret == -1) {
-
-                op_errno = ENOMEM;
-                goto err;
-        }
+        local->fop_succeeded = 0;
 
         local->flags = flags;
 
@@ -4422,7 +4956,7 @@ dht_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
                 STACK_WIND (frame, dht_rmdir_opendir_cbk,
                             conf->subvolumes[i],
                             conf->subvolumes[i]->fops->opendir,
-                            loc, local->fd);
+                            loc, local->fd, NULL);
         }
 
         return 0;
@@ -4430,240 +4964,42 @@ dht_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
         DHT_STACK_UNWIND (rmdir, frame, -1, op_errno,
-                          NULL, NULL);
+                          NULL, NULL, NULL);
 
         return 0;
 }
 
-
 int
-dht_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, dict_t *dict)
-{
-        DHT_STACK_UNWIND (xattrop, frame, op_ret, op_errno, dict);
-        return 0;
-}
-
-
-int
-dht_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             gf_xattrop_flags_t flags, dict_t *dict)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        subvol = dht_subvol_get_cached (this, loc->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
-
-        local->inode = inode_ref (loc->inode);
-        local->call_cnt = 1;
-
-        STACK_WIND (frame,
-                    dht_xattrop_cbk,
-                    subvol, subvol->fops->xattrop,
-                    loc, flags, dict);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (xattrop, frame, -1, op_errno, NULL);
-
-        return 0;
-}
-
+dht_entrylk_cbk (call_frame_t *frame, void *cookie,
+                 xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
-int
-dht_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, dict_t *dict)
 {
-        DHT_STACK_UNWIND (fxattrop, frame, op_ret, op_errno, dict);
+        DHT_STACK_UNWIND (entrylk, frame, op_ret, op_errno, xdata);
         return 0;
 }
 
 
 int
-dht_fxattrop (call_frame_t *frame, xlator_t *this,
-              fd_t *fd, gf_xattrop_flags_t flags, dict_t *dict)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        STACK_WIND (frame,
-                    dht_fxattrop_cbk,
-                    subvol, subvol->fops->fxattrop,
-                    fd, flags, dict);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fxattrop, frame, -1, op_errno, NULL);
-
-        return 0;
-}
-
-
-int
-dht_inodelk_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-        DHT_STACK_UNWIND (inodelk, frame, op_ret, op_errno);
-        return 0;
-}
-
-
-int32_t
-dht_inodelk (call_frame_t *frame, xlator_t *this,
-             const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *lock)
+dht_entrylk (call_frame_t *frame, xlator_t *this,
+             const char *volume, loc_t *loc, const char *basename,
+             entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
 {
         xlator_t     *subvol = NULL;
         int           op_errno = -1;
         dht_local_t  *local = NULL;
 
-
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        subvol = dht_subvol_get_cached (this, loc->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_ENTRYLK);
         if (!local) {
                 op_errno = ENOMEM;
-
                 goto err;
         }
 
-        local->inode = inode_ref (loc->inode);
-        local->call_cnt = 1;
-
-        STACK_WIND (frame,
-                    dht_inodelk_cbk,
-                    subvol, subvol->fops->inodelk,
-                    volume, loc, cmd, lock);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (inodelk, frame, -1, op_errno);
-
-        return 0;
-}
-
-
-int
-dht_finodelk_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-        DHT_STACK_UNWIND (finodelk, frame, op_ret, op_errno);
-        return 0;
-}
-
-
-int
-dht_finodelk (call_frame_t *frame, xlator_t *this,
-              const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *lock)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-
-        STACK_WIND (frame,
-                    dht_finodelk_cbk,
-                    subvol, subvol->fops->finodelk,
-                    volume, fd, cmd, lock);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (finodelk, frame, -1, op_errno);
-
-        return 0;
-}
-
-
-int
-dht_entrylk_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-        DHT_STACK_UNWIND (entrylk, frame, op_ret, op_errno);
-        return 0;
-}
-
-
-int
-dht_entrylk (call_frame_t *frame, xlator_t *this,
-             const char *volume, loc_t *loc, const char *basename,
-             entrylk_cmd cmd, entrylk_type type)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        subvol = dht_subvol_get_cached (this, loc->inode);
+        subvol = local->cached_subvol;
         if (!subvol) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "no cached subvolume for path=%s", loc->path);
@@ -4671,25 +5007,17 @@ dht_entrylk (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
-
-        local->inode = inode_ref (loc->inode);
         local->call_cnt = 1;
 
         STACK_WIND (frame, dht_entrylk_cbk,
                     subvol, subvol->fops->entrylk,
-                    volume, loc, basename, cmd, type);
+                    volume, loc, basename, cmd, type, xdata);
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (entrylk, frame, -1, op_errno);
+        DHT_STACK_UNWIND (entrylk, frame, -1, op_errno, NULL);
 
         return 0;
 }
@@ -4697,10 +5025,10 @@ err:
 
 int
 dht_fentrylk_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno)
+                  xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
 {
-        DHT_STACK_UNWIND (fentrylk, frame, op_ret, op_errno);
+        DHT_STACK_UNWIND (fentrylk, frame, op_ret, op_errno, NULL);
         return 0;
 }
 
@@ -4708,7 +5036,7 @@ dht_fentrylk_cbk (call_frame_t *frame, void *cookie,
 int
 dht_fentrylk (call_frame_t *frame, xlator_t *this,
               const char *volume, fd_t *fd, const char *basename,
-              entrylk_cmd cmd, entrylk_type type)
+              entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
 {
         xlator_t     *subvol = NULL;
         int           op_errno = -1;
@@ -4727,168 +5055,13 @@ dht_fentrylk (call_frame_t *frame, xlator_t *this,
 
         STACK_WIND (frame, dht_fentrylk_cbk,
                     subvol, subvol->fops->fentrylk,
-                    volume, fd, basename, cmd, type);
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fentrylk, frame, -1, op_errno);
-
-        return 0;
-}
-
-
-int
-dht_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, struct iatt *statpre,
-                 struct iatt *statpost)
-{
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
-
-
-        local = frame->local;
-        prev = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
-
-                dht_iatt_merge (this, &local->prebuf, statpre, prev->this);
-                dht_iatt_merge (this, &local->stbuf, statpost, prev->this);
-
-                local->op_ret = 0;
-        }
-unlock:
-        UNLOCK (&frame->lock);
-
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_UNWIND (setattr, frame, local->op_ret, local->op_errno,
-                                  &local->prebuf, &local->stbuf);
-
-        return 0;
-}
-
-
-int
-dht_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             struct iatt *stbuf, int32_t valid)
-{
-        dht_layout_t *layout = NULL;
-        dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-        int           i = -1;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "memory allocation failed :(");
-                goto err;
-        }
-
-        local->layout = layout = dht_layout_get (this, loc->inode);
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        if (!layout_is_sane (layout)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "layout is not sane for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local->inode = inode_ref (loc->inode);
-        local->call_cnt = layout->cnt;
-
-        for (i = 0; i < layout->cnt; i++) {
-                STACK_WIND (frame, dht_setattr_cbk,
-                            layout->list[i].xlator,
-                            layout->list[i].xlator->fops->setattr,
-                            loc, stbuf, valid);
-        }
+                    volume, fd, basename, cmd, type, xdata);
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL);
-
-        return 0;
-}
-
-
-int
-dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
-              int32_t valid)
-{
-        dht_layout_t *layout = NULL;
-        dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-        int           i = -1;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame);
-        if (!local) {
-                op_errno = ENOMEM;
-
-                goto err;
-        }
-
-        local->layout = layout = dht_layout_get (this, fd->inode);
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        if (!layout_is_sane (layout)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "layout is not sane for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        local->inode = inode_ref (fd->inode);
-        local->call_cnt = layout->cnt;
-
-        for (i = 0; i < layout->cnt; i++) {
-                STACK_WIND (frame, dht_setattr_cbk,
-                            layout->list[i].xlator,
-                            layout->list[i].xlator->fops->fsetattr,
-                            fd, stbuf, valid);
-        }
-
-        return 0;
-
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL);
+        DHT_STACK_UNWIND (fentrylk, frame, -1, op_errno, NULL);
 
         return 0;
 }
@@ -4897,64 +5070,22 @@ err:
 int
 dht_forget (xlator_t *this, inode_t *inode)
 {
-        uint64_t      tmp_layout = 0;
+        uint64_t        ctx_int = 0;
+        dht_inode_ctx_t *ctx    = NULL;
         dht_layout_t *layout = NULL;
 
-        inode_ctx_del (inode, this, &tmp_layout);
+        inode_ctx_del (inode, this, &ctx_int);
 
-        if (!tmp_layout)
+        if (!ctx_int)
                 return 0;
 
-        layout = (dht_layout_t *)(long)tmp_layout;
-        dht_layout_unref (this, layout);
-
-        return 0;
-}
-
-
-
-int
-dht_init_subvolumes (xlator_t *this, dht_conf_t *conf)
-{
-        xlator_list_t *subvols = NULL;
-        int            cnt = 0;
-
-        if (!conf)
-                return -1;
-
-        for (subvols = this->children; subvols; subvols = subvols->next)
-                cnt++;
+        ctx = (dht_inode_ctx_t *) (long) ctx_int;
 
-        conf->subvolumes = GF_CALLOC (cnt, sizeof (xlator_t *),
-                                      gf_dht_mt_xlator_t);
-        if (!conf->subvolumes) {
-
-                return -1;
-        }
-        conf->subvolume_cnt = cnt;
-
-        cnt = 0;
-        for (subvols = this->children; subvols; subvols = subvols->next)
-                conf->subvolumes[cnt++] = subvols->xlator;
-
-        conf->subvolume_status = GF_CALLOC (cnt, sizeof (char),
-                                            gf_dht_mt_char);
-        if (!conf->subvolume_status) {
-
-                return -1;
-        }
-
-        conf->last_event = GF_CALLOC (cnt, sizeof (int),
-                                      gf_dht_mt_char);
-        if (!conf->last_event) {
+        layout = ctx->layout;
+        ctx->layout = NULL;
+        dht_layout_unref (this, layout);
+        GF_FREE (ctx);
 
-                return -1;
-        }
-        conf->subvol_up_time = GF_CALLOC (cnt, sizeof (time_t),
-                                          gf_dht_mt_subvol_time);
-        if (!conf->subvol_up_time) {
-                return -1;
-        }
         return 0;
 }
 
@@ -4962,16 +5093,22 @@ dht_init_subvolumes (xlator_t *this, dht_conf_t *conf)
 int
 dht_notify (xlator_t *this, int event, void *data, ...)
 {
-        xlator_t   *subvol = NULL;
-        int         cnt    = -1;
-        int         i      = -1;
-        dht_conf_t *conf   = NULL;
-        int         ret    = -1;
-        int         propagate = 0;
+        xlator_t                *subvol = NULL;
+        int                      cnt    = -1;
+        int                      i      = -1;
+        dht_conf_t              *conf   = NULL;
+        int                      ret    = -1;
+        int                      propagate = 0;
+
+        int                      had_heard_from_all = 0;
+        int                      have_heard_from_all = 0;
+        struct timeval           time = {0,};
+        gf_defrag_info_t        *defrag = NULL;
+        dict_t                  *dict   = NULL;
+        gf_defrag_type           cmd    = 0;
+        dict_t                  *output = NULL;
+        va_list                  ap;
 
-        int         had_heard_from_all = 0;
-        int         have_heard_from_all = 0;
-        struct timeval  time = {0,};
 
         conf = this->private;
         if (!conf)
@@ -5033,7 +5170,12 @@ dht_notify (xlator_t *this, int event, void *data, ...)
                 if (conf->assert_no_child_down) {
                         gf_log (this->name, GF_LOG_WARNING,
                                 "Received CHILD_DOWN. Exiting");
-                        exit(0);
+                        if (conf->defrag) {
+                                gf_defrag_stop (conf->defrag,
+                                                GF_DEFRAG_STATUS_FAILED, NULL);
+                        } else {
+                                kill (getpid(), SIGTERM);
+                        }
                 }
 
                 for (i = 0; i < conf->subvolume_cnt; i++) {
@@ -5084,6 +5226,37 @@ dht_notify (xlator_t *this, int event, void *data, ...)
                 UNLOCK (&conf->subvolume_lock);
 
                 break;
+        case GF_EVENT_VOLUME_DEFRAG:
+        {
+                if (!conf->defrag) {
+                        return ret;
+                }
+                defrag = conf->defrag;
+
+                dict = data;
+                va_start (ap, data);
+                output = va_arg (ap, dict_t*);
+
+                ret = dict_get_int32 (dict, "rebalance-command",
+                                      (int32_t*)&cmd);
+                if (ret)
+                        return ret;
+                LOCK (&defrag->lock);
+                {
+                        if (defrag->is_exiting)
+                                goto unlock;
+                        if (cmd == GF_DEFRAG_CMD_STATUS)
+                                gf_defrag_status_get (defrag, output);
+                        else if (cmd == GF_DEFRAG_CMD_STOP)
+                                gf_defrag_stop (defrag,
+                                                GF_DEFRAG_STATUS_STOPPED, output);
+                }
+unlock:
+                UNLOCK (&defrag->lock);
+                return 0;
+                break;
+        }
+
         default:
                 propagate = 1;
                 break;
@@ -5099,9 +5272,12 @@ dht_notify (xlator_t *this, int event, void *data, ...)
 
         /* if all subvols have reported status, no need to hide anything
            or wait for anything else. Just propagate blindly */
-        if (have_heard_from_all)
+        if (have_heard_from_all) {
                 propagate = 1;
 
+        }
+
+
         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
@@ -5120,6 +5296,19 @@ dht_notify (xlator_t *this, int event, void *data, ...)
                                 /* continue to check other events for CHILD_UP */
                         }
                 }
+
+                /* rebalance is started with assert_no_child_down. So we do
+                 * not need to handle CHILD_DOWN event here.
+                 */
+                if (conf->defrag) {
+                        ret = gf_thread_create (&conf->defrag->th, NULL,
+						gf_defrag_start, this);
+                        if (ret) {
+                                conf->defrag = NULL;
+                                GF_FREE (conf->defrag);
+                                kill (getpid(), SIGTERM);
+                        }
+                }
         }
 
         ret = 0;
@@ -5128,3 +5317,24 @@ dht_notify (xlator_t *this, int event, void *data, ...)
 
         return ret;
 }
+
+int
+dht_inode_ctx_layout_get (inode_t *inode, xlator_t *this, dht_layout_t **layout)
+{
+        dht_inode_ctx_t         *ctx            = NULL;
+        int                      ret            = -1;
+
+        ret = dht_inode_ctx_get (inode, this, &ctx);
+
+        if (!ret && ctx) {
+                if (ctx->layout) {
+                        if (layout)
+                                *layout = ctx->layout;
+                        ret = 0;
+                } else {
+                        ret = -1;
+                }
+        }
+
+        return ret;
+}
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index 2fe85d00e..2ece28a61 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef _CONFIG_H
@@ -22,6 +13,9 @@
 #include "config.h"
 #endif
 
+#include <regex.h>
+#include <signal.h>
+
 #include "dht-mem-types.h"
 #include "libxlator.h"
 #include "syncop.h"
@@ -29,45 +23,81 @@
 #ifndef _DHT_H
 #define _DHT_H
 
-#define GF_XATTR_FIX_LAYOUT_KEY   "trusted.distribute.fix.layout"
+#define GF_XATTR_FIX_LAYOUT_KEY     "distribute.fix.layout"
 #define GF_DHT_LOOKUP_UNHASHED_ON   1
 #define GF_DHT_LOOKUP_UNHASHED_AUTO 2
-#define DHT_PATHINFO_HEADER "DISTRIBUTE:"
+#define DHT_PATHINFO_HEADER         "DISTRIBUTE:"
 
 #include <fnmatch.h>
 
 typedef int (*dht_selfheal_dir_cbk_t) (call_frame_t *frame, void *cookie,
-                                       xlator_t *this,
-                                       int32_t op_ret, int32_t op_errno);
+                                       xlator_t     *this,
+                                       int32_t       op_ret, int32_t op_errno,
+                                       dict_t *xdata);
+typedef int (*dht_defrag_cbk_fn_t) (xlator_t        *this, call_frame_t *frame,
+                                    int              ret);
 
 
 struct dht_layout {
-        int               spread_cnt;  /* layout spread count per directory,
-                                          is controlled by 'setxattr()' with
-                                          special key */
-        int               cnt;
-        int               preset;
-        int               gen;
-        int               type;
-        int               ref;   /* use with dht_conf_t->layout_lock */
-        int               search_unhashed;
+        int                spread_cnt;  /* layout spread count per directory,
+                                           is controlled by 'setxattr()' with
+                                           special key */
+        int                cnt;
+        int                preset;
+        int                gen;
+        int                type;
+        int                ref; /* use with dht_conf_t->layout_lock */
+        gf_boolean_t       search_unhashed;
         struct {
-                int       err;   /* 0 = normal
-                                    -1 = dir exists and no xattr
-                                    >0 = dir lookup failed with errno
-                                 */
-                uint32_t  start;
-                uint32_t  stop;
-                xlator_t *xlator;
-        } list[0];
+                int        err;   /* 0 = normal
+                                     -1 = dir exists and no xattr
+                                     >0 = dir lookup failed with errno
+                                  */
+                uint32_t   start;
+                uint32_t   stop;
+                xlator_t  *xlator;
+        } list[];
+};
+typedef struct dht_layout  dht_layout_t;
+
+struct dht_stat_time {
+        uint32_t        atime;
+        uint32_t        atime_nsec;
+        uint32_t        ctime;
+        uint32_t        ctime_nsec;
+        uint32_t        mtime;
+        uint32_t        mtime_nsec;
+};
+
+typedef struct dht_stat_time dht_stat_time_t;
+
+struct dht_inode_ctx {
+        dht_layout_t    *layout;
+        dht_stat_time_t  time;
 };
-typedef struct dht_layout dht_layout_t;
+
+typedef struct dht_inode_ctx dht_inode_ctx_t;
 
 
 typedef enum {
         DHT_HASH_TYPE_DM,
+        DHT_HASH_TYPE_DM_USER,
 } dht_hashfn_type_t;
 
+/* rebalance related */
+struct dht_rebalance_ {
+        xlator_t            *from_subvol;
+        xlator_t            *target_node;
+        off_t                offset;
+        size_t               size;
+        int32_t              flags;
+        int                  count;
+        struct iobref       *iobref;
+        struct iovec        *vector;
+        struct iatt          stbuf;
+        dht_defrag_cbk_fn_t  target_op_fn;
+        dict_t              *xdata;
+};
 
 struct dht_local {
         int                      call_cnt;
@@ -101,6 +131,7 @@ struct dht_local {
         int                      file_count;
         int                      dir_count;
         call_frame_t            *main_frame;
+        int                      fop_succeeded;
         struct {
                 fop_mknod_cbk_t  linkfile_cbk;
                 struct iatt      stbuf;
@@ -112,7 +143,6 @@ struct dht_local {
         struct {
                 uint32_t         hole_cnt;
                 uint32_t         overlaps_cnt;
-                uint32_t         missing;
                 uint32_t         down;
                 uint32_t         misc;
                 dht_selfheal_dir_cbk_t   dir_cbk;
@@ -125,11 +155,16 @@ struct dht_local {
         int32_t flags;
         mode_t  mode;
         dev_t   rdev;
+        mode_t  umask;
 
         /* need for file-info */
-        char   *pathinfo;
+        char   *xattr_val;
         char   *key;
 
+        /* which xattr request? */
+        char xsel[256];
+        int32_t alloc_len;
+
         char   *newpath;
 
         /* gfid related */
@@ -140,22 +175,82 @@ struct dht_local {
 
         /* flag used to make sure we need to return estale in
            {lookup,revalidate}_cbk */
-        char    return_estale;
+        char return_estale;
+        char need_lookup_everywhere;
+
+        glusterfs_fop_t      fop;
+
+        gf_boolean_t     linked;
+        xlator_t        *link_subvol;
+
+        struct dht_rebalance_ rebalance;
+        xlator_t        *first_up_subvol;
 
-        /* rebalance related */
-#define to_subvol    hashed_subvol
-#define from_subvol  cached_subvol
 };
 typedef struct dht_local dht_local_t;
 
 /* du - disk-usage */
 struct dht_du {
         double   avail_percent;
+	double   avail_inodes;
         uint64_t avail_space;
         uint32_t log;
 };
 typedef struct dht_du dht_du_t;
 
+enum gf_defrag_type {
+        GF_DEFRAG_CMD_START = 1,
+        GF_DEFRAG_CMD_STOP = 1 + 1,
+        GF_DEFRAG_CMD_STATUS = 1 + 2,
+        GF_DEFRAG_CMD_START_LAYOUT_FIX = 1 + 3,
+        GF_DEFRAG_CMD_START_FORCE = 1 + 4,
+};
+typedef enum gf_defrag_type gf_defrag_type;
+
+enum gf_defrag_status_t {
+        GF_DEFRAG_STATUS_NOT_STARTED,
+        GF_DEFRAG_STATUS_STARTED,
+        GF_DEFRAG_STATUS_STOPPED,
+        GF_DEFRAG_STATUS_COMPLETE,
+        GF_DEFRAG_STATUS_FAILED,
+        GF_DEFRAG_STATUS_LAYOUT_FIX_STARTED,
+        GF_DEFRAG_STATUS_LAYOUT_FIX_STOPPED,
+        GF_DEFRAG_STATUS_LAYOUT_FIX_COMPLETE,
+        GF_DEFRAG_STATUS_LAYOUT_FIX_FAILED,
+};
+typedef enum gf_defrag_status_t gf_defrag_status_t;
+
+typedef struct gf_defrag_pattern_list gf_defrag_pattern_list_t;
+
+struct gf_defrag_pattern_list {
+        char                       path_pattern[256];
+        uint64_t                   size;
+        gf_defrag_pattern_list_t  *next;
+};
+
+struct gf_defrag_info_ {
+        uint64_t                     total_files;
+        uint64_t                     total_data;
+        uint64_t                     num_files_lookedup;
+        uint64_t                     total_failures;
+        uint64_t                     skipped;
+        gf_lock_t                    lock;
+        int                          cmd;
+        pthread_t                    th;
+        gf_defrag_status_t           defrag_status;
+        struct rpc_clnt             *rpc;
+        uint32_t                     connected;
+        uint32_t                     is_exiting;
+        pid_t                        pid;
+        inode_t                     *root_inode;
+        uuid_t                       node_uuid;
+        struct timeval               start_time;
+        gf_boolean_t                 stats;
+        gf_defrag_pattern_list_t    *defrag_pattern;
+};
+
+typedef struct gf_defrag_info_ gf_defrag_info_t;
+
 struct dht_conf {
         gf_lock_t      subvolume_lock;
         int            subvolume_cnt;
@@ -167,7 +262,8 @@ struct dht_conf {
         gf_boolean_t   search_unhashed;
         int            gen;
         dht_du_t      *du_stats;
-        uint64_t       min_free_disk;
+        double         min_free_disk;
+        double         min_free_inodes;
         char           disk_unit;
         int32_t        refresh_interval;
         gf_boolean_t   unhashed_sticky_bit;
@@ -184,7 +280,28 @@ struct dht_conf {
         /* Will be a global flag to control the layout spread count */
         uint32_t       dir_spread_cnt;
 
-	struct syncenv *env; /* The env pointer to the rebalance synctask */
+        /* to keep track of nodes which are decomissioned */
+        xlator_t     **decommissioned_bricks;
+        int            decommission_in_progress;
+        int            decommission_subvols_cnt;
+
+        /* defrag related */
+        gf_defrag_info_t *defrag;
+
+        /* Request to filter directory entries in readdir request */
+
+        gf_boolean_t    readdir_optimize;
+
+        /* Support regex-based name reinterpretation. */
+        regex_t         rsync_regex;
+        gf_boolean_t    rsync_regex_valid;
+        regex_t         extra_regex;
+        gf_boolean_t    extra_regex_valid;
+
+        /* Support variable xattr names. */
+        char            *xattr_name;
+        char            *link_xattr_name;
+        char            *wild_xattr_name;
 };
 typedef struct dht_conf dht_conf_t;
 
@@ -199,21 +316,42 @@ struct dht_disk_layout {
 };
 typedef struct dht_disk_layout dht_disk_layout_t;
 
-#define WIPE(statp) do { typeof(*statp) z = {0,}; if (statp) *statp = z; } while (0)
+typedef enum {
+        GF_DHT_MIGRATE_DATA,
+        GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS,
+        GF_DHT_MIGRATE_HARDLINK,
+        GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS
+} gf_dht_migrate_data_type_t;
 
 #define ENTRY_MISSING(op_ret, op_errno) (op_ret == -1 && op_errno == ENOENT)
 
-#define is_fs_root(loc) (strcmp (loc->path, "/") == 0)
-
-#define is_revalidate(loc) (inode_ctx_get (loc->inode, this, NULL) == 0)
+#define is_revalidate(loc) (dht_inode_ctx_layout_get (loc->inode, this, NULL) == 0)
 
 #define is_last_call(cnt) (cnt == 0)
 
-#define DHT_LINKFILE_MODE (S_ISVTX)
-#define check_is_linkfile(i,s,x) (                                      \
-                ((st_mode_from_ia (s->ia_prot, s->ia_type) & ~S_IFMT)   \
-                 == DHT_LINKFILE_MODE) &&                               \
-                (s->ia_size == 0))
+#define DHT_MIGRATION_IN_PROGRESS 1
+#define DHT_MIGRATION_COMPLETED   2
+
+#define check_is_linkfile(i,s,x,n) (IS_DHT_LINKFILE_MODE (s) && dict_get (x, n))
+
+#define IS_DHT_MIGRATION_PHASE2(buf)  (                                 \
+                IA_ISREG ((buf)->ia_type) &&                            \
+                ((st_mode_from_ia ((buf)->ia_prot, (buf)->ia_type) &    \
+                  ~S_IFMT) == DHT_LINKFILE_MODE))
+
+#define IS_DHT_MIGRATION_PHASE1(buf)  (                                 \
+                IA_ISREG ((buf)->ia_type) &&                            \
+                ((buf)->ia_prot.sticky == 1) &&                         \
+                ((buf)->ia_prot.sgid == 1))
+
+#define DHT_STRIP_PHASE1_FLAGS(buf)  do {                       \
+                if ((buf) && IS_DHT_MIGRATION_PHASE1(buf)) {    \
+                        (buf)->ia_prot.sticky = 0;              \
+                        (buf)->ia_prot.sgid = 0;                \
+                }                                               \
+        } while (0)
+
+#define dht_inode_missing(op_errno) (op_errno == ENOENT || op_errno == ESTALE)
 
 #define check_is_dir(i,s,x) (IA_ISDIR(s->ia_type))
 
@@ -221,121 +359,429 @@ typedef struct dht_disk_layout dht_disk_layout_t;
 
 #define DHT_STACK_UNWIND(fop, frame, params ...) do {           \
                 dht_local_t *__local = NULL;                    \
-                xlator_t *__xl = NULL;                          \
+                xlator_t    *__xl    = NULL;                    \
                 if (frame) {                                    \
-                        __xl = frame->this;                     \
-                        __local = frame->local;                 \
+                        __xl         = frame->this;             \
+                        __local      = frame->local;            \
                         frame->local = NULL;                    \
                 }                                               \
                 STACK_UNWIND_STRICT (fop, frame, params);       \
                 dht_local_wipe (__xl, __local);                 \
         } while (0)
 
-#define DHT_STACK_DESTROY(frame) do {           \
-                dht_local_t *__local = NULL;    \
-                xlator_t *__xl = NULL;          \
-                __xl = frame->this;             \
-                __local = frame->local;         \
-                frame->local = NULL;            \
-                STACK_DESTROY (frame->root);    \
-                dht_local_wipe (__xl, __local); \
+#define DHT_STACK_DESTROY(frame) do {                   \
+                dht_local_t *__local = NULL;            \
+                xlator_t    *__xl    = NULL;            \
+                __xl                 = frame->this;     \
+                __local              = frame->local;    \
+                frame->local         = NULL;            \
+                STACK_DESTROY (frame->root);            \
+                dht_local_wipe (__xl, __local);         \
+        } while (0)
+
+#define DHT_UPDATE_TIME(ctx_sec, ctx_nsec, new_sec, new_nsec, inode, post) do {\
+                int32_t sec = 0;                                        \
+                sec = new_sec;                                          \
+                LOCK (&inode->lock);                                    \
+                {                                                       \
+                        new_sec = max(new_sec, ctx_sec);                \
+                        if (sec < new_sec)                              \
+                                new_nsec = ctx_nsec;                    \
+                        if (sec == new_sec)                             \
+                                new_nsec = max (new_nsec, ctx_nsec);    \
+                        if (post) {                                     \
+                                ctx_sec = new_sec;                      \
+                                ctx_nsec = new_nsec;                    \
+                        }                                               \
+                }                                                       \
+                UNLOCK (&inode->lock);                                  \
         } while (0)
 
-dht_layout_t *dht_layout_new (xlator_t *this, int cnt);
-dht_layout_t *dht_layout_get (xlator_t *this, inode_t *inode);
-dht_layout_t *dht_layout_for_subvol (xlator_t *this, xlator_t *subvol);
-xlator_t *dht_layout_search (xlator_t *this, dht_layout_t *layout,
+#define is_greater_time(a, an, b, bn) (((a) < (b)) || (((a) == (b)) && ((an) < (bn))))
+dht_layout_t                            *dht_layout_new (xlator_t *this, int cnt);
+dht_layout_t                            *dht_layout_get (xlator_t *this, inode_t *inode);
+dht_layout_t                            *dht_layout_for_subvol (xlator_t *this, xlator_t *subvol);
+xlator_t *dht_layout_search (xlator_t   *this, dht_layout_t *layout,
                              const char *name);
-int dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout);
-int dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
-                          uint32_t *holes_p, uint32_t *overlaps_p,
-                          uint32_t *missing_p, uint32_t *down_p,
-                          uint32_t *misc_p);
-int dht_layout_dir_mismatch (xlator_t *this, dht_layout_t *layout,
-                             xlator_t *subvol, loc_t *loc, dict_t *xattr);
+int                                      dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout);
+int dht_layout_anomalies (xlator_t      *this, loc_t *loc, dht_layout_t *layout,
+                          uint32_t      *holes_p, uint32_t *overlaps_p,
+                          uint32_t      *missing_p, uint32_t *down_p,
+                          uint32_t      *misc_p, uint32_t *no_space_p);
+int dht_layout_dir_mismatch (xlator_t   *this, dht_layout_t *layout,
+                             xlator_t   *subvol, loc_t *loc, dict_t *xattr);
 
 xlator_t *dht_linkfile_subvol (xlator_t *this, inode_t *inode,
                                struct iatt *buf, dict_t *xattr);
-int dht_linkfile_unlink (call_frame_t *frame, xlator_t *this,
-                         xlator_t *subvol, loc_t *loc);
+int dht_linkfile_unlink (call_frame_t      *frame, xlator_t *this,
+                         xlator_t          *subvol, loc_t *loc);
 
 int dht_layouts_init (xlator_t *this, dht_conf_t *conf);
 int dht_layout_merge (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
-                      int op_ret, int op_errno, dict_t *xattr);
+                      int       op_ret, int op_errno, dict_t *xattr);
 
 int dht_disk_layout_extract (xlator_t *this, dht_layout_t *layout,
-                             int pos, int32_t **disk_layout_p);
-int dht_disk_layout_merge (xlator_t *this, dht_layout_t *layout,
-                           int pos, void *disk_layout_raw);
+                             int       pos, int32_t **disk_layout_p);
+int dht_disk_layout_merge (xlator_t   *this, dht_layout_t *layout,
+                           int         pos, void *disk_layout_raw, int disk_layout_len);
 
 
 int dht_frame_return (call_frame_t *frame);
 
-int dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y);
+int                             dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y);
 int dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol,
                       uint64_t *x);
 
 void dht_local_wipe (xlator_t *this, dht_local_t *local);
-dht_local_t *dht_local_init (call_frame_t *frame);
-int dht_iatt_merge (xlator_t *this, struct iatt *to, struct iatt *from,
-                    xlator_t *subvol);
+dht_local_t *dht_local_init (call_frame_t    *frame, loc_t *loc, fd_t *fd,
+                             glusterfs_fop_t  fop);
+int dht_iatt_merge (xlator_t                 *this, struct iatt *to, struct iatt *from,
+                    xlator_t                 *subvol);
 
 xlator_t *dht_subvol_get_hashed (xlator_t *this, loc_t *loc);
 xlator_t *dht_subvol_get_cached (xlator_t *this, inode_t *inode);
 xlator_t *dht_subvol_next (xlator_t *this, xlator_t *prev);
-int dht_subvol_cnt (xlator_t *this, xlator_t *subvol);
+xlator_t *dht_subvol_next_available (xlator_t *this, xlator_t *prev);
+int       dht_subvol_cnt (xlator_t *this, xlator_t *subvol);
 
-int dht_hash_compute (int type, const char *name, uint32_t *hash_p);
+int dht_hash_compute (xlator_t *this, int type, const char *name, uint32_t *hash_p);
 
-int dht_linkfile_create (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
-                         xlator_t *tovol, xlator_t *fromvol, loc_t *loc);
-int dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc);
-int dht_lookup_everywhere (call_frame_t *frame, xlator_t *this, loc_t *loc);
+int dht_linkfile_create (call_frame_t    *frame, fop_mknod_cbk_t linkfile_cbk,
+                         xlator_t        *this, xlator_t *tovol,
+                         xlator_t        *fromvol, loc_t *loc);
+int                                       dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc);
+int                                       dht_lookup_everywhere (call_frame_t *frame, xlator_t *this, loc_t *loc);
 int
-dht_selfheal_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
-                        loc_t *loc, dht_layout_t *layout);
+dht_selfheal_directory (call_frame_t     *frame, dht_selfheal_dir_cbk_t cbk,
+                        loc_t            *loc, dht_layout_t *layout);
 int
 dht_selfheal_new_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
                             dht_layout_t *layout);
 int
-dht_selfheal_restore (call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
-                      loc_t *loc, dht_layout_t *layout);
+dht_selfheal_restore (call_frame_t       *frame, dht_selfheal_dir_cbk_t cbk,
+                      loc_t              *loc, dht_layout_t *layout);
 int
 dht_layout_sort_volname (dht_layout_t *layout);
 
-int dht_rename (call_frame_t *frame, xlator_t *this,
-                loc_t *oldloc, loc_t *newloc);
-
 int dht_get_du_info (call_frame_t *frame, xlator_t *this, loc_t *loc);
 
-int dht_is_subvol_filled (xlator_t *this, xlator_t *subvol);
-xlator_t *dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol);
-int dht_get_du_info_for_subvol (xlator_t *this, int subvol_idx);
+gf_boolean_t dht_is_subvol_filled (xlator_t *this, xlator_t *subvol);
+xlator_t *dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol,
+                                          dht_local_t *layout);
+int       dht_get_du_info_for_subvol (xlator_t *this, int subvol_idx);
 
 int dht_layout_preset (xlator_t *this, xlator_t *subvol, inode_t *inode);
-int dht_layout_set (xlator_t *this, inode_t *inode, dht_layout_t *layout);
-void dht_layout_unref (xlator_t *this, dht_layout_t *layout);
+int           dht_layout_set (xlator_t *this, inode_t *inode, dht_layout_t *layout);;
+void          dht_layout_unref (xlator_t *this, dht_layout_t *layout);
 dht_layout_t *dht_layout_ref (xlator_t *this, dht_layout_t *layout);
-xlator_t *dht_first_up_subvol (xlator_t *this);
-xlator_t *dht_last_up_subvol (xlator_t *this);
+xlator_t     *dht_first_up_subvol (xlator_t *this);
+xlator_t     *dht_last_up_subvol (xlator_t *this);
 
 int dht_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name);
 
-int dht_filter_loc_subvol_key (xlator_t *this, loc_t *loc, loc_t *new_loc,
+int dht_filter_loc_subvol_key (xlator_t  *this, loc_t *loc, loc_t *new_loc,
                                xlator_t **subvol);
 
-int dht_rename_cleanup (call_frame_t *frame);
+int                                     dht_rename_cleanup (call_frame_t *frame);
 int dht_rename_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      inode_t *inode, struct iatt *stbuf,
-                      struct iatt *preparent, struct iatt *postparent);
-
-int dht_linkfile_recreate(call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
-                         xlator_t *tovol, xlator_t *fromvol, loc_t *loc);
+                          int32_t           op_ret, int32_t op_errno,
+                          inode_t          *inode, struct iatt *stbuf,
+                          struct iatt      *preparent, struct iatt *postparent,
+                          dict_t *xdata);
 
 int dht_fix_directory_layout (call_frame_t *frame,
-                              dht_selfheal_dir_cbk_t dir_cbk,
-                              dht_layout_t *layout);
+                              dht_selfheal_dir_cbk_t  dir_cbk,
+                              dht_layout_t           *layout);
 
+int dht_init_subvolumes (xlator_t *this, dht_conf_t *conf);
+
+/* migration/rebalance */
 int dht_start_rebalance_task (xlator_t *this, call_frame_t *frame);
-#endif /* _DHT_H */
+
+int dht_rebalance_in_progress_check (xlator_t *this, call_frame_t *frame);
+int dht_rebalance_complete_check (xlator_t *this, call_frame_t *frame);
+
+
+/* FOPS */
+int32_t dht_lookup (call_frame_t *frame,
+                    xlator_t *this,
+                    loc_t    *loc,
+                    dict_t   *xattr_req);
+
+int32_t dht_stat (call_frame_t *frame,
+                  xlator_t *this,
+                  loc_t    *loc, dict_t *xdata);
+
+int32_t dht_fstat (call_frame_t *frame,
+                   xlator_t *this,
+                   fd_t     *fd, dict_t *xdata);
+
+int32_t dht_truncate (call_frame_t *frame,
+                      xlator_t *this,
+                      loc_t    *loc,
+                      off_t     offset, dict_t *xdata);
+
+int32_t dht_ftruncate (call_frame_t *frame,
+                       xlator_t *this,
+                       fd_t     *fd,
+                       off_t     offset, dict_t *xdata);
+
+int32_t dht_access (call_frame_t *frame,
+                    xlator_t *this,
+                    loc_t    *loc,
+                    int32_t   mask, dict_t *xdata);
+
+int32_t dht_readlink (call_frame_t *frame,
+                      xlator_t *this,
+                      loc_t    *loc,
+                      size_t    size, dict_t *xdata);
+
+int32_t dht_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata);
+
+int32_t dht_mkdir (call_frame_t *frame, xlator_t *this,
+                   loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata);
+
+int32_t dht_unlink (call_frame_t *frame,
+                    xlator_t *this,
+                    loc_t    *loc, int xflag, dict_t *xdata);
+
+int32_t dht_rmdir (call_frame_t *frame, xlator_t *this,
+                   loc_t    *loc, int flags, dict_t *xdata);
+
+int32_t dht_symlink (call_frame_t   *frame, xlator_t *this,
+                     const char *linkpath, loc_t *loc, mode_t umask,
+                     dict_t *xdata);
+
+int32_t dht_rename (call_frame_t *frame,
+                    xlator_t *this,
+                    loc_t    *oldloc,
+                    loc_t    *newloc, dict_t *xdata);
+
+int32_t dht_link (call_frame_t *frame,
+                  xlator_t *this,
+                  loc_t    *oldloc,
+                  loc_t    *newloc, dict_t *xdata);
+
+int32_t dht_create (call_frame_t *frame, xlator_t *this,
+                    loc_t    *loc, int32_t flags, mode_t mode,
+                    mode_t umask, fd_t     *fd, dict_t *params);
+
+int32_t dht_open (call_frame_t *frame,
+                  xlator_t *this,
+                  loc_t    *loc,
+                  int32_t   flags, fd_t *fd, dict_t *xdata);
+
+int32_t dht_readv (call_frame_t *frame,
+                   xlator_t *this,
+                   fd_t     *fd,
+                   size_t    size,
+                   off_t     offset, uint32_t flags, dict_t *xdata);
+
+int32_t dht_writev (call_frame_t      *frame,
+                    xlator_t      *this,
+                    fd_t          *fd,
+                    struct iovec  *vector,
+                    int32_t        count,
+                    off_t          offset,
+                    uint32_t       flags,
+                    struct iobref *iobref, dict_t *xdata);
+
+int32_t dht_flush (call_frame_t *frame,
+                   xlator_t *this,
+                   fd_t     *fd, dict_t *xdata);
+
+int32_t dht_fsync (call_frame_t *frame,
+                   xlator_t *this,
+                   fd_t     *fd,
+                   int32_t   datasync, dict_t *xdata);
+
+int32_t dht_opendir (call_frame_t *frame,
+                     xlator_t *this,
+                     loc_t    *loc, fd_t *fd, dict_t *xdata);
+
+int32_t dht_fsyncdir (call_frame_t *frame,
+                      xlator_t *this,
+                      fd_t     *fd,
+                      int32_t   datasync, dict_t *xdata);
+
+int32_t dht_statfs (call_frame_t *frame,
+                    xlator_t *this,
+                    loc_t    *loc, dict_t *xdata);
+
+int32_t dht_setxattr (call_frame_t *frame,
+                      xlator_t *this,
+                      loc_t    *loc,
+                      dict_t   *dict,
+                      int32_t   flags, dict_t *xdata);
+
+int32_t dht_getxattr (call_frame_t   *frame,
+                      xlator_t   *this,
+                      loc_t      *loc,
+                      const char *name, dict_t *xdata);
+
+int32_t dht_fsetxattr (call_frame_t *frame,
+                       xlator_t *this,
+                       fd_t     *fd,
+                       dict_t   *dict,
+                       int32_t   flags, dict_t *xdata);
+
+int32_t dht_fgetxattr (call_frame_t   *frame,
+                       xlator_t   *this,
+                       fd_t       *fd,
+                       const char *name, dict_t *xdata);
+
+int32_t dht_removexattr (call_frame_t   *frame,
+                         xlator_t   *this,
+                         loc_t      *loc,
+                         const char *name, dict_t *xdata);
+int32_t dht_fremovexattr (call_frame_t   *frame,
+                          xlator_t   *this,
+                          fd_t      *fd,
+                          const char *name, dict_t *xdata);
+
+int32_t dht_lk (call_frame_t        *frame,
+                xlator_t        *this,
+                fd_t            *fd,
+                int32_t          cmd,
+                struct gf_flock *flock, dict_t *xdata);
+
+int32_t dht_inodelk (call_frame_t *frame, xlator_t *this,
+                     const char      *volume, loc_t *loc, int32_t cmd,
+                     struct gf_flock *flock, dict_t *xdata);
+
+int32_t dht_finodelk (call_frame_t        *frame, xlator_t *this,
+                      const char      *volume, fd_t *fd, int32_t cmd,
+                      struct gf_flock *flock, dict_t *xdata);
+
+int32_t dht_entrylk (call_frame_t    *frame, xlator_t *this,
+                     const char  *volume, loc_t *loc, const char *basename,
+                     entrylk_cmd  cmd, entrylk_type type, dict_t *xdata);
+
+int32_t dht_fentrylk (call_frame_t    *frame, xlator_t *this,
+                      const char  *volume, fd_t *fd, const char *basename,
+                      entrylk_cmd  cmd, entrylk_type type, dict_t *xdata);
+
+int32_t dht_readdir (call_frame_t  *frame,
+                     xlator_t *this,
+                     fd_t     *fd,
+                     size_t    size, off_t off, dict_t *xdata);
+
+int32_t dht_readdirp (call_frame_t *frame,
+                      xlator_t *this,
+                      fd_t     *fd,
+                      size_t    size, off_t off, dict_t *dict);
+
+int32_t dht_xattrop (call_frame_t           *frame,
+                     xlator_t           *this,
+                     loc_t              *loc,
+                     gf_xattrop_flags_t  flags,
+                     dict_t             *dict, dict_t *xdata);
+
+int32_t dht_fxattrop (call_frame_t *frame,
+                      xlator_t           *this,
+                      fd_t               *fd,
+                      gf_xattrop_flags_t  flags,
+                      dict_t             *dict, dict_t *xdata);
+
+int32_t dht_forget (xlator_t *this, inode_t *inode);
+int32_t dht_setattr (call_frame_t  *frame, xlator_t *this, loc_t *loc,
+                     struct iatt   *stbuf, int32_t valid, dict_t *xdata);
+int32_t dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      struct iatt  *stbuf, int32_t valid, dict_t *xdata);
+int32_t dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd,
+		      int32_t mode, off_t offset, size_t len, dict_t *xdata);
+int32_t dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd,
+		    off_t offset, size_t len, dict_t *xdata);
+int32_t dht_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                    off_t offset, off_t len, dict_t *xdata);
+
+int32_t dht_init (xlator_t *this);
+void    dht_fini (xlator_t *this);
+int     dht_reconfigure (xlator_t *this, dict_t *options);
+int32_t dht_notify (xlator_t *this, int32_t event, void *data, ...);
+
+/* definitions for nufa/switch */
+int dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent);
+int dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent);
+int dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
+                             xlator_t *this, int op_ret, int op_errno,
+                             inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                             struct iatt *postparent);
+int dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno,
+                    inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                    struct iatt *postparent);
+int dht_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno,
+                    fd_t *fd, inode_t *inode, struct iatt *stbuf,
+                    struct iatt *preparent, struct iatt *postparent,
+                    dict_t *xdata);
+int dht_newfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno,
+                     inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata);
+
+int
+gf_defrag_status_get (gf_defrag_info_t *defrag, dict_t *dict);
+
+int
+gf_defrag_stop (gf_defrag_info_t *defrag, gf_defrag_status_t status,
+                dict_t *output);
+
+void*
+gf_defrag_start (void *this);
+
+int32_t
+gf_defrag_handle_hardlink (xlator_t *this, loc_t *loc, dict_t  *xattrs,
+                           struct iatt *stbuf);
+int
+dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
+                 int flag);
+int
+dht_inode_ctx_layout_get (inode_t *inode, xlator_t *this,
+                          dht_layout_t **layout_int);
+int
+dht_inode_ctx_layout_set (inode_t *inode, xlator_t *this,
+                          dht_layout_t* layout_int);
+int
+dht_inode_ctx_time_update (inode_t *inode, xlator_t *this, struct iatt *stat,
+                           int32_t update_ctx);
+void dht_inode_ctx_time_set (inode_t *inode, xlator_t *this, struct iatt *stat);
+
+int dht_inode_ctx_get (inode_t *inode, xlator_t *this, dht_inode_ctx_t **ctx);
+int dht_inode_ctx_set (inode_t *inode, xlator_t *this, dht_inode_ctx_t *ctx);
+int
+dht_dir_attr_heal (void *data);
+int
+dht_dir_attr_heal_done (int ret, call_frame_t *sync_frame, void *data);
+int
+dht_dir_has_layout (dict_t *xattr, char *name);
+gf_boolean_t
+dht_is_subvol_in_layout (dht_layout_t *layout, xlator_t *xlator);
+xlator_t *
+dht_subvol_with_free_space_inodes (xlator_t *this, xlator_t *subvol,
+                                   dht_layout_t *layout);
+xlator_t *
+dht_subvol_maxspace_nonzeroinode (xlator_t *this, xlator_t *subvol,
+                                  dht_layout_t *layout);
+int
+dht_linkfile_attr_heal (call_frame_t *frame, xlator_t *this);
+
+void
+dht_layout_dump (dht_layout_t  *layout, const char *prefix);
+int32_t
+dht_priv_dump (xlator_t *this);
+int32_t
+dht_inodectx_dump (xlator_t *this, inode_t *inode);
+
+int
+dht_inode_ctx_get1 (xlator_t *this, inode_t *inode, xlator_t **subvol);
+
+#endif/* _DHT_H */
diff --git a/xlators/cluster/dht/src/dht-diskusage.c b/xlators/cluster/dht/src/dht-diskusage.c
index 2ff470ec8..fe3955ecb 100644
--- a/xlators/cluster/dht/src/dht-diskusage.c
+++ b/xlators/cluster/dht/src/dht-diskusage.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -35,224 +26,389 @@
 
 int
 dht_du_info_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, struct statvfs *statvfs)
+		 int op_ret, int op_errno, struct statvfs *statvfs,
+                 dict_t *xdata)
 {
-        dht_conf_t    *conf         = NULL;
-        call_frame_t  *prev          = NULL;
-        int            this_call_cnt = 0;
-        int            i = 0;
-        double         percent = 0;
-        uint64_t       bytes = 0;
-
-        conf = this->private;
-        prev = cookie;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "failed to get disk info from %s", prev->this->name);
-                goto out;
-        }
-
-        if (statvfs && statvfs->f_blocks) {
-                percent = (statvfs->f_bavail * 100) / statvfs->f_blocks;
-                bytes = (statvfs->f_bavail * statvfs->f_frsize);
-        }
-
-        LOCK (&conf->subvolume_lock);
-        {
-                for (i = 0; i < conf->subvolume_cnt; i++)
-                        if (prev->this == conf->subvolumes[i]) {
-                                conf->du_stats[i].avail_percent = percent;
-                                conf->du_stats[i].avail_space   = bytes;
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "on subvolume '%s': avail_percent is: "
-                                        "%.2f and avail_space is: %"PRIu64"",
-                                        prev->this->name,
-                                        conf->du_stats[i].avail_percent,
-                                        conf->du_stats[i].avail_space);
-                        }
-        }
-        UNLOCK (&conf->subvolume_lock);
+	dht_conf_t    *conf         = NULL;
+	call_frame_t  *prev          = NULL;
+	int            this_call_cnt = 0;
+	int            i = 0;
+	double         percent = 0;
+	double         percent_inodes = 0;
+	uint64_t       bytes = 0;
+
+	conf = this->private;
+	prev = cookie;
+
+	if (op_ret == -1) {
+		gf_log (this->name, GF_LOG_WARNING,
+			"failed to get disk info from %s", prev->this->name);
+		goto out;
+	}
+
+	if (statvfs && statvfs->f_blocks) {
+		percent = (statvfs->f_bavail * 100) / statvfs->f_blocks;
+		bytes = (statvfs->f_bavail * statvfs->f_frsize);
+	}
+
+	if (statvfs && statvfs->f_files) {
+		percent_inodes = (statvfs->f_ffree * 100) / statvfs->f_files;
+	} else {
+		/* set percent inodes to 100 for dynamically allocated inode filesystems
+		   this logic holds good so that, distribute has nothing to worry about
+		   total inodes rather let the 'create()' to be scheduled on the hashed
+		   subvol regardless of the total inodes. since we have no awareness on
+		   loosing inodes this logic fits well
+		*/
+		percent_inodes = 100;
+	}
+
+	LOCK (&conf->subvolume_lock);
+	{
+		for (i = 0; i < conf->subvolume_cnt; i++)
+			if (prev->this == conf->subvolumes[i]) {
+				conf->du_stats[i].avail_percent = percent;
+				conf->du_stats[i].avail_space   = bytes;
+				conf->du_stats[i].avail_inodes  = percent_inodes;
+				gf_log (this->name, GF_LOG_DEBUG,
+					"on subvolume '%s': avail_percent is: "
+					"%.2f and avail_space is: %"PRIu64" "
+					"and avail_inodes is: %.2f",
+					prev->this->name,
+					conf->du_stats[i].avail_percent,
+					conf->du_stats[i].avail_space,
+					conf->du_stats[i].avail_inodes);
+			}
+	}
+	UNLOCK (&conf->subvolume_lock);
 
 out:
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_DESTROY (frame);
+	this_call_cnt = dht_frame_return (frame);
+	if (is_last_call (this_call_cnt))
+		DHT_STACK_DESTROY (frame);
 
-        return 0;
+	return 0;
 }
 
 int
 dht_get_du_info_for_subvol (xlator_t *this, int subvol_idx)
 {
-        dht_conf_t    *conf         = NULL;
-        call_frame_t  *statfs_frame = NULL;
-        dht_local_t   *statfs_local = NULL;
-        call_pool_t   *pool         = NULL;
-
-        conf = this->private;
-        pool = this->ctx->pool;
-
-        statfs_frame = create_frame (this, pool);
-        if (!statfs_frame) {
-                goto err;
-        }
-
-        statfs_local = dht_local_init (statfs_frame);
-        if (!statfs_local) {
-                goto err;
-        }
-
-        loc_t tmp_loc = { .inode = NULL,
-                          .path = "/",
-        };
-
-        statfs_local->call_cnt = 1;
-        STACK_WIND (statfs_frame, dht_du_info_cbk,
-                    conf->subvolumes[subvol_idx],
-                    conf->subvolumes[subvol_idx]->fops->statfs,
-                    &tmp_loc);
-
-        return 0;
+	dht_conf_t    *conf         = NULL;
+	call_frame_t  *statfs_frame = NULL;
+	dht_local_t   *statfs_local = NULL;
+	call_pool_t   *pool         = NULL;
+	loc_t          tmp_loc      = {0,};
+
+	conf = this->private;
+	pool = this->ctx->pool;
+
+	statfs_frame = create_frame (this, pool);
+	if (!statfs_frame) {
+		goto err;
+	}
+
+	/* local->fop value is not used in this case */
+	statfs_local = dht_local_init (statfs_frame, NULL, NULL,
+				       GF_FOP_MAXVALUE);
+	if (!statfs_local) {
+		goto err;
+	}
+
+        /* make it root gfid, should be enough to get the proper info back */
+        tmp_loc.gfid[15] = 1;
+
+	statfs_local->call_cnt = 1;
+	STACK_WIND (statfs_frame, dht_du_info_cbk,
+		    conf->subvolumes[subvol_idx],
+		    conf->subvolumes[subvol_idx]->fops->statfs,
+		    &tmp_loc, NULL);
+
+	return 0;
 err:
-        if (statfs_frame)
-                DHT_STACK_DESTROY (statfs_frame);
+	if (statfs_frame)
+		DHT_STACK_DESTROY (statfs_frame);
 
-        return -1;
+	return -1;
 }
 
 int
 dht_get_du_info (call_frame_t *frame, xlator_t *this, loc_t *loc)
 {
-        int            i = 0;
-        dht_conf_t    *conf         = NULL;
-        call_frame_t  *statfs_frame = NULL;
-        dht_local_t   *statfs_local = NULL;
-        struct timeval tv = {0,};
+	int            i            = 0;
+	dht_conf_t    *conf         = NULL;
+	call_frame_t  *statfs_frame = NULL;
+	dht_local_t   *statfs_local = NULL;
+	struct timeval tv           = {0,};
+        loc_t          tmp_loc      = {0,};
+
+	conf  = this->private;
+
+	gettimeofday (&tv, NULL);
+
+        /* make it root gfid, should be enough to get the proper
+           info back */
+        tmp_loc.gfid[15] = 1;
+
+	if (tv.tv_sec > (conf->refresh_interval
+			 + conf->last_stat_fetch.tv_sec)) {
+
+		statfs_frame = copy_frame (frame);
+		if (!statfs_frame) {
+			goto err;
+		}
+
+		/* In this case, 'local->fop' is not used */
+		statfs_local = dht_local_init (statfs_frame, loc, NULL,
+					       GF_FOP_MAXVALUE);
+		if (!statfs_local) {
+			goto err;
+		}
+
+		statfs_local->call_cnt = conf->subvolume_cnt;
+		for (i = 0; i < conf->subvolume_cnt; i++) {
+			STACK_WIND (statfs_frame, dht_du_info_cbk,
+				    conf->subvolumes[i],
+				    conf->subvolumes[i]->fops->statfs,
+				    &tmp_loc, NULL);
+		}
+
+		conf->last_stat_fetch.tv_sec = tv.tv_sec;
+	}
+	return 0;
+err:
+	if (statfs_frame)
+		DHT_STACK_DESTROY (statfs_frame);
 
-        conf  = this->private;
+	return -1;
+}
 
-        gettimeofday (&tv, NULL);
-        if (tv.tv_sec > (conf->refresh_interval
-                         + conf->last_stat_fetch.tv_sec)) {
 
-                statfs_frame = copy_frame (frame);
-                if (!statfs_frame) {
-                        goto err;
-                }
+gf_boolean_t
+dht_is_subvol_filled (xlator_t *this, xlator_t *subvol)
+{
+	int         i = 0;
+	dht_conf_t *conf = NULL;
+	gf_boolean_t subvol_filled_inodes = _gf_false;
+	gf_boolean_t subvol_filled_space = _gf_false;
+	gf_boolean_t is_subvol_filled = _gf_false;
+
+	conf = this->private;
+
+	/* Check for values above specified percent or free disk */
+	LOCK (&conf->subvolume_lock);
+	{
+		for (i = 0; i < conf->subvolume_cnt; i++) {
+			if (subvol == conf->subvolumes[i]) {
+				if (conf->disk_unit == 'p') {
+					if (conf->du_stats[i].avail_percent <
+					    conf->min_free_disk) {
+						subvol_filled_space = _gf_true;
+						break;
+					}
+
+				} else {
+					if (conf->du_stats[i].avail_space <
+					    conf->min_free_disk) {
+						subvol_filled_space = _gf_true;
+						break;
+					}
+				}
+				if (conf->du_stats[i].avail_inodes <
+				    conf->min_free_inodes) {
+					subvol_filled_inodes = _gf_true;
+					break;
+				}
+			}
+		}
+	}
+	UNLOCK (&conf->subvolume_lock);
+
+	if (subvol_filled_space && conf->subvolume_status[i]) {
+		if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
+			gf_log (this->name, GF_LOG_WARNING,
+				"disk space on subvolume '%s' is getting "
+				"full (%.2f %%), consider adding more nodes",
+				subvol->name,
+				(100 - conf->du_stats[i].avail_percent));
+		}
+	}
+
+	if (subvol_filled_inodes && conf->subvolume_status[i]) {
+		if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
+			gf_log (this->name, GF_LOG_CRITICAL,
+				"inodes on subvolume '%s' are at "
+				"(%.2f %%), consider adding more nodes",
+				subvol->name,
+				(100 - conf->du_stats[i].avail_inodes));
+		}
+	}
+
+	is_subvol_filled = (subvol_filled_space || subvol_filled_inodes);
+
+	return is_subvol_filled;
+}
+
+
+/*Get the best subvolume to create the file in*/
+xlator_t *
+dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol,
+                                dht_local_t *local)
+{
+	xlator_t   *avail_subvol = NULL;
+	dht_conf_t *conf = NULL;
+        dht_layout_t *layout = NULL;
+        loc_t      *loc = NULL;
 
-                statfs_local = dht_local_init (statfs_frame);
-                if (!statfs_local) {
-                        goto err;
+	conf = this->private;
+        if (!local)
+                goto out;
+        loc = &local->loc;
+        if (!local->layout) {
+                layout = dht_layout_get (this, loc->parent);
+
+                if (!layout) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "layout missing path=%s parent=%s",
+                                loc->path, uuid_utoa (loc->parent->gfid));
+                        goto out;
                 }
+        } else {
+                layout = dht_layout_ref (this, local->layout);
+        }
 
-                loc_copy (&statfs_local->loc, loc);
-                loc_t tmp_loc = { .inode = NULL,
-                                  .path = "/",
-                };
-
-                statfs_local->call_cnt = conf->subvolume_cnt;
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        STACK_WIND (statfs_frame, dht_du_info_cbk,
-                                    conf->subvolumes[i],
-                                    conf->subvolumes[i]->fops->statfs,
-                                    &tmp_loc);
+        LOCK (&conf->subvolume_lock);
+	{
+                avail_subvol = dht_subvol_with_free_space_inodes(this, subvol,
+                                                                 layout);
+                if(!avail_subvol)
+                {
+                        avail_subvol = dht_subvol_maxspace_nonzeroinode(this,
+                                                                        subvol,
+                                                                        layout);
                 }
 
-                conf->last_stat_fetch.tv_sec = tv.tv_sec;
-        }
-        return 0;
-err:
-        if (statfs_frame)
-                DHT_STACK_DESTROY (statfs_frame);
+	}
+	UNLOCK (&conf->subvolume_lock);
+out:
+	if (!avail_subvol) {
+		gf_log (this->name,
+                        GF_LOG_DEBUG,
+			"no subvolume has enough free space and/or inodes\
+                         to create");
+                avail_subvol = subvol;
+	}
 
-        return -1;
+        if (layout)
+                dht_layout_unref (this, layout);
+	return avail_subvol;
 }
 
+static inline
+int32_t dht_subvol_has_err (xlator_t *this, dht_layout_t *layout)
+{
+        int ret = -1;
+        int i   = 0;
 
-int
-dht_is_subvol_filled (xlator_t *this, xlator_t *subvol)
+        if (!this || !layout)
+                goto out;
+
+        /* check if subvol has layout errors, before selecting it */
+        for (i = 0; i < layout->cnt; i++) {
+                if (!strcmp (layout->list[i].xlator->name, this->name) &&
+                     (layout->list[i].err != 0)) {
+                        ret = -1;
+                        goto out;
+                }
+        }
+        ret = 0;
+out:
+        return ret;
+}
+
+/*Get subvolume which has both space and inodes more than the min criteria*/
+xlator_t *
+dht_subvol_with_free_space_inodes(xlator_t *this, xlator_t *subvol,
+                                  dht_layout_t *layout)
 {
-        int         i = 0;
-        int         subvol_filled = 0;
+        int i = 0;
+        double max = 0;
+        double max_inodes = 0;
+        int    ignore_subvol = 0;
+
+        xlator_t *avail_subvol = NULL;
         dht_conf_t *conf = NULL;
 
         conf = this->private;
 
-        /* Check for values above specified percent or free disk */
-        LOCK (&conf->subvolume_lock);
-        {
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (subvol == conf->subvolumes[i]) {
-                                if (conf->disk_unit == 'p') {
-                                        if (conf->du_stats[i].avail_percent <
-                                            conf->min_free_disk) {
-                                                subvol_filled = 1;
-                                                break;
-                                        }
-                                } else {
-                                        if (conf->du_stats[i].avail_space <
-                                            conf->min_free_disk) {
-                                                subvol_filled = 1;
-                                                break;
-                                        }
-                                }
+        for(i=0; i < conf->subvolume_cnt; i++) {
+                /* check if subvol has layout errors, before selecting it */
+                ignore_subvol = dht_subvol_has_err (conf->subvolumes[i],
+                                                    layout);
+                if (ignore_subvol)
+                        continue;
+
+                if ((conf->disk_unit == 'p') &&
+                    (conf->du_stats[i].avail_percent > conf->min_free_disk) &&
+                    (conf->du_stats[i].avail_inodes  > conf->min_free_inodes)) {
+                        if ((conf->du_stats[i].avail_inodes > max_inodes) ||
+                            (conf->du_stats[i].avail_percent > max)) {
+                                max = conf->du_stats[i].avail_percent;
+                                max_inodes = conf->du_stats[i].avail_inodes;
+                                avail_subvol = conf->subvolumes[i];
                         }
                 }
-        }
-        UNLOCK (&conf->subvolume_lock);
-
-        if (subvol_filled && conf->subvolume_status[i]) {
-                if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "disk space on subvolume '%s' is getting "
-                                "full (%.2f %%), consider adding more nodes",
-                                subvol->name,
-                                (100 - conf->du_stats[i].avail_percent));
+
+                if ((conf->disk_unit != 'p') &&
+                    (conf->du_stats[i].avail_space > conf->min_free_disk) &&
+                    (conf->du_stats[i].avail_inodes  > conf->min_free_inodes)) {
+                        if ((conf->du_stats[i].avail_inodes > max_inodes) ||
+                            (conf->du_stats[i].avail_space > max)) {
+                                max = conf->du_stats[i].avail_space;
+                                max_inodes = conf->du_stats[i].avail_inodes;
+                                avail_subvol = conf->subvolumes[i];
+                        }
                 }
         }
 
-        return subvol_filled;
+        return avail_subvol;
 }
 
+
+/* Get subvol which has atleast one inode and maximum space */
 xlator_t *
-dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol)
+dht_subvol_maxspace_nonzeroinode (xlator_t *this, xlator_t *subvol,
+                                  dht_layout_t *layout)
 {
         int         i = 0;
-        double      max= 0;
+        double      max = 0;
+        int         ignore_subvol = 0;
+
         xlator_t   *avail_subvol = NULL;
         dht_conf_t *conf = NULL;
 
         conf = this->private;
 
-        LOCK (&conf->subvolume_lock);
-        {
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (conf->disk_unit == 'p') {
-                                if (conf->du_stats[i].avail_percent > max) {
-                                        max = conf->du_stats[i].avail_percent;
-                                        avail_subvol = conf->subvolumes[i];
-                                }
-                        } else {
-                                if (conf->du_stats[i].avail_space > max) {
-                                        max = conf->du_stats[i].avail_space;
-                                        avail_subvol = conf->subvolumes[i];
-                                }
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                /* check if subvol has layout errors, before selecting it */
+                ignore_subvol = dht_subvol_has_err (conf->subvolumes[i],
+                                                    layout);
+                if (ignore_subvol)
+                        continue;
+
+                if (conf->disk_unit == 'p') {
+                        if ((conf->du_stats[i].avail_percent > max)
+                            && (conf->du_stats[i].avail_inodes > 0 )) {
+                                max = conf->du_stats[i].avail_percent;
+                                avail_subvol = conf->subvolumes[i];
                         }
-                }
-        }
-        UNLOCK (&conf->subvolume_lock);
-
-        if (!avail_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume has enough free space to create");
+               } else {
+                         if ((conf->du_stats[i].avail_space > max)
+                            && (conf->du_stats[i].avail_inodes > 0)) {
+                                 max = conf->du_stats[i].avail_space;
+                                 avail_subvol = conf->subvolumes[i];
+                         }
+               }
         }
 
-        if (max < conf->min_free_disk)
-                avail_subvol = subvol;
-
-        if (!avail_subvol)
-                avail_subvol = subvol;
-
         return avail_subvol;
 }
diff --git a/xlators/cluster/dht/src/dht-hashfn.c b/xlators/cluster/dht/src/dht-hashfn.c
index 99bb13265..656cf23a0 100644
--- a/xlators/cluster/dht/src/dht-hashfn.c
+++ b/xlators/cluster/dht/src/dht-hashfn.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef _CONFIG_H
@@ -37,6 +28,7 @@ dht_hash_compute_internal (int type, const char *name, uint32_t *hash_p)
 
         switch (type) {
         case DHT_HASH_TYPE_DM:
+        case DHT_HASH_TYPE_DM_USER:
                 hash = gf_dm_hashfn (name, strlen (name));
                 break;
         default:
@@ -52,30 +44,68 @@ dht_hash_compute_internal (int type, const char *name, uint32_t *hash_p)
 }
 
 
-#define MAKE_RSYNC_FRIENDLY_NAME(rsync_frndly_name, name) do {          \
-                rsync_frndly_name = (char *) name;                      \
-                if (name[0] == '.') {                                   \
-                        char *dot   = 0;                                \
-                        int namelen = 0;                                \
-                                                                        \
-                        dot = strrchr (name, '.');                      \
-                        if (dot && dot > (name + 1) && *(dot + 1)) {    \
-                                namelen = (dot - name);                 \
-                                rsync_frndly_name = alloca (namelen);   \
-                                strncpy (rsync_frndly_name, name + 1,   \
-                                         namelen);                      \
-                                rsync_frndly_name[namelen - 1] = 0;     \
-                        }                                               \
-                }                                                       \
-        } while (0);
+static inline
+gf_boolean_t
+dht_munge_name (const char *original, char *modified, size_t len, regex_t *re)
+{
+        regmatch_t      matches[2];
+        size_t          new_len;
+
+        if (regexec(re,original,2,matches,0) != REG_NOMATCH) {
+                if (matches[1].rm_so != -1) {
+                        new_len = matches[1].rm_eo - matches[1].rm_so;
+                        /* Equal would fail due to the NUL at the end. */
+                        if (new_len < len) {
+                                memcpy (modified,original+matches[1].rm_so,
+                                        new_len);
+                                modified[new_len] = '\0';
+                                return _gf_true;
+                        }
+                }
+        }
 
+        /* This is guaranteed safe because of how the dest was allocated. */
+        strcpy(modified,original);
+        return _gf_false;
+}
 
 int
-dht_hash_compute (int type, const char *name, uint32_t *hash_p)
+dht_hash_compute (xlator_t *this, int type, const char *name, uint32_t *hash_p)
 {
-        char     *rsync_friendly_name = NULL;
+        char            *rsync_friendly_name    = NULL;
+        dht_conf_t      *priv                   = this->private;
+        size_t           len                    = 0;
+        gf_boolean_t     munged                 = _gf_false;
+
+        /*
+         * It wouldn't be safe to use alloca in an inline function that doesn't
+         * actually get inlined, and it wouldn't be efficient to do a real
+         * allocation, so we use alloca here (if needed) and pass that to the
+         * inline.
+         */
+
+        if (priv->extra_regex_valid) {
+                len = strlen(name) + 1;
+                rsync_friendly_name = alloca(len);
+                munged = dht_munge_name (name, rsync_friendly_name, len,
+                                         &priv->extra_regex);
+        }
+
+        if (!munged && priv->rsync_regex_valid) {
+                len = strlen(name) + 1;
+                rsync_friendly_name = alloca(len);
+                gf_log (this->name, GF_LOG_TRACE, "trying regex for %s", name);
+                munged = dht_munge_name (name, rsync_friendly_name, len,
+                                         &priv->rsync_regex);
+                if (munged) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "munged down to %s", rsync_friendly_name);
+                }
+        }
 
-        MAKE_RSYNC_FRIENDLY_NAME (rsync_friendly_name, name);
+        if (!munged) {
+                rsync_friendly_name = (char *)name;
+        }
 
         return dht_hash_compute_internal (type, rsync_friendly_name, hash_p);
 }
diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c
index bb7ae0476..f1dc5072f 100644
--- a/xlators/cluster/dht/src/dht-helper.c
+++ b/xlators/cluster/dht/src/dht-helper.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef _CONFIG_H
@@ -27,6 +18,28 @@
 #include "xlator.h"
 #include "dht-common.h"
 
+static inline int
+dht_inode_ctx_set1 (xlator_t *this, inode_t *inode, xlator_t *subvol)
+{
+        uint64_t tmp_subvol = 0;
+
+        tmp_subvol = (long)subvol;
+        return inode_ctx_set1 (inode, this, &tmp_subvol);
+}
+
+int
+dht_inode_ctx_get1 (xlator_t *this, inode_t *inode, xlator_t **subvol)
+{
+        int ret = -1;
+        uint64_t tmp_subvol = 0;
+
+        ret =  inode_ctx_get1 (inode, this, &tmp_subvol);
+        if (tmp_subvol && subvol)
+                *subvol = (xlator_t *)tmp_subvol;
+
+        return ret;
+}
+
 
 int
 dht_frame_return (call_frame_t *frame)
@@ -49,6 +62,43 @@ dht_frame_return (call_frame_t *frame)
 }
 
 
+static uint64_t
+dht_bits_for (uint64_t num)
+{
+	uint64_t bits = 0, ctrl = 1;
+
+	while (ctrl < num) {
+		ctrl *= 2;
+		bits ++;
+	}
+
+	return bits;
+}
+
+/*
+ * A slightly "updated" version of the algorithm described in the commit log
+ * is used here.
+ *
+ * The only enhancement is that:
+ *
+ * - The number of bits used by the backend filesystem for HUGE d_off which
+ *   is described as 63, and
+ * - The number of bits used by the d_off presented by the transformation
+ *   upwards which is described as 64, are both made "configurable."
+ */
+
+
+#define BACKEND_D_OFF_BITS 63
+#define PRESENT_D_OFF_BITS 63
+
+#define ONE 1ULL
+#define MASK (~0ULL)
+#define PRESENT_MASK (MASK >> (64 - PRESENT_D_OFF_BITS))
+#define BACKEND_MASK (MASK >> (64 - BACKEND_D_OFF_BITS))
+
+#define TOP_BIT (ONE << (PRESENT_D_OFF_BITS - 1))
+#define SHIFT_BITS (max (0, (BACKEND_D_OFF_BITS - PRESENT_D_OFF_BITS + 1)))
+
 int
 dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
 {
@@ -56,6 +106,9 @@ dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
         int         cnt = 0;
         int         max = 0;
         uint64_t    y = 0;
+        uint64_t    hi_mask = 0;
+        uint64_t    off_mask = 0;
+        int         max_bits = 0;
 
         if (x == ((uint64_t) -1)) {
                 y = (uint64_t) -1;
@@ -69,7 +122,23 @@ dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
         max = conf->subvolume_cnt;
         cnt = dht_subvol_cnt (this, subvol);
 
-        y = ((x * max) + cnt);
+	if (max == 1) {
+		y = x;
+		goto out;
+	}
+
+        max_bits = dht_bits_for (max);
+
+        hi_mask = ~(PRESENT_MASK >> (max_bits + 1));
+
+        if (x & hi_mask) {
+                /* HUGE d_off */
+                off_mask = MASK << max_bits;
+                y = TOP_BIT | ((x >> SHIFT_BITS) & off_mask) | cnt;
+        } else {
+                /* small d_off */
+                y = ((x * max) + cnt);
+        }
 
 out:
         if (y_p)
@@ -89,7 +158,7 @@ dht_filter_loc_subvol_key (xlator_t *this, loc_t *loc, loc_t *new_loc,
         int            ret       = 0; /* not found */
 
         /* Why do other tasks if first required 'char' itself is not there */
-        if (loc->name && !strchr (loc->name, '@'))
+        if (!new_loc || !loc || !loc->name || !strchr (loc->name, '@'))
                 goto out;
 
         trav = this->children;
@@ -117,7 +186,6 @@ dht_filter_loc_subvol_key (xlator_t *this, loc_t *loc, loc_t *new_loc,
                                 new_loc->path   = ((new_path) ? new_path:
                                                    gf_strdup (loc->path));
                                 new_loc->name   = new_name;
-                                new_loc->ino    = loc->ino;
                                 new_loc->inode  = inode_ref (loc->inode);
                                 new_loc->parent = inode_ref (loc->parent);
                         }
@@ -130,10 +198,8 @@ dht_filter_loc_subvol_key (xlator_t *this, loc_t *loc, loc_t *new_loc,
 out:
         if (!ret) {
                 /* !success */
-                if (new_path)
-                        GF_FREE (new_path);
-                if (new_name)
-                        GF_FREE (new_name);
+                GF_FREE (new_path);
+                GF_FREE (new_name);
         }
         return ret;
 }
@@ -147,16 +213,38 @@ dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol_p,
         int         max = 0;
         uint64_t    x = 0;
         xlator_t   *subvol = 0;
+        int         max_bits = 0;
+        uint64_t    off_mask = 0;
+        uint64_t    host_mask = 0;
 
         if (!this->private)
-                goto out;
+                return -1;
 
         conf = this->private;
         max = conf->subvolume_cnt;
 
-        cnt = y % max;
-        x   = y / max;
+	if (max == 1) {
+		x = y;
+		cnt = 0;
+		goto out;
+	}
+
+        if (y & TOP_BIT) {
+                /* HUGE d_off */
+                max_bits = dht_bits_for (max);
+                off_mask = (MASK << max_bits);
+                host_mask = ~(off_mask);
+
+                x = ((y & ~TOP_BIT) & off_mask) << SHIFT_BITS;
+
+                cnt = y & host_mask;
+	} else {
+                /* small d_off */
+                cnt = y % max;
+                x = y / max;
+        }
 
+out:
         subvol = conf->subvolumes[cnt];
 
         if (subvol_p)
@@ -165,7 +253,6 @@ dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol_p,
         if (x_p)
                 *x_p = x;
 
-out:
         return 0;
 }
 
@@ -216,52 +303,65 @@ dht_local_wipe (xlator_t *this, dht_local_t *local)
                 local->selfheal.layout = NULL;
         }
 
-        if (local->newpath) {
-                GF_FREE (local->newpath);
-        }
+        GF_FREE (local->newpath);
 
-        if (local->key) {
-                GF_FREE (local->key);
-        }
+        GF_FREE (local->key);
+
+        GF_FREE (local->rebalance.vector);
 
-        GF_FREE (local);
+        if (local->rebalance.iobref)
+                iobref_unref (local->rebalance.iobref);
+
+        mem_put (local);
 }
 
 
 dht_local_t *
-dht_local_init (call_frame_t *frame)
+dht_local_init (call_frame_t *frame, loc_t *loc, fd_t *fd, glusterfs_fop_t fop)
 {
         dht_local_t *local = NULL;
+        inode_t     *inode = NULL;
+        int          ret   = 0;
 
-        /* TODO: use mem-pool */
-        local = GF_CALLOC (1, sizeof (*local),
-                           gf_dht_mt_dht_local_t);
-
+        local = mem_get0 (THIS->local_pool);
         if (!local)
-                return NULL;
+                goto out;
 
-        local->op_ret = -1;
-        local->op_errno = EUCLEAN;
+        if (loc) {
+                ret = loc_copy (&local->loc, loc);
+                if (ret)
+                        goto out;
 
-        frame->local = local;
+                inode = loc->inode;
+        }
 
-        return local;
-}
+        if (fd) {
+                local->fd = fd_ref (fd);
+                if (!inode)
+                        inode = fd->inode;
+        }
 
+        local->op_ret   = -1;
+        local->op_errno = EUCLEAN;
+        local->fop      = fop;
 
-char *
-basestr (const char *str)
-{
-        char *basestr = NULL;
+        if (inode) {
+                local->layout   = dht_layout_get (frame->this, inode);
+                local->cached_subvol = dht_subvol_get_cached (frame->this,
+                                                              inode);
+        }
 
-        basestr = strrchr (str, '/');
-        if (basestr)
-                basestr ++;
+        frame->local = local;
 
-        return basestr;
+out:
+        if (ret) {
+                if (local)
+                        mem_put (local);
+                local = NULL;
+        }
+        return local;
 }
 
-
 xlator_t *
 dht_first_up_subvol (xlator_t *this)
 {
@@ -326,17 +426,23 @@ dht_subvol_get_hashed (xlator_t *this, loc_t *loc)
         dht_layout_t *layout = NULL;
         xlator_t     *subvol = NULL;
 
-        if (is_fs_root (loc)) {
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO (this->name, loc, out);
+
+        if (__is_root_gfid (loc->gfid)) {
                 subvol = dht_first_up_subvol (this);
                 goto out;
         }
 
+        GF_VALIDATE_OR_GOTO (this->name, loc->parent, out);
+        GF_VALIDATE_OR_GOTO (this->name, loc->name, out);
+
         layout = dht_layout_get (this, loc->parent);
 
         if (!layout) {
                 gf_log (this->name, GF_LOG_DEBUG,
-                        "layout missing path=%s parent=%"PRId64,
-                        loc->path, loc->parent->ino);
+                        "layout missing path=%s parent=%s",
+                        loc->path, uuid_utoa (loc->parent->gfid));
                 goto out;
         }
 
@@ -364,6 +470,8 @@ dht_subvol_get_cached (xlator_t *this, inode_t *inode)
         dht_layout_t *layout = NULL;
         xlator_t     *subvol = NULL;
 
+        GF_VALIDATE_OR_GOTO (this->name, this, out);
+        GF_VALIDATE_OR_GOTO (this->name, inode, out);
 
         layout = dht_layout_get (this, inode);
 
@@ -405,7 +513,36 @@ out:
         return next;
 }
 
+/* This func wraps around, if prev is actually the last subvol.
+ */
+xlator_t *
+dht_subvol_next_available (xlator_t *this, xlator_t *prev)
+{
+        dht_conf_t *conf = NULL;
+        int         i = 0;
+        xlator_t   *next = NULL;
+
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (conf->subvolumes[i] == prev) {
+                        /* if prev is last in conf->subvolumes, then wrap
+                         * around.
+                         */
+                        if ((i + 1) < conf->subvolume_cnt) {
+                                next = conf->subvolumes[i + 1];
+                        } else {
+                                next = conf->subvolumes[0];
+                        }
+                        break;
+                }
+        }
 
+out:
+        return next;
+}
 int
 dht_subvol_cnt (xlator_t *this, xlator_t *subvol)
 {
@@ -434,6 +571,15 @@ out:
                         (a) = (b);              \
         } while (0)
 
+
+#define set_if_greater_time(a, an, b, bn) do {                          \
+                if (((a) < (b)) || (((a) == (b)) && ((an) < (bn)))){    \
+                        (a) = (b);                                      \
+                        (an) = (bn);                                    \
+                }                                                       \
+        } while (0)                                                     \
+
+
 int
 dht_iatt_merge (xlator_t *this, struct iatt *to,
                 struct iatt *from, xlator_t *subvol)
@@ -457,9 +603,12 @@ dht_iatt_merge (xlator_t *this, struct iatt *to,
         set_if_greater (to->ia_uid, from->ia_uid);
         set_if_greater (to->ia_gid, from->ia_gid);
 
-        set_if_greater (to->ia_atime, from->ia_atime);
-        set_if_greater (to->ia_mtime, from->ia_mtime);
-        set_if_greater (to->ia_ctime, from->ia_ctime);
+        set_if_greater_time(to->ia_atime, to->ia_atime_nsec,
+                            from->ia_atime, from->ia_atime_nsec);
+        set_if_greater_time (to->ia_mtime, to->ia_mtime_nsec,
+                             from->ia_mtime, from->ia_mtime_nsec);
+        set_if_greater_time (to->ia_ctime, to->ia_ctime_nsec,
+                             from->ia_ctime, from->ia_ctime_nsec);
 
         return 0;
 }
@@ -496,3 +645,562 @@ err:
         loc_wipe (child);
         return -1;
 }
+
+
+
+int
+dht_init_subvolumes (xlator_t *this, dht_conf_t *conf)
+{
+        xlator_list_t *subvols = NULL;
+        int            cnt = 0;
+
+        if (!conf)
+                return -1;
+
+        for (subvols = this->children; subvols; subvols = subvols->next)
+                cnt++;
+
+        conf->subvolumes = GF_CALLOC (cnt, sizeof (xlator_t *),
+                                      gf_dht_mt_xlator_t);
+        if (!conf->subvolumes) {
+                return -1;
+        }
+        conf->subvolume_cnt = cnt;
+
+        cnt = 0;
+        for (subvols = this->children; subvols; subvols = subvols->next)
+                conf->subvolumes[cnt++] = subvols->xlator;
+
+        conf->subvolume_status = GF_CALLOC (cnt, sizeof (char),
+                                            gf_dht_mt_char);
+        if (!conf->subvolume_status) {
+                return -1;
+        }
+
+        conf->last_event = GF_CALLOC (cnt, sizeof (int),
+                                      gf_dht_mt_char);
+        if (!conf->last_event) {
+                return -1;
+        }
+
+        conf->subvol_up_time = GF_CALLOC (cnt, sizeof (time_t),
+                                          gf_dht_mt_subvol_time);
+        if (!conf->subvol_up_time) {
+                return -1;
+        }
+
+        conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
+                                    gf_dht_mt_dht_du_t);
+        if (!conf->du_stats) {
+                return -1;
+        }
+
+        conf->decommissioned_bricks = GF_CALLOC (cnt, sizeof (xlator_t *),
+                                                 gf_dht_mt_xlator_t);
+        if (!conf->decommissioned_bricks) {
+                return -1;
+        }
+
+        return 0;
+}
+
+
+
+
+static int
+dht_migration_complete_check_done (int op_ret, call_frame_t *frame, void *data)
+{
+        dht_local_t *local = NULL;
+
+        local = frame->local;
+
+        local->rebalance.target_op_fn (THIS, frame, op_ret);
+
+        return 0;
+}
+
+
+int
+dht_migration_complete_check_task (void *data)
+{
+        int           ret      = -1;
+        xlator_t     *src_node = NULL;
+        xlator_t     *dst_node = NULL;
+        dht_local_t  *local    = NULL;
+        dict_t       *dict     = NULL;
+        dht_layout_t *layout   = NULL;
+        struct iatt   stbuf    = {0,};
+        xlator_t     *this     = NULL;
+        call_frame_t *frame    = NULL;
+        loc_t         tmp_loc  = {0,};
+        char         *path     = NULL;
+        dht_conf_t   *conf     = NULL;
+        inode_t      *inode    = NULL;
+        fd_t         *iter_fd  = NULL;
+        uint64_t      tmp_subvol = 0;
+        int           open_failed = 0;
+
+        this  = THIS;
+        frame = data;
+        local = frame->local;
+        conf = this->private;
+
+        src_node = local->cached_subvol;
+
+        if (!local->loc.inode && !local->fd) {
+                local->op_errno = EINVAL;
+                goto out;
+        }
+
+        inode = (!local->fd) ? local->loc.inode : local->fd->inode;
+
+        /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
+         * as root:root. If a fd is already open, access check wont be done*/
+
+        if (!local->loc.inode) {
+                ret = syncop_fgetxattr (src_node, local->fd, &dict,
+                                        conf->link_xattr_name);
+        } else {
+                SYNCTASK_SETID (0, 0);
+                ret = syncop_getxattr (src_node, &local->loc, &dict,
+                                       conf->link_xattr_name);
+                SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+        }
+
+        if (!ret)
+                dst_node = dht_linkfile_subvol (this, NULL, NULL, dict);
+
+        if (ret) {
+                if (!dht_inode_missing(-ret) || (!local->loc.inode)) {
+                        local->op_errno = -ret;
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: failed to get the 'linkto' xattr %s",
+                                local->loc.path, strerror (-ret));
+                        ret = -1;
+                        goto out;
+                }
+
+                /* Need to do lookup on hashed subvol, then get the file */
+                ret = syncop_lookup (this, &local->loc, NULL, &stbuf, NULL,
+                                     NULL);
+                if (ret) {
+                        local->op_errno = -ret;
+                        ret = -1;
+                        goto out;
+                }
+
+                dst_node = dht_subvol_get_cached (this, local->loc.inode);
+        }
+
+        if (!dst_node) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to get the destination node",
+                        local->loc.path);
+                ret = -1;
+                local->op_errno = EINVAL;
+                goto out;
+        }
+
+        /* lookup on dst */
+        if (local->loc.inode) {
+                ret = syncop_lookup (dst_node, &local->loc, NULL, &stbuf, NULL,
+                                     NULL);
+
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: failed to lookup the file on %s",
+                                local->loc.path, dst_node->name);
+                        local->op_errno = -ret;
+                        ret = -1;
+                        goto out;
+                }
+
+                if (uuid_compare (stbuf.ia_gfid, local->loc.inode->gfid)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: gfid different on the target file on %s",
+                                local->loc.path, dst_node->name);
+                        ret = -1;
+                        local->op_errno = EIO;
+                        goto out;
+                }
+        }
+
+        /* update inode ctx (the layout) */
+        dht_layout_unref (this, local->layout);
+
+        ret = dht_layout_preset (this, dst_node, inode);
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "%s: could not set preset layout for subvol %s",
+                        local->loc.path, dst_node->name);
+                ret   = -1;
+                local->op_errno = EINVAL;
+                goto out;
+        }
+
+        layout = dht_layout_for_subvol (this, dst_node);
+        if (!layout) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "%s: no pre-set layout for subvolume %s",
+                        local->loc.path, dst_node ? dst_node->name : "<nil>");
+                ret = -1;
+                local->op_errno = EINVAL;
+                goto out;
+        }
+
+        ret = dht_layout_set (this, inode, layout);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to set the new layout",
+                        local->loc.path);
+                local->op_errno = EINVAL;
+                goto out;
+        }
+
+        local->cached_subvol = dst_node;
+        ret = 0;
+
+        /* once we detect the migration complete, the inode-ctx2 is no more
+           required.. delete the ctx and also, it means, open() already
+           done on all the fd of inode */
+        ret = inode_ctx_reset1 (inode, this, &tmp_subvol);
+        if (tmp_subvol)
+                goto out;
+
+        if (list_empty (&inode->fd_list))
+                goto out;
+
+        /* perform open as root:root. There is window between linkfile
+         * creation(root:root) and setattr with the correct uid/gid
+         */
+        SYNCTASK_SETID(0, 0);
+
+        /* perform 'open()' on all the fd's present on the inode */
+        tmp_loc.inode = inode;
+        inode_path (inode, NULL, &path);
+        if (path)
+                tmp_loc.path = path;
+        list_for_each_entry (iter_fd, &inode->fd_list, inode_list) {
+                if (fd_is_anonymous (iter_fd))
+                        continue;
+
+                /* flags for open are stripped down to allow following the
+                 * new location of the file, otherwise we can get EEXIST or
+                 * truncate the file again as rebalance is moving the data */
+                ret = syncop_open (dst_node, &tmp_loc,
+                                   (iter_fd->flags &
+                                   ~(O_CREAT | O_EXCL | O_TRUNC)), iter_fd);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR, "failed to open "
+                                "the fd (%p, flags=0%o) on file %s @ %s",
+                                iter_fd, iter_fd->flags, path, dst_node->name);
+                        open_failed = 1;
+                        local->op_errno = -ret;
+                        ret = -1;
+                }
+        }
+        GF_FREE (path);
+
+        SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+
+        if (open_failed) {
+                ret = -1;
+                goto out;
+        }
+        ret = 0;
+out:
+
+        return ret;
+}
+
+int
+dht_rebalance_complete_check (xlator_t *this, call_frame_t *frame)
+{
+        int         ret     = -1;
+
+        ret = synctask_new (this->ctx->env, dht_migration_complete_check_task,
+                            dht_migration_complete_check_done,
+                            frame, frame);
+        return ret;
+}
+
+/* During 'in-progress' state, both nodes should have the file */
+static int
+dht_inprogress_check_done (int op_ret, call_frame_t *sync_frame, void *data)
+{
+        dht_local_t *local = NULL;
+
+        local = sync_frame->local;
+
+        local->rebalance.target_op_fn (THIS, sync_frame, op_ret);
+
+        return 0;
+}
+
+static int
+dht_rebalance_inprogress_task (void *data)
+{
+        int           ret      = -1;
+        xlator_t     *src_node = NULL;
+        xlator_t     *dst_node = NULL;
+        dht_local_t  *local    = NULL;
+        dict_t       *dict     = NULL;
+        call_frame_t *frame    = NULL;
+        xlator_t     *this     = NULL;
+        char         *path     = NULL;
+        struct iatt   stbuf    = {0,};
+        loc_t         tmp_loc  = {0,};
+        dht_conf_t   *conf     = NULL;
+        inode_t      *inode    = NULL;
+        fd_t         *iter_fd  = NULL;
+        int           open_failed = 0;
+
+        this  = THIS;
+        frame = data;
+        local = frame->local;
+        conf = this->private;
+
+        src_node = local->cached_subvol;
+
+        if (!local->loc.inode && !local->fd)
+                goto out;
+
+        inode = (!local->fd) ? local->loc.inode : local->fd->inode;
+
+        /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
+         * as root:root. If a fd is already open, access check wont be done*/
+        if (local->loc.inode) {
+                SYNCTASK_SETID (0, 0);
+                ret = syncop_getxattr (src_node, &local->loc, &dict,
+                                       conf->link_xattr_name);
+                SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+        } else {
+                ret = syncop_fgetxattr (src_node, local->fd, &dict,
+                                        conf->link_xattr_name);
+        }
+
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to get the 'linkto' xattr %s",
+                        local->loc.path, strerror (-ret));
+                ret = -1;
+                goto out;
+        }
+
+        dst_node = dht_linkfile_subvol (this, NULL, NULL, dict);
+        if (!dst_node) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to get the 'linkto' xattr from dict",
+                        local->loc.path);
+                ret = -1;
+                goto out;
+        }
+
+        local->rebalance.target_node = dst_node;
+
+        if (local->loc.inode) {
+                /* lookup on dst */
+                ret = syncop_lookup (dst_node, &local->loc, NULL,
+                                     &stbuf, NULL, NULL);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: failed to lookup the file on %s",
+                                local->loc.path, dst_node->name);
+                        ret = -1;
+                        goto out;
+                }
+
+                if (uuid_compare (stbuf.ia_gfid, local->loc.inode->gfid)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: gfid different on the target file on %s",
+                                local->loc.path, dst_node->name);
+                        ret = -1;
+                        goto out;
+                }
+        }
+
+        ret = 0;
+
+        if (list_empty (&inode->fd_list))
+                goto done;
+
+        /* perform open as root:root. There is window between linkfile
+         * creation(root:root) and setattr with the correct uid/gid
+         */
+        SYNCTASK_SETID (0, 0);
+
+        tmp_loc.inode = inode;
+        inode_path (inode, NULL, &path);
+        if (path)
+                tmp_loc.path = path;
+
+        list_for_each_entry (iter_fd, &inode->fd_list, inode_list) {
+                if (fd_is_anonymous (iter_fd))
+                        continue;
+
+                /* flags for open are stripped down to allow following the
+                 * new location of the file, otherwise we can get EEXIST or
+                 * truncate the file again as rebalance is moving the data */
+                ret = syncop_open (dst_node, &tmp_loc,
+                                   (iter_fd->flags &
+                                   ~(O_CREAT | O_EXCL | O_TRUNC)), iter_fd);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR, "failed to send open "
+                                "the fd (%p, flags=0%o) on file %s @ %s",
+                                iter_fd, iter_fd->flags, path, dst_node->name);
+                        ret = -1;
+                        open_failed = 1;
+                }
+        }
+        GF_FREE (path);
+
+        SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+
+        if (open_failed) {
+                ret = -1;
+                goto out;
+        }
+
+done:
+        ret = dht_inode_ctx_set1 (this, inode, dst_node);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to set inode-ctx target file at %s",
+                        local->loc.path, dst_node->name);
+                goto out;
+        }
+
+        ret = 0;
+out:
+        return ret;
+}
+
+int
+dht_rebalance_in_progress_check (xlator_t *this, call_frame_t *frame)
+{
+
+        int         ret     = -1;
+
+        ret = synctask_new (this->ctx->env, dht_rebalance_inprogress_task,
+                            dht_inprogress_check_done,
+                            frame, frame);
+        return ret;
+}
+
+int
+dht_inode_ctx_layout_set (inode_t *inode, xlator_t *this,
+                          dht_layout_t *layout_int)
+{
+        dht_inode_ctx_t         *ctx            = NULL;
+        int                      ret            = -1;
+
+        ret = dht_inode_ctx_get (inode, this, &ctx);
+        if (!ret && ctx) {
+                ctx->layout = layout_int;
+        } else {
+                ctx = GF_CALLOC (1, sizeof (*ctx), gf_dht_mt_inode_ctx_t);
+                if (!ctx)
+                        return ret;
+                ctx->layout = layout_int;
+        }
+
+        ret = dht_inode_ctx_set (inode, this, ctx);
+
+        return ret;
+}
+
+
+void
+dht_inode_ctx_time_set (inode_t *inode, xlator_t *this, struct iatt *stat)
+{
+        dht_inode_ctx_t         *ctx            = NULL;
+        dht_stat_time_t         *time           = 0;
+        int                      ret            = -1;
+
+        ret = dht_inode_ctx_get (inode, this, &ctx);
+
+        if (ret)
+		return;
+
+        time = &ctx->time;
+
+	time->mtime      = stat->ia_mtime;
+	time->mtime_nsec = stat->ia_mtime_nsec;
+
+	time->ctime      = stat->ia_ctime;
+	time->ctime_nsec = stat->ia_ctime_nsec;
+
+	time->atime      = stat->ia_atime;
+	time->atime_nsec = stat->ia_atime_nsec;
+
+	return;
+}
+
+
+int
+dht_inode_ctx_time_update (inode_t *inode, xlator_t *this, struct iatt *stat,
+                           int32_t post)
+{
+        dht_inode_ctx_t         *ctx            = NULL;
+        dht_stat_time_t         *time           = 0;
+        int                      ret            = -1;
+
+        GF_VALIDATE_OR_GOTO (this->name, stat, out);
+        GF_VALIDATE_OR_GOTO (this->name, inode, out);
+
+        ret = dht_inode_ctx_get (inode, this, &ctx);
+
+        if (ret) {
+                ctx = GF_CALLOC (1, sizeof (*ctx), gf_dht_mt_inode_ctx_t);
+                if (!ctx)
+                        return -1;
+        }
+
+        time = &ctx->time;
+
+        DHT_UPDATE_TIME(time->mtime, time->mtime_nsec,
+                        stat->ia_mtime, stat->ia_mtime_nsec, inode, post);
+        DHT_UPDATE_TIME(time->ctime, time->ctime_nsec,
+                        stat->ia_ctime, stat->ia_ctime_nsec, inode, post);
+        DHT_UPDATE_TIME(time->atime, time->atime_nsec,
+                        stat->ia_atime, stat->ia_atime_nsec, inode, post);
+
+        ret = dht_inode_ctx_set (inode, this, ctx);
+out:
+        return 0;
+}
+
+int
+dht_inode_ctx_get (inode_t *inode, xlator_t *this, dht_inode_ctx_t **ctx)
+{
+        int             ret     = -1;
+        uint64_t        ctx_int = 0;
+
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO (this->name, inode, out);
+
+        ret = inode_ctx_get (inode, this, &ctx_int);
+
+        if (ret)
+                return ret;
+
+        if (ctx)
+                *ctx = (dht_inode_ctx_t *) ctx_int;
+out:
+        return ret;
+}
+
+int dht_inode_ctx_set (inode_t *inode, xlator_t *this, dht_inode_ctx_t *ctx)
+{
+        int             ret = -1;
+        uint64_t        ctx_int = 0;
+
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO (this->name, inode, out);
+        GF_VALIDATE_OR_GOTO (this->name, ctx, out);
+
+        ctx_int = (long)ctx;
+        ret = inode_ctx_set (inode, this, &ctx_int);
+out:
+        return ret;
+}
diff --git a/xlators/cluster/dht/src/dht-inode-read.c b/xlators/cluster/dht/src/dht-inode-read.c
new file mode 100644
index 000000000..e8a9a7196
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-inode-read.c
@@ -0,0 +1,1139 @@
+/*
+  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 _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "dht-common.h"
+
+int dht_access2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_readv2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_attr2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_open2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_flush2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_lk2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_fsync2 (xlator_t *this, call_frame_t *frame, int ret);
+
+int
+dht_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+              int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = 0;
+
+        local = frame->local;
+        prev = cookie;
+
+        local->op_errno = op_errno;
+        if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        if (!op_ret || (local->call_cnt != 1))
+                goto out;
+
+        /* rebalance would have happened */
+        local->rebalance.target_op_fn = dht_open2;
+        ret = dht_rebalance_complete_check (this, frame);
+        if (!ret)
+                return 0;
+
+out:
+        DHT_STACK_UNWIND (open, frame, op_ret, op_errno, local->fd, xdata);
+
+        return 0;
+}
+
+int
+dht_open2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t *local  = NULL;
+        xlator_t    *subvol = NULL;
+        int          op_errno = EINVAL;
+
+        local = frame->local;
+        if (!local)
+                goto out;
+
+        op_errno = ENOENT;
+        if (op_ret)
+                goto out;
+
+        local->call_cnt = 2;
+        subvol = local->cached_subvol;
+
+        STACK_WIND (frame, dht_open_cbk, subvol, subvol->fops->open,
+                    &local->loc, local->rebalance.flags, local->fd,
+                    NULL);
+        return 0;
+
+out:
+        DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
+
+int
+dht_open (call_frame_t *frame, xlator_t *this,
+          loc_t *loc, int flags, fd_t *fd, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, loc, fd, GF_FOP_OPEN);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        local->rebalance.flags = flags;
+        local->call_cnt = 1;
+
+        STACK_WIND (frame, dht_open_cbk, subvol, subvol->fops->open,
+                    loc, flags, fd, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (open, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+int
+dht_file_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, struct iatt *stbuf, dict_t *xdata)
+{
+        xlator_t     *subvol = 0;
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        inode_t      *inode = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+                local->op_errno = op_errno;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        if (local->call_cnt != 1)
+                goto out;
+
+        local->op_errno = op_errno;
+        /* Check if the rebalance phase2 is true */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (stbuf)) {
+                inode = (local->fd) ? local->fd->inode : local->loc.inode;
+                ret = dht_inode_ctx_get1 (this, inode, &subvol);
+                if (!subvol) {
+                        /* Phase 2 of migration */
+                        local->rebalance.target_op_fn = dht_attr2;
+                        ret = dht_rebalance_complete_check (this, frame);
+                        if (!ret)
+                                return 0;
+                } else {
+                        /* value is already set in fd_ctx, that means no need
+                           to check for whether its complete or not. */
+                        dht_attr2 (this, frame, 0);
+                        return 0;
+                }
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (stat, frame, op_ret, op_errno, stbuf, xdata);
+err:
+        return 0;
+}
+
+int
+dht_attr2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t *local  = NULL;
+        xlator_t    *subvol = NULL;
+        int          op_errno = EINVAL;
+
+        local = frame->local;
+        if (!local)
+                goto out;
+
+        op_errno = local->op_errno;
+        if (op_ret == -1)
+                goto out;
+
+        subvol = local->cached_subvol;
+        local->call_cnt = 2;
+
+        if (local->fop == GF_FOP_FSTAT) {
+                STACK_WIND (frame, dht_file_attr_cbk, subvol,
+                            subvol->fops->fstat, local->fd, NULL);
+        } else {
+                STACK_WIND (frame, dht_file_attr_cbk, subvol,
+                            subvol->fops->stat, &local->loc, NULL);
+        }
+        return 0;
+
+out:
+        DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
+
+int
+dht_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+              int op_ret, int op_errno, struct iatt *stbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "subvolume %s returned -1 (%s)",
+                                prev->this->name, strerror (op_errno));
+
+                        goto unlock;
+                }
+
+                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+
+                local->op_ret = 0;
+        }
+unlock:
+        UNLOCK (&frame->lock);
+out:
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+                DHT_STACK_UNWIND (stat, frame, local->op_ret, local->op_errno,
+                                  &local->stbuf, xdata);
+        }
+err:
+        return 0;
+}
+
+int
+dht_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+        dht_layout_t *layout = NULL;
+        int           i = 0;
+        int           call_cnt = 0;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_STAT);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (IA_ISREG (loc->inode->ia_type)) {
+                local->call_cnt = 1;
+
+                subvol = local->cached_subvol;
+
+                STACK_WIND (frame, dht_file_attr_cbk, subvol,
+                            subvol->fops->stat, loc, xdata);
+
+                return 0;
+        }
+
+        local->call_cnt = call_cnt = layout->cnt;
+
+        for (i = 0; i < call_cnt; i++) {
+                subvol = layout->list[i].xlator;
+
+                STACK_WIND (frame, dht_attr_cbk,
+                            subvol, subvol->fops->stat,
+                            loc, xdata);
+        }
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+        dht_layout_t *layout = NULL;
+        int           i = 0;
+        int           call_cnt = 0;
+
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FSTAT);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (IA_ISREG (fd->inode->ia_type)) {
+                local->call_cnt = 1;
+
+                subvol = local->cached_subvol;
+
+                STACK_WIND (frame, dht_file_attr_cbk, subvol,
+                            subvol->fops->fstat, fd, xdata);
+
+                return 0;
+        }
+
+        local->call_cnt = call_cnt = layout->cnt;
+
+        for (i = 0; i < call_cnt; i++) {
+                subvol = layout->list[i].xlator;
+                STACK_WIND (frame, dht_attr_cbk,
+                            subvol, subvol->fops->fstat,
+                            fd, xdata);
+        }
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fstat, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+int
+dht_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int op_ret, int op_errno,
+               struct iovec *vector, int count, struct iatt *stbuf,
+               struct iobref *iobref, dict_t *xdata)
+{
+        dht_local_t *local      = NULL;
+        int          ret        = 0;
+        inode_t     *inode      = NULL;
+        xlator_t    *subvol = 0;
+
+        local = frame->local;
+        if (!local) {
+                op_ret = -1;
+                op_errno = EINVAL;
+                goto out;
+        }
+
+        /* This is already second try, no need for re-check */
+        if (local->call_cnt != 1)
+                goto out;
+
+        if ((op_ret == -1) && !dht_inode_missing(op_errno))
+                goto out;
+
+        local->op_errno = op_errno;
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (stbuf)) {
+                /* File would be migrated to other node */
+                ret = dht_inode_ctx_get1 (this, inode, &subvol);
+                if (!subvol) {
+                        local->rebalance.target_op_fn = dht_readv2;
+                        ret = dht_rebalance_complete_check (this, frame);
+                        if (!ret)
+                                return 0;
+                } else {
+                        /* value is already set in fd_ctx, that means no need
+                           to check for whether its complete or not. */
+                        dht_readv2 (this, frame, 0);
+                        return 0;
+                }
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (readv, frame, op_ret, op_errno, vector, count, stbuf,
+                          iobref, xdata);
+
+        return 0;
+}
+
+int
+dht_readv2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t *local  = NULL;
+        xlator_t    *subvol = NULL;
+        int          op_errno = EINVAL;
+
+        local = frame->local;
+        if (!local)
+                goto out;
+
+        op_errno = local->op_errno;
+        if (op_ret == -1)
+                goto out;
+
+        local->call_cnt = 2;
+        subvol = local->cached_subvol;
+
+        STACK_WIND (frame, dht_readv_cbk, subvol, subvol->fops->readv,
+                    local->fd, local->rebalance.size, local->rebalance.offset,
+                    local->rebalance.flags, NULL);
+
+        return 0;
+
+out:
+        DHT_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
+        return 0;
+}
+
+int
+dht_readv (call_frame_t *frame, xlator_t *this,
+           fd_t *fd, size_t size, off_t off, uint32_t flags, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_READ);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        local->rebalance.offset = off;
+        local->rebalance.size   = size;
+        local->rebalance.flags  = flags;
+        local->call_cnt = 1;
+
+        STACK_WIND (frame, dht_readv_cbk,
+                    subvol, subvol->fops->readv,
+                    fd, size, off, flags, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
+
+        return 0;
+}
+
+int
+dht_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                int op_ret, int op_errno, dict_t *xdata)
+{
+        int          ret = -1;
+        dht_local_t *local = NULL;
+        xlator_t    *subvol = NULL;
+        call_frame_t *prev = NULL;
+
+        local = frame->local;
+        prev = cookie;
+
+        if (!prev || !prev->this)
+                goto out;
+        if (local->call_cnt != 1)
+                goto out;
+        if ((op_ret == -1) && (op_errno == ENOTCONN) &&
+            IA_ISDIR(local->loc.inode->ia_type)) {
+
+                subvol = dht_subvol_next_available (this, prev->this);
+                if (!subvol)
+                        goto out;
+
+                /* check if we are done with visiting every node */
+                if (subvol == local->cached_subvol) {
+                        goto out;
+                }
+
+                STACK_WIND (frame, dht_access_cbk, subvol, subvol->fops->access,
+                            &local->loc, local->rebalance.flags, NULL);
+                return 0;
+        }
+        if ((op_ret == -1) && dht_inode_missing(op_errno)) {
+                /* File would be migrated to other node */
+                local->op_errno = op_errno;
+                local->rebalance.target_op_fn = dht_access2;
+                ret = dht_rebalance_complete_check (frame->this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STACK_UNWIND (access, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+int
+dht_access2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t *local  = NULL;
+        xlator_t    *subvol = NULL;
+        int          op_errno = EINVAL;
+
+        local = frame->local;
+        if (!local)
+                goto out;
+
+        op_errno = local->op_errno;
+        if (op_ret == -1)
+                goto out;
+
+        local->call_cnt = 2;
+        subvol = local->cached_subvol;
+
+        STACK_WIND (frame, dht_access_cbk, subvol, subvol->fops->access,
+                    &local->loc, local->rebalance.flags, NULL);
+
+        return 0;
+
+out:
+        DHT_STACK_UNWIND (access, frame, -1, op_errno, NULL);
+        return 0;
+}
+
+
+int
+dht_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+            dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_ACCESS);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->rebalance.flags = mask;
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_access_cbk, subvol, subvol->fops->access,
+                    loc, mask, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (access, frame, -1, op_errno, NULL);
+
+        return 0;
+}
+
+
+int
+dht_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int op_ret, int op_errno, dict_t *xdata)
+{
+        dht_local_t  *local  = NULL;
+        inode_t      *inode  = NULL;
+        xlator_t     *subvol = 0;
+
+        local = frame->local;
+
+        local->op_errno = op_errno;
+
+        if (local->call_cnt != 1)
+                goto out;
+
+        /* If context is set, then send flush() it to the destination */
+        dht_inode_ctx_get1 (this, inode, &subvol);
+        if (subvol) {
+                dht_flush2 (this, frame, 0);
+                return 0;
+        }
+
+out:
+        DHT_STACK_UNWIND (flush, frame, op_ret, op_errno, xdata);
+
+        return 0;
+}
+
+int
+dht_flush2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+
+        local = frame->local;
+
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        STACK_WIND (frame, dht_flush_cbk,
+                    subvol, subvol->fops->flush, local->fd, NULL);
+
+        return 0;
+}
+
+
+int
+dht_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FLUSH);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        local->call_cnt = 1;
+
+        STACK_WIND (frame, dht_flush_cbk,
+                    subvol, subvol->fops->flush, fd, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (flush, frame, -1, op_errno, NULL);
+
+        return 0;
+}
+
+
+int
+dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+               dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        inode_t      *inode = NULL;
+        xlator_t     *subvol = 0;
+
+        local = frame->local;
+        prev = cookie;
+
+        local->op_errno = op_errno;
+        if (op_ret == -1 && !dht_inode_missing(op_errno)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+
+        local->op_errno = op_errno;
+        dht_inode_ctx_get1 (this, inode, &subvol);
+        if (!subvol) {
+                local->rebalance.target_op_fn = dht_fsync2;
+
+                /* Check if the rebalance phase1 is true */
+                if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                        dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                        dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+
+                        ret = dht_rebalance_in_progress_check (this, frame);
+                }
+
+                /* Check if the rebalance phase2 is true */
+                if (IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                        ret = dht_rebalance_complete_check (this, frame);
+                }
+                if (!ret)
+                        return 0;
+        } else {
+                dht_fsync2 (this, frame, 0);
+                return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (fsync, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+
+        return 0;
+}
+
+int
+dht_fsync2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+
+        local = frame->local;
+
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        STACK_WIND (frame, dht_fsync_cbk, subvol, subvol->fops->fsync,
+                    local->fd, local->rebalance.flags, NULL);
+
+        return 0;
+}
+
+int
+dht_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync,
+           dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FSYNC);
+        if (!local) {
+                op_errno = ENOMEM;
+
+                goto err;
+        }
+
+        local->call_cnt = 1;
+        local->rebalance.flags = datasync;
+
+        subvol = local->cached_subvol;
+
+        STACK_WIND (frame, dht_fsync_cbk, subvol, subvol->fops->fsync,
+                    fd, datasync, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+/* TODO: for 'lk()' call, we need some other special error, may be ESTALE to
+   indicate that lock migration happened on the fd, so we can consider it as
+   phase 2 of migration */
+int
+dht_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+            int op_ret, int op_errno, struct gf_flock *flock, dict_t *xdata)
+{
+        DHT_STACK_UNWIND (lk, frame, op_ret, op_errno, flock, xdata);
+
+        return 0;
+}
+
+
+int
+dht_lk (call_frame_t *frame, xlator_t *this,
+        fd_t *fd, int cmd, struct gf_flock *flock, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        subvol = dht_subvol_get_cached (this, fd->inode);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        /* TODO: for rebalance, we need to preserve the fop arguments */
+        STACK_WIND (frame, dht_lk_cbk, subvol, subvol->fops->lk, fd,
+                    cmd, flock, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (lk, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+/* Symlinks are currently not migrated, so no need for any check here */
+int
+dht_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int op_ret, int op_errno, const char *path,
+                  struct iatt *stbuf, dict_t *xdata)
+{
+        dht_local_t *local = NULL;
+
+        local = frame->local;
+        if (op_ret == -1)
+                goto err;
+
+        if (!local) {
+                op_ret = -1;
+                op_errno = EINVAL;
+        }
+
+err:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (readlink, frame, op_ret, op_errno, path, stbuf, xdata);
+
+        return 0;
+}
+
+
+int
+dht_readlink (call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+              dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_READLINK);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_readlink_cbk,
+                    subvol, subvol->fops->readlink,
+                    loc, size, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (readlink, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+/* Currently no translators on top of 'distribute' will be using
+ * below fops, hence not implementing 'migration' related checks
+ */
+
+int
+dht_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+        DHT_STACK_UNWIND (xattrop, frame, op_ret, op_errno, dict, xdata);
+        return 0;
+}
+
+
+int
+dht_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
+             gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_XATTROP);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        local->call_cnt = 1;
+
+        STACK_WIND (frame,
+                    dht_xattrop_cbk,
+                    subvol, subvol->fops->xattrop,
+                    loc, flags, dict, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (xattrop, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+        DHT_STACK_UNWIND (fxattrop, frame, op_ret, op_errno, dict, xdata);
+        return 0;
+}
+
+
+int
+dht_fxattrop (call_frame_t *frame, xlator_t *this,
+              fd_t *fd, gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        subvol = dht_subvol_get_cached (this, fd->inode);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame,
+                    dht_fxattrop_cbk,
+                    subvol, subvol->fops->fxattrop,
+                    fd, flags, dict, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fxattrop, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_inodelk_cbk (call_frame_t *frame, void *cookie,
+                 xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+        DHT_STACK_UNWIND (inodelk, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+
+int32_t
+dht_inodelk (call_frame_t *frame, xlator_t *this, const char *volume,
+             loc_t *loc, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_INODELK);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        local->call_cnt = 1;
+
+        STACK_WIND (frame,
+                    dht_inodelk_cbk,
+                    subvol, subvol->fops->inodelk,
+                    volume, loc, cmd, lock, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (inodelk, frame, -1, op_errno, NULL);
+
+        return 0;
+}
+
+
+int
+dht_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+        DHT_STACK_UNWIND (finodelk, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+
+int
+dht_finodelk (call_frame_t *frame, xlator_t *this, const char *volume,
+              fd_t *fd, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        subvol = dht_subvol_get_cached (this, fd->inode);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+
+        STACK_WIND (frame, dht_finodelk_cbk, subvol, subvol->fops->finodelk,
+                    volume, fd, cmd, lock, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (finodelk, frame, -1, op_errno, NULL);
+
+        return 0;
+}
diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c
new file mode 100644
index 000000000..576f007e5
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-inode-write.c
@@ -0,0 +1,1017 @@
+/*
+  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 _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "dht-common.h"
+
+int dht_writev2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_truncate2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_setattr2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_fallocate2(xlator_t *this, call_frame_t *frame, int op_ret);
+int dht_discard2(xlator_t *this, call_frame_t *frame, int op_ret);
+int dht_zerofill2(xlator_t *this, call_frame_t *frame, int op_ret);
+
+int
+dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                int op_ret, int op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t *local = NULL;
+        int          ret   = -1;
+        xlator_t    *subvol = NULL;
+
+        if (op_ret == -1 && !dht_inode_missing(op_errno)) {
+                goto out;
+        }
+
+        local = frame->local;
+        if (!local) {
+                op_ret = -1;
+                op_errno = EINVAL;
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                /* preserve the modes of source */
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+
+        local->rebalance.target_op_fn = dht_writev2;
+
+        local->op_errno = op_errno;
+        /* Phase 2 of migration */
+        if (IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* Check if the rebalance phase1 is true */
+        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+
+                ret = dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+                if (subvol) {
+                        dht_writev2 (this, frame, 0);
+                        return 0;
+                }
+                ret = dht_rebalance_in_progress_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+
+        DHT_STACK_UNWIND (writev, frame, op_ret, op_errno, prebuf, postbuf,
+                          xdata);
+
+        return 0;
+}
+
+int
+dht_writev2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+
+        local = frame->local;
+
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        STACK_WIND (frame, dht_writev_cbk,
+                    subvol, subvol->fops->writev,
+                    local->fd, local->rebalance.vector, local->rebalance.count,
+                    local->rebalance.offset, local->rebalance.flags,
+                    local->rebalance.iobref, NULL);
+
+        return 0;
+}
+
+int
+dht_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
+            struct iovec *vector, int count, off_t off, uint32_t flags,
+            struct iobref *iobref, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_WRITE);
+        if (!local) {
+
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+
+        local->rebalance.vector = iov_dup (vector, count);
+        local->rebalance.offset = off;
+        local->rebalance.count = count;
+        local->rebalance.flags = flags;
+        local->rebalance.iobref = iobref_ref (iobref);
+        local->call_cnt = 1;
+
+        STACK_WIND (frame, dht_writev_cbk,
+                    subvol, subvol->fops->writev,
+                    fd, vector, count, off, flags, iobref, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+
+int
+dht_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int op_ret, int op_errno, struct iatt *prebuf,
+                  struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        xlator_t    *subvol = NULL;
+        inode_t      *inode = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+
+        local->rebalance.target_op_fn = dht_truncate2;
+
+        local->op_errno = op_errno;
+        /* Phase 2 of migration */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* Check if the rebalance phase1 is true */
+        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+                inode = (local->fd) ? local->fd->inode : local->loc.inode;
+                dht_inode_ctx_get1 (this, inode, &subvol);
+                if (subvol) {
+                        dht_truncate2 (this, frame, 0);
+                        return 0;
+                }
+                ret = dht_rebalance_in_progress_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (truncate, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+err:
+        return 0;
+}
+
+
+int
+dht_truncate2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+        inode_t      *inode = NULL;
+
+        local = frame->local;
+
+        inode = local->fd ? local->fd->inode : local->loc.inode;
+
+        dht_inode_ctx_get1 (this, inode, &subvol);
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        if (local->fop == GF_FOP_TRUNCATE) {
+                STACK_WIND (frame, dht_truncate_cbk, subvol,
+                            subvol->fops->truncate, &local->loc,
+                            local->rebalance.offset, NULL);
+        } else {
+                STACK_WIND (frame, dht_truncate_cbk, subvol,
+                            subvol->fops->ftruncate, local->fd,
+                            local->rebalance.offset, NULL);
+        }
+
+        return 0;
+}
+
+int
+dht_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+              dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_TRUNCATE);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->rebalance.offset = offset;
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_truncate_cbk,
+                    subvol, subvol->fops->truncate,
+                    loc, offset, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+int
+dht_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FTRUNCATE);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->rebalance.offset = offset;
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_truncate_cbk,
+                    subvol, subvol->fops->ftruncate,
+                    fd, offset, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int op_ret, int op_errno, struct iatt *prebuf,
+                  struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        xlator_t    *subvol = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+        local->rebalance.target_op_fn = dht_fallocate2;
+
+        /* Phase 2 of migration */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* Check if the rebalance phase1 is true */
+        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+                dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+                if (subvol) {
+                        dht_fallocate2 (this, frame, 0);
+                        return 0;
+                }
+                ret = dht_rebalance_in_progress_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (fallocate, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+err:
+        return 0;
+}
+
+int
+dht_fallocate2(xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+
+        local = frame->local;
+
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+	STACK_WIND(frame, dht_fallocate_cbk, subvol, subvol->fops->fallocate,
+		   local->fd, local->rebalance.flags, local->rebalance.offset,
+		   local->rebalance.size, NULL);
+
+        return 0;
+}
+
+int
+dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+	      off_t offset, size_t len, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FALLOCATE);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+	local->rebalance.flags = mode;
+        local->rebalance.offset = offset;
+	local->rebalance.size = len;
+
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_fallocate_cbk,
+                    subvol, subvol->fops->fallocate,
+                    fd, mode, offset, len, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int op_ret, int op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        xlator_t    *subvol = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+        local->rebalance.target_op_fn = dht_discard2;
+
+        /* Phase 2 of migration */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* Check if the rebalance phase1 is true */
+        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+                dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+                if (subvol) {
+                        dht_discard2 (this, frame, 0);
+                        return 0;
+                }
+                ret = dht_rebalance_in_progress_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (discard, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+err:
+        return 0;
+}
+
+int
+dht_discard2(xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+
+        local = frame->local;
+
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+	STACK_WIND(frame, dht_discard_cbk, subvol, subvol->fops->discard,
+		   local->fd, local->rebalance.offset, local->rebalance.size,
+		   NULL);
+
+        return 0;
+}
+
+int
+dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+	    size_t len, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_DISCARD);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->rebalance.offset = offset;
+	local->rebalance.size = len;
+
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_discard_cbk, subvol, subvol->fops->discard,
+                    fd, offset, len, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+int
+dht_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int op_ret, int op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev  = NULL;
+        int           ret   = -1;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+        local->rebalance.target_op_fn = dht_zerofill2;
+        /* Phase 2 of migration */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* Check if the rebalance phase1 is true */
+        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+                ret = fd_ctx_get (local->fd, this, NULL);
+                if (!ret) {
+                        dht_zerofill2 (this, frame, 0);
+                        return 0;
+                }
+                ret = dht_rebalance_in_progress_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (zerofill, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+err:
+        return 0;
+}
+
+int
+dht_zerofill2(xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local          = NULL;
+        xlator_t     *subvol         = NULL;
+        uint64_t      tmp_subvol     = 0;
+        int           ret            = -1;
+
+        local = frame->local;
+
+        if (local->fd)
+                ret = fd_ctx_get (local->fd, this, &tmp_subvol);
+        if (!ret)
+                subvol = (xlator_t *)(long)tmp_subvol;
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        STACK_WIND(frame, dht_zerofill_cbk, subvol, subvol->fops->zerofill,
+                   local->fd, local->rebalance.offset, local->rebalance.size,
+                   NULL);
+
+        return 0;
+}
+
+int
+dht_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            off_t len, dict_t *xdata)
+{
+        xlator_t     *subvol       = NULL;
+        int           op_errno     = -1;
+        dht_local_t  *local        = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_ZEROFILL);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->rebalance.offset = offset;
+        local->rebalance.size = len;
+
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_zerofill_cbk, subvol, subvol->fops->zerofill,
+                    fd, offset, len, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+
+/* handle cases of migration here for 'setattr()' calls */
+int
+dht_file_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+
+        local = frame->local;
+        prev = cookie;
+
+        local->op_errno = op_errno;
+        if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        if (local->call_cnt != 1)
+                goto out;
+
+        local->rebalance.target_op_fn = dht_setattr2;
+
+        /* Phase 2 of migration */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* At the end of the migration process, whatever 'attr' we
+           have on source file will be migrated to destination file
+           in one shot, hence we don't need to check for in progress
+           state here (ie, PHASE1) */
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (setattr, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+
+        return 0;
+}
+
+int
+dht_setattr2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+        inode_t      *inode = NULL;
+
+        local = frame->local;
+
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
+
+        dht_inode_ctx_get1 (this, inode, &subvol);
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        if (local->fop == GF_FOP_SETATTR) {
+                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
+                            subvol->fops->setattr, &local->loc,
+                            &local->rebalance.stbuf, local->rebalance.flags,
+                            NULL);
+        } else {
+                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
+                            subvol->fops->fsetattr, local->fd,
+                            &local->rebalance.stbuf, local->rebalance.flags,
+                            NULL);
+        }
+
+        return 0;
+}
+
+
+/* Keep the existing code same for all the cases other than regular file */
+int
+dht_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                 int op_ret, int op_errno, struct iatt *statpre,
+                 struct iatt *statpost, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
+
+
+        local = frame->local;
+        prev = cookie;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "subvolume %s returned -1 (%s)",
+                                prev->this->name, strerror (op_errno));
+                        goto unlock;
+                }
+
+                dht_iatt_merge (this, &local->prebuf, statpre, prev->this);
+                dht_iatt_merge (this, &local->stbuf, statpost, prev->this);
+
+                local->op_ret = 0;
+        }
+unlock:
+        UNLOCK (&frame->lock);
+
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+		if (local->op_ret == 0)
+			dht_inode_ctx_time_set (local->loc.inode, this,
+						&local->stbuf);
+                DHT_STACK_UNWIND (setattr, frame, local->op_ret, local->op_errno,
+                                  &local->prebuf, &local->stbuf, xdata);
+	}
+
+        return 0;
+}
+
+
+int
+dht_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
+             struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        dht_layout_t *layout = NULL;
+        dht_local_t  *local  = NULL;
+        int           op_errno = -1;
+        int           i = -1;
+        int           call_cnt = 0;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_SETATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (!layout_is_sane (layout)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "layout is not sane for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (IA_ISREG (loc->inode->ia_type)) {
+                /* in the regular file _cbk(), we need to check for
+                   migration possibilities */
+                local->rebalance.stbuf = *stbuf;
+                local->rebalance.flags = valid;
+                local->call_cnt = 1;
+                subvol = local->cached_subvol;
+
+                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
+                            subvol->fops->setattr,
+                            loc, stbuf, valid, xdata);
+
+                return 0;
+        }
+
+        local->call_cnt = call_cnt = layout->cnt;
+
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (frame, dht_setattr_cbk,
+                            layout->list[i].xlator,
+                            layout->list[i].xlator->fops->setattr,
+                            loc, stbuf, valid, xdata);
+        }
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+              int32_t valid, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        dht_layout_t *layout = NULL;
+        dht_local_t  *local  = NULL;
+        int           op_errno = -1;
+        int           i = -1;
+        int           call_cnt = 0;
+
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FSETATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (!layout_is_sane (layout)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "layout is not sane for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (IA_ISREG (fd->inode->ia_type)) {
+                /* in the regular file _cbk(), we need to check for
+                   migration possibilities */
+                local->rebalance.stbuf = *stbuf;
+                local->rebalance.flags = valid;
+                local->call_cnt = 1;
+                subvol = local->cached_subvol;
+
+                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
+                            subvol->fops->fsetattr,
+                            fd, stbuf, valid, xdata);
+
+                return 0;
+        }
+
+        local->call_cnt = call_cnt = layout->cnt;
+
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (frame, dht_setattr_cbk,
+                            layout->list[i].xlator,
+                            layout->list[i].xlator->fops->fsetattr,
+                            fd, stbuf, valid, xdata);
+        }
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
diff --git a/xlators/cluster/dht/src/dht-layout.c b/xlators/cluster/dht/src/dht-layout.c
index e1ab6c7f1..e1a37b77c 100644
--- a/xlators/cluster/dht/src/dht-layout.c
+++ b/xlators/cluster/dht/src/dht-layout.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef _CONFIG_H
@@ -34,14 +25,12 @@
 
 #define layout_size(cnt) (layout_base_size + (cnt * layout_entry_size))
 
-
 dht_layout_t *
 dht_layout_new (xlator_t *this, int cnt)
 {
         dht_layout_t *layout = NULL;
         dht_conf_t   *conf = NULL;
 
-
         conf = this->private;
 
         layout = GF_CALLOC (1, layout_size (cnt),
@@ -59,6 +48,7 @@ dht_layout_new (xlator_t *this, int cnt)
         }
 
         layout->ref = 1;
+
 out:
         return layout;
 }
@@ -68,9 +58,7 @@ dht_layout_t *
 dht_layout_get (xlator_t *this, inode_t *inode)
 {
         dht_conf_t   *conf = NULL;
-        uint64_t      layout_int = 0;
         dht_layout_t *layout = NULL;
-        int           ret    = -1;
 
         conf = this->private;
         if (!conf)
@@ -78,9 +66,8 @@ dht_layout_get (xlator_t *this, inode_t *inode)
 
         LOCK (&conf->layout_lock);
         {
-                ret = inode_ctx_get (inode, this, &layout_int);
-                if (ret == 0) {
-                        layout = (dht_layout_t *) (unsigned long) layout_int;
+                dht_inode_ctx_layout_get (inode, this, &layout);
+                if (layout) {
                         layout->ref++;
                 }
         }
@@ -98,7 +85,6 @@ dht_layout_set (xlator_t *this, inode_t *inode, dht_layout_t *layout)
         int           oldret = -1;
         int           ret = 0;
         dht_layout_t *old_layout;
-        uint64_t      old_layout_int;
 
         conf = this->private;
         if (!conf)
@@ -106,16 +92,13 @@ dht_layout_set (xlator_t *this, inode_t *inode, dht_layout_t *layout)
 
         LOCK (&conf->layout_lock);
         {
-                oldret = inode_ctx_get (inode, this, &old_layout_int);
-
+                oldret = dht_inode_ctx_layout_get (inode, this, &old_layout);
                 layout->ref++;
-                ret = inode_ctx_put (inode, this, (uint64_t) (unsigned long)
-                                     layout);
+                dht_inode_ctx_layout_set (inode, this, layout);
         }
         UNLOCK (&conf->layout_lock);
 
-        if (oldret == 0) {
-                old_layout = (dht_layout_t *) (unsigned long) old_layout_int;
+        if (!oldret) {
                 dht_layout_unref (this, old_layout);
         }
 
@@ -130,7 +113,7 @@ dht_layout_unref (xlator_t *this, dht_layout_t *layout)
         dht_conf_t  *conf = NULL;
         int          ref = 0;
 
-        if (layout->preset || !this->private)
+        if (!layout || layout->preset || !this->private)
                 return;
 
         conf = this->private;
@@ -174,9 +157,9 @@ dht_layout_search (xlator_t *this, dht_layout_t *layout, const char *name)
         int        ret = 0;
 
 
-        ret = dht_hash_compute (layout->type, name, &hash);
+        ret = dht_hash_compute (this, layout->type, name, &hash);
         if (ret != 0) {
-                gf_log (this->name, GF_LOG_INFO,
+                gf_log (this->name, GF_LOG_WARNING,
                         "hash computation failed for type=%d name=%s",
                         layout->type, name);
                 goto out;
@@ -191,7 +174,7 @@ dht_layout_search (xlator_t *this, dht_layout_t *layout, const char *name)
         }
 
         if (!subvol) {
-                gf_log (this->name, GF_LOG_INFO,
+                gf_log (this->name, GF_LOG_WARNING,
                         "no subvolume for hash (value) = %u", hash);
         }
 
@@ -280,6 +263,9 @@ dht_disk_layout_extract (xlator_t *this, dht_layout_t *layout,
 
         if (disk_layout_p)
                 *disk_layout_p = disk_layout;
+        else
+                GF_FREE (disk_layout);
+
         ret = 0;
 
 out:
@@ -289,7 +275,7 @@ out:
 
 int
 dht_disk_layout_merge (xlator_t *this, dht_layout_t *layout,
-                       int pos, void *disk_layout_raw)
+		       int pos, void *disk_layout_raw, int disk_layout_len)
 {
         int      cnt = 0;
         int      type = 0;
@@ -297,19 +283,38 @@ dht_disk_layout_merge (xlator_t *this, dht_layout_t *layout,
         int      stop_off = 0;
         int      disk_layout[4];
 
-        /* TODO: assert disk_layout_ptr is of required length */
+	if (!disk_layout_raw) {
+		gf_log (this->name, GF_LOG_CRITICAL,
+                        "error no layout on disk for merge");
+		return -1;
+	}
 
-        memcpy (disk_layout, disk_layout_raw, sizeof (disk_layout));
+	GF_ASSERT (disk_layout_len == sizeof (disk_layout));
+
+        memcpy (disk_layout, disk_layout_raw, disk_layout_len);
 
         cnt  = ntoh32 (disk_layout[0]);
         if (cnt != 1) {
-                gf_log (this->name, GF_LOG_INFO,
+                gf_log (this->name, GF_LOG_ERROR,
                         "disk layout has invalid count %d", cnt);
                 return -1;
         }
 
-        /* TODO: assert type is compatible */
-        type      = ntoh32 (disk_layout[1]);
+        type = ntoh32 (disk_layout[1]);
+	switch (type) {
+        case DHT_HASH_TYPE_DM_USER:
+                gf_log (this->name, GF_LOG_DEBUG, "found user-set layout");
+                layout->type = type;
+                /* Fall through. */
+	case DHT_HASH_TYPE_DM:
+		break;
+        default:
+		gf_log (this->name, GF_LOG_CRITICAL,
+			"Catastrophic error layout with unknown type found %d",
+			disk_layout[1]);
+		return -1;
+	}
+
         start_off = ntoh32 (disk_layout[2]);
         stop_off  = ntoh32 (disk_layout[3]);
 
@@ -329,11 +334,12 @@ int
 dht_layout_merge (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
                   int op_ret, int op_errno, dict_t *xattr)
 {
-        int      i     = 0;
-        int      ret   = -1;
-        int      err   = -1;
-        void    *disk_layout_raw = NULL;
-
+        int         i     = 0;
+        int         ret   = -1;
+        int         err   = -1;
+        void       *disk_layout_raw = NULL;
+        int         disk_layout_len = 0;
+        dht_conf_t *conf = this->private;
 
         if (op_ret != 0) {
                 err = op_errno;
@@ -354,12 +360,12 @@ dht_layout_merge (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
 
         if (xattr) {
                 /* during lookup and not mkdir */
-                ret = dict_get_ptr (xattr, "trusted.glusterfs.dht",
-                                    &disk_layout_raw);
+                ret = dict_get_ptr_and_len (xattr, conf->xattr_name,
+					    &disk_layout_raw, &disk_layout_len);
         }
 
         if (ret != 0) {
-                layout->list[i].err = -1;
+                layout->list[i].err = 0;
                 gf_log (this->name, GF_LOG_TRACE,
                         "missing disk layout on %s. err = %d",
                         subvol->name, err);
@@ -367,9 +373,10 @@ dht_layout_merge (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
                 goto out;
         }
 
-        ret = dht_disk_layout_merge (this, layout, i, disk_layout_raw);
+        ret = dht_disk_layout_merge (this, layout, i, disk_layout_raw,
+				     disk_layout_len);
         if (ret != 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
+                gf_log (this->name, GF_LOG_WARNING,
                         "layout merge from subvolume %s failed",
                         subvol->name);
                 goto out;
@@ -405,6 +412,22 @@ dht_layout_entry_swap (dht_layout_t *layout, int i, int j)
         layout->list[j].err    = err_swap;
 }
 
+void
+dht_layout_range_swap (dht_layout_t *layout, int i, int j)
+{
+        uint32_t  start_swap = 0;
+        uint32_t  stop_swap = 0;
+
+        start_swap  = layout->list[i].start;
+        stop_swap   = layout->list[i].stop;
+
+        layout->list[i].start  = layout->list[j].start;
+        layout->list[i].stop   = layout->list[j].stop;
+
+        layout->list[j].start  = start_swap;
+        layout->list[j].stop   = stop_swap;
+}
+
 int64_t
 dht_layout_entry_cmp_volname (dht_layout_t *layout, int i, int j)
 {
@@ -412,17 +435,39 @@ dht_layout_entry_cmp_volname (dht_layout_t *layout, int i, int j)
                         layout->list[j].xlator->name));
 }
 
+
+gf_boolean_t
+dht_is_subvol_in_layout (dht_layout_t *layout, xlator_t *xlator)
+{
+        int i = 0;
+
+        for (i = 0; i < layout->cnt; i++) {
+                /* Check if xlator is already part of layout, and layout is
+                 * non-zero. */
+                if (!strcmp (layout->list[i].xlator->name, xlator->name)) {
+                        if (layout->list[i].start != layout->list[i].stop)
+                                return _gf_true;
+                        break;
+                }
+        }
+        return _gf_false;
+}
+
 int64_t
 dht_layout_entry_cmp (dht_layout_t *layout, int i, int j)
 {
         int64_t diff = 0;
 
-        if (layout->list[i].err || layout->list[j].err)
-                diff = layout->list[i].err - layout->list[j].err;
-        else
-                diff = (int64_t) layout->list[i].start
-                        - (int64_t) layout->list[j].start;
+        /* swap zero'ed out layouts to front, if needed */
+        if (!layout->list[j].start && !layout->list[j].stop) {
+                diff = (int64_t) layout->list[i].stop
+                       - (int64_t) layout->list[j].stop;
+                       goto out;
+        }
+        diff = (int64_t) layout->list[i].start
+                - (int64_t) layout->list[j].start;
 
+out:
         return diff;
 }
 
@@ -471,7 +516,8 @@ dht_layout_sort_volname (dht_layout_t *layout)
 int
 dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
                       uint32_t *holes_p, uint32_t *overlaps_p,
-                      uint32_t *missing_p, uint32_t *down_p, uint32_t *misc_p)
+                      uint32_t *missing_p, uint32_t *down_p, uint32_t *misc_p,
+                      uint32_t *no_space_p)
 {
         uint32_t    overlaps = 0;
         uint32_t    missing  = 0;
@@ -484,30 +530,51 @@ dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
         uint32_t    prev_stop = 0;
         uint32_t    last_stop = 0;
         char        is_virgin = 1;
-
-        /* TODO: explain WTF is happening */
-
+        uint32_t    no_space  = 0;
+
+        /* This funtion scans through the layout spread of a directory to
+           check if there are any anomalies. Prior to calling this function
+           the layout entries should be sorted in the ascending order.
+
+           If the layout entry has err != 0
+                then increment the corresponding anomaly.
+           else
+                if (start of the current layout entry > stop + 1 of previous
+                   non erroneous layout entry)
+                        then it indicates a hole in the layout
+                if (start of the current layout entry < stop + 1 of previous
+                    non erroneous layout entry)
+                         then it indicates an overlap in the layout
+        */
         last_stop = layout->list[0].start - 1;
         prev_stop = last_stop;
 
         for (i = 0; i < layout->cnt; i++) {
-                if (layout->list[i].err) {
-                        switch (layout->list[i].err) {
-                        case -1:
-                        case ENOENT:
-                                missing++;
-                                break;
-                        case ENOTCONN:
-                                down++;
-                                break;
-                        case ENOSPC:
-                                down++;
-                                break;
-                        default:
-                                misc++;
+                switch (layout->list[i].err) {
+                case -1:
+                case ENOENT:
+                case ESTALE:
+                        missing++;
+                        continue;
+                case ENOTCONN:
+                        down++;
+                        continue;
+                case ENOSPC:
+                        no_space++;
+                        continue;
+                case 0:
+                        /* if err == 0 and start == stop, then it is a non misc++;
+                         * participating subvolume(spread-cnt). Then, do not
+                         * check for anomalies. If start != stop, then treat it
+                         * as misc err */
+                        if (layout->list[i].start == layout->list[i].stop) {
+                                continue;
                         }
+                        break;
+                default:
+                        misc++;
                         continue;
-                }
+                 }
 
                 is_virgin = 0;
 
@@ -540,6 +607,9 @@ dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
         if (misc_p)
                 *misc_p = misc;
 
+        if (no_space_p)
+                *no_space_p = no_space;
+
         return ret;
 }
 
@@ -555,7 +625,6 @@ dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout)
         uint32_t     down = 0;
         uint32_t     misc = 0;
 
-
         ret = dht_layout_sort (layout);
         if (ret == -1) {
                 gf_log (this->name, GF_LOG_WARNING,
@@ -565,7 +634,7 @@ dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout)
 
         ret = dht_layout_anomalies (this, loc, layout,
                                     &holes, &overlaps,
-                                    &missing, &down, &misc);
+                                    &missing, &down, &misc, NULL);
         if (ret == -1) {
                 gf_log (this->name, GF_LOG_WARNING,
                         "error while finding anomalies in %s -- not good news",
@@ -583,43 +652,56 @@ dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout)
                                 "found anomalies in %s. holes=%d overlaps=%d",
                                 loc->path, holes, overlaps);
                 }
-                ret = 1;
+                ret = -1;
         }
 
         for (i = 0; i < layout->cnt; i++) {
-                /* TODO During DHT selfheal rewrite (almost) find a better place to
-                 * detect this - probably in dht_layout_anomalies()
+                /* TODO During DHT selfheal rewrite (almost) find a better place
+                 * to detect this - probably in dht_layout_anomalies()
                  */
                 if (layout->list[i].err > 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "path=%s err=%s on subvol=%s",
-                                loc->path, strerror (layout->list[i].err),
-                                (layout->list[i].xlator ?
-                                 layout->list[i].xlator->name : "<>"));
-                        if (layout->list[i].err == ENOENT)
-                                ret = 1;
+                        gf_log_callingfn (this->name, GF_LOG_DEBUG,
+                                          "path=%s err=%s on subvol=%s",
+                                          loc->path,
+                                          strerror (layout->list[i].err),
+                                          (layout->list[i].xlator ?
+                                           layout->list[i].xlator->name
+                                           : "<>"));
+                        if ((layout->list[i].err == ENOENT) && (ret >= 0)) {
+                                ret++;
+                        }
                 }
         }
 
+
 out:
         return ret;
 }
 
+int
+dht_dir_has_layout (dict_t *xattr, char *name)
+{
+
+        void     *disk_layout_raw = NULL;
+
+        return dict_get_ptr (xattr, name, &disk_layout_raw);
+}
 
 int
 dht_layout_dir_mismatch (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
                          loc_t *loc, dict_t *xattr)
 {
-        int       idx = 0;
-        int       pos = -1;
-        int       ret = 0;
-        int       err = 0;
-        int       dict_ret = 0;
-        int32_t   disk_layout[4];
-        void     *disk_layout_raw = NULL;
-        int32_t   count = -1;
-        uint32_t  start_off = -1;
-        uint32_t  stop_off = -1;
+        int         idx = 0;
+        int         pos = -1;
+        int         ret = 0;
+        int         err = 0;
+        int         dict_ret = 0;
+        int32_t     disk_layout[4];
+        void       *disk_layout_raw = NULL;
+        int32_t     count = -1;
+        uint32_t    start_off = -1;
+        uint32_t    stop_off = -1;
+        dht_conf_t *conf = this->private;
 
 
         for (idx = 0; idx < layout->cnt; idx++) {
@@ -649,11 +731,11 @@ dht_layout_dir_mismatch (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
                 goto out;
         }
 
-        dict_ret = dict_get_ptr (xattr, "trusted.glusterfs.dht",
+        dict_ret = dict_get_ptr (xattr, conf->xattr_name,
                                  &disk_layout_raw);
 
         if (dict_ret < 0) {
-                if (err == 0) {
+                if (err == 0 && layout->list[pos].stop) {
                         gf_log (this->name, GF_LOG_INFO,
                                 "%s - disk layout missing", loc->path);
                         ret = -1;
@@ -665,7 +747,7 @@ dht_layout_dir_mismatch (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
 
         count  = ntoh32 (disk_layout[0]);
         if (count != 1) {
-                gf_log (this->name, GF_LOG_INFO,
+                gf_log (this->name, GF_LOG_ERROR,
                         "%s - disk layout has invalid count %d",
                         loc->path, count);
                 ret = -1;
@@ -714,7 +796,7 @@ dht_layout_preset (xlator_t *this, xlator_t *subvol, inode_t *inode)
 
         LOCK (&conf->layout_lock);
         {
-                inode_ctx_put (inode, this, (uint64_t)(long)layout);
+                dht_inode_ctx_layout_set (inode, this, layout);
         }
         UNLOCK (&conf->layout_lock);
 
diff --git a/xlators/cluster/dht/src/dht-linkfile.c b/xlators/cluster/dht/src/dht-linkfile.c
index e936f3119..dbc9d0b3c 100644
--- a/xlators/cluster/dht/src/dht-linkfile.c
+++ b/xlators/cluster/dht/src/dht-linkfile.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef _CONFIG_H
@@ -28,37 +19,106 @@
 #include "compat.h"
 #include "dht-common.h"
 
+int
+dht_linkfile_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno,
+                         inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                         struct iatt *postparent)
+{
+        char          is_linkfile   = 0;
+        dht_conf_t   *conf          = NULL;
+        dht_local_t  *local         = NULL;
+        call_frame_t *prev          = NULL;
 
+        local = frame->local;
+        prev = cookie;
+        conf = this->private;
 
+        if (op_ret)
+                goto out;
+
+        is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+                                         conf->link_xattr_name);
+        if (!is_linkfile)
+                gf_log (this->name, GF_LOG_WARNING, "got non-linkfile %s:%s",
+                        prev->this->name, local->loc.path);
+out:
+        local->linkfile.linkfile_cbk (frame, cookie, this, op_ret, op_errno,
+                                      inode, stbuf, postparent, postparent,
+                                      xattr);
+        return 0;
+}
+
+#define is_equal(a, b) (a == b)
 int
 dht_linkfile_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                          int op_ret, int op_errno, inode_t *inode,
                          struct iatt *stbuf, struct iatt *preparent,
-                         struct iatt *postparent)
+                         struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
+        xlator_t     *subvol = NULL;
+        call_frame_t *prev = NULL;
+        dict_t       *xattrs = NULL;
+        dht_conf_t   *conf = NULL;
+        int           ret = -1;
 
         local = frame->local;
 
+        if (!op_ret)
+                local->linked = _gf_true;
+
+        FRAME_SU_UNDO (frame, dht_local_t);
+
+        if (op_ret && (op_errno == EEXIST)) {
+                conf = this->private;
+                prev = cookie;
+                subvol = prev->this;
+                if (!subvol)
+                        goto out;
+                xattrs = dict_new ();
+                if (!xattrs)
+                        goto out;
+                ret = dict_set_uint32 (xattrs, conf->link_xattr_name, 256);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Failed to set linkto key");
+                        goto out;
+                }
+
+                STACK_WIND (frame, dht_linkfile_lookup_cbk, subvol,
+                            subvol->fops->lookup, &local->loc, xattrs);
+                if (xattrs)
+                        dict_unref (xattrs);
+                return 0;
+        }
+out:
         local->linkfile.linkfile_cbk (frame, cookie, this, op_ret, op_errno,
-                                      inode, stbuf, preparent, postparent);
+                                      inode, stbuf, preparent, postparent,
+                                      xdata);
+        if (xattrs)
+                dict_unref (xattrs);
         return 0;
 }
 
 
 int
 dht_linkfile_create (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
+                     xlator_t *this,
                      xlator_t *tovol, xlator_t *fromvol, loc_t *loc)
 {
         dht_local_t *local = NULL;
         dict_t      *dict = NULL;
         int          need_unref = 0;
         int          ret = 0;
+        dht_conf_t  *conf = this->private;
 
         local = frame->local;
         local->linkfile.linkfile_cbk = linkfile_cbk;
         local->linkfile.srcvol = tovol;
 
+        local->linked = _gf_false;
+
         dict = local->params;
         if (!dict) {
                 dict = dict_new ();
@@ -74,8 +134,12 @@ dht_linkfile_create (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
                                 "%s: gfid set failed", loc->path);
         }
 
-        ret = dict_set_str (dict, "trusted.glusterfs.dht.linkto",
-                                   tovol->name);
+        ret = dict_set_str (dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+        if (ret)
+                gf_log ("dht-linkfile", GF_LOG_INFO,
+                        "%s: internal-fop set failed", loc->path);
+
+        ret = dict_set_str (dict, conf->link_xattr_name, tovol->name);
 
         if (ret < 0) {
                 gf_log (frame->this->name, GF_LOG_INFO,
@@ -84,9 +148,13 @@ dht_linkfile_create (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
                 goto out;
         }
 
+        local->link_subvol = fromvol;
+        /* Always create as root:root. dht_linkfile_attr_heal fixes the
+         * ownsership */
+        FRAME_SU_DO (frame, dht_local_t);
         STACK_WIND (frame, dht_linkfile_create_cbk,
                     fromvol, fromvol->fops->mknod, loc,
-                    S_IFREG | DHT_LINKFILE_MODE, 0, dict);
+                    S_IFREG | DHT_LINKFILE_MODE, 0, 0, dict);
 
         if (need_unref && dict)
                 dict_unref (dict);
@@ -94,7 +162,7 @@ dht_linkfile_create (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
         return 0;
 out:
         local->linkfile.linkfile_cbk (frame, NULL, frame->this, -1, ENOMEM,
-                                      loc->inode, NULL, NULL, NULL);
+                                      loc->inode, NULL, NULL, NULL, NULL);
 
         if (need_unref && dict)
                 dict_unref (dict);
@@ -106,7 +174,8 @@ out:
 int
 dht_linkfile_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                          int32_t op_ret, int32_t op_errno,
-                         struct iatt *preparent, struct iatt *postparent)
+                         struct iatt *preparent, struct iatt *postparent,
+                         dict_t *xdata)
 {
         dht_local_t   *local = NULL;
         call_frame_t  *prev = NULL;
@@ -140,16 +209,17 @@ dht_linkfile_unlink (call_frame_t *frame, xlator_t *this,
                 goto err;
         }
 
-        unlink_local = dht_local_init (unlink_frame);
+        /* Using non-fop value here, as anyways, 'local->fop' is not used in
+           this particular case */
+        unlink_local = dht_local_init (unlink_frame, loc, NULL,
+                                       GF_FOP_MAXVALUE);
         if (!unlink_local) {
                 goto err;
         }
 
-        loc_copy (&unlink_local->loc, loc);
-
         STACK_WIND (unlink_frame, dht_linkfile_unlink_cbk,
                     subvol, subvol->fops->unlink,
-                    &unlink_local->loc);
+                    &unlink_local->loc, 0, NULL);
 
         return 0;
 err:
@@ -174,7 +244,7 @@ dht_linkfile_subvol (xlator_t *this, inode_t *inode, struct iatt *stbuf,
         if (!xattr)
                 goto out;
 
-        ret = dict_get_ptr (xattr, "trusted.glusterfs.dht.linkto", &volname);
+        ret = dict_get_ptr (xattr, conf->link_xattr_name, &volname);
 
         if ((-1 == ret) || !volname)
                 goto out;
@@ -191,141 +261,68 @@ out:
 }
 
 int
-dht_recreate_linkfile_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno,
-                         struct iatt *preparent, struct iatt *postparent)
+dht_linkfile_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, struct iatt *statpre,
+                          struct iatt *statpost, dict_t *xdata)
 {
-        dht_local_t   *local            = NULL;
-        xlator_t      *subvol           = NULL;
-        xlator_t      *src_cached       = NULL;
-        xlator_t      *dst_hashed       = NULL;
-        call_frame_t  *prev             = NULL;
-        loc_t         *loc              = NULL;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, err);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, err);
+        dht_local_t *local = NULL;
+        loc_t *loc = NULL;
 
         local = frame->local;
-        prev = cookie;
-        subvol = prev->this;
-        src_cached = local->src_cached;
-        dst_hashed = local->dst_hashed;
         loc = &local->loc;
 
-        if (!src_cached || !dst_hashed) {
-                gf_log (this->name, GF_LOG_ERROR, "src_cached or dst_hashed"
-                        "subvol is null");
-                local->op_ret = -1;
-                local->op_errno = EINVAL;
-                goto out;
-        }
+        if (op_ret)
+                gf_log (this->name, GF_LOG_ERROR, "setattr of uid/gid on %s"
+                        " :<gfid:%s> failed (%s)",
+                        (loc->path? loc->path: "NULL"),
+                        uuid_utoa(local->gfid), strerror(op_errno));
 
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "unlinking linkfile %s on %s failed (%s)",
-                        loc->path, subvol->name, strerror (op_errno));
-                local->op_ret = op_ret;
-                local->op_errno = op_errno;
-                goto out;
-        }
-        gf_log (this->name, GF_LOG_DEBUG, "unlink successfull. Proceeding with"
-                " creation of link file %s", loc->path);
-        dht_linkfile_create (frame, local->linkfile.linkfile_cbk,
-                             src_cached, dst_hashed, loc);
+        DHT_STACK_DESTROY (frame);
 
-        return 0;
-out:
-        local->linkfile.linkfile_cbk (frame, NULL, frame->this, -1, EINVAL,
-                                      local->loc.inode, NULL, NULL, NULL);
-err:
-        return -1;
+       return 0;
 }
 
 int
-dht_linkfile_recreate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno,
-                inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                struct iatt *postparent)
+dht_linkfile_attr_heal (call_frame_t *frame, xlator_t *this)
 {
-        dht_local_t  *local         = NULL;
-        xlator_t     *src_cached    = NULL;
-        xlator_t     *dst_hashed    = NULL;
-        loc_t        *loc           = NULL;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, err);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, err);
+        int     ret = -1;
+        call_frame_t *copy = NULL;
+        dht_local_t  *local = NULL;
+        dht_local_t  *copy_local = NULL;
+        xlator_t     *subvol = NULL;
+        struct iatt   stbuf = {0,};
 
         local = frame->local;
-        loc = &local->loc;
 
-        src_cached = local->src_cached;
-        dst_hashed = local->dst_hashed;
+        GF_VALIDATE_OR_GOTO ("dht", local, out);
+        GF_VALIDATE_OR_GOTO ("dht", local->link_subvol, out);
 
-        if (!src_cached || !dst_hashed) {
-                gf_log (this->name, GF_LOG_ERROR, "src_cached or dst_hashed"
-                        " or hashed_subvol is null");
-                local->op_ret = -1;
-                local->op_errno = EINVAL;
-                goto out;
-        }
+        if (local->stbuf.ia_type == IA_INVAL)
+                return 0;
 
-        /* if link_file exists, remove it, else create it*/
-        if (!op_ret) {
-               if (!check_is_linkfile (inode, stbuf, xattr))
-                       goto out;
-               gf_log (this->name, GF_LOG_DEBUG, "link file exists."
-                        " Calling unlink on %s", loc->path);
-                STACK_WIND (frame, dht_recreate_linkfile_unlink_cbk,
-                            dst_hashed, dst_hashed->fops->unlink,
-                            loc);
-
-        } else if (op_errno == ENOENT) {
-                gf_log (this->name, GF_LOG_DEBUG, "link file does not exist."
-                        " Proceeding to creation of linkfile %s", loc->path);
-                dht_linkfile_create (frame, local->linkfile.linkfile_cbk,
-                                     src_cached, dst_hashed, loc);
-        } else {
-                gf_log (this->name, GF_LOG_ERROR, "returned error %s",
-                        strerror(op_errno));
-                local->op_ret = op_ret;
-                local->op_errno = op_errno;
-                goto out;
-        }
+        uuid_copy (local->loc.gfid, local->stbuf.ia_gfid);
 
-        return 0;
-out:
-        local->linkfile.linkfile_cbk (frame, NULL, frame->this, -1, EINVAL,
-                                      loc->inode, NULL, NULL, NULL);
-err:
-        return -1;
-}
+        copy = copy_frame (frame);
 
-int
-dht_linkfile_recreate (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
-                       xlator_t *tovol, xlator_t *fromvol, loc_t *loc)
-{
-        dht_local_t     *local          = NULL;
+        if (!copy)
+                goto out;
 
-        GF_VALIDATE_OR_GOTO ("dht", frame, out);
-        GF_VALIDATE_OR_GOTO ("dht", loc, out);
-        GF_VALIDATE_OR_GOTO ("dht", tovol, out);
-        GF_VALIDATE_OR_GOTO ("dht", fromvol, out);
+       copy_local = dht_local_init (copy, &local->loc, NULL, 0);
 
-        local = frame->local;
-        local->linkfile.linkfile_cbk = linkfile_cbk;
-        local->linkfile.srcvol = tovol;
-        loc_copy (&local->linkfile.loc, loc);
+       if (!copy_local)
+               goto out;
 
-        STACK_WIND (frame, dht_linkfile_recreate_cbk,
-                    fromvol, fromvol->fops->lookup, loc, NULL);
+        stbuf = local->stbuf;
+        subvol = local->link_subvol;
 
-        return 0;
+        copy->local = copy_local;
 
-out:
-        linkfile_cbk (frame, NULL, frame->this, -1, EINVAL, loc->inode, NULL,
-                      NULL, NULL);
+        FRAME_SU_DO (copy, dht_local_t);
 
-        return -1;
+        STACK_WIND (copy, dht_linkfile_setattr_cbk, subvol,
+                    subvol->fops->setattr, &copy_local->loc,
+                    &stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID), NULL);
+        ret = 0;
+out:
+        return ret;
 }
diff --git a/xlators/cluster/dht/src/dht-mem-types.h b/xlators/cluster/dht/src/dht-mem-types.h
index 9dc2648a2..e893eb48f 100644
--- a/xlators/cluster/dht/src/dht-mem-types.h
+++ b/xlators/cluster/dht/src/dht-mem-types.h
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -28,7 +19,6 @@ enum gf_dht_mem_types_ {
         gf_dht_mt_dht_conf_t,
         gf_dht_mt_char,
         gf_dht_mt_int32_t,
-        gf_dht_mt_dht_local_t,
         gf_dht_mt_xlator_t,
         gf_dht_mt_dht_layout_t,
         gf_switch_mt_dht_conf_t,
@@ -37,6 +27,9 @@ enum gf_dht_mem_types_ {
         gf_switch_mt_switch_struct,
         gf_dht_mt_subvol_time,
         gf_dht_mt_loc_t,
+        gf_defrag_info_mt,
+        gf_dht_mt_inode_ctx_t,
+        gf_dht_mt_ctx_stat_time_t,
         gf_dht_mt_end
 };
 #endif
diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c
index 6dea40f0b..4f78f5203 100644
--- a/xlators/cluster/dht/src/dht-rebalance.c
+++ b/xlators/cluster/dht/src/dht-rebalance.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2011 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -24,11 +15,13 @@
 #endif
 
 #include "dht-common.h"
+#include "xlator.h"
+#include <signal.h>
+#include <fnmatch.h>
 
 #define GF_DISK_SECTOR_SIZE             512
 #define DHT_REBALANCE_PID               4242 /* Change it if required */
 #define DHT_REBALANCE_BLKSIZE           (128 * 1024)
-#define DHT_MIGRATE_EVEN_IF_LINK_EXISTS 1
 
 static int
 dht_write_with_holes (xlator_t *to, fd_t *fd, struct iovec *vec, int count,
@@ -60,9 +53,15 @@ dht_write_with_holes (xlator_t *to, fd_t *fd, struct iovec *vec, int count,
                                 ret = syncop_write (to, fd, (buf + tmp_offset),
                                                     (start_idx - tmp_offset),
                                                     (offset + tmp_offset),
-                                                    iobref);
-                                if (ret < 0)
+                                                    iobref, 0);
+                                /* 'path' will be logged in calling function */
+                                if (ret < 0) {
+                                        gf_log (THIS->name, GF_LOG_WARNING,
+                                                "failed to write (%s)",
+                                                strerror (-ret));
+                                        ret = -1;
                                         goto out;
+                                }
 
                                 write_needed = 0;
                         }
@@ -73,9 +72,15 @@ dht_write_with_holes (xlator_t *to, fd_t *fd, struct iovec *vec, int count,
                         /* This means, last chunk is not yet written.. write it */
                         ret = syncop_write (to, fd, (buf + tmp_offset),
                                             (buf_len - tmp_offset),
-                                            (offset + tmp_offset), iobref);
-                        if (ret < 0)
+                                            (offset + tmp_offset), iobref, 0);
+                        if (ret < 0) {
+                                /* 'path' will be logged in calling function */
+                                gf_log (THIS->name, GF_LOG_WARNING,
+                                        "failed to write (%s)",
+                                        strerror (-ret));
+                                ret = -1;
                                 goto out;
+                        }
                 }
 
                 size_pending = (size - buf_len);
@@ -83,52 +88,207 @@ dht_write_with_holes (xlator_t *to, fd_t *fd, struct iovec *vec, int count,
                         break;
         }
 
-        /* do it regardless of all the above cases as we had to 'write' the
-           given number of bytes */
-        ret = syncop_ftruncate (to, fd, offset + size);
+        ret = size;
+out:
+        return ret;
+
+}
+
+/*
+   return values:
+   -1 : failure
+   -2 : success
+
+Hard link migration is carried out in three stages.
+
+(Say there are n hardlinks)
+Stage 1: Setting the new hashed subvol information on the 1st hardlink
+         encountered (linkto setxattr)
+
+Stage 2: Creating hardlinks on new hashed subvol for the 2nd to (n-1)th
+         hardlink
+
+Stage 3: Physical migration of the data file for nth hardlink
+
+Why to deem "-2" as success and not "0":
+
+   dht_migrate_file expects return value "0" from _is_file_migratable if
+the file has to be migrated.
+
+   _is_file_migratable returns zero only when it is called with the
+flag "GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS".
+
+   gf_defrag_handle_hardlink calls dht_migrate_file for physical migration
+of the data file with the flag "GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS"
+
+Hence, gf_defrag_handle_hardlink returning "0" for success will force
+"dht_migrate_file" to migrate each of the hardlink which is not intended.
+
+For each of the three stage mentioned above "-2" will be returned and will
+be converted to "0" in dht_migrate_file.
+
+*/
+
+int32_t
+gf_defrag_handle_hardlink (xlator_t *this, loc_t *loc, dict_t  *xattrs,
+                           struct iatt *stbuf)
+{
+        int32_t                 ret             = -1;
+        xlator_t               *cached_subvol   = NULL;
+        xlator_t               *hashed_subvol   = NULL;
+        xlator_t               *linkto_subvol   = NULL;
+        data_t                 *data            = NULL;
+        struct iatt             iatt            = {0,};
+        int32_t                 op_errno        = 0;
+        dht_conf_t             *conf            = NULL;
+
+        GF_VALIDATE_OR_GOTO ("defrag", loc, out);
+        GF_VALIDATE_OR_GOTO ("defrag", loc->name, out);
+        GF_VALIDATE_OR_GOTO ("defrag", stbuf, out);
+        GF_VALIDATE_OR_GOTO ("defrag", this, out);
+        GF_VALIDATE_OR_GOTO ("defrag", xattrs, out);
+        GF_VALIDATE_OR_GOTO ("defrag", this->private, out);
+
+        conf = this->private;
+
+        if (uuid_is_null (loc->pargfid)) {
+                gf_log ("", GF_LOG_ERROR, "loc->pargfid is NULL for "
+                        "%s", loc->path);
+                goto out;
+        }
+
+        if (uuid_is_null (loc->gfid)) {
+                gf_log ("", GF_LOG_ERROR, "loc->gfid is NULL for "
+                        "%s", loc->path);
+                goto out;
+        }
+
+        cached_subvol = dht_subvol_get_cached (this, loc->inode);
+        if (!cached_subvol) {
+                gf_log (this->name, GF_LOG_ERROR, "Failed to get cached subvol"
+                        " for %s on %s", loc->name, this->name);
+                goto out;
+        }
+
+        hashed_subvol = dht_subvol_get_hashed (this, loc);
+        if (!hashed_subvol) {
+                gf_log (this->name, GF_LOG_ERROR, "Failed to get hashed subvol"
+                        " for %s on %s", loc->name, this->name);
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_INFO, "Attempting to migrate hardlink %s "
+                "with gfid %s from %s -> %s", loc->name, uuid_utoa (loc->gfid),
+                cached_subvol->name, hashed_subvol->name);
+        data = dict_get (xattrs, conf->link_xattr_name);
+        /* set linkto on cached -> hashed if not present, else link it */
+        if (!data) {
+                ret = dict_set_str (xattrs, conf->link_xattr_name,
+                                    hashed_subvol->name);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to set "
+                                "linkto xattr in dict for %s", loc->name);
+                        goto out;
+                }
+
+                ret = syncop_setxattr (cached_subvol, loc, xattrs, 0);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR, "Linkto setxattr "
+                                "failed %s -> %s (%s)", cached_subvol->name,
+                                loc->name, strerror (-ret));
+                        ret = -1;
+                        goto out;
+                }
+                ret = -2;
+                goto out;
+        } else {
+                linkto_subvol = dht_linkfile_subvol (this, NULL, NULL, xattrs);
+                if (!linkto_subvol) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to get "
+                                "linkto subvol for %s", loc->name);
+                } else {
+                        hashed_subvol = linkto_subvol;
+                }
+
+                ret = syncop_link (hashed_subvol, loc, loc);
+                if  (ret) {
+                        op_errno = -ret;
+                        ret = -1;
+                        gf_log (this->name, GF_LOG_ERROR, "link of %s -> %s"
+                                " failed on  subvol %s (%s)", loc->name,
+                                uuid_utoa(loc->gfid),
+                                hashed_subvol->name, strerror (op_errno));
+                        if (op_errno != EEXIST)
+                                goto out;
+                }
+        }
+        ret = syncop_lookup (hashed_subvol, loc, NULL, &iatt, NULL, NULL);
         if (ret) {
-                gf_log (THIS->name, GF_LOG_WARNING,
-                        "failed to perform truncate on %s", to->name);
+                gf_log (this->name, GF_LOG_ERROR, "Failed lookup %s on %s (%s)"
+                        , loc->name, hashed_subvol->name, strerror (-ret));
+                ret = -1;
                 goto out;
         }
 
-        ret = size;
+        if (iatt.ia_nlink == stbuf->ia_nlink) {
+                ret = dht_migrate_file (this, loc, cached_subvol, hashed_subvol,
+                                        GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS);
+                if (ret)
+                        goto out;
+        }
+        ret = -2;
 out:
         return ret;
-
 }
 
+/*
+     return values
+     0 : File will be migrated
+    -2 : File will not be migrated
+         (This is the return value from gf_defrag_handle_hardlink. Checkout
+         gf_defrag_handle_hardlink for description of "returning -2")
+    -1 : failure
+*/
 static inline int
-__is_file_migratable (xlator_t *this, loc_t *loc, dict_t *rsp_dict,
-                      struct iatt *stbuf)
+__is_file_migratable (xlator_t *this, loc_t *loc,
+                      struct iatt *stbuf, dict_t *xattrs, int flags)
 {
-        int ret           = -1;
-        int open_fd_count = 0;
+        int ret = -1;
 
-        if (!IA_ISREG (stbuf->ia_type)) {
+        if (IA_ISDIR (stbuf->ia_type)) {
                 gf_log (this->name, GF_LOG_WARNING,
-                        "%s: migrate-file called on non-regular entry (0%o)",
-                        loc->path, stbuf->ia_type);
+                        "%s: migrate-file called on directory", loc->path);
                 ret = -1;
                 goto out;
         }
 
-        if (stbuf->ia_nlink > 1) {
-                /* TODO : support migrating hardlinks */
-                gf_log (this->name, GF_LOG_WARNING, "%s: file has hardlinks",
-                        loc->path);
-                ret = -ENOTSUP;
+        if (flags == GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS) {
+                ret = 0;
                 goto out;
         }
-
-        ret = dict_get_int32 (rsp_dict, GLUSTERFS_OPEN_FD_COUNT, &open_fd_count);
-        if (!ret && (open_fd_count > 0)) {
-                /* TODO: support migration of files with open fds */
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: file has open fds, not attempting migration",
-                        loc->path);
+        if (stbuf->ia_nlink > 1) {
+                /* support for decomission */
+                if (flags == GF_DHT_MIGRATE_HARDLINK) {
+                        ret = gf_defrag_handle_hardlink (this, loc,
+                                                         xattrs, stbuf);
+
+                   /*
+                   Returning zero will force the file to be remigrated.
+                   Checkout gf_defrag_handle_hardlink for more information.
+                   */
+                        if (ret && ret != -2) {
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "%s: failed to migrate file with link",
+                                        loc->path);
+                        }
+                } else {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: file has hardlinks", loc->path);
+                        ret = -ENOTSUP;
+                }
                 goto out;
         }
+
         ret = 0;
 
 out:
@@ -136,16 +296,17 @@ out:
 }
 
 static inline int
-__dht_rebalance_create_dst_file (xlator_t *to, loc_t *loc, struct iatt *stbuf,
-                                 dict_t *dict, fd_t **dst_fd, int *need_unlink)
+__dht_rebalance_create_dst_file (xlator_t *to, xlator_t *from, loc_t *loc, struct iatt *stbuf,
+                                 dict_t *dict, fd_t **dst_fd, dict_t *xattr)
 {
-        xlator_t *this = NULL;
-        int       ret  = -1;
-        mode_t    mode = 0;
-        fd_t     *fd   = NULL;
-        struct iatt new_stbuf = {0,};
+        xlator_t    *this = NULL;
+        int          ret  = -1;
+        fd_t        *fd   = NULL;
+        struct iatt  new_stbuf = {0,};
+        dht_conf_t  *conf = NULL;
 
         this = THIS;
+        conf = this->private;
 
         ret = dict_set_static_bin (dict, "gfid-req", stbuf->ia_gfid, 16);
         if (ret) {
@@ -154,49 +315,74 @@ __dht_rebalance_create_dst_file (xlator_t *to, loc_t *loc, struct iatt *stbuf,
                 goto out;
         }
 
+        ret = dict_set_str (dict, conf->link_xattr_name, from->name);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to set gfid in dict for create", loc->path);
+                goto out;
+        }
+
         fd = fd_create (loc->inode, DHT_REBALANCE_PID);
         if (!fd) {
                 gf_log (this->name, GF_LOG_ERROR,
-                        "%s: fd create failed (destination)", loc->path);
+                        "%s: fd create failed (destination) (%s)",
+                        loc->path, strerror (errno));
                 ret = -1;
                 goto out;
         }
 
         ret = syncop_lookup (to, loc, NULL, &new_stbuf, NULL, NULL);
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG, "failed to lookup %s on %s",
-                        loc->path, to->name);
-
-                mode = st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type);
-                ret = syncop_create (to, loc, O_WRONLY, mode, fd, dict);
-                if (ret < 0) {
+        if (!ret) {
+                /* File exits in the destination, check if gfid matches */
+                if (uuid_compare (stbuf->ia_gfid, new_stbuf.ia_gfid) != 0) {
                         gf_log (this->name, GF_LOG_ERROR,
-                                "failed to create %s on %s", loc->path, to->name);
+                                "file %s exits in %s with different gfid",
+                                loc->path, to->name);
+                        fd_unref (fd);
                         goto out;
                 }
-
-                *need_unlink = 1;
-                goto done;
         }
-
-        /* File exits in the destination, just do the open if gfid matches */
-        if (uuid_compare (stbuf->ia_gfid, new_stbuf.ia_gfid) != 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "file %s exits in %s with different gfid",
-                        loc->path, to->name);
-                fd_unref (fd);
+        if ((ret < 0) && (-ret != ENOENT)) {
+                /* File exists in destination, but not accessible */
+                gf_log (THIS->name, GF_LOG_WARNING,
+                        "%s: failed to lookup file (%s)",
+                        loc->path, strerror (-ret));
+                ret = -1;
                 goto out;
         }
 
-        ret = syncop_open (to, loc, O_WRONLY, fd);
+        /* Create the destination with LINKFILE mode, and linkto xattr,
+           if the linkfile already exists, it will just open the file */
+        ret = syncop_create (to, loc, O_RDWR, DHT_LINKFILE_MODE, fd,
+                             dict, &new_stbuf);
         if (ret < 0) {
                 gf_log (this->name, GF_LOG_ERROR,
-                        "failed to open file %s on %s",
-                        loc->path, to->name);
-                fd_unref (fd);
+                        "failed to create %s on %s (%s)",
+                        loc->path, to->name, strerror (-ret));
+                ret = -1;
                 goto out;
         }
-done:
+
+        ret = syncop_fsetxattr (to, fd, xattr, 0);
+        if (ret < 0)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to set xattr on %s (%s)",
+                        loc->path, to->name, strerror (-ret));
+
+        ret = syncop_ftruncate (to, fd, stbuf->ia_size);
+        if (ret < 0)
+                gf_log (this->name, GF_LOG_ERROR,
+                        "ftruncate failed for %s on %s (%s)",
+                        loc->path, to->name, strerror (-ret));
+
+        ret = syncop_fsetattr (to, fd, stbuf,
+                               (GF_SET_ATTR_UID | GF_SET_ATTR_GID),
+                                NULL, NULL);
+        if (ret < 0)
+                gf_log (this->name, GF_LOG_ERROR,
+                        "chown failed for %s on %s (%s)",
+                        loc->path, to->name, strerror (-ret));
+
         if (dst_fd)
                 *dst_fd = fd;
 
@@ -209,41 +395,77 @@ out:
 
 static inline int
 __dht_check_free_space (xlator_t *to, xlator_t *from, loc_t *loc,
-                        struct iatt *stbuf)
+                        struct iatt *stbuf, int flag)
 {
         struct statvfs  src_statfs = {0,};
         struct statvfs  dst_statfs = {0,};
         int             ret        = -1;
         xlator_t       *this       = NULL;
 
+        uint64_t        src_statfs_blocks = 1;
+        uint64_t        dst_statfs_blocks = 1;
+
         this = THIS;
 
         ret = syncop_statfs (from, loc, &src_statfs);
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR,
-                        "failed to get statfs of %s on %s",
-                        loc->path, from->name);
+                        "failed to get statfs of %s on %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
                 goto out;
         }
 
         ret = syncop_statfs (to, loc, &dst_statfs);
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR,
-                        "failed to get statfs of %s on %s",
-                        loc->path, to->name);
+                        "failed to get statfs of %s on %s (%s)",
+                        loc->path, to->name, strerror (-ret));
+                ret = -1;
                 goto out;
         }
-        if (((dst_statfs.f_bavail *
-              dst_statfs.f_bsize) / GF_DISK_SECTOR_SIZE) >
-            (((src_statfs.f_bavail * src_statfs.f_bsize) /
-              GF_DISK_SECTOR_SIZE) - stbuf->ia_blocks)) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "data movement attempted from node (%s) with"
-                        " higher disk space to a node (%s) with "
-                        "lesser disk space (%s)", from->name,
-                        to->name, loc->path);
 
-                ret = -1;
+        /* if force option is given, do not check for space @ dst.
+         * Check only if space is avail for the file */
+        if (flag != GF_DHT_MIGRATE_DATA)
+                goto check_avail_space;
+
+        /* Check:
+           During rebalance `migrate-data` - Destination subvol experiences
+           a `reduction` in 'blocks' of free space, at the same time source
+           subvol gains certain 'blocks' of free space. A valid check is
+           necessary here to avoid errorneous move to destination where
+           the space could be scantily available.
+         */
+        if (stbuf) {
+                dst_statfs_blocks = ((dst_statfs.f_bavail *
+                                      dst_statfs.f_bsize) /
+                                     GF_DISK_SECTOR_SIZE);
+                src_statfs_blocks = ((src_statfs.f_bavail *
+                                      src_statfs.f_bsize) /
+                                     GF_DISK_SECTOR_SIZE);
+                if ((dst_statfs_blocks - stbuf->ia_blocks) <
+                    (src_statfs_blocks + stbuf->ia_blocks)) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "data movement attempted from node (%s) with"
+                                " higher disk space to a node (%s) with "
+                                "lesser disk space (%s)", from->name,
+                                to->name, loc->path);
+
+                        /* this is not a 'failure', but we don't want to
+                           consider this as 'success' too :-/ */
+                        ret = 1;
+                        goto out;
+                }
+        }
+check_avail_space:
+        if (((dst_statfs.f_bavail * dst_statfs.f_bsize) /
+              GF_DISK_SECTOR_SIZE) < stbuf->ia_blocks) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "data movement attempted from node (%s) with "
+                        "to node (%s) which does not have required free space"
+                        " for %s", from->name, to->name, loc->path);
+                ret = 1;
                 goto out;
         }
 
@@ -253,18 +475,23 @@ out:
 }
 
 static inline int
-__dht_rebalane_migrate_data (xlator_t *from, xlator_t *to, fd_t *src, fd_t *dst,
-                             int hole_exists)
+__dht_rebalance_migrate_data (xlator_t *from, xlator_t *to, fd_t *src, fd_t *dst,
+                             uint64_t ia_size, int hole_exists)
 {
-        int            ret    = -1;
+        int            ret    = 0;
         int            count  = 0;
         off_t          offset = 0;
         struct iovec  *vector = NULL;
         struct iobref *iobref = NULL;
-
-        while (1) {
-                ret = syncop_readv (from, src, DHT_REBALANCE_BLKSIZE,
-                                    offset, &vector, &count, &iobref);
+        uint64_t       total  = 0;
+        size_t         read_size = 0;
+
+        /* if file size is '0', no need to enter this loop */
+        while (total < ia_size) {
+                read_size = (((ia_size - total) > DHT_REBALANCE_BLKSIZE) ?
+                             DHT_REBALANCE_BLKSIZE : (ia_size - total));
+                ret = syncop_readv (from, src, read_size,
+                                    offset, 0, &vector, &count, &iobref);
                 if (!ret || (ret < 0)) {
                         break;
                 }
@@ -274,14 +501,14 @@ __dht_rebalane_migrate_data (xlator_t *from, xlator_t *to, fd_t *src, fd_t *dst,
                                                     ret, offset, iobref);
                 else
                         ret = syncop_writev (to, dst, vector, count,
-                                             offset, iobref);
+                                             offset, iobref, 0);
                 if (ret < 0) {
                         break;
                 }
                 offset += ret;
+                total += ret;
 
-                if (vector)
-                        GF_FREE (vector);
+                GF_FREE (vector);
                 if (iobref)
                         iobref_unref (iobref);
                 iobref = NULL;
@@ -289,15 +516,233 @@ __dht_rebalane_migrate_data (xlator_t *from, xlator_t *to, fd_t *src, fd_t *dst,
         }
         if (iobref)
                 iobref_unref (iobref);
-        if (vector)
-                GF_FREE (vector);
+        GF_FREE (vector);
 
         if (ret >= 0)
                 ret = 0;
+        else
+                ret = -1;
+
+        return ret;
+}
+
+
+static inline int
+__dht_rebalance_open_src_file (xlator_t *from, xlator_t *to, loc_t *loc,
+                               struct iatt *stbuf, fd_t **src_fd)
+{
+        int          ret  = 0;
+        fd_t        *fd   = NULL;
+        dict_t      *dict = NULL;
+        xlator_t    *this = NULL;
+        struct iatt  iatt = {0,};
+        dht_conf_t  *conf = NULL;
+
+        this = THIS;
+        conf = this->private;
+
+        fd = fd_create (loc->inode, DHT_REBALANCE_PID);
+        if (!fd) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: fd create failed (source)", loc->path);
+                ret = -1;
+                goto out;
+        }
+
+        ret = syncop_open (from, loc, O_RDWR, fd);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to open file %s on %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
+                goto out;
+        }
+
+        ret = -1;
+        dict = dict_new ();
+        if (!dict)
+                goto out;
+
+        ret = dict_set_str (dict, conf->link_xattr_name, to->name);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to set xattr in dict for %s (linkto:%s)",
+                        loc->path, to->name);
+                goto out;
+        }
+
+        /* Once the migration starts, the source should have 'linkto' key set
+           to show which is the target, so other clients can work around it */
+        ret = syncop_setxattr (from, loc, dict, 0);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to set xattr on %s in %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
+                goto out;
+        }
+
+        /* mode should be (+S+T) to indicate migration is in progress */
+        iatt.ia_prot = stbuf->ia_prot;
+        iatt.ia_type = stbuf->ia_type;
+        iatt.ia_prot.sticky = 1;
+        iatt.ia_prot.sgid = 1;
+
+        ret = syncop_setattr (from, loc, &iatt, GF_SET_ATTR_MODE, NULL, NULL);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to set mode on %s in %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
+                goto out;
+        }
+
+        if (src_fd)
+                *src_fd = fd;
+
+        /* success */
+        ret = 0;
+out:
+        if (dict)
+                dict_unref (dict);
+
+        return ret;
+}
+
+int
+migrate_special_files (xlator_t *this, xlator_t *from, xlator_t *to, loc_t *loc,
+                       struct iatt *buf)
+{
+        int          ret      = -1;
+        dict_t      *rsp_dict = NULL;
+        dict_t      *dict     = NULL;
+        char        *link     = NULL;
+        struct iatt  stbuf    = {0,};
+        dht_conf_t  *conf     = this->private;
+
+        dict = dict_new ();
+        if (!dict)
+                goto out;
+
+        ret = dict_set_int32 (dict, conf->link_xattr_name, 256);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to set 'linkto' key in dict", loc->path);
+                goto out;
+        }
+
+        /* check in the destination if the file is link file */
+        ret = syncop_lookup (to, loc, dict, &stbuf, &rsp_dict, NULL);
+        if ((ret < 0) && (-ret != ENOENT)) {
+                gf_log (this->name, GF_LOG_WARNING, "%s: lookup failed (%s)",
+                        loc->path, strerror (-ret));
+                ret = -1;
+                goto out;
+        }
+
+        /* we no more require this key */
+        dict_del (dict, conf->link_xattr_name);
+
+        /* file exists in target node, only if it is 'linkfile' its valid,
+           otherwise, error out */
+        if (!ret) {
+                if (!check_is_linkfile (loc->inode, &stbuf, rsp_dict,
+                                        conf->link_xattr_name)) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: file exists in destination", loc->path);
+                        ret = -1;
+                        goto out;
+                }
+
+                /* as file is linkfile, delete it */
+                ret = syncop_unlink (to, loc);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: failed to delete the linkfile (%s)",
+                                loc->path, strerror (-ret));
+                        ret = -1;
+                        goto out;
+                }
+        }
+
+        /* Set the gfid of the source file in dict */
+        ret = dict_set_static_bin (dict, "gfid-req", buf->ia_gfid, 16);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to set gfid in dict for create", loc->path);
+                goto out;
+        }
+
+        /* Create the file in target */
+        if (IA_ISLNK (buf->ia_type)) {
+                /* Handle symlinks separately */
+                ret = syncop_readlink (from, loc, &link, buf->ia_size);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: readlink on symlink failed (%s)",
+                                loc->path, strerror (-ret));
+                        ret = -1;
+                        goto out;
+                }
+
+                ret = syncop_symlink (to, loc, link, dict, 0);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: creating symlink failed (%s)",
+                                loc->path, strerror (-ret));
+                        ret = -1;
+                        goto out;
+                }
+
+                goto done;
+        }
+
+        ret = syncop_mknod (to, loc, st_mode_from_ia (buf->ia_prot,
+                                                      buf->ia_type),
+                            makedev (ia_major (buf->ia_rdev),
+                                     ia_minor (buf->ia_rdev)), dict, 0);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING, "%s: mknod failed (%s)",
+                        loc->path, strerror (-ret));
+                ret = -1;
+                goto out;
+        }
+
+done:
+        ret = syncop_setattr (to, loc, buf,
+                              (GF_SET_ATTR_UID | GF_SET_ATTR_GID |
+                               GF_SET_ATTR_MODE), NULL, NULL);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to perform setattr on %s (%s)",
+                        loc->path, to->name, strerror (-ret));
+                ret = -1;
+        }
+
+        ret = syncop_unlink (from, loc);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING, "%s: unlink failed (%s)",
+                        loc->path, strerror (-ret));
+                ret = -1;
+        }
+
+out:
+        if (dict)
+                dict_unref (dict);
+
+        if (rsp_dict)
+                dict_unref (rsp_dict);
 
         return ret;
 }
 
+/*
+  return values:
+
+   -1 : failure
+    0 : successfully migrated data
+    1 : not a failure, but we can't migrate data as of now
+*/
 int
 dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
                   int flag)
@@ -305,13 +750,15 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
         int             ret            = -1;
         struct iatt     new_stbuf      = {0,};
         struct iatt     stbuf          = {0,};
+        struct iatt     empty_iatt     = {0,};
+        ia_prot_t       src_ia_prot    = {0,};
         fd_t           *src_fd         = NULL;
         fd_t           *dst_fd         = NULL;
         dict_t         *dict           = NULL;
         dict_t         *xattr          = NULL;
-        dict_t         *rsp_dict       = NULL;
+        dict_t         *xattr_rsp      = NULL;
         int             file_has_holes = 0;
-        int             need_unlink    = 0;
+        dht_conf_t     *conf           = this->private;
 
         gf_log (this->name, GF_LOG_INFO, "%s: attempting to move from %s to %s",
                 loc->path, from->name, to->name);
@@ -320,112 +767,218 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
         if (!dict)
                 goto out;
 
-        ret = dict_set_int32 (dict, GLUSTERFS_OPEN_FD_COUNT, 4);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to set fd-count key in dict, may attempt "
-                        "migration of file which has open fds", loc->path);
+        ret = dict_set_int32 (dict, conf->link_xattr_name, 256);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to set 'linkto' key in dict", loc->path);
+                goto out;
+        }
 
-        ret = syncop_lookup (from, loc, dict, &stbuf, &rsp_dict, NULL);
+        /* Phase 1 - Data migration is in progress from now on */
+        ret = syncop_lookup (from, loc, dict, &stbuf, &xattr_rsp, NULL);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "failed to lookup %s on %s",
-                        loc->path, from->name);
+                gf_log (this->name, GF_LOG_ERROR, "%s: lookup failed on %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
                 goto out;
         }
 
+        /* we no more require this key */
+        dict_del (dict, conf->link_xattr_name);
+
+        /* preserve source mode, so set the same to the destination */
+        src_ia_prot = stbuf.ia_prot;
+
         /* Check if file can be migrated */
-        ret = __is_file_migratable (this, loc, rsp_dict, &stbuf);
-        if (ret)
+        ret = __is_file_migratable (this, loc, &stbuf, xattr_rsp, flag);
+        if (ret) {
+                if (ret == -2)
+                        ret = 0;
                 goto out;
-
-        /* create the destination */
-        ret = __dht_rebalance_create_dst_file (to, loc, &stbuf, dict, &dst_fd,
-                                               &need_unlink);
-        if (ret)
+        }
+        /* Take care of the special files */
+        if (!IA_ISREG (stbuf.ia_type)) {
+                /* Special files */
+                ret = migrate_special_files (this, from, to, loc, &stbuf);
                 goto out;
+        }
 
-        /* Should happen on all files when 'force' option is not given */
-        if (flag != DHT_MIGRATE_EVEN_IF_LINK_EXISTS) {
-                ret = __dht_check_free_space (to, from, loc, &stbuf);
-                if (ret)
-                        goto out;
+        /* TODO: move all xattr related operations to fd based operations */
+        ret = syncop_listxattr (from, loc, &xattr);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to get xattr from %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
         }
 
-        /* Try to preserve 'holes' while migrating data */
-        if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE))
-                file_has_holes = 1;
+        /* create the destination, with required modes/xattr */
+        ret = __dht_rebalance_create_dst_file (to, from, loc, &stbuf,
+                                               dict, &dst_fd, xattr);
+        if (ret)
+                goto out;
 
-        src_fd = fd_create (loc->inode, DHT_REBALANCE_PID);
-        if (!src_fd) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: fd create failed (source)", loc->path);
-                ret = -1;
+        ret = __dht_check_free_space (to, from, loc, &stbuf, flag);
+        if (ret) {
                 goto out;
         }
 
-        ret = syncop_open (from, loc, O_RDONLY, src_fd);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to open file %s on %s",
+        /* Open the source, and also update mode/xattr */
+        ret = __dht_rebalance_open_src_file (from, to, loc, &stbuf, &src_fd);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "failed to open %s on %s",
                         loc->path, from->name);
                 goto out;
         }
 
+
+        ret = syncop_fstat (from, src_fd, &stbuf);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "failed to lookup %s on %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
+                goto out;
+        }
+
+        /* Try to preserve 'holes' while migrating data */
+        if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE))
+                file_has_holes = 1;
+
         /* All I/O happens in this function */
-        ret = __dht_rebalane_migrate_data (from, to, src_fd, dst_fd,
-                                           file_has_holes);
+        ret = __dht_rebalance_migrate_data (from, to, src_fd, dst_fd,
+					    stbuf.ia_size, file_has_holes);
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR, "%s: failed to migrate data",
                         loc->path);
+                /* reset the destination back to 0 */
+                ret = syncop_ftruncate (to, dst_fd, 0);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: failed to reset target size back to 0 (%s)",
+                                loc->path, strerror (-ret));
+                }
+
+                ret = -1;
                 goto out;
         }
 
-        ret = syncop_lookup (from, loc, NULL, &new_stbuf, NULL, NULL);
+        /* TODO: Sync the locks */
+
+        ret = syncop_fsync (to, dst_fd, 0);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to fsync on %s (%s)",
+                        loc->path, to->name, strerror (-ret));
+                ret = -1;
+        }
+
+
+        /* Phase 2 - Data-Migration Complete, Housekeeping updates pending */
+
+        ret = syncop_fstat (from, src_fd, &new_stbuf);
         if (ret < 0) {
                 /* Failed to get the stat info */
                 gf_log (this->name, GF_LOG_ERROR,
-                        "failed to lookup file %s on %s",
-                        loc->path, from->name);
-                need_unlink = 0;
+                        "failed to fstat file %s on %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
                 goto out;
         }
 
-        /* No need to rebalance, if there is some
-           activity on source file */
-        if (new_stbuf.ia_mtime != stbuf.ia_mtime) {
+        /* source would have both sticky bit and sgid bit set, reset it to 0,
+           and set the source permission on destination, if it was not set
+           prior to setting rebalance-modes in source  */
+        if (!src_ia_prot.sticky)
+                new_stbuf.ia_prot.sticky = 0;
+
+        if (!src_ia_prot.sgid)
+                new_stbuf.ia_prot.sgid = 0;
+
+        /* TODO: if the source actually had sticky bit, or sgid bit set,
+           we are not handling it */
+
+        ret = syncop_fsetattr (to, dst_fd, &new_stbuf,
+                               (GF_SET_ATTR_UID | GF_SET_ATTR_GID |
+                                GF_SET_ATTR_MODE), NULL, NULL);
+        if (ret) {
                 gf_log (this->name, GF_LOG_WARNING,
-                        "%s: ignoring destination file as source has "
-                        "undergone some changes while migration was happening",
-                        loc->path);
+                        "%s: failed to perform setattr on %s (%s)",
+                        loc->path, to->name, strerror (-ret));
                 ret = -1;
                 goto out;
         }
 
+        /* Because 'futimes' is not portable */
         ret = syncop_setattr (to, loc, &new_stbuf,
-                              (GF_SET_ATTR_UID | GF_SET_ATTR_GID |
-                               GF_SET_ATTR_MODE | GF_SET_ATTR_ATIME |
-                               GF_SET_ATTR_MTIME), NULL, NULL);
+                              (GF_SET_ATTR_MTIME | GF_SET_ATTR_ATIME),
+                              NULL, NULL);
         if (ret) {
                 gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to perform setattr on %s",
-                        loc->path, to->name);
+                        "%s: failed to perform setattr on %s (%s)",
+                        loc->path, to->name, strerror (-ret));
+                ret = -1;
         }
 
-        ret = syncop_listxattr (from, loc, &xattr);
-        if (ret == -1)
+        /* Make the source as a linkfile first before deleting it */
+        empty_iatt.ia_prot.sticky = 1;
+        ret = syncop_fsetattr (from, src_fd, &empty_iatt,
+                               GF_SET_ATTR_MODE, NULL, NULL);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING,             \
+                        "%s: failed to perform setattr on %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
+                goto out;
+        }
+
+       /* Free up the data blocks on the source node, as the whole
+           file is migrated */
+        ret = syncop_ftruncate (from, src_fd, 0);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to perform truncate on %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
+        }
+
+        /* remove the 'linkto' xattr from the destination */
+        ret = syncop_fremovexattr (to, dst_fd, conf->link_xattr_name, 0);
+        if (ret) {
                 gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to get xattr from %s", loc->path, from->name);
+                        "%s: failed to perform removexattr on %s (%s)",
+                        loc->path, to->name, strerror (-ret));
+                ret = -1;
+        }
 
-        ret = syncop_setxattr (to, loc, xattr, 0);
-        if (ret == -1)
+        /* Do a stat and check the gfid before unlink */
+        ret = syncop_stat (from, loc, &empty_iatt);
+        if (ret) {
                 gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to set xattr on %s", loc->path, to->name);
+                        "%s: failed to do a stat on %s (%s)",
+                        loc->path, from->name, strerror (-ret));
+                ret = -1;
+                goto out;
+        }
 
-        /* rebalance complete */
-        syncop_close (dst_fd);
-        syncop_close (src_fd);
-        syncop_unlink (from, loc);
-        need_unlink = 0;
+        if (uuid_compare (empty_iatt.ia_gfid, loc->gfid) == 0) {
+                /* take out the source from namespace */
+                ret = syncop_unlink (from, loc);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: failed to perform unlink on %s (%s)",
+                                loc->path, from->name, strerror (-ret));
+                        ret = -1;
+                        goto out;
+                }
+        }
+
+        ret = syncop_lookup (this, loc, NULL, NULL, NULL, NULL);
+        if (ret) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "%s: failed to lookup the file on subvolumes (%s)",
+                        loc->path, strerror (-ret));
+                ret = -1;
+        }
 
         gf_log (this->name, GF_LOG_INFO,
                 "completed migration of %s from subvolume %s to %s",
@@ -436,14 +989,15 @@ out:
         if (dict)
                 dict_unref (dict);
 
-        if (ret) {
-                if (dst_fd)
-                        syncop_close (dst_fd);
-                if (src_fd)
-                        syncop_close (src_fd);
-                if (need_unlink)
-                        syncop_unlink (to, loc);
-        }
+        if (xattr)
+                dict_unref (xattr);
+        if (xattr_rsp)
+                dict_unref (xattr_rsp);
+
+        if (dst_fd)
+                syncop_close (dst_fd);
+        if (src_fd)
+                syncop_close (src_fd);
 
         return ret;
 }
@@ -452,19 +1006,17 @@ static int
 rebalance_task (void *data)
 {
         int           ret   = -1;
-        xlator_t     *this  = NULL;
         dht_local_t  *local = NULL;
         call_frame_t *frame = NULL;
 
         frame = data;
-        this = THIS;
 
         local = frame->local;
 
         /* This function is 'synchrounous', hence if it returns,
            we are done with the task */
-        ret = dht_migrate_file (THIS, &local->loc, local->from_subvol,
-                                local->to_subvol, local->flags);
+        ret = dht_migrate_file (THIS, &local->loc, local->rebalance.from_subvol,
+                                local->rebalance.target_node, local->flags);
 
         return ret;
 }
@@ -477,6 +1029,7 @@ rebalance_task_completion (int op_ret, call_frame_t *sync_frame, void *data)
         dht_layout_t *layout     = 0;
         xlator_t     *this       = NULL;
         dht_local_t  *local      = NULL;
+        int32_t       op_errno   = EINVAL;
 
         this = THIS;
         local = sync_frame->local;
@@ -490,16 +1043,31 @@ rebalance_task_completion (int op_ret, call_frame_t *sync_frame, void *data)
                         dht_layout_unref (this, layout);
                 }
 
-                ret = dht_layout_preset (this, local->to_subvol,
+                ret = dht_layout_preset (this, local->rebalance.target_node,
                                          local->loc.inode);
                 if (ret)
                         gf_log (this->name, GF_LOG_WARNING,
                                 "%s: failed to set inode ctx", local->loc.path);
         }
 
-        /* if success, errno is not checked,
-           if ret is -1, then let errno be 'ENOTSUP' */
-        DHT_STACK_UNWIND (setxattr, sync_frame, op_ret, ENOTSUP);
+        if (op_ret == -1) {
+                /* Failure of migration process, mostly due to write process.
+                   as we can't preserve the exact errno, lets say there was
+                   no space to migrate-data
+                */
+                op_errno = ENOSPC;
+        }
+
+        if (op_ret == 1) {
+                /* migration didn't happen, but is not a failure, let the user
+                   understand that he doesn't have permission to migrate the
+                   file.
+                */
+                op_ret = -1;
+                op_errno = EPERM;
+        }
+
+        DHT_STACK_UNWIND (setxattr, sync_frame, op_ret, op_errno, NULL);
         return 0;
 }
 
@@ -507,12 +1075,834 @@ int
 dht_start_rebalance_task (xlator_t *this, call_frame_t *frame)
 {
         int         ret     = -1;
-        dht_conf_t *conf    = NULL;
 
-        conf = this->private;
-
-        ret = synctask_new (conf->env, rebalance_task,
+        ret = synctask_new (this->ctx->env, rebalance_task,
                             rebalance_task_completion,
                             frame, frame);
         return ret;
 }
+
+int
+gf_listener_stop (xlator_t *this)
+{
+        glusterfs_ctx_t  *ctx = NULL;
+        cmd_args_t       *cmd_args = NULL;
+        int              ret = 0;
+
+        ctx = this->ctx;
+        GF_ASSERT (ctx);
+        cmd_args = &ctx->cmd_args;
+        if (cmd_args->sock_file) {
+                ret = unlink (cmd_args->sock_file);
+                if (ret && (ENOENT == errno)) {
+                        ret = 0;
+                }
+        }
+
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Failed to unlink listener "
+                        "socket %s, error: %s", cmd_args->sock_file,
+                        strerror (errno));
+        }
+        return ret;
+}
+
+void
+dht_build_root_inode (xlator_t *this, inode_t **inode)
+{
+        inode_table_t    *itable        = NULL;
+        uuid_t            root_gfid     = {0, };
+
+        itable = inode_table_new (0, this);
+        if (!itable)
+                return;
+
+        root_gfid[15] = 1;
+        *inode = inode_find (itable, root_gfid);
+}
+
+void
+dht_build_root_loc (inode_t *inode, loc_t *loc)
+{
+        loc->path = "/";
+        loc->inode = inode;
+        loc->inode->ia_type = IA_IFDIR;
+        memset (loc->gfid, 0, 16);
+        loc->gfid[15] = 1;
+}
+
+
+/* return values: 1 -> error, bug ignore and continue
+                  0 -> proceed
+                 -1 -> error, handle it */
+int32_t
+gf_defrag_handle_migrate_error (int32_t op_errno, gf_defrag_info_t *defrag)
+{
+        /* if errno is not ENOSPC or ENOTCONN, we can still continue
+           with rebalance process */
+        if ((op_errno != ENOSPC) || (op_errno != ENOTCONN))
+                return 1;
+
+        if (op_errno == ENOTCONN) {
+                /* Most probably mount point went missing (mostly due
+                   to a brick down), say rebalance failure to user,
+                   let him restart it if everything is fine */
+                defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+                return -1;
+        }
+
+        if (op_errno == ENOSPC) {
+                /* rebalance process itself failed, may be
+                   remote brick went down, or write failed due to
+                   disk full etc etc.. */
+                defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+                return -1;
+        }
+
+        return 0;
+}
+
+static gf_boolean_t
+gf_defrag_pattern_match (gf_defrag_info_t *defrag, char *name, uint64_t size)
+{
+        gf_defrag_pattern_list_t *trav = NULL;
+        gf_boolean_t               match = _gf_false;
+        gf_boolean_t               ret = _gf_false;
+
+        GF_VALIDATE_OR_GOTO ("dht", defrag, out);
+
+        trav = defrag->defrag_pattern;
+        while (trav) {
+                if (!fnmatch (trav->path_pattern, name, FNM_NOESCAPE)) {
+                        match = _gf_true;
+                        break;
+                }
+                trav = trav->next;
+        }
+
+        if ((match == _gf_true) && (size >= trav->size))
+                ret = _gf_true;
+
+ out:
+        return ret;
+}
+
+/* We do a depth first traversal of directories. But before we move into
+ * subdirs, we complete the data migration of those directories whose layouts
+ * have been fixed
+ */
+
+int
+gf_defrag_migrate_data (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
+                        dict_t *migrate_data)
+{
+        int                      ret            = -1;
+        loc_t                    entry_loc      = {0,};
+        fd_t                    *fd             = NULL;
+        gf_dirent_t              entries;
+        gf_dirent_t             *tmp            = NULL;
+        gf_dirent_t             *entry          = NULL;
+        gf_boolean_t             free_entries   = _gf_false;
+        off_t                    offset         = 0;
+        dict_t                  *dict           = NULL;
+        struct iatt              iatt           = {0,};
+        int32_t                  op_errno       = 0;
+        char                    *uuid_str       = NULL;
+        uuid_t                   node_uuid      = {0,};
+        struct timeval           dir_start      = {0,};
+        struct timeval           end            = {0,};
+        double                   elapsed        = {0,};
+        struct timeval           start          = {0,};
+        int32_t                  err            = 0;
+        int                      loglevel       = GF_LOG_TRACE;
+
+        gf_log (this->name, GF_LOG_INFO, "migrate data called on %s",
+                loc->path);
+        gettimeofday (&dir_start, NULL);
+
+        fd = fd_create (loc->inode, defrag->pid);
+        if (!fd) {
+                gf_log (this->name, GF_LOG_ERROR, "Failed to create fd");
+                goto out;
+        }
+
+        ret = syncop_opendir (this, loc, fd);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Failed to open dir %s",
+                        loc->path);
+                ret = -1;
+                goto out;
+        }
+
+        INIT_LIST_HEAD (&entries.list);
+
+        while ((ret = syncop_readdirp (this, fd, 131072, offset, NULL,
+                                       &entries)) != 0) {
+
+                if (ret < 0) {
+
+                        gf_log (this->name, GF_LOG_ERROR, "Readdir returned %s."
+                                " Aborting migrate-data",
+                                strerror(-ret));
+                        ret = -1;
+                        goto out;
+                }
+
+                if (list_empty (&entries.list))
+                        break;
+
+                free_entries = _gf_true;
+
+                list_for_each_entry_safe (entry, tmp, &entries.list, list) {
+                        if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+                                ret = 1;
+                                goto out;
+                        }
+
+                        offset = entry->d_off;
+
+                        if (!strcmp (entry->d_name, ".") ||
+                            !strcmp (entry->d_name, ".."))
+                                continue;
+
+                        if (IA_ISDIR (entry->d_stat.ia_type))
+                                continue;
+
+                        defrag->num_files_lookedup++;
+                        if (defrag->stats == _gf_true) {
+                                gettimeofday (&start, NULL);
+                        }
+                        if (defrag->defrag_pattern &&
+                            (gf_defrag_pattern_match (defrag, entry->d_name,
+                                                      entry->d_stat.ia_size)
+                             == _gf_false)) {
+                                continue;
+                        }
+                        loc_wipe (&entry_loc);
+                        ret =dht_build_child_loc (this, &entry_loc, loc,
+                                                  entry->d_name);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR, "Child loc"
+                                        " build failed");
+                                goto out;
+                        }
+
+                        if (uuid_is_null (entry->d_stat.ia_gfid)) {
+                                gf_log (this->name, GF_LOG_ERROR, "%s/%s"
+                                        " gfid not present", loc->path,
+                                         entry->d_name);
+                                continue;
+                        }
+
+                        uuid_copy (entry_loc.gfid, entry->d_stat.ia_gfid);
+
+                        if (uuid_is_null (loc->gfid)) {
+                                gf_log (this->name, GF_LOG_ERROR, "%s/%s"
+                                        " gfid not present", loc->path,
+                                         entry->d_name);
+                                continue;
+                        }
+
+                        uuid_copy (entry_loc.pargfid, loc->gfid);
+
+                        entry_loc.inode->ia_type = entry->d_stat.ia_type;
+
+                        ret = syncop_lookup (this, &entry_loc, NULL, &iatt,
+                                             NULL, NULL);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR, "%s"
+                                        " lookup failed", entry_loc.path);
+                                ret = -1;
+                                continue;
+                        }
+
+                        ret = syncop_getxattr (this, &entry_loc, &dict,
+                                               GF_XATTR_NODE_UUID_KEY);
+                        if(ret < 0) {
+                                gf_log (this->name, GF_LOG_ERROR, "Failed to "
+                                        "get node-uuid for %s", entry_loc.path);
+                                ret = -1;
+                                continue;
+                        }
+
+                        ret = dict_get_str (dict, GF_XATTR_NODE_UUID_KEY,
+                                            &uuid_str);
+                        if(ret < 0) {
+                                gf_log (this->name, GF_LOG_ERROR, "Failed to "
+                                        "get node-uuid from dict for %s",
+                                        entry_loc.path);
+                                ret = -1;
+                                continue;
+                        }
+
+                        if (uuid_parse (uuid_str, node_uuid)) {
+                                gf_log (this->name, GF_LOG_ERROR, "uuid_parse "
+                                        "failed for %s", entry_loc.path);
+                                continue;
+                        }
+
+                        /* if file belongs to different node, skip migration
+                         * the other node will take responsibility of migration
+                         */
+                        if (uuid_compare (node_uuid, defrag->node_uuid)) {
+                                gf_log (this->name, GF_LOG_TRACE, "%s does not"
+                                        "belong to this node", entry_loc.path);
+                                continue;
+                        }
+
+                        uuid_str = NULL;
+
+                        dict_del (dict, GF_XATTR_NODE_UUID_KEY);
+
+
+                        /* if distribute is present, it will honor this key.
+                         * -1, ENODATA is returned if distribute is not present
+                         * or file doesn't have a link-file. If file has
+                         * link-file, the path of link-file will be the value,
+                         * and also that guarantees that file has to be mostly
+                         * migrated */
+
+                        ret = syncop_getxattr (this, &entry_loc, &dict,
+                                               GF_XATTR_LINKINFO_KEY);
+                        if (ret < 0) {
+                                if (-ret != ENODATA) {
+                                        loglevel = GF_LOG_ERROR;
+                                        defrag->total_failures += 1;
+                                } else {
+                                        loglevel = GF_LOG_TRACE;
+                                }
+                                gf_log (this->name, loglevel, "%s: failed to "
+                                        "get "GF_XATTR_LINKINFO_KEY" key - %s",
+                                        entry_loc.path, strerror (-ret));
+                                ret = -1;
+                                continue;
+                        }
+
+                        ret = syncop_setxattr (this, &entry_loc, migrate_data,
+                                               0);
+                        if (ret) {
+                                err = op_errno;
+                                /* errno is overloaded. See
+                                 * rebalance_task_completion () */
+                                if (err != ENOSPC) {
+                                        gf_log (this->name, GF_LOG_DEBUG,
+                                                "migrate-data skipped for %s"
+                                                " due to space constraints",
+                                                entry_loc.path);
+                                        defrag->skipped +=1;
+                                } else{
+                                        gf_log (this->name, GF_LOG_ERROR,
+                                                "migrate-data failed for %s",
+                                                entry_loc.path);
+                                        defrag->total_failures +=1;
+                                }
+                        }
+
+                        if (ret < 0) {
+                                op_errno = -ret;
+                                ret = gf_defrag_handle_migrate_error (op_errno,
+                                                                      defrag);
+
+                                if (!ret)
+                                        gf_log (this->name, GF_LOG_DEBUG,
+                                                "migrate-data on %s failed: %s",
+                                                entry_loc.path,
+                                                strerror (op_errno));
+                                else if (ret == 1)
+                                        continue;
+                                else if (ret == -1)
+                                        goto out;
+                        }
+
+                        LOCK (&defrag->lock);
+                        {
+                                defrag->total_files += 1;
+                                defrag->total_data += iatt.ia_size;
+                        }
+                        UNLOCK (&defrag->lock);
+                        if (defrag->stats == _gf_true) {
+                                gettimeofday (&end, NULL);
+                                elapsed = (end.tv_sec - start.tv_sec) * 1e6 +
+                                          (end.tv_usec - start.tv_usec);
+                                gf_log (this->name, GF_LOG_INFO, "Migration of "
+                                        "file:%s size:%"PRIu64" bytes took %.2f"
+                                        "secs", entry_loc.path, iatt.ia_size,
+                                         elapsed/1e6);
+                        }
+                }
+
+                gf_dirent_free (&entries);
+                free_entries = _gf_false;
+                INIT_LIST_HEAD (&entries.list);
+        }
+
+        gettimeofday (&end, NULL);
+        elapsed = (end.tv_sec - dir_start.tv_sec) * 1e6 +
+                  (end.tv_usec - dir_start.tv_usec);
+        gf_log (this->name, GF_LOG_INFO, "Migration operation on dir %s took "
+                "%.2f secs", loc->path, elapsed/1e6);
+        ret = 0;
+out:
+        if (free_entries)
+                gf_dirent_free (&entries);
+
+        loc_wipe (&entry_loc);
+
+        if (dict)
+                dict_unref(dict);
+
+        if (fd)
+                fd_unref (fd);
+        return ret;
+
+}
+
+int
+gf_defrag_fix_layout (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
+                  dict_t *fix_layout, dict_t *migrate_data)
+{
+        int                      ret            = -1;
+        loc_t                    entry_loc      = {0,};
+        fd_t                    *fd             = NULL;
+        gf_dirent_t              entries;
+        gf_dirent_t             *tmp            = NULL;
+        gf_dirent_t             *entry          = NULL;
+        gf_boolean_t             free_entries   = _gf_false;
+        dict_t                  *dict           = NULL;
+        off_t                    offset         = 0;
+        struct iatt              iatt           = {0,};
+
+        ret = syncop_lookup (this, loc, NULL, &iatt, NULL, NULL);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Lookup failed on %s",
+                        loc->path);
+                ret = -1;
+                goto out;
+        }
+
+        if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) {
+                ret = gf_defrag_migrate_data (this, defrag, loc, migrate_data);
+                if (ret)
+                        goto out;
+        }
+
+        gf_log (this->name, GF_LOG_TRACE, "fix layout called on %s", loc->path);
+
+        fd = fd_create (loc->inode, defrag->pid);
+        if (!fd) {
+                gf_log (this->name, GF_LOG_ERROR, "Failed to create fd");
+                ret = -1;
+                goto out;
+        }
+
+        ret = syncop_opendir (this, loc, fd);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Failed to open dir %s",
+                        loc->path);
+                ret = -1;
+                goto out;
+        }
+
+        INIT_LIST_HEAD (&entries.list);
+        while ((ret = syncop_readdirp (this, fd, 131072, offset, NULL,
+                &entries)) != 0)
+        {
+
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR, "Readdir returned %s"
+                                ". Aborting fix-layout",strerror(-ret));
+                        ret = -1;
+                        goto out;
+                }
+
+                if (list_empty (&entries.list))
+                        break;
+
+                free_entries = _gf_true;
+
+                list_for_each_entry_safe (entry, tmp, &entries.list, list) {
+                        if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+                                ret = 1;
+                                goto out;
+                        }
+
+                        offset = entry->d_off;
+
+                        if (!strcmp (entry->d_name, ".") ||
+                            !strcmp (entry->d_name, ".."))
+                                continue;
+
+                        if (!IA_ISDIR (entry->d_stat.ia_type))
+                                continue;
+
+                        loc_wipe (&entry_loc);
+                        ret =dht_build_child_loc (this, &entry_loc, loc,
+                                                  entry->d_name);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR, "Child loc"
+                                        " build failed");
+                                goto out;
+                        }
+
+                        if (uuid_is_null (entry->d_stat.ia_gfid)) {
+                                gf_log (this->name, GF_LOG_ERROR, "%s/%s"
+                                        " gfid not present", loc->path,
+                                         entry->d_name);
+                                continue;
+                        }
+
+                        entry_loc.inode->ia_type = entry->d_stat.ia_type;
+
+                        uuid_copy (entry_loc.gfid, entry->d_stat.ia_gfid);
+                        if (uuid_is_null (loc->gfid)) {
+                                gf_log (this->name, GF_LOG_ERROR, "%s/%s"
+                                        " gfid not present", loc->path,
+                                         entry->d_name);
+                                continue;
+                        }
+
+                        uuid_copy (entry_loc.pargfid, loc->gfid);
+
+                        ret = syncop_lookup (this, &entry_loc, NULL, &iatt,
+                                             NULL, NULL);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR, "%s"
+                                        " lookup failed", entry_loc.path);
+                                ret = -1;
+                                continue;
+                        }
+
+                        ret = syncop_setxattr (this, &entry_loc, fix_layout,
+                                               0);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR, "Setxattr "
+                                        "failed for %s", entry_loc.path);
+                                defrag->defrag_status =
+                                GF_DEFRAG_STATUS_FAILED;
+                                defrag->total_failures ++;
+                                ret = -1;
+                                goto out;
+                        }
+                        ret = gf_defrag_fix_layout (this, defrag, &entry_loc,
+                                                    fix_layout, migrate_data);
+
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR, "Fix layout "
+                                        "failed for %s", entry_loc.path);
+                                defrag->total_failures++;
+                                goto out;
+                        }
+
+                }
+                gf_dirent_free (&entries);
+                free_entries = _gf_false;
+                INIT_LIST_HEAD (&entries.list);
+        }
+
+        ret = 0;
+out:
+        if (free_entries)
+                gf_dirent_free (&entries);
+
+        loc_wipe (&entry_loc);
+
+        if (dict)
+                dict_unref(dict);
+
+        if (fd)
+                fd_unref (fd);
+
+        return ret;
+
+}
+
+
+int
+gf_defrag_start_crawl (void *data)
+{
+        xlator_t                *this   = NULL;
+        dht_conf_t              *conf   = NULL;
+        gf_defrag_info_t        *defrag = NULL;
+        int                      ret    = -1;
+        loc_t                    loc    = {0,};
+        struct iatt              iatt   = {0,};
+        struct iatt              parent = {0,};
+        dict_t                  *fix_layout = NULL;
+        dict_t                  *migrate_data = NULL;
+        dict_t                  *status = NULL;
+        glusterfs_ctx_t         *ctx = NULL;
+
+        this = data;
+        if (!this)
+                goto out;
+
+        ctx = this->ctx;
+        if (!ctx)
+                goto out;
+
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        defrag = conf->defrag;
+        if (!defrag)
+                goto out;
+
+        gettimeofday (&defrag->start_time, NULL);
+        dht_build_root_inode (this, &defrag->root_inode);
+        if (!defrag->root_inode)
+                goto out;
+
+        dht_build_root_loc (defrag->root_inode, &loc);
+
+        /* fix-layout on '/' first */
+
+        ret = syncop_lookup (this, &loc, NULL, &iatt, NULL, &parent);
+
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "look up on / failed");
+                ret = -1;
+                goto out;
+        }
+
+        fix_layout = dict_new ();
+        if (!fix_layout) {
+                ret = -1;
+                goto out;
+        }
+
+        ret = dict_set_str (fix_layout, GF_XATTR_FIX_LAYOUT_KEY, "yes");
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Failed to set dict str");
+                goto out;
+        }
+
+        ret = syncop_setxattr (this, &loc, fix_layout, 0);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "fix layout on %s failed",
+                        loc.path);
+                defrag->total_failures++;
+                ret = -1;
+                goto out;
+        }
+
+        if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) {
+                migrate_data = dict_new ();
+                if (!migrate_data) {
+                        ret = -1;
+                        goto out;
+                }
+                if (defrag->cmd == GF_DEFRAG_CMD_START_FORCE)
+                        ret = dict_set_str (migrate_data,
+                                            "distribute.migrate-data", "force");
+                else
+                        ret = dict_set_str (migrate_data,
+                                            "distribute.migrate-data",
+                                            "non-force");
+                if (ret)
+                        goto out;
+        }
+        ret = gf_defrag_fix_layout (this, defrag, &loc, fix_layout,
+                                    migrate_data);
+        if ((defrag->defrag_status != GF_DEFRAG_STATUS_STOPPED) &&
+            (defrag->defrag_status != GF_DEFRAG_STATUS_FAILED)) {
+                defrag->defrag_status = GF_DEFRAG_STATUS_COMPLETE;
+        }
+
+
+
+out:
+        LOCK (&defrag->lock);
+        {
+                status = dict_new ();
+                gf_defrag_status_get (defrag, status);
+                if (ctx->notify)
+                        ctx->notify (GF_EN_DEFRAG_STATUS, status);
+                if (status)
+                        dict_unref (status);
+                defrag->is_exiting = 1;
+        }
+        UNLOCK (&defrag->lock);
+
+        if (defrag) {
+                GF_FREE (defrag);
+                conf->defrag = NULL;
+        }
+
+        return ret;
+}
+
+
+static int
+gf_defrag_done  (int ret, call_frame_t *sync_frame, void *data)
+{
+        gf_listener_stop (sync_frame->this);
+
+        STACK_DESTROY (sync_frame->root);
+        kill (getpid(), SIGTERM);
+        return 0;
+}
+
+void *
+gf_defrag_start (void *data)
+{
+        int                      ret    = -1;
+        call_frame_t            *frame  = NULL;
+        dht_conf_t              *conf   = NULL;
+        gf_defrag_info_t        *defrag = NULL;
+        xlator_t                *this  = NULL;
+
+        this = data;
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        defrag = conf->defrag;
+        if (!defrag)
+                goto out;
+
+        frame = create_frame (this, this->ctx->pool);
+        if (!frame)
+                goto out;
+
+        frame->root->pid = GF_CLIENT_PID_DEFRAG;
+
+        defrag->pid = frame->root->pid;
+
+        defrag->defrag_status = GF_DEFRAG_STATUS_STARTED;
+
+        ret = synctask_new (this->ctx->env, gf_defrag_start_crawl,
+                            gf_defrag_done, frame, this);
+
+        if (ret)
+                gf_log (this->name, GF_LOG_ERROR, "Could not create"
+                        " task for rebalance");
+out:
+        return NULL;
+}
+
+int
+gf_defrag_status_get (gf_defrag_info_t *defrag, dict_t *dict)
+{
+        int      ret    = 0;
+        uint64_t files  = 0;
+        uint64_t size   = 0;
+        uint64_t lookup = 0;
+        uint64_t failures = 0;
+        uint64_t skipped = 0;
+        char     *status = "";
+        double   elapsed = 0;
+        struct timeval end = {0,};
+
+
+        if (!defrag)
+                goto out;
+
+        ret = 0;
+        if (defrag->defrag_status == GF_DEFRAG_STATUS_NOT_STARTED)
+                goto out;
+
+        files  = defrag->total_files;
+        size   = defrag->total_data;
+        lookup = defrag->num_files_lookedup;
+        failures = defrag->total_failures;
+        skipped = defrag->skipped;
+
+        gettimeofday (&end, NULL);
+
+        elapsed = end.tv_sec - defrag->start_time.tv_sec;
+
+        if (!dict)
+                goto log;
+
+        ret = dict_set_uint64 (dict, "files", files);
+        if (ret)
+                gf_log (THIS->name, GF_LOG_WARNING,
+                        "failed to set file count");
+
+        ret = dict_set_uint64 (dict, "size", size);
+        if (ret)
+                gf_log (THIS->name, GF_LOG_WARNING,
+                        "failed to set size of xfer");
+
+        ret = dict_set_uint64 (dict, "lookups", lookup);
+        if (ret)
+                gf_log (THIS->name, GF_LOG_WARNING,
+                        "failed to set lookedup file count");
+
+
+        ret = dict_set_int32 (dict, "status", defrag->defrag_status);
+        if (ret)
+                gf_log (THIS->name, GF_LOG_WARNING,
+                        "failed to set status");
+        if (elapsed) {
+                ret = dict_set_double (dict, "run-time", elapsed);
+                if (ret)
+                        gf_log (THIS->name, GF_LOG_WARNING,
+                                "failed to set run-time");
+        }
+
+        ret = dict_set_uint64 (dict, "failures", failures);
+        if (ret)
+                gf_log (THIS->name, GF_LOG_WARNING,
+                        "failed to set failure count");
+
+        ret = dict_set_uint64 (dict, "skipped", skipped);
+        if (ret)
+                gf_log (THIS->name, GF_LOG_WARNING,
+                        "failed to set skipped file count");
+log:
+        switch (defrag->defrag_status) {
+        case GF_DEFRAG_STATUS_NOT_STARTED:
+                status = "not started";
+                break;
+        case GF_DEFRAG_STATUS_STARTED:
+                status = "in progress";
+                break;
+        case GF_DEFRAG_STATUS_STOPPED:
+                status = "stopped";
+                break;
+        case GF_DEFRAG_STATUS_COMPLETE:
+                status = "completed";
+                break;
+        case GF_DEFRAG_STATUS_FAILED:
+                status = "failed";
+                break;
+        default:
+                break;
+        }
+
+        gf_log (THIS->name, GF_LOG_INFO, "Rebalance is %s. Time taken is %.2f "
+                "secs", status, elapsed);
+        gf_log (THIS->name, GF_LOG_INFO, "Files migrated: %"PRIu64", size: %"
+                PRIu64", lookups: %"PRIu64", failures: %"PRIu64", skipped: "
+                "%"PRIu64, files, size, lookup, failures, skipped);
+
+
+out:
+        return 0;
+}
+
+int
+gf_defrag_stop (gf_defrag_info_t *defrag, gf_defrag_status_t status,
+                dict_t *output)
+{
+        /* TODO: set a variable 'stop_defrag' here, it should be checked
+           in defrag loop */
+        int     ret = -1;
+        GF_ASSERT (defrag);
+
+        if (defrag->defrag_status == GF_DEFRAG_STATUS_NOT_STARTED) {
+                goto out;
+        }
+
+        gf_log ("", GF_LOG_INFO, "Received stop command on rebalance");
+        defrag->defrag_status = status;
+
+        if (output)
+                gf_defrag_status_get (defrag, output);
+        ret = 0;
+out:
+        gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
+        return ret;
+}
diff --git a/xlators/cluster/dht/src/dht-rename.c b/xlators/cluster/dht/src/dht-rename.c
index 8110387d5..925538cc8 100644
--- a/xlators/cluster/dht/src/dht-rename.c
+++ b/xlators/cluster/dht/src/dht-rename.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 /* TODO: link(oldpath, newpath) fails if newpath already exists. DHT should
@@ -35,7 +26,8 @@ int
 dht_rename_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                     int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
                     struct iatt *preoldparent, struct iatt *postoldparent,
-                    struct iatt *prenewparent, struct iatt *postnewparent)
+                    struct iatt *prenewparent, struct iatt *postnewparent,
+                    dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         int           this_call_cnt = 0;
@@ -80,10 +72,11 @@ unwind:
                 WIPE (&local->preparent);
                 WIPE (&local->postparent);
 
+                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
                 DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
                                   &local->stbuf, &local->preoldparent,
                                   &local->postoldparent,
-                                  &local->preparent, &local->postparent);
+                                  &local->preparent, &local->postparent, xdata);
         }
 
         return 0;
@@ -96,7 +89,7 @@ dht_rename_hashed_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                            struct iatt *preoldparent,
                            struct iatt *postoldparent,
                            struct iatt *prenewparent,
-                           struct iatt *postnewparent)
+                           struct iatt *postnewparent, dict_t *xdata)
 {
         dht_conf_t   *conf = NULL;
         dht_local_t  *local = NULL;
@@ -146,7 +139,7 @@ dht_rename_hashed_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 STACK_WIND (frame, dht_rename_dir_cbk,
                             conf->subvolumes[i],
                             conf->subvolumes[i]->fops->rename,
-                            &local->loc, &local->loc2);
+                            &local->loc, &local->loc2, NULL);
                 if (!--call_cnt)
                         break;
         }
@@ -159,10 +152,11 @@ unwind:
         WIPE (&local->preparent);
         WIPE (&local->postparent);
 
+        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
         DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
                           &local->stbuf, &local->preoldparent,
                           &local->postoldparent,
-                          &local->preparent, &local->postparent);
+                          &local->preparent, &local->postparent, NULL);
 
         return 0;
 }
@@ -172,9 +166,7 @@ int
 dht_rename_dir_do (call_frame_t *frame, xlator_t *this)
 {
         dht_local_t  *local = NULL;
-        dht_conf_t   *conf = NULL;
 
-        conf = this->private;
         local = frame->local;
 
         if (local->op_ret == -1)
@@ -185,19 +177,20 @@ dht_rename_dir_do (call_frame_t *frame, xlator_t *this)
         STACK_WIND (frame, dht_rename_hashed_dir_cbk,
                     local->dst_hashed,
                     local->dst_hashed->fops->rename,
-                    &local->loc, &local->loc2);
+                    &local->loc, &local->loc2, NULL);
         return 0;
 
 err:
         DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno, NULL, NULL,
-                          NULL, NULL, NULL);
+                          NULL, NULL, NULL, NULL);
         return 0;
 }
 
 
 int
 dht_rename_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int op_ret, int op_errno, gf_dirent_t *entries)
+                        int op_ret, int op_errno, gf_dirent_t *entries,
+                        dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         int           this_call_cnt = -1;
@@ -226,7 +219,7 @@ dht_rename_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 int
 dht_rename_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int op_ret, int op_errno, fd_t *fd)
+                        int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         int           this_call_cnt = -1;
@@ -246,7 +239,7 @@ dht_rename_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         STACK_WIND (frame, dht_rename_readdir_cbk,
                     prev->this, prev->this->fops->readdir,
-                    local->fd, 4096, 0);
+                    local->fd, 4096, 0, NULL);
 
         return 0;
 
@@ -302,22 +295,70 @@ dht_rename_dir (call_frame_t *frame, xlator_t *this)
                 STACK_WIND (frame, dht_rename_opendir_cbk,
                             conf->subvolumes[i],
                             conf->subvolumes[i]->fops->opendir,
-                            &local->loc2, local->fd);
+                            &local->loc2, local->fd, NULL);
         }
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+        DHT_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                          NULL, NULL);
         return 0;
 }
+#define DHT_MARK_FOP_INTERNAL(xattr) do {                                      \
+                int tmp = -1;                                                  \
+                if (!xattr) {                                                  \
+                        xattr = dict_new ();                                   \
+                        if (!xattr)                                            \
+                                break;                                         \
+                }                                                              \
+                tmp = dict_set_str (xattr, GLUSTERFS_INTERNAL_FOP_KEY, "yes"); \
+                if (tmp) {                                                     \
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to set"      \
+                                " internal dict key for %s", local->loc.path); \
+                }                                                              \
+        }while (0)
+
+#define DHT_MARKER_DONT_ACCOUNT(xattr) do {                             \
+                int tmp = -1;                                                  \
+                if (!xattr) {                                                  \
+                        xattr = dict_new ();                                   \
+                        if (!xattr)                                            \
+                                break;                                         \
+                }                                                              \
+                tmp = dict_set_str (xattr, GLUSTERFS_MARKER_DONT_ACCOUNT_KEY,  \
+                                    "yes");                                    \
+                if (tmp) {                                                     \
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to set"      \
+                                " marker dont account key for %s", local->loc.path); \
+                }                                                              \
+        }while (0)
+
+int
+dht_rename_done (call_frame_t *frame, xlator_t *this)
+{
+        dht_local_t     *local = NULL;
+
+        local = frame->local;
+
+        if (local->linked == _gf_true) {
+                local->linked = _gf_false;
+                dht_linkfile_attr_heal (frame, this);
+        }
+        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
+        DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
+                          &local->stbuf, &local->preoldparent,
+                          &local->postoldparent, &local->preparent,
+                          &local->postparent, NULL);
 
+        return 0;
+}
 
 int
 dht_rename_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                        int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                       struct iatt *postparent)
+                       struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         call_frame_t *prev = NULL;
@@ -346,10 +387,7 @@ dht_rename_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         WIPE (&local->postparent);
 
         if (is_last_call (this_call_cnt)) {
-                DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-                                  &local->stbuf, &local->preoldparent,
-                                  &local->postoldparent, &local->preparent,
-                                  &local->postparent);
+                dht_rename_done (frame, this);
         }
 
 out:
@@ -367,7 +405,7 @@ dht_rename_cleanup (call_frame_t *frame)
         xlator_t    *dst_hashed = NULL;
         xlator_t    *dst_cached = NULL;
         int          call_cnt = 0;
-
+        dict_t      *xattr      = NULL;
 
         local = frame->local;
         this  = frame->this;
@@ -391,24 +429,53 @@ dht_rename_cleanup (call_frame_t *frame)
         if (!call_cnt)
                 goto nolinks;
 
+        DHT_MARK_FOP_INTERNAL (xattr);
+
         if (dst_hashed != src_hashed && dst_hashed != src_cached) {
+                dict_t *xattr_new = NULL;
+
                 gf_log (this->name, GF_LOG_TRACE,
                         "unlinking linkfile %s @ %s => %s",
                         local->loc.path, dst_hashed->name, src_cached->name);
+
+                xattr_new = dict_copy_with_ref (xattr, NULL);
+
+
+                DHT_MARKER_DONT_ACCOUNT(xattr_new);
+
                 STACK_WIND (frame, dht_rename_unlink_cbk,
                             dst_hashed, dst_hashed->fops->unlink,
-                            &local->loc);
+                            &local->loc, 0, xattr_new);
+
+                dict_unref (xattr_new);
+                xattr_new = NULL;
         }
 
         if (src_cached != dst_hashed) {
+                dict_t *xattr_new = NULL;
+
                 gf_log (this->name, GF_LOG_TRACE,
                         "unlinking link %s => %s (%s)", local->loc.path,
                         local->loc2.path, src_cached->name);
+
+                xattr_new = dict_copy_with_ref (xattr, NULL);
+
+                if (uuid_compare (local->loc.pargfid,
+                                  local->loc2.pargfid) == 0) {
+                        DHT_MARKER_DONT_ACCOUNT(xattr_new);
+                }
+
                 STACK_WIND (frame, dht_rename_unlink_cbk,
                             src_cached, src_cached->fops->unlink,
-                            &local->loc2);
+                            &local->loc2, 0, xattr_new);
+
+                dict_unref (xattr_new);
+                xattr_new = NULL;
         }
 
+        if (xattr)
+                dict_unref (xattr);
+
         return 0;
 
 nolinks:
@@ -417,10 +484,40 @@ nolinks:
         WIPE (&local->preparent);
         WIPE (&local->postparent);
 
+        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
         DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
                           &local->stbuf, &local->preoldparent,
                           &local->postoldparent, &local->preparent,
-                          &local->postparent);
+                          &local->postparent, NULL);
+
+        return 0;
+}
+
+
+int
+dht_rename_links_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno,
+                             inode_t *inode, struct iatt *stbuf,
+                             struct iatt *preparent, struct iatt *postparent,
+                             dict_t *xdata)
+{
+        call_frame_t *prev = NULL;
+        dht_local_t  *local = NULL;
+
+        prev = cookie;
+        local = frame->local;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "link/file %s on %s failed (%s)",
+                        local->loc.path, prev->this->name, strerror (op_errno));
+        }
+
+        if (local->linked == _gf_true) {
+                local->linked = _gf_false;
+                dht_linkfile_attr_heal (frame, this);
+        }
+        DHT_STACK_DESTROY (frame);
 
         return 0;
 }
@@ -430,7 +527,8 @@ int
 dht_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
                 struct iatt *preoldparent, struct iatt *postoldparent,
-                struct iatt *prenewparent, struct iatt *postnewparent)
+                struct iatt *prenewparent, struct iatt *postnewparent,
+                dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         call_frame_t *prev = NULL;
@@ -439,6 +537,9 @@ dht_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         xlator_t     *dst_hashed = NULL;
         xlator_t     *dst_cached = NULL;
         xlator_t     *rename_subvol = NULL;
+        call_frame_t *link_frame = NULL;
+        dht_local_t *link_local = NULL;
+        dict_t       *xattr     = NULL;
 
         local = frame->local;
         prev = cookie;
@@ -448,6 +549,8 @@ dht_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         dst_hashed = local->dst_hashed;
         dst_cached = local->dst_cached;
 
+        if (local->linked == _gf_true)
+                FRAME_SU_UNDO (frame, dht_local_t);
         if (op_ret == -1) {
                 gf_log (this->name, GF_LOG_WARNING,
                         "%s: rename on %s failed (%s)", local->loc.path,
@@ -457,11 +560,45 @@ dht_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 goto cleanup;
         }
 
-        dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-        dht_iatt_merge (this, &local->preoldparent, preoldparent, prev->this);
-        dht_iatt_merge (this, &local->postoldparent, postoldparent, prev->this);
-        dht_iatt_merge (this, &local->preparent, prenewparent, prev->this);
-        dht_iatt_merge (this, &local->postparent, postnewparent, prev->this);
+        if ((src_cached == dst_cached) && (dst_hashed != dst_cached)) {
+                link_frame = copy_frame (frame);
+                if (!link_frame) {
+                        goto err;
+                }
+
+                /* fop value sent as maxvalue because it is not used
+                   anywhere in this case */
+                link_local = dht_local_init (link_frame, &local->loc2, NULL,
+                                             GF_FOP_MAXVALUE);
+                if (!link_local) {
+                        goto err;
+                }
+
+                if (link_local->loc.inode)
+                        inode_unref (link_local->loc.inode);
+                link_local->loc.inode = inode_ref (local->loc.inode);
+                uuid_copy (link_local->gfid, local->loc.inode->gfid);
+
+                dht_linkfile_create (link_frame, dht_rename_links_create_cbk,
+                                     this, src_cached, dst_hashed,
+                                     &link_local->loc);
+        }
+
+err:
+        /* Merge attrs only from src_cached. In case there of src_cached !=
+         * dst_hashed, this ignores linkfile attrs. */
+        if (prev->this == src_cached) {
+                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+                dht_iatt_merge (this, &local->preoldparent, preoldparent,
+                                prev->this);
+                dht_iatt_merge (this, &local->postoldparent, postoldparent,
+                                prev->this);
+                dht_iatt_merge (this, &local->preparent, prenewparent,
+                                prev->this);
+                dht_iatt_merge (this, &local->postparent, postnewparent,
+                                prev->this);
+        }
+
 
         /* NOTE: rename_subvol is the same subvolume from which dht_rename_cbk
          *       is called. since rename has already happened on rename_subvol,
@@ -486,24 +623,47 @@ dht_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         if (local->call_cnt == 0)
                 goto unwind;
 
+        DHT_MARK_FOP_INTERNAL (xattr);
+
         if (src_cached != dst_hashed && src_cached != dst_cached) {
+                dict_t *xattr_new = NULL;
+
+                xattr_new = dict_copy_with_ref (xattr, NULL);
+
                 gf_log (this->name, GF_LOG_TRACE,
                         "deleting old src datafile %s @ %s",
                         local->loc.path, src_cached->name);
 
+                if (uuid_compare (local->loc.pargfid,
+                                  local->loc2.pargfid) == 0) {
+                        DHT_MARKER_DONT_ACCOUNT(xattr_new);
+                }
+
                 STACK_WIND (frame, dht_rename_unlink_cbk,
                             src_cached, src_cached->fops->unlink,
-                            &local->loc);
+                            &local->loc, 0, xattr_new);
+
+                dict_unref (xattr_new);
+                xattr_new = NULL;
         }
 
         if (src_hashed != rename_subvol && src_hashed != src_cached) {
+                dict_t *xattr_new = NULL;
+
+                xattr_new = dict_copy_with_ref (xattr, NULL);
+
                 gf_log (this->name, GF_LOG_TRACE,
                         "deleting old src linkfile %s @ %s",
                         local->loc.path, src_hashed->name);
 
+                DHT_MARKER_DONT_ACCOUNT(xattr_new);
+
                 STACK_WIND (frame, dht_rename_unlink_cbk,
                             src_hashed, src_hashed->fops->unlink,
-                            &local->loc);
+                            &local->loc, 0, xattr_new);
+
+                dict_unref (xattr_new);
+                xattr_new = NULL;
         }
 
         if (dst_cached
@@ -515,8 +675,10 @@ dht_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
                 STACK_WIND (frame, dht_rename_unlink_cbk,
                             dst_cached, dst_cached->fops->unlink,
-                            &local->loc2);
+                            &local->loc2, 0, xattr);
         }
+        if (xattr)
+                dict_unref (xattr);
         return 0;
 
 unwind:
@@ -524,15 +686,16 @@ unwind:
         WIPE (&local->postoldparent);
         WIPE (&local->preparent);
         WIPE (&local->postparent);
+        if (xattr)
+                dict_unref (xattr);
 
-        DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-                          &local->stbuf, &local->preoldparent,
-                          &local->postoldparent, &local->preparent,
-                          &local->postparent);
+        dht_rename_done (frame, this);
 
         return 0;
 
 cleanup:
+        if (xattr)
+                dict_unref (xattr);
         dht_rename_cleanup (frame);
 
         return 0;
@@ -542,12 +705,13 @@ cleanup:
 int
 dht_do_rename (call_frame_t *frame)
 {
-        dht_local_t *local = NULL;
-        xlator_t    *dst_hashed = NULL;
-        xlator_t    *src_cached = NULL;
-        xlator_t    *dst_cached = NULL;
-        xlator_t    *this = NULL;
+        dht_local_t *local         = NULL;
+        xlator_t    *dst_hashed    = NULL;
+        xlator_t    *src_cached    = NULL;
+        xlator_t    *dst_cached    = NULL;
+        xlator_t    *this          = NULL;
         xlator_t    *rename_subvol = NULL;
+        dict_t      *dict          = NULL;
 
 
         local = frame->local;
@@ -562,13 +726,19 @@ dht_do_rename (call_frame_t *frame)
         else
                 rename_subvol = dst_hashed;
 
+        if ((src_cached != dst_hashed) && (rename_subvol == dst_hashed)) {
+                DHT_MARKER_DONT_ACCOUNT(dict);
+        }
+
         gf_log (this->name, GF_LOG_TRACE,
                 "renaming %s => %s (%s)",
                 local->loc.path, local->loc2.path, rename_subvol->name);
 
+        if (local->linked == _gf_true)
+                FRAME_SU_DO (frame, dht_local_t);
         STACK_WIND (frame, dht_rename_cbk,
                     rename_subvol, rename_subvol->fops->rename,
-                    &local->loc, &local->loc2);
+                    &local->loc, &local->loc2, dict);
 
         return 0;
 }
@@ -578,7 +748,8 @@ int
 dht_rename_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                       int32_t op_ret, int32_t op_errno,
                       inode_t *inode, struct iatt *stbuf,
-                      struct iatt *preparent, struct iatt *postparent)
+                      struct iatt *preparent, struct iatt *postparent,
+                      dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         call_frame_t *prev = NULL;
@@ -593,7 +764,11 @@ dht_rename_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         "link/file on %s failed (%s)",
                         prev->this->name, strerror (op_errno));
                 local->op_ret   = -1;
-                local->op_errno = op_errno;
+                if (op_errno != ENOENT)
+                        local->op_errno = op_errno;
+        } else if (local->src_cached == prev->this) {
+                /* merge of attr returned only from linkfile creation */
+                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
         }
 
         this_call_cnt = dht_frame_return (frame);
@@ -616,7 +791,8 @@ cleanup:
 int
 dht_rename_unlink_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                              int32_t op_ret, int32_t op_errno,
-                             struct iatt *preparent, struct iatt *postparent)
+                             struct iatt *preparent, struct iatt *postparent,
+                             dict_t *xdata)
 {
 	dht_local_t  *local = NULL;
 	call_frame_t *prev = NULL;
@@ -625,7 +801,7 @@ dht_rename_unlink_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 	local = frame->local;
 	prev = cookie;
 
-	if (op_ret == -1) {
+	if ((op_ret == -1) && (op_errno != ENOENT)) {
 		gf_log (this->name, GF_LOG_DEBUG,
 			"unlink of %s on %s failed (%s)",
 			local->loc2.path, prev->this->name,
@@ -658,6 +834,7 @@ dht_rename_create_links (call_frame_t *frame)
         xlator_t    *dst_hashed = NULL;
         xlator_t    *dst_cached = NULL;
         int          call_cnt = 0;
+        dict_t      *xattr = NULL;
 
 
         local = frame->local;
@@ -668,18 +845,27 @@ dht_rename_create_links (call_frame_t *frame)
         dst_hashed = local->dst_hashed;
         dst_cached = local->dst_cached;
 
+        DHT_MARK_FOP_INTERNAL (xattr);
 
         if (src_cached == dst_cached) {
+                dict_t *xattr_new = NULL;
+
                 if (dst_hashed == dst_cached)
                         goto nolinks;
 
+                xattr_new = dict_copy_with_ref (xattr, NULL);
+
 		gf_log (this->name, GF_LOG_TRACE,
 			"unlinking dst linkfile %s @ %s",
 			local->loc2.path, dst_hashed->name);
 
+                DHT_MARKER_DONT_ACCOUNT(xattr_new);
+
 		STACK_WIND (frame, dht_rename_unlink_links_cbk,
 			    dst_hashed, dst_hashed->fops->unlink,
-			    &local->loc2);
+			    &local->loc2, 0, xattr_new);
+
+                dict_unref (xattr_new);
                 return 0;
         }
 
@@ -696,17 +882,28 @@ dht_rename_create_links (call_frame_t *frame)
                         "linkfile %s @ %s => %s",
                         local->loc.path, dst_hashed->name, src_cached->name);
                 memcpy (local->gfid, local->loc.inode->gfid, 16);
-		dht_linkfile_create (frame, dht_rename_links_cbk,
+		dht_linkfile_create (frame, dht_rename_links_cbk, this,
 				     src_cached, dst_hashed, &local->loc);
 	}
 
 	if (src_cached != dst_hashed) {
+                dict_t *xattr_new = NULL;
+
+                xattr_new = dict_copy_with_ref (xattr, NULL);
+
 		gf_log (this->name, GF_LOG_TRACE,
 			"link %s => %s (%s)", local->loc.path,
 			local->loc2.path, src_cached->name);
+                if (uuid_compare (local->loc.pargfid,
+                                  local->loc2.pargfid) == 0) {
+                        DHT_MARKER_DONT_ACCOUNT(xattr_new);
+                }
+
 		STACK_WIND (frame, dht_rename_links_cbk,
 			    src_cached, src_cached->fops->link,
-			    &local->loc, &local->loc2);
+			    &local->loc, &local->loc2, xattr_new);
+
+                dict_unref (xattr_new);
 	}
 
 nolinks:
@@ -714,6 +911,8 @@ nolinks:
                 /* skip to next step */
                 dht_do_rename (frame);
         }
+        if (xattr)
+                dict_unref (xattr);
 
         return 0;
 }
@@ -721,7 +920,7 @@ nolinks:
 
 int
 dht_rename (call_frame_t *frame, xlator_t *this,
-            loc_t *oldloc, loc_t *newloc)
+            loc_t *oldloc, loc_t *newloc, dict_t *xdata)
 {
         xlator_t    *src_cached = NULL;
         xlator_t    *src_hashed = NULL;
@@ -766,17 +965,14 @@ dht_rename (call_frame_t *frame, xlator_t *this,
         if (newloc->inode)
                 dst_cached = dht_subvol_get_cached (this, newloc->inode);
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, oldloc, NULL, GF_FOP_RENAME);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
-
-        ret = loc_copy (&local->loc, oldloc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+        /* cached_subvol will be set from dht_local_init, reset it to NULL,
+           as the logic of handling rename is different  */
+        local->cached_subvol = NULL;
 
         ret = loc_copy (&local->loc2, newloc);
         if (ret == -1) {
@@ -806,7 +1002,8 @@ dht_rename (call_frame_t *frame, xlator_t *this,
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+        DHT_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                          NULL, NULL);
 
         return 0;
 }
diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c
index 8245cacdc..0e6527544 100644
--- a/xlators/cluster/dht/src/dht-selfheal.c
+++ b/xlators/cluster/dht/src/dht-selfheal.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef _CONFIG_H
@@ -26,7 +17,7 @@
 #include "glusterfs.h"
 #include "xlator.h"
 #include "dht-common.h"
-
+#include "glusterfs-acl.h"
 
 #define DHT_SET_LAYOUT_RANGE(layout,i,srt,chunk,cnt,path)    do {       \
                 layout->list[i].start = srt;                            \
@@ -38,43 +29,40 @@
                         layout->list[i].xlator->name, path);            \
         } while (0)
 
+#define DHT_RESET_LAYOUT_RANGE(layout)    do {                          \
+                int cnt = 0;                                            \
+                for (cnt = 0; cnt < layout->cnt; cnt++ ) {              \
+                        layout->list[cnt].start = 0;                    \
+                        layout->list[cnt].stop  = 0;                    \
+                }                                                       \
+        } while (0)
 
-static inline uint32_t
-dht_find_overlap (int idx, int cnk_idx, uint32_t start, uint32_t stop,
-                  uint32_t chunk_size)
+static uint32_t
+dht_overlap_calc (dht_layout_t *old, int o, dht_layout_t *new, int n)
 {
-        uint32_t overlap = 0;
-        uint32_t chunk_begin = 0;
-
-        chunk_begin = cnk_idx * chunk_size;
+	if (o >= old->cnt || n >= new->cnt)
+		return 0;
 
-        /* There is no chance of overlap */
-        if ((chunk_begin > stop) ||
-            ((chunk_begin + chunk_size) < start))
-                goto out;
+	if (old->list[o].err > 0 || new->list[n].err > 0)
+		return 0;
 
-        if ((chunk_begin <= start) &&
-            ((chunk_begin + chunk_size) <= stop)) {
-                overlap = ((chunk_begin + chunk_size) - start);
-                goto out;
+        if (old->list[o].start == old->list[o].stop) {
+                return 0;
         }
 
-        if ((chunk_begin <= start) &&
-            ((chunk_begin + chunk_size) >= stop)) {
-                overlap = (stop - start);
-                goto out;
+        if (new->list[n].start == new->list[n].stop) {
+                return 0;
         }
 
-        if ((chunk_begin < stop) &&
-            ((chunk_begin + chunk_size) >= stop)) {
-                overlap = (stop - chunk_begin);
-                goto out;
-        }
+	if ((old->list[o].start > new->list[n].stop) ||
+	    (old->list[o].stop < new->list[n].start))
+		return 0;
 
-out:
-        return overlap;
+	return min (old->list[o].stop, new->list[n].stop) -
+	        max (old->list[o].start, new->list[n].start) + 1;
 }
 
+
 int
 dht_selfheal_dir_finish (call_frame_t *frame, xlator_t *this, int ret)
 {
@@ -82,7 +70,7 @@ dht_selfheal_dir_finish (call_frame_t *frame, xlator_t *this, int ret)
 
         local = frame->local;
         local->selfheal.dir_cbk (frame, NULL, frame->this, ret,
-                                 local->op_errno);
+                                 local->op_errno, NULL);
 
         return 0;
 }
@@ -90,7 +78,7 @@ dht_selfheal_dir_finish (call_frame_t *frame, xlator_t *this, int ret)
 
 int
 dht_selfheal_dir_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                            int op_ret, int op_errno)
+                            int op_ret, int op_errno, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
         call_frame_t *prev = NULL;
@@ -129,18 +117,33 @@ dht_selfheal_dir_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 int
 dht_selfheal_dir_xattr_persubvol (call_frame_t *frame, loc_t *loc,
-                                  dht_layout_t *layout, int i)
+                                  dht_layout_t *layout, int i,
+                                  xlator_t *req_subvol)
 {
         xlator_t          *subvol = NULL;
         dict_t            *xattr = NULL;
         int                ret = 0;
         xlator_t          *this = NULL;
         int32_t           *disk_layout = NULL;
+        dht_local_t       *local = NULL;
+        dht_conf_t        *conf = NULL;
+        data_t            *data = NULL;
 
-
-        subvol = layout->list[i].xlator;
+        local = frame->local;
+        if (req_subvol)
+                subvol = req_subvol;
+        else
+                subvol = layout->list[i].xlator;
         this = frame->this;
 
+        GF_VALIDATE_OR_GOTO ("", this, err);
+        GF_VALIDATE_OR_GOTO (this->name, layout, err);
+        GF_VALIDATE_OR_GOTO (this->name, local, err);
+        GF_VALIDATE_OR_GOTO (this->name, subvol, err);
+        VALIDATE_OR_GOTO (this->private, err);
+
+        conf = this->private;
+
         xattr = get_new_dict ();
         if (!xattr) {
                 goto err;
@@ -154,8 +157,7 @@ dht_selfheal_dir_xattr_persubvol (call_frame_t *frame, loc_t *loc,
                 goto err;
         }
 
-        ret = dict_set_bin (xattr, "trusted.glusterfs.dht",
-                            disk_layout, 4 * 4);
+        ret = dict_set_bin (xattr, conf->xattr_name, disk_layout, 4 * 4);
         if (ret == -1) {
                 gf_log (this->name, GF_LOG_WARNING,
                         "%s: (subvol %s) failed to set xattr dictionary",
@@ -170,10 +172,22 @@ dht_selfheal_dir_xattr_persubvol (call_frame_t *frame, loc_t *loc,
                 layout->type, subvol->name, loc->path);
 
         dict_ref (xattr);
+        if (local->xattr) {
+                data = dict_get (local->xattr, QUOTA_LIMIT_KEY);
+                if (data) {
+                        ret = dict_add (xattr, QUOTA_LIMIT_KEY, data);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR, "Failed to "
+                                        "set quota limit key on %s",loc->path);
+                        }
+                }
+        }
+        if (!uuid_is_null (local->gfid))
+                uuid_copy (loc->gfid, local->gfid);
 
         STACK_WIND (frame, dht_selfheal_dir_xattr_cbk,
                     subvol, subvol->fops->setxattr,
-                    loc, xattr, 0);
+                    loc, xattr, 0, NULL);
 
         dict_unref (xattr);
 
@@ -183,11 +197,10 @@ err:
         if (xattr)
                 dict_destroy (xattr);
 
-        if (disk_layout)
-                GF_FREE (disk_layout);
+        GF_FREE (disk_layout);
 
         dht_selfheal_dir_xattr_cbk (frame, subvol, frame->this,
-                                    -1, ENOMEM);
+                                    -1, ENOMEM, NULL);
         return 0;
 }
 
@@ -198,21 +211,42 @@ dht_fix_dir_xattr (call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
         int          i = 0;
         int          count = 0;
         xlator_t    *this = NULL;
+        dht_conf_t  *conf = NULL;
+        dht_layout_t *dummy = NULL;
 
         local = frame->local;
         this = frame->this;
+        conf = this->private;
 
         gf_log (this->name, GF_LOG_DEBUG,
                 "writing the new range for all subvolumes");
 
-        local->call_cnt = count = layout->cnt;
+        local->call_cnt = count = conf->subvolume_cnt;
 
         for (i = 0; i < layout->cnt; i++) {
-                dht_selfheal_dir_xattr_persubvol (frame, loc, layout, i);
+                dht_selfheal_dir_xattr_persubvol (frame, loc, layout, i, NULL);
 
                 if (--count == 0)
-                        break;
+                        goto out;
+        }
+        /* if we are here, subvolcount > layout_count. subvols-per-directory
+         * option might be set here. We need to clear out layout from the
+         * non-participating subvolumes, else it will result in overlaps */
+        dummy = dht_layout_new (this, 1);
+        if (!dummy)
+                goto out;
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (_gf_false ==
+                    dht_is_subvol_in_layout (layout, conf->subvolumes[i])) {
+                        dht_selfheal_dir_xattr_persubvol (frame, loc, dummy, 0,
+                                                          conf->subvolumes[i]);
+                        if (--count == 0)
+                                break;
+                }
         }
+
+        dht_layout_unref (this, dummy);
+out:
         return 0;
 }
 
@@ -223,14 +257,17 @@ dht_selfheal_dir_xattr (call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
         int          missing_xattr = 0;
         int          i = 0;
         xlator_t    *this = NULL;
+        dht_conf_t   *conf = NULL;
+        dht_layout_t *dummy = NULL;
 
         local = frame->local;
         this = frame->this;
+        conf = this->private;
 
         for (i = 0; i < layout->cnt; i++) {
                 if (layout->list[i].err != -1 || !layout->list[i].stop) {
                         /* err != -1 would mean xattr present on the directory
-                         * or the directory is itself non existant.
+                         * or the directory is non existent.
                          * !layout->list[i].stop would mean layout absent
                          */
 
@@ -238,7 +275,14 @@ dht_selfheal_dir_xattr (call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
                 }
                 missing_xattr++;
         }
-
+        /* Also account for subvolumes with no-layout. Used for zero'ing out
+         * the layouts and for setting quota key's if present */
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (_gf_false ==
+                    dht_is_subvol_in_layout (layout, conf->subvolumes[i])) {
+                        missing_xattr++;
+                }
+        }
         gf_log (this->name, GF_LOG_TRACE,
                 "%d subvolumes missing xattr for %s",
                 missing_xattr, loc->path);
@@ -249,23 +293,36 @@ dht_selfheal_dir_xattr (call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
         }
 
         local->call_cnt = missing_xattr;
-
         for (i = 0; i < layout->cnt; i++) {
                 if (layout->list[i].err != -1 || !layout->list[i].stop)
                         continue;
 
-                dht_selfheal_dir_xattr_persubvol (frame, loc, layout, i);
+                dht_selfheal_dir_xattr_persubvol (frame, loc, layout, i, NULL);
 
                 if (--missing_xattr == 0)
                         break;
         }
+        dummy = dht_layout_new (this, 1);
+        if (!dummy)
+                goto out;
+        for (i = 0; i < conf->subvolume_cnt && missing_xattr; i++) {
+                if (_gf_false ==
+                    dht_is_subvol_in_layout (layout, conf->subvolumes[i])) {
+                        dht_selfheal_dir_xattr_persubvol (frame, loc, dummy, 0,
+                                                          conf->subvolumes[i]);
+                        missing_xattr--;
+                }
+        }
+
+        dht_layout_unref (this, dummy);
+out:
         return 0;
 }
 
 int
 dht_selfheal_dir_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                               int op_ret, int op_errno, struct iatt *statpre,
-                              struct iatt *statpost)
+                              struct iatt *statpost, dict_t *xdata)
 {
         dht_local_t   *local = NULL;
         dht_layout_t  *layout = NULL;
@@ -306,6 +363,9 @@ dht_selfheal_dir_setattr (call_frame_t *frame, loc_t *loc, struct iatt *stbuf,
                 return 0;
         }
 
+        if (!uuid_is_null (local->gfid))
+                uuid_copy (loc->gfid, local->gfid);
+
         local->call_cnt = missing_attr;
         for (i = 0; i < layout->cnt; i++) {
                 if (layout->list[i].err == -1) {
@@ -316,7 +376,7 @@ dht_selfheal_dir_setattr (call_frame_t *frame, loc_t *loc, struct iatt *stbuf,
                         STACK_WIND (frame, dht_selfheal_dir_setattr_cbk,
                                     layout->list[i].xlator,
                                     layout->list[i].xlator->fops->setattr,
-                                    loc, stbuf, valid);
+                                    loc, stbuf, valid, NULL);
                 }
         }
 
@@ -327,7 +387,8 @@ int
 dht_selfheal_dir_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                             int op_ret, int op_errno,
                             inode_t *inode, struct iatt *stbuf,
-                            struct iatt *preparent, struct iatt *postparent)
+                            struct iatt *preparent, struct iatt *postparent,
+                            dict_t *xdata)
 {
         dht_local_t   *local = NULL;
         dht_layout_t  *layout = NULL;
@@ -373,6 +434,46 @@ out:
         return 0;
 }
 
+void
+dht_selfheal_dir_mkdir_setacl (dict_t *xattr, dict_t *dict)
+{
+        data_t          *acl_default = NULL;
+        data_t          *acl_access = NULL;
+        xlator_t        *this = NULL;
+        int     ret = -1;
+
+        GF_ASSERT (xattr);
+        GF_ASSERT (dict);
+
+        this = THIS;
+        GF_ASSERT (this);
+
+        acl_default = dict_get (xattr, POSIX_ACL_DEFAULT_XATTR);
+
+        if (!acl_default) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "ACL_DEFAULT xattr not present");
+                goto cont;
+        }
+        ret = dict_set (dict, POSIX_ACL_DEFAULT_XATTR, acl_default);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Could not set ACL_DEFAULT xattr");
+cont:
+        acl_access = dict_get (xattr, POSIX_ACL_ACCESS_XATTR);
+        if (!acl_access) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "ACL_ACCESS xattr not present");
+                goto out;
+        }
+        ret = dict_set (dict, POSIX_ACL_ACCESS_XATTR, acl_access);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Could not set ACL_ACCESS xattr");
+
+out:
+        return;
+}
 
 int
 dht_selfheal_dir_mkdir (call_frame_t *frame, loc_t *loc,
@@ -406,16 +507,19 @@ dht_selfheal_dir_mkdir (call_frame_t *frame, loc_t *loc,
 
                 ret = dict_set_static_bin (dict, "gfid-req", local->gfid, 16);
                 if (ret)
-                        gf_log (this->name, GF_LOG_INFO,
+                        gf_log (this->name, GF_LOG_WARNING,
                                 "%s: failed to set gfid in dict", loc->path);
         } else if (local->params) {
                 /* Send the dictionary from higher layers directly */
                 dict = dict_ref (local->params);
         }
+        /* Set acls */
+        if (local->xattr && dict)
+                dht_selfheal_dir_mkdir_setacl (local->xattr, dict);
 
         if (!dict)
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "dict is NULL, need to make sure gfid's are same");
+                gf_log (this->name, GF_LOG_WARNING,
+                        "dict is NULL, need to make sure gfids are same");
 
         for (i = 0; i < layout->cnt; i++) {
                 if (layout->list[i].err == ENOENT || force) {
@@ -429,7 +533,7 @@ dht_selfheal_dir_mkdir (call_frame_t *frame, loc_t *loc,
                                     loc,
                                     st_mode_from_ia (local->stbuf.ia_prot,
                                                      local->stbuf.ia_type),
-                                    dict);
+                                    0, dict);
                 }
         }
 
@@ -448,7 +552,7 @@ dht_selfheal_layout_alloc_start (xlator_t *this, loc_t *loc,
         uint32_t      hashval = 0;
         int           ret = 0;
 
-        ret = dht_hash_compute (layout->type, loc->path, &hashval);
+        ret = dht_hash_compute (this, layout->type, loc->path, &hashval);
         if (ret == 0) {
                 start = (hashval % layout->cnt);
         }
@@ -460,14 +564,52 @@ static inline int
 dht_get_layout_count (xlator_t *this, dht_layout_t *layout, int new_layout)
 {
         int i = 0;
+        int j = 0;
         int err = 0;
         int count = 0;
+        dht_conf_t *conf = NULL;
+
+        /* Gets in use only for replace-brick, remove-brick */
+        conf = this->private;
+        for (i = 0; i < layout->cnt; i++) {
+                for (j = 0; j < conf->subvolume_cnt; j++) {
+                        if (conf->decommissioned_bricks[j] &&
+                            conf->decommissioned_bricks[j] == layout->list[i].xlator) {
+                                layout->list[i].err = EINVAL;
+                                break;
+                        }
+                }
+        }
 
         for (i = 0; i < layout->cnt; i++) {
                 err = layout->list[i].err;
-                if (err == -1 || err == 0) {
-                        layout->list[i].err = -1;
+                if (err == -1 || err == 0 || err == ENOENT) {
+			/* Setting list[i].err = -1 is an indication for
+			   dht_selfheal_layout_new_directory() to assign
+			   a range. We set it to -1 based on any one of
+			   the three criteria:
+
+			   - err == -1 already, which means directory
+			     existed but layout was not set on it.
+
+			   - err == 0, which means directory exists and
+			     has an old layout piece which will be
+			     overwritten now.
+
+			   - err == ENOENT, which means directory does
+			     not exist (possibly racing with mkdir or
+			     finishing half done mkdir). The missing
+			     directory will be attempted to be recreated.
+
+			     It is important to note that it is safe
+			     to race with mkdir() as self-heal and
+			     mkdir are idempotent operations. Both will
+			     strive to set the directory and layouts to
+			     the same final state.
+			*/
                         count++;
+			if (!err)
+				layout->list[i].err = -1;
                 }
         }
 
@@ -482,49 +624,126 @@ dht_get_layout_count (xlator_t *this, dht_layout_t *layout, int new_layout)
                 }
         }
 
-        count = ((layout->spread_cnt) ? layout->spread_cnt :
-                 ((count) ? count : 1));
+        /* if layout->spread_cnt is set, check if it is <= available
+         * subvolumes (down brick and decommissioned bricks are considered
+         * un-availbale). Else return count (available up bricks) */
+        count = ((layout->spread_cnt &&
+                 (layout->spread_cnt <= count)) ?
+                 layout->spread_cnt : ((count) ? count : 1));
 
         return count;
 }
 
 
+void dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc,
+					dht_layout_t *new_layout);
+
+void dht_layout_entry_swap (dht_layout_t *layout, int i, int j);
+void dht_layout_range_swap (dht_layout_t *layout, int i, int j);
+
+/*
+ * It's a bit icky using local variables in a macro, but it makes the rest
+ * of the code a lot clearer.
+ */
+#define OV_ENTRY(x,y)      table[x*new->cnt+y]
+
+void
+dht_selfheal_layout_maximize_overlap (call_frame_t *frame, loc_t *loc,
+				      dht_layout_t *new, dht_layout_t *old)
+{
+	int           i            = 0;
+	int           j            = 0;
+	uint32_t      curr_overlap = 0;
+	uint32_t      max_overlap  = 0;
+	int           max_overlap_idx = -1;
+	uint32_t      overlap      = 0;
+        uint32_t     *table = NULL;
+
+	dht_layout_sort_volname (old);
+	/* Now both old_layout->list[] and new_layout->list[]
+	   are match the same xlators/subvolumes. i.e,
+	   old_layout->[i] and new_layout->[i] are referring
+	   to the same subvolumes
+	*/
+
+        /* Build a table of overlaps between new[i] and old[j]. */
+        table = alloca(sizeof(overlap)*old->cnt*new->cnt);
+        if (!table) {
+                return;
+        }
+        memset(table,0,sizeof(overlap)*old->cnt*new->cnt);
+        for (i = 0; i < new->cnt; ++i) {
+                for (j = 0; j < old->cnt; ++j) {
+                        OV_ENTRY(i,j) = dht_overlap_calc(old,j,new,i);
+                }
+        }
+
+	for (i = 0; i < new->cnt; i++) {
+		if (new->list[i].err > 0) {
+			/* Subvol might be marked for decommission
+			   with EINVAL, or some other serious error
+			   marked with positive errno.
+			*/
+			continue;
+                }
+
+                max_overlap = 0;
+                max_overlap_idx = i;
+                for (j = (i + 1); j < new->cnt; ++j) {
+                        if (new->list[j].err > 0) {
+			        /* Subvol might be marked for decommission
+			        with EINVAL, or some other serious error
+			        marked with positive errno.
+			        */
+			        continue;
+                        }
+                        /* Calculate the overlap now. */
+                        curr_overlap = OV_ENTRY(i,i) + OV_ENTRY(j,j);
+                        /* Calculate the overlap after the proposed swap. */
+                        overlap = OV_ENTRY(i,j) + OV_ENTRY(j,i);
+                        /* Are we better than status quo? */
+                        if (overlap > curr_overlap) {
+                                overlap -= curr_overlap;
+                                /* Are we better than the previous choice? */
+                                if (overlap > max_overlap) {
+                                        max_overlap = overlap;
+                                        max_overlap_idx = j;
+                                }
+                        }
+                }
+
+		if (max_overlap_idx != i) {
+ 			dht_layout_range_swap (new, i, max_overlap_idx);
+                        /* Need to swap the table values too. */
+                        for (j = 0; j < old->cnt; ++j) {
+                                overlap = OV_ENTRY(i,j);
+                                OV_ENTRY(i,j) = OV_ENTRY(max_overlap_idx,j);
+                                OV_ENTRY(max_overlap_idx,j) = overlap;
+                        }
+                }
+	}
+}
+
+
 dht_layout_t *
 dht_fix_layout_of_directory (call_frame_t *frame, loc_t *loc,
                              dht_layout_t *layout)
 {
-        uint32_t      chunk        = 0;
-        uint32_t      start        = 0;
-        uint32_t      stop         = 0;
-        uint32_t      overlap      = 0;
-        uint32_t      max_overlap  = 0;
-        uint32_t      chunk_begin  = 0;
-        int           count        = 0;
-        int           cnt          = 0;
         int           i            = 0;
-        int           j            = 0;
-        int           k            = 0;
-        int           loop_cnt     = 0;
-        int           start_subvol = 0;
-        int          *fix_array    = NULL;
         xlator_t     *this         = NULL;
         dht_layout_t *new_layout   = NULL;
         dht_conf_t   *priv         = NULL;
         dht_local_t  *local        = NULL;
+        uint32_t      subvol_down  = 0;
+        int           ret          = 0;
 
         this  = frame->this;
         priv  = this->private;
         local = frame->local;
 
-        count = cnt = dht_get_layout_count (this, layout, 0);
-
-        chunk = ((unsigned long) 0xffffffff) / ((cnt) ? cnt : 1);
-
-        start_subvol = dht_selfheal_layout_alloc_start (this, loc, layout);
-
-        fix_array = GF_CALLOC (sizeof (int), layout->cnt, gf_common_mt_char);
-        if (!fix_array) {
-                /* No fix, use the existing layout itself */
+        if (layout->type == DHT_HASH_TYPE_DM_USER) {
+                gf_log (THIS->name, GF_LOG_DEBUG, "leaving %s alone",
+                        loc->path);
                 goto done;
         }
 
@@ -532,98 +751,33 @@ dht_fix_layout_of_directory (call_frame_t *frame, loc_t *loc,
         if (!new_layout)
                 goto done;
 
-        for (i = 0; i < new_layout->cnt; i++) {
-                /* TODO: fix this in layout_alloc() itself */
-                new_layout->list[i].err = -ENOENT;
-                if (i < layout->cnt)
-                        new_layout->list[i].xlator = layout->list[i].xlator;
-        }
+        /* If a subvolume is down, do not re-write the layout. */
+        ret = dht_layout_anomalies (this, loc, layout, NULL, NULL, NULL,
+                                    &subvol_down, NULL, NULL);
 
-        /* Check if there are any overlap in layout, and give the proper fix */
-        for (i = 0; i < layout->cnt; i++) {
-                /* No need to fix if 'err' is not '-1' */
-                if (layout->list[i].err != -1)
-                        continue;
-
-                /* If already existing layout is having no range, skip it */
-                start = layout->list[i].start;
-                stop  = layout->list[i].stop;
-                if ((stop - start) == 0)
-                        continue;
-
-                max_overlap = 0;
-
-                /* 'j' is used as starting point of each chunk */
-                for (j = 1; j <= count; j++) {
-                        /* if chunk is already used, don't use it again */
-                        for (k = 0; k < i; k++)
-                                if (j == fix_array[k])
-                                        break;
-                        if (k < i)
-                                continue;
-
-                        overlap = dht_find_overlap (i, (j-1), start, stop, chunk);
-                        if (max_overlap < overlap) {
-                                max_overlap = overlap;
-                                fix_array[i] = j;
-                        }
-                }
+        if (subvol_down || (ret == -1)) {
+                gf_log (this->name, GF_LOG_WARNING, "%u subvolume(s) are down"
+                        ". Skipping fix layout.", subvol_down);
+                GF_FREE (new_layout);
+                return NULL;
+        }
 
-                /* If we have any overlap, then use that itself as new
-                   layout for the subvolume */
-                if (fix_array[i]) {
-                        chunk_begin = chunk * (fix_array[i] - 1);
-                        new_layout->list[i].err = -1;
-                        DHT_SET_LAYOUT_RANGE (new_layout, i, chunk_begin,
-                                              chunk, cnt, loc->path);
-                        /* make sure to give (max - 1) as 'stop' range,
-                           if it is last chunk */
-                        if (fix_array[i] == count)
-                                new_layout->list[i].stop = 0xffffffff;
-                        if (--cnt == 0)
-                                goto done;
+        for (i = 0; i < new_layout->cnt; i++) {
+		if (layout->list[i].err != ENOSPC)
+			new_layout->list[i].err = layout->list[i].err;
+		else
+			new_layout->list[i].err = -1;
 
-                }
+		new_layout->list[i].xlator = layout->list[i].xlator;
         }
 
-        /* Now, look for layouts which are not having any overlaps
-           and give it a fix */
-        for (loop_cnt = 0, i = start_subvol; loop_cnt < new_layout->cnt;
-             i++, loop_cnt++) {
-                if (i == new_layout->cnt)
-                        i = 0;
+	/* First give it a layout as though it is a new directory. This
+	   ensures rotation to kick in */
+        dht_layout_sort_volname (new_layout);
+	dht_selfheal_layout_new_directory (frame, loc, new_layout);
 
-                /* If 'fix_array[i]' is set, the layout is already fixed. */
-                if (fix_array[i])
-                        continue;
-
-                if (layout->list[i].err != -1) {
-                        new_layout->list[i].err = layout->list[i].err;
-                        continue;
-                }
-
-                for (k = 1; k <= count; k++) {
-                        for (j = 0; j < new_layout->cnt; j++) {
-                                if (k == fix_array[j])
-                                        break;
-                        }
-                        /* Didn't find any of the list begining with 'k' */
-                        if (j == new_layout->cnt)
-                                break;
-                }
-
-                fix_array[i] = k;
-                chunk_begin = (k - 1) * chunk;
-                new_layout->list[i].err = -1;
-                DHT_SET_LAYOUT_RANGE (new_layout, i, chunk_begin, chunk, cnt,
-                                      loc->path);
-                /* make sure to give (max - 1) as 'stop' range,
-                   if it is last chunk */
-                if (k == count)
-                        new_layout->list[i].stop = 0xffffffff;
-                if (--cnt == 0)
-                        goto done;
-        }
+	/* Now selectively re-assign ranges only when it helps */
+	dht_selfheal_layout_maximize_overlap (frame, loc, new_layout, layout);
 
 done:
         if (new_layout) {
@@ -637,7 +791,7 @@ done:
                 local->layout = new_layout;
         }
 
-        return new_layout;
+        return local->layout;
 }
 
 
@@ -661,9 +815,11 @@ dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc,
 
         start_subvol = dht_selfheal_layout_alloc_start (this, loc, layout);
 
+        /* clear out the range, as we are re-computing here */
+        DHT_RESET_LAYOUT_RANGE (layout);
         for (i = start_subvol; i < layout->cnt; i++) {
                 err = layout->list[i].err;
-                if (err == -1) {
+                if (err == -1 || err == ENOENT) {
                         DHT_SET_LAYOUT_RANGE(layout, i, start, chunk,
                                              cnt, loc->path);
                         if (--cnt == 0) {
@@ -676,7 +832,7 @@ dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc,
 
         for (i = 0; i < start_subvol; i++) {
                 err = layout->list[i].err;
-                if (err == -1) {
+                if (err == -1 || err == ENOENT) {
                         DHT_SET_LAYOUT_RANGE(layout, i, start, chunk,
                                              cnt, loc->path);
                         if (--cnt == 0) {
@@ -695,35 +851,17 @@ int
 dht_selfheal_dir_getafix (call_frame_t *frame, loc_t *loc,
                           dht_layout_t *layout)
 {
-        dht_conf_t  *conf = NULL;
-        xlator_t    *this = NULL;
         dht_local_t *local = NULL;
-        int          missing = -1;
-        int          down = -1;
-        int          holes = -1;
+        uint32_t     holes = 0;
         int          ret = -1;
         int          i = -1;
-        int          overlaps = -1;
+        uint32_t     overlaps = 0;
 
-        this = frame->this;
-        conf = this->private;
         local = frame->local;
 
-        missing = local->selfheal.missing;
-        down = local->selfheal.down;
         holes = local->selfheal.hole_cnt;
         overlaps = local->selfheal.overlaps_cnt;
 
-        if ((missing + down) == conf->subvolume_cnt) {
-                dht_selfheal_layout_new_directory (frame, loc, layout);
-                ret = 0;
-        }
-
-        if (holes <= down) {
-                /* the down subvol might fill up the holes */
-                ret = 0;
-        }
-
         if (holes || overlaps) {
                 dht_selfheal_layout_new_directory (frame, loc, layout);
                 ret = 0;
@@ -775,6 +913,9 @@ dht_fix_directory_layout (call_frame_t *frame,
 
         /* No layout sorting required here */
         tmp_layout = dht_fix_layout_of_directory (frame, &local->loc, layout);
+        if (!tmp_layout) {
+                return -1;
+        }
         dht_fix_dir_xattr (frame, &local->loc, tmp_layout);
 
         return 0;
@@ -786,7 +927,6 @@ dht_selfheal_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
                         loc_t *loc, dht_layout_t *layout)
 {
         dht_local_t *local    = NULL;
-        uint32_t     holes    = 0;
         uint32_t     down     = 0;
         uint32_t     misc     = 0;
         int          ret      = 0;
@@ -798,11 +938,9 @@ dht_selfheal_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
         dht_layout_anomalies (this, loc, layout,
                               &local->selfheal.hole_cnt,
                               &local->selfheal.overlaps_cnt,
-                              &local->selfheal.missing,
-                              &local->selfheal.down,
-                              &local->selfheal.misc);
+                              NULL, &local->selfheal.down,
+                              &local->selfheal.misc, NULL);
 
-        holes    = local->selfheal.hole_cnt;
         down     = local->selfheal.down;
         misc     = local->selfheal.misc;
 
@@ -810,14 +948,14 @@ dht_selfheal_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
         local->selfheal.layout = dht_layout_ref (this, layout);
 
         if (down) {
-                gf_log (this->name, GF_LOG_INFO,
+                gf_log (this->name, GF_LOG_WARNING,
                         "%d subvolumes down -- not fixing", down);
                 ret = 0;
                 goto sorry_no_fix;
         }
 
         if (misc) {
-                gf_log (this->name, GF_LOG_INFO,
+                gf_log (this->name, GF_LOG_WARNING,
                         "%d subvolumes have unrecoverable errors", misc);
                 ret = 0;
                 goto sorry_no_fix;
@@ -827,7 +965,7 @@ dht_selfheal_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
         ret = dht_selfheal_dir_getafix (frame, loc, layout);
 
         if (ret == -1) {
-                gf_log (this->name, GF_LOG_INFO,
+                gf_log (this->name, GF_LOG_WARNING,
                         "not able to form layout for the directory");
                 goto sorry_no_fix;
         }
@@ -860,3 +998,51 @@ dht_selfheal_restore (call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
 
         return ret;
 }
+
+int
+dht_dir_attr_heal (void *data)
+{
+        call_frame_t    *frame = NULL;
+        dht_local_t     *local = NULL;
+        xlator_t        *subvol = NULL;
+        xlator_t        *this  = NULL;
+        dht_conf_t      *conf  = NULL;
+        int              call_cnt = 0;
+        int              ret   = -1;
+        int              i     = 0;
+
+        GF_VALIDATE_OR_GOTO ("dht", data, out);
+
+        frame = data;
+        local = frame->local;
+        this = frame->this;
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", local, out);
+        conf = this->private;
+        GF_VALIDATE_OR_GOTO ("dht", conf, out);
+
+        call_cnt = conf->subvolume_cnt;
+
+        for (i = 0; i < call_cnt; i++) {
+                subvol = conf->subvolumes[i];
+                if (!subvol || (subvol == dht_first_up_subvol (this)))
+                        continue;
+                ret = syncop_setattr (subvol, &local->loc, &local->stbuf,
+                                      (GF_SET_ATTR_UID | GF_SET_ATTR_GID),
+                                      NULL, NULL);
+                if (ret) {
+                        gf_log ("dht", GF_LOG_ERROR, "Failed to set uid/gid on"
+                                " %s on %s subvol (%s)", local->loc.path,
+                                subvol->name, strerror (-ret));
+                }
+        }
+out:
+        return 0;
+}
+
+int
+dht_dir_attr_heal_done (int ret, call_frame_t *sync_frame, void *data)
+{
+        DHT_STACK_DESTROY (sync_frame);
+        return 0;
+}
diff --git a/xlators/cluster/dht/src/dht-shared.c b/xlators/cluster/dht/src/dht-shared.c
new file mode 100644
index 000000000..f2e7467ab
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-shared.c
@@ -0,0 +1,780 @@
+/*
+  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 _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+/* TODO: add NS locking */
+
+#include "statedump.h"
+#include "dht-common.h"
+
+/* TODO:
+   - use volumename in xattr instead of "dht"
+   - use NS locks
+   - handle all cases in self heal layout reconstruction
+   - complete linkfile selfheal
+*/
+struct volume_options options[];
+
+void
+dht_layout_dump (dht_layout_t  *layout, const char *prefix)
+{
+
+        char    key[GF_DUMP_MAX_BUF_LEN];
+        int     i = 0;
+
+        if (!layout)
+                goto out;
+        if (!prefix)
+                goto out;
+
+        gf_proc_dump_build_key(key, prefix, "cnt");
+        gf_proc_dump_write(key, "%d", layout->cnt);
+        gf_proc_dump_build_key(key, prefix, "preset");
+        gf_proc_dump_write(key, "%d", layout->preset);
+        gf_proc_dump_build_key(key, prefix, "gen");
+        gf_proc_dump_write(key, "%d", layout->gen);
+        if (layout->type != IA_INVAL) {
+                gf_proc_dump_build_key(key, prefix, "inode type");
+                gf_proc_dump_write(key, "%d", layout->type);
+        }
+
+        if  (!IA_ISDIR (layout->type))
+                goto out;
+
+        for (i = 0; i < layout->cnt; i++) {
+                gf_proc_dump_build_key(key, prefix,"list[%d].err", i);
+                gf_proc_dump_write(key, "%d", layout->list[i].err);
+                gf_proc_dump_build_key(key, prefix,"list[%d].start", i);
+                gf_proc_dump_write(key, "%u", layout->list[i].start);
+                gf_proc_dump_build_key(key, prefix,"list[%d].stop", i);
+                gf_proc_dump_write(key, "%u", layout->list[i].stop);
+                if (layout->list[i].xlator) {
+                        gf_proc_dump_build_key(key, prefix,
+                                               "list[%d].xlator.type", i);
+                        gf_proc_dump_write(key, "%s",
+                                           layout->list[i].xlator->type);
+                        gf_proc_dump_build_key(key, prefix,
+                                               "list[%d].xlator.name", i);
+                        gf_proc_dump_write(key, "%s",
+                                           layout->list[i].xlator->name);
+                }
+        }
+
+out:
+        return;
+}
+
+
+int32_t
+dht_priv_dump (xlator_t *this)
+{
+        char            key_prefix[GF_DUMP_MAX_BUF_LEN];
+        char            key[GF_DUMP_MAX_BUF_LEN];
+        int             i = 0;
+        dht_conf_t      *conf = NULL;
+        int             ret = -1;
+
+        if (!this)
+                goto out;
+
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        ret = TRY_LOCK(&conf->subvolume_lock);
+        if (ret != 0) {
+                return ret;
+        }
+
+        gf_proc_dump_add_section("xlator.cluster.dht.%s.priv", this->name);
+        gf_proc_dump_build_key(key_prefix,"xlator.cluster.dht","%s.priv",
+                               this->name);
+        gf_proc_dump_write("subvol_cnt","%d", conf->subvolume_cnt);
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                snprintf (key, sizeof (key), "subvolumes[%d]", i);
+                gf_proc_dump_write(key, "%s.%s", conf->subvolumes[i]->type,
+                                   conf->subvolumes[i]->name);
+                if (conf->file_layouts && conf->file_layouts[i]){
+                        snprintf (key, sizeof (key), "file_layouts[%d]", i);
+                        dht_layout_dump(conf->file_layouts[i], key);
+                }
+                if (conf->dir_layouts && conf->dir_layouts[i]) {
+                        snprintf (key, sizeof (key), "dir_layouts[%d]", i);
+                        dht_layout_dump(conf->dir_layouts[i], key);
+                }
+                if (conf->subvolume_status) {
+
+                        snprintf (key, sizeof (key), "subvolume_status[%d]", i);
+                        gf_proc_dump_write(key, "%d",
+                                           (int)conf->subvolume_status[i]);
+                }
+
+        }
+
+        gf_proc_dump_write("search_unhashed", "%d", conf->search_unhashed);
+        gf_proc_dump_write("gen", "%d", conf->gen);
+        gf_proc_dump_write("min_free_disk", "%lf", conf->min_free_disk);
+	gf_proc_dump_write("min_free_inodes", "%lf", conf->min_free_inodes);
+        gf_proc_dump_write("disk_unit", "%c", conf->disk_unit);
+        gf_proc_dump_write("refresh_interval", "%d", conf->refresh_interval);
+        gf_proc_dump_write("unhashed_sticky_bit", "%d", conf->unhashed_sticky_bit);
+
+        if (conf->du_stats) {
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                        if (!conf->subvolume_status[i])
+                                continue;
+
+                        snprintf (key, sizeof (key), "subvolumes[%d]", i);
+                        gf_proc_dump_write (key, "%s",
+                                            conf->subvolumes[i]->name);
+
+                        snprintf (key, sizeof (key),
+                                  "du_stats[%d].avail_percent", i);
+                        gf_proc_dump_write (key, "%lf",
+                                            conf->du_stats[i].avail_percent);
+
+                        snprintf (key, sizeof (key), "du_stats[%d].avail_space",
+                                  i);
+                        gf_proc_dump_write (key, "%lu",
+                                            conf->du_stats[i].avail_space);
+
+                        snprintf (key, sizeof (key),
+                                  "du_stats[%d].avail_inodes", i);
+                        gf_proc_dump_write (key, "%lf",
+                                            conf->du_stats[i].avail_inodes);
+
+                        snprintf (key, sizeof (key), "du_stats[%d].log", i);
+                        gf_proc_dump_write (key, "%lu",
+                                            conf->du_stats[i].log);
+                }
+        }
+
+        if (conf->last_stat_fetch.tv_sec)
+                gf_proc_dump_write("last_stat_fetch", "%s",
+                                    ctime(&conf->last_stat_fetch.tv_sec));
+
+        UNLOCK(&conf->subvolume_lock);
+
+out:
+        return ret;
+}
+
+int32_t
+dht_inodectx_dump (xlator_t *this, inode_t *inode)
+{
+        int             ret = -1;
+        dht_layout_t    *layout = NULL;
+
+        if (!this)
+                goto out;
+        if (!inode)
+                goto out;
+
+        ret = dht_inode_ctx_layout_get (inode, this, &layout);
+
+        if ((ret != 0) || !layout)
+                return ret;
+
+        gf_proc_dump_add_section("xlator.cluster.dht.%s.inode", this->name);
+        dht_layout_dump(layout, "layout");
+
+out:
+        return ret;
+}
+
+void
+dht_fini (xlator_t *this)
+{
+        int         i = 0;
+        dht_conf_t *conf = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+
+        conf = this->private;
+        this->private = NULL;
+        if (conf) {
+                if (conf->file_layouts) {
+                        for (i = 0; i < conf->subvolume_cnt; i++) {
+                                GF_FREE (conf->file_layouts[i]);
+                        }
+                        GF_FREE (conf->file_layouts);
+                }
+
+                GF_FREE (conf->subvolumes);
+
+                GF_FREE (conf->subvolume_status);
+
+                GF_FREE (conf);
+        }
+out:
+        return;
+}
+
+int32_t
+mem_acct_init (xlator_t *this)
+{
+        int     ret = -1;
+
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+
+        ret = xlator_mem_acct_init (this, gf_dht_mt_end + 1);
+
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
+                        "failed");
+                return ret;
+        }
+out:
+        return ret;
+}
+
+
+int
+dht_parse_decommissioned_bricks (xlator_t *this, dht_conf_t *conf,
+                                 const char *bricks)
+{
+        int         i  = 0;
+        int         ret  = -1;
+        char       *tmpstr = NULL;
+        char       *dup_brick = NULL;
+        char       *node = NULL;
+
+        if (!conf || !bricks)
+                goto out;
+
+        dup_brick = gf_strdup (bricks);
+        node = strtok_r (dup_brick, ",", &tmpstr);
+        while (node) {
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                        if (!strcmp (conf->subvolumes[i]->name, node)) {
+                                conf->decommissioned_bricks[i] =
+                                        conf->subvolumes[i];
+                                        conf->decommission_subvols_cnt++;
+                                gf_log (this->name, GF_LOG_INFO,
+                                        "decommissioning subvolume %s",
+                                        conf->subvolumes[i]->name);
+                                break;
+                        }
+                }
+                if (i == conf->subvolume_cnt) {
+                        /* Wrong node given. */
+                        goto out;
+                }
+                node = strtok_r (NULL, ",", &tmpstr);
+        }
+
+        ret = 0;
+        conf->decommission_in_progress = 1;
+out:
+        GF_FREE (dup_brick);
+
+        return ret;
+}
+
+
+int
+dht_decommissioned_remove (xlator_t *this, dht_conf_t *conf)
+{
+        int         i  = 0;
+        int         ret  = -1;
+
+        if (!conf)
+                goto out;
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (conf->decommissioned_bricks[i]) {
+                        conf->decommissioned_bricks[i] = NULL;
+                        conf->decommission_subvols_cnt--;
+                }
+        }
+
+        ret = 0;
+out:
+
+        return ret;
+}
+void
+dht_init_regex (xlator_t *this, dict_t *odict, char *name,
+                regex_t *re, gf_boolean_t *re_valid)
+{
+        char    *temp_str;
+
+        if (dict_get_str (odict, name, &temp_str) != 0) {
+                if (strcmp(name,"rsync-hash-regex")) {
+                        return;
+                }
+                temp_str = "^\\.(.+)\\.[^.]+$";
+        }
+
+        if (*re_valid) {
+                regfree(re);
+                *re_valid = _gf_false;
+        }
+
+        if (!strcmp(temp_str,"none")) {
+                return;
+        }
+
+        if (regcomp(re,temp_str,REG_EXTENDED) == 0) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "using regex %s = %s", name, temp_str);
+                *re_valid = _gf_true;
+        }
+        else {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "compiling regex %s failed", temp_str);
+        }
+}
+
+int
+dht_reconfigure (xlator_t *this, dict_t *options)
+{
+        dht_conf_t      *conf = NULL;
+        char            *temp_str = NULL;
+        gf_boolean_t     search_unhashed;
+        int              ret = -1;
+
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", options, out);
+
+        conf = this->private;
+        if (!conf)
+                return 0;
+
+        if (dict_get_str (options, "lookup-unhashed", &temp_str) == 0) {
+                /* If option is not "auto", other options _should_ be boolean*/
+                if (strcasecmp (temp_str, "auto")) {
+                        if (!gf_string2boolean (temp_str, &search_unhashed)) {
+                                gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
+                                       " lookup-unhashed reconfigured (%s)",
+                                       temp_str);
+                                conf->search_unhashed = search_unhashed;
+                        } else {
+                                gf_log(this->name, GF_LOG_ERROR, "Reconfigure:"
+                                       " lookup-unhashed should be boolean,"
+                                       " not (%s), defaulting to (%d)",
+                                       temp_str, conf->search_unhashed);
+                                ret = -1;
+                                goto out;
+                        }
+                } else {
+                        gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
+                               " lookup-unhashed reconfigured auto ");
+                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
+                }
+        }
+
+	GF_OPTION_RECONF ("min-free-disk", conf->min_free_disk, options,
+                          percent_or_size, out);
+        /* option can be any one of percent or bytes */
+        conf->disk_unit = 0;
+        if (conf->min_free_disk < 100.0)
+                conf->disk_unit = 'p';
+
+	GF_OPTION_RECONF ("min-free-inodes", conf->min_free_inodes, options,
+                          percent, out);
+
+        GF_OPTION_RECONF ("directory-layout-spread", conf->dir_spread_cnt,
+                          options, uint32, out);
+
+        GF_OPTION_RECONF ("readdir-optimize", conf->readdir_optimize, options,
+                          bool, out);
+        if (conf->defrag) {
+                GF_OPTION_RECONF ("rebalance-stats", conf->defrag->stats,
+                                  options, bool, out);
+        }
+
+        if (dict_get_str (options, "decommissioned-bricks", &temp_str) == 0) {
+                ret = dht_parse_decommissioned_bricks (this, conf, temp_str);
+                if (ret == -1)
+                        goto out;
+        } else {
+                ret = dht_decommissioned_remove (this, conf);
+                if (ret == -1)
+                        goto out;
+        }
+
+        dht_init_regex (this, options, "rsync-hash-regex",
+                        &conf->rsync_regex, &conf->rsync_regex_valid);
+        dht_init_regex (this, options, "extra-hash-regex",
+                        &conf->extra_regex, &conf->extra_regex_valid);
+
+        ret = 0;
+out:
+        return ret;
+}
+
+static int
+gf_defrag_pattern_list_fill (xlator_t *this, gf_defrag_info_t *defrag, char *data)
+{
+        int                    ret = -1;
+        char                  *tmp_str = NULL;
+        char                  *tmp_str1 = NULL;
+        char                  *dup_str = NULL;
+        char                  *num = NULL;
+        char                  *pattern_str = NULL;
+        char                  *pattern = NULL;
+        gf_defrag_pattern_list_t *temp_list = NULL;
+        gf_defrag_pattern_list_t *pattern_list = NULL;
+
+        if (!this || !defrag || !data)
+                goto out;
+
+        /* Get the pattern for pattern list. "pattern:<optional-size>"
+         * eg: *avi, *pdf:10MB, *:1TB
+         */
+        pattern_str = strtok_r (data, ",", &tmp_str);
+        while (pattern_str) {
+                dup_str = gf_strdup (pattern_str);
+                pattern_list = GF_CALLOC (1, sizeof (gf_defrag_pattern_list_t),
+                                        1);
+                if (!pattern_list) {
+                        goto out;
+                }
+                pattern = strtok_r (dup_str, ":", &tmp_str1);
+                num = strtok_r (NULL, ":", &tmp_str1);
+                if (!pattern)
+                        goto out;
+                if (!num) {
+                        if (gf_string2bytesize_uint64(pattern, &pattern_list->size)
+                             == 0) {
+                                pattern = "*";
+                        }
+                } else if (gf_string2bytesize_uint64 (num, &pattern_list->size) != 0) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "invalid number format \"%s\"", num);
+                        goto out;
+                }
+                memcpy (pattern_list->path_pattern, pattern, strlen (dup_str));
+
+                if (!defrag->defrag_pattern)
+                        temp_list = NULL;
+                else
+                        temp_list = defrag->defrag_pattern;
+
+                pattern_list->next = temp_list;
+
+                defrag->defrag_pattern = pattern_list;
+                pattern_list = NULL;
+
+                GF_FREE (dup_str);
+                dup_str = NULL;
+
+                pattern_str = strtok_r (NULL, ",", &tmp_str);
+        }
+
+        ret = 0;
+out:
+        if (ret)
+                GF_FREE (pattern_list);
+        GF_FREE (dup_str);
+
+        return ret;
+}
+
+int
+dht_init (xlator_t *this)
+{
+        dht_conf_t                      *conf           = NULL;
+        char                            *temp_str       = NULL;
+        int                              ret            = -1;
+        int                              i              = 0;
+        gf_defrag_info_t                *defrag         = NULL;
+        int                              cmd            = 0;
+        char                            *node_uuid      = NULL;
+
+
+        GF_VALIDATE_OR_GOTO ("dht", this, err);
+
+        if (!this->children) {
+                gf_log (this->name, GF_LOG_CRITICAL,
+                        "Distribute needs more than one subvolume");
+                return -1;
+        }
+
+        if (!this->parents) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "dangling volume. check volfile");
+        }
+
+        conf = GF_CALLOC (1, sizeof (*conf), gf_dht_mt_dht_conf_t);
+        if (!conf) {
+                goto err;
+        }
+
+        ret = dict_get_int32 (this->options, "rebalance-cmd", &cmd);
+
+        if (cmd) {
+                defrag = GF_CALLOC (1, sizeof (gf_defrag_info_t),
+                                    gf_defrag_info_mt);
+
+                GF_VALIDATE_OR_GOTO (this->name, defrag, err);
+
+                LOCK_INIT (&defrag->lock);
+
+                defrag->is_exiting = 0;
+
+                conf->defrag = defrag;
+
+                ret = dict_get_str (this->options, "node-uuid", &node_uuid);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR, "node-uuid not "
+                                "specified");
+                        goto err;
+                }
+
+                if (uuid_parse (node_uuid, defrag->node_uuid)) {
+                        gf_log (this->name, GF_LOG_ERROR, "Cannot parse "
+                                "glusterd node uuid");
+                        goto err;
+                }
+
+                defrag->cmd = cmd;
+
+                defrag->stats = _gf_false;
+        }
+
+        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
+        if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
+                /* If option is not "auto", other options _should_ be boolean */
+                if (strcasecmp (temp_str, "auto"))
+                        gf_string2boolean (temp_str, &conf->search_unhashed);
+                else
+                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
+        }
+
+        GF_OPTION_INIT ("unhashed-sticky-bit", conf->unhashed_sticky_bit, bool,
+                        err);
+
+        GF_OPTION_INIT ("use-readdirp", conf->use_readdirp, bool, err);
+
+	GF_OPTION_INIT ("min-free-disk", conf->min_free_disk, percent_or_size,
+			err);
+
+        GF_OPTION_INIT ("min-free-inodes", conf->min_free_inodes, percent,
+                        err);
+
+        conf->dir_spread_cnt = conf->subvolume_cnt;
+        GF_OPTION_INIT ("directory-layout-spread", conf->dir_spread_cnt,
+                        uint32, err);
+
+        GF_OPTION_INIT ("assert-no-child-down", conf->assert_no_child_down,
+                        bool, err);
+
+        GF_OPTION_INIT ("readdir-optimize", conf->readdir_optimize, bool, err);
+
+        if (defrag) {
+                GF_OPTION_INIT ("rebalance-stats", defrag->stats, bool, err);
+                if (dict_get_str (this->options, "rebalance-filter", &temp_str)
+                    == 0) {
+                        if (gf_defrag_pattern_list_fill (this, defrag, temp_str)
+                            == -1) {
+                                gf_log (this->name, GF_LOG_ERROR, "Cannot parse"
+                                        " rebalance-filter (%s)", temp_str);
+                                goto err;
+                        }
+                }
+        }
+
+        /* option can be any one of percent or bytes */
+        conf->disk_unit = 0;
+        if (conf->min_free_disk < 100)
+                conf->disk_unit = 'p';
+
+        ret = dht_init_subvolumes (this, conf);
+        if (ret == -1) {
+                goto err;
+        }
+
+        if (dict_get_str (this->options, "decommissioned-bricks", &temp_str) == 0) {
+                ret = dht_parse_decommissioned_bricks (this, conf, temp_str);
+                if (ret == -1)
+                        goto err;
+        }
+
+        dht_init_regex (this, this->options, "rsync-hash-regex",
+                        &conf->rsync_regex, &conf->rsync_regex_valid);
+        dht_init_regex (this, this->options, "extra-hash-regex",
+                        &conf->extra_regex, &conf->extra_regex_valid);
+
+        ret = dht_layouts_init (this, conf);
+        if (ret == -1) {
+                goto err;
+        }
+
+        LOCK_INIT (&conf->subvolume_lock);
+        LOCK_INIT (&conf->layout_lock);
+
+        conf->gen = 1;
+
+        this->local_pool = mem_pool_new (dht_local_t, 512);
+        if (!this->local_pool) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to create local_t's memory pool");
+                goto err;
+        }
+
+        GF_OPTION_INIT ("xattr-name", conf->xattr_name, str, err);
+        gf_asprintf (&conf->link_xattr_name, "%s."DHT_LINKFILE_STR,
+                     conf->xattr_name);
+        gf_asprintf (&conf->wild_xattr_name, "%s*", conf->xattr_name);
+        if (!conf->link_xattr_name || !conf->wild_xattr_name) {
+                goto err;
+        }
+
+        this->private = conf;
+
+        return 0;
+
+err:
+        if (conf) {
+                if (conf->file_layouts) {
+                        for (i = 0; i < conf->subvolume_cnt; i++) {
+                                GF_FREE (conf->file_layouts[i]);
+                        }
+                        GF_FREE (conf->file_layouts);
+                }
+
+                GF_FREE (conf->subvolumes);
+
+                GF_FREE (conf->subvolume_status);
+
+                GF_FREE (conf->du_stats);
+
+                GF_FREE (conf->defrag);
+
+                GF_FREE (conf->xattr_name);
+                GF_FREE (conf->link_xattr_name);
+                GF_FREE (conf->wild_xattr_name);
+
+                GF_FREE (conf);
+        }
+
+        return -1;
+}
+
+
+struct volume_options options[] = {
+        { .key  = {"lookup-unhashed"},
+          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
+                    "on", "off"},
+          .type = GF_OPTION_TYPE_STR,
+          .default_value = "on",
+          .description = "This option if set to ON, does a lookup through "
+          "all the sub-volumes, in case a lookup didn't return any result "
+          "from the hash subvolume. If set to OFF, it does not do a lookup "
+          "on the remaining subvolumes."
+        },
+        { .key  = {"min-free-disk"},
+          .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
+          .default_value = "10%",
+          .description = "Percentage/Size of disk space, after which the "
+          "process starts balancing out the cluster, and logs will appear "
+          "in log files",
+        },
+	{ .key  = {"min-free-inodes"},
+          .type = GF_OPTION_TYPE_PERCENT,
+          .default_value = "5%",
+          .description = "after system has only N% of inodes, warnings "
+          "starts to appear in log files",
+        },
+        { .key = {"unhashed-sticky-bit"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+        },
+        { .key = {"use-readdirp"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "on",
+          .description = "This option if set to ON, forces the use of "
+          "readdirp, and hence also displays the stats of the files."
+        },
+        { .key = {"assert-no-child-down"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+          .description = "This option if set to ON, in the event of "
+          "CHILD_DOWN, will call exit."
+        },
+        { .key  = {"directory-layout-spread"},
+          .type = GF_OPTION_TYPE_INT,
+          .min  = 1,
+          .validate = GF_OPT_VALIDATE_MIN,
+          .description = "Specifies the directory layout spread. Takes number "
+                         "of subvolumes as default value."
+        },
+        { .key  = {"decommissioned-bricks"},
+          .type = GF_OPTION_TYPE_ANY,
+          .description = "This option if set to ON, decommissions "
+          "the brick, so that no new data is allowed to be created "
+          "on that brick."
+        },
+        { .key  = {"rebalance-cmd"},
+          .type = GF_OPTION_TYPE_INT,
+        },
+        { .key = {"node-uuid"},
+          .type = GF_OPTION_TYPE_STR,
+        },
+        { .key = {"rebalance-stats"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+          .description = "This option if set to ON displays and logs the "
+          " time taken for migration of each file, during the rebalance "
+          "process. If set to OFF, the rebalance logs will only display the "
+          "time spent in each directory."
+        },
+        { .key = {"readdir-optimize"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+          .description = "This option if set to ON enables the optimization "
+          "that allows DHT to requests non-first subvolumes to filter out "
+          "directory entries."
+        },
+        { .key = {"rsync-hash-regex"},
+          .type = GF_OPTION_TYPE_STR,
+          /* Setting a default here doesn't work.  See dht_init_regex. */
+          .description = "Regular expression for stripping temporary-file "
+          "suffix and prefix used by rsync, to prevent relocation when the "
+          "file is renamed."
+        },
+        { .key = {"extra-hash-regex"},
+          .type = GF_OPTION_TYPE_STR,
+          /* Setting a default here doesn't work.  See dht_init_regex. */
+          .description = "Regular expression for stripping temporary-file "
+          "suffix and prefix used by an application, to prevent relocation when "
+          "the file is renamed."
+        },
+        { .key = {"rebalance-filter"},
+          .type = GF_OPTION_TYPE_STR,
+        },
+
+        { .key = {"xattr-name"},
+          .type = GF_OPTION_TYPE_STR,
+          .default_value = "trusted.glusterfs.dht",
+          .description = "Base for extended attributes used by this "
+          "translator instance, to avoid conflicts with others above or "
+          "below it."
+        },
+
+        /* NUFA option */
+        { .key  = {"local-volume-name"},
+          .type = GF_OPTION_TYPE_XLATOR
+        },
+
+        /* switch option */
+        { .key  = {"pattern.switch.case"},
+          .type = GF_OPTION_TYPE_ANY
+        },
+
+        { .key  = {NULL} },
+};
diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c
index e78ee1b45..fc0ca2f77 100644
--- a/xlators/cluster/dht/src/dht.c
+++ b/xlators/cluster/dht/src/dht.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -23,544 +14,23 @@
 #include "config.h"
 #endif
 
-/* TODO: add NS locking */
-
 #include "statedump.h"
-#include "dht-common.c"
-
-/* TODO:
-   - use volumename in xattr instead of "dht"
-   - use NS locks
-   - handle all cases in self heal layout reconstruction
-   - complete linkfile selfheal
-*/
-struct volume_options options[];
-
-void
-dht_layout_dump (dht_layout_t  *layout, const char *prefix)
-{
-
-        char    key[GF_DUMP_MAX_BUF_LEN];
-        int     i = 0;
-
-        GF_VALIDATE_OR_GOTO ("dht", layout, out);
-        GF_VALIDATE_OR_GOTO ("dht", prefix, out);
-
-        gf_proc_dump_build_key(key, prefix, "cnt");
-        gf_proc_dump_write(key, "%d", layout->cnt);
-        gf_proc_dump_build_key(key, prefix, "preset");
-        gf_proc_dump_write(key, "%d", layout->preset);
-        gf_proc_dump_build_key(key, prefix, "gen");
-        gf_proc_dump_write(key, "%d", layout->gen);
-        gf_proc_dump_build_key(key, prefix, "type");
-        gf_proc_dump_write(key, "%d", layout->type);
-
-        for (i = 0; i < layout->cnt; i++) {
-                gf_proc_dump_build_key(key, prefix,"list[%d].err", i);
-                gf_proc_dump_write(key, "%d", layout->list[i].err);
-                gf_proc_dump_build_key(key, prefix,"list[%d].start", i);
-                gf_proc_dump_write(key, "%u", layout->list[i].start);
-                gf_proc_dump_build_key(key, prefix,"list[%d].stop", i);
-                gf_proc_dump_write(key, "%u", layout->list[i].stop);
-                if (layout->list[i].xlator) {
-                        gf_proc_dump_build_key(key, prefix,
-                                               "list[%d].xlator.type", i);
-                        gf_proc_dump_write(key, "%s",
-                                           layout->list[i].xlator->type);
-                        gf_proc_dump_build_key(key, prefix,
-                                               "list[%d].xlator.name", i);
-                        gf_proc_dump_write(key, "%s",
-                                           layout->list[i].xlator->name);
-                }
-        }
-
-out:
-        return;
-}
-
-
-int32_t
-dht_priv_dump (xlator_t *this)
-{
-        char            key_prefix[GF_DUMP_MAX_BUF_LEN];
-        char            key[GF_DUMP_MAX_BUF_LEN];
-        int             i = 0;
-        dht_conf_t      *conf = NULL;
-        int             ret = -1;
-
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-
-        conf = this->private;
-
-        if (!conf)
-                return -1;
-
-        ret = TRY_LOCK(&conf->subvolume_lock);
-
-        if (ret != 0) {
-                gf_log("", GF_LOG_WARNING, "Unable to lock dht subvolume %s",
-                       this->name);
-                return ret;
-        }
-
-        gf_proc_dump_add_section("xlator.cluster.dht.%s.priv", this->name);
-        gf_proc_dump_build_key(key_prefix,"xlator.cluster.dht","%s.priv",
-                               this->name);
-        gf_proc_dump_build_key(key, key_prefix, "subvolume_cnt");
-        gf_proc_dump_write(key,"%d", conf->subvolume_cnt);
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                gf_proc_dump_build_key(key, key_prefix, "subvolumes[%d]", i);
-                gf_proc_dump_write(key, "%s.%s", conf->subvolumes[i]->type,
-                                   conf->subvolumes[i]->name);
-                if (conf->file_layouts && conf->file_layouts[i]){
-                        gf_proc_dump_build_key(key, key_prefix,
-                                               "file_layouts[%d]",i);
-                        dht_layout_dump(conf->file_layouts[i], key);
-                }
-                if (conf->dir_layouts && conf->dir_layouts[i]) {
-                        gf_proc_dump_build_key(key, key_prefix,
-                                               "dir_layouts[%d]",i);
-                        dht_layout_dump(conf->dir_layouts[i], key);
-                }
-                if (conf->subvolume_status) {
-                        gf_proc_dump_build_key(key, key_prefix,
-                                               "subvolume_status[%d]", i);
-                        gf_proc_dump_write(key, "%d",
-                                           (int)conf->subvolume_status[i]);
-                }
-
-        }
-
-        gf_proc_dump_build_key(key, key_prefix, "search_unhashed");
-        gf_proc_dump_write(key, "%d", conf->search_unhashed);
-        gf_proc_dump_build_key(key, key_prefix, "gen");
-        gf_proc_dump_write(key, "%d", conf->gen);
-        gf_proc_dump_build_key(key, key_prefix, "min_free_disk");
-        gf_proc_dump_write(key, "%lu", conf->min_free_disk);
-        gf_proc_dump_build_key(key, key_prefix, "disk_unit");
-        gf_proc_dump_write(key, "%c", conf->disk_unit);
-        gf_proc_dump_build_key(key, key_prefix, "refresh_interval");
-        gf_proc_dump_write(key, "%d", conf->refresh_interval);
-        gf_proc_dump_build_key(key, key_prefix, "unhashed_sticky_bit");
-        gf_proc_dump_write(key, "%d", conf->unhashed_sticky_bit);
-        if (conf ->du_stats) {
-                gf_proc_dump_build_key(key, key_prefix,
-                                       "du_stats.avail_percent");
-                gf_proc_dump_write(key, "%lf", conf->du_stats->avail_percent);
-                gf_proc_dump_build_key(key, key_prefix,
-                                       "du_stats.avail_space");
-                gf_proc_dump_write(key, "%lu", conf->du_stats->avail_space);
-                gf_proc_dump_build_key(key, key_prefix,
-                                       "du_stats.log");
-                gf_proc_dump_write(key, "%lu", conf->du_stats->log);
-        }
-        gf_proc_dump_build_key(key, key_prefix, "last_stat_fetch");
-        gf_proc_dump_write(key, "%s", ctime(&conf->last_stat_fetch.tv_sec));
-
-        UNLOCK(&conf->subvolume_lock);
-
-out:
-        return ret;
-}
-
-int32_t
-dht_inodectx_dump (xlator_t *this, inode_t *inode)
-{
-        int             ret = -1;
-        char            key_prefix[GF_DUMP_MAX_BUF_LEN];
-        dht_layout_t    *layout = NULL;
-        uint64_t        tmp_layout = 0;
-
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", inode, out);
-
-        ret = inode_ctx_get (inode, this, &tmp_layout);
-
-        if (ret != 0)
-                return ret;
-
-        layout = (dht_layout_t *)(long)tmp_layout;
-
-        if (!layout)
-                return -1;
-
-        gf_proc_dump_build_key(key_prefix, "xlator.cluster.dht",
-                               "%s.inode.%ld", this->name, inode->ino);
-        dht_layout_dump(layout, key_prefix);
-
-out:
-        return ret;
-}
-
-int
-notify (xlator_t *this, int event, void *data, ...)
-{
-        int ret = -1;
-
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-
-        ret = dht_notify (this, event, data);
-
-out:
-        return ret;
-}
-
-void
-fini (xlator_t *this)
-{
-        int         i = 0;
-        dht_conf_t *conf = NULL;
-
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-
-        conf = this->private;
-        this->private = NULL;
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
-
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
-
-                if (conf->subvolume_status)
-                        GF_FREE (conf->subvolume_status);
-
-                GF_FREE (conf);
-        }
-out:
-        return;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-
-        ret = xlator_mem_acct_init (this, gf_dht_mt_end + 1);
-
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        "failed");
-                return ret;
-        }
-out:
-        return ret;
-}
-
-int
-validate_options (xlator_t *this, char **op_errstr)
-{
-        int                 ret = 0;
-        volume_opt_list_t  *vol_opt = NULL;
-        volume_opt_list_t  *tmp;
-
-        if (!this) {
-                gf_log (this->name, GF_LOG_DEBUG, "'this' not a valid ptr");
-                ret =-1;
-                goto out;
-        }
-
-        if (list_empty (&this->volume_options))
-                goto out;
-
-        vol_opt = list_entry (this->volume_options.next,
-                                      volume_opt_list_t, list);
-         list_for_each_entry_safe (vol_opt, tmp, &this->volume_options, list) {
-                ret = validate_xlator_volume_options_attacherr (this,
-                                                                vol_opt->given_opt,
-                                                                op_errstr);
-        }
-
-out:
-
-        return ret;
-}
-
-int
-reconfigure (xlator_t *this, dict_t *options)
-{
-        dht_conf_t      *conf = NULL;
-        char            *temp_str = NULL;
-        gf_boolean_t     search_unhashed;
-        uint32_t         temp_free_disk = 0;
-        int              ret = -1;
-        uint32_t         dir_spread = 0;
-
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", options, out);
-
-        conf = this->private;
-        if (!conf)
-                return 0;
-
-        if (dict_get_str (options, "lookup-unhashed", &temp_str) == 0) {
-                /* If option is not "auto", other options _should_ be boolean*/
-                if (strcasecmp (temp_str, "auto")) {
-                        if (!gf_string2boolean (temp_str, &search_unhashed)) {
-                                gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
-                                       " lookup-unahashed reconfigured (%s)",
-                                       temp_str);
-                                conf->search_unhashed = search_unhashed;
-                        } else {
-                                gf_log(this->name, GF_LOG_ERROR, "Reconfigure:"
-                                       " lookup-unahashed should be boolean,"
-                                       " not (%s), defaulting to (%d)",
-                                       temp_str, conf->search_unhashed);
-                                //return -1;
-                                ret = -1;
-                                goto out;
-                        }
-                } else {
-                        gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
-                               " lookup-unahashed reconfigured auto ");
-                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
-                }
-        }
-
-        if (dict_get_str (options, "min-free-disk", &temp_str) == 0) {
-                if (gf_string2percent (temp_str, &temp_free_disk) == 0) {
-                        if (temp_free_disk > 100) {
-                                gf_string2bytesize (temp_str,
-                                                    &conf->min_free_disk);
-                                conf->disk_unit = 'b';
-                        } else {
-                                conf->min_free_disk = (uint64_t)temp_free_disk;
-                        }
-                } else {
-                        gf_string2bytesize (temp_str, &conf->min_free_disk);
-                        conf->disk_unit = 'b';
-                }
-
-                gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
-                       " min-free-disk reconfigured to %s",
-                       temp_str);
-        }
-
-        if (dict_get_str (options, "directory-layout-spread", &temp_str) == 0) {
-                ret = gf_string2uint32 (temp_str, &dir_spread);
-                if (ret ||
-                    (dir_spread > conf->subvolume_cnt) ||
-                    (dir_spread < 1)) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "wrong 'directory-layout-spread' option given "
-                                "(%s). setting to earlier value (%d)",
-                                temp_str, conf->dir_spread_cnt);
-                        ret = -1;
-                        goto out;
-                }
-                conf->dir_spread_cnt = dir_spread;
-        }
-
-        ret = 0;
-out:
-        return ret;
-}
-
-int
-init (xlator_t *this)
-{
-        dht_conf_t    *conf = NULL;
-        char          *temp_str = NULL;
-        int            ret = -1;
-        int            i = 0;
-        uint32_t       temp_free_disk = 0;
-        char          *def_val = NULL;
-
-        GF_VALIDATE_OR_GOTO ("dht", this, err);
-
-        if (!this->children) {
-                gf_log (this->name, GF_LOG_CRITICAL,
-                        "Distribute needs more than one subvolume");
-                return -1;
-        }
-
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile");
-        }
-
-        conf = GF_CALLOC (1, sizeof (*conf), gf_dht_mt_dht_conf_t);
-        if (!conf) {
-                goto err;
-        }
-
-        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
-        if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
-                /* If option is not "auto", other options _should_ be boolean */
-                if (strcasecmp (temp_str, "auto"))
-                        gf_string2boolean (temp_str, &conf->search_unhashed);
-                else
-                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
-        }
-
-        conf->unhashed_sticky_bit = 0;
-
-        if (dict_get_str (this->options, "unhashed-sticky-bit",
-                          &temp_str) == 0) {
-                gf_string2boolean (temp_str, &conf->unhashed_sticky_bit);
-        }
-
-        conf->use_readdirp = 1;
-
-        if (dict_get_str (this->options, "use-readdirp",
-                          &temp_str) == 0) {
-                gf_string2boolean (temp_str, &conf->use_readdirp);
-        }
-
-        if (xlator_get_volopt_info (&this->volume_options, "min-free-disk",
-                                    &def_val, NULL)) {
-                gf_log (this->name, GF_LOG_ERROR, "Default value of "
-                         " min-free-disk not found");
-                ret = -1;
-                goto err;
-        } else {
-                if (gf_string2percent (def_val, &temp_free_disk) == 0) {
-                        if (temp_free_disk > 100) {
-                                gf_string2bytesize (temp_str,
-                                                    &conf->min_free_disk);
-                                conf->disk_unit = 'b';
-                        } else {
-                                conf->min_free_disk = (uint64_t)temp_free_disk;
-                                conf->disk_unit = 'p';
-                        }
-                } else {
-                        gf_string2bytesize (temp_str, &conf->min_free_disk);
-                        conf->disk_unit = 'b';
-                }
-        }
-
-        if (dict_get_str (this->options, "min-free-disk", &temp_str) == 0) {
-                if (gf_string2percent (temp_str, &temp_free_disk) == 0) {
-                        if (temp_free_disk > 100) {
-                                gf_string2bytesize (temp_str,
-                                                    &conf->min_free_disk);
-                                conf->disk_unit = 'b';
-                        } else {
-                                conf->min_free_disk = (uint64_t)temp_free_disk;
-                        }
-                } else {
-                        gf_string2bytesize (temp_str, &conf->min_free_disk);
-                        conf->disk_unit = 'b';
-                }
-        }
-
-        conf->dir_spread_cnt = conf->subvolume_cnt;
-        if (dict_get_str (this->options, "directory-layout-spread",
-                          &temp_str) == 0) {
-                ret = gf_string2uint32 (temp_str, &conf->dir_spread_cnt);
-                if (ret ||
-                    (conf->dir_spread_cnt > conf->subvolume_cnt) ||
-                    (conf->dir_spread_cnt < 1)) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "wrong 'directory-layout-spread' option given "
-                                "(%s). setting it to subvolume count",
-                                temp_str);
-                        conf->dir_spread_cnt = conf->subvolume_cnt;
-                }
-        }
-
-        conf->assert_no_child_down = 0;
-
-        ret = dict_get_str_boolean (this->options, "assert-no-child-down", 0);
-        if (ret != -1) {
-                if (conf->assert_no_child_down != ret) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Changing assert-no-child-down from %d to %d",
-                                conf->assert_no_child_down, ret);
-                }
-                conf->assert_no_child_down = ret;
-        } else {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "'assert-no-child-down' takes only boolean arguments");
-        }
-
-        ret = dht_init_subvolumes (this, conf);
-        if (ret == -1) {
-                goto err;
-        }
-
-        ret = dht_layouts_init (this, conf);
-        if (ret == -1) {
-                goto err;
-        }
-
-        conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
-                                    gf_dht_mt_dht_du_t);
-        if (!conf->du_stats) {
-                goto err;
-        }
-
-        LOCK_INIT (&conf->subvolume_lock);
-        LOCK_INIT (&conf->layout_lock);
-
-        conf->gen = 1;
-
-        /* Create 'syncop' environment */
-	conf->env = syncenv_new (0);
-        if (!conf->env) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to create sync environment %s",
-                        strerror (errno));
-                goto err;
-        }
-
-        this->private = conf;
-
-        return 0;
-
-err:
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
-
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
-
-                if (conf->subvolume_status)
-                        GF_FREE (conf->subvolume_status);
-
-                if (conf->du_stats)
-                        GF_FREE (conf->du_stats);
-
-                GF_FREE (conf);
-        }
-
-        return -1;
-}
+#include "dht-common.h"
 
+class_methods_t class_methods = {
+        .init           = dht_init,
+        .fini           = dht_fini,
+        .reconfigure    = dht_reconfigure,
+        .notify         = dht_notify
+};
 
 struct xlator_fops fops = {
         .lookup      = dht_lookup,
         .mknod       = dht_mknod,
         .create      = dht_create,
 
-        .stat        = dht_stat,
-        .fstat       = dht_fstat,
-        .truncate    = dht_truncate,
-        .ftruncate   = dht_ftruncate,
-        .access      = dht_access,
-        .readlink    = dht_readlink,
-        .setxattr    = dht_setxattr,
-        .fsetxattr   = dht_fsetxattr,
-        .getxattr    = dht_getxattr,
-        .removexattr = dht_removexattr,
         .open        = dht_open,
-        .readv       = dht_readv,
-        .writev      = dht_writev,
-        .flush       = dht_flush,
-        .fsync       = dht_fsync,
         .statfs      = dht_statfs,
-        .lk          = dht_lk,
         .opendir     = dht_opendir,
         .readdir     = dht_readdir,
         .readdirp    = dht_readdirp,
@@ -571,14 +41,38 @@ struct xlator_fops fops = {
         .mkdir       = dht_mkdir,
         .rmdir       = dht_rmdir,
         .rename      = dht_rename,
-        .inodelk     = dht_inodelk,
-        .finodelk    = dht_finodelk,
         .entrylk     = dht_entrylk,
         .fentrylk    = dht_fentrylk,
+
+        /* Inode read operations */
+        .stat        = dht_stat,
+        .fstat       = dht_fstat,
+        .access      = dht_access,
+        .readlink    = dht_readlink,
+        .getxattr    = dht_getxattr,
+        .fgetxattr    = dht_fgetxattr,
+        .readv       = dht_readv,
+        .flush       = dht_flush,
+        .fsync       = dht_fsync,
+        .inodelk     = dht_inodelk,
+        .finodelk    = dht_finodelk,
+        .lk          = dht_lk,
+
+        /* Inode write operations */
+        .fremovexattr = dht_fremovexattr,
+        .removexattr = dht_removexattr,
+        .setxattr    = dht_setxattr,
+        .fsetxattr   = dht_fsetxattr,
+        .truncate    = dht_truncate,
+        .ftruncate   = dht_ftruncate,
+        .writev      = dht_writev,
         .xattrop     = dht_xattrop,
         .fxattrop    = dht_fxattrop,
         .setattr     = dht_setattr,
         .fsetattr    = dht_fsetattr,
+	.fallocate   = dht_fallocate,
+	.discard     = dht_discard,
+        .zerofill    = dht_zerofill,
 };
 
 struct xlator_dumpops dumpops = {
@@ -592,31 +86,4 @@ struct xlator_cbks cbks = {
 //      .releasedir = dht_releasedir,
         .forget     = dht_forget
 };
-
-
-struct volume_options options[] = {
-        { .key  = {"lookup-unhashed"},
-          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
-                    "on", "off"},
-          .type = GF_OPTION_TYPE_STR
-        },
-        { .key  = {"min-free-disk"},
-          .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
-          .default_value = "10%",
-          .description = "Percentage/Size of disk space that must be "
-                         "kept free."
-        },
-        { .key = {"unhashed-sticky-bit"},
-          .type = GF_OPTION_TYPE_BOOL
-        },
-        { .key = {"use-readdirp"},
-          .type = GF_OPTION_TYPE_BOOL
-        },
-        { .key = {"assert-no-child-down"},
-          .type = GF_OPTION_TYPE_BOOL
-        },
-        { .key  = {"directory-layout-spread"},
-          .type = GF_OPTION_TYPE_INT,
-        },
-        { .key  = {NULL} },
-};
+;
diff --git a/xlators/cluster/dht/src/nufa.c b/xlators/cluster/dht/src/nufa.c
index b3d17351a..e934acdf0 100644
--- a/xlators/cluster/dht/src/nufa.c
+++ b/xlators/cluster/dht/src/nufa.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -23,10 +14,12 @@
 #include "config.h"
 #endif
 
-#include "dht-common.c"
+#include "dht-common.h"
 
 /* TODO: all 'TODO's in dht.c holds good */
 
+extern struct volume_options options[];
+
 int
 nufa_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                        int op_ret, int op_errno,
@@ -44,7 +37,6 @@ nufa_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         int           call_cnt    = 0;
         int           ret         = 0;
 
-
         conf  = this->private;
 
         prev  = cookie;
@@ -62,7 +54,8 @@ nufa_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         if (op_ret == -1)
                 goto out;
 
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr);
+        is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+                                         conf->link_xattr_name);
         is_dir      = check_is_dir (inode, stbuf, xattr);
 
         if (!is_dir && !is_linkfile) {
@@ -141,7 +134,7 @@ out:
 
 err:
         DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno,
-                          inode, stbuf, xattr, NULL);
+                          inode, stbuf, xattr, postparent);
         return 0;
 }
 
@@ -150,7 +143,6 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
              loc_t *loc, dict_t *xattr_req)
 {
         xlator_t     *hashed_subvol = NULL;
-        xlator_t     *cached_subvol = NULL;
         xlator_t     *subvol = NULL;
         dht_local_t  *local  = NULL;
         dht_conf_t   *conf = NULL;
@@ -169,21 +161,12 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
 
         conf = this->private;
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_LOOKUP);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
 
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = errno;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "copying location failed for path=%s",
-                        loc->path);
-                goto err;
-        }
-
         if (xattr_req) {
                 local->xattr_req = dict_ref (xattr_req);
         } else {
@@ -191,14 +174,11 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
         }
 
         hashed_subvol = dht_subvol_get_hashed (this, &local->loc);
-        cached_subvol = dht_subvol_get_cached (this, local->loc.inode);
 
-        local->cached_subvol = cached_subvol;
         local->hashed_subvol = hashed_subvol;
 
         if (is_revalidate (loc)) {
-                local->layout = layout = dht_layout_get (this, loc->inode);
-
+                layout = local->layout;
                 if (!layout) {
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "revalidate without cache. path=%s",
@@ -215,7 +195,7 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
                         goto do_fresh_lookup;
                 }
 
-                local->inode    = inode_ref (loc->inode);
+                local->inode = inode_ref (loc->inode);
 
                 local->call_cnt = layout->cnt;
                 call_cnt = local->call_cnt;
@@ -224,7 +204,7 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
                  *       revalidates directly go to the cached-subvolume.
                  */
                 ret = dict_set_uint32 (local->xattr_req,
-                                       "trusted.glusterfs.dht", 4 * 4);
+                                       conf->xattr_name, 4 * 4);
                 if (ret < 0) {
                         gf_log (this->name, GF_LOG_ERROR,
                                 "Failed to set dict value.");
@@ -245,7 +225,7 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
         } else {
         do_fresh_lookup:
                 ret = dict_set_uint32 (local->xattr_req,
-                                       "trusted.glusterfs.dht", 4 * 4);
+                                       conf->xattr_name, 4 * 4);
                 if (ret < 0) {
                         gf_log (this->name, GF_LOG_ERROR,
                                 "Failed to set dict value.");
@@ -254,7 +234,7 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
                 }
 
                 ret = dict_set_uint32 (local->xattr_req,
-                                       "trusted.glusterfs.dht.linkto", 256);
+                                       conf->link_xattr_name, 256);
                 if (ret < 0) {
                         gf_log (this->name, GF_LOG_ERROR,
                                 "Failed to set dict value.");
@@ -273,7 +253,8 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL);
         return 0;
 }
 
@@ -282,7 +263,7 @@ nufa_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
                                  xlator_t *this, int op_ret, int op_errno,
                                  inode_t *inode, struct iatt *stbuf,
                                  struct iatt *preparent,
-                                 struct iatt *postparent)
+                                 struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
 
@@ -293,28 +274,27 @@ nufa_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
 
         STACK_WIND (frame, dht_create_cbk,
                     local->cached_subvol, local->cached_subvol->fops->create,
-                    &local->loc, local->flags, local->mode, local->fd,
-                    local->params);
+                    &local->loc, local->flags, local->mode, local->umask,
+                    local->fd, local->params);
 
         return 0;
 
 err:
         DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
+                          NULL, NULL, NULL, NULL, NULL, NULL);
         return 0;
 }
 
 int
 nufa_create (call_frame_t *frame, xlator_t *this,
              loc_t *loc, int32_t flags, mode_t mode,
-             fd_t *fd, dict_t *params)
+             mode_t umask, fd_t *fd, dict_t *params)
 {
         dht_local_t *local = NULL;
         dht_conf_t  *conf  = NULL;
         xlator_t    *subvol = NULL;
         xlator_t    *avail_subvol = NULL;
         int          op_errno = -1;
-        int          ret = -1;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -324,7 +304,7 @@ nufa_create (call_frame_t *frame, xlator_t *this,
 
         dht_get_du_info (frame, this, loc);
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, fd, GF_FOP_CREATE);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -343,26 +323,19 @@ nufa_create (call_frame_t *frame, xlator_t *this,
         if (dht_is_subvol_filled (this, (xlator_t *)conf->private)) {
                 avail_subvol =
                         dht_free_disk_available_subvol (this,
-                                                        (xlator_t *)conf->private);
+                                                        (xlator_t *)conf->private,
+                                                        local);
         }
 
         if (subvol != avail_subvol) {
                 /* create a link file instead of actual file */
-                ret = loc_copy (&local->loc, loc);
-                if (ret == -1) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
-
-                local->fd = fd_ref (fd);
                 local->params = dict_ref (params);
                 local->mode = mode;
                 local->flags = flags;
-
+                local->umask = umask;
                 local->cached_subvol = avail_subvol;
-                dht_linkfile_create (frame,
-                                     nufa_create_linkfile_create_cbk,
-                                     avail_subvol, subvol, loc);
+                dht_linkfile_create (frame, nufa_create_linkfile_create_cbk,
+                                     this, avail_subvol, subvol, loc);
                 return 0;
         }
 
@@ -371,14 +344,14 @@ nufa_create (call_frame_t *frame, xlator_t *this,
 
         STACK_WIND (frame, dht_create_cbk,
                     subvol, subvol->fops->create,
-                    loc, flags, mode, fd, params);
+                    loc, flags, mode, umask, fd, params);
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
         DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
+                          NULL, NULL, NULL, NULL, NULL, NULL);
 
         return 0;
 }
@@ -387,41 +360,45 @@ int
 nufa_mknod_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                          int op_ret, int op_errno, inode_t *inode,
                          struct iatt *stbuf, struct iatt *preparent,
-                         struct iatt *postparent)
+                         struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
 
         local = frame->local;
+        if (!local || !local->cached_subvol) {
+                op_errno = EINVAL;
+                op_ret = -1;
+                goto err;
+        }
 
         if (op_ret >= 0) {
-                STACK_WIND (frame, dht_newfile_cbk,
-                            local->cached_subvol,
+                STACK_WIND_COOKIE (frame, dht_newfile_cbk,
+                            (void *)local->cached_subvol, local->cached_subvol,
                             local->cached_subvol->fops->mknod,
                             &local->loc, local->mode, local->rdev,
-                            local->params);
+                            local->umask, local->params);
 
                 return 0;
         }
-
+err:
         WIPE (postparent);
         WIPE (preparent);
 
         DHT_STACK_UNWIND (link, frame, op_ret, op_errno,
-                          inode, stbuf, preparent, postparent);
+                          inode, stbuf, preparent, postparent, xdata);
         return 0;
 }
 
 
 int
 nufa_mknod (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, mode_t mode, dev_t rdev, dict_t *params)
+            loc_t *loc, mode_t mode, dev_t rdev, mode_t umask, dict_t *params)
 {
         dht_local_t *local = NULL;
         dht_conf_t  *conf  = NULL;
         xlator_t    *subvol = NULL;
         xlator_t    *avail_subvol = NULL;
         int          op_errno = -1;
-        int          ret = -1;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -431,7 +408,7 @@ nufa_mknod (call_frame_t *frame, xlator_t *this,
 
         dht_get_du_info (frame, this, loc);
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_MKNOD);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -451,23 +428,20 @@ nufa_mknod (call_frame_t *frame, xlator_t *this,
         if (dht_is_subvol_filled (this, (xlator_t *)conf->private)) {
                 avail_subvol =
                         dht_free_disk_available_subvol (this,
-                                                        (xlator_t *)conf->private);
+                                                        (xlator_t *)conf->private,
+                                                        local);
         }
 
         if (avail_subvol != subvol) {
                 /* Create linkfile first */
-                ret = loc_copy (&local->loc, loc);
-                if (ret == -1) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
 
                 local->params = dict_ref (params);
                 local->mode = mode;
+                local->umask = umask;
                 local->rdev = rdev;
                 local->cached_subvol = avail_subvol;
 
-                dht_linkfile_create (frame, nufa_mknod_linkfile_cbk,
+                dht_linkfile_create (frame, nufa_mknod_linkfile_cbk, this,
                                      avail_subvol, subvol, loc);
                 return 0;
         }
@@ -475,211 +449,185 @@ nufa_mknod (call_frame_t *frame, xlator_t *this,
         gf_log (this->name, GF_LOG_TRACE,
                 "creating %s on %s", loc->path, subvol->name);
 
-        STACK_WIND (frame, dht_newfile_cbk,
-                    subvol, subvol->fops->mknod,
-                    loc, mode, rdev, params);
+        STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol, subvol,
+                           subvol->fops->mknod, loc, mode, rdev, umask,
+                           params);
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
         DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL);
+                          NULL, NULL, NULL, NULL, NULL);
 
         return 0;
 }
 
 
-int
-notify (xlator_t *this, int event, void *data, ...)
+gf_boolean_t
+same_first_part (char *str1, char term1, char *str2, char term2)
 {
-        int ret = -1;
-
-        ret = dht_notify (this, event, data);
-
-        return ret;
+        gf_boolean_t    ended1;
+        gf_boolean_t    ended2;
+
+        for (;;) {
+                ended1 = ((*str1 == '\0') || (*str1 == term1));
+                ended2 = ((*str2 == '\0') || (*str2 == term2));
+                if (ended1 && ended2) {
+                        return _gf_true;
+                }
+                if (ended1 || ended2 || (*str1 != *str2)) {
+                        return _gf_false;
+                }
+                ++str1;
+                ++str2;
+        }
 }
 
-void
-fini (xlator_t *this)
-{
-        int         i = 0;
-        dht_conf_t *conf = NULL;
+typedef struct nufa_args {
+        xlator_t    *this;
+        char        *volname;
+        gf_boolean_t addr_match;
+} nufa_args_t;
 
-        conf = this->private;
+static void
+nufa_find_local_brick (xlator_t *xl, void *data)
+{
+        nufa_args_t     *args = data;
+        xlator_t        *this = args->this;
+        char            *local_volname = args->volname;
+        gf_boolean_t    addr_match = args->addr_match;
+        char            *brick_host = NULL;
+        dht_conf_t      *conf = this->private;
+        int             ret = -1;
+
+        /*This means a local subvol was already found. We pick the first brick
+         * that is local*/
+        if (conf->private)
+                return;
+
+        if (strcmp (xl->name, local_volname) == 0) {
+                conf->private = xl;
+                gf_log (this->name, GF_LOG_INFO, "Using specified subvol %s",
+                        local_volname);
+                return;
+        }
 
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
+        if (!addr_match)
+                return;
 
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
+        ret = dict_get_str (xl->options, "remote-host", &brick_host);
+        if ((ret == 0) &&
+            (gf_is_same_address (local_volname, brick_host) ||
+             gf_is_local_addr (brick_host))) {
+                conf->private = xl;
+                gf_log (this->name, GF_LOG_INFO, "Using the first local "
+                        "subvol %s", xl->name);
+                return;
+        }
 
-                if (conf->subvolume_status)
-                        GF_FREE (conf->subvolume_status);
+}
 
-                GF_FREE (conf);
-        }
+static void
+nufa_to_dht (xlator_t *this)
+{
+        GF_ASSERT (this);
+        GF_ASSERT (this->fops);
 
-        return;
+        this->fops->lookup = dht_lookup;
+        this->fops->create = dht_create;
+        this->fops->mknod  = dht_mknod;
 }
 
 int
-init (xlator_t *this)
+nufa_find_local_subvol (xlator_t *this,
+                        void (*fn) (xlator_t *each, void* data), void *data)
 {
-        dht_conf_t    *conf = NULL;
-        xlator_list_t *trav = NULL;
-        data_t        *data = NULL;
-        char          *local_volname = NULL;
-        char          *temp_str = NULL;
-        int            ret = -1;
-        int            i = 0;
-        char           my_hostname[256];
-        uint32_t       temp_free_disk = 0;
-
-        if (!this->children) {
-                gf_log (this->name, GF_LOG_CRITICAL,
-                        "NUFA needs more than one subvolume");
+        int             ret = -1;
+        dht_conf_t      *conf = this->private;
+        xlator_list_t   *trav = NULL;
+        xlator_t        *parent = NULL;
+        xlator_t        *candidate = NULL;
+
+        xlator_foreach_depth_first (this, fn, data);
+        if (!conf->private) {
+                gf_log (this->name, GF_LOG_ERROR, "Couldn't find a local "
+                        "brick");
                 return -1;
         }
 
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile");
-        }
-
-        conf = GF_CALLOC (1, sizeof (*conf),
-                          gf_dht_mt_dht_conf_t);
-        if (!conf) {
-                goto err;
-        }
-
-        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
-        if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
-                /* If option is not "auto", other options _should_ be boolean */
-                if (strcasecmp (temp_str, "auto"))
-                        gf_string2boolean (temp_str, &conf->search_unhashed);
-                else
-                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
-        }
+        candidate = conf->private;
+        trav = candidate->parents;
+        while (trav) {
 
-        ret = dht_init_subvolumes (this, conf);
-        if (ret == -1) {
-                goto err;
-        }
+                parent = trav->xlator;
+                if (strcmp (parent->type, "cluster/nufa") == 0) {
+                        gf_log (this->name, GF_LOG_INFO, "Found local subvol, "
+                                "%s", candidate->name);
+                        ret = 0;
+                        conf->private = candidate;
+                        break;
+                }
 
-        ret = dht_layouts_init (this, conf);
-        if (ret == -1) {
-                goto err;
+                candidate = parent;
+                trav = parent->parents;
         }
 
-        LOCK_INIT (&conf->subvolume_lock);
-        LOCK_INIT (&conf->layout_lock);
+        return ret;
+}
 
-        conf->gen = 1;
+int
+nufa_init (xlator_t *this)
+{
+        data_t        *data = NULL;
+        char          *local_volname = NULL;
+        int            ret = -1;
+        char           my_hostname[256];
+        gf_boolean_t   addr_match = _gf_false;
+        nufa_args_t    args = {0, };
 
-        local_volname = "localhost";
-        ret = gethostname (my_hostname, 256);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "could not find hostname (%s)",
-                        strerror (errno));
+        ret = dht_init(this);
+        if (ret) {
+                return ret;
         }
 
-        if (ret == 0)
-                local_volname = my_hostname;
-
-        data = dict_get (this->options, "local-volume-name");
-        if (data) {
+        if ((data = dict_get (this->options, "local-volume-name"))) {
                 local_volname = data->data;
-        }
 
-        trav = this->children;
-        while (trav) {
-                if (strcmp (trav->xlator->name, local_volname) == 0)
-                        break;
-                trav = trav->next;
-        }
+        } else {
+                addr_match = _gf_true;
+                local_volname = "localhost";
+                ret = gethostname (my_hostname, 256);
+                if (ret == 0)
+                        local_volname = my_hostname;
 
-        if (!trav) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not find subvolume named '%s'. "
-                        "Please define volume with the name as the hostname "
-                        "or override it with 'option local-volume-name'",
-                        local_volname);
-                goto err;
-        }
-        /* The volume specified exists */
-        conf->private = trav->xlator;
-
-        conf->min_free_disk = 10;
-        conf->disk_unit = 'p';
-
-        if (dict_get_str (this->options, "min-free-disk",
-                          &temp_str) == 0) {
-                if (gf_string2percent (temp_str,
-                                       &temp_free_disk) == 0) {
-                        if (temp_free_disk > 100) {
-                                gf_string2bytesize (temp_str,
-                                                    &conf->min_free_disk);
-                                conf->disk_unit = 'b';
-                        } else {
-                                conf->min_free_disk = (uint64_t)temp_free_disk;
-                                conf->disk_unit = 'p';
-                        }
-                } else {
-                        gf_string2bytesize (temp_str,
-                                            &conf->min_free_disk);
-                        conf->disk_unit = 'b';
-                }
-        }
+                else
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "could not find hostname (%s)",
+                                strerror (errno));
 
-        conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
-                                    gf_dht_mt_dht_du_t);
-        if (!conf->du_stats) {
-                goto err;
         }
 
-        /* Create 'syncop' environment */
-	conf->env = syncenv_new (0);
-        if (!conf->env) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to create sync environment %s",
-                        strerror (errno));
-                goto err;
+        args.this = this;
+        args.volname = local_volname;
+        args.addr_match = addr_match;
+        ret = nufa_find_local_subvol (this, nufa_find_local_brick, &args);
+        if (ret) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Unable to find local subvolume, switching "
+                        "to dht mode");
+                nufa_to_dht (this);
         }
-
-        this->private = conf;
-
         return 0;
+}
 
-err:
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
-
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
-
-                if (conf->subvolume_status)
-                        GF_FREE (conf->subvolume_status);
-
-                if (conf->du_stats)
-                        GF_FREE (conf->du_stats);
-
-                GF_FREE (conf);
-        }
 
-        return -1;
-}
+class_methods_t class_methods = {
+        .init           = nufa_init,
+        .fini           = dht_fini,
+        .reconfigure    = dht_reconfigure,
+        .notify         = dht_notify
+};
 
 
 struct xlator_fops fops = {
@@ -726,19 +674,3 @@ struct xlator_fops fops = {
 struct xlator_cbks cbks = {
         .forget     = dht_forget
 };
-
-
-struct volume_options options[] = {
-        { .key  = {"lookup-unhashed"},
-          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
-                    "on", "off"},
-          .type = GF_OPTION_TYPE_STR
-        },
-        { .key  = {"local-volume-name"},
-          .type = GF_OPTION_TYPE_XLATOR
-        },
-        { .key  = {"min-free-disk"},
-          .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
-        },
-        { .key  = {NULL} },
-};
diff --git a/xlators/cluster/dht/src/switch.c b/xlators/cluster/dht/src/switch.c
index 211834fb5..2717ce975 100644
--- a/xlators/cluster/dht/src/switch.c
+++ b/xlators/cluster/dht/src/switch.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -23,7 +14,7 @@
 #include "config.h"
 #endif
 
-#include "dht-common.c"
+#include "dht-common.h"
 #include "dht-mem-types.h"
 
 #include <sys/time.h>
@@ -31,6 +22,8 @@
 #include <fnmatch.h>
 #include <string.h>
 
+extern struct volume_options options[];
+
 struct switch_sched_array {
         xlator_t *xl;
         int32_t   eligible;
@@ -76,29 +69,37 @@ get_switch_matching_subvol (const char *path, dht_conf_t *conf,
         struct switch_struct *cond      = NULL;
         struct switch_struct *trav      = NULL;
         char                 *pathname  = NULL;
-        int                   idx     = 0;
+        int                   idx       = 0;
+        xlator_t             *subvol    = NULL;
 
         cond = conf->private;
+        subvol = hashed_subvol;
         if (!cond)
-                return hashed_subvol;
+                goto out;
 
-        trav = cond;
         pathname = gf_strdup (path);
+        if (!pathname)
+                goto out;
+
+        trav = cond;
         while (trav) {
                 if (fnmatch (trav->path_pattern,
                              pathname, FNM_NOESCAPE) == 0) {
                         for (idx = 0; idx < trav->num_child; idx++) {
                                 if (trav->array[idx].xl == hashed_subvol)
-                                        return hashed_subvol;
+                                        goto out;
                         }
                         idx = trav->node_index++;
                         trav->node_index %= trav->num_child;
-                        return trav->array[idx].xl;
+                        subvol = trav->array[idx].xl;
+                        goto out;
                 }
                 trav = trav->next;
         }
+out:
         GF_FREE (pathname);
-        return hashed_subvol;
+
+        return subvol;
 }
 
 
@@ -136,7 +137,8 @@ switch_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         if (op_ret == -1)
                 goto out;
 
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr);
+        is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+                                         conf->link_xattr_name);
         is_dir      = check_is_dir (inode, stbuf, xattr);
 
         if (!is_dir && !is_linkfile) {
@@ -246,21 +248,12 @@ switch_lookup (call_frame_t *frame, xlator_t *this,
 
         conf = this->private;
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_LOOKUP);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
 
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = errno;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "copying location failed for path=%s",
-                        loc->path);
-                goto err;
-        }
-
         if (xattr_req) {
                 local->xattr_req = dict_ref (xattr_req);
         } else {
@@ -268,14 +261,12 @@ switch_lookup (call_frame_t *frame, xlator_t *this,
         }
 
         hashed_subvol = dht_subvol_get_hashed (this, &local->loc);
-        cached_subvol = dht_subvol_get_cached (this, local->loc.inode);
+        cached_subvol = local->cached_subvol;
 
-        local->cached_subvol = cached_subvol;
         local->hashed_subvol = hashed_subvol;
 
         if (is_revalidate (loc)) {
-                local->layout = layout = dht_layout_get (this, loc->inode);
-
+                layout = local->layout;
                 if (!layout) {
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "revalidate without cache. path=%s",
@@ -301,11 +292,11 @@ switch_lookup (call_frame_t *frame, xlator_t *this,
                  * attribute, revalidates directly go to the cached-subvolume.
                  */
                 ret = dict_set_uint32 (local->xattr_req,
-                                       "trusted.glusterfs.dht", 4 * 4);
+                                       conf->xattr_name, 4 * 4);
                 if (ret < 0)
                         gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set dict value for "
-                                "trusted.glusterfs.dht");
+                                "failed to set dict value for %s",
+                                conf->xattr_name);
 
                 for (i = 0; i < layout->cnt; i++) {
                         subvol = layout->list[i].xlator;
@@ -320,18 +311,18 @@ switch_lookup (call_frame_t *frame, xlator_t *this,
         } else {
         do_fresh_lookup:
                 ret = dict_set_uint32 (local->xattr_req,
-                                       "trusted.glusterfs.dht", 4 * 4);
+                                       conf->xattr_name, 4 * 4);
                 if (ret < 0)
                         gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set dict value for "
-                                "trusted.glusterfs.dht");
+                                "failed to set dict value for %s",
+                                conf->xattr_name);
 
                 ret = dict_set_uint32 (local->xattr_req,
-                                       "trusted.glusterfs.dht.linkto", 256);
+                                       conf->link_xattr_name, 256);
                 if (ret < 0)
                         gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set dict value for "
-                                "trusted.glusterfs.dht.linkto");
+                                "failed to set dict value for %s",
+                                conf->link_xattr_name);
 
                 if (!hashed_subvol) {
                         gf_log (this->name, GF_LOG_DEBUG,
@@ -377,7 +368,8 @@ switch_lookup (call_frame_t *frame, xlator_t *this,
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+        DHT_STACK_UNWIND (lookup, frame, -1, op_errno,
+                          NULL, NULL, NULL, NULL);
         return 0;
 }
 
@@ -386,7 +378,7 @@ switch_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
                                    xlator_t *this, int op_ret, int op_errno,
                                    inode_t *inode, struct iatt *stbuf,
                                    struct iatt *preparent,
-                                   struct iatt *postparent)
+                                   struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
 
@@ -397,28 +389,27 @@ switch_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
 
         STACK_WIND (frame, dht_create_cbk,
                     local->cached_subvol, local->cached_subvol->fops->create,
-                    &local->loc, local->flags, local->mode, local->fd,
-                    local->params);
+                    &local->loc, local->flags, local->mode, local->umask,
+                    local->fd, local->params);
 
         return 0;
 
 err:
         DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
+                          NULL, NULL, NULL, NULL, NULL, NULL);
         return 0;
 }
 
 int
 switch_create (call_frame_t *frame, xlator_t *this,
                loc_t *loc, int32_t flags, mode_t mode,
-               fd_t *fd, dict_t *params)
+               mode_t umask, fd_t *fd, dict_t *params)
 {
         dht_local_t *local = NULL;
         dht_conf_t  *conf  = NULL;
         xlator_t    *subvol = NULL;
         xlator_t    *avail_subvol = NULL;
         int          op_errno = -1;
-        int          ret = -1;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -428,7 +419,7 @@ switch_create (call_frame_t *frame, xlator_t *this,
 
         dht_get_du_info (frame, this, loc);
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, fd, GF_FOP_CREATE);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -446,25 +437,18 @@ switch_create (call_frame_t *frame, xlator_t *this,
         avail_subvol = get_switch_matching_subvol (loc->path, conf, subvol);
         if (dht_is_subvol_filled (this, avail_subvol)) {
                 avail_subvol =
-                        dht_free_disk_available_subvol (this, avail_subvol);
+                        dht_free_disk_available_subvol (this, avail_subvol,
+                                                        local);
         }
 
         if (subvol != avail_subvol) {
                 /* create a link file instead of actual file */
-                ret = loc_copy (&local->loc, loc);
-                if (ret == -1) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
-
-                local->fd = fd_ref (fd);
                 local->mode = mode;
                 local->flags = flags;
-
+                local->umask = umask;
                 local->cached_subvol = avail_subvol;
-                dht_linkfile_create (frame,
-                                     switch_create_linkfile_create_cbk,
-                                     avail_subvol, subvol, loc);
+                dht_linkfile_create (frame, switch_create_linkfile_create_cbk,
+                                     this, avail_subvol, subvol, loc);
                 return 0;
         }
 
@@ -473,14 +457,14 @@ switch_create (call_frame_t *frame, xlator_t *this,
 
         STACK_WIND (frame, dht_create_cbk,
                     subvol, subvol->fops->create,
-                    loc, flags, mode, fd, params);
+                    loc, flags, mode, umask, fd, params);
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
         DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
+                          NULL, NULL, NULL, NULL, NULL, NULL);
 
         return 0;
 }
@@ -489,38 +473,42 @@ int
 switch_mknod_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                            int op_ret, int op_errno, inode_t *inode,
                            struct iatt *stbuf, struct iatt *preparent,
-                           struct iatt *postparent)
+                           struct iatt *postparent, dict_t *xdata)
 {
         dht_local_t  *local = NULL;
 
         local = frame->local;
+        if (!local || !local->cached_subvol) {
+                op_errno = EINVAL;
+                op_ret = -1;
+                goto err;
+        }
 
         if (op_ret >= 0) {
-                STACK_WIND (frame, dht_newfile_cbk,
-                            local->cached_subvol,
+                STACK_WIND_COOKIE (frame, dht_newfile_cbk,
+                            (void *)local->cached_subvol, local->cached_subvol,
                             local->cached_subvol->fops->mknod,
                             &local->loc, local->mode, local->rdev,
-                            local->params);
+                            local->umask, local->params);
 
                 return 0;
         }
-
+err:
         DHT_STACK_UNWIND (link, frame, op_ret, op_errno,
-                          inode, stbuf, preparent, postparent);
+                          inode, stbuf, preparent, postparent, xdata);
         return 0;
 }
 
 
 int
-switch_mknod (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, mode_t mode, dev_t rdev, dict_t *params)
+switch_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+              dev_t rdev, mode_t umask, dict_t *params)
 {
         dht_local_t *local = NULL;
         dht_conf_t  *conf  = NULL;
         xlator_t    *subvol = NULL;
         xlator_t    *avail_subvol = NULL;
         int          op_errno = -1;
-        int          ret = -1;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -530,7 +518,7 @@ switch_mknod (call_frame_t *frame, xlator_t *this,
 
         dht_get_du_info (frame, this, loc);
 
-        local = dht_local_init (frame);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_MKNOD);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -549,59 +537,45 @@ switch_mknod (call_frame_t *frame, xlator_t *this,
         avail_subvol = get_switch_matching_subvol (loc->path, conf, subvol);
         if (dht_is_subvol_filled (this, avail_subvol)) {
                 avail_subvol =
-                        dht_free_disk_available_subvol (this, avail_subvol);
+                        dht_free_disk_available_subvol (this, avail_subvol,
+                                                        local);
         }
 
         if (avail_subvol != subvol) {
                 /* Create linkfile first */
-                ret = loc_copy (&local->loc, loc);
-                if (ret == -1) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
 
                 local->params = dict_ref (params);
                 local->mode = mode;
+                local->umask = umask;
                 local->rdev = rdev;
                 local->cached_subvol = avail_subvol;
 
                 dht_linkfile_create (frame, switch_mknod_linkfile_cbk,
-                                     avail_subvol, subvol, loc);
+                                     this, avail_subvol, subvol, loc);
                 return 0;
         }
 
         gf_log (this->name, GF_LOG_TRACE,
                 "creating %s on %s", loc->path, subvol->name);
 
-        STACK_WIND (frame, dht_newfile_cbk,
-                    subvol, subvol->fops->mknod,
-                    loc, mode, rdev, params);
+        STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol, subvol,
+                           subvol->fops->mknod, loc, mode, rdev, umask,
+                           params);
 
         return 0;
 
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
         DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL);
+                          NULL, NULL, NULL, NULL, NULL);
 
         return 0;
 }
 
 
-int
-notify (xlator_t *this, int event, void *data, ...)
-{
-        int ret = -1;
-
-        ret = dht_notify (this, event, data);
-
-        return ret;
-}
-
 void
-fini (xlator_t *this)
+switch_fini (xlator_t *this)
 {
-        int                   i = 0;
         dht_conf_t           *conf = NULL;
         struct switch_struct *trav = NULL;
         struct switch_struct *prev = NULL;
@@ -612,30 +586,14 @@ fini (xlator_t *this)
                 trav = (struct switch_struct *)conf->private;
                 conf->private = NULL;
                 while (trav) {
-                        if (trav->array)
-                                GF_FREE (trav->array);
+                        GF_FREE (trav->array);
                         prev = trav;
                         trav = trav->next;
                         GF_FREE (prev);
                 }
-
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
-
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
-
-                if (conf->subvolume_status)
-                        GF_FREE (conf->subvolume_status);
-
-                GF_FREE (conf);
         }
 
-        return;
+        dht_fini(this);
 }
 
 int
@@ -695,8 +653,10 @@ set_switch_pattern (xlator_t *this, dht_conf_t *conf,
                 dup_str = gf_strdup (switch_str);
                 switch_opt = GF_CALLOC (1, sizeof (struct switch_struct),
                                         gf_switch_mt_switch_struct);
-                if (!switch_opt)
+                if (!switch_opt) {
+                        GF_FREE (dup_str);
                         goto err;
+                }
 
                 pattern = strtok_r (dup_str, ":", &tmp_str1);
                 childs = strtok_r (NULL, ":", &tmp_str1);
@@ -706,9 +666,11 @@ set_switch_pattern (xlator_t *this, dht_conf_t *conf,
                                 "for all the unconfigured child nodes,"
                                 " hence neglecting current option");
                         switch_str = strtok_r (NULL, ";", &tmp_str);
+                        GF_FREE (switch_opt);
                         GF_FREE (dup_str);
                         continue;
                 }
+                GF_FREE (dup_str);
                 memcpy (switch_opt->path_pattern, pattern, strlen (pattern));
                 if (childs) {
                         dup_childs = gf_strdup (childs);
@@ -765,7 +727,6 @@ set_switch_pattern (xlator_t *this, dht_conf_t *conf,
                                 "option in unify volume. Exiting");
                         goto err;
                 }
-                GF_FREE (dup_str);
 
                 /* Link it to the main structure */
                 if (switch_buf) {
@@ -778,6 +739,7 @@ set_switch_pattern (xlator_t *this, dht_conf_t *conf,
                         /* First entry */
                         switch_buf = switch_opt;
                 }
+                switch_opt = NULL;
                 switch_str = strtok_r (NULL, ";", &tmp_str);
         }
 
@@ -834,19 +796,20 @@ set_switch_pattern (xlator_t *this, dht_conf_t *conf,
                         /* First entry */
                         switch_buf = switch_opt;
                 }
+                switch_opt = NULL;
         }
         /* */
         conf->private = switch_buf;
 
         return 0;
 err:
+        GF_FREE (switch_buf_array);
+        GF_FREE (switch_opt);
+
         if (switch_buf) {
-                if (switch_buf_array)
-                        GF_FREE (switch_buf_array);
                 trav = switch_buf;
                 while (trav) {
-                        if (trav->array)
-                                GF_FREE (trav->array);
+                        GF_FREE (trav->array);
                         switch_opt = trav;
                         trav = trav->next;
                         GF_FREE (switch_opt);
@@ -856,68 +819,18 @@ err:
 }
 
 
-int
-init (xlator_t *this)
+int32_t
+switch_init (xlator_t *this)
 {
         dht_conf_t            *conf = NULL;
         data_t                *data = NULL;
-        char                  *temp_str = NULL;
         int                    ret = -1;
-        int                    i = 0;
-        uint32_t               temp_free_disk = 0;
-
-        if (!this->children) {
-                gf_log (this->name, GF_LOG_CRITICAL,
-                        "SWITCH needs more than one subvolume");
-                return -1;
-        }
-
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile");
-        }
-
-        conf = GF_CALLOC (1, sizeof (*conf), gf_switch_mt_dht_conf_t);
-        if (!conf) {
-                goto err;
-        }
-
-        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
-        if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
-                /* If option is not "auto", other options _should_ be boolean */
-                if (strcasecmp (temp_str, "auto"))
-                        gf_string2boolean (temp_str, &conf->search_unhashed);
-                else
-                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
-        }
 
-        conf->unhashed_sticky_bit = 0;
-        if (dict_get_str (this->options, "unhashed-sticky-bit",
-                          &temp_str) == 0) {
-                gf_string2boolean (temp_str, &conf->unhashed_sticky_bit);
-        }
-
-        conf->min_free_disk = 10;
-        conf->disk_unit = 'p';
-
-        if (dict_get_str (this->options, "min-free-disk",
-                          &temp_str) == 0) {
-                if (gf_string2percent (temp_str,
-                                       &temp_free_disk) == 0) {
-                        if (temp_free_disk > 100) {
-                                gf_string2bytesize (temp_str,
-                                                    &conf->min_free_disk);
-                                conf->disk_unit = 'b';
-                        } else {
-                                conf->min_free_disk = (uint64_t)temp_free_disk;
-                                conf->disk_unit = 'p';
-                        }
-                } else {
-                        gf_string2bytesize (temp_str,
-                                            &conf->min_free_disk);
-                        conf->disk_unit = 'b';
-                }
+        ret = dht_init(this);
+        if (ret) {
+                return ret;
         }
+        conf = this->private;
 
         data = dict_get (this->options, "pattern.switch.case");
         if (data) {
@@ -928,65 +841,23 @@ init (xlator_t *this)
                 }
         }
 
-        ret = dht_init_subvolumes (this, conf);
-        if (ret == -1) {
-                goto err;
-        }
-
-        ret = dht_layouts_init (this, conf);
-        if (ret == -1) {
-                goto err;
-        }
-
-        LOCK_INIT (&conf->subvolume_lock);
-        LOCK_INIT (&conf->layout_lock);
-
-        conf->gen = 1;
-
-        conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
-                                    gf_switch_mt_dht_du_t);
-        if (!conf->du_stats) {
-                goto err;
-        }
-
-        /* Create 'syncop' environment */
-	conf->env = syncenv_new (0);
-        if (!conf->env) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to create sync environment %s",
-                        strerror (errno));
-                goto err;
-        }
-
         this->private = conf;
-
         return 0;
 
 err:
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
-
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
-
-                if (conf->subvolume_status)
-                        GF_FREE (conf->subvolume_status);
-
-                if (conf->du_stats)
-                        GF_FREE (conf->du_stats);
-
-                GF_FREE (conf);
-        }
-
+        dht_fini(this);
         return -1;
 }
 
 
+class_methods_t class_methods = {
+        .init           = switch_init,
+        .fini           = switch_fini,
+        .reconfigure    = dht_reconfigure,
+        .notify         = dht_notify
+};
+
+
 struct xlator_fops fops = {
         .lookup      = switch_lookup,
         .create      = switch_create,
@@ -1031,19 +902,3 @@ struct xlator_fops fops = {
 struct xlator_cbks cbks = {
         .forget     = dht_forget
 };
-
-
-struct volume_options options[] = {
-        { .key  = {"lookup-unhashed"},
-          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
-                    "on", "off"},
-          .type = GF_OPTION_TYPE_STR
-        },
-        { .key  = {"pattern.switch.case"},
-          .type = GF_OPTION_TYPE_ANY
-        },
-        { .key  = {"min-free-disk"},
-          .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
-        },
-        { .key  = {NULL} },
-};
diff --git a/xlators/cluster/dht/src/unittest/dht_layout_mock.c b/xlators/cluster/dht/src/unittest/dht_layout_mock.c
new file mode 100644
index 000000000..aa19ddc57
--- /dev/null
+++ b/xlators/cluster/dht/src/unittest/dht_layout_mock.c
@@ -0,0 +1,63 @@
+/*
+  Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "glusterfs.h"
+#include "xlator.h"
+#include "dht-common.h"
+#include "byte-order.h"
+
+int
+dht_hash_compute (xlator_t *this, int type, const char *name, uint32_t *hash_p)
+{
+    return 0;
+}
+
+int
+dht_inode_ctx_layout_get (inode_t *inode, xlator_t *this, dht_layout_t **layout)
+{
+    return 0;
+}
+
+int
+dht_inode_ctx_layout_set (inode_t *inode, xlator_t *this,
+                          dht_layout_t *layout_int)
+{
+    return 0;
+}
+
+int
+dict_get_ptr (dict_t *this, char *key, void **ptr)
+{
+    return 0;
+}
+
+int
+dict_get_ptr_and_len (dict_t *this, char *key, void **ptr, int *len)
+{
+    return 0;
+}
+
+int _gf_log (const char *domain, const char *file,
+             const char *function, int32_t line, gf_loglevel_t level,
+             const char *fmt, ...)
+{
+    return 0;
+}
+
+int _gf_log_callingfn (const char *domain, const char *file,
+                       const char *function, int32_t line, gf_loglevel_t level,
+                       const char *fmt, ...)
+{
+    return 0;
+}
diff --git a/xlators/cluster/dht/src/unittest/dht_layout_unittest.c b/xlators/cluster/dht/src/unittest/dht_layout_unittest.c
new file mode 100644
index 000000000..b5233d235
--- /dev/null
+++ b/xlators/cluster/dht/src/unittest/dht_layout_unittest.c
@@ -0,0 +1,124 @@
+/*
+  Copyright (c) 2008-2014 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "dht-common.h"
+#include "logging.h"
+#include "xlator.h"
+
+#include <stdarg.h>
+#include <stddef.h>
+#include <setjmp.h>
+#include <inttypes.h>
+#include <cmockery/pbc.h>
+#include <cmockery/cmockery.h>
+
+/*
+ * Helper functions
+ */
+
+static xlator_t *
+helper_xlator_init(uint32_t num_types)
+{
+    xlator_t *xl;
+    int i, ret;
+
+    REQUIRE(num_types > 0);
+
+    xl = test_calloc(1, sizeof(xlator_t));
+    assert_non_null(xl);
+    xl->mem_acct.num_types = num_types;
+    xl->mem_acct.rec = test_calloc(num_types, sizeof(struct mem_acct_rec));
+    assert_non_null(xl->mem_acct.rec);
+
+    xl->ctx = test_calloc(1, sizeof(glusterfs_ctx_t));
+    assert_non_null(xl->ctx);
+
+    for (i = 0; i < num_types; i++) {
+            ret = LOCK_INIT(&(xl->mem_acct.rec[i].lock));
+            assert_false(ret);
+    }
+
+    ENSURE(num_types == xl->mem_acct.num_types);
+    ENSURE(NULL != xl);
+
+    return xl;
+}
+
+static int
+helper_xlator_destroy(xlator_t *xl)
+{
+    int i, ret;
+
+    for (i = 0; i < xl->mem_acct.num_types; i++) {
+            ret = LOCK_DESTROY(&(xl->mem_acct.rec[i].lock));
+            assert_int_equal(ret, 0);
+    }
+
+    free(xl->mem_acct.rec);
+    free(xl->ctx);
+    free(xl);
+    return 0;
+}
+
+/*
+ * Unit tests
+ */
+static void
+test_dht_layout_new(void **state)
+{
+    xlator_t *xl;
+    dht_layout_t *layout;
+    dht_conf_t   *conf;
+    int cnt;
+
+    expect_assert_failure(dht_layout_new(NULL, 0));
+    expect_assert_failure(dht_layout_new((xlator_t *)0x12345, -1));
+    xl = helper_xlator_init(10);
+
+    // xl->private is NULL
+    assert_null(xl->private);
+    cnt = 100;
+    layout = dht_layout_new(xl, cnt);
+    assert_non_null(layout);
+    assert_int_equal(layout->type, DHT_HASH_TYPE_DM);
+    assert_int_equal(layout->cnt, cnt);
+    assert_int_equal(layout->ref, 1);
+    assert_int_equal(layout->gen, 0);
+    assert_int_equal(layout->spread_cnt, 0);
+    free(layout);
+
+    // xl->private is not NULL
+    cnt = 110;
+    conf = (dht_conf_t *)test_calloc(1, sizeof(dht_conf_t));
+    assert_non_null(conf);
+    conf->dir_spread_cnt = 12345;
+    conf->gen = -123;
+    xl->private = conf;
+
+    layout = dht_layout_new(xl, cnt);
+    assert_non_null(layout);
+    assert_int_equal(layout->type, DHT_HASH_TYPE_DM);
+    assert_int_equal(layout->cnt, cnt);
+    assert_int_equal(layout->ref, 1);
+    assert_int_equal(layout->gen, conf->gen);
+    assert_int_equal(layout->spread_cnt, conf->dir_spread_cnt);
+    free(layout);
+
+    free(conf);
+    helper_xlator_destroy(xl);
+}
+
+int main(void) {
+    const UnitTest tests[] = {
+        unit_test(test_dht_layout_new),
+    };
+
+    return run_tests(tests, "xlator_dht_layout");
+}
diff --git a/xlators/cluster/ha/src/Makefile.am b/xlators/cluster/ha/src/Makefile.am
index 5f78a2965..5c1364b7f 100644
--- a/xlators/cluster/ha/src/Makefile.am
+++ b/xlators/cluster/ha/src/Makefile.am
@@ -1,15 +1,16 @@
 xlator_LTLIBRARIES = ha.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/testing/cluster
 
-ha_la_LDFLAGS = -module -avoidversion 
+ha_la_LDFLAGS = -module -avoid-version 
 
 ha_la_SOURCES = ha-helpers.c ha.c
 ha_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
 noinst_HEADERS = ha.h
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	    -I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 CLEANFILES = 
 
diff --git a/xlators/cluster/ha/src/ha-helpers.c b/xlators/cluster/ha/src/ha-helpers.c
index 7c361547a..19be1ed27 100644
--- a/xlators/cluster/ha/src/ha-helpers.c
+++ b/xlators/cluster/ha/src/ha-helpers.c
@@ -1,22 +1,12 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 #include "xlator.h"
 #include "call-stub.h"
 #include "defaults.h"
diff --git a/xlators/cluster/ha/src/ha-mem-types.h b/xlators/cluster/ha/src/ha-mem-types.h
index b460588aa..e5e97d237 100644
--- a/xlators/cluster/ha/src/ha-mem-types.h
+++ b/xlators/cluster/ha/src/ha-mem-types.h
@@ -1,24 +1,13 @@
-
 /*
-   Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __HA_MEM_TYPES_H__
 #define __HA_MEM_TYPES_H__
 
diff --git a/xlators/cluster/ha/src/ha.c b/xlators/cluster/ha/src/ha.c
index 7bb1824a1..3eccb516b 100644
--- a/xlators/cluster/ha/src/ha.c
+++ b/xlators/cluster/ha/src/ha.c
@@ -1,22 +1,12 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 /* generate errors randomly, code is simple now, better alogorithm
  * can be written to decide what error to be returned and when
  */
@@ -1876,13 +1866,9 @@ err:
         }
 
         if (hafdp) {
-                if (hafdp->fdstate) {
-                        GF_FREE (hafdp->fdstate);
-                }
+                GF_FREE (hafdp->fdstate);
 
-                if (hafdp->path) {
-                        GF_FREE (hafdp->path);
-                }
+                GF_FREE (hafdp->path);
 
                 GF_FREE (hafdp);
         }
diff --git a/xlators/cluster/ha/src/ha.h b/xlators/cluster/ha/src/ha.h
index d6a519fa8..e2ed7eaa6 100644
--- a/xlators/cluster/ha/src/ha.h
+++ b/xlators/cluster/ha/src/ha.h
@@ -1,22 +1,12 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 #ifndef __HA_H_
 #define __HA_H_
 
diff --git a/xlators/cluster/map/src/Makefile.am b/xlators/cluster/map/src/Makefile.am
index 26e19137a..a278b05e2 100644
--- a/xlators/cluster/map/src/Makefile.am
+++ b/xlators/cluster/map/src/Makefile.am
@@ -1,15 +1,16 @@
 xlator_LTLIBRARIES = map.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/testing/cluster
 
-map_la_LDFLAGS = -module -avoidversion 
+map_la_LDFLAGS = -module -avoid-version 
 
 map_la_SOURCES = map.c map-helper.c
 map_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
 noinst_HEADERS = map.h
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	    -I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 CLEANFILES = 
 
diff --git a/xlators/cluster/map/src/map-helper.c b/xlators/cluster/map/src/map-helper.c
index 24ab6eb50..851397b68 100644
--- a/xlators/cluster/map/src/map-helper.c
+++ b/xlators/cluster/map/src/map-helper.c
@@ -1,22 +1,12 @@
 /*
-  Copyright (c) 2009-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2009-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 #ifndef _CONFIG_H
 #define _CONFIG_H
 #include "config.h"
diff --git a/xlators/cluster/map/src/map-mem-types.h b/xlators/cluster/map/src/map-mem-types.h
index 23c798dea..3e89f4736 100644
--- a/xlators/cluster/map/src/map-mem-types.h
+++ b/xlators/cluster/map/src/map-mem-types.h
@@ -1,24 +1,13 @@
-
 /*
-   Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __MAP_MEM_TYPES_H__
 #define __MAP_MEM_TYPES_H__
 
diff --git a/xlators/cluster/map/src/map.c b/xlators/cluster/map/src/map.c
index 73186e8cb..6150a33ce 100644
--- a/xlators/cluster/map/src/map.c
+++ b/xlators/cluster/map/src/map.c
@@ -1,22 +1,12 @@
 /*
-  Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2010-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 #ifndef _CONFIG_H
 #define _CONFIG_H
 #include "config.h"
@@ -2375,8 +2365,7 @@ fini (xlator_t *this)
 	priv = this->private;
 
 	if (priv) {
-		if (priv->xlarray)
-			GF_FREE (priv->xlarray);
+		GF_FREE (priv->xlarray);
 
 		trav_map = priv->map;
 		while (trav_map) {
diff --git a/xlators/cluster/map/src/map.h b/xlators/cluster/map/src/map.h
index 44ba3ee12..7703a543e 100644
--- a/xlators/cluster/map/src/map.h
+++ b/xlators/cluster/map/src/map.h
@@ -1,22 +1,12 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
-
 #ifndef __MAP_H__
 #define __MAP_H__
 
diff --git a/xlators/cluster/unify/Makefile.am b/xlators/cluster/nsr-client/Makefile.am
index d471a3f92..d471a3f92 100644
--- a/xlators/cluster/unify/Makefile.am
+++ b/xlators/cluster/nsr-client/Makefile.am
diff --git a/xlators/cluster/nsr-client/src/Makefile.am b/xlators/cluster/nsr-client/src/Makefile.am
new file mode 100644
index 000000000..4541ea01a
--- /dev/null
+++ b/xlators/cluster/nsr-client/src/Makefile.am
@@ -0,0 +1,33 @@
+noinst_PYTHON = gen-fops.py
+
+xlator_LTLIBRARIES = nsrc.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
+
+nsrc_la_LDFLAGS = -module -avoid-version
+nsrc_la_SOURCES = nsrc.c
+
+nsrc_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = fop-template.c \
+		 $(top_srcdir)/xlators/lib/src/libxlator.h \
+		 $(top_srcdir)/glusterfsd/src/glusterfsd.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) \
+	-I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/xlators/lib/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+XLATOR_HEADER	= $(top_srcdir)/libglusterfs/src/xlator.h
+
+CLEANFILES = nsrc-cg.c
+
+CODEGEN_DIR = ../../nsr-server/src/codegen.py
+
+nsrc-cg.c: gen-fops.py $(CODEGEN) $(XLATOR_HEADER) fop-template.c
+	$(PYTHON) ./gen-fops.py $(XLATOR_HEADER) fop-template.c > $@
+
+nsrc.lo: nsrc-cg.c
+
+uninstall-local:
+	rm -f $(DESTDIR)$(xlatordir)/nsr.so
diff --git a/xlators/cluster/nsr-client/src/fop-template.c b/xlators/cluster/nsr-client/src/fop-template.c
new file mode 100644
index 000000000..699b07d40
--- /dev/null
+++ b/xlators/cluster/nsr-client/src/fop-template.c
@@ -0,0 +1,113 @@
+// template-name fop
+$TYPE$
+nsrc_$NAME$ (call_frame_t *frame, xlator_t *this,
+             $ARGS_LONG$)
+{
+        nsrc_local_t    *local          = NULL;
+        xlator_t        *target_xl      = ACTIVE_CHILD(this);
+
+        local = mem_get(this->local_pool);
+        if (!local) {
+                goto err;
+        }
+
+        local->stub = fop_$NAME$_stub (frame, nsrc_$NAME$_continue,
+                                       $ARGS_SHORT$);
+        if (!local->stub) {
+                goto err;
+        }
+        local->curr_xl = target_xl;
+        local->scars = 0;
+
+        frame->local = local;
+        STACK_WIND_COOKIE (frame, nsrc_$NAME$_cbk, target_xl,
+                    target_xl, target_xl->fops->$NAME$,
+                    $ARGS_SHORT$);
+        return 0;
+
+err:
+        if (local) {
+                mem_put(local);
+        }
+        STACK_UNWIND_STRICT ($NAME$, frame, -1, ENOMEM,
+                             $DEFAULTS$);
+        return 0;
+}
+
+// template-name cbk
+$TYPE$
+nsrc_$NAME$_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno,
+                 $ARGS_LONG$)
+{
+        nsrc_local_t    *local          = frame->local;
+        xlator_t        *last_xl        = cookie;
+        xlator_t        *next_xl;
+        nsrc_private_t  *priv           = this->private;
+        struct timespec spec;
+
+        if (op_ret != (-1)) {
+                if (local->scars) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                HILITE("retried %p OK"), frame->local);
+                }
+                priv->active = last_xl;
+                goto unwind;
+        }
+        if ((op_errno != EREMOTE) && (op_errno != ENOTCONN)) {
+                goto unwind;
+        }
+
+        /* TBD: get leader ID from xdata? */
+        next_xl = next_xlator(this,last_xl);
+        /*
+         * We can't just give up after we've tried all bricks, because it's
+         * quite likely that a new leader election just hasn't finished yet.
+         * We also shouldn't retry endlessly, and especially not at a high
+         * rate, but that's good enough while we work on other things.
+         * 
+         * TBD: implement slow/finite retry via a worker thread
+         */
+        if (!next_xl || (local->scars >= SCAR_LIMIT)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        HILITE("ran out of retries for %p"), frame->local);
+                goto unwind;
+        }
+
+        local->curr_xl = next_xl;
+        local->scars += 1;
+        spec.tv_sec = 1;
+        spec.tv_nsec = 0;
+        /*
+         * WARNING
+         *
+         * Just calling gf_timer_call_after like this leaves open the
+         * possibility that writes will get reordered, if a first write is
+         * rescheduled and then a second comes along to find an updated
+         * priv->active before the first actually executes.  We might need to
+         * implement a stricter (and more complicated) queuing mechanism to
+         * ensure absolute consistency in this case.
+         */
+        if (gf_timer_call_after(this->ctx,spec,nsrc_retry_cb,local)) {
+                return 0;
+        }
+
+unwind:
+        call_stub_destroy(local->stub);
+        STACK_UNWIND_STRICT ($NAME$, frame, op_ret, op_errno,
+                                     $ARGS_SHORT$);
+        return 0;
+}
+
+// template-name cont-func
+$TYPE$
+nsrc_$NAME$_continue (call_frame_t *frame, xlator_t *this,
+                      $ARGS_LONG$)
+{
+        nsrc_local_t    *local  = frame->local;
+
+        STACK_WIND_COOKIE (frame, nsrc_$NAME$_cbk, local->curr_xl,
+                           local->curr_xl, local->curr_xl->fops->$NAME$,
+                           $ARGS_SHORT$);
+        return 0;
+}
diff --git a/xlators/cluster/nsr-client/src/gen-fops.py b/xlators/cluster/nsr-client/src/gen-fops.py
new file mode 100755
index 000000000..b07b3c5b1
--- /dev/null
+++ b/xlators/cluster/nsr-client/src/gen-fops.py
@@ -0,0 +1,57 @@
+#!/usr/bin/python
+
+# This script generates the boilerplate versions of most fops in the client,
+# mostly so that we can use STACK_WIND instead of STACK_WIND_TAIL (see
+# fop-template.c for the details).  The problem we're solving is that we sit
+# under DHT, which makes assumptions about getting callbacks only from its
+# direct children.  If we didn't define our own versions of these fops, the
+# default versions would use STACK_WIND_TAIL and the callbacks would come from
+# DHT's grandchildren.  The code-generation approach allows us to handle this
+# with a minimum of code, and also keep up with any changes to the fop table.
+
+import sys
+sys.path.append("../../nsr-server/src") # Blech.
+import codegen
+
+type_re = "([a-z_0-9]+)"
+name_re = "\(\*fop_([a-z0-9]+)_t\)"
+full_re = type_re + " *" + name_re
+fop_cg = codegen.CodeGenerator()
+fop_cg.skip = 2
+fop_cg.parse_decls(sys.argv[1],full_re)
+fop_cg.load_templates(sys.argv[2])
+
+# Use the multi-template feature to generate multiple callbacks from the same
+# parsed declarations.
+type_re = "([a-z_0-9]+)"
+name_re = "\(\*fop_([a-z0-9]+)_cbk_t\)"
+full_re = type_re + " *" + name_re
+cbk_cg = codegen.CodeGenerator()
+cbk_cg.skip = 5
+cbk_cg.parse_decls(sys.argv[1],full_re)
+cbk_cg.load_templates(sys.argv[2])
+
+# This is a nasty little trick to handle the case where a generated fop needs
+# a set of default arguments for the corresponding callback.
+#
+# Yes, it's ironic that I'm copying and pasting the generator code.
+fop_cg.make_defaults = cbk_cg.make_defaults
+
+# Sorry, getspec, you're not a real fop until someone writes a stub function
+# for you.
+del fop_cg.decls["getspec"]
+del cbk_cg.decls["getspec"]
+
+# cbk is used by both fop and continue, so emit first
+for f_name in cbk_cg.decls.keys():
+	cbk_cg.emit(f_name,"cbk")
+	print("")
+
+# continue is used by fop, so emit next
+for f_name in fop_cg.decls.keys():
+	fop_cg.emit(f_name,"cont-func")
+	print("")
+
+for f_name in fop_cg.decls.keys():
+	fop_cg.emit(f_name,"fop")
+	print("")
diff --git a/xlators/cluster/nsr-client/src/nsrc.c b/xlators/cluster/nsr-client/src/nsrc.c
new file mode 100644
index 000000000..4551a1432
--- /dev/null
+++ b/xlators/cluster/nsr-client/src/nsrc.c
@@ -0,0 +1,243 @@
+/*
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "call-stub.h"
+#include "defaults.h"
+#include "timer.h"
+#include "xlator.h"
+
+#define SCAR_LIMIT      20
+#define HILITE(x)       (""x"")
+
+/*
+ * The fops are actually generated by gen-fops.py; the rest was mostly copied
+ * from defaults.c (commit cd253754 on 27 August 2013).
+ */
+
+enum gf_dht_mem_types_ {
+        gf_mt_nsrc_private_t = gf_common_mt_end + 1,
+        gf_mt_nsrc_end
+};
+
+typedef struct {
+        xlator_t        *active;
+} nsrc_private_t;
+
+typedef struct {
+        call_stub_t     *stub;
+        xlator_t        *curr_xl;
+        uint16_t        scars;
+} nsrc_local_t;
+
+char            *NSRC_XATTR     = "user.nsr.active";
+
+static inline
+xlator_t *
+ACTIVE_CHILD (xlator_t *parent)
+{
+        nsrc_private_t  *priv   = parent->private;
+
+        return priv ? priv->active : FIRST_CHILD(parent);
+}
+
+xlator_t *
+next_xlator (xlator_t *this, xlator_t *prev)
+{
+        xlator_list_t   *trav;
+
+        for (trav = this->children; trav; trav = trav->next) {
+                if (trav->xlator == prev) {
+                        return trav->next ? trav->next->xlator
+                                          : this->children->xlator;
+                }
+        }
+
+        return NULL;
+}
+
+void
+nsrc_retry_cb (void *cb_arg)
+{
+        nsrc_local_t    *local  = cb_arg;
+
+        gf_log (__func__, GF_LOG_INFO, HILITE("retrying %p"), local);
+        call_resume_wind(local->stub);
+}
+
+#include "nsrc-cg.c"
+
+int32_t
+nsrc_forget (xlator_t *this, inode_t *inode)
+{
+        gf_log_callingfn (this->name, GF_LOG_WARNING, "xlator does not "
+                          "implement forget_cbk");
+        return 0;
+}
+
+
+int32_t
+nsrc_releasedir (xlator_t *this, fd_t *fd)
+{
+        gf_log_callingfn (this->name, GF_LOG_WARNING, "xlator does not "
+                          "implement releasedir_cbk");
+        return 0;
+}
+
+int32_t
+nsrc_release (xlator_t *this, fd_t *fd)
+{
+        gf_log_callingfn (this->name, GF_LOG_WARNING, "xlator does not "
+                          "implement release_cbk");
+        return 0;
+}
+
+struct xlator_fops fops = {
+        .lookup         = nsrc_lookup,
+        .stat           = nsrc_stat,
+        .fstat          = nsrc_fstat,
+        .truncate       = nsrc_truncate,
+        .ftruncate      = nsrc_ftruncate,
+        .access         = nsrc_access,
+        .readlink       = nsrc_readlink,
+        .mknod          = nsrc_mknod,
+        .mkdir          = nsrc_mkdir,
+        .unlink         = nsrc_unlink,
+        .rmdir          = nsrc_rmdir,
+        .symlink        = nsrc_symlink,
+        .rename         = nsrc_rename,
+        .link           = nsrc_link,
+        .create         = nsrc_create,
+        .open           = nsrc_open,
+        .readv          = nsrc_readv,
+        .writev         = nsrc_writev,
+        .flush          = nsrc_flush,
+        .fsync          = nsrc_fsync,
+        .opendir        = nsrc_opendir,
+        .readdir        = nsrc_readdir,
+        .readdirp       = nsrc_readdirp,
+        .fsyncdir       = nsrc_fsyncdir,
+        .statfs         = nsrc_statfs,
+        .setxattr       = nsrc_setxattr,
+        .getxattr       = nsrc_getxattr,
+        .fsetxattr      = nsrc_fsetxattr,
+        .fgetxattr      = nsrc_fgetxattr,
+        .removexattr    = nsrc_removexattr,
+        .fremovexattr   = nsrc_fremovexattr,
+        .lk             = nsrc_lk,
+        .inodelk        = nsrc_inodelk,
+        .finodelk       = nsrc_finodelk,
+        .entrylk        = nsrc_entrylk,
+        .fentrylk       = nsrc_fentrylk,
+        .rchecksum      = nsrc_rchecksum,
+        .xattrop        = nsrc_xattrop,
+        .fxattrop       = nsrc_fxattrop,
+        .setattr        = nsrc_setattr,
+        .fsetattr       = nsrc_fsetattr,
+	.fallocate	= nsrc_fallocate,
+	.discard        = nsrc_discard,
+};
+
+struct xlator_cbks cbks = {
+};
+
+
+int32_t
+mem_acct_init (xlator_t *this)
+{
+        int     ret = -1;
+
+        GF_VALIDATE_OR_GOTO ("nsrc", this, out);
+
+        ret = xlator_mem_acct_init (this, gf_mt_nsrc_end + 1);
+
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Memory accounting init" "failed");
+                return ret;
+        }
+out:
+        return ret;
+}
+
+
+int32_t
+nsrc_init (xlator_t *this)
+{
+        nsrc_private_t  *priv   = NULL;
+
+        this->local_pool = mem_pool_new (nsrc_local_t, 128);
+        if (!this->local_pool) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to create nsrc_local_t pool");
+                goto err;
+        }
+
+        priv = GF_CALLOC (1, sizeof (*priv), gf_mt_nsrc_private_t);
+        if (!priv) {
+               goto err;
+        }
+
+        priv->active = FIRST_CHILD(this);
+        this->private = priv;
+        return 0;
+
+err:
+        if (priv) {
+                GF_FREE(priv);
+        }
+        return -1;
+}
+
+void
+nsrc_fini (xlator_t *this)
+{
+        GF_FREE(this->private);
+}
+
+int32_t
+nsrc_notify (xlator_t *this, int32_t event, void *data, ...)
+{
+        int32_t         ret     = 0;
+
+        switch (event) {
+        case GF_EVENT_CHILD_DOWN:
+                /*
+                 * TBD: handle this properly
+                 *
+                 * What we really should do is propagate this only if it caused
+                 * us to lose quorum, and likewise for GF_EVENT_CHILD_UP only
+                 * if it caused us to gain quorum.  However, that requires
+                 * tracking child states and for now it's easier to swallow
+                 * these unconditionally.  The consequence of failing to do
+                 * this is that DHT sees the first GF_EVENT_CHILD_DOWN and gets
+                 * confused, so it doesn't call us and doesn't get up-to-date
+                 * directory listings etc.
+                 */
+                break;
+        default:
+                ret = default_notify (this, event, data);
+        }
+
+        return ret;
+}
+
+class_methods_t class_methods = {
+        .init           = nsrc_init,
+        .fini           = nsrc_fini,
+        .notify         = nsrc_notify,
+};
+
+struct volume_options options[] = {
+	{ .key = {NULL} },
+};
diff --git a/xlators/protocol/legacy/client/Makefile.am b/xlators/cluster/nsr-recon/Makefile.am
index d471a3f92..d471a3f92 100644
--- a/xlators/protocol/legacy/client/Makefile.am
+++ b/xlators/cluster/nsr-recon/Makefile.am
diff --git a/xlators/cluster/nsr-recon/src/Makefile.am b/xlators/cluster/nsr-recon/src/Makefile.am
new file mode 100644
index 000000000..e639e4437
--- /dev/null
+++ b/xlators/cluster/nsr-recon/src/Makefile.am
@@ -0,0 +1,23 @@
+xlator_LTLIBRARIES = nsr_recon.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
+
+nsr_recon_la_LDFLAGS = -module -avoid-version
+nsr_recon_la_SOURCES = recon_driver.c recon_xlator.c
+
+nsr_recon_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+				$(top_builddir)/api/src/libgfapi.la
+
+noinst_HEADERS = recon_driver.h recon_xlator.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) \
+	-I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/xlators/lib/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+XLATOR_HEADER	= $(top_srcdir)/libglusterfs/src/xlator.h
+
+CLEANFILES =
+
+uninstall-local:
+	rm -f $(DESTDIR)$(xlatordir)/nsr.so
diff --git a/xlators/cluster/nsr-recon/src/recon_driver.c b/xlators/cluster/nsr-recon/src/recon_driver.c
new file mode 100644
index 000000000..8c7622a02
--- /dev/null
+++ b/xlators/cluster/nsr-recon/src/recon_driver.c
@@ -0,0 +1,3130 @@
+/*
+   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.
+*/
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include <sys/types.h>
+#include <fcntl.h>
+#include <string.h>
+#include <unistd.h>
+#include <fnmatch.h>
+
+
+#include "call-stub.h"
+#include "defaults.h"
+#include "xlator.h"
+
+
+#include "recon_driver.h"
+#include "recon_xlator.h"
+#include "api/src/glfs-internal.h"
+#include "api/src/glfs-handles.h"
+
+/* TBD: move declarations here and nsr.c into a common place */
+#define NSR_TERM_XATTR          "trusted.nsr.term"
+#define RECON_TERM_XATTR        "trusted.nsr.recon-term"
+#define RECON_INDEX_XATTR       "trusted.nsr.recon-index"
+
+/*
+ * Execution architecture for the NSR reconciliation driver.  The driver runs
+ * as a seperate process in each node where the brick is.  The main function of
+ * the driver is nsr_reconciliation_driver() (last function below) The driver
+ * just sits in a tight loop waiting for state changes. When a brick becomes a
+ * replica leader, it fences IO, contacts this process and waits for
+ * reconciliation to finish.
+ *
+ * The replica leader talks to other bricks in replica group which are alive
+ * and gets the last term info using which it decides which has the latest
+ * data.  That brick is referred to as the "reconciliator"; leader sends a
+ * message to reconciliator to freeze its data (by reading any incomplete data
+ * from other nodes from that term if required)
+ *
+ * Once that is done leader sends a message to all nodes except the
+ * reconciliator to sync themselves with the reconciliator. This process is
+ * referred to as "resolution".
+ *
+ * Hence the reconciliation processes need to talk to each other to get a given
+ * term info.  This is implemented using the recon translator IOs which
+ * implements a bare bone RPC by exposing a file interface to which
+ * reads/writes are done to pass control messages.  This is referred to as the
+ * "control plane". This implementation allows the control plane to be
+ * implemented as a bunch of threads for each of the nodes.
+ *
+ * The reconciliation process also needs to talk to the brick process on that
+ * node to actually write the data as part of reconciliation/resolution. This
+ * is referred to as the "data plane". Again there are a bunch of threads that
+ * do this work.
+ *
+ * The way the worker threads are organised is that main driver context has a
+ * pointer to contexts for each of these thread contexts. The thread context at
+ * index 0 always refers to talking with local recon process/brick. So the
+ * control worker at index 0 will get the local changelog info and data worker
+ * at index 0 will talk to local brick.
+ *
+ * All the ops from the control/data planes are implemented using the glfs
+ * APIs.
+ */
+
+#if defined(NSR_DEBUG)
+
+/* This lets us change on the fly even if NSR_DEBUG is defined. */
+int nsr_debug_level = GF_LOG_TRACE;
+
+FILE *
+recon_create_log (char *member, char *module)
+{
+        char    *dpath  = NULL;
+        char    *p;
+        char    *fpath  = NULL;
+        FILE    *fp     = NULL;
+        int     fd      = -1;
+
+        (void)mkdir(NSR_LOG_DIR,0777);
+        (void)asprintf(&dpath,NSR_LOG_DIR"/%s",member);
+        if (dpath) {
+                for (p = dpath + strlen(NSR_LOG_DIR) + 1; *p; ++p) {
+                        if (*p == '/') {
+                                *p = '-';
+                        }
+                }
+                (void)mkdir(dpath,0777);
+                (void)asprintf(&fpath,"%s/%s",dpath,module);
+                if (fpath) {
+                        fd = open(fpath,O_WRONLY|O_CREAT|O_APPEND|O_SYNC,0666);
+                        if (fd >= 0) {
+                                fp = fdopen(fd,"a");
+                                if (!fp) {
+                                        close(fd);
+                                }
+                        }
+                        if (fp) {
+                                if (setvbuf (fp, NULL, _IONBF, 0)) {
+                                        /*
+                                         * Might as well take advantage of it
+                                         * to log the error.
+                                         */
+                                        fprintf (fp,
+                                                 "setvbuf failed for log\n");
+                                        fprintf (fp,
+                                                 "log output may be async\n");
+                                        fflush(fp);
+                                }
+                        }
+                        free(fpath);
+                }
+                free(dpath);
+        }
+
+        return fp;
+}
+
+void
+_nsr_driver_log (const char *func, int line, char *member, FILE *fp,
+                 char *fmt, ...)
+{
+        va_list         ap;
+        char            *buf    = NULL;
+        int             retval;
+
+        if (!fp) {
+                fp = recon_create_log(member,"nsr-driver-log");
+                if (!fp) {
+                        return;
+                }
+        }
+
+        va_start(ap,fmt);
+        retval = vasprintf(&buf,fmt,ap);
+        if (buf) {
+                fprintf(fp,"[%s:%d] %.*s\n",func,line,retval,buf);
+                free(buf);
+        }
+        va_end(ap);
+}
+
+void
+_nsr_worker_log (const char *func, int line, char *member,
+                 char *type, uint32_t index, FILE *fp,
+                 char *fmt, ...)
+{
+        va_list         ap;
+        char            *buf    = NULL;
+        int             retval;
+
+        if (!fp) {
+                char *name;
+                if (asprintf(&name,"%s-%u",type,index) < 1) {
+                        return;
+                }
+                fp = recon_create_log (member, name);
+                if (!fp) {
+                        return;
+                }
+        }
+
+        va_start(ap,fmt);
+        retval = vasprintf(&buf,fmt,ap);
+        if (buf) {
+                fprintf(fp,"[%s:%d] %.*s\n",func,line,retval,buf);
+                free(buf);
+        }
+        va_end(ap);
+}
+
+#endif
+
+/*
+ * Recon Driver Calloc
+ *
+ * We need this because all of this messing about with gfapi from within a
+ * translator keeps scrambling THIS (only one reason it's a terrible idea) and
+ * we need THIS to have a value that represents our initialization with our
+ * memory types.
+ *
+ * Note that the macro requires "this" to be defined in the current scope.
+ */
+
+#define RD_CALLOC(x,y,z) ({THIS = this; GF_CALLOC(x,y,z); })
+
+/*
+ * This function gets the size of all the extended attributes for a file.
+ * This is used so that caller knows how much to allocate for key-value storage.
+ *
+ * Input Arguments:
+ * fd - the file opened using glfs API.
+ * dict - passed so that NSR translator can get this from the required brick
+ *
+ * Output Arguments:
+ * b - pointer to the buffer where the attributes are filled up.
+ * key_size - the size of all keys
+ * val_size - the size of all values
+ * num - number of key/values
+ */
+static int32_t
+get_xattr_total_size( struct glfs_fd *fd,
+                     char **b,
+                     uint32_t *key_size,
+                     uint32_t *val_size,
+                     uint32_t* num,
+                     dict_t *dict)
+{
+        int32_t s = -1, ret = -1;
+        char *c = NULL;
+
+        *key_size = 0;
+        *val_size = 0;
+        *num = 0;
+
+        // First get the size of the keys
+        s = glfs_flistxattr_with_xdata(fd, NULL,0, dict);
+        if (s == -1) {
+                goto out;
+        }
+        *key_size = s;
+
+        // TBD - use the regular calloc
+        (*b) = c = calloc(s+1,1);
+
+        // get the keys themselves
+        if (glfs_flistxattr_with_xdata(fd, c, s+1, dict) == -1) {
+                goto out;
+        }
+        do {
+                int32_t r;
+                uint32_t len = 0;
+                // for each key get the size of the value
+                r = glfs_fgetxattr_with_xdata(fd, c, NULL, 0, dict);
+                if (r == -1)
+                        goto out;
+                (*val_size) += r;
+                len = strlen(c) + 1;
+                c += len;
+                s -= len;
+                (*num)++;
+        } while(s);
+        ret = 0;
+out:
+        return ret;
+}
+
+/*
+ * This function gets bunch of xattr values given set of keys.
+ *
+ * Input Arguments:
+ * fd - the file opened using glfs API.
+ * keys - the bunch of keys
+ * size - size of values
+ * num - number of keys
+ * dict - passed so that NSR translator can get this from the required brick
+ *
+ * Output Arguments:
+ * buf - where the values are written one after the other (NULL seperated)
+ */
+static int32_t
+get_xattr(struct glfs_fd *fd,
+          char *keys,
+          char *buf,
+          uint32_t size,
+          uint32_t num,
+          dict_t *dict)
+{
+        while(num--) {
+                int32_t r;
+                uint32_t len = 0;
+
+                // copy the key
+                strcpy(buf, keys);
+                len = strlen(keys);
+                len++;
+                buf += len;
+
+                // get the value and copy the value after incrementing buf after the key
+                r = glfs_fgetxattr_with_xdata(fd, keys, buf, size, dict);
+
+                // TBD - handle error
+                if (r == -1)
+                        return -1;
+
+                // increment the key to next value
+                keys += len;
+
+                // increment buf to hold the next key
+                buf += strlen(buf) + 1;
+        }
+        return 0;
+}
+
+/*
+ * Function deletes a bunch of key values in extended attributes of a file.
+ * Input Arguments:
+ * fd - the file opened using glfs API.
+ * dict - passed so that NSR translator can do this from the required brick
+ * keys - bunch of NULL seperated key names
+ * num - number of keys
+ */
+static int32_t delete_xattr(struct glfs_fd *fd,
+                         dict_t *dict_t,
+                         char *keys,
+                         uint32_t num)
+{
+        while(num--) {
+                // get the value and copy the value
+                // TBD - handle failure cases when calling glfs_fremovexattr_with_xdata()
+                if (glfs_fremovexattr_with_xdata(fd, keys, dict_t) == -1)
+			return -1;
+                keys += strlen(keys) +1;
+        }
+        return 0;
+}
+
+/*
+ * Given a bunch of key value pairs, fill them as xattrs for a file
+ *
+ * Input Arguments:
+ * fd - the file opened using glfs API.
+ * dict - passed so that NSR translator can do this from the required brick
+ * buf - buffer containing the keys-values pairs. The key value are NULL seperated.
+ *       Each of the key-value is seperated by NULL in turn.
+ * num - Number of such key value pairs.
+ */
+static int32_t
+fill_xattr(struct glfs_fd *fd,
+           dict_t *dict,
+           char *buf,
+           uint32_t num)
+{
+        char *k = buf, *val = NULL;
+
+        while(num--) {
+                int32_t r;
+
+                val = k + strlen(k) + 1;
+
+                // TBD - handle failure cases when calling glfs_fsetxattr_with_xdata()
+                r = glfs_fsetxattr_with_xdata(fd, k, val, strlen(val), 0, dict);
+                if (r == -1)
+                        return -1;
+                k = val + strlen(val) + 1;
+        }
+        return 0;
+}
+
+/*
+ * This function gets a file that can be used for doing glfs_init later.
+ * The control file is used by control thread(function) to talk to peer reconciliation process.
+ * The data file is used by the data thread(function) to talk to the bricks.
+ * The control file is of name such as con:gfs1:-mnt-a1 where "gfs1" is name of host
+ * and the brick path is "/mnt/a1". 
+ * The data file is of name such as data:gfs1:-mnt-a1.
+ *
+ * Input Arguments:
+ * vol - name of the volume. This is used to build the full path of the control and data file
+ * such as /var/lib/glusterd/vols/test/bricks/gfs2:-mnt-test1-nsr-recon.vol.
+ * In above example the volume name is test and brick on gfs2 is on path /mnt/test1
+ *
+ * worker - The worker for a given node. This worker has 2 threads - one on the data plane
+ * and one on the control plane. The worker->name is already filled with hostname:brickname
+ * in the function nsr_reconciliation_driver(). Use that to build the volume file.
+ * So if worker->name has  gfs1:/mnt/a1, control file is con:gfs1:-mnt-a1 
+ * and data file is data:gfs1:-mnt-a1.
+ * All these files are under the bricks directory. TBD - move this to a NSR recon directory later.
+ */
+static void
+nsr_recon_get_file(char *vol, nsr_replica_worker_t *worker)
+{
+        char *ptr;
+        char tr[256];
+
+        // Replace the "/" to -
+        strcpy(tr, worker->name);
+        ptr = strchr (tr, '/');
+        while (ptr) {
+                *ptr = '-';
+                ptr = strchr (tr, '/');
+        }
+
+        // Build the base directory such as "/var/lib/glusterd/vols/test/bricks/"
+        sprintf(worker->control_worker->vol_file,
+                "/%s/%s/%s/%s/",
+                GLUSTERD_DEFAULT_WORKDIR,
+                GLUSTERD_VOLUME_DIR_PREFIX,
+                vol,
+                GLUSTERD_BRICK_INFO_DIR);
+
+        strcat(worker->control_worker->vol_file, "con:");
+        strcat(worker->control_worker->vol_file, tr);
+
+        sprintf(worker->data_worker->vol_file,
+                "/%s/%s/%s/%s/",
+                GLUSTERD_DEFAULT_WORKDIR,
+                GLUSTERD_VOLUME_DIR_PREFIX,
+                vol,
+                GLUSTERD_BRICK_INFO_DIR);
+        strcat(worker->data_worker->vol_file, "data:");
+        strcat(worker->data_worker->vol_file, tr);
+}
+
+/*
+ * This function does all the glfs initialisation 
+ * so that reconciliation process can talk to other recon processes/bricks
+ * for the control/data messages.
+ * This will be done everytime a worker needs to be kicked off to talk
+ * across any plane.
+ *
+ * Input arguments:
+ * ctx - The per worker based context
+ * control - set to true if this worker is for the control plane
+ */
+static int32_t
+nsr_recon_start_work(nsr_per_node_worker_t *ctx,
+           gf_boolean_t control)
+{
+        glfs_t    *fs = NULL;
+        xlator_t *this = ctx->driver_ctx->this;
+        int32_t ret = 0;
+        glfs_fd_t *aux_fd = NULL; // fd of auxilary log
+        char lf[256];
+        nsr_recon_private_t *priv = NULL;
+        char *my_name = NULL;
+        char *morph_name = NULL, *ptr = NULL;
+
+        priv = this->private;
+        my_name = RD_CALLOC (1,
+                             strlen (priv->replica_group_members[0]) + 1,
+                             gf_mt_recon_member_name_t);
+        strcpy (my_name, priv->replica_group_members[0]);
+
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "starting work with volfile %s\n",
+                       ctx->vol_file);
+
+        fs = glfs_new(ctx->id);
+        if (!fs) {
+                glusterfs_this_set(this);
+                nsr_worker_log(this->name, GF_LOG_ERROR,
+                       "cannot create gfls context for thread %s\n",ctx->id);
+                return -1;
+        }
+
+        // For some vague reason, glfs init APIs seem to be clobbering "this". hence resetting it.
+        glusterfs_this_set(this);
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "init done. setting volfile %s\n",
+                       ctx->vol_file);
+
+        ret = glfs_set_volfile(fs, ctx->vol_file);
+        if (ret != 0) {
+                glusterfs_this_set(this);
+                nsr_worker_log(this->name, GF_LOG_ERROR,
+                       "cannot set volfile %s for thread %s\n",ctx->vol_file, ctx->id);
+                return -1;
+        }
+
+        morph_name = RD_CALLOC (1, strlen (my_name) + 1,
+                                gf_mt_recon_member_name_t);
+        strcpy (morph_name, my_name);
+
+        ptr = strchr (morph_name, '/');
+        while (ptr)
+        {
+                *ptr = '-';
+                 ptr = strchr (morph_name, '/');
+        }
+        // TBD - convert this to right /usr/local/var/log based log files.
+
+        sprintf(lf, NSR_LOG_DIR"/%s/%s-%"PRIu32, morph_name,
+                (control == _gf_true)?"glfs-con":"glfs-data", ctx->index);
+        ret =  glfs_set_logging (fs, lf, 7);
+        if (ret) {
+                glusterfs_this_set(this);
+                gf_log (this->name, GF_LOG_ERROR, "glfs logging set failed (%s)",
+                        strerror (errno));
+                return -1;
+        }
+
+        ret = glfs_init (fs);
+        if (ret != 0) {
+                glusterfs_this_set(this);
+                nsr_worker_log(this->name, GF_LOG_ERROR, "cannot do init for thread %s with volfile %s\n",ctx->id, ctx->vol_file);
+                return -1;
+        }
+        glusterfs_this_set(this);
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "setting volfile %s done\n",
+                       ctx->vol_file);
+
+        // If it is control thread, open the "/" as the aux_fd.
+        // All IOs happening via the fd will do the RPCs across the reconciliation
+        // processes. For some vague reason, the root seems to be open'able like a file.
+        // TBD - try to clean this up. (implement a virtual file???)
+        if (control == _gf_true) {
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "doing open for / \n");
+                aux_fd = glfs_open (fs, "/", O_RDWR);
+                // TBD - proper error handling. Stall reconciliation if such a thing happens?
+                if (aux_fd == NULL) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                               "cannot open  aux log file for thread %s\n",ctx->id);
+			return -1;
+                } else {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                               "---opened aux log file for thread %s\n",ctx->id);
+		}
+                ctx->aux_fd = aux_fd;
+        }
+        glusterfs_this_set(this);
+        ctx->fs = fs;
+        return 0;
+}
+
+/*
+ *
+ * This function does the cleanup after reconciliation is done
+ * or before we start a new reconciliation.
+ *
+ * Input arguments:
+ * ctx - The per worker based context
+ * control - set to true if this worker is for the control plane
+ */
+static int32_t
+nsr_recon_end_work(nsr_per_node_worker_t *ctx,
+         gf_boolean_t control)
+{
+        int32_t ret = 0;
+        xlator_t *this = ctx->driver_ctx->this;
+
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "doing fini for recon worker\n");
+
+        ret = glfs_fini(ctx->fs);
+        if (ret != 0) {
+                glusterfs_this_set(this);
+                nsr_worker_log(this->name, GF_LOG_ERROR, "cannot do fini for thread %s with volfile %s\n",ctx->id, ctx->vol_file);
+                return -1;
+        }
+        glusterfs_this_set(this);
+        ctx->fs = NULL;
+        if (control == _gf_true) {
+                glfs_close (ctx->aux_fd);
+                ctx->aux_fd = NULL;
+        }
+        return 0;
+}
+
+//called in case all worker functions run as sepeerate threads
+static void
+init_worker(nsr_per_node_worker_t *ctx, gf_boolean_t control)
+{
+        pthread_mutex_init(&(ctx->mutex), NULL);
+        pthread_cond_init(&(ctx->cv), NULL);
+        INIT_LIST_HEAD(&(ctx->head.list));
+}
+
+
+/*
+ * Control worker funct for getting changelog info on this node.
+ * calls directly functions to parse the changelog.
+ *
+ * Input arguments:
+ * ctx - The per worker based context
+ * control - set to true if this worker is for the control plane
+ */
+static void
+control_worker_func_0(nsr_per_node_worker_t *ctx,
+                      nsr_recon_work_t *work)
+{
+        unsigned int index = ctx->index;
+        nsr_replica_worker_t *rw = &(ctx->driver_ctx->workers[index]);
+        nsr_recon_driver_ctx_t *dr = ctx->driver_ctx;
+        xlator_t *this = dr->this;
+        nsr_recon_private_t *priv = this->private;
+
+        ctx->is_control = _gf_true;
+
+        switch (work->req_id) {
+        case NSR_WORK_ID_INI:
+            {
+                break;
+            }
+        case NSR_WORK_ID_FINI:
+            {
+                break;
+            }
+        case NSR_WORK_ID_GET_LAST_TERM_INFO:
+            {
+                nsr_recon_last_term_info_t lt;
+                nsr_reconciliator_info_t *recon_info = rw->recon_info;
+                // term is stuffed inside work->index. overloading.
+                int32_t term = work->index;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "trying to get last term info for  node %d with current term %d\n",index, term);
+
+                // TBD - handle errors
+                // This is called by the leader after it gets the current term.
+                // Makes searching easier.
+                nsr_recon_libchangelog_get_last_term_info(this, priv->changelog_base_path, term, &lt);
+                recon_info->last_term = lt.last_term;
+                recon_info->commited_ops = lt.commited_ops;
+                recon_info->last_index = lt.last_index;
+                recon_info->first_index = lt.first_index;
+
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "out of get last term info with current term %d. got ops %d with first %d and last %d \n",
+                               recon_info->last_term, recon_info->commited_ops,
+                               recon_info->first_index, recon_info->last_index);
+                break;
+            }
+        case NSR_WORK_ID_GET_GIVEN_TERM_INFO:
+            {
+                nsr_recon_last_term_info_t lt;
+                nsr_reconciliator_info_t *recon_info = rw->recon_info;
+                // term is stuffed inside work->index. overloading.
+                int32_t term = work->index;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "trying to get term info for  node %d for term %d\n",index, term);
+
+                // TBD - handle errors
+                nsr_recon_libchangelog_get_this_term_info(this,priv->changelog_base_path, term, &lt);
+
+                recon_info->last_term = lt.last_term;
+                recon_info->commited_ops = lt.commited_ops;
+                recon_info->last_index = lt.last_index;
+                recon_info->first_index = lt.first_index;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "out of get term info for term %d. got ops %d with first %d and last %d \n",
+                               recon_info->last_term, recon_info->commited_ops,
+                               recon_info->first_index, recon_info->last_index);
+
+                break;
+            }
+        case NSR_WORK_ID_RECONCILIATOR_DO_WORK:
+            {
+                // For local resolution, the main driver thread does it.
+                // SO there is no way we can have this message for this node.
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "this message should not be sent \n");
+                break;
+            }
+        case NSR_WORK_ID_RESOLUTION_DO_WORK:
+            {
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "this message should not be sent \n");
+                ctx->result = -1;
+                break;
+            }
+        case NSR_WORK_ID_GET_RECONCILATION_WINDOW:
+            {
+                nsr_reconciliator_info_t *recon_info = rw->recon_info;
+                // first_index and last_index at 0 indicates empty log.
+                // For non empty log, the first_index always starts at 1.
+                uint32_t num = (dr->workers[index].recon_info->last_index -
+                               dr->workers[index].recon_info->first_index + 1);
+                nsr_recon_record_details_t *rd;
+                uint32_t i=0;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "trying to get reconciliation window records for node %d for term %d with first %d last %d\n",
+                               index, recon_info->last_term, recon_info->first_index, recon_info->last_index);
+
+
+                // TBD - handle buffer allocation errors
+                rd = RD_CALLOC(num,
+                               sizeof(nsr_recon_record_details_t),
+                               gf_mt_recon_record_details_t);
+                if (rd == NULL) {
+                        ctx->result = -1;
+                        return;
+                }
+
+                recon_info->records = RD_CALLOC(num,
+                               sizeof(nsr_reconciliation_record_t),
+                               gf_mt_recon_record_t);
+                if (recon_info->records == NULL) {
+                        ctx->result = -1;
+                        return;
+                }
+
+                // TBD - handle errors
+                if (nsr_recon_libchangelog_get_records(this, priv->changelog_base_path,
+                                                   recon_info->last_term,
+                                                   recon_info->first_index,
+                                                   recon_info->last_index,
+                                                   rd) == _gf_false) {
+                        ctx->result = -1;
+                        return;
+                }
+
+                // The above function writes into rd from 0 to (num -1)
+                // We need to take care of this whenever we deal with records
+                for (i=0; i < num; i++) {
+                        ENDIAN_CONVERSION_RD(rd[i], _gf_true); //ntohl
+                        memcpy(&(recon_info->records[i].rec),
+                        &(rd[i]),
+                        sizeof(nsr_recon_record_details_t));
+                }
+
+                GF_FREE(rd);
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "got reconciliation window records for node %d for term %d \n",
+                               index, recon_info->last_term);
+                break;
+            }
+
+            default:
+                nsr_worker_log (this->name, GF_LOG_ERROR,
+                                "bad req id %u", work->req_id);
+        }
+
+        return;
+}
+
+// Control worker thread
+static void*
+control_worker_main_0(nsr_per_node_worker_t *ctx)
+{
+
+        ctx->is_control = _gf_true;
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "starting control worker func 0\n");
+
+        init_worker(ctx, 1);
+
+        while(1)
+        {
+                nsr_recon_work_t *work = NULL;
+                nsr_recon_driver_ctx_t *dr = ctx->driver_ctx;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "waiting for work\n");
+
+                pthread_mutex_lock(&ctx->mutex);
+                while (list_empty(&(ctx->head.list))) {
+                        pthread_cond_wait(&ctx->cv, &ctx->mutex);
+                }
+                pthread_mutex_unlock(&ctx->mutex);
+
+
+                list_for_each_entry(work, &(ctx->head.list), list) {
+                        nsr_worker_log(this->name, GF_LOG_INFO,
+                                       "got work with id %d\n", work->req_id);
+			work->in_use = _gf_false;
+
+                        // Call the main function.
+                        control_worker_func_0(ctx, work);
+
+                        atomic_dec(&(dr->outstanding));
+                        break;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,"deleting work item\n");
+		list_del_init (&work->list);
+                GF_FREE(work);
+                nsr_worker_log(this->name, GF_LOG_INFO,"finished deleting work item\n");
+        }
+
+        return NULL;
+}
+
+static void
+control_worker_do_reconciliation (nsr_per_node_worker_t *ctx,
+                                  nsr_recon_work_t *work)
+{
+        unsigned int index = ctx->index;
+        nsr_recon_driver_ctx_t *dr = ctx->driver_ctx;
+        nsr_recon_role_t rr;
+        uint32_t i=0;
+        uint32_t num=0;
+        uint32_t idx = dr->reconciliator_index;
+        uint32_t term = dr->workers[idx].recon_info->last_term;
+
+        GF_ASSERT(idx == index);
+
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "trying to make this index %d as reconciliator for term %d\n", index, term);
+
+        // TBD - error handling for all the glfs APIs
+        if (glfs_lseek(ctx->aux_fd,
+                   nsr_recon_xlator_sector_1,
+                   SEEK_SET) == -1) {
+                ctx->result = -1;
+                return;
+        }
+
+        // We have all the info for all other nodes.
+        // Fill all that info when sending data to that process.
+        for (i=0; i <  dr->replica_group_size; i++) {
+                if ( dr->workers[i].in_use &&
+                    (dr->workers[i].recon_info->last_term == term)) {
+                    rr.info[num].last_term =
+                        dr->workers[i].recon_info->last_term;
+                    rr.info[num].commited_ops =
+                        dr->workers[i].recon_info->commited_ops;
+                    rr.info[num].last_index =
+                        dr->workers[i].recon_info->last_index;
+                    rr.info[num].first_index =
+                        dr->workers[i].recon_info->first_index;
+                    strcpy(rr.info[num].name,
+                           dr->workers[i].name);
+                }
+                num++;
+        }
+        rr.num = num;
+        rr.role = reconciliator;
+        ENDIAN_CONVERSION_RR(rr, _gf_false); //htonl
+        if (glfs_write(ctx->aux_fd, &rr, sizeof(rr), 0) == -1) {
+                ctx->result = -1;
+                // Put the errno only for this case since we are bothered about
+                // retrying only for this case.  For rest of the cases we will
+                // just return EIO in errno.
+                ctx->op_errno = errno;
+                return;
+        }
+
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "sent reconciliator info for term %d with node count as %d\n", term, num);
+}
+
+static void
+control_worker_do_resolution (nsr_per_node_worker_t *ctx,
+                              nsr_recon_work_t *work)
+{
+        unsigned int index = ctx->index;
+        nsr_recon_driver_ctx_t *dr = ctx->driver_ctx;
+        nsr_recon_role_t rr;
+        unsigned int i=0, j=0;
+        unsigned int rec = dr->reconciliator_index;
+
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "trying to make this index %d as resolutor with reconciliator as %d\n",index, rec);
+
+        // TBD - error handling for all the glfs APIs
+        if (glfs_lseek(ctx->aux_fd,
+                   nsr_recon_xlator_sector_1,
+                   SEEK_SET) == -1) {
+                ctx->result = -1;
+                return;
+        }
+
+        rr.num = 2;
+
+        // Fill in info[0] as info for the node for which we are seeking
+        // resolution.  Fill in info[1] as info of the reconciliator node.  The
+        // function nsr_recon_driver_get_role() that will be called when this
+        // message reaches the node will look at index 1 for term information
+        // related to the reconciliator.
+        for (i=0; i <  2; i++) {
+                (i == 0) ? (j = index) : (j = rec);
+                rr.info[i].last_term =
+                        dr->workers[j].recon_info->last_term;
+                rr.info[i].commited_ops =
+                        dr->workers[j].recon_info->commited_ops;
+                rr.info[i].last_index =
+                        dr->workers[j].recon_info->last_index;
+                rr.info[i].first_index =
+                        dr->workers[j].recon_info->first_index;
+                // The name is used as the key to convert indices since the
+                // reconciliator index could be different across the nodes.
+                strcpy(rr.info[i].name,
+                        dr->workers[j].name);
+                if (i == 0) {
+                        nsr_worker_log(this->name, GF_LOG_INFO,
+                                       "this node info term=%d, ops=%d, first=%d, last=%d\n",
+                                       rr.info[i].last_term, rr.info[i].commited_ops,
+                                       rr.info[i].first_index,rr.info[i].last_index);
+                } else {
+                        nsr_worker_log(this->name, GF_LOG_INFO,
+                                       "reconciliator node info term=%d, ops=%d, first=%d, last=%d\n",
+                                       rr.info[i].last_term, rr.info[i].commited_ops,
+                                       rr.info[i].first_index,rr.info[i].last_index);
+                }
+        }
+        rr.role = resolutor;
+        ENDIAN_CONVERSION_RR(rr, _gf_false); //htonl
+        if (glfs_write(ctx->aux_fd, &rr, sizeof(rr), 0) == -1) {
+                ctx->result = -1;
+                // Put the errno only for this case since we are bothered about
+                // retrying only for this case.  For rest of the cases we will
+                // just return EIO in errno.
+                ctx->op_errno = errno;
+                return;
+        }
+
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "sent message to this node %d resolutor with reconciliator as %d\n", index, rec);
+}
+
+static void
+control_worker_get_window (nsr_per_node_worker_t *ctx, nsr_recon_work_t *work)
+{
+        unsigned int index = ctx->index;
+        nsr_replica_worker_t *rw = &(ctx->driver_ctx->workers[index]);
+        nsr_recon_driver_ctx_t *dr = ctx->driver_ctx;
+        xlator_t *this = dr->this;
+        nsr_recon_log_info_t li;
+        nsr_reconciliator_info_t *recon_info = rw->recon_info;
+        uint32_t i = 0;
+        uint32_t num = (dr->workers[index].recon_info->last_index -
+                       dr->workers[index].recon_info->first_index +1);
+        nsr_recon_record_details_t *rd;
+
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "trying to get reconciliation window records for node %d for term %d with first %d last %d\n",
+                       index, recon_info->last_term, recon_info->first_index, recon_info->last_index);
+
+        // TBD - error handling for all the glfs APIs
+        if (glfs_lseek(ctx->aux_fd, nsr_recon_xlator_sector_2, SEEK_SET) == -1) {
+                ctx->result = -1;
+                return;
+        }
+
+        // write to node what term & indices we are interested
+        li.term = recon_info->last_term;
+        li.first_index = recon_info->first_index;
+        li.last_index = recon_info->last_index;
+        ENDIAN_CONVERSION_LI(li, _gf_false); //htonl
+        if (glfs_write(ctx->aux_fd, &li, sizeof(li), 0) == -1) {
+                ctx->result = -1;
+                return;
+        }
+
+        // then read
+        rd = RD_CALLOC(num,
+                       sizeof(nsr_recon_record_details_t),
+                       gf_mt_recon_private_t);
+        if (rd == NULL) {
+                ctx->result = -1;
+                return;
+        }
+        recon_info->records = RD_CALLOC(num,
+                       sizeof(nsr_reconciliation_record_t),
+                       gf_mt_recon_private_t);
+        if (recon_info->records == NULL) {
+                ctx->result = -1;
+                goto err;
+        }
+
+        if (glfs_read(ctx->aux_fd, rd, num * sizeof(nsr_recon_record_details_t), 0) == -1) {
+                ctx->result = -1;
+                goto err;
+        }
+
+        for (i=0; i < num; i++) {
+                ENDIAN_CONVERSION_RD(rd[i], _gf_true); //ntohl
+                memcpy (&(recon_info->records[i].rec), &(rd[i]),
+                        sizeof(nsr_recon_record_details_t));
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "get_reconcilaition_window:Got %d at index %d\n",
+                               recon_info->records[i].rec.type,
+                               i + recon_info->first_index);
+        }
+
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "got reconciliation window records for node %d for term %d \n",
+                       index, recon_info->last_term);
+
+err:
+        GF_FREE(rd);
+}
+
+/*
+ * Control worker funct for getting changelog info on some other node.
+ * calls glfs functions to seek/read/write on aux_fd.
+ *
+ * Input arguments:
+ * ctx - The per worker based context
+ * control - set to true if this worker is for the control plane
+ */
+static void
+control_worker_func(nsr_per_node_worker_t *ctx,
+                    nsr_recon_work_t *work)
+{
+        unsigned int index = ctx->index;
+        nsr_replica_worker_t *rw = &(ctx->driver_ctx->workers[index]);
+        nsr_recon_last_term_info_t lt;
+        nsr_reconciliator_info_t *recon_info = rw->recon_info;
+        int32_t term = htonl(work->index); // overloading it
+
+        ctx->is_control = _gf_true;
+
+        switch (work->req_id){
+
+        case NSR_WORK_ID_INI:
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "calling nsr_recon_start_work\n");
+
+                // TBD - handle error in case nsr_recon_start_work gives error
+                if (nsr_recon_start_work(ctx, _gf_true) != 0) {
+                        ctx->result = -1;
+                        return;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "finished nsr_recon_start_work\n");
+                break;
+
+        case NSR_WORK_ID_FINI:
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "calling nsr_recon_end_work\n");
+
+                // TBD - handle error in case nsr_recon_end_work gives error
+                if (nsr_recon_end_work(ctx, _gf_true) != 0) {
+                        ctx->result = -1;
+                        return;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "finished nsr_recon_end_work\n");
+                break;
+
+        case NSR_WORK_ID_GET_LAST_TERM_INFO:
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "trying to get last term info for  node %d with current term %d\n",index, work->index);
+
+                // first write the current term term number
+                // TBD - error handling for all the glfs APIs
+                if (glfs_lseek(ctx->aux_fd, nsr_recon_xlator_sector_4, SEEK_SET) == -1) {
+                        ctx->result = -1;
+                        return;
+                }
+                if (glfs_write(ctx->aux_fd, &term, sizeof(term), 0) == -1) {
+                        ctx->result = -1;
+                        return;
+                }
+                if (glfs_read(ctx->aux_fd, &lt, sizeof(lt), 0) == -1) {
+                        ctx->result = -1;
+                        return;
+                }
+                ENDIAN_CONVERSION_LT(lt, _gf_true); //ntohl
+                recon_info->last_term = lt.last_term;
+                recon_info->commited_ops = lt.commited_ops;
+                recon_info->last_index = lt.last_index;
+                recon_info->first_index = lt.first_index;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "out of get last term info with current term %d. got ops %d with first %d and last %d \n",
+                               recon_info->last_term, recon_info->commited_ops,
+                               recon_info->first_index, recon_info->last_index);
+
+                break;
+
+        case NSR_WORK_ID_GET_GIVEN_TERM_INFO:
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "trying to get term info for  node %d for term %d\n",index, work->index);
+
+                // first write the term number
+                // TBD - error handling for all the glfs APIs
+                if (glfs_lseek(ctx->aux_fd, nsr_recon_xlator_sector_3, SEEK_SET) == -1) {
+                        ctx->result = -1;
+                        return;
+                }
+                if (glfs_write(ctx->aux_fd, &term, sizeof(term), 0) == -1) {
+                        ctx->result = -1;
+                        return;
+                }
+                if (glfs_read(ctx->aux_fd, &lt, sizeof(lt), 0) == -1) {
+                        ctx->result = -1;
+                        return;
+                }
+                ENDIAN_CONVERSION_LT(lt, _gf_true); //ntohl
+                recon_info->last_term = lt.last_term;
+                recon_info->commited_ops = lt.commited_ops;
+                recon_info->last_index = lt.last_index;
+                recon_info->first_index = lt.first_index;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "out of get term info for term %d. got ops %d with first %d and last %d \n",
+                               recon_info->last_term, recon_info->commited_ops,
+                               recon_info->first_index, recon_info->last_index);
+
+                break;
+
+        case NSR_WORK_ID_RECONCILIATOR_DO_WORK:
+                control_worker_do_reconciliation(ctx,work);
+                break;
+
+        case NSR_WORK_ID_RESOLUTION_DO_WORK:
+                control_worker_do_resolution(ctx,work);
+                break;
+
+        case NSR_WORK_ID_GET_RECONCILATION_WINDOW:
+                control_worker_get_window(ctx,work);
+                break;
+
+        default:
+                nsr_worker_log (this->name, GF_LOG_ERROR,
+                                "bad work type %d", work->req_id);
+        }
+
+        return;
+}
+
+// Control worker thread
+static void*
+control_worker_main(nsr_per_node_worker_t *ctx)
+{
+        unsigned int index = ctx->index;
+
+        ctx->is_control = _gf_true;
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "starting control worker func\n");
+
+        // if this is for local processing, call the changelog parsing calls directly
+        if (index == 0) {
+                control_worker_main_0(ctx);
+                return NULL;
+        }
+
+        init_worker(ctx, 1);
+
+
+        while(1)
+        {
+                nsr_recon_work_t *work = NULL;
+                nsr_recon_driver_ctx_t *dr = ctx->driver_ctx;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "waiting for work\n");
+
+                pthread_mutex_lock(&ctx->mutex);
+                while (list_empty(&(ctx->head.list))) {
+                        pthread_cond_wait(&ctx->cv, &ctx->mutex);
+                }
+                pthread_mutex_unlock(&ctx->mutex);
+
+
+                list_for_each_entry(work, &(ctx->head.list), list) {
+                        nsr_worker_log(this->name, GF_LOG_INFO,
+                                       "got work with id %d\n", work->req_id);
+			work->in_use = _gf_false;
+                        control_worker_func(ctx,work);
+                        atomic_dec(&(dr->outstanding));
+			break;
+                }
+                nsr_worker_log(this->name, GF_LOG_INFO,"deleting work item\n");
+		list_del_init (&work->list);
+                GF_FREE(work);
+                nsr_worker_log(this->name, GF_LOG_INFO,"finished deleting work item\n");
+        }
+
+        return NULL;
+}
+
+/*
+ * This function gets called if this process is chosen as the reconciliator
+ * for this replica group. It would have already got the records for the last term
+ * for the indices that are required (from the first HOLE to last index) from
+ * all other nodes that also witnessed that term. COmpare all the records and
+ * compute the work required.
+ *
+ * Input arguments
+ * ctx - driver context. All recon work is stored in workers[0].recon_info
+ */
+static void
+compute_reconciliation_work(nsr_recon_driver_ctx_t *ctx)
+{
+        uint32_t i=0, j=0;
+        nsr_reconciliator_info_t *my_recon = ctx->workers[0].recon_info;
+        uint32_t num = (my_recon->last_index - my_recon->first_index + 1);
+
+        for (i=0; i < num; i++) {
+                nsr_log_type_t orig, new;
+                unsigned int src = 0;
+                orig = new = my_recon->records[i].rec.type;
+                nsr_recon_work_type_t tw = NSR_RECON_WORK_NONE;
+                // index 0 means this node. Look at all other nodes.
+                for (j=1; j < ctx->replica_group_size; j++) {
+                        if (ctx->workers[j].in_use) {
+                                nsr_log_type_t pr = ctx->workers[j].recon_info->records[i].work.type;
+                                if ((new != pr) && (pr > new)) {
+                                        src = j;
+                                        new = (new | pr);
+                                }
+                        }
+                }
+                // TBD - compare data if new and orig are all FILLs. (can detect changelog corruption)
+                // Right now we compare if both orig and new are psuedo holes since
+                // only that is of interest to us.
+                if (orig != new) {
+                        if ((orig == NSR_LOG_HOLE) && (new == NSR_LOG_PSEUDO_HOLE))
+                                tw = NSR_RECON_WORK_HOLE_TO_PSEUDO_HOLE;
+                        else if ((orig == NSR_LOG_HOLE) && (new == NSR_LOG_FILL))
+                                tw = NSR_RECON_WORK_HOLE_TO_FILL;
+                        else if ((orig == NSR_LOG_PSEUDO_HOLE) && (new == NSR_LOG_PSEUDO_HOLE))
+                                tw = NSR_RECON_WORK_COMPARE_PSEUDO_HOLE;
+                        else if ((orig == NSR_LOG_PSEUDO_HOLE) && (new == NSR_LOG_FILL))
+                                tw = NSR_RECON_WORK_HOLE_TO_FILL;
+                }
+                if (tw != NSR_RECON_WORK_NONE) {
+                        my_recon->records[i].work.type = tw;
+                        my_recon->records[i].work.source = src;
+                        // Overwrite the record
+                        memcpy(&(my_recon->records[i].rec),
+                               &(ctx->workers[src].recon_info->records[i].rec),
+                               sizeof(nsr_recon_record_details_t));
+               }
+       }
+       return;
+}
+
+static int32_t
+nsr_recon_in_use(nsr_recon_driver_ctx_t *ctx,
+                 uint32_t i,
+                 gf_boolean_t in_use);
+
+/*
+ * Write the role and associated information to the node.
+ * This gets called from recon xlator indicating node is either
+ * leader, reconciliator or should do resolution.
+ */
+gf_boolean_t
+nsr_recon_driver_set_role(nsr_recon_driver_ctx_t *ctx,
+                          nsr_recon_role_t *rr,
+                          uint32_t term)
+{
+        nsr_role_work_t *rw;
+        xlator_t *this = ctx->this;
+
+        nsr_driver_log(this->name, GF_LOG_INFO, "set role called \n");
+        rw = RD_CALLOC(1, sizeof (nsr_role_work_t), gf_mt_recon_role_work_t);
+        memcpy(&rw->role, rr, sizeof(nsr_recon_role_t));
+        rw->term = term;
+        INIT_LIST_HEAD(&(rw->list));
+        pthread_mutex_lock(&(ctx->mutex));
+        list_add_tail(&rw->list, &ctx->role_head.list);
+        pthread_cond_signal(&(ctx->cv));
+        pthread_mutex_unlock(&(ctx->mutex));
+        nsr_driver_log(this->name, GF_LOG_INFO, "set role returns \n");
+        return _gf_true;
+}
+
+/*
+ * First we undo the last role to make sure we clean up.
+ *
+ * Input arguments
+ * ctx - driver context.
+ * rr - Role information. 
+ *      If leader, the thread now sends the list of all nodes that are part of
+ *      the current replica group. Use that to find out the activate the
+ *      required worker threads.
+ *      If reconciliator, the leader node would have sent information  about
+ *      all nodes which saw last term as the reconciliator.
+ *      If resolution to be done, then rr.info[0] will have this node's info
+ *      which the leader would have got earlier. rr[1].info will have the
+ *      info regarding the reconciliator.
+ * term - leader's term that is causing this role
+ */
+nsr_recon_driver_state_t
+nsr_recon_driver_get_role(int32_t *status,
+			  nsr_recon_driver_ctx_t *ctx,
+                          nsr_role_work_t *rw)
+{
+        uint8_t i=0, j=0;
+        nsr_recon_role_t *rr = &(rw->role);
+        nsr_reconciliator_info_t *tmp;
+        xlator_t *this = ctx->this;
+
+        // First make all the threads uninitialise
+        for (i = 0; i < ctx->replica_group_size; i++) {
+                if (nsr_recon_in_use(ctx, i, _gf_false) == -1) {
+			*status = -1;
+			return 0;
+		}
+        }
+
+        switch (rr->role) {
+        case leader:
+        case joiner:
+
+                // First set info this node
+                tmp = RD_CALLOC (1, sizeof (nsr_reconciliator_info_t),
+                                 gf_mt_recon_reconciliator_info_t);
+                if (!tmp) {
+			*status = -1;
+                        return 0;
+                }
+                ctx->workers[0].recon_info = tmp;
+                if (nsr_recon_in_use(ctx, 0, _gf_true) == -1) {
+			*status = -1;
+			return 0;
+		}
+                ctx->current_term = rr->current_term;
+
+                // Find rest of the nodes
+                for (i=1; i < ctx->replica_group_size; i++) {
+                        for (j=0 ; /* nothing */; j++) {
+                                if (j >= rr->num) {
+                                        nsr_driver_log (this->name, GF_LOG_ERROR,
+                                                        "failed to find %s",
+                                                        ctx->workers[i].name);
+                                        break;
+                                }
+                                if (strcmp(ctx->workers[i].name,
+                                           rr->info[j].name)) {
+                                        continue;
+                                }
+                                nsr_driver_log (this->name, GF_LOG_INFO,
+                                               "nsr_recon_driver_get_role: this as %s. found other server %s\n",
+                                               (rr->role == leader) ? "leader"
+                                                                    : "joiner",
+                                               ctx->workers[i].name);
+
+                                // Allocate this here. This will get later
+                                // filled when the leader tries to get last term
+                                // information from all the nodes
+                                tmp = RD_CALLOC (1,
+                                        sizeof (nsr_reconciliator_info_t),
+                                        gf_mt_recon_reconciliator_info_t);
+                                if (!tmp) {
+                                        *status = -1;
+                                        return 0;
+                                }
+                                ctx->workers[i].recon_info = tmp;
+                                if (nsr_recon_in_use(ctx, i, _gf_true) == -1) {
+                                        *status = -1;
+                                        return 0;
+                                }
+                                break;
+                        }
+                }
+                // If leader, reconciliator has to be chosen.
+                // If joiner, we are the reconciliator.
+                if (rr->role == leader)
+                        ctx->reconciliator_index = -1;
+                else
+                        ctx->reconciliator_index = 0;
+                break;
+
+        case reconciliator:
+                ctx->reconciliator_index = 0;
+                // Copy information about all the other members which had the
+                // same term
+                for (i=0; i < rr->num; i++) {
+                        for (j=0; /* nothing */; j++) {
+                                if (j >= ctx->replica_group_size) {
+                                        nsr_driver_log (this->name, GF_LOG_ERROR,
+                                                        "failed to find %s",
+                                                        rr->info[i].name);
+                                        break;
+                                }
+                                if (strcmp(rr->info[i].name,
+                                           ctx->workers[j].name)) {
+                                        continue;
+                                }
+                                nsr_driver_log(this->name, GF_LOG_INFO,
+                                               "nsr_recon_driver_get_role: this as reconciliator. found other server %s\n",
+                                               ctx->workers[j].name);
+                                tmp = RD_CALLOC (1,
+                                        sizeof (nsr_reconciliator_info_t),
+                                        gf_mt_recon_reconciliator_info_t);
+                                if (!tmp) {
+                                        *status = -1;
+                                        return 0;
+                                }
+                                tmp->last_term = rr->info[i].last_term;
+                                tmp->commited_ops = rr->info[i].commited_ops;
+                                tmp->last_index = rr->info[i].last_index;
+                                tmp->first_index = rr->info[i].first_index;
+                                ctx->workers[j].recon_info = tmp;
+                                if (nsr_recon_in_use(ctx, j, _gf_true) == -1) {
+                                        *status = -1;
+                                        return 0;
+                                }
+                                break;
+                        }
+                 }
+                 break;
+
+        case resolutor:
+                for (j=0; /* nothing */; j++) {
+                        // info[1] has the information regarding the
+                        // reconciliator
+                        if (j >= ctx->replica_group_size) {
+                                nsr_driver_log (this->name, GF_LOG_ERROR,
+                                                "failed to find %s",
+                                                rr->info[1].name);
+                                break;
+                        }
+                        if (strcmp(rr->info[1].name,
+                                   ctx->workers[j].name)) {
+                                continue;
+                        }
+                        nsr_driver_log(this->name, GF_LOG_INFO,
+                                       "nsr_recon_driver_get_role: this as resolutor. found other server %s as reconciliator\n",
+                                       ctx->workers[j].name);
+                        tmp = RD_CALLOC (1,
+                                sizeof (nsr_reconciliator_info_t),
+                                gf_mt_recon_reconciliator_info_t);
+                        if (!tmp) {
+                                 *status = -1;
+                                 return 0;
+                        }
+                        tmp->last_term = rr->info[1].last_term;
+                        tmp->commited_ops = rr->info[1].commited_ops;
+                        tmp->last_index = rr->info[1].last_index;
+                        tmp->first_index = rr->info[1].first_index;
+                        ctx->reconciliator_index = j;
+                        ctx->workers[j].recon_info = tmp;
+                        if (nsr_recon_in_use(ctx, j, _gf_true) == -1) {
+                                 *status = -1;
+                                 return 0;
+                        }
+                        GF_ASSERT(ctx->reconciliator_index != 0);
+                        break;
+                 }
+                 tmp = RD_CALLOC (1,
+                         sizeof (nsr_reconciliator_info_t),
+                         gf_mt_recon_reconciliator_info_t);
+                 if (!tmp) {
+			 *status = -1;
+			 return 0;
+                 }
+                 // info[0] has all info for this node
+                 tmp->last_term = rr->info[0].last_term;
+                 tmp->commited_ops = rr->info[0].commited_ops;
+                 tmp->last_index = rr->info[0].last_index;
+                 tmp->first_index = rr->info[0].first_index;
+                 ctx->workers[0].recon_info = tmp;
+                 if (nsr_recon_in_use(ctx, 0, _gf_true) == -1) {
+			 *status = -1;
+			 return 0;
+		 }
+        }
+
+        ctx->term = rw->term;
+
+	*status = 0;
+        return rr->role;
+}
+
+
+/*
+ * This function gets called if this process is chosen to sync itself with 
+ * the reconciliator.
+ *
+ * Input arguments
+ * ctx - driver context.
+ * my_info - local changelog info that has all the local records for indices that require work
+ * his_info - reconciliator's info that has all the golden copies
+ * invalidate - if set to true, then do not consult local records
+ */
+
+static void
+compute_resolution_work(nsr_recon_driver_ctx_t *ctx,
+                        nsr_reconciliator_info_t *my_info,
+                        nsr_reconciliator_info_t *his_info,
+                        gf_boolean_t invalidate)
+{
+        uint32_t i=0;
+        uint32_t num = (my_info->last_index - my_info->first_index + 1);
+        xlator_t *this = ctx->this;
+
+        if (invalidate) {
+                if (my_info->records) {
+                        GF_FREE(my_info->records);
+		}
+		my_info->records = RD_CALLOC(num,
+					     sizeof(nsr_reconciliation_record_t),
+					     gf_mt_recon_record_t);
+        }
+
+        for (i=0; i < num; i++) {
+                nsr_log_type_t orig, new;
+                nsr_recon_work_type_t tw = NSR_RECON_WORK_NONE;
+                orig = my_info->records[i].rec.type;
+                if (invalidate)
+                        orig = NSR_LOG_HOLE;
+                new  = his_info->records[i].rec.type;
+                // TBD - we can never have PSUEDO_HOLE in reconciliator's info
+                // We should have taken care of that during reconciliation.
+                // Put an assert to validate that.
+                if (new != orig) {
+                        if ((orig != NSR_LOG_FILL) && (new == NSR_LOG_FILL))
+                                tw = NSR_RECON_WORK_HOLE_TO_FILL;
+                        else if ((orig != NSR_LOG_HOLE) && (new == NSR_LOG_HOLE))
+                                tw = NSR_RECON_WORK_UNDO_FILL;
+                }
+                // copy the records anyway
+                my_info->records[i].work.type = tw;
+                my_info->records[i].work.source = ctx->reconciliator_index;
+                memcpy(&(my_info->records[i].rec),
+                       &(his_info->records[i].rec),
+                       sizeof(nsr_recon_record_details_t));
+       }
+       return;
+}
+
+
+// Create an glfs object
+static struct glfs_object *
+create_obj(nsr_per_node_worker_t *ctx, char  *gfid_str)
+{
+       struct glfs_object *obj = NULL;
+       uuid_t gfid;
+
+       uuid_parse(gfid_str, gfid);
+
+       obj = glfs_h_create_from_handle(ctx->fs, gfid, GFAPI_HANDLE_LENGTH, NULL);
+       if (obj == NULL) {
+                GF_ASSERT(obj != NULL);
+                nsr_worker_log(this->name, GF_LOG_ERROR,
+                               "creating of handle failed\n");
+                return NULL;
+        }
+        return obj;
+}
+
+/*
+ * Function to apply the actual record onto the local brick.
+ * prior to this we should have read all the data from the
+ * brick that has the data.
+ *
+ * Input parameters:
+ * ctx - per node worker context that has the fs for communicating to brick
+ * ri - Reconciliation record that needs fixup
+ * dict - So that NSR server translator on brick applis fixup only on this brick
+ *        and the changelog translator consumes term and index.
+ */
+
+static gf_boolean_t
+apply_record(nsr_per_node_worker_t *ctx,
+             nsr_reconciliation_record_t *ri,
+             dict_t * dict)
+{
+        struct glfs_fd *fd = NULL;
+        struct glfs_object *obj = NULL;
+        struct glfs_object *to_obj = NULL;
+        gf_boolean_t retval = _gf_false;
+
+        if (ri->rec.op == GF_FOP_WRITE) {
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "DOing write for file %s  @offset %d for len %d\n",
+                               ri->rec.gfid, ri->rec.offset, ri->rec.len);
+
+                // The file has got deleted on the source. Hence just ignore
+                // this.
+                // TBD - get a way to just stuff the log entry without writing
+                // the data so that changelogs remain identical.
+                if (ri->work.data == NULL) {
+                        return _gf_true;
+                }
+
+                if ((obj = create_obj(ctx,ri->rec.gfid)) == NULL)
+		        goto err;
+
+                fd = glfs_h_open_with_xdata(ctx->fs, obj, O_RDWR, dict);
+                if (fd == NULL) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "open for file %s failed\n",
+                                       ri->rec.gfid);
+                        goto err;
+                }
+                if (glfs_lseek_with_xdata(fd, ri->rec.offset, SEEK_SET, dict) != ri->rec.offset) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "lseek for file %s failed at offset %d\n",
+                                       ri->rec.gfid, ri->rec.offset);
+                        goto err;
+                }
+                if (glfs_write_with_xdata(fd, ri->work.data, ri->rec.len, 0, dict) != ri->rec.len) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "write for file %s failed for bytes %d\n",
+                                       ri->rec.gfid, ri->rec.len);
+                        goto err;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Finished DOing write for gfid %s  @offset %d for len %d\n",
+                               ri->rec.gfid, ri->rec.offset, ri->rec.len);
+
+        } else if (ri->rec.op == GF_FOP_FTRUNCATE) {
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "DOing truncate for file %s  @offset %d \n",
+                               ri->rec.gfid, ri->rec.offset);
+
+                if ((obj = create_obj(ctx, ri->rec.gfid)) == NULL) {
+                       goto err;
+                }
+
+                fd = glfs_h_open_with_xdata(ctx->fs, obj, O_RDWR, dict);
+                if (fd == NULL) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "open for file %s failed\n",
+                                       ri->rec.gfid);
+                        goto err;
+                }
+                if (glfs_ftruncate_with_xdata(fd, ri->rec.offset, dict) == -1) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "trunctae for file %s failed @offset %d\n",
+                                       ri->rec.gfid,ri->rec.offset );
+                        goto err;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Finished DOing truncate for gfid %s  @offset %d \n",
+                               ri->rec.gfid, ri->rec.offset);
+
+        } else if ((ri->rec.op == GF_FOP_FREMOVEXATTR) ||
+                   (ri->rec.op == GF_FOP_REMOVEXATTR) ||
+                   (ri->rec.op == GF_FOP_SETXATTR) ||
+                   (ri->rec.op == GF_FOP_FSETXATTR)) {
+
+                uint32_t k_s = 0, v_s = 0;
+                char *t_b= NULL;
+                uint32_t num = 0;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Doing set extended attr for file %s \n",
+                               ri->rec.gfid);
+
+                // The file has got deleted on the source. Hence just ignore
+                // this.  TBD - get a way to just stuff the log entry without
+                // writing the data so that changelogs remain identical.
+                if (ri->work.data == NULL) {
+                        return  _gf_true;
+                }
+
+                if ((obj = create_obj(ctx, ri->rec.gfid)) == NULL) {
+                       goto err;
+                }
+
+                if (obj->inode->ia_type  == IA_IFDIR)
+                        fd = glfs_h_opendir_with_xdata(ctx->fs, obj, dict);
+                else
+                        fd = glfs_h_open_with_xdata(ctx->fs, obj, O_RDWR, dict);
+                if (fd == NULL) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "open for file %s failed\n",
+                                       ri->rec.gfid);
+                        goto err;
+                }
+
+                if(get_xattr_total_size(fd, &t_b, &k_s, &v_s, &num, dict) == -1) {
+                        if (t_b) free(t_b);
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "list of xattr of  %s failed\n", ri->rec.gfid);
+                        goto err;
+                }
+
+                if (delete_xattr(fd, dict, t_b, num) == -1) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "deleting xattrs failed\n");
+                        goto err;
+		}
+
+                // Set one special dict flag to indicate the opcode so that
+                // the opcode gets set to this
+                if (dict_set_int32(dict,"recon-xattr-opcode",ri->rec.op)) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "setting opcode to %d failed\n",ri->rec.op);
+                        goto err;
+                }
+
+                if (fill_xattr(fd, dict, ri->work.data,  ri->work.num) == -1) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "filling xattrs failed\n");
+                        goto err;
+		}
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Finsihed Doing set extended attr for  %s \n",
+                               ri->rec.gfid);
+
+        } else if (ri->rec.op == GF_FOP_CREATE) {
+
+                uuid_t gfid;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Doing create for file %s \n",
+                               ri->rec.gfid);
+
+                // TBD - add mode and flags later
+                uuid_parse(ri->rec.gfid, gfid);
+                if ((obj = create_obj(ctx, ri->rec.pargfid)) == NULL) {
+                       goto err;
+                }
+
+                nsr_worker_log (this->name, GF_LOG_INFO,
+                                "creating with mode 0%o", ri->rec.mode);
+                if (glfs_h_creat_with_xdata(ctx->fs, obj, ri->rec.entry, O_RDWR, ri->rec.mode,  NULL, gfid, dict) == NULL) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "Failure for Doing create for file %s\n",
+                                       ri->rec.entry);
+                        goto err;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Finished Doing create for file %s \n",
+                               ri->rec.entry);
+
+        } else if (ri->rec.op == GF_FOP_MKNOD) {
+
+                uuid_t gfid;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Doing mknod for file %s \n",
+                               ri->rec.entry);
+
+                // TBD - add mode and flags later
+                uuid_parse(ri->rec.gfid, gfid);
+                if ((obj = create_obj(ctx, ri->rec.pargfid)) == NULL) {
+                       goto err;
+                }
+
+                if (glfs_h_mknod_with_xdata(ctx->fs, obj,  ri->rec.entry, O_RDWR, 0777, NULL, gfid, dict) == NULL) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "Failure for Doing mknod for file %s\n",
+                                       ri->rec.entry);
+                        goto err;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Finished Doing mknod for file %s \n",
+                               ri->rec.entry);
+
+        } else if (ri->rec.op == GF_FOP_MKDIR) {
+
+                uuid_t gfid;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Doing mkdir for dir %s \n",
+                               ri->rec.gfid);
+
+                // TBD - add mode and flags later
+                uuid_parse(ri->rec.gfid, gfid);
+                if ((obj = create_obj(ctx, ri->rec.pargfid)) == NULL) {
+                       goto err;
+                }
+
+                if (glfs_h_mkdir_with_xdata(ctx->fs, obj, ri->rec.entry, 0777, NULL, gfid, dict) != 0) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "Failure for Doing mkdir for file %s\n",
+                                       ri->rec.entry);
+                        goto err;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Finished Doing mkdir for file %s \n",
+                               ri->rec.entry);
+
+        } else if ((ri->rec.op == GF_FOP_RMDIR) || (ri->rec.op == GF_FOP_UNLINK)) {
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Doing rmdir/ublink for dir %s \n",
+                               ri->rec.entry);
+
+                if ((obj = create_obj(ctx, ri->rec.pargfid)) == NULL) {
+                       goto err;
+                }
+                if (glfs_h_unlink_with_xdata(ctx->fs, obj, ri->rec.entry, dict) != 0) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "Failure for Doing rmdir/unlink for file %s\n",
+                                       ri->rec.entry);
+                        goto err;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Finished Doing rmdir/unlink for file %s \n",
+                               ri->rec.entry);
+
+        } else if (ri->rec.op == GF_FOP_SYMLINK) {
+
+                uuid_t gfid;
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Doing symlink for file %s to file  %s \n",
+                               ri->rec.entry, ri->rec.link_path);
+
+                if ((obj = create_obj(ctx, ri->rec.pargfid)) == NULL) {
+                       goto err;
+                }
+                uuid_parse(ri->rec.gfid, gfid);
+
+                if (glfs_h_symlink_with_xdata(ctx->fs, obj,  ri->rec.entry, ri->rec.link_path, NULL, gfid, dict) == NULL) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "Failed to Doing symlink for file %s to file  %s \n",
+                                        ri->rec.entry, ri->rec.link_path);
+                        goto err;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Finished Doing symlink for file %s to file  %s \n",
+                               ri->rec.entry, ri->rec.link_path);
+
+        } else if (ri->rec.op == GF_FOP_LINK) {
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Doing hard link for file %s to file  %s \n",
+                               ri->rec.entry, ri->rec.gfid);
+
+                if ((obj = create_obj(ctx, ri->rec.pargfid)) == NULL) {
+                       goto err;
+                }
+                if ((to_obj = create_obj(ctx, ri->rec.gfid)) == NULL) {
+                       goto err;
+                }
+
+                if (glfs_h_link_with_xdata(ctx->fs, to_obj, obj, ri->rec.entry, dict) == -1) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "Failed to Doing hard link for file %s to file  %s \n",
+                                        ri->rec.entry, ri->rec.gfid);
+                        goto err;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Finsihed doing hard link for file %s to file  %s \n",
+                               ri->rec.entry, ri->rec.gfid);
+
+        } else if (ri->rec.op == GF_FOP_RENAME) {
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Doing rename for file %s to file  %s \n",
+                               ri->rec.entry, ri->rec.newloc);
+
+                if ((obj = create_obj(ctx, ri->rec.pargfid)) == NULL) {
+                       goto err;
+                }
+                if ((to_obj = create_obj(ctx, ri->rec.gfid)) == NULL) {
+                       goto err;
+                }
+
+                if (glfs_h_rename_with_xdata(ctx->fs, obj, ri->rec.entry, to_obj, ri->rec.newloc, dict) == -1) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "Failed to Doing rename for file %s to file  %s \n",
+                                       ri->rec.entry, ri->rec.newloc);
+                        goto err;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Finsihed doing renam for file %s to file  %s \n",
+                               ri->rec.entry, ri->rec.newloc);
+
+
+        } else if ((ri->rec.op == GF_FOP_SETATTR) || (ri->rec.op == GF_FOP_FSETATTR)) {
+
+                struct iatt      iatt = {0, };
+                int              valid = 0;
+                int              ret = -1;
+
+                // TBD - do the actual settings once we do that
+                // right now we just set the mode so that changelog gets filled
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Doing attr for file %s \n",
+                               ri->rec.gfid);
+
+                if ((obj = create_obj(ctx, ri->rec.gfid)) == NULL) {
+                       goto err;
+                }
+
+                if (obj->inode->ia_type  == IA_IFDIR)
+                        fd = glfs_h_opendir_with_xdata(ctx->fs, obj, dict);
+                else
+                        fd = glfs_h_open_with_xdata(ctx->fs, obj, O_RDWR, dict);
+                if (fd == NULL) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "open for file %s failed\n",
+                                       ri->rec.gfid);
+                        goto err;
+                }
+
+                iatt.ia_prot = ia_prot_from_st_mode(777);
+                valid = GF_SET_ATTR_MODE;
+
+
+                // Set one special dict flag to indicate the opcode so that
+                // the opcode gets set to this
+                if (dict_set_int32(dict,"recon-attr-opcode",ri->rec.op)) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "setting opcode to %d failed\n",ri->rec.op);
+                        goto err;
+                }
+
+                ret = glfs_fsetattr_with_xdata(fd, &iatt, valid, dict);
+                if (ret == -1) {
+                        nsr_worker_log(this->name, GF_LOG_INFO,
+                                       "failed Doing attr for file %s \n",
+                                       ri->rec.gfid);
+                        goto err;
+                }
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                               "Doing attr for file %s \n",
+                               ri->rec.gfid);
+
+        }
+
+        retval = _gf_true;
+
+err:
+        if (fd) {
+                /*
+                 * It's not clear that we should be passing the same dict to
+                 * glfs_close that was passed to us for glfs_open, but that's
+                 * the prior behavior so let's preserve it for now.
+                 */
+                if (glfs_close_with_xdata(fd, dict) == -1) {
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "close failed\n");
+		}
+        }
+        if (obj) {
+                /*
+                 * AFAICT fd operations do not borrow this reference, so we
+                 * still need to drop it ourselves.
+                 */
+                glfs_h_close(obj);
+        }
+        if (to_obj) {
+                /*
+                 * AFAICT fd operations do not borrow this reference, so we
+                 * still need to drop it ourselves.
+                 */
+                glfs_h_close(to_obj);
+        }
+        return retval;
+}
+
+//return back opcodes that requires reading from source
+static gf_boolean_t
+recon_check_changelog(nsr_recon_record_details_t *rd)
+{
+        return((rd->op == GF_FOP_WRITE) ||
+               (rd->op == GF_FOP_FSETATTR) ||
+               (rd-> op == GF_FOP_SETATTR) ||
+               (rd->op == GF_FOP_FREMOVEXATTR) ||
+               (rd->op == GF_FOP_SETXATTR) ||
+               (rd->op == GF_FOP_FSETXATTR) ||
+               (rd->op == GF_FOP_SYMLINK));
+
+}
+
+// TBD
+static gf_boolean_t
+recon_compute_undo(nsr_recon_record_details_t *rd)
+{
+        return(_gf_false);
+}
+
+
+/*
+ * Function that talks to the brick for data tranfer.
+ *
+ * Input arguments:
+ * ctx - worker context
+ * work - pointer to work object
+ */
+static void
+data_worker_func(nsr_per_node_worker_t *ctx,
+                 nsr_recon_work_t *work)
+{
+        nsr_recon_driver_ctx_t *dr = ctx->driver_ctx;
+        xlator_t *this = dr->this;
+        nsr_reconciliation_record_t *ri = NULL;
+        nsr_recon_record_details_t *rd = NULL;
+        int wip = 0;
+        dict_t * dict = NULL;
+        struct glfs_fd *fd = NULL;
+        struct glfs_object *obj = NULL;
+        uuid_t gfid;
+        uint32_t k_s = 0, v_s = 0;
+        char *t_b= NULL;
+        uint32_t num=0;
+
+        switch (work->req_id){
+        case NSR_WORK_ID_INI:
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                                "started data ini \n");
+
+                if (nsr_recon_start_work(ctx, _gf_false) != 0) {
+			ctx->result = -1;
+			return;
+		}
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                                "finished data ini \n");
+                break;
+        case NSR_WORK_ID_FINI:
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                                "started data fini \n");
+
+                if (nsr_recon_end_work(ctx, _gf_false) != 0) {
+			ctx->result = -1;
+			return;
+		}
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                                "finished data fini \n");
+                break;
+        case NSR_WORK_ID_SINGLE_RECONCILIATION_READ:
+                // first_index always starts with 1 but records starts at 0.
+                wip = work->index - (dr->workers[0].recon_info->first_index);
+                ri = &(dr->workers[0].recon_info->records[wip]);
+                rd = &(ri->rec);
+
+                dict = dict_new ();
+                if (!dict) {
+			ctx->result = -1;
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "failed allocating for dictionary\n");
+                        break;
+                }
+                if (dict_set_int32(dict,RECON_TERM_XATTR,ri->work.term)) {
+			ctx->result = -1;
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "error setting term in dict\n");
+                        break;
+                }
+                if (dict_set_int32(dict,RECON_INDEX_XATTR,ri->work.index)) {
+			ctx->result = -1;
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "error setting term in dict\n");
+                        break;
+                }
+
+                switch (rd->op) {
+                case GF_FOP_WRITE:
+
+                        // record already copied.
+                        // copy data to this node's info.
+
+                        uuid_parse(ri->rec.gfid, gfid);
+
+                        nsr_worker_log(this->name, GF_LOG_INFO,
+                                       "started recon read for file %s at offset %d at len %d\n",
+                                       ri->rec.gfid, rd->offset, rd->len);
+
+                        obj = glfs_h_create_from_handle (ctx->fs, gfid,
+                                                         GFAPI_HANDLE_LENGTH,
+                                                         NULL);
+                        if (obj == NULL) {
+                                GF_ASSERT(obj != NULL);
+                                nsr_worker_log(this->name, GF_LOG_ERROR,
+                                               "creating of handle failed\n");
+                                break;
+                        }
+
+                        // The file has probably got deleted.
+                        fd = glfs_h_open_with_xdata (ctx->fs, obj, O_RDONLY,
+                                                     dict);
+                        if (fd  == NULL) {
+                                GF_ASSERT(fd != NULL);
+                                nsr_worker_log(this->name, GF_LOG_ERROR,
+                                               "opening of file failed\n");
+                                break;
+                        }
+
+                        if (glfs_lseek_with_xdata (fd, rd->offset, SEEK_SET,
+                                                   dict) != rd->offset) {
+                                nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "lseek of file failed to offset %d\n",
+                                       rd->offset);
+                                break;
+                        }
+
+                        ri->work.data = RD_CALLOC (rd->len , sizeof(char),
+                                                   gf_mt_recon_work_data_t);
+                        if (glfs_read_with_xdata (fd, ri->work.data, rd->len,
+                                                  0, dict) != rd->len) {
+                                nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "read of file failed to offset %d for bytes %d\n",
+                                       rd->offset, rd->len);
+                                break;
+                        }
+                        break;
+
+                case GF_FOP_FTRUNCATE:
+                case GF_FOP_SYMLINK:
+                case GF_FOP_RMDIR:
+                case GF_FOP_UNLINK:
+                case GF_FOP_MKNOD:
+                case GF_FOP_CREATE:
+                case GF_FOP_LINK:
+                case GF_FOP_MKDIR:
+                case GF_FOP_RENAME:
+                        nsr_worker_log (this->name, GF_LOG_ERROR,
+                                        "unimplemented fop %u\n", rd->op);
+                        break;
+
+                case GF_FOP_FREMOVEXATTR:
+                case GF_FOP_REMOVEXATTR:
+                case GF_FOP_SETXATTR:
+                case GF_FOP_FSETXATTR:
+
+                        uuid_parse(ri->rec.gfid, gfid);
+
+
+                        // This is for all the set attribute/extended
+                        // attributes commands.  Get all the attributes from
+                        // the source and fill it in the buffer as a NULL
+                        // seperated key and value which are in turn seperated
+                        // by NULL.
+
+                        nsr_worker_log(this->name, GF_LOG_INFO,
+                                       "doing getattr for gfid %s \n",
+                                       ri->rec.gfid);
+
+                        obj = glfs_h_create_from_handle (ctx->fs, gfid,
+                                                         GFAPI_HANDLE_LENGTH,
+                                                         NULL);
+                        if (obj == NULL) {
+                                GF_ASSERT(fd != NULL);
+                                nsr_worker_log(this->name, GF_LOG_ERROR,
+                                               "creating of handle failed\n");
+                                break;
+                        }
+
+                        if (obj->inode->ia_type  == IA_IFDIR)
+                                fd = glfs_h_opendir_with_xdata (ctx->fs, obj,
+                                                                dict);
+                        else
+                                fd = glfs_h_open_with_xdata (ctx->fs, obj,
+                                                             O_RDONLY, dict);
+
+                        if (fd  == NULL) {
+                                GF_ASSERT(fd != NULL);
+                                nsr_worker_log(this->name, GF_LOG_ERROR,
+                                               "opening of file failed\n");
+                                break;
+                        }
+
+                        if (get_xattr_total_size (fd, &t_b, &k_s, &v_s, &num,
+                                                  dict) == -1) {
+                                if (t_b) free(t_b);
+                                nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "list of xattr of gfid %s failed\n",
+                                       rd->gfid);
+                                break;
+                        }
+                        ri->work.data = RD_CALLOC ((k_s + v_s) , sizeof(char),
+                                                   gf_mt_recon_work_data_t);
+                        if (get_xattr(fd, t_b, ri->work.data, v_s, num, dict) == -1) {
+				nsr_worker_log(this->name, GF_LOG_ERROR,
+					       "get xattr of gfid %s failed\n", rd->gfid);
+				break;
+			}
+                        ri->work.num = num;
+                        nsr_worker_log(this->name, GF_LOG_INFO,
+                                       "finished getattr for gfid %s \n",
+                                       ri->rec.gfid);
+                        free(t_b);
+                        break;
+
+                case GF_FOP_FSETATTR:
+                case GF_FOP_SETATTR:
+                        //TBD - to get the actual attrbutes and fill
+                        // mode, uid, gid, size, atime, mtime
+                        nsr_worker_log (this->name, GF_LOG_ERROR,
+                                        "unimplemented fop %u\n", rd->op);
+                        break;
+                default:
+                        nsr_worker_log (this->name, GF_LOG_ERROR,
+                                        "unrecognized fop %u\n", rd->op);
+
+                }
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                       "finished recon read for gfid %s at offset %d for %d bytes \n",
+                       rd->gfid, rd->offset, rd->len);
+                break;
+
+        case NSR_WORK_ID_SINGLE_RECONCILIATION_COMMIT:
+                // first_index always starts with 1 but records starts at 0.
+                wip = work->index - (dr->workers[0].recon_info->first_index);
+                ri = &(dr->workers[0].recon_info->records[wip]);
+                rd = &(ri->rec);
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                        "got recon commit for index %d that has gfid %s \n",
+                        wip, rd->gfid);
+                dict = dict_new ();
+                if (!dict) {
+			ctx->result = -1;
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "failed allocating for dictionary\n");
+                        break;
+                }
+                if (dict_set_int32(dict,RECON_TERM_XATTR,ri->work.term)) {
+			ctx->result = -1;
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "error setting term in dict\n");
+                        break;
+                }
+                if (dict_set_int32(dict,RECON_INDEX_XATTR,ri->work.index)) {
+			ctx->result = -1;
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "error setting term in dict\n");
+                        break;
+                }
+                if (apply_record(ctx, ri, dict) == _gf_false) {
+			nsr_worker_log(this->name, GF_LOG_ERROR,
+				       "apply_record fails\n");
+		}
+
+                nsr_worker_log(this->name, GF_LOG_INFO,
+                        "finished recon commit for gfid %s \n",
+                        rd->gfid);
+                break;
+
+        case NSR_WORK_ID_SINGLE_RECONCILIATION_FLUSH:
+                dict = dict_new ();
+                if (!dict) {
+			ctx->result = -1;
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "failed allocating for dictionary\n");
+                        break;
+                }
+                if (dict_set_int32(dict,RECON_TERM_XATTR,ri->work.term)) {
+			ctx->result = -1;
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "error setting term in dict\n");
+                        break;
+                }
+                if (dict_set_int32(dict,RECON_INDEX_XATTR,ri->work.index)) {
+			ctx->result = -1;
+                        nsr_worker_log(this->name, GF_LOG_ERROR,
+                                       "error setting term in dict\n");
+                        break;
+                }
+
+                // Increment work index with the start index
+                wip = work->index - (dr->workers[0].recon_info->first_index);
+                ri = &(dr->workers[0].recon_info->records[wip]);
+                rd = &(ri->rec);
+
+                glfs_fsync_with_xdata(fd, dict);
+                break;
+
+        default:
+                nsr_worker_log (this->name, GF_LOG_ERROR,
+                                "unrecognized request id %u\n", work->req_id);
+        }
+
+        if (fd) {
+                glfs_close_with_xdata(fd, dict);
+        }
+        if (obj) {
+                glfs_h_close(obj);
+        }
+        if (dict) {
+                dict_unref(dict);
+        }
+}
+
+// thread for doing data work
+static void *
+data_worker_main(nsr_per_node_worker_t *ctx)
+{
+        nsr_worker_log(this->name, GF_LOG_INFO,
+                       "starting data worker func\n");
+        init_worker(ctx, 0);
+
+        while(1) {
+                  nsr_recon_work_t *work = NULL;
+                  nsr_recon_driver_ctx_t *dr = ctx->driver_ctx;
+
+                  nsr_worker_log(this->name, GF_LOG_INFO,
+                                 "waiting for work\n");
+
+                  pthread_mutex_lock(&(ctx->mutex));
+                  while (list_empty(&(ctx->head.list))) {
+                        pthread_cond_wait(&(ctx->cv), &(ctx->mutex));
+                  }
+                  pthread_mutex_unlock(&(ctx->mutex));
+                  list_for_each_entry(work, &(ctx->head.list), list) {
+                        nsr_worker_log(this->name, GF_LOG_INFO,
+                                       "got work with id %d\n",work->req_id);
+			work->in_use = _gf_false;
+                        data_worker_func(ctx, work);
+                        atomic_dec(&(dr->outstanding));
+			break;
+                }
+                nsr_worker_log(this->name, GF_LOG_INFO,"deleting work item\n");
+		list_del_init (&work->list);
+                GF_FREE(work);
+                nsr_worker_log(this->name, GF_LOG_INFO,"finished deleting work item\n");
+        }
+
+        return NULL;
+}
+
+
+//make recon work 
+static void
+recon_make_work(nsr_recon_driver_ctx_t *ctx,
+                nsr_recon_work_t **work,
+                nsr_recon_work_req_id_t req_id,
+                int32_t i)
+{
+        xlator_t *this = ctx->this;
+
+        // TBD - change this to get from a static pool
+        // This cannot fail
+        (*work) =  RD_CALLOC (1, sizeof (nsr_recon_work_t), gf_mt_recon_work_t);
+        (*work)->req_id = req_id;
+        (*work)->index = i;
+	(*work)->in_use = _gf_true;
+        INIT_LIST_HEAD(&((*work)->list));
+        return;
+}
+
+// Schedule a work object to a worker thread.
+static void
+recon_queue_to_worker(nsr_recon_driver_ctx_t *ctx,
+                      nsr_recon_work_t *work,
+                      unsigned int id,
+                      nsr_recon_queue_type_t type)
+{
+        nsr_per_node_worker_t *worker;
+        if (type == NSR_RECON_QUEUE_TO_CONTROL) {
+		worker = ctx->workers[id].control_worker;
+                nsr_driver_log(this->name, GF_LOG_INFO,
+                               "queueing work to control index %d\n",id);
+        } else {
+		 worker= ctx->workers[id].data_worker;
+                nsr_driver_log(this->name, GF_LOG_INFO,
+                               "queueing work to data index %d\n",id);
+        }
+        pthread_mutex_lock(&worker->mutex);
+        list_add_tail(&work->list, &worker->head.list);
+        pthread_cond_signal(&worker->cv);
+        pthread_mutex_unlock(&worker->mutex);
+        return;
+}
+
+typedef void * (*F_t) (void *);
+
+// In case  mode is set to NSR_USE_THREADS, create worker threads.
+static gf_boolean_t
+create_worker_threads(nsr_recon_private_t *priv,
+                       nsr_recon_driver_ctx_t *ctx,
+                       nsr_per_node_worker_t *w,
+                       gf_boolean_t control_or_data,
+                       F_t f,
+                       uint32_t num)
+{
+        uint32_t i;
+        nsr_per_node_worker_t *worker = w;
+        xlator_t *this = ctx->this;
+
+        for (i=0; i < num; i++) {
+                worker->id = RD_CALLOC(1, 10, gf_mt_recon_id_t);
+                if (!worker->id) {
+                        nsr_driver_log (priv->this->name, GF_LOG_ERROR, "memory allocation error \n");
+                        return _gf_false;
+                }
+                sprintf(worker->id,"recon_%d", i);
+                worker->driver_ctx = ctx ;
+
+                if (ctx->mode == NSR_USE_THREADS) {
+                        if (pthread_create(&worker->thread_id, NULL, f, worker)) {
+                                nsr_driver_log (ctx->this->name, GF_LOG_ERROR, "control work thread creation error \n");
+                                return  _gf_false;
+                        }
+                }
+		worker->index = i;
+                worker++;
+        }
+        return _gf_true;
+}
+
+/*
+ * In case of thread, send the work item; else call the function directly.
+ *
+ * Input arguments:
+ * bm - bitmap containing indices of nodes we want to send work
+ * num - number of such indices
+ * ctx - driver context from where we derive per worker context
+ * id - request ID
+ * q - control or data
+ * misc - used to overload such as index.
+ */
+static void
+send_and_wait(int32_t *result,
+	      int32_t *op_errno,
+	      int32_t bm,
+              uint32_t num,
+              nsr_recon_driver_ctx_t *ctx,
+              nsr_recon_work_req_id_t id,
+              nsr_recon_queue_type_t q,
+              int32_t misc)
+{
+        uint32_t i = 0;
+	nsr_recon_work_t *work;
+
+#define CONTROL_WORKER(i) ctx->workers[i].control_worker
+#define DATA_WORKER(i) ctx->workers[i].data_worker
+#define WORKER(i) ((q == NSR_RECON_QUEUE_TO_CONTROL) ? (CONTROL_WORKER(i)) : (DATA_WORKER(i)))
+
+	*result = *op_errno = 0;
+
+	for (i=0; i < num; i++) {
+                if ((bm & (1 << i)) && ctx->workers[i].in_use) {
+			WORKER(i)->result = 0;
+			WORKER(i)->op_errno = 0;
+		}
+	}
+        if (ctx->mode == NSR_SEQ) {
+                for (i=0; i < num; i++) {
+                        if ((bm & (1 << i)) && ctx->workers[i].in_use) {
+                                recon_make_work(ctx,&work, id, misc);
+                                if (q == NSR_RECON_QUEUE_TO_CONTROL) {
+                                        if (i == 0)
+                                                control_worker_func_0(ctx->workers[0].control_worker, work);
+                                        else
+                                                control_worker_func(ctx->workers[i].control_worker, work);
+                                } else {
+                                        data_worker_func(ctx->workers[i].data_worker, work);
+                                }
+                                GF_FREE(work);
+                        }
+                }
+		goto out;
+        }
+
+        for (i=0; i < num; i++) {
+                if ((bm & (1 << i)) && ctx->workers[i].in_use) {
+			recon_make_work(ctx,&work, id, misc);
+                        atomic_inc(&(ctx->outstanding));
+                        recon_queue_to_worker(ctx, work, i, q);
+                }
+        }
+
+        nsr_driver_log(this->name, GF_LOG_INFO, "send_and_wait: waiting\n");
+        while (ctx->outstanding) {
+                pthread_yield();
+        }
+out:
+	for (i=0; i < num; i++) {
+                if ((bm & (1 << i)) && ctx->workers[i].in_use) {
+			if (WORKER(i)->result == -1) {
+				*result = -1;
+			}
+		}
+	}
+	if (*result == -1) {
+		for (i=0; i < num; i++) {
+			if ((bm & (1 << i)) && ctx->workers[i].in_use) {
+				if (WORKER(i)->op_errno == EAGAIN) {
+					*op_errno = EAGAIN;
+					break;
+				} else {
+					*op_errno = EIO;
+				}
+			}
+		}
+	}
+
+        nsr_driver_log(this->name, GF_LOG_INFO, "send_and_wait: all workers have returned with result: %d errno:%d\n", *result, *op_errno);
+        return;
+}
+
+// send INI or FINI
+static int32_t
+nsr_recon_in_use(nsr_recon_driver_ctx_t *ctx,
+                 uint32_t i,
+                 gf_boolean_t in_use)
+{
+        uint32_t bm = 1 << i;
+	gf_boolean_t send = _gf_false;
+	int32_t status =0, op_errno = 0;
+
+        send = (ctx->workers[i].in_use != in_use);
+        ctx->workers[i].in_use = in_use;
+
+	if (!send) {
+                return 0;
+        }
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "sending %s to index %d\n",in_use?"INI":"FINI",i);
+
+        send_and_wait(&status, &op_errno, bm, ctx->replica_group_size, ctx,
+              (in_use == _gf_true) ? NSR_WORK_ID_INI : NSR_WORK_ID_FINI,
+              NSR_RECON_QUEUE_TO_CONTROL, -1);
+        if (status == -1)
+                goto err;
+
+        send_and_wait(&status, &op_errno, bm, ctx->replica_group_size, ctx,
+              (in_use == _gf_true) ? NSR_WORK_ID_INI : NSR_WORK_ID_FINI,
+              NSR_RECON_QUEUE_TO_DATA, -1);
+        if (status == -1)
+                goto err;
+
+        /*
+         * We really need better error recovery.  To activate a worker, we
+         * allocate memory and send two messages.  If any of those actions
+         * fail, we should undo the others.  It would probably be good to
+         * collapse the two messages into one, because it's pretty trivial to
+         * allocate a temporary structure and either link it in or free it
+         * depending on the result here.
+         */
+
+        if (in_use == _gf_false) {
+                GF_FREE(ctx->workers[i].recon_info);
+        }
+
+	return 0;
+
+err:
+        GF_FREE(ctx->workers[i].recon_info);
+        ctx->workers[i].recon_info = NULL;
+        return -1;
+}
+
+gf_boolean_t
+nsr_recon_driver_reconciliator (nsr_recon_private_t *priv)
+{
+	uint32_t replica_group_size = priv->replica_group_size;
+        uint32_t i;
+        nsr_recon_driver_ctx_t *ctx = priv->driver_thread_context;
+        int32_t bm;
+	int32_t status = 0;
+	int32_t op_errno = 0;
+        gf_boolean_t do_recon = _gf_false;
+        uint32_t start_index = ctx->workers[0].recon_info->first_index;
+        uint32_t end_index = ctx->workers[0].recon_info->last_index;
+        uint32_t num = ((start_index == 0) && (end_index == 0)) ? 0 : (end_index - start_index + 1);
+
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "starting reconciliation work as reconciliator \n");
+
+        // nothing to be done? signal back to the recon translator that this
+        // phase done.
+        bm = 1;
+        for (i=1; i < replica_group_size; i++) {
+                if (ctx->workers[i].in_use &&
+                   (ctx->workers[0].recon_info->last_term == ctx->workers[i].recon_info->last_term)) {
+                        ctx->workers[i].recon_info->last_index = end_index;
+                        ctx->workers[i].recon_info->first_index = start_index;
+                        bm |= (1 << i);
+                        do_recon = _gf_true;
+                }
+        }
+
+        if (!do_recon || !num) {
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "nothing needs to be done as resolutor \n");
+                return _gf_true;
+        }
+
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "getting reconciliation window for term %d from %dto %d  \n",
+                        ctx->workers[0].recon_info->last_term,
+                        start_index, end_index);
+        // We have set the bm in the above for loop where we go thru all nodes
+        // including this node that have seen the last term.
+        send_and_wait(&status, &op_errno, bm,
+                      replica_group_size,
+                      ctx,
+                      NSR_WORK_ID_GET_RECONCILATION_WINDOW,
+                      NSR_RECON_QUEUE_TO_CONTROL, -1);
+        if (status == -1)
+                return _gf_false;
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "finished getting reconciliation window for term %d from %dto %d  \n",
+                        ctx->workers[0].recon_info->last_term,
+                        start_index, end_index);
+
+
+        // from the changelogs, calculate the entries that need action and the
+        // source for each of these entries
+        compute_reconciliation_work(ctx);
+
+        // for each of the entries that need fixup, issue IO
+        for (i=start_index; i < (start_index + num); i++) {
+                nsr_reconciliator_info_t *my_recon_info =
+                        ctx->workers[0].recon_info;
+                nsr_reconciliation_record_t *record =
+                        &(my_recon_info->records[i - start_index]);
+
+                record->work.term = ctx->workers[0].recon_info->last_term;
+                record->work.index = i;
+
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "fixing index %d\n",i);
+                if ((record->work.type == NSR_RECON_WORK_HOLE_TO_PSEUDO_HOLE) ||
+                    (record->work.type  == NSR_RECON_WORK_HOLE_TO_FILL)) {
+                        // 1st case (RECON_WORK_HOLE_TO_PSEUDO_HOLE): If there
+                        // are only pseudo_holes in others, it is best effort.
+                        // Just pick from the first node that has it and
+                        // proceed.
+                        // 2nd case (RECON_WORK_HOLE_TO_FILL): this node has
+                        // either a HOLE or PSUEDO_HOLE and some one else has a
+                        // FILL(source).  analyse the changelog to check if data
+                        // needs to be read or if the log has all the data
+                        // required
+
+                        if (recon_check_changelog(&record->rec)) {
+                                bm = (1 << record->work.source);
+                                nsr_driver_log (this->name, GF_LOG_INFO,
+                                                "reading data from source %d\n",record->work.source);
+                                send_and_wait(&status, &op_errno, bm,
+                                              replica_group_size,
+                                              ctx,
+                                              NSR_WORK_ID_SINGLE_RECONCILIATION_READ,
+                                              NSR_RECON_QUEUE_TO_DATA,
+                                              i);
+                                if (status == -1)
+                                         return _gf_false;
+                                nsr_driver_log (this->name, GF_LOG_INFO,
+                                                "got data from source %d\n",record->work.source);
+                        }
+
+                        nsr_driver_log (this->name, GF_LOG_INFO,
+                                        "fixing local data as part of reconciliation\n");
+
+                        bm = 1;
+                        send_and_wait(&status, &op_errno, bm,
+                                      replica_group_size,
+                                      ctx,
+                                      NSR_WORK_ID_SINGLE_RECONCILIATION_COMMIT,
+                                      NSR_RECON_QUEUE_TO_DATA,
+                                      i);
+                        if (status == -1)
+                                 return _gf_false;
+
+                        nsr_driver_log (this->name, GF_LOG_INFO,
+                                        "finished fixing local data as part of reconciliation\n");
+
+                } else if (record->work.type == NSR_RECON_WORK_COMPARE_PSEUDO_HOLE) {
+                        // this node has a pseudo_hole and some others have just
+                        // that too. Just convert this to FILL.  let others
+                        // blindly pick it from here.
+                        nsr_driver_log (this->name, GF_LOG_INFO,
+                                        "fixing this record as a fill\n");
+                        bm = 1;
+                        send_and_wait(&status, &op_errno, bm,
+                                      replica_group_size,
+                                      ctx,
+                                      NSR_WORK_ID_SINGLE_RECONCILIATION_FLUSH,
+                                      NSR_RECON_QUEUE_TO_DATA,
+                                      i);
+                        if (status == -1)
+                                 return _gf_false;
+                        nsr_driver_log (this->name, GF_LOG_INFO,
+                                        "finished fixing this record as a fill\n");
+                }
+        }
+
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "finished reconciliation work as reconciliator \n");
+
+        // tbd - mark this term golden in the reconciliator
+        return _gf_true;
+}
+
+gf_boolean_t
+nsr_recon_driver_resolutor (nsr_recon_private_t *priv)
+{
+	uint32_t replica_group_size = priv->replica_group_size;
+        uint32_t i;
+        nsr_recon_driver_ctx_t *ctx = priv->driver_thread_context;
+        int32_t bm;
+	int32_t status = 0;
+	int32_t op_errno = 0;
+        // This node's last term is filled when it gets a message from the
+        // leader to act as a reconciliator.
+        uint32_t recon_index = ctx->reconciliator_index;
+        nsr_reconciliator_info_t *my_info =
+                ctx->workers[0].recon_info;
+        nsr_reconciliator_info_t *his_info =
+                ctx->workers[recon_index].recon_info;
+        uint32_t my_last_term = my_info->last_term;
+        uint32_t to_do_term = his_info->last_term;
+        uint32_t my_start_index = 1, my_end_index = 1;
+        uint32_t his_start_index = 1, his_end_index = 1;
+        uint32_t num = 0;
+        gf_boolean_t fl = _gf_true;
+
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "starting resolutor work with reconciliator as %d from term %d to term %d \n",
+                        recon_index, my_last_term, to_do_term);
+
+        do {
+
+                if (!fl) {
+                        (his_info->last_term)++;
+                        (my_info->last_term)++;
+                } else {
+                        his_info->last_term = my_last_term;
+                }
+
+                nsr_driver_log (this->name, GF_LOG_INFO, "resolving term %d \n", my_info->last_term);
+
+                // Get reconciliator's term information for that term
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "getting info from reconciliator for term %d \n", my_info->last_term);
+                bm = (1 << recon_index);
+                send_and_wait(&status, &op_errno, bm,
+                              replica_group_size,
+                              ctx,
+                              NSR_WORK_ID_GET_GIVEN_TERM_INFO,
+                              NSR_RECON_QUEUE_TO_CONTROL, his_info->last_term);
+                if (status == -1)
+                         return _gf_false;
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "finished getting info from reconciliator for term %d \n", my_info->last_term);
+
+
+                // empty term
+                if (!his_info->commited_ops) {
+                        nsr_driver_log (this->name, GF_LOG_INFO,
+                                "reconciliator for term %d is empty. moving to next term. \n", my_info->last_term);
+                        // TBD - mark the term golden
+                        fl = _gf_false;
+                        continue;
+                }
+
+                // calculate the resolution window boundary.  for the last term
+                // this node saw, we compare the resolution window of this and
+                // reconciliator.  for the rest of the nodes, we just accept the
+                // reconciliator info.
+                if (fl) {
+                        my_start_index = my_info->first_index;
+                        my_end_index = my_info->last_index;
+                        his_start_index = his_info->first_index;
+                        his_end_index = his_info->last_index;
+                        my_info->first_index = (my_start_index < his_start_index) ? my_start_index : his_start_index;
+                        my_info->last_index = (my_end_index >  his_end_index) ? my_end_index : his_end_index;
+                } else {
+                        my_info->first_index = his_info->first_index;
+                        my_info->last_index = his_info->last_index;
+                        my_info->commited_ops = his_info->commited_ops;
+                }
+                if (my_info->first_index == 0)
+                        my_info->first_index = 1;
+                num = (my_info->last_index - my_info->first_index) + 1;
+
+
+                // Get the logs from the reconciliator (and this node for this
+                // term)
+                if (fl)
+                        bm = ((1 << recon_index) | 1);
+                else
+                        bm = (1 << recon_index);
+
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "getting reconciliation window for term %d from %d to %d  \n",
+                                my_info->last_term,
+                                my_info->first_index, my_info->last_index);
+                send_and_wait(&status, &op_errno, bm,
+                              replica_group_size,
+                              ctx,
+                              NSR_WORK_ID_GET_RECONCILATION_WINDOW,
+                              NSR_RECON_QUEUE_TO_CONTROL, -1);
+                if (status == -1)
+                         return _gf_false;
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "finished getting reconciliation window for term %d from %d to %d  \n",
+                                my_info->last_term,
+                                my_info->first_index, my_info->last_index);
+
+                // from the changelogs, calculate the entries that need action
+                compute_resolution_work(ctx, my_info, his_info, !fl);
+
+
+                // for each of the entries that need fixup, issue IO
+                for (i=my_info->first_index; i < (my_info->first_index + num); i++) {
+                        nsr_reconciliation_record_t *record =
+                                &(my_info->records[i - my_info->first_index]);
+
+                        record->work.term = my_info->last_term;
+                        record->work.index = i;
+
+                        nsr_driver_log (this->name, GF_LOG_INFO,
+                                        "fixing index %d\n",i);
+                        if ((record->work.type == NSR_RECON_WORK_HOLE_TO_FILL) ||
+                            (record->work.type == NSR_RECON_WORK_UNDO_FILL)) {
+                                if (((record->work.type == NSR_RECON_WORK_HOLE_TO_FILL) &&
+                                      recon_check_changelog(&record->rec)) ||
+                                    ((record->work.type == NSR_RECON_WORK_UNDO_FILL) &&
+                                      recon_compute_undo(&record->rec))) {
+                                        nsr_driver_log (this->name, GF_LOG_INFO,
+                                                        "reading data from source %d\n",recon_index);
+                                        bm = (1 << recon_index);
+                                        send_and_wait(&status, &op_errno, bm,
+                                                      replica_group_size,
+                                                      ctx,
+                                                      NSR_WORK_ID_SINGLE_RECONCILIATION_READ,
+                                                      NSR_RECON_QUEUE_TO_DATA,
+                                                      i);
+                                        if (status == -1)
+                                                return _gf_false;
+                                        nsr_driver_log (this->name, GF_LOG_INFO,
+                                                        "finished reading data from source %d\n",recon_index);
+                                }
+
+                                nsr_driver_log (this->name, GF_LOG_INFO,
+                                                "fixing local data as part of resolutor\n");
+
+                                bm = 1;
+                                send_and_wait(&status, &op_errno, bm,
+                                              replica_group_size,
+                                              ctx,
+                                              NSR_WORK_ID_SINGLE_RECONCILIATION_COMMIT,
+                                              NSR_RECON_QUEUE_TO_DATA,
+                                              i);
+                                if (status == -1)
+                                        return _gf_false;
+
+                                nsr_driver_log (this->name, GF_LOG_INFO,
+                                                "finished fixing local data as part of resolutor\n");
+                        }
+                }
+                fl = _gf_false;
+
+                // tbd - mark this term golden in the reconciliator
+        } while (my_last_term++ != to_do_term);
+
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "finished resolutor work \n");
+        return _gf_true;
+}
+
+gf_boolean_t
+nsr_recon_driver_leader (nsr_recon_private_t *priv)
+{
+	uint32_t replica_group_size = priv->replica_group_size;
+        uint32_t i;
+        nsr_recon_driver_ctx_t *ctx = priv->driver_thread_context;
+        int32_t bm;
+	int32_t status = 0;
+	int32_t op_errno = 0;
+        int32_t chosen = -1;
+        int32_t last_term = -1, last_ops = -1;
+
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "getting last term info from all members of this group\n");
+        // Get last term info from all members for this group
+        send_and_wait(&status, &op_errno, -1,
+                      replica_group_size,
+                      ctx,
+                      NSR_WORK_ID_GET_LAST_TERM_INFO,
+                      NSR_RECON_QUEUE_TO_CONTROL, ctx->current_term);
+        if (status == -1)
+                return _gf_false;
+
+
+        // compare all the info received and choose the reconciliator First
+        // choose all with latest term
+        for (i=0; i < replica_group_size; i++) {
+                if (ctx->workers[i].in_use) {
+                        if (ctx->workers[i].recon_info->last_term > last_term) {
+                                last_term = ctx->workers[i].recon_info->last_term;
+                        }
+                }
+        }
+        // First choose all with latest term and highest ops
+        for (i=0; i < replica_group_size; i++) {
+                if ((ctx->workers[i].in_use) && (last_term == ctx->workers[i].recon_info->last_term)) {
+                        if (ctx->workers[i].recon_info->commited_ops > last_ops) {
+                                last_ops = ctx->workers[i].recon_info->commited_ops;
+                        }
+                }
+        }
+        // choose the first among the lot
+        for (i=0; i < replica_group_size; i++) {
+                if ((ctx->workers[i].in_use) &&
+                    (last_term == ctx->workers[i].recon_info->last_term) &&
+                    (last_ops == ctx->workers[i].recon_info->commited_ops)) {
+                    chosen = i;
+                    break;
+                }
+        }
+
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "reconciliator chosen is %d\n", chosen);
+        ctx->reconciliator_index = chosen;
+        GF_ASSERT(chosen != -1);
+        if (chosen == -1) {
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "no reconciliatior chosen\n");
+                return _gf_false;
+        }
+
+        // send the message to reconciliator to do reconciliation with list of
+        // nodes that are part of this quorum
+        if (chosen != 0) {
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "sending reconciliation work to %d\n", chosen);
+                 bm = 1 << ctx->reconciliator_index;
+                 send_and_wait(&status, &op_errno, bm,
+                      replica_group_size,
+                      ctx,
+                      NSR_WORK_ID_RECONCILIATOR_DO_WORK,
+                      NSR_RECON_QUEUE_TO_CONTROL, -1);
+                if (status == -1)
+                      return _gf_false;
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "finished reconciliation work to %d\n", chosen);
+        } else {
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "local node is reconciliator. before set jmp\n");
+                nsr_recon_driver_reconciliator(priv);
+        }
+
+        // send message to all other nodes to sync up with the reconciliator
+        // including itself if required
+        // requires optimisation - TBD
+        if (chosen != 0) {
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                                "local node resolution needs to be done. before set jmp\n");
+                nsr_recon_driver_resolutor(priv);
+        }
+
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "sending resolution work to all nodes except this node and reconciliator\n");
+        bm = ~((1 << ctx->reconciliator_index) || 1);
+        send_and_wait(&status, &op_errno, bm,
+                      replica_group_size,
+                      ctx,
+                      NSR_WORK_ID_RESOLUTION_DO_WORK,
+                      NSR_RECON_QUEUE_TO_CONTROL, -1);
+        if (status == -1)
+                return _gf_false;
+
+        nsr_driver_log (this->name, GF_LOG_INFO,
+                        "finished reconciliation work as leader \n");
+        return _gf_true;
+}
+
+// main recon driver thread
+void *
+nsr_reconciliation_driver(void *arg)
+{
+        nsr_recon_private_t *priv = (nsr_recon_private_t *) arg;
+	uint32_t replica_group_size = priv->replica_group_size;
+        uint32_t i;
+        nsr_per_node_worker_t *control_s, *data_s;
+        nsr_recon_driver_ctx_t **driver_ctx, *ctx;
+        int32_t bm;
+        xlator_t *this = priv->this;
+        char *con_name, *data_name;
+	int32_t status = 0;
+	int32_t op_errno = 0;
+
+        driver_ctx = &priv->driver_thread_context;
+        (*driver_ctx) = GF_CALLOC (1,
+                                   sizeof (nsr_recon_driver_ctx_t),
+                                   gf_mt_recon_driver_ctx_t);
+        if (!driver_ctx) {
+                gf_log (this->name, GF_LOG_ERROR, "memory allocation error \n");
+                return NULL;
+        }
+        ctx = *driver_ctx;
+        ctx->this = priv->this;
+        ctx->replica_group_size = replica_group_size;
+
+        ctx->fp = recon_create_log (priv->replica_group_members[0], "nsr-driver-log");
+        if (!ctx->fp)
+                return NULL;
+
+        if ((pthread_mutex_init(&(ctx->mutex), NULL)) ||
+            (pthread_cond_init(&(ctx->cv), NULL))){
+                nsr_driver_log (this->name, GF_LOG_ERROR, "mutex init error \n");
+                return NULL;
+        }
+        INIT_LIST_HEAD(&(ctx->role_head.list));
+
+        ctx->workers = RD_CALLOC (replica_group_size,
+                                  sizeof(nsr_replica_worker_t),
+                                  gf_mt_recon_worker_t);
+        if (!ctx->workers) {
+                nsr_driver_log (this->name, GF_LOG_ERROR, "memory allocation error \n");
+                return NULL;
+        }
+        for (i=0; i < replica_group_size; i++) {
+		strcpy(ctx->workers[i].name, priv->replica_group_members[i]);
+        }
+
+        control_s = RD_CALLOC (replica_group_size,
+                               sizeof(nsr_per_node_worker_t),
+                               gf_mt_recon_per_node_worker_t);
+        if (!control_s) {
+                nsr_driver_log (this->name, GF_LOG_ERROR, "memory allocation error \n");
+                return NULL;
+        }
+
+        data_s = RD_CALLOC (replica_group_size,
+                            sizeof(nsr_per_node_worker_t),
+                            gf_mt_recon_per_node_worker_t);
+        if (!data_s) {
+                nsr_driver_log (this->name, GF_LOG_ERROR, "memory allocation error \n");
+                return NULL;
+        }
+        for (i=0; i < replica_group_size; i++) {
+                ctx->workers[i].control_worker = &control_s[i];
+                if (asprintf(&con_name,"recon-con-%u",i) < 1) {
+                        return NULL;
+                }
+                ctx->workers[i].control_worker->fp = recon_create_log
+                                (priv->replica_group_members[0], con_name);
+                if (!ctx->workers[i].control_worker->fp)
+                        return NULL;
+                ctx->workers[i].data_worker = &data_s[i];
+                if (asprintf (&data_name,"recon-data-%u",i) <1) {
+                        return NULL;
+                }
+                ctx->workers[i].data_worker->fp = recon_create_log
+                                (priv->replica_group_members[0], data_name);
+                if (!ctx->workers[i].data_worker->fp)
+                        return NULL;
+       }
+
+         nsr_driver_log (this->name, GF_LOG_INFO, "creating threads \n");
+        // Create the worker threads
+        // For every brick including itself there will be 2 worker threads:
+        // one for data and one for control
+        if (!create_worker_threads(priv, ctx, control_s, _gf_true,
+                                   (F_t) control_worker_main, replica_group_size) ||
+            !create_worker_threads(priv, ctx, data_s, _gf_false,
+                                   (F_t) data_worker_main, replica_group_size)) {
+                return NULL;
+        }
+
+        for (i=0; i < replica_group_size; i++) {
+                nsr_recon_get_file(priv->volname, &(ctx->workers[i]));
+        }
+
+        while (1) {
+
+            nsr_role_work_t *rr;
+
+            nsr_driver_log (this->name, GF_LOG_INFO, "waiting for role to be queued \n");
+            pthread_mutex_lock(&(ctx->mutex));
+            while (list_empty(&(ctx->role_head.list))) {
+                  pthread_cond_wait(&(ctx->cv), &(ctx->mutex));
+            }
+            pthread_mutex_unlock(&(ctx->mutex));
+
+            list_for_each_entry(rr, &(ctx->role_head.list), list) {
+                nsr_recon_driver_state_t state;
+                state = nsr_recon_driver_get_role(&status, ctx, rr);
+
+		if (status == -1) {
+			op_errno = EIO;
+			goto out;
+		}
+
+                switch (state) {
+
+                case leader:
+                        if (!nsr_recon_driver_leader(priv)) {
+                                goto out;
+                        }
+                        break;
+                case reconciliator:
+                        if (!nsr_recon_driver_reconciliator(priv)) {
+                                goto out;
+                        }
+                        break;
+                case resolutor:
+                        if (!nsr_recon_driver_resolutor(priv)) {
+                                goto out;
+                        }
+                        break;
+
+                case joiner:
+
+                        nsr_driver_log (this->name, GF_LOG_INFO, "getting last term info from all members of this group\n");
+                        // Get last term info from all members for this group
+                        // which will be the leader(this node) and the node that wants to join.
+                        send_and_wait(&status, &op_errno, -1,
+                                      replica_group_size,
+                                      ctx,
+                                      NSR_WORK_ID_GET_LAST_TERM_INFO,
+                                      NSR_RECON_QUEUE_TO_CONTROL, ctx->current_term);
+			if (status == -1)
+				goto out;
+
+
+                        // send message to other node that just joined to sync up with this node which is also the leader
+                        nsr_driver_log (this->name, GF_LOG_INFO, "sending resolution work to all nodes except this\n");
+                        bm = ~(1);
+                        send_and_wait(&status, &op_errno, bm,
+                                      replica_group_size,
+                                      ctx,
+                                      NSR_WORK_ID_RESOLUTION_DO_WORK,
+                                      NSR_RECON_QUEUE_TO_CONTROL, -1);
+			if (status == -1)
+				goto out;
+
+                        nsr_driver_log (this->name, GF_LOG_INFO,
+                                        "finished recon work as joiner \n");
+                        break;
+
+                default:
+                        nsr_driver_log (this->name, GF_LOG_ERROR,
+                                        "bad state %d", state);
+                }
+
+
+                // free the asasociated recon_info contexts created as part of this role
+
+out:
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                               "sending end of reconciliation message \n");
+                nsr_recon_return_back(priv, ctx->term, status, op_errno);
+                nsr_driver_log (this->name, GF_LOG_INFO,
+                               "finished sending end of reconciliation message \n");
+           }
+           list_del_init (&rr->list);
+        }
+
+        return NULL;
+}
diff --git a/xlators/cluster/nsr-recon/src/recon_driver.h b/xlators/cluster/nsr-recon/src/recon_driver.h
new file mode 100644
index 000000000..3efb26269
--- /dev/null
+++ b/xlators/cluster/nsr-recon/src/recon_driver.h
@@ -0,0 +1,325 @@
+/*
+  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 __RECON_DRIVER_H__
+#define __RECON_DRIVER_H__
+
+
+#include "api/src/glfs.h"
+
+#define MAX_HOSTNAME_LEN 32
+#define MAXIMUM_REPLICA_STRENGTH 8
+#define MAX_RECONCILIATION_WINDOW_SIZE 10000
+
+#define GLUSTERD_DEFAULT_WORKDIR "/var/lib/glusterd"
+#define GLUSTERD_VOLUME_DIR_PREFIX "vols"
+#define GLUSTERD_BRICK_INFO_DIR "bricks"
+
+/*
+ * Even with the names fixed, the non-NSR_DEBUG definitions of nsr_*_log don't
+ * work because many callers don't have "this" defined.
+ *
+ * TBD: use gf_log, fix "this" problem, eliminate extra fields and newlines.
+ */
+#define NSR_DEBUG
+
+typedef enum nsr_recon_work_req_id_t {
+       NSR_WORK_ID_GET_NONE = 0,
+       NSR_WORK_ID_GET_LAST_TERM_INFO = NSR_WORK_ID_GET_NONE + 1,
+       NSR_WORK_ID_GET_GIVEN_TERM_INFO = NSR_WORK_ID_GET_LAST_TERM_INFO + 1,
+       NSR_WORK_ID_RECONCILIATOR_DO_WORK = NSR_WORK_ID_GET_GIVEN_TERM_INFO + 1,
+       NSR_WORK_ID_RESOLUTION_DO_WORK = NSR_WORK_ID_RECONCILIATOR_DO_WORK + 1,
+       NSR_WORK_ID_GET_RECONCILATION_WINDOW = NSR_WORK_ID_RESOLUTION_DO_WORK + 1,
+       NSR_WORK_ID_SINGLE_RECONCILIATION_READ = NSR_WORK_ID_GET_RECONCILATION_WINDOW + 1,
+       NSR_WORK_ID_SINGLE_RECONCILIATION_COMMIT = NSR_WORK_ID_SINGLE_RECONCILIATION_READ + 1,
+       NSR_WORK_ID_SINGLE_RECONCILIATION_FLUSH = NSR_WORK_ID_SINGLE_RECONCILIATION_COMMIT + 1,
+       NSR_WORK_ID_GET_RESOLUTION_WINDOW = NSR_WORK_ID_SINGLE_RECONCILIATION_FLUSH + 1,
+       NSR_WORK_ID_END_RECONCILIATION = NSR_WORK_ID_GET_RESOLUTION_WINDOW + 1,
+       NSR_WORK_ID_INI = NSR_WORK_ID_END_RECONCILIATION + 1,
+       NSR_WORK_ID_FINI = NSR_WORK_ID_INI + 1
+} nsr_recon_work_req_id_t;
+
+typedef enum nsr_recon_queue_type_t {
+       NSR_RECON_QUEUE_TO_CONTROL = 0,
+       NSR_RECON_QUEUE_TO_DATA =NSR_RECON_QUEUE_TO_CONTROL + 1,
+} nsr_recon_queue_type_t;
+
+typedef enum nsr_log_type_t {
+       NSR_LOG_HOLE = 0b0,
+       NSR_LOG_PSEUDO_HOLE = 0b1,
+       NSR_LOG_FILL = 0b11
+} nsr_log_type_t;
+
+typedef enum nsr_mode_t {
+        NSR_SEQ = 0,
+        NSR_USE_THREADS = 1,
+        NSR_ASYNC = 2
+} nsr_mode_t;
+
+typedef enum nsr_recon_work_type_t {
+       NSR_RECON_WORK_NONE = 0,
+       NSR_RECON_WORK_HOLE_TO_NOOP = NSR_RECON_WORK_NONE + 1,
+       NSR_RECON_WORK_HOLE_TO_PSEUDO_HOLE = NSR_RECON_WORK_HOLE_TO_NOOP + 1,
+       NSR_RECON_WORK_COMPARE_PSEUDO_HOLE = NSR_RECON_WORK_HOLE_TO_PSEUDO_HOLE + 1,
+       NSR_RECON_WORK_HOLE_TO_FILL = NSR_RECON_WORK_COMPARE_PSEUDO_HOLE + 1,
+       NSR_RECON_WORK_UNDO_FILL = NSR_RECON_WORK_HOLE_TO_FILL + 1,
+} nsr_recon_work_type_t;
+
+typedef enum nsr_recon_driver_state_t {
+        none = 0,
+        leader = 1,
+        reconciliator = 2,
+        resolutor = 3,
+        joiner = 4,
+} nsr_recon_driver_state_t;
+
+// role structure
+#pragma pack(push, 1)
+typedef struct _nsr_recon_role_s {
+         uint32_t role; // leader, reconciliator, resolutor
+         uint32_t num; // required in case state is reconciliator
+         uint32_t current_term; // current term used in case of leader
+         // In case this is reconciliator, num is set to nodes that were part
+         // of previous term.
+         // In case this is resolutor, num is set to 2.
+         // info[0] - information for this node.
+         // info[1] - information of the reconciliator.
+         // In case this is leader, num is set to this term's membership list
+         // set info.name to all members including the leader
+         struct {
+                int32_t last_term;
+                int32_t commited_ops;
+                uint32_t last_index;
+                uint32_t first_index;
+                char name[MAX_HOSTNAME_LEN];
+        } info[MAXIMUM_REPLICA_STRENGTH];
+} nsr_recon_role_t;
+#pragma pack(pop)
+
+#define ENDIAN_CONVERSION_RR(rr, is_true) \
+{ \
+        uint32_t i=0; \
+        uint32_t (*f)(uint32_t) = ((is_true == _gf_true) ? ntohl : htonl); \
+        if (is_true == _gf_true) rr.num = f(rr.num); \
+        rr.current_term = f(rr.current_term); \
+        for (i=0; i < rr.num; i++) { \
+                rr.info[i].last_term = f(rr.info[i].last_term); \
+                rr.info[i].commited_ops = f(rr.info[i].commited_ops); \
+                rr.info[i].last_index = f(rr.info[i].last_index); \
+                rr.info[i].first_index = f(rr.info[i].first_index); \
+        } \
+        if (is_true == _gf_false) rr.num = f(rr.num); \
+}
+
+//  last term info structure
+#pragma pack(push, 1)
+typedef struct _nsr_recon_last_term_info_s {
+        int32_t last_term;
+        int32_t commited_ops;
+        uint32_t last_index;
+        uint32_t first_index;
+} nsr_recon_last_term_info_t;
+#pragma pack(pop)
+
+#define ENDIAN_CONVERSION_LT(lt, is_true) \
+{ \
+        uint32_t (*f)(uint32_t) = ((is_true == _gf_true) ? ntohl : htonl); \
+        lt.last_term = f(lt.last_term); \
+        lt.commited_ops = f(lt.commited_ops); \
+        lt.last_index = f(lt.last_index); \
+        lt.first_index = f(lt.first_index); \
+}
+
+// log information
+#pragma pack(push, 1)
+typedef struct _nsr_recon_log_info_s {
+        uint32_t term;
+        uint32_t first_index;
+        uint32_t last_index;
+} nsr_recon_log_info_t;
+#pragma pack(pop)
+
+#define ENDIAN_CONVERSION_LI(li, is_true) \
+{ \
+        uint32_t (*f)(uint32_t) = ((is_true == _gf_true) ? ntohl : htonl); \
+        li.term = f(li.term); \
+        li.first_index = f(li.first_index); \
+        li.last_index = f(li.last_index); \
+}
+
+#pragma pack(push, 1)
+typedef struct nsr_recon_record_details_s {
+       uint32_t type;
+       uint32_t op;
+       char gfid[36+1];
+       char pargfid[36+1];
+       char link_path[256]; // should it be PATH_MAX?
+       uint32_t offset;
+       uint32_t len;
+       char entry[128];
+       char newloc[128]; // for rename. can you overload link_path for this? TBD
+       mode_t mode;
+} nsr_recon_record_details_t;
+#pragma pack(pop)
+
+#define ENDIAN_CONVERSION_RD(rd, is_true) \
+{ \
+        uint32_t (*f)(uint32_t) = ((is_true == _gf_true) ? ntohl : htonl); \
+        rd.type = f(rd.type); \
+        rd.op = f(rd.op); \
+        rd.offset = f(rd.offset); \
+        rd.len = f(rd.len); \
+}
+
+typedef struct _nsr_role_work_s {
+        nsr_recon_role_t role;
+        uint32_t term;
+        struct list_head list;
+} nsr_role_work_t;
+        
+typedef struct _nsr_recon_work_s {
+	gf_boolean_t in_use;
+        uint32_t index;
+        uint32_t req_id;
+        struct list_head list;
+} nsr_recon_work_t;
+
+typedef struct _nsr_reconciliation_work_s {
+       uint32_t term;
+       uint32_t index;
+       uint32_t type;
+       uint32_t source;
+       void *data;
+
+       uint32_t num; // used for xattr
+
+} nsr_reconciliation_work_t;
+
+typedef struct _nsr_reconciliation_record_s {
+        nsr_reconciliation_work_t work; // will store the computed work
+        nsr_recon_record_details_t rec;
+} nsr_reconciliation_record_t;
+
+typedef struct _nsr_reconciliator_info {
+        uint32_t reconcilator_index;
+        int32_t last_term;
+        int32_t commited_ops;
+        uint32_t last_index;
+        uint32_t first_index;
+        //nsr_reconciliation_record_t records[MAX_RECONCILIATION_WINDOW_SIZE];
+        nsr_reconciliation_record_t *records;
+} nsr_reconciliator_info_t;
+
+typedef struct _nsr_per_node_worker_s {
+       char *id; // identifier
+       char vol_file[256]; //volfile that will be used by this thread
+       glfs_t *fs;
+       glfs_fd_t *aux_fd;
+       uint32_t index; // index into array of workers
+       pthread_t thread_id; // thread id
+       void * context; // thread context
+       struct _nsr_recon_driver_ctxt *driver_ctx;
+       char local; // local data worker
+       //struct list_head  list; //list of work items
+       nsr_recon_work_t head;
+       pthread_mutex_t mutex; //mutex to guard the state
+       pthread_cond_t cv; //condition variable for signaling the worker thread
+       gf_boolean_t is_control;
+#if defined(NSR_DEBUG)
+       FILE *fp;
+#endif
+       int32_t result; // result of latest work
+       int32_t op_errno; // errno
+} nsr_per_node_worker_t;
+
+typedef struct _nsr_replica_worker_s {
+       char name[256];
+       nsr_per_node_worker_t *control_worker;
+       nsr_per_node_worker_t *data_worker;
+       gf_boolean_t in_use;
+       nsr_reconciliator_info_t *recon_info; // Bunch of infos kept for this reconciliation
+} nsr_replica_worker_t;
+
+typedef struct _nsr_recon_driver_ctxt {
+       xlator_t *this;
+       uint32_t replica_group_size; // number of static members of replica group
+       nsr_replica_worker_t *workers; // worker info
+       int32_t reconciliator;
+       pthread_mutex_t mutex; 
+       pthread_cond_t cv;
+       nsr_role_work_t role_head;
+       volatile int32_t outstanding;
+       uint32_t reconciliator_index;
+       uint32_t term;
+       uint32_t current_term;
+       nsr_mode_t mode; // default set to seq
+#if defined(NSR_DEBUG)
+       FILE *fp;
+#endif
+} nsr_recon_driver_ctx_t;
+
+void *
+nsr_reconciliation_driver(void *);
+
+gf_boolean_t
+nsr_recon_driver_set_role(nsr_recon_driver_ctx_t *ctx, nsr_recon_role_t *rr, uint32_t term);
+
+#define atomic_inc(ptr)        ((void) __sync_fetch_and_add(ptr, 1))
+#define atomic_dec(ptr)        ((void) __sync_fetch_and_add(ptr, -1))
+#define atomic_fetch_and       __sync_fetch_and_and
+#define atomic_fetch_or        __sync_fetch_and_or
+
+#if defined(NSR_DEBUG)
+
+#define NSR_LOG_DIR "/var/log/nsr-logs"
+
+extern int      nsr_debug_level;
+extern FILE     *recon_create_log (char *member, char *module);
+
+extern void
+_nsr_driver_log (const char *func, int line, char *member, FILE *fp,
+                 char *fmt, ...);
+
+#define nsr_driver_log(dom, levl, fmt...) do {                          \
+        FMT_WARN (fmt);                                                 \
+        if (levl <= nsr_debug_level) {                                  \
+                nsr_recon_private_t     *priv   = ctx->this->private;   \
+                _nsr_driver_log (__FUNCTION__, __LINE__,                \
+                                   priv->replica_group_members[0],      \
+                                   ctx->fp,                             \
+                                   ##fmt);                              \
+        }                                                               \
+} while (0)
+
+extern void
+_nsr_worker_log (const char *func, int line, char *member,
+                 char *type, uint32_t index, FILE *fp,
+                 char *fmt, ...);
+
+#define nsr_worker_log(dom, levl, fmt...) do {                          \
+        FMT_WARN (fmt);                                                 \
+        if (levl <= nsr_debug_level) {                                  \
+                nsr_recon_private_t *priv;                              \
+                priv = ctx->driver_ctx->this->private;                  \
+                _nsr_worker_log (__FUNCTION__, __LINE__,                \
+                                 priv->replica_group_members[0],        \
+                                 ctx->is_control ? "recon-con" :        \
+                                 "recon-data",                          \
+                                 ctx->index, ctx->fp,                   \
+                                 ##fmt);                                \
+        }                                                               \
+} while (0)
+
+#else
+#define nsr_driver_log(dom, levl, fmt...) gf_log(dom, levl, fmt)
+#define nsr_worker_log(dom, levl, fmt...) gf_log(dom, levl, fmt)
+#endif
+
+#endif /* #ifndef __RECON_DRIVER_H__ */
diff --git a/xlators/cluster/nsr-recon/src/recon_xlator.c b/xlators/cluster/nsr-recon/src/recon_xlator.c
new file mode 100644
index 000000000..272c35dc2
--- /dev/null
+++ b/xlators/cluster/nsr-recon/src/recon_xlator.c
@@ -0,0 +1,1010 @@
+/*
+   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.
+*/
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include <sys/types.h>
+#include <fcntl.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "call-stub.h"
+#include "defaults.h"
+#include "xlator.h"
+
+#include "recon_driver.h"
+#include "recon_xlator.h"
+
+typedef struct _nsr_recon_fd_s {
+        int32_t term;
+        nsr_recon_driver_state_t state;
+        uint32_t first_index;
+        uint32_t last_index;
+        call_frame_t *frame;
+} nsr_recon_fd_t;
+
+#if defined(NSR_DEBUG)
+
+void
+_recon_main_log (const char *func, int line, char *member, FILE *fp,
+                 char *fmt, ...)
+{
+        va_list         ap;
+        char            *buf    = NULL;
+        int             retval;
+
+        if (!fp) {
+                fp = recon_create_log(member,"recon-main-log");
+                if (!fp) {
+                        return;
+                }
+        }
+
+        va_start(ap,fmt);
+        retval = vasprintf(&buf,fmt,ap);
+        if (buf) {
+                fprintf(fp,"[%s:%d] %.*s\n",func,line,retval,buf);
+                free(buf);
+        }
+        va_end(ap);
+}
+
+#endif
+
+// Given fd, get back the NSR based fd context.
+static int32_t this_fd_ctx_get(fd_t *fd, xlator_t *this, nsr_recon_fd_t **rfd)
+{
+        uint64_t tmp = 0;
+        int32_t ret = -1;
+
+        if ((ret = fd_ctx_get(fd, this, &tmp)) != 0) {
+                return ret;
+        } else {
+                *rfd = (nsr_recon_fd_t *)tmp;
+                return 0;
+        }
+}
+
+// Add the frame in q after associating with term
+// term usage tbd
+static void put_frame(nsr_recon_private_t *priv,
+                      call_frame_t *frame,
+                      uint32_t term)
+{
+	xlator_t *this = priv->this;
+        recon_main_log (this->name, GF_LOG_INFO, "adding frame for term %d \n", term);
+	priv->frame = frame;
+	return;
+}
+
+// get the frame from the queue given the term
+// term usage tbd
+static void get_frame(nsr_recon_private_t *priv,
+                      call_frame_t **frame,
+                      uint32_t term)
+{
+	if (frame != NULL)
+		*frame = priv->frame;
+	priv->frame = NULL;
+	return;
+}
+
+// check if there are outstanding frames
+static gf_boolean_t is_frame(nsr_recon_private_t *priv)
+{
+	return((priv->frame != NULL) ? _gf_true : _gf_false);
+}
+
+#define ENTRY_SIZE 128
+
+long
+get_entry_count (char *path)
+{
+        int             fd;
+        struct stat     buf;
+        unsigned long   entries         = -1;
+        long            min;            /* last entry not known to be empty */
+        long            max;            /* first entry known to be empty */
+        long            curr;
+        char            entry[ENTRY_SIZE];
+        void            *err_label      = &&done;
+
+        fd = open(path,O_RDONLY);
+        if (fd < 0) {
+                goto *err_label;
+        }
+        err_label = &&close_fd;
+
+        if (fstat(fd,&buf) < 0) {
+                goto *err_label;
+        }
+
+        min = 0;
+        max = buf.st_size / ENTRY_SIZE;
+        printf("max = %ld\n",max);
+
+        while ((min+1) < max) {
+                curr = (min + max) / 2;
+                printf("trying entry %ld\n",curr);
+                if (lseek(fd,curr*ENTRY_SIZE,SEEK_SET) < 0) {
+                        goto *err_label;
+                }
+                if (read(fd,entry,sizeof(entry)) != sizeof(entry)) {
+                        goto *err_label;
+                }
+                if ((entry[0] == '_') && (entry[1] == 'P')) {
+                        min = curr;
+                }
+                else {
+                        max = curr;
+                }
+        }
+
+        entries = max;
+
+close_fd:
+        close(fd);
+done:
+        return entries;
+}
+
+// Get the term info for the term number specified
+void nsr_recon_libchangelog_get_this_term_info(xlator_t *this, char *bp, int32_t term, nsr_recon_last_term_info_t *lt)
+{
+        char path[PATH_MAX];
+        long entries;
+
+        bzero(lt, sizeof(nsr_recon_last_term_info_t));
+        lt->last_term = term;
+        sprintf(path,"%s/%s%d",bp,"TERM.",term);
+        entries = get_entry_count(path);
+        if (entries > 1) {
+                /* The first entry is actually a header. */
+                lt->first_index = 1;
+                /*
+                 * This seems wrong, because it means that last_index*128 will
+                 * be exactly at EOF and commited_ops will be one greater than
+                 * it should be.  Maybe some other code makes the exact
+                 * opposite mistake to compensate.
+                 */
+                lt->last_index = lt->commited_ops = (int)entries;
+        }
+        recon_main_log (this->name, GF_LOG_INFO, "for term=%d got first_index=%d last_index=%d commited_ops=%d\n",
+                        term, lt->first_index, lt->last_index, lt->commited_ops);
+        return;
+}
+
+// Given the term number, find the last term in the changelogs
+void nsr_recon_libchangelog_get_last_term_info(xlator_t *this, char *bp, int32_t term, nsr_recon_last_term_info_t *lt)
+{
+        uint32_t t = term;
+        struct stat buf;
+        char path[PATH_MAX];
+        bzero(lt, sizeof(nsr_recon_last_term_info_t));
+        while(t) {
+                // journal file is of type TERM-1.jnl
+                sprintf(path,"%s/%s%d",bp,"TERM.",t);
+                if (!stat(path, &buf)) {
+                        nsr_recon_libchangelog_get_this_term_info(this, bp, t, lt);
+                        recon_main_log (this->name, GF_LOG_INFO, "got last term given current term %d as %d\n", term, t);
+                        return;
+                }
+                t--;
+        }
+        recon_main_log (this->name, GF_LOG_INFO, "got no last term given current term %d \n", term);
+
+        return;
+}
+
+// Return back the frame stored against the term
+void nsr_recon_return_back(nsr_recon_private_t *priv, uint32_t term, int32_t status, int32_t op_errno)
+{
+        call_frame_t *old_frame = NULL;
+        xlator_t *this = priv->this;
+
+        get_frame(priv, &old_frame, term);
+        if (old_frame) {
+                recon_main_log (this->name, GF_LOG_INFO, "nsr_recon_writev returns old frame \n");
+                                // first return the original write for which this ack was sent
+                STACK_UNWIND_STRICT (writev, old_frame, status, op_errno, NULL, NULL, NULL);
+        } else {
+                recon_main_log (this->name, GF_LOG_ERROR, "EIII---nsr_recon_writev cnnot return old frame \n");
+        }
+}
+
+typedef enum records_type_t {
+        fop_gfid_pgfid_oldloc_newloc = 1,
+        fop_gfid_pgfid_entry = fop_gfid_pgfid_oldloc_newloc + 1,
+        fop_gfid = fop_gfid_pgfid_entry + 1 ,
+        fop_gfid_offset = fop_gfid + 1,
+        fop_gfid_offset_len = fop_gfid_offset + 1,
+} records_type_t;
+
+// Get the backend ./glusterfs/xx/xx/<...> path
+static void
+get_gfid_path(nsr_recon_private_t *priv, char *gfid, char *path)
+{
+        strcpy(path, priv->base_dir);
+        strcat(path, "/.glusterfs/");
+        strncat(path,gfid,2);
+        strcat(path,"/");
+        strncat(path,gfid+2,2);
+        strcat(path,"/");
+        strcat(path,gfid);
+}
+
+
+// Get the link to which backend points to
+static gf_boolean_t
+get_link_using_gfid(nsr_recon_private_t *priv, char *gfid, char *path)
+{
+        char lp[PATH_MAX];
+        xlator_t *this = priv->this;
+        get_gfid_path(priv,gfid, lp);
+        if (readlink(lp, path, 255) == -1) {
+                GF_ASSERT(0);
+                recon_main_log(priv->this, GF_LOG_ERROR,
+                               "cannot get readlink  for %s\n",lp);
+                return _gf_false;
+        }
+        return _gf_true;
+}
+
+// Get the list of changelog records given a term , first and last index.
+//
+// TBD: rewrite this hideous ball of mud in at least the following ways:
+//
+//      (1) Break out the code for handling a single record into a separate
+//          function, to make error handling easier and reduce "indentation
+//          creep" so the code's readable.
+//
+//      (2) Change all of the fop_xxx_yyy nonsense to OR together values
+//          like FOP_HAS_FIELD_XXX and FOP_HAS_FIELD_YYY, to reduce code
+//          duplication and facilitate the addition of new fields.
+//
+//      (3) Stop making so many assumptions about the underlying formats.
+//          The code as it is won't even work for the existing binary format,
+//          let alone as changelog evolves over time.
+//
+// Really, 90% of this code should just GO AWAY in favor of using
+// libgfchangelog, enhanced as necessary to support our needs.
+
+/*
+ * Use this macro to skip over a field we're not using yet.
+ * NB: the body is a null statement on purpose
+ * TBD: all instances of this should be removed eventually!
+ */
+#define SKIP_FIELD do /* nothing */ ; while (*(start++) != '\0')
+
+#define SKIP_OVER
+gf_boolean_t nsr_recon_libchangelog_get_records(xlator_t *this, char *bp, int32_t term, uint32_t first, uint32_t last, void *buf)
+{
+        // do a mmap; seek into the first and read all records till last.
+        // TBD - right now all records are pseudo holes but mark them as fills.
+        // TBD - pseudo hole to be implemented when actual fsync gets done on data.
+        char *rb = NULL, *orig = NULL;
+        char path[PATH_MAX];
+        int fd;
+        uint32_t index = 0;
+
+        recon_main_log (this->name, GF_LOG_INFO,
+                        "libchangelog_get_records called for term %d  index from %d to %d \n",
+                        term, first, last );
+
+        orig = rb = GF_CALLOC(128, ((last - first) + 1),
+                              gf_mt_recon_changelog_buf_t);
+
+        sprintf(path,"%s/%s%d",bp,"TERM.",term);
+        fd = open(path, O_RDONLY);
+        if (fd == -1) {
+		return _gf_false;
+	} else {
+                char *start = NULL;
+                nsr_recon_record_details_t * rec = (nsr_recon_record_details_t *)buf;
+
+                if (first == 0)
+                        lseek(fd, 128, SEEK_SET);
+                else
+                        lseek(fd, first * 128, SEEK_SET);
+                if (read(fd, rb, (last - first  + 1) * 128) == -1) {
+			return _gf_false;
+		}
+                start = rb;
+                index = first;
+                do {
+                        recon_main_log (this->name, GF_LOG_INFO,
+                                        "libchangelog_get_records start inspecting records at index %d \n",
+                                        index );
+                        if (!strncmp(start, "_PRE_", 5)) {
+                                uint32_t i;
+                                uint32_t opcode = 0;
+                                records_type_t type;
+
+                                start += 5;
+                                // increment by the NULLs after the PRE
+                                start += 4;
+                                SKIP_FIELD;     // real index
+                                // now we have the opcode
+                                while (*start != '\0') {
+                                        opcode *= 10;
+                                        opcode += (*(start++) - '0');
+                                }
+                                ++start;
+                                recon_main_log (this->name, GF_LOG_ERROR,
+                                                "libchangelog_get_records: got opcode %d @index %d\n", opcode, index);
+                                if ((opcode == GF_FOP_RENAME)) {
+                                        type = fop_gfid_pgfid_oldloc_newloc;
+                                } else if ((opcode == GF_FOP_UNLINK) ||
+                                    (opcode == GF_FOP_RMDIR) ||
+                                    (opcode == GF_FOP_LINK) ||
+                                    (opcode == GF_FOP_MKDIR) ||
+                                    (opcode == GF_FOP_SYMLINK) ||
+                                    (opcode == GF_FOP_MKNOD) ||
+                                    (opcode == GF_FOP_CREATE)) {
+                                        type = fop_gfid_pgfid_entry;
+                                } else if ((opcode == GF_FOP_FSETATTR) ||
+                                    (opcode == GF_FOP_SETATTR) ||
+                                    (opcode == GF_FOP_FREMOVEXATTR) ||
+                                    (opcode == GF_FOP_REMOVEXATTR) ||
+                                    (opcode == GF_FOP_SETXATTR) ||
+                                    (opcode == GF_FOP_FSETXATTR)) {
+                                        type = fop_gfid;
+                                } else if ((opcode == GF_FOP_TRUNCATE) ||
+                                    (opcode == GF_FOP_FTRUNCATE)) {
+                                        type = fop_gfid_offset;
+                                } else if (opcode == GF_FOP_WRITE) {
+                                        type = fop_gfid_offset_len;
+                                } else {
+                                        recon_main_log (this->name,
+                                                        GF_LOG_ERROR,
+                                                        "libchangelog_get_records:got no proper opcode %d @index %d\n",
+                                                        opcode, index);
+                                        //GF_ASSERT(0);
+                                        // make this as a hole.
+                                        // TBD - check this logic later. maybe we should raise alarm here because
+                                        // this means that changelog is corrupted. We are not handling changelog
+                                        // corruptions as of now.
+                                        rec->type = NSR_LOG_HOLE;
+                                        goto finish;
+                                }
+                                // TBD - handle psuedo holes once that logic is in.
+                                rec->type = NSR_LOG_FILL;
+                                recon_main_log (this->name, GF_LOG_ERROR,
+                                               "libchangelog_get_records:got type %d at index %d \n",
+                                                rec->type, index);
+                                rec->op = opcode;
+
+                                // Now get the gfid and parse it
+                                // before that increment the pointer
+                                for (i=0; i < 36; i++) {
+                                        rec->gfid[i] = (*start);
+                                        start++;
+                                }
+                                rec->gfid[i] = '\0';
+
+                                GF_ASSERT(*start == 0);
+                                start ++;
+
+                                if (opcode == GF_FOP_SYMLINK) {
+                                        i = 0;
+                                        do {
+                                                if (i >= 256) {
+                                                        goto finish;
+                                                }
+                                                rec->link_path[i++] = *start;
+                                        } while (*(start++) != '\0');
+                                }
+
+                                i = 0;
+                                // If type is fop_gfid_offset+_len, get offset
+                                if ((type == fop_gfid_offset) || (type == fop_gfid_offset_len)) {
+                                        char offset_str[128];
+                                        while(*start != 0) {
+                                                offset_str[i++] = *start;
+                                                start ++;
+                                        }
+                                        offset_str[i] = '\0';
+                                        // get over the 0
+                                        start++;
+                                        rec->offset = strtoul(offset_str, NULL, 10);
+                                        recon_main_log (this->name,
+                                                        GF_LOG_ERROR,
+                                                        "libchangelog_get_records:got offset %d @index %d \n", rec->offset, index);
+
+                                }
+                                i = 0;
+                                if (type == fop_gfid_offset_len) {
+                                        char len_str[128];
+                                        while(*start != 0) {
+                                                len_str[i++] = *start;
+                                                start ++;
+                                        }
+                                        len_str[i] = '\0';
+                                        // get over the 0
+                                        start++;
+                                        rec->len = strtoul(len_str, NULL, 10);
+                                        recon_main_log (this->name,
+                                                        GF_LOG_ERROR,
+                                                        "libchangelog_get_records:got length %d @index %d \n", rec->len, index);
+                                }
+                                i = 0;
+                                if (type == fop_gfid_pgfid_entry) {
+                                        switch (opcode) {
+                                        case GF_FOP_CREATE:
+                                        case GF_FOP_MKDIR:
+                                        case GF_FOP_MKNOD:
+                                                SKIP_FIELD;     // mode
+                                                break;
+                                        /* TBD: handle GF_FOP_SYMLINK target */
+                                        default:
+                                                ;
+                                        }
+                                        SKIP_FIELD;             // uid
+                                        SKIP_FIELD;             // gid
+                                        if (opcode == GF_FOP_MKNOD) {
+                                                SKIP_FIELD;     // dev
+                                        }
+                                        // first get the gfid and then the path
+                                        for (i=0; i < 36; i++) {
+                                                rec->pargfid[i] = (*start);
+                                                start++;
+                                        }
+                                        rec->pargfid[i] = '\0';
+                                        GF_ASSERT(*start == '/');
+                                        start ++;
+
+                                        i = 0;
+                                        while(*start != 0) {
+                                                rec->entry[i++] = *start;
+                                                start ++;
+                                        }
+                                        rec->entry[i] = '\0';
+                                        // get over the 0
+                                        start++;
+                                        /*
+                                         * Having to add this as a special case
+                                         * is awful.  See the function header
+                                         * comment for the real solution.
+                                         */
+                                        if (opcode == GF_FOP_CREATE) {
+                                                rec->mode = 0;
+                                                while (*start != '\0') {
+                                                        rec->mode *= 10;
+                                                        rec->mode += *start
+                                                                   - '0';
+                                                        ++start;
+                                                }
+                                                ++start;
+                                        }
+                                        recon_main_log (this->name,
+                                                        GF_LOG_ERROR,
+                                                        "libchangelog_get_records:got entry %s @index %d \n", rec->entry, index);
+
+                                }
+                                i = 0;
+                                if (type == fop_gfid_pgfid_oldloc_newloc) {
+
+                                        // first get the source and then the destination
+                                        // source stuff gets stored in pargfid/entry
+                                        for (i=0; i < 36; i++) {
+                                                rec->pargfid[i] = (*start);
+                                                start++;
+                                        }
+                                        rec->pargfid[i] = '\0';
+                                        GF_ASSERT(*start == '/');
+                                        start ++;
+
+                                        i=0;
+                                        while(*start != 0) {
+                                                rec->entry[i++] = *start;
+                                                start ++;
+                                        }
+                                        rec->entry[i] = '\0';
+                                        // get over the 0
+                                        start++;
+
+                                        // dst stuff gets stored in gfid/newloc
+                                        for (i=0; i < 36; i++) {
+                                                rec->gfid[i] = (*start);
+                                                start++;
+                                        }
+                                        rec->gfid[i] = '\0';
+                                        GF_ASSERT(*start == '/');
+                                        start ++;
+                                        i = 0;
+                                        while(*start != 0) {
+                                                rec->newloc[i++] = *start;
+                                                start ++;
+                                        }
+                                        rec->newloc[i] = '\0';
+                                        // get over the 0
+                                        start++;
+
+                                }
+                                ENDIAN_CONVERSION_RD((*rec), _gf_false); //htonl
+                        }
+finish:
+                        if (index == last)
+                                break;
+                        index++;
+                        rb += 128;
+                        start = rb;
+                        rec++;
+                } while(1);
+        }
+        GF_FREE(orig);
+        close(fd);
+
+        recon_main_log (this->name, GF_LOG_INFO,
+                        "libchangelog_get_records finsihed inspecting records for  term %d \n",
+                        term);
+        return _gf_true;
+}
+
+int32_t
+nsr_recon_open (call_frame_t *frame, xlator_t *this,
+                loc_t *loc, int32_t flags, fd_t *fd, dict_t *xdata)
+{
+        int32_t op_ret       = 0;
+        int32_t op_errno     = 0;
+        nsr_recon_fd_t *rfd = NULL;
+
+        recon_main_log (this->name, GF_LOG_INFO, "nsr_recon_open called for path %s \n",loc->path );
+        rfd = GF_CALLOC (1, sizeof (*rfd), gf_mt_recon_fd_t);
+        if (!rfd) {
+                op_ret = -1;
+                op_errno = ENOMEM;
+        }
+
+        op_ret = fd_ctx_set (fd, this, (uint64_t)(long)rfd);
+        if (op_ret) {
+                op_ret = -1;
+                op_errno = EINVAL;
+        }
+        recon_main_log (this->name, GF_LOG_INFO, "nsr_recon_open returns with %d for path %s \n",op_ret,loc->path );
+        STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd, NULL);
+        return 0;
+}
+
+int32_t
+nsr_recon_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)
+{
+        nsr_recon_fd_t *rfd = NULL;
+        nsr_recon_private_t *priv = NULL;
+        int32_t op_ret = 0;
+        int32_t op_errno = 0;
+        int32_t ret = 0;
+
+        ret = this_fd_ctx_get (fd, this, &rfd);
+        if (ret < 0) {
+                return -1;
+        }
+        priv = (nsr_recon_private_t *)this->private;
+
+        recon_main_log (this->name, GF_LOG_INFO, "nsr_recon_writev called for offset %d \n",(unsigned int)offset );
+        GF_ASSERT(count == 1);
+        switch (offset) {
+                // client(brick, leader) writes the role of the node
+                case nsr_recon_xlator_sector_1 :
+                {
+                        nsr_recon_role_t rr;
+                        memcpy((void *)&rr, (void *)vector[0].iov_base, sizeof(rr));
+                        ENDIAN_CONVERSION_RR(rr, _gf_true); //ntohl
+
+                        recon_main_log (this->name, GF_LOG_INFO, "nsr_recon_writev called to set role %d\n", rr.role);
+                        if ((rr.role != leader) &&
+                            (rr.role != reconciliator) &&
+                            (rr.role != resolutor) &&
+                            (rr.role != joiner)) {
+                                recon_main_log (this->name, GF_LOG_ERROR,
+                                                "EIII---nsr_recon_writev cannot set state \n");
+                                STACK_UNWIND_STRICT (writev, frame, -1, op_errno,
+                                                     NULL, NULL, NULL);
+                        }
+
+                        GF_ASSERT(rr.num <= MAXIMUM_REPLICA_STRENGTH);
+
+			// Check if already a role play is going on. If yes return with EAGAIN.
+			// Ideally we should check if we have got a higher term number while
+			// servicing a lower term number; if so abort the older one.
+			// However the abort infrastructure needs to be sketched properly; TBD.
+			if (is_frame(priv) == _gf_true) {
+                                recon_main_log (this->name, GF_LOG_ERROR,
+                                                "nsr_recon_writev set_role - already role play \n");
+                                STACK_UNWIND_STRICT (writev, frame, -1, EAGAIN,
+                                                     NULL, NULL, NULL);
+			} else {
+
+				// Store the stack frame so that when the actual job gets finished
+				// we send the response back to the brick.
+				put_frame(priv, frame, rr.current_term);
+				if (nsr_recon_driver_set_role(priv->driver_thread_context,
+					                      &rr,
+							       rr.current_term) == _gf_false) {
+					 get_frame(priv, NULL, rr.current_term);
+					 recon_main_log (this->name, GF_LOG_ERROR,
+                                                         "nsr_recon_writev set_role - cannot seem to set role \n");
+					 STACK_UNWIND_STRICT (writev, frame, -1, op_errno,
+							       NULL, NULL, NULL);
+				} else {
+					 recon_main_log (this->name, GF_LOG_INFO,
+                                                         "nsr_recon_writev set_role - set role succesfully \n");
+				}
+			}
+                        break;
+                }
+                // client(reconciliator) writes how much it needs for the read
+                case nsr_recon_xlator_sector_2 :
+                {
+                        nsr_recon_log_info_t li;
+                        memcpy((void *)&li, (void *)vector[0].iov_base, sizeof(li));
+                        ENDIAN_CONVERSION_LI(li, _gf_true); //ntohl
+
+                        recon_main_log (this->name, GF_LOG_INFO,
+                                        "nsr_recon_writev - setting term info for reconcilation info. term=%d, first_index=%d,start_index=%d \n",
+                                        li.term, li.first_index, li.last_index);
+                        rfd->term = li.term;
+                        rfd->last_index = li.last_index;
+                        rfd->first_index = li.first_index;
+                        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno,
+                                             NULL, NULL, NULL);
+                        break;
+                }
+                // client(reconciliator) writes term for which it needs info
+                case nsr_recon_xlator_sector_3 :
+                {
+                        int32_t term;
+
+                        memcpy((void *)&term, (void *)vector[0].iov_base, sizeof(term));
+                        term = ntohl(term); //ntohl
+                        recon_main_log (this->name, GF_LOG_INFO,
+                                        "nsr_recon_writev - setting term info for term info. term=%d\n",
+                                        term);
+                        rfd->term = term;
+                        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno,
+                                             NULL, NULL, NULL);
+                        break;
+                }
+                // client(reconciliator) writes current term so that it gets last term info later
+                case nsr_recon_xlator_sector_4 :
+                {
+                        int32_t term;
+
+                        memcpy((void *)&term, (void *)vector[0].iov_base, sizeof(term));
+                        term = ntohl(term); //ntohl
+                        recon_main_log (this->name, GF_LOG_INFO,
+                                        "nsr_recon_writev - setting term info for last term info  given current term=%d\n",
+                                        term);
+                        rfd->term = term;
+                        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno,
+                                             NULL, NULL, NULL);
+                        break;
+                }
+		default:
+		{
+                        recon_main_log (this->name, GF_LOG_ERROR,
+                                       "nsr_recon_writev called with wrong offset\n");
+                        STACK_UNWIND_STRICT (writev, frame, -1, op_errno,
+                                             NULL, NULL, NULL);
+			break;
+		}
+        }
+
+        return 0;
+}
+
+int
+nsr_recon_readv (call_frame_t *frame, xlator_t *this,
+                 fd_t *fd, size_t size, off_t offset, uint32_t flags, dict_t *xdata)
+{
+        nsr_recon_fd_t *rfd = NULL;
+        int32_t op_errno = 0;
+        // copied stuff from quick-read.c and posix.c
+        struct iobuf *iobuf = NULL;
+        struct iobref    *iobref = NULL;
+        struct iovec      iov = {0, };
+        int32_t ret = -1;
+        nsr_recon_private_t *priv = NULL;
+
+        iobuf = iobuf_get2 (this->ctx->iobuf_pool, size);
+        if (!iobuf) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        iobref = iobref_new ();
+        if (!iobref) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        iobref_add (iobref, iobuf);
+
+        ret = this_fd_ctx_get (fd, this, &rfd);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
+        priv = (nsr_recon_private_t *)this->private;
+
+        recon_main_log (this->name, GF_LOG_INFO, "nsr_recon_readv called for offset %d \n",(unsigned int)offset );
+        switch (offset) {
+                // client(leader) reads from here to get info for this term on this node
+                // invole libchagelog to get the information
+                case nsr_recon_xlator_sector_3 :
+                {
+                        nsr_recon_last_term_info_t lt;
+                        GF_ASSERT(size == sizeof(lt));
+                        nsr_recon_libchangelog_get_this_term_info(this,priv->changelog_base_path, rfd->term, &lt);
+                        recon_main_log (this->name, GF_LOG_INFO,
+                                        "nsr_recon_readv - getting term info for term=%d, ops=%d, first=%d, last=%d\n",
+                                        rfd->term, lt.commited_ops, lt.first_index, lt.last_index);
+                        ENDIAN_CONVERSION_LT(lt, _gf_false); //htonl
+                        memcpy(iobuf->ptr, &lt, size);
+                        goto out;
+                }
+                // client(reconciliator) reads individual record information
+                case nsr_recon_xlator_sector_2 :
+                {
+                        uint32_t num = (rfd->last_index - rfd->first_index + 1);
+                        recon_main_log (this->name, GF_LOG_INFO,
+                                        "nsr_recon_readv - expected size %lu got size %lu\n",
+                                        (num * sizeof(nsr_recon_record_details_t)), size);
+
+                        GF_ASSERT(size == (num * sizeof(nsr_recon_record_details_t)));
+                        bzero(iobuf->ptr, size);
+                        recon_main_log (this->name, GF_LOG_INFO,
+                                        "nsr_recon_readv - getting records for term=%d from %d to %d\n",
+                                        rfd->term, rfd->first_index, rfd->last_index);
+                        nsr_recon_libchangelog_get_records(this, priv->changelog_base_path,
+                                                           rfd->term, rfd->first_index, rfd->last_index, iobuf->ptr);
+                        goto out;
+                }
+                // read last term info
+                case nsr_recon_xlator_sector_4 :
+                {
+                        nsr_recon_last_term_info_t lt;
+                        GF_ASSERT(size == sizeof(lt));
+                        nsr_recon_libchangelog_get_last_term_info(this, priv->changelog_base_path, rfd->term, &lt);
+                        recon_main_log (this->name, GF_LOG_INFO,
+                                        "nsr_recon_readv - getting last term info given current term=%d. last term = %d ops=%d, first=%d, last=%d\n",
+                                        rfd->term, lt.last_term,  lt.commited_ops, lt.first_index, lt.last_index);
+                        ENDIAN_CONVERSION_LT(lt, _gf_false); //htonl
+                        memcpy(iobuf->ptr, &lt, size);
+                        goto out;
+                }
+		default:
+		{
+                        recon_main_log (this->name, GF_LOG_ERROR,
+                                       "nsr_recon_readv called with wrong offset\n");
+			op_errno = -1;
+			break;
+		}
+        }
+
+out:
+        if (op_errno == 0) {
+                iov.iov_base = iobuf->ptr;
+                ret = iov.iov_len = size;
+        }
+
+        STACK_UNWIND_STRICT (readv, frame, ret, op_errno, &iov, 1, NULL, iobref , NULL);
+
+        if (iobref)
+                iobref_unref (iobref);
+        if (iobuf)
+                iobuf_unref (iobuf);
+        return 0;
+}
+
+int
+nsr_recon_lookup (call_frame_t *frame, xlator_t *this,
+                  loc_t *loc, dict_t *xdata)
+{
+        struct iatt buf = {0, };
+        // dirty hack to set root as regular but seems to work.
+        buf.ia_type = IA_IFREG;
+        recon_main_log (this->name, GF_LOG_INFO, "nsr_recon_lookup called \n");
+
+        STACK_UNWIND_STRICT (lookup, frame, 0, 0, this->itable->root, &buf, NULL, NULL);
+        return 0;
+}
+
+
+int32_t
+nsr_recon_flush (call_frame_t *frame, xlator_t *this,
+             fd_t *fd, dict_t *xdata)
+{
+        STACK_UNWIND_STRICT (flush, frame, 0, 0, NULL);
+        return 0;
+}
+
+
+int32_t
+mem_acct_init (xlator_t *this)
+{
+        int     ret = -1;
+
+        GF_VALIDATE_OR_GOTO ("recon", this, out);
+
+        ret = xlator_mem_acct_init (this, gf_mt_recon_end + 1);
+
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Memory accounting init" "failed");
+                return ret;
+        }
+out:
+        return ret;
+}
+
+
+int32_t
+init (xlator_t *this)
+{
+        nsr_recon_private_t *priv        = NULL;
+        char  *local, *members;
+        unsigned int i=0;
+
+        priv = GF_CALLOC (1, sizeof (*priv), gf_mt_recon_private_t);
+        if (!priv) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "priv allocation error\n");
+                return -1;
+        }
+        GF_OPTION_INIT ("replica-group-size", priv->replica_group_size, uint32, err);
+        GF_OPTION_INIT ("vol-name", priv->volname, str, err);
+        if (!priv->volname) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "missing volname option (required)");
+                return -1;
+        }
+        GF_OPTION_INIT ("changelog-dir", priv->changelog_base_path, str, err);
+        if (!priv->changelog_base_path) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "missing changelog directory option (required)");
+                return -1;
+        }
+        GF_OPTION_INIT ("base-dir", priv->base_dir, str, err);
+        if (!priv->base_dir) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "missing brick base directory option (required)");
+                return -1;
+        }
+        GF_OPTION_INIT ("replica-group-members", members, str, err);
+        if (!members) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "missing membership option (required)");
+                return -1;
+        }
+        GF_OPTION_INIT ("local-member", local, str, err);
+        if (!local) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "missing local member option (required)");
+                return -1;
+        }
+
+        priv->replica_group_members = GF_CALLOC (priv->replica_group_size,
+                                                 sizeof(char *),
+                                                 gf_mt_recon_members_list_t);
+        priv->replica_group_members[0] = GF_CALLOC (1,
+                                                    strlen(local),
+                                                    gf_mt_recon_member_name_t);
+        if (!priv->replica_group_members || !(priv->replica_group_members[0])) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "str allocation error\n");
+                return -1;
+        }
+        strcpy(priv->replica_group_members[0], local);
+        for (i=1; i < priv->replica_group_size; i++) {
+                char *member;
+                if (i == 1)
+                        member = strtok(members, ",");
+                else
+                        member = strtok(NULL, ",");
+                priv->replica_group_members[i] = GF_CALLOC (1,
+                        strlen(member) + 1, gf_mt_recon_member_name_t);
+                if (!priv->replica_group_members[i]) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "str allocation error\n");
+                        return -1;
+                }
+                strcpy(priv->replica_group_members[i], member);
+        }
+
+
+        priv->this = this;
+        this->private = (void *)priv;
+
+        priv->fp = recon_create_log (priv->replica_group_members[0], "recon-main-log");
+        if (!priv->fp)
+                return -1;
+
+        recon_main_log (this->name, GF_LOG_INFO, "creating reconciliation driver \n");
+
+        if (pthread_create(&priv->thread_id, NULL, nsr_reconciliation_driver, priv)) {
+                recon_main_log (this->name, GF_LOG_ERROR,
+                        "pthread creation error \n");
+                return -1;
+        }
+
+        INIT_LIST_HEAD(&(priv->list));
+
+
+	return 0;
+
+err:
+        return -1;
+}
+
+
+void
+fini (xlator_t *this)
+{
+        nsr_recon_private_t *priv        = NULL;
+        void *ret = NULL;
+
+        priv = (nsr_recon_private_t *)this->private;
+
+        pthread_cancel(priv->thread_id);
+        pthread_join(priv->thread_id, &ret);
+}
+
+
+struct xlator_fops fops = {
+        .open = nsr_recon_open,
+        .readv = nsr_recon_readv,
+        .writev = nsr_recon_writev,
+        .lookup = nsr_recon_lookup,
+        .flush = nsr_recon_flush
+};
+
+struct xlator_cbks cbks = {
+};
+
+struct volume_options options[] = {
+        { .key = {"replica-group-size"},
+          .type = GF_OPTION_TYPE_INT,
+          .min = 2,
+          .max = INT_MAX,
+          .default_value = "2",
+          .description = "Number of bricks in replica group. can be derived but putting it here for testing."
+        },
+        {
+          .key = {"vol-name"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "volume name"
+        },
+        {
+          .key = {"local-member"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "member(brick) for which this translator is responsible."
+        },
+        {
+          .key = {"replica-group-members"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "Comma seperated member names other than local."
+        },
+        {
+          .key = {"changelog-dir"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "Base directory where per term changelogs are maintained."
+        },
+        {
+          .key = {"base-dir"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "Base directory for this brick. This should go away once we fix gfid based lookups"
+        },
+	{ .key = {NULL} },
+};
diff --git a/xlators/cluster/nsr-recon/src/recon_xlator.h b/xlators/cluster/nsr-recon/src/recon_xlator.h
new file mode 100644
index 000000000..d9692a632
--- /dev/null
+++ b/xlators/cluster/nsr-recon/src/recon_xlator.h
@@ -0,0 +1,92 @@
+/*
+  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 __RECON_XLATOR_H__
+#define __RECON_XLATOR_H__
+
+#include <semaphore.h>
+#include <pthread.h>
+
+enum gf_dht_mem_types_ {
+        gf_mt_recon_changelog_buf_t = gf_common_mt_end + 1,
+        gf_mt_recon_driver_ctx_t,
+        gf_mt_recon_fd_t,
+        gf_mt_recon_id_t,
+        gf_mt_recon_member_name_t,
+        gf_mt_recon_members_list_t,
+        gf_mt_recon_per_node_worker_t,
+        gf_mt_recon_private_t,
+        gf_mt_recon_reconciliator_info_t,
+        gf_mt_recon_record_t,
+        gf_mt_recon_record_details_t,
+        gf_mt_recon_role_work_t,
+        gf_mt_recon_work_t,
+        gf_mt_recon_work_data_t,
+        gf_mt_recon_worker_t,
+        gf_mt_recon_end,
+};
+
+enum nsr_recon_xlator_sector_t {
+        nsr_recon_xlator_sector_0 = 0, // to report back the status of given transaction ids
+        nsr_recon_xlator_sector_1 = 512, //  to write here information about leadership changes from the brick
+        nsr_recon_xlator_sector_2 = (512 * 2), // to write here individual roles and wait for that role to be done
+        nsr_recon_xlator_sector_3 = (512 *3), // read from here to get term info for given term
+        nsr_recon_xlator_sector_4 = (512 * 4), // read from here to get last term info
+};
+
+
+typedef struct _nsr_recon_private_s {
+        xlator_t *this; //back pointer
+        unsigned int replica_group_size; // number of static members of replica group
+        char **replica_group_members; // replica group members (including itself in first slot)
+        pthread_t thread_id; // driver thread id
+        nsr_recon_driver_ctx_t *driver_thread_context; //driver thread context
+        unsigned int outstanding; // for communicating with driver thread
+        call_frame_t *frame; // old frame that is pending (just one as of now)
+        struct list_head list;
+        char *volname;
+        uint32_t txn_id;
+        char *changelog_base_path;
+        char *base_dir;
+#if defined(NSR_DEBUG)
+        FILE *fp;
+#endif
+} nsr_recon_private_t;
+
+#define atomic_cmpxchg         __sync_val_compare_and_swap
+
+#if defined(NSR_DEBUG)
+
+extern void
+_recon_main_log (const char *func, int line, char *member, FILE *fp,
+                 char *fmt, ...);
+
+#define recon_main_log(dom, levl, fmt...) do {                          \
+        FMT_WARN (fmt);                                                 \
+        if (levl <= nsr_debug_level) {                                  \
+                nsr_recon_private_t     *priv   = this->private;        \
+                _recon_main_log (__FUNCTION__, __LINE__,                \
+                                 priv->replica_group_members[0],        \
+                                 priv->fp,                              \
+                                 ##fmt);                                \
+        }                                                               \
+} while (0)
+
+#else
+#define recon_main_log(dom, levl, fmt...) gf_log(dom, levl, fmt)
+#endif
+
+void nsr_recon_libchangelog_get_this_term_info(xlator_t *this, char *bp, int32_t term, nsr_recon_last_term_info_t *lt);
+void nsr_recon_libchangelog_get_last_term_info(xlator_t *this, char *bp, int32_t term, nsr_recon_last_term_info_t *lt);
+void nsr_recon_return_back(nsr_recon_private_t *priv, uint32_t term, int32_t status, int32_t op_errno);
+gf_boolean_t nsr_recon_libchangelog_get_records(xlator_t *this, char *bp, int32_t term, uint32_t first, uint32_t last, void *buf);
+
+
+#endif /* #ifndef __RECON_XLATOR_H__ */
diff --git a/xlators/protocol/legacy/lib/Makefile.am b/xlators/cluster/nsr-server/Makefile.am
index d471a3f92..d471a3f92 100644
--- a/xlators/protocol/legacy/lib/Makefile.am
+++ b/xlators/cluster/nsr-server/Makefile.am
diff --git a/xlators/cluster/nsr-server/src/Makefile.am b/xlators/cluster/nsr-server/src/Makefile.am
new file mode 100644
index 000000000..0092aad4f
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/Makefile.am
@@ -0,0 +1,43 @@
+noinst_PYTHON = codegen.py gen-fops.py
+
+xlator_LTLIBRARIES = nsr.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
+
+nsr_la_LDFLAGS = -module -avoid-version -lcurl
+
+if ENABLE_ETCD_SIM
+nsr_la_SOURCES = nsr.c leader.c recon_notify.c etcd-sim.c
+else
+nsr_la_SOURCES = nsr.c leader.c recon_notify.c etcd-api.c \
+		 yajl.c yajl_alloc.c yajl_buf.c yajl_encode.c yajl_gen.c \
+		 yajl_lex.c yajl_parser.c yajl_tree.c yajl_version.c
+endif
+
+
+nsr_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+				$(top_builddir)/api/src/libgfapi.la
+
+noinst_HEADERS = nsr-internal.h etcd-api.h all-templates.c \
+	         yajl_alloc.h yajl_buf.h yajl_bytestack.h yajl_encode.h \
+		 yajl_lex.h yajl_parser.h yajl/yajl_common.h yajl/yajl_gen.h \
+		 yajl/yajl_parse.h yajl/yajl_tree.h yajl/yajl_version.h \
+		 $(top_srcdir)/xlators/lib/src/libxlator.h \
+		 $(top_srcdir)/glusterfsd/src/glusterfsd.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) \
+	-I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/xlators/lib/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src -DSBIN_DIR=\"$(sbindir)\"
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+XLATOR_HEADER	= $(top_srcdir)/libglusterfs/src/xlator.h
+
+CLEANFILES = nsr-cg.c
+
+nsr-cg.c: gen-fops.py codegen.py $(XLATOR_HEADER) all-templates.c
+	$(PYTHON) ./gen-fops.py $(XLATOR_HEADER) all-templates.c > $@
+
+nsr.lo: nsr-cg.c
+
+uninstall-local:
+	rm -f $(DESTDIR)$(xlatordir)/nsr.so
diff --git a/xlators/cluster/nsr-server/src/all-templates.c b/xlators/cluster/nsr-server/src/all-templates.c
new file mode 100644
index 000000000..fa29de7b2
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/all-templates.c
@@ -0,0 +1,345 @@
+/*
+ * You can put anything here - it doesn't even have to be a comment - and it
+ * will be ignored until we reach the first template-name comment.
+ */
+
+
+// template-name read-fop
+$TYPE$
+nsr_$NAME$ (call_frame_t *frame, xlator_t *this,
+            $ARGS_LONG$)
+{
+        nsr_private_t   *priv   = this->private;
+        gf_boolean_t in_recon = _gf_false;
+        int32_t recon_term, recon_index;
+
+        // allow reads during reconciliation
+        // TBD: allow "dirty" reads on non-leaders
+        if (xdata &&
+            (dict_get_int32(xdata, RECON_TERM_XATTR, &recon_term) == 0) &&
+            (dict_get_int32(xdata, RECON_INDEX_XATTR, &recon_index) == 0)) {
+            in_recon = _gf_true;
+        }
+
+        if ((!priv->leader) && (in_recon == _gf_false)) {
+                goto err;
+        }
+
+        STACK_WIND (frame, default_$NAME$_cbk,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->$NAME$,
+                    $ARGS_SHORT$);
+        return 0;
+
+err:
+        STACK_UNWIND_STRICT ($NAME$, frame, -1, EREMOTE,
+                             $DEFAULTS$);
+        return 0;
+}
+
+// template-name read-dispatch
+/* No "dispatch" function needed for $NAME$ */
+
+// template-name read-fan-in
+/* No "fan-in" function needed for $NAME$ */
+
+// template-name read-continue
+/* No "continue" function needed for $NAME$ */
+
+// template-name read-complete
+/* No "complete" function needed for $NAME$ */
+
+// template-name write-fop
+$TYPE$
+nsr_$NAME$ (call_frame_t *frame, xlator_t *this,
+            $ARGS_LONG$)
+{
+        nsr_local_t     *local          = NULL;
+        nsr_private_t   *priv           = this->private;
+        int             op_errno        = ENOMEM;
+        int             from_leader;
+        int             from_recon;
+	uint32_t 	ti = 0;
+        double          must_be_up;
+        double          are_up;
+
+        /*
+         * Our first goal here is to avoid "split brain surprise" for users who
+         * specify exactly 50% with two- or three-way replication.  That means
+         * either a more-than check against half the total replicas or an
+         * at-least check against half of our peers (one less).  Of the two,
+         * only an at-least check supports the intuitive use of 100% to mean
+         * all replicas must be present, because "more than 100%" will never
+         * succeed regardless of which count we use.  This leaves us with a
+         * slightly non-traditional definition of quorum ("at least X% of peers
+         * not including ourselves") but one that's useful enough to be worth
+         * it.
+         *
+         * Note that n_children and up_children *do* include the local
+         * subvolume, so we need to subtract one in each case.
+         */
+        must_be_up = ((double)(priv->n_children - 1)) * priv->quorum_pct;
+        are_up = ((double)(priv->up_children - 1)) * 100.0;
+        if (are_up < must_be_up) {
+                /* Emulate the AFR client-side-quorum behavior. */
+                op_errno = EROFS;
+                goto err;
+        }
+
+        local = mem_get0(this->local_pool);
+        if (!local) {
+                goto err;
+        }
+#if defined(NSR_CG_NEED_FD)
+        local->fd = fd_ref(fd);
+#else
+        local->fd = NULL;
+#endif
+        INIT_LIST_HEAD(&local->qlinks);
+        frame->local = local;
+
+        if (xdata) {
+                from_leader = !!dict_get(xdata,NSR_TERM_XATTR);
+                from_recon = !!dict_get(xdata,RECON_TERM_XATTR)
+                          && !!dict_get(xdata,RECON_INDEX_XATTR);
+        }
+        else {
+                from_leader = from_recon = _gf_false;
+        }
+
+        // follower/recon path
+        // just send it to local node
+        if (from_leader || from_recon) {
+                atomic_inc(&priv->ops_in_flight);
+                STACK_WIND (frame, nsr_$NAME$_complete,
+                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->$NAME$,
+                            $ARGS_SHORT$);
+                return 0;
+        }
+
+
+        if (!priv->leader/* || priv->fence_io*/) {
+                op_errno = EREMOTE;
+                goto err;
+        }
+
+
+        if (!xdata) {
+                xdata = dict_new();
+                if (!xdata) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "failed to allocate xdata");
+                        goto err;
+                }
+        }
+
+        if (dict_set_int32(xdata,NSR_TERM_XATTR,priv->current_term) != 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to set nsr-term");
+                goto err;
+        }
+
+        LOCK(&priv->index_lock);
+        ti = ++(priv->index);
+        UNLOCK(&priv->index_lock);
+        if (dict_set_int32(xdata,NSR_INDEX_XATTR,ti) != 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to set index");
+                goto err;
+        }
+
+        local->stub = fop_$NAME$_stub (frame,nsr_$NAME$_continue,
+                                       $ARGS_SHORT$);
+        if (!local->stub) {
+                goto err;
+        }
+
+
+#if defined(NSR_CG_QUEUE)
+        nsr_inode_ctx_t         *ictx   = nsr_get_inode_ctx(this,fd->inode);
+        if (!ictx) {
+                op_errno = EIO;
+                goto err;
+        }
+        LOCK(&ictx->lock);
+                if (ictx->active) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "queuing request due to conflict");
+                        /*
+                         * TBD: enqueue only for real conflict
+                         *
+                         * Currently we just act like all writes are in
+                         * conflict with one another.  What we should really do
+                         * is check the active/pending queues and defer only if
+                         * there's a conflict there.
+                         *
+                         * It's important to check the pending queue because we
+                         * might have an active request X which conflicts with
+                         * a pending request Y, and this request Z might
+                         * conflict with Y but not X.  If we checked only the
+                         * active queue then Z could jump ahead of Y, which
+                         * would be incorrect.
+                         */
+                        local->qstub = fop_$NAME$_stub (frame,
+                                                        nsr_$NAME$_dispatch,
+                                                        $ARGS_SHORT$);
+                        if (!local->qstub) {
+                                UNLOCK(&ictx->lock);
+                                goto err;
+                        }
+                        list_add_tail(&local->qlinks,&ictx->pqueue);
+                        ++(ictx->pending);
+                        UNLOCK(&ictx->lock);
+                        return 0;
+                }
+                else {
+                        list_add_tail(&local->qlinks,&ictx->aqueue);
+                        ++(ictx->active);
+                }
+        UNLOCK(&ictx->lock);
+#endif
+
+        return nsr_$NAME$_dispatch (frame, this, $ARGS_SHORT$);
+
+err:
+        if (local) {
+                if (local->stub) {
+                        call_stub_destroy(local->stub);
+                }
+                if (local->qstub) {
+                        call_stub_destroy(local->qstub);
+                }
+                if (local->fd) {
+                        fd_unref(local->fd);
+                }
+                mem_put(local);
+        }
+        STACK_UNWIND_STRICT ($NAME$, frame, -1, op_errno,
+                             $DEFAULTS$);
+        return 0;
+}
+
+// template-name write-dispatch
+$TYPE$
+nsr_$NAME$_dispatch (call_frame_t *frame, xlator_t *this,
+                     $ARGS_LONG$)
+{
+        nsr_local_t     *local  = frame->local;
+        nsr_private_t   *priv   = this->private;
+        xlator_list_t   *trav;
+
+        atomic_inc(&priv->ops_in_flight);
+
+        /*
+         * TBD: unblock pending request(s) if we fail after this point but
+         * before we get to nsr_$NAME$_complete (where that code currently
+         * resides).
+         */
+
+        local->call_count = priv->n_children - 1;
+        for (trav = this->children->next; trav; trav = trav->next) {
+                STACK_WIND (frame, nsr_$NAME$_fan_in,
+                            trav->xlator, trav->xlator->fops->$NAME$,
+                            $ARGS_SHORT$);
+        }
+
+        // TBD: variable Issue count
+        return 0;
+}
+
+// template-name write-fan-in
+$TYPE$
+nsr_$NAME$_fan_in (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno,
+                   $ARGS_LONG$)
+{
+        nsr_local_t     *local  = frame->local;
+        uint8_t         call_count;
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "op_ret = %d, op_errno = %d\n", op_ret, op_errno);
+
+        LOCK(&frame->lock);
+        call_count = --(local->call_count);
+        UNLOCK(&frame->lock);
+
+        // TBD: variable Completion count
+        if (call_count == 0) {
+                call_resume(local->stub);
+        }
+
+        return 0;
+}
+
+// template-name write-continue
+$TYPE$
+nsr_$NAME$_continue (call_frame_t *frame, xlator_t *this,
+                     $ARGS_LONG$)
+{
+        STACK_WIND (frame, nsr_$NAME$_complete,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->$NAME$,
+                    $ARGS_SHORT$);
+        return 0;
+}
+
+// template-name write-complete
+$TYPE$
+nsr_$NAME$_complete (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno,
+                     $ARGS_LONG$)
+{
+        nsr_private_t   *priv           = this->private;
+#if defined(NSR_CG_NEED_FD)
+        nsr_local_t *local = frame->local;
+#endif
+
+#if defined(NSR_CG_QUEUE)
+        nsr_inode_ctx_t *ictx;
+        nsr_local_t     *next;
+        if (local->qlinks.next != &local->qlinks) {
+                list_del(&local->qlinks);
+                ictx = nsr_get_inode_ctx(this,local->fd->inode);
+                if (ictx) {
+                    LOCK(&ictx->lock);
+                        if (ictx->pending) {
+                                /*
+                                 * TBD: dequeue *all* non-conflicting reqs
+                                 *
+                                 * With the stub implementation there can only
+                                 * be one request active at a time (zero here)
+                                 * so it's not an issue.  In a real
+                                 * implementation there might still be other
+                                 * active requests to check against, and
+                                 * multiple pending requests that could
+                                 * continue.
+                                 */
+                                gf_log (this->name, GF_LOG_DEBUG,
+                                        "unblocking next request");
+                                --(ictx->pending);
+                                next = list_entry (ictx->pqueue.next,
+                                                   nsr_local_t, qlinks);
+                                list_del(&next->qlinks);
+                                list_add_tail(&next->qlinks,&ictx->aqueue);
+                                call_resume(next->qstub);
+                        }
+                        else {
+                                --(ictx->active);
+                        }
+                    UNLOCK(&ictx->lock);
+                }
+        }
+#endif
+
+#if defined(NSR_CG_FSYNC)
+        nsr_mark_fd_dirty(this,local);
+#endif
+
+#if defined(NSR_CG_NEED_FD)
+        fd_unref(local->fd);
+#endif
+
+        STACK_UNWIND_STRICT ($NAME$, frame, op_ret, op_errno,
+                                     $ARGS_SHORT$);
+        atomic_dec(&priv->ops_in_flight);
+        return 0;
+        
+}
diff --git a/xlators/cluster/nsr-server/src/codegen.py b/xlators/cluster/nsr-server/src/codegen.py
new file mode 100755
index 000000000..709f5662f
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/codegen.py
@@ -0,0 +1,174 @@
+#!/usr/bin/python
+
+# This module lets us auto-generate boilerplate versions of fops and cbks,
+# both for the client side and (eventually) on the server side as well.  This
+# allows us to implement common logic (e.g. leader fan-out and sequencing)
+# once, without all the problems that come with copying and pasting the same
+# code into dozens of functions (or failing to).
+#
+# I've tried to make this code pretty generic, since it's already likely to
+# be used multiple ways within NSR.  Really, we should use something like this
+# to generate defaults.[ch] as well, to avoid the same sorts of mismatches
+# that we've already seen and to which this approach makes NSR immune.  That
+# would require using something other than defaults.h as the input, but that
+# format could be even simpler so that's a good thing too.
+
+
+import re
+import sys
+
+decl_re = re.compile("([a-z0-9_]+)$")
+tmpl_re = re.compile("// template-name (.*)")
+
+class CodeGenerator:
+
+	def __init__ (self):
+		self.decls = {}
+		self.skip = 0
+		self.templates = {}
+		self.make_defaults = self._make_defaults
+
+	# Redefine this to preprocess the name in a declaration, e.g.
+	#	fop_lookup_t => nsrc_lookup
+	def munge_name (self, orig):
+		return orig
+
+	# By default, this will convert the argument string into a sequence of
+	# (type, name) tuples minus the first self.skip (default zero) arguments.
+	# You can redefine it to skip the conversion, do a different conversion,
+	# or rearrange the arguments however you like.
+	def munge_args (self, orig):
+		args = []
+		for decl in orig.strip("(); ").split(","):
+			m = decl_re.search(decl)
+			if m:
+				args.append((m.group(1),decl[:m.start(1)].strip()))
+			else:
+				raise RuntimeError("can't split %s into type+name"%decl)
+		return args[self.skip:]
+
+	def add_decl (self, fname, ftype, fargs):
+		self.decls[self.munge_name(fname)] = (ftype, self.munge_args(fargs))
+
+	def parse_decls (self, path, pattern):
+		regex = re.compile(pattern)
+		f = open(path,"r")
+		have_decl = False
+		while True:
+			line = f.readline()
+			if not line:
+				break
+			m = regex.search(line)
+			if m:
+				if have_decl:
+					self.add_decl(f_name,f_type,f_args)
+				f_name = m.group(2)
+				f_type = m.group(1)
+				f_args = line[m.end(0):-1].strip()
+				if f_args.rfind(")") >= 0:
+					self.add_decl(f_name,f_type,f_args)
+				else:
+					have_decl = True
+			elif have_decl:
+				if line.strip() == "":
+					self.add_decl(f_name,f_type,f_args)
+					have_decl = False
+				else:
+					f_args += " "
+					f_args += line[:-1].strip()
+		if have_decl:
+			self.add_decl(f_name,f_type,f_args)
+
+	# Legacy function (yeah, already) to load a single template.  If you're
+	# using multiple templates, you're better off loading them all from one
+	# file using load_templates (note plural) instead.
+	def load_template (self, name, path):
+		self.templates[name] = open(path,"r").readlines()
+
+	# Load multiple templates.  Each is introduced by a special comment of
+	# the form
+	#
+	#	// template-name xyz
+	#
+	# One side effect is that the block before the first such comment will be
+	# ignored.  This seems like it might be useful some day so I'll leave it
+	# in, but if people trip over it maybe it will change.
+	#
+	# It is recommended to define templates in expected execution order, to
+	# make the result more readable than the inverted order (e.g. callback
+	# then fop) common in the rest of our code.
+	def load_templates (self, path):
+		t_name = None
+		for line in open(path,"r").readlines():
+			if not line:
+				break
+			m = tmpl_re.match(line)
+			if m:
+				if t_name:
+					self.templates[t_name] = t_contents
+				t_name = m.group(1).strip()
+				t_contents = []
+			elif t_name:
+				t_contents.append(line)
+		if t_name:
+			self.templates[t_name] = t_contents
+
+	# Emit the template, with the following expansions:
+	#
+	#	$NAME$			=> function name (as passed in)
+	#	$TYPE$			=> function return value
+	#	$ARGS_SHORT$	=> argument list, including types
+	#	$ARGS_LONG$		=> argument list, *not* including types
+	#	$DEFAULTS$		=> default callback args (see below)
+	#
+	# The $DEFAULTS$ substitution is for the case where a fop (which has one
+	# set of arguments) needs to signal an error via STACK_UNWIND (which
+	# requires a different set of arguments).  In this case we look up the
+	# argument list for the opposite direction, using self.make_defaults which
+	# the user must explicitly set to the method for the opposite direction.
+	# If an argument is a pointer, we replace it with NULL; otherwise we
+	# replace it with zero.  It's a hack, but it's the only thing we do that
+	# doesn't require specific knowledge of our environment and the specific
+	# call we're handling.  If this doesn't suffice, we'll have to add
+	# something like $ARG0$ which can be passed in for specific cases.
+	def emit (self, f_name, tmpl):
+		args = self.decls[f_name][1]
+		zipper = lambda x: x[0]
+		a_short = ", ".join(map(zipper,args))
+		zipper = lambda x: x[1] + " " + x[0]
+		a_long = ", ".join(map(zipper,args))
+		for line in self.templates[tmpl]:
+			line = line.replace("$NAME$",f_name)
+			line = line.replace("$TYPE$",self.decls[f_name][0])
+			line = line.replace("$ARGS_SHORT$",a_short)
+			line = line.replace("$ARGS_LONG$",a_long)
+			line = line.replace("$DEFAULTS$",self.make_defaults(f_name))
+			print(line.rstrip())
+
+	def _make_defaults (self, f_name):
+		result = []
+		for arg in self.decls[f_name][1]:
+			if arg[1][-1] == "*":
+				result.append("NULL")
+			else:
+				result.append("0")
+		return ", ".join(result)
+
+if __name__ == "__main__":
+	type_re = "([a-z_0-9]+)"
+	name_re = "\(\*fop_([a-z0-9]+)_t\)"
+	full_re = type_re + " *" + name_re
+	cg = CodeGenerator()
+	cg.skip = 2
+	cg.parse_decls(sys.argv[1],full_re)
+	"""
+	for k, v in cg.decls.iteritems():
+		print("=== %s" % k)
+		print("  return type %s" % v[0])
+		for arg in v[1]:
+			print("  arg %s (type %s)" % arg)
+	"""
+	cg.load_template("fop",sys.argv[2])
+	cg.emit("lookup","fop")
+	cg.emit("rename","fop")
+	cg.emit("setxattr","fop")
diff --git a/xlators/cluster/nsr-server/src/etcd-api.c b/xlators/cluster/nsr-server/src/etcd-api.c
new file mode 100644
index 000000000..a07019244
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/etcd-api.c
@@ -0,0 +1,831 @@
+/*
+ * Copyright (c) 2013, Red Hat
+ * All rights reserved.
+
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+ * list of conditions and the following disclaimer.  Redistributions in binary
+ * form must reproduce the above copyright notice, this list of conditions and
+ * the following disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/* For asprintf */
+#if !defined(_GNU_SOURCE)
+#define _GNU_SOURCE
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <curl/curl.h>
+#include <yajl/yajl_tree.h>
+#include "etcd-api.h"
+
+
+#define DEFAULT_ETCD_PORT       4001
+#define SL_DELIM                "\n\r\t ,;"
+
+typedef struct {
+        etcd_server     *servers;
+} _etcd_session;
+
+typedef struct {
+        char            *key;
+        char            *value;
+        int             *index_in;      /* pointer so NULL can be special */
+        int             index_out;      /* NULL would be meaningless */
+} etcd_watch_t;
+
+typedef size_t curl_callback_t (void *, size_t, size_t, void *);
+
+int             g_inited        = 0;
+const char      *value_path[]   = { "node", "value", NULL };
+const char      *nodes_path[]   = { "node", "nodes", NULL };
+const char      *entry_path[]   = { "key", NULL };
+
+/*
+ * We only call this in case where it should be safe, but gcc doesn't know
+ * that so we use this to shut it up.
+ */
+char *
+MY_YAJL_GET_STRING (yajl_val x)
+{
+        char *y = YAJL_GET_STRING(x);
+
+        return y ? y : "bogus";
+}
+
+#if defined(DEBUG)
+void
+print_curl_error (char *intro, CURLcode res)
+{
+        printf("%s: %s\n",intro,curl_easy_strerror(res));
+}
+#else
+#define print_curl_error(intro,res)
+#endif
+
+ 
+etcd_session
+etcd_open (etcd_server *server_list)
+{
+        _etcd_session   *session;
+
+        if (!g_inited) {
+                curl_global_init(CURL_GLOBAL_ALL);
+                g_inited = 1;
+        }
+
+        session = malloc(sizeof(*session));
+        if (!session) {
+                return NULL;
+        }
+
+        /*
+         * Some day we'll set up more persistent connections, and keep track
+         * (via redirects) of which server is leader so that we can always
+         * try it first.  For now we just push that to the individual request
+         * functions, which do the most brain-dead thing that can work.
+         */
+
+        session->servers = server_list;
+        return session;
+}
+
+
+void
+etcd_close (etcd_session session)
+{
+        free(session);
+}
+
+/*
+ * Normal yajl_tree_get is returning NULL for these paths even when I can
+ * verify (in gdb) that they exist.  I suppose I could debug this for them, but
+ * this is way easier.
+ *
+ * TBD: see if common distros are packaging a JSON library that isn't total
+ * crap.
+ */
+yajl_val
+my_yajl_tree_get (yajl_val root, char const **path, yajl_type type)
+{
+        yajl_val        obj    = root;
+        int             i;
+
+        for (;;) {
+                if (!*path) {
+                        if (obj && (obj->type != type)) {
+                                return NULL;
+                        }
+                        return obj;
+                }
+                if (obj->type != yajl_t_object) {
+                        return NULL;
+                }
+                for (i = 0; /* nothing */; ++i) {
+                        if (i >= obj->u.object.len) {
+                                return NULL;
+                        }
+                        if (!strcmp(obj->u.object.keys[i],*path)) {
+                                obj = obj->u.object.values[i];
+                                ++path;
+                                break;
+                        }
+                }
+        }
+}
+
+
+/*
+ * Looking directly at node->u.array seems terribly un-modular, but the YAJL
+ * tree interface doesn't seem to have any exposed API for iterating over the
+ * elements of an array.  I tried using yajl_tree_get with an index in the
+ * path, either as a type-casted integer or as a string, but that didn't work.
+ */
+char *
+parse_array_response (yajl_val parent)
+{
+        size_t          i;
+        yajl_val        item;
+        yajl_val        value;
+        char            *retval = NULL;
+        char            *saved;
+        yajl_val        node;
+
+        node = my_yajl_tree_get(parent,nodes_path,yajl_t_array);
+        if (!node) {
+                return NULL;
+        }
+
+        for (i = 0; i < node->u.array.len; ++i) {
+                item = node->u.array.values[i];
+                if (!item) {
+                        break;
+                }
+                value = my_yajl_tree_get(item,entry_path,yajl_t_string);
+                if (!value) {
+                        break;
+                }
+                if (retval) {
+                        saved = retval;
+                        retval = NULL;
+                        (void)asprintf (&retval, "%s\n%s",
+                                        saved, MY_YAJL_GET_STRING(value));
+                        free(saved);
+                }
+                else {
+                        retval = strdup(MY_YAJL_GET_STRING(value));
+                }
+                if (!retval) {
+                        break;
+                }
+        }
+
+        return retval;
+}
+
+size_t
+parse_get_response (void *ptr, size_t size, size_t nmemb, void *stream)
+{
+        yajl_val        node;
+        yajl_val        value;
+
+        node = yajl_tree_parse(ptr,NULL,0);
+        if (node) {
+                value = my_yajl_tree_get(node,value_path,yajl_t_string);
+                if (value) {
+                        /* 
+                         * YAJL probably copied it once, now we're going to
+                         * copy it again.  If anybody really cares for such
+                         * small and infrequently used values, we'd have to do
+                         * do something much more complicated (like using the
+                         * stream interface) to avoid the copy.  Right now it's
+                         * just not worth it.
+                         */
+                        *((char **)stream) = strdup(MY_YAJL_GET_STRING(value));
+                }
+                else {
+                        /* Might as well try this. */
+                        *((char **)stream) = parse_array_response(node);
+                }
+                yajl_tree_free(node);
+        }
+
+        return size*nmemb;
+}
+
+
+etcd_result
+etcd_get_one (_etcd_session *session, char *key, etcd_server *srv, char *prefix,
+              char *post, curl_callback_t cb, char **stream)
+{
+        char            *url;
+        CURL            *curl;
+        CURLcode        curl_res;
+        etcd_result     res             = ETCD_WTF;
+        void            *err_label      = &&done;
+
+        if (asprintf(&url,"http://%s:%u/v2/%s%s",
+                     srv->host,srv->port,prefix,key) < 0) {
+                goto *err_label;
+        }
+        printf("url = %s\n",url);
+        err_label = &&free_url;
+
+        curl = curl_easy_init();
+        if (!curl) {
+                goto *err_label;
+        }
+        err_label = &&cleanup_curl;
+
+        /* TBD: add error checking for these */
+        curl_easy_setopt(curl,CURLOPT_URL,url);
+        curl_easy_setopt(curl,CURLOPT_FOLLOWLOCATION,1L);
+        curl_easy_setopt(curl,CURLOPT_WRITEFUNCTION,cb);
+        curl_easy_setopt(curl,CURLOPT_WRITEDATA,stream);
+        if (post) {
+                curl_easy_setopt(curl,CURLOPT_POST,1L);
+                curl_easy_setopt(curl,CURLOPT_POSTFIELDS,post);
+        }
+#if defined(DEBUG)
+        curl_easy_setopt(curl,CURLOPT_VERBOSE,1L);
+#endif
+
+        curl_res = curl_easy_perform(curl);
+        if (curl_res != CURLE_OK) {
+                print_curl_error("perform",curl_res);
+                goto *err_label;
+        }
+
+        res = ETCD_OK;
+
+cleanup_curl:
+        curl_easy_cleanup(curl);
+free_url:
+        free(url);
+done:
+        return res;
+}
+
+
+char *
+etcd_get (etcd_session session_as_void, char *key)
+{
+        _etcd_session   *session   = session_as_void;
+        etcd_server     *srv;
+        etcd_result     res;
+        char            *value  = NULL;
+
+        for (srv = session->servers; srv->host; ++srv) {
+                res = etcd_get_one(session,key,srv,"keys/",NULL,
+                                   parse_get_response,&value);
+                if ((res == ETCD_OK) && value) {
+                        return value;
+                }
+        }
+
+        return NULL;
+}
+
+
+size_t
+parse_watch_response (void *ptr, size_t size, size_t nmemb, void *stream)
+{
+        yajl_val                node;
+        yajl_val                value;
+        etcd_watch_t            *watch  = stream;
+        static const char       *i_path[] = { "node", "modifiedIndex", NULL };
+        static const char       *k_path[] = { "node", "key", NULL };
+        static const char       *v_path[] = { "node", "value", NULL };
+
+        node = yajl_tree_parse(ptr,NULL,0);
+        if (node) {
+                value = my_yajl_tree_get(node,i_path,yajl_t_number);
+                if (value) {
+                        watch->index_out = strtoul(YAJL_GET_NUMBER(value),
+                                                   NULL,10);
+                }
+                value = my_yajl_tree_get(node,k_path,yajl_t_string);
+                if (value) {
+                        watch->key = strdup(MY_YAJL_GET_STRING(value));
+                }
+                value = my_yajl_tree_get(node,v_path,yajl_t_string);
+                if (value) {
+                        watch->value = strdup(MY_YAJL_GET_STRING(value));
+                }
+        }
+
+        return size*nmemb;
+}
+
+
+etcd_result
+etcd_watch (etcd_session session_as_void, char *pfx,
+            char **keyp, char **valuep, int *index_in, int *index_out)
+{
+        _etcd_session   *session   = session_as_void;
+        etcd_server     *srv;
+        etcd_result     res;
+        etcd_watch_t    watch;
+        char            *path;
+
+        if (index_in) {
+                if (asprintf(&path,"%s?wait=true&recursive=true&waitIndex=%d",
+                             pfx,*index_in) < 0) {
+                        return ETCD_WTF;
+                }
+        }
+        else {
+                if (asprintf(&path,"%s?wait=true&recursive=true",pfx) < 0) {
+                        return ETCD_WTF;
+                }
+        }
+
+        memset(&watch,0,sizeof(watch));
+        watch.index_in = index_in;
+
+        for (srv = session->servers; srv->host; ++srv) {
+                res = etcd_get_one(session,path,srv,"keys/",NULL,
+                                   parse_watch_response,(char **)&watch);
+                if (res == ETCD_OK) {
+                        if (keyp) {
+                                *keyp = watch.key;
+                        }
+                        if (valuep) {
+                                *valuep = watch.value;
+                        }
+                        if (index_out) {
+                                *index_out = watch.index_out;
+                        }
+                        break;
+                }
+        }
+
+        free(path);
+        return res;
+}
+
+
+size_t
+parse_set_response (void *ptr, size_t size, size_t nmemb, void *stream)
+{
+        yajl_val        node;
+        yajl_val        value;
+        etcd_result     res     = ETCD_PROTOCOL_ERROR;
+        /*
+         * Success responses contain prevValue and index.  Failure responses
+         * contain errorCode and cause.  Among all these, index seems to be the
+         * one we're most likely to need later, so look for that.
+         */
+        static const char       *path[] = { "node", "modifiedIndex", NULL };
+
+        node = yajl_tree_parse(ptr,NULL,0);
+        if (node) {
+                value = my_yajl_tree_get(node,path,yajl_t_number);
+                if (value) {
+                        res = ETCD_OK;
+                }
+        }
+
+        *((etcd_result *)stream) = res;
+        return size*nmemb;
+}
+
+
+size_t
+parse_lock_response (void *ptr, size_t size, size_t nmemb, void *stream)
+{
+        *((char **)stream) = strdup(ptr);
+        return size*nmemb;
+}
+
+
+/* 
+ * There are two use cases, based on is_lock.
+ *
+ * If is_lock is null, we use the "keys" namespace.  A null value means an
+ * HTTP DELETE; precond and ttl are both ignored.  Otherwise we're setting a
+ * value, with *optional* precond and ttl.
+ *
+ * If is_lock is set, we use the "locks" namespace.  A null value means an
+ * HTTP DELETE as before, and we still ignore ttl as before, but now precond
+ * must be set to represent the lock index.  Otherwise ttl must be present,
+ * and we decide what to do based on precond.  If it's null, this is an
+ * initial lock so we use an HTTP POST.  Otherwise it's a renewal so we use
+ * an HTTP PUT instead.
+ */
+etcd_result
+etcd_set_one (_etcd_session *session, char *key, char *value,
+              char *precond, unsigned int ttl, etcd_server *srv,
+              char **is_lock)
+{
+        char                    *url;
+        char                    *contents       = NULL;
+        CURL                    *curl;
+        etcd_result             res             = ETCD_WTF;
+        CURLcode                curl_res;
+        void                    *err_label      = &&done;
+        char                    *namespace;
+        char                    *http_cmd;
+        char                    *orig_index;
+
+        if (is_lock) {
+                namespace = "mod/v2/lock";
+                if (value) {
+                        if (!ttl) {
+                                /* Lock/renew must specify ttl. */
+                                return ETCD_WTF;
+                        }
+                        http_cmd = precond ? "PUT" : "POST";
+                }
+                else {
+                        if (!precond) {
+                                /* Unlock must specify index. */
+                                return ETCD_WTF;
+                        }
+                        http_cmd = "DELETE";
+                }
+                orig_index = *is_lock;
+        }
+        else {
+                namespace = "v2/keys";
+                http_cmd = value ? "PUT" : "DELETE";
+        }
+
+        if (asprintf(&url,"http://%s:%u/%s/%s",
+                     srv->host,srv->port,namespace,key) < 0) {
+                goto *err_label;
+        }
+        err_label = &&free_url;
+
+        if (is_lock) {
+                if (precond) {
+                        if (asprintf(&contents,"index=%s",precond) < 0) {
+                                goto *err_label;
+                        }
+                        err_label = &&free_contents;
+                }
+                if (ttl) {
+                        if (contents) {
+                                char *c2;
+                                if (asprintf(&c2,"ttl=%u;%s",ttl,contents) < 0) {
+                                        goto *err_label;
+                                }
+                                free(contents);
+                                contents = c2;
+                        }
+                        else {
+                                if (asprintf(&contents,"ttl=%u",ttl) < 0) {
+                                        goto *err_label;
+                                }
+                        }
+                        err_label = &&free_contents;
+                }
+        }
+        else {
+                if (value) {
+                        if (asprintf(&contents,"value=%s",value) < 0) {
+                                goto *err_label;
+                        }
+                        err_label = &&free_contents;
+                }
+                if (precond) {
+                        char *c2;
+                        if (asprintf(&c2,"%s;prevValue=%s",contents,
+                                     precond) < 0) {
+                                goto *err_label;
+                        }
+                        free(contents);
+                        contents = c2;
+                        err_label = &&free_contents;
+                }
+                if (ttl) {
+                        char *c2;
+                        if (asprintf(&c2,"%s;ttl=%u",contents,ttl) < 0) {
+                                goto *err_label;
+                        }
+                        free(contents);
+                        contents = c2;
+                        err_label = &&free_contents;
+                }
+        }
+
+        curl = curl_easy_init();
+        if (!curl) {
+                goto *err_label;
+        }
+        err_label = &&cleanup_curl;
+
+        /* TBD: add error checking for these */
+        curl_easy_setopt(curl,CURLOPT_CUSTOMREQUEST,http_cmd);
+        curl_easy_setopt(curl,CURLOPT_URL,url);
+        curl_easy_setopt(curl,CURLOPT_FOLLOWLOCATION,1L);
+        curl_easy_setopt(curl,CURLOPT_POSTREDIR,CURL_REDIR_POST_ALL);
+
+        if (is_lock && value && !precond) {
+                /* Only do this for an initial lock, not a renewal. */
+                curl_easy_setopt (curl, CURLOPT_WRITEFUNCTION,
+                                  parse_lock_response);
+                curl_easy_setopt(curl,CURLOPT_WRITEDATA,is_lock);
+        }
+        else {
+                curl_easy_setopt (curl, CURLOPT_WRITEFUNCTION,
+                                  parse_set_response);
+                curl_easy_setopt(curl,CURLOPT_WRITEDATA,&res);
+        }
+
+        /*
+         * CURLOPT_HTTPPOST would be easier, but it looks like etcd will barf on
+         * that.  Sigh.
+         */
+        if (contents) {
+                curl_easy_setopt(curl,CURLOPT_POST,1L);
+                curl_easy_setopt(curl,CURLOPT_POSTFIELDS,contents);
+        }
+#if defined(DEBUG)
+        curl_easy_setopt(curl,CURLOPT_VERBOSE,1L);
+#endif
+
+        curl_res = curl_easy_perform(curl);
+        if (curl_res != CURLE_OK) {
+                print_curl_error("perform",curl_res);
+                goto *err_label;
+        }
+
+        if (is_lock && value) {
+                if (!precond) {
+                        /*
+                         * If this is an initial lock, parse_lock_response would
+                         * have been unable to set "res" for us.  Instead, we
+                         * set it here if the index string got updated.
+                         */
+                        if (*is_lock != orig_index) {
+                                res = ETCD_OK;
+                        }
+                }
+                else {
+                        /*
+                         * If this is a lock renewal, then a successful call
+                         * will pass through neither parse_lock_response nor
+                         * parse_get_response.  The curl response code alone
+                         * is sufficient.
+                         */
+                        res = ETCD_OK;
+                }
+        }
+
+        /*
+         * If the request succeeded, or at least got to the server and failed
+         * there, parse_set_response should have set res appropriately.
+         */
+
+cleanup_curl:
+        curl_easy_cleanup(curl);
+free_contents:
+        free(contents); /* might already be NULL for delete, but that's OK */
+free_url:
+        free(url);
+done:
+        return res;
+}
+
+
+etcd_result
+etcd_set (etcd_session session_as_void, char *key, char *value,
+          char *precond, unsigned int ttl)
+{
+        _etcd_session   *session   = session_as_void;
+        etcd_server     *srv;
+        etcd_result     res;
+
+        for (srv = session->servers; srv->host; ++srv) {
+                res = etcd_set_one(session,key,value,precond,ttl,srv,NULL);
+                /*
+                 * Protocol errors are likely to be things like precondition
+                 * failures, which won't be helped by retrying on another
+                 * server.
+                 */
+                if ((res == ETCD_OK) || (res == ETCD_PROTOCOL_ERROR)) {
+                        return res;
+                }
+        }
+
+        return ETCD_WTF;
+}
+
+
+/*
+ * This uses the same path and status checks as SET, but with a different HTTP
+ * command instead of data.  Precondition and TTL are obviously not used in
+ * this case, though a conditional delete would be a cool feature for etcd.  I
+ * think you can get a timed delete by doing a conditional set to the current
+ * value with a TTL, but I haven't actually tried it.
+ */
+etcd_result
+etcd_delete (etcd_session session_as_void, char *key)
+{
+        _etcd_session   *session   = session_as_void;
+        etcd_server     *srv;
+        etcd_result     res;
+
+        for (srv = session->servers; srv->host; ++srv) {
+                res = etcd_set_one(session,key,NULL,NULL,0,srv,NULL);
+                if (res == ETCD_OK) {
+                        break;
+                }
+        }
+
+        return res;
+}
+
+
+etcd_result
+etcd_lock (etcd_session session_as_void, char *key, unsigned int ttl,
+           char *index_in, char **index_out)
+{
+        _etcd_session   *session        = session_as_void;
+        etcd_server     *srv;
+        etcd_result     res;
+        char            *tmp            = NULL;
+
+        for (srv = session->servers; srv->host; ++srv) {
+                res = etcd_set_one(session,key,"hack",index_in,ttl,srv,&tmp);
+                if (res == ETCD_OK) {
+                        if (index_out) {
+                                *index_out = tmp;
+                        }
+                        break;
+                }
+        }
+
+        return res;
+}
+
+
+etcd_result
+etcd_unlock (etcd_session session_as_void, char *key, char *index)
+{
+        _etcd_session   *session   = session_as_void;
+        etcd_server     *srv;
+        etcd_result     res;
+        char            *tmp            = NULL;
+
+        for (srv = session->servers; srv->host; ++srv) {
+                res = etcd_set_one(session,key,NULL,index,0,srv,&tmp);
+                if (res == ETCD_OK) {
+                        break;
+                }
+        }
+
+        return res;
+}
+size_t
+store_leader (void *ptr, size_t size, size_t nmemb, void *stream)
+{
+        *((char **)stream) = strdup(ptr);
+        return size * nmemb;
+}
+
+
+char *
+etcd_leader (etcd_session session_as_void)
+{
+        _etcd_session   *session   = session_as_void;
+        etcd_server     *srv;
+        etcd_result     res;
+        char            *value  = NULL;
+
+        for (srv = session->servers; srv->host; ++srv) {
+                res = etcd_get_one(session,"leader",srv,"",NULL,
+                                   store_leader,&value);
+                if ((res == ETCD_OK) && value) {
+                        return value;
+                }
+        }
+
+        return NULL;
+}
+
+
+void
+free_sl (etcd_server *server_list)
+{
+        size_t          num_servers;
+
+        for (num_servers = 0; server_list[num_servers].host; ++num_servers) {
+                free(server_list[num_servers].host);
+        }
+        free(server_list);
+}
+
+
+int
+_count_matching (char *text, char *cset, int result)
+{
+        char    *t;
+        int     res     = 0;
+
+        for (t = text; *t; ++t) {
+                if ((strchr(cset,*t) != NULL) != result) {
+                        break;
+                }
+                ++res;
+        }
+
+        return res;
+}
+
+#define count_matching(t,cs)    _count_matching(t,cs,1)
+#define count_nonmatching(t,cs) _count_matching(t,cs,0)
+
+
+etcd_session
+etcd_open_str (char *server_names)
+{
+        char            *snp;
+        int             run_len;
+        int             host_len;
+        size_t           num_servers;
+        etcd_server     *server_list;
+        etcd_session    *session;
+
+        /*
+         * Yeah, we iterate over the string twice so we can allocate an
+         * appropriately sized array instead of turning it into a linked list.
+         * Unfortunately this means we can't use strtok* which is destructive
+         * with no platform-independent way to reverse the destructive effects.
+         */
+
+        num_servers = 0;
+        snp = server_names;
+        while (*snp) {
+                run_len = count_nonmatching(snp,SL_DELIM);
+                if (!run_len) {
+                        snp += count_matching(snp,SL_DELIM);
+                        continue;
+                }
+                ++num_servers;
+                snp += run_len;
+        }
+
+        if (!num_servers) {
+                return NULL;
+        }
+
+        server_list = calloc(num_servers+1,sizeof(*server_list));
+        if (!server_list) {
+                return NULL;
+        }
+        num_servers = 0;
+
+        snp = server_names;
+        while (*snp) {
+                run_len = count_nonmatching(snp,SL_DELIM);
+                if (!run_len) {
+                        snp += count_matching(snp,SL_DELIM);
+                        continue;
+                }
+                host_len = count_nonmatching(snp,":");
+                if ((run_len - host_len) > 1) {
+                        server_list[num_servers].host = strndup(snp,host_len);
+                        server_list[num_servers].port = (unsigned short)
+                                strtoul(snp+host_len+1,NULL,10);
+                }
+                else {
+                        server_list[num_servers].host = strndup(snp,run_len);
+                        server_list[num_servers].port = DEFAULT_ETCD_PORT;
+                }
+                ++num_servers;
+                snp += run_len;
+        }
+
+        session = etcd_open(server_list);
+        if (!session) {
+                free_sl(server_list);
+        }
+        return session;
+}
+
+
+void
+etcd_close_str (etcd_session session)
+{
+        free_sl(((_etcd_session *)session)->servers);
+        etcd_close(session);
+}
diff --git a/xlators/cluster/nsr-server/src/etcd-api.h b/xlators/cluster/nsr-server/src/etcd-api.h
new file mode 100644
index 000000000..66275d40d
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/etcd-api.h
@@ -0,0 +1,214 @@
+/*
+ * Copyright (c) 2013, Red Hat
+ * All rights reserved.
+
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+ * list of conditions and the following disclaimer.  Redistributions in binary
+ * form must reproduce the above copyright notice, this list of conditions and
+ * the following disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/*
+ * Description of an etcd server.  For now it just includes the name and
+ * port, but some day it might include other stuff like SSL certificate
+ * information.
+ */
+
+typedef enum {
+        ETCD_OK = 0,
+        ETCD_PROTOCOL_ERROR,
+                                /* TBD: add other error categories here */
+        ETCD_WTF                /* anything we can't easily categorize */
+} etcd_result;
+
+typedef struct {
+        char            *host;
+        unsigned short  port;
+} etcd_server;
+
+typedef void *etcd_session;
+
+/*
+ * etcd_open
+ *
+ * Establish a session to an etcd cluster, with automatic reconnection and
+ * so on.
+ *
+ *      server_list
+ *      Array of etcd_server structures, with the last having host=NULL.  The
+ *      caller is responsible for ensuring that this remains valid as long as
+ *      the session exists.
+ */
+etcd_session    etcd_open       (etcd_server *server_list);
+
+
+/*
+ * etcd_open_str
+ *
+ * Same as etcd_open, except that the servers are specified as a list of
+ * host:port strings, separated by comma/semicolon or whitespace.
+ */
+etcd_session    etcd_open_str   (char *server_names);
+
+
+/*
+ * etcd_close
+ *
+ * Terminate a session, closing connections and freeing memory (or any other
+ * resources) associated with it.
+ */
+void            etcd_close      (etcd_session session);
+
+
+/*
+ * etcd_close
+ *
+ * Same as etcd_close, but also free the server list as etcd_open_str would
+ * have allocated it.
+ */
+void            etcd_close_str  (etcd_session session);
+
+
+/*
+ * etcd_get
+ *
+ * Fetch a key from one of the servers in a session.  The return value is a
+ * newly allocated string, which must be freed by the caller.
+ *
+ *      key
+ *      The etcd key (path) to fetch.
+ */
+char *          etcd_get (etcd_session session, char *key);
+
+
+/*
+ * etcd_watch
+ * Watch the set of keys matching a prefix.
+ *
+ *      pfx
+ *      The etcd key prefix (like a path) to watch.
+ *
+ *      keyp
+ *      Space for a pointer to the key that was added/modified/deleted.
+ *
+ *      valuep
+ *      Space for a pointer to the value if a key was added/modified.  A delete
+ *      is signified by this being set to NULL.
+ *
+ *      index_in
+ *      Pointer to an index to be used for *issuing* the watch request, or
+ *      NULL for a watch without an index.
+ *
+ *      index_out
+ *      Pointer to space for an index *returned* by etcd, or NULL to mean don't
+ *      bother.
+ *
+ * In normal usage, index_in will be NULL and index_out will be set to receive
+ * the index for the first watch.  Subsequently, index_in will be set to
+ * provide the previous index (plus one) and index_out will be set to receive
+ * the next.  It's entirely legitimate to point both at the same variable.
+ */
+
+etcd_result     etcd_watch (etcd_session session, char *pfx,
+                            char **keyp, char **valuep,
+                            int *index_in, int *index_out);
+
+
+/*
+ * etcd_set
+ *
+ * Write a key, with optional TTL and/or previous value (as a precondition).
+ *
+ *      key
+ *      The etcd key (path) to set.
+ *
+ *      value
+ *      New value as a null-terminated string.  Unlike etcd_get, we can derive
+ *      the length ourselves instead of needing it to be passed in separately.
+ *
+ *      precond
+ *      Required previous value as a null-terminated string, or NULL to mean
+ *      an unconditional set.
+ *
+ *      ttl
+ *      Time in seconds after which the value will automatically expire and be
+ *      deleted, or zero to mean no auto-expiration.
+ */
+
+etcd_result     etcd_set        (etcd_session session, char *key, char *value,
+                                 char *precond, unsigned int ttl);
+
+
+/*
+ * etcd_delete
+ *
+ * Delete a key from one of the servers in a session.
+ *
+ *      key
+ *      The etcd key (path) to delete.
+ */
+
+etcd_result     etcd_delete     (etcd_session session, char *key);
+
+
+/*
+ * etcd_leader
+ *
+ * Get the identify of the current leader.
+ */
+
+char *          etcd_leader     (etcd_session session);
+
+/*
+ * etcd_lock
+ *
+ * Take or renew a lock - really a lease but the etcd folks call it a lock so
+ * we'll follow suit.
+ *
+ *      key
+ *      The path (in the "locks" namespace) for the lock.
+ *
+ *      ttl
+ *      Time in seconds for the lock.
+ *
+ *      index_in (optional, indicates renewal)
+ *      Lock index from previous lock call.
+ *
+ *      index_out (only used for initial lock)
+ *      Place for the new lock index.  You must free this.
+ */
+
+etcd_result     etcd_lock (etcd_session session_as_void, char *key,
+                           unsigned int ttl, char *index_in, char **index_out);
+
+/*
+ * etcd_unlock
+ *
+ * Release a lock (see etcd_lock regarding terminology).
+ *
+ *      key
+ *      The path (in the "locks" namespace) for the lock.
+ *
+ *      index
+ *      Lock index from previous lock call.
+ */
+
+etcd_result     etcd_unlock (etcd_session session_as_void, char *key,
+                             char *index);
+
diff --git a/xlators/cluster/nsr-server/src/etcd-sim.c b/xlators/cluster/nsr-server/src/etcd-sim.c
new file mode 100644
index 000000000..d0bea12c7
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/etcd-sim.c
@@ -0,0 +1,280 @@
+/*
+ * Copyright (c) 2014, Red Hat
+ * All rights reserved.
+
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+ * list of conditions and the following disclaimer.  Redistributions in binary
+ * form must reproduce the above copyright notice, this list of conditions and
+ * the following disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+#include <unistd.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <sys/file.h>
+
+#include "mem-pool.h"
+
+/*
+ * Mock implementation of etcd
+ * The etcd file is simulated in /tmp/<server-names>
+ * Writes from Multiple writers are protected using file lock.
+*/
+
+#include "etcd-api.h"
+#define MAX_KEY_LEN 64
+#define MAX_VALUE_LEN 64
+#define MAX_EXPIRE_LEN 16
+
+etcd_session
+etcd_open (etcd_server *server_list)
+{
+        return NULL;
+}
+
+typedef struct _etcd_sim_s {
+        char *path;
+} etcd_sim_t;
+
+void
+etcd_close (etcd_session this)
+{
+        etcd_sim_t *sim = (etcd_sim_t *)this;
+        free(sim->path);
+        free(this);
+}
+
+
+char *
+etcd_get_1 (FILE *stream, char *key)
+{
+        char *str = NULL;
+        size_t len;
+        unsigned long expires;
+	char *ret;
+
+        // Read the file
+        while(1) {
+                if(str) {
+                        free(str);
+                        str = NULL;
+                }
+                if (getline((char **)&str, &len,stream) == -1) {
+                        break;
+                }
+                if (!strncmp(str, key, strlen(key))) {
+			char k[256], s[256];
+			sscanf(str,"%s %s %lu",k, s, &expires);
+			// check if key is expired.
+			if (time(NULL) > expires) {
+				/* Keep looking for an unexpired entry. */
+				continue;
+			}
+		        ret = calloc(1, strlen(s) + 1);
+		        strcpy(ret,s);
+		        free(str);
+		        return(ret);
+                }
+        }
+        return NULL;
+}
+
+
+char *
+etcd_get (etcd_session this, char *key)
+{
+        etcd_sim_t      *sim            = (etcd_sim_t *)this;
+        int             fd;
+        FILE            *stream;
+        char            *retval;
+
+        fd = open(sim->path,O_RDONLY);
+        if (!fd) {
+                return NULL;
+        }
+
+        stream = fdopen(fd,"r");
+        (void)flock(fd,LOCK_SH);
+        retval = etcd_get_1(stream,key);
+        (void)flock(fd,LOCK_UN);
+        fclose(stream); /* closes fd as well */
+
+        return retval;
+}
+
+
+etcd_result
+etcd_set_1 (FILE *stream, char *key, char *value,
+            char *precond, unsigned int ttl)
+{
+        char *str = NULL;
+        char tp[255];
+        size_t len;
+        unsigned long expires;
+
+        while(1) {
+                if(str) {
+                        free(str);
+                        str = NULL;
+                }
+                if (getline((char **)&str, &len,stream) == -1) {
+                        break;
+                }
+                if (!strncmp(str, key, strlen(key))) {
+                        char k[256], s[256];
+                        sscanf(str,"%s %s %lu",k, s, &expires);
+                        // check if the present key is expired
+                        if (time(NULL) > expires) {
+				/* Keep looking for an unexpired entry. */
+				continue;
+			}
+			/*
+			 * The only case in which we should fail here is if a
+			 * precondition was specified and does not match the
+                         * current (non-expired) value.
+			 */
+			if (precond && strcmp(precond, s)) {
+				free(str);
+				return ETCD_WTF;
+			}
+                        fseek(stream, -strlen(str), SEEK_CUR);               
+                        free(str);
+                        goto here;
+                }
+        }
+here:
+        memset(tp, 0, 255);
+        sprintf(tp,"%*s %*s %*lu\n",
+                -MAX_KEY_LEN, key, -MAX_VALUE_LEN, value,
+                -MAX_EXPIRE_LEN, ttl ? time(NULL) + ttl : ~0);
+        if (fwrite(tp, 1,strlen(tp), stream) != strlen(tp)) {
+                return ETCD_WTF;
+        }
+        fflush(stream);
+        fsync(fileno(stream));
+        return ETCD_OK;
+}
+
+
+etcd_result
+etcd_set (etcd_session this, char *key, char *value,
+          char *precond, unsigned int ttl)
+{
+        etcd_sim_t      *sim = (etcd_sim_t *)this;
+        int             fd;
+        FILE            *stream;
+        etcd_result     retval;
+
+        fd = open(sim->path,O_RDWR);
+        if (fd < 0) {
+                return ETCD_WTF;
+        }
+
+        stream = fdopen(fd,"r+");
+        (void)flock(fd,LOCK_EX);
+        retval = etcd_set_1(stream,key,value,precond,ttl);
+        (void)flock(fd,LOCK_UN);
+        fclose(stream); /* closes fd as well */
+
+        return retval;
+}
+
+
+etcd_session
+etcd_open_str (char *server_names)
+{
+        etcd_sim_t      *sim;
+        int             fd;
+
+        sim = calloc(1, sizeof(etcd_sim_t));
+        (void)asprintf(&sim->path,"/tmp/%s",server_names);
+
+        fd = open(sim->path, O_RDWR | O_CREAT, 0777);
+        if (fd == -1) {
+                free(sim->path);
+                free(sim);
+                return NULL;
+        }
+
+        close(fd);
+        return ((void *)sim);
+}
+
+
+void
+etcd_close_str (etcd_session this)
+{
+        etcd_close(this);
+}
+
+etcd_result
+etcd_delete (etcd_session this, char *key)
+{
+	return ETCD_WTF;
+}
+
+char *
+etcd_leader (etcd_session this_as_void)
+{
+	return NULL;
+}
+
+etcd_result
+etcd_watch (etcd_session this, char *pfx, char **keyp, char **valuep,
+	    int *index_in, int *index_out)
+{
+	return ETCD_WTF;
+}
+
+etcd_result
+etcd_lock (etcd_session session_as_void, char *key, unsigned int ttl,
+           char *index_in, char **index_out)
+{
+        char    *path;
+        int     fd;
+
+        if (!index_in) {
+                if (gf_asprintf(&path,"/var/tmp/%s",key) < 0) {
+                        return ETCD_WTF;
+                }
+                fd = open(path,O_RDWR|O_CREAT,0666);
+                GF_FREE(path);
+                if (fd < 0) {
+                        return ETCD_WTF;
+                }
+                if (flock(fd,LOCK_EX) < 0) {
+                        close(fd);
+                        return ETCD_WTF;
+                }
+                *index_out = strdup("42");
+        }
+
+        /*
+         * Yes, we leak an fd by not closing it here (and nobody else even
+         * knows about it).  That would be awful in any other context, but
+         * for test scripts it won't matter.
+         */
+        return ETCD_OK;
+}
+
diff --git a/xlators/cluster/nsr-server/src/gen-fops.py b/xlators/cluster/nsr-server/src/gen-fops.py
new file mode 100755
index 000000000..1639f489c
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/gen-fops.py
@@ -0,0 +1,120 @@
+#!/usr/bin/python
+
+# This script generates the boilerplate versions of most fops and cbks in the
+# server.  This allows the details of leadership-status checking, sequencing
+# between leader and followers (including fan-out), and basic error checking
+# to be centralized one place, with per-operation code kept to a minimum.
+
+import sys
+import codegen
+
+type_re = "([a-z_0-9]+)"
+name_re = "\(\*fop_([a-z0-9]+)_t\)"
+full_re = type_re + " *" + name_re
+fop_cg = codegen.CodeGenerator()
+fop_cg.skip = 2
+fop_cg.parse_decls(sys.argv[1],full_re)
+fop_cg.load_templates(sys.argv[2])
+
+# Use the multi-template feature to generate multiple callbacks from the same
+# parsed declarations.
+type_re = "([a-z_0-9]+)"
+name_re = "\(\*fop_([a-z0-9]+)_cbk_t\)"
+full_re = type_re + " *" + name_re
+cbk_cg = codegen.CodeGenerator()
+cbk_cg.skip = 5
+cbk_cg.parse_decls(sys.argv[1],full_re)
+cbk_cg.load_templates(sys.argv[2])
+
+# This is a nasty little trick to handle the case where a generated fop needs
+# a set of default arguments for the corresponding callback.
+fop_cg.make_defaults = cbk_cg.make_defaults
+
+# We need two types of templates.  The first, for pure read operations, just
+# needs to do a simple am-i-leader check (augmented to allow dirty reads).
+# The second, for pure writes, needs to do fan-out to followers between those
+# initial checks and local execution.  There are other operations that don't
+# fit neatly into either category - e.g. lock ops or fsync - so we'll just have
+# to handle those manually.  The table thus includes entries only for those we
+# can categorize.  The special cases, plus any new operations we've never even
+# heard of, aren't in there.
+#
+# Various keywords can be used to define/undefine preprocessor symbols used
+# in the templates, on a per-function basis.  For example, if the keyword here
+# is "fsync" (lowercase word or abbreviation) that will cause NSR_CG_FSYNC
+# (prefix plus uppercase version) to be defined above all of the generated code
+# for that fop.
+
+fop_table = {
+	"access":		"read",
+	"create":		"write",
+	"discard":		"write",
+#	"entrylk":		"read",
+	"fallocate":	"write",
+#	"fentrylk":		"read",
+	"fgetxattr":	"read",
+#	"finodelk":		"read",
+#	"flush":		"read",
+	"fremovexattr":	"write",
+	"fsetattr":		"write",
+	"fsetxattr":	"write",
+	"fstat":		"read",
+#	"fsync":		"read",
+#	"fsyncdir":		"read",
+	"ftruncate":	"write",
+	"fxattrop":		"write",
+	"getxattr":		"read",
+#	"inodelk":		"read",
+	"link":			"write",
+#	"lk":			"read",
+#	"lookup":		"read",
+	"mkdir":		"write",
+	"mknod":		"write",
+	"open":			"write",
+	"opendir":		"read",
+	"rchecksum":	"read",
+	"readdir":		"read",
+	"readdirp":		"read",
+	"readlink":		"read",
+	"readv":		"read",
+	"removexattr":	"write",
+	"rename":		"write",
+	"rmdir":		"write",
+	"setattr":		"write",
+	"setxattr":		"write",
+	"stat":			"read",
+	"statfs":		"read",
+	"symlink":		"write",
+	"truncate":		"write",
+	"unlink":		"write",
+	"writev":		"write,fsync,queue",
+	"xattrop":		"write",
+}
+
+fops_done = []
+for x in sorted(fop_cg.decls.keys()):
+	if x in fop_table.keys():
+		info = fop_table[x].split(",")
+		kind = info[0]
+		flags = info[1:]
+		if ("fsync" in flags) or ("queue" in flags):
+			flags.append("need_fd")
+		for fname in flags:
+			print "#define NSR_CG_%s" % fname.upper()
+		cbk_cg.emit(x,kind+"-complete")
+		fop_cg.emit(x,kind+"-continue")
+		cbk_cg.emit(x,kind+"-fan-in")
+		fop_cg.emit(x,kind+"-dispatch")
+		fop_cg.emit(x,kind+"-fop")
+		for fname in flags:
+			print "#undef NSR_CG_%s" % fname.upper()
+		fops_done.append(x)
+	else:
+		print("/* No code emitted for %s */"%x)
+		print("")
+
+# Just for fun, emit the fops table too.
+print("struct xlator_fops fops = {")
+for x in fops_done:
+	print("	.%s = nsr_%s,"%(x,x))
+print("};")
diff --git a/xlators/cluster/nsr-server/src/leader.c b/xlators/cluster/nsr-server/src/leader.c
new file mode 100644
index 000000000..02a2609c8
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/leader.c
@@ -0,0 +1,138 @@
+/*
+   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 <regex.h>
+//#include <stdlib.h>
+#include <string.h>
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "call-stub.h"
+#include "defaults.h"
+#include "xlator.h"
+#include "api/src/glfs.h"
+#include "api/src/glfs-internal.h"
+
+#ifndef NSR_SIM_ETCD
+#include "etcd-api.h"
+#endif
+#include "nsr-internal.h"
+#include "../../nsr-recon/src/recon_driver.h"
+#include "../../nsr-recon/src/recon_xlator.h"
+
+#define NSR_TTL         5
+
+static void
+nsr_set_leader (xlator_t *this, etcd_session etcd)
+{
+        long            term = 0;
+        etcd_result     res;
+        nsr_private_t   *priv   = this->private;
+        char             n_t[sizeof(long)+1];
+        char            *text   = NULL;
+
+        gf_log (this->name, GF_LOG_INFO, "Just became leader");
+
+        text = etcd_get(etcd, priv->term_key);
+        if(text == NULL) {
+                term = 0;
+        } else {
+                term = strtol(text, NULL, 10);
+        }
+        sprintf(n_t,"%ld",term+1);
+        res = etcd_set(etcd, priv->term_key,n_t,text,0);
+        if(res != ETCD_OK) {
+		gf_log (this->name, GF_LOG_ERROR, "failed to set term");
+		return;
+	}
+        priv->leader = _gf_true;
+
+        priv->current_term = term + 1;
+
+        if (priv->nsr_recon_start == _gf_false) {
+                atomic_fetch_and(&(priv->fence_io), 0);
+                return;
+        }
+
+        // Move this inside recon notify???
+        atomic_fetch_or(&(priv->fence_io), 1);
+
+        nsr_recon_notify_event_set_leader(priv);
+
+        return;
+}
+
+void *
+nsr_leader_thread (void *arg)
+{
+        xlator_t        *this           = (xlator_t *) arg;
+        nsr_private_t   *priv           = this->private;
+        etcd_result     res;
+        char            *index_in       = NULL;
+        char            *index_out      = NULL;
+
+        gf_log (this->name, GF_LOG_INFO,
+                "calling etcd_open_str on servers %s", priv->etcd_servers);
+
+        priv->etcd = etcd_open_str(priv->etcd_servers);
+        if (!(priv->etcd)) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to open etcd session\n");
+                return NULL;
+        }
+
+        priv->leader_inited = 1;
+
+        for (;;) {
+                /* Not leader yet.  Try to become leader. */
+                for (;;) {
+                        res = etcd_lock (priv->etcd, priv->leader_key, NSR_TTL,
+                                         index_in, &index_out);
+                        if (res == ETCD_OK) {
+                                break;
+                        }
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "etcd_lock failed (%d)", res);
+                        sleep(1);
+                }
+                /* We're there.  Notify other parts of the code. */
+                nsr_set_leader(this,priv->etcd);
+                /* Try to retain leadership. */
+                index_in = index_out;
+                index_out = NULL;
+                for (;;) {
+                        res = etcd_lock (priv->etcd, priv->leader_key, NSR_TTL,
+                                         index_in, &index_out);
+                        if (index_out && (index_in != index_out)) {
+                                if (index_in) {
+                                        free(index_in);
+                                }
+                                index_in = index_out;
+                                index_out = NULL;
+                        }
+                        if (res != ETCD_OK) {
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "lost leadership (%d)", res);
+                                if (index_out) {
+                                        free(index_out);
+                                }
+                                break;
+                        }
+                        sleep(1);
+                }
+        }
+
+        etcd_close_str(priv->etcd);
+        return NULL;
+}
+
diff --git a/xlators/cluster/nsr-server/src/nsr-internal.h b/xlators/cluster/nsr-server/src/nsr-internal.h
new file mode 100644
index 000000000..72b61bfa5
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/nsr-internal.h
@@ -0,0 +1,101 @@
+/*
+   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 <sys/stat.h>
+#include <sys/types.h>
+
+#define LEADER_XATTR            "user.nsr.leader"
+#define SECOND_CHILD(xl)        (xl->children->next->xlator)
+
+enum {
+        gf_mt_nsr_private_t = gf_common_mt_end + 1,
+        gf_mt_nsr_fd_ctx_t,
+        gf_mt_nsr_inode_ctx_t,
+        gf_mt_nsr_dirty_t,
+        gf_mt_nsr_end
+};
+
+typedef enum nsr_recon_notify_ev_id_t {
+        NSR_RECON_SET_LEADER = 1,
+        NSR_RECON_ADD_CHILD = 2
+} nsr_recon_notify_ev_id_t;
+
+typedef struct _nsr_recon_notify_ev_s {
+        nsr_recon_notify_ev_id_t id;
+        uint32_t index; // in case of add
+        struct list_head list;
+} nsr_recon_notify_ev_t;
+
+typedef struct {
+        char                    *etcd_servers;
+        char                    *subvol_uuid;
+        char                    *leader_key;
+        char                    *term_key;
+        char                    *brick_uuid;
+        gf_boolean_t            leader;
+        uint8_t                 up_children;
+        uint8_t                 n_children;
+        char                    *vol_file;
+        etcd_session            etcd;
+        volatile unsigned int   fence_io;
+        uint32_t                current_term;
+#ifdef  NSR_DEBUG
+        uint32_t                leader_log_fd;
+#endif
+        volatile int            recon_notify_inited;
+        volatile int            leader_inited;
+        uint32_t                kid_state;
+        gf_lock_t               dirty_lock;
+        struct list_head        dirty_fds;
+        gf_boolean_t            nsr_recon_start;
+        void *                  recon_ctx;
+        volatile uint32_t       ops_in_flight;
+	uint32_t                index;
+	gf_lock_t               index_lock;
+        double                  quorum_pct;
+} nsr_private_t;
+
+typedef struct {
+        call_stub_t             *stub;
+        call_stub_t             *qstub;
+        uint8_t                 call_count;
+        fd_t                    *fd;
+        struct list_head        qlinks;
+} nsr_local_t;
+
+/*
+ * This should match whatever changelog returns on the pre-op for us to pass
+ * when we're ready for our post-op.
+ */
+typedef uint32_t log_id_t;
+
+typedef struct {
+        struct list_head        links;
+        log_id_t                id;
+} nsr_dirty_list_t;
+
+typedef struct {
+        fd_t                    *fd;
+        struct list_head        dirty_list;
+        struct list_head        fd_list;
+} nsr_fd_ctx_t;
+
+typedef struct {
+        gf_lock_t               lock;
+        uint32_t                active;
+        struct list_head        aqueue;
+        uint32_t                pending;
+        struct list_head        pqueue;
+} nsr_inode_ctx_t;
+
+void nsr_recon_notify_event_set_leader(nsr_private_t *priv);
+void nsr_recon_notify_event_add_child(nsr_private_t *priv, uint32_t index);
+void* nsr_recon_notify_thread (void *this);
+
diff --git a/xlators/cluster/nsr-server/src/nsr.c b/xlators/cluster/nsr-server/src/nsr.c
new file mode 100644
index 000000000..85eba09b5
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/nsr.c
@@ -0,0 +1,812 @@
+/*
+   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.
+*/
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "call-stub.h"
+#include "defaults.h"
+#include "xlator.h"
+#include "api/src/glfs.h"
+#include "api/src/glfs-internal.h"
+#include "run.h"
+#include "common-utils.h"
+#include "syncop.h"
+
+#include "etcd-api.h"
+#include "nsr-internal.h"
+#include "../../nsr-recon/src/recon_driver.h"
+#include "../../nsr-recon/src/recon_xlator.h"
+
+
+#define GLUSTERD_DEFAULT_WORKDIR "/var/lib/glusterd"
+#define GLUSTERD_VOLUME_DIR_PREFIX "vols"
+#define GLUSTERD_BRICK_INFO_DIR "bricks"
+
+#define NSR_FLUSH_INTERVAL      5
+
+nsr_inode_ctx_t *
+nsr_get_inode_ctx (xlator_t *this, inode_t *inode)
+{
+        uint64_t                ctx_int         = 0LL;
+        nsr_inode_ctx_t         *ctx_ptr;
+
+        if (__inode_ctx_get(inode,this,&ctx_int) == 0) {
+                ctx_ptr = (nsr_inode_ctx_t *)(long)ctx_int;
+        }
+        else {
+                ctx_ptr = GF_CALLOC (1, sizeof(*ctx_ptr),
+                                     gf_mt_nsr_inode_ctx_t);
+                if (ctx_ptr) {
+                        ctx_int = (uint64_t)(long)ctx_ptr;
+                        if (__inode_ctx_set(inode,this,&ctx_int) == 0) {
+                                LOCK_INIT(&ctx_ptr->lock);
+                                INIT_LIST_HEAD(&ctx_ptr->aqueue);
+                                INIT_LIST_HEAD(&ctx_ptr->pqueue);
+                        }
+                        else {
+                                GF_FREE(ctx_ptr);
+                                ctx_ptr = NULL;
+                        }
+                }
+                                     
+        }
+
+        return ctx_ptr;
+}
+
+nsr_fd_ctx_t *
+nsr_get_fd_ctx (xlator_t *this, fd_t *fd)
+{
+        uint64_t                ctx_int         = 0LL;
+        nsr_fd_ctx_t            *ctx_ptr;
+
+        if (__fd_ctx_get(fd,this,&ctx_int) == 0) {
+                ctx_ptr = (nsr_fd_ctx_t *)(long)ctx_int;
+        }
+        else {
+                ctx_ptr = GF_CALLOC (1, sizeof(*ctx_ptr), gf_mt_nsr_fd_ctx_t);
+                if (ctx_ptr) {
+                        if (__fd_ctx_set(fd,this,(uint64_t)ctx_ptr) == 0) {
+                                INIT_LIST_HEAD(&ctx_ptr->dirty_list);
+                                INIT_LIST_HEAD(&ctx_ptr->fd_list);
+                        }
+                        else {
+                                GF_FREE(ctx_ptr);
+                                ctx_ptr = NULL;
+                        }
+                }
+                                     
+        }
+
+        return ctx_ptr;
+}
+
+void
+nsr_mark_fd_dirty (xlator_t *this, nsr_local_t *local)
+{
+        fd_t                    *fd             = local->fd;
+        nsr_fd_ctx_t            *ctx_ptr;
+        nsr_dirty_list_t        *dirty;
+        nsr_private_t           *priv           = this->private;
+
+        /*
+         * TBD: don't do any of this for O_SYNC/O_DIRECT writes.
+         * Unfortunately, that optimization requires that we distinguish
+         * between writev and other "write" calls, saving the original flags
+         * and checking them in the callback.  Too much work for too little
+         * gain right now.
+         */
+
+        LOCK(&fd->lock);
+                ctx_ptr = nsr_get_fd_ctx(this,fd);
+                dirty = GF_CALLOC(1,sizeof(*dirty),gf_mt_nsr_dirty_t);
+                if (ctx_ptr && dirty) {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "marking fd %p as dirty (%p)", fd, dirty);
+                        /* TBD: fill dirty->id from what changelog gave us */
+                        list_add_tail(&dirty->links,&ctx_ptr->dirty_list);
+                        if (list_empty(&ctx_ptr->fd_list)) {
+                                /* Add a ref so _release doesn't get called. */
+                                ctx_ptr->fd = fd_ref(fd);
+                                LOCK(&priv->dirty_lock);
+                                        list_add_tail (&ctx_ptr->fd_list,
+                                                       &priv->dirty_fds);
+                                UNLOCK(&priv->dirty_lock);
+                        }
+                }
+                else {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "could not mark %p dirty", fd);
+                        if (ctx_ptr) {
+                                GF_FREE(ctx_ptr);
+                        }
+                        if (dirty) {
+                                GF_FREE(dirty);
+                        }
+                }
+        UNLOCK(&fd->lock);
+}
+
+#define NSR_TERM_XATTR          "trusted.nsr.term"
+#define NSR_INDEX_XATTR         "trusted.nsr.index"
+#define RECON_TERM_XATTR        "trusted.nsr.recon-term"
+#define RECON_INDEX_XATTR       "trusted.nsr.recon-index"
+#define NSR_REP_COUNT_XATTR     "trusted.nsr.rep-count"
+#include "nsr-cg.c"
+
+uint8_t
+nsr_count_up_kids (nsr_private_t *priv)
+{
+        uint8_t         retval  = 0;
+        uint8_t         i;
+
+        for (i = 0; i < priv->n_children; ++i) {
+                if (priv->kid_state & (1 << i)) {
+                        ++retval;
+                }
+        }
+
+        return retval;
+}
+
+/*
+ * The fsync machinery looks a lot like that for any write call, but there are
+ * some important differences that are easy to miss.  First, we don't care
+ * about the xdata that shows whether the call came from a leader or
+ * reconciliation process.  If we're the leader we fan out; if we're not we
+ * don't.  Second, we don't wait for followers before we issue the local call.
+ * The code generation system could be updated to handle this, and still might
+ * if we need to implement other "almost identical" paths (e.g. for open), but
+ * a copy is more readable as long as it's just one.
+ */
+
+int32_t
+nsr_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)
+{
+        nsr_local_t     *local  = frame->local;
+        gf_boolean_t    unwind;
+
+        LOCK(&frame->lock);
+                unwind = !--(local->call_count);
+        UNLOCK(&frame->lock);
+
+        if (unwind) {
+                STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf,
+                                     postbuf, xdata);
+        }
+        return 0;
+}
+
+int32_t
+nsr_fsync_local_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)
+{
+        nsr_dirty_list_t        *dirty;
+        nsr_dirty_list_t        *dtmp;
+        nsr_local_t             *local  = frame->local;
+
+        list_for_each_entry_safe (dirty, dtmp, &local->qlinks, links) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "sending post-op on %p (%p)", local->fd, dirty);
+                GF_FREE(dirty);
+        }
+
+        return nsr_fsync_cbk (frame, cookie, this, op_ret, op_errno,
+                              prebuf, postbuf, xdata);
+}
+
+int32_t
+nsr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+           dict_t *xdata)
+{
+        nsr_private_t   *priv   = this->private;
+        nsr_local_t     *local;
+        uint64_t        ctx_int         = 0LL;
+        nsr_fd_ctx_t    *ctx_ptr;
+        xlator_list_t   *trav;
+
+        local = mem_get0(this->local_pool);
+        if (!local) {
+                STACK_UNWIND_STRICT(fsync,frame,-1,ENOMEM,NULL,NULL,xdata);
+                return 0;
+        }
+        INIT_LIST_HEAD(&local->qlinks);
+        frame->local = local;
+
+        /* Move the dirty list from the fd to the fsync request. */
+        LOCK(&fd->lock);
+                if (__fd_ctx_get(fd,this,&ctx_int) == 0) {
+                        ctx_ptr = (nsr_fd_ctx_t *)(long)ctx_int;
+                        list_splice_init (&ctx_ptr->dirty_list,
+                                          &local->qlinks);
+                }
+        UNLOCK(&fd->lock);
+
+        /* Issue the local call. */
+        local->call_count = priv->leader ? priv->n_children : 1;
+        STACK_WIND (frame, nsr_fsync_local_cbk,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsync,
+                    fd, flags, xdata);
+
+        /* Issue remote calls if we're the leader. */
+        if (priv->leader) {
+                for (trav = this->children->next; trav; trav = trav->next) {
+                        STACK_WIND (frame, nsr_fsync_cbk,
+                                    FIRST_CHILD(this),
+                                    FIRST_CHILD(this)->fops->fsync,
+                                    fd, flags, xdata);
+                }
+        }
+
+        return 0;
+}
+
+int32_t
+nsr_getxattr_special (call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      const char *name, dict_t *xdata)
+{
+        dict_t          *result;
+        nsr_private_t   *priv   = this->private;
+
+        if (!priv->leader) {
+                STACK_UNWIND_STRICT (getxattr, frame, -1, EREMOTE, NULL, NULL);
+                return 0;
+        }
+
+        if (!name || (strcmp(name,NSR_REP_COUNT_XATTR) != 0)) {
+                STACK_WIND_TAIL (frame,
+                                 FIRST_CHILD(this),
+                                 FIRST_CHILD(this)->fops->getxattr,
+                                 loc, name, xdata);
+                return 0;
+        }
+
+        result = dict_new();
+        if (!result) {
+                goto dn_failed;
+        }
+
+        priv->up_children = nsr_count_up_kids(this->private);
+        if (dict_set_uint32(result,NSR_REP_COUNT_XATTR,priv->up_children) != 0) {
+                goto dsu_failed;
+        }
+
+        STACK_UNWIND_STRICT (getxattr, frame, 0, 0, result, NULL);
+        dict_destroy(result);
+        return 0;
+
+dsu_failed:
+        dict_destroy(result);
+dn_failed:
+        STACK_UNWIND_STRICT (getxattr, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+}
+
+void
+nsr_flush_fd (xlator_t *this, nsr_fd_ctx_t *fd_ctx)
+{
+        nsr_dirty_list_t        *dirty;
+        nsr_dirty_list_t        *dtmp;
+
+        list_for_each_entry_safe (dirty, dtmp, &fd_ctx->dirty_list, links) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "sending post-op on %p (%p)", fd_ctx->fd, dirty);
+                GF_FREE(dirty);
+        }
+
+        INIT_LIST_HEAD(&fd_ctx->dirty_list);
+}
+
+void *
+nsr_flush_thread (void *ctx)
+{
+        xlator_t                *this   = ctx;
+        nsr_private_t           *priv   = this->private;
+        struct list_head        dirty_fds;
+        nsr_fd_ctx_t            *fd_ctx;
+        nsr_fd_ctx_t            *fd_tmp;
+        int                     ret;
+
+        for (;;) {
+                /*
+                 * We have to be very careful to avoid lock inversions here, so
+                 * we can't just hold priv->dirty_lock while we take and
+                 * release locks for each fd.  Instead, we only hold dirty_lock
+                 * at the beginning of each iteration, as we (effectively) make
+                 * a copy of the current list head and then clear the original.
+                 * This leads to four scenarios for adding the first entry to
+                 * an fd and potentially putting it on the global list.
+                 *
+                 * (1) While we're asleep.  No lock contention, it just gets
+                 *     added and will be processed on the next iteration.
+                 *
+                 * (2) After we've made a local copy, but before we've started
+                 *     processing that fd.  The new entry will be added to the
+                 *     fd (under its lock), and we'll process it on the current
+                 *     iteration.
+                 *
+                 * (3) While we're processing the fd.  They'll block on the fd
+                 *     lock, then see that the list is empty and put it on the
+                 *     global list.  We'll process it here on the next
+                 *     iteration.
+                 *
+                 * (4) While we're working, but after we've processed that fd.
+                 *     Same as (1) as far as that fd is concerned.
+                 */
+                INIT_LIST_HEAD(&dirty_fds);
+                LOCK(&priv->dirty_lock);
+                list_splice_init(&priv->dirty_fds,&dirty_fds);
+                UNLOCK(&priv->dirty_lock);
+
+                list_for_each_entry_safe (fd_ctx, fd_tmp, &dirty_fds, fd_list) {
+                        ret = syncop_fsync(FIRST_CHILD(this),fd_ctx->fd,0);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "failed to fsync %p (%d)",
+                                        fd_ctx->fd, -ret);
+                        }
+
+                        LOCK(&fd_ctx->fd->lock);
+                                nsr_flush_fd(this,fd_ctx);
+                                list_del_init(&fd_ctx->fd_list);
+                        UNLOCK(&fd_ctx->fd->lock);
+                        fd_unref(fd_ctx->fd);
+                }
+                                          
+                sleep(NSR_FLUSH_INTERVAL);
+        }
+
+        return NULL;
+}
+
+int32_t
+nsr_forget (xlator_t *this, inode_t *inode)
+{
+        uint64_t        ctx     = 0LL;
+
+        if ((inode_ctx_del(inode,this,&ctx) == 0) && ctx) {
+                GF_FREE((void *)(long)ctx);
+        }
+
+        return 0;
+}
+
+int32_t
+nsr_release (xlator_t *this, fd_t *fd)
+{
+        uint64_t        ctx     = 0LL;
+
+        if ((fd_ctx_del(fd,this,&ctx) == 0) && ctx) {
+                GF_FREE((void *)(long)ctx);
+        }
+
+        return 0;
+}
+
+struct xlator_cbks cbks = {
+        .forget  = nsr_forget,
+        .release = nsr_release,
+};
+
+int
+nsr_reconfigure (xlator_t *this, dict_t *options)
+{
+        nsr_private_t   *priv   = this->private;
+
+	GF_OPTION_RECONF ("leader", priv->leader, options, bool, err);
+        gf_log (this->name, GF_LOG_INFO,
+                "reconfigure called. setting priv->leader to %d\n", priv->leader);
+        return 0;
+
+err:
+        return -1;
+}
+
+int
+nsr_get_child_index (xlator_t *this, xlator_t *kid)
+{
+        xlator_list_t   *trav;
+        int             retval = -1;
+
+        for (trav = this->children; trav; trav = trav->next) {
+                ++retval;
+                if (trav->xlator == kid) {
+                        return retval;
+                }
+        }
+
+        return -1;
+}
+
+/*
+ * Child notify handling is unreasonably FUBAR.  Sometimes we'll get a
+ * CHILD_DOWN for a protocol/client child before we ever got a CHILD_UP for it.
+ * Other times we won't.  Because it's effectively random (probably racy), we
+ * can't just maintain a count.  We actually have to keep track of the state
+ * for each child separately, to filter out the bogus CHILD_DOWN events, and
+ * then generate counts on demand.
+ */
+int
+nsr_notify (xlator_t *this, int event, void *data, ...)
+{
+        nsr_private_t   *priv   = this->private;
+        int             index;
+
+        switch (event) {
+        case GF_EVENT_CHILD_UP:
+                index = nsr_get_child_index(this,data);
+                if (index >= 0) {
+                        priv->kid_state |= (1 << index);
+                        priv->up_children = nsr_count_up_kids(priv);
+                        gf_log (this->name, GF_LOG_INFO,
+                                "got CHILD_UP for %s, now %u kids",
+                                ((xlator_t *)data)->name,
+                                priv->up_children);
+                        if (priv->nsr_recon_start == _gf_true) {
+                                nsr_recon_notify_event_add_child(priv, index);
+                        }
+                }
+                break;
+        case GF_EVENT_CHILD_DOWN:
+                index = nsr_get_child_index(this,data);
+                if (index >= 0) {
+                        priv->kid_state &= ~(1 << index);
+                        priv->up_children = nsr_count_up_kids(priv);
+                        gf_log (this->name, GF_LOG_INFO,
+                                "got CHILD_DOWN for %s, now %u kids",
+                                ((xlator_t *)data)->name,
+                                priv->up_children);
+                }
+                break;
+        default:
+                ;
+        }
+
+        return default_notify(this,event,data);
+}
+
+
+int32_t
+mem_acct_init (xlator_t *this)
+{
+        int     ret = -1;
+
+        GF_VALIDATE_OR_GOTO ("nsr", this, out);
+
+        ret = xlator_mem_acct_init (this, gf_mt_nsr_end + 1);
+
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Memory accounting init" "failed");
+                return ret;
+        }
+out:
+        return ret;
+}
+
+
+extern void *nsr_leader_thread (void *);
+
+void
+nsr_deallocate_priv (nsr_private_t *priv)
+{
+        if (!priv) {
+                return;
+        }
+
+        if (priv->leader_key) {
+                GF_FREE(priv->leader_key);
+        }
+
+        if (priv->term_key) {
+                GF_FREE(priv->term_key);
+        }
+
+        GF_FREE(priv);
+}
+
+
+int32_t
+nsr_init (xlator_t *this)
+{
+        xlator_list_t   *remote;
+        xlator_list_t   *local;
+        nsr_private_t   *priv   = NULL;
+        xlator_list_t   *trav;
+        pthread_t       kid;
+        uuid_t          tmp_uuid;
+        char            *my_name = NULL, *morph_name = NULL, *recon_file = NULL, *recon_pid_file = NULL, *ptr = NULL;
+        char *volname;
+        extern xlator_t global_xlator;
+        glusterfs_ctx_t *oldctx = global_xlator.ctx;
+        runner_t runner  = {0,};
+        int32_t ret  = -1;
+        struct stat buf;
+        char *recon_log = NULL, *recon_log_dir = NULL;
+
+        /*
+         * Any fop that gets special treatment has to be patched in here,
+         * because the compiled-in table is produced by the code generator and
+         * only contains generated functions.  Note that we have to go through
+         * this->fops because of some dynamic-linking strangeness; modifying
+         * the static table doesn't work.
+         */
+        this->fops->getxattr = nsr_getxattr_special;
+        this->fops->fsync = nsr_fsync;
+
+        local = this->children;
+        if (!local) {
+                gf_log (this->name, GF_LOG_ERROR, "no local subvolume");
+                goto err;
+        }
+
+        remote = local->next;
+        if (!remote) {
+                gf_log (this->name, GF_LOG_ERROR, "no remote subvolumes");
+                goto err;
+        }
+
+        this->local_pool = mem_pool_new (nsr_local_t, 128);
+        if (!this->local_pool) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to create nsr_local_t pool");
+                goto err;
+        }
+
+        priv = GF_CALLOC (1, sizeof(*priv), gf_mt_nsr_private_t);
+        if (!priv) {
+                gf_log (this->name, GF_LOG_ERROR, "could not allocate priv");
+                goto err;
+        }
+
+        // set this so that unless leader election is done, IO is fenced
+        priv->fence_io = 1;
+
+        for (trav = this->children; trav; trav = trav->next) {
+                ++(priv->n_children);
+        }
+
+        LOCK_INIT(&priv->dirty_lock);
+	LOCK_INIT(&priv->index_lock);
+        INIT_LIST_HEAD(&priv->dirty_fds);
+
+        this->private = priv;
+
+        GF_OPTION_INIT ("etcd-servers", priv->etcd_servers, str, err);
+        if (!priv->etcd_servers) {
+                gf_log (this->name, GF_LOG_ERROR, "etcd servers not generated. ???");
+                goto err;
+        }
+
+
+        GF_OPTION_INIT ("quorum-percent", priv->quorum_pct, percent, err);
+
+        GF_OPTION_INIT ("subvol-uuid", priv->subvol_uuid, str, err);
+        gf_log (this->name, GF_LOG_INFO, "subvol_uuid = %s", priv->subvol_uuid);
+        if (gf_asprintf(&priv->leader_key,"%s:leader",priv->subvol_uuid) <= 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "could not generate leader key");
+                goto err;
+        }
+        if (gf_asprintf(&priv->term_key,"%s:term",priv->subvol_uuid) <= 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "could not generate term key");
+                goto err;
+        }
+        uuid_generate(tmp_uuid);
+        priv->brick_uuid = strdup(uuid_utoa(tmp_uuid));
+        gf_log (this->name, GF_LOG_INFO, "brick_uuid = %s\n", priv->brick_uuid);
+
+        GF_OPTION_INIT ("my-name", my_name, str, err);
+        if (!my_name) {
+                gf_log (this->name, GF_LOG_ERROR, "brick name not generated. ???");
+                goto err;
+        }
+        GF_OPTION_INIT ("vol-name", volname, str, err);
+        if (!volname) {
+                gf_log (this->name, GF_LOG_ERROR, "vol name not generated. ???");
+                goto err;
+        }
+        
+        morph_name = GF_CALLOC (1, strlen(my_name) + 1, gf_mt_nsr_private_t);
+        strcpy(morph_name, my_name);
+        recon_file = GF_CALLOC (1,PATH_MAX + strlen(morph_name) + strlen("con") +1, gf_mt_nsr_private_t);
+        recon_pid_file = GF_CALLOC (1,PATH_MAX + strlen(morph_name) + strlen("recon") +1, gf_mt_nsr_private_t);
+        if ((!recon_file) || (!recon_pid_file)) {
+                gf_log (this->name, GF_LOG_ERROR, "could not allocate reconciliation file name");
+                goto err;
+        }
+        ptr = strchr (morph_name, '/');
+        while (ptr) {
+                *ptr = '-';
+                ptr = strchr (morph_name, '/');
+        }
+
+        sprintf(recon_file,"/%s/%s/%s/%s/",GLUSTERD_DEFAULT_WORKDIR,
+                                     GLUSTERD_VOLUME_DIR_PREFIX,
+                                     volname,
+                                     GLUSTERD_BRICK_INFO_DIR);
+        strcat(recon_file, morph_name);
+        strcat(recon_file, "-nsr-recon.vol");
+
+        sprintf(recon_pid_file,"/%s/%s/%s/%s/",GLUSTERD_DEFAULT_WORKDIR,
+                                     GLUSTERD_VOLUME_DIR_PREFIX,
+                                     volname,
+                                     "run");
+        strcat(recon_pid_file, morph_name);
+        strcat(recon_pid_file, "-recon.pid");
+
+        priv->vol_file = GF_CALLOC (1,PATH_MAX + strlen(morph_name) + strlen("con") +1, gf_mt_nsr_private_t);
+        if (!priv->vol_file) {
+                gf_log (this->name, GF_LOG_ERROR, "could not allocate reconciliation file name");
+                goto err;
+        }
+        sprintf(priv->vol_file,"%s/%s/%s/%s/",
+                GLUSTERD_DEFAULT_WORKDIR,
+                GLUSTERD_VOLUME_DIR_PREFIX,
+                volname,
+                GLUSTERD_BRICK_INFO_DIR);
+        strcat(priv->vol_file, "con:");
+        strcat(priv->vol_file, morph_name);
+
+        if (pthread_create(&kid,NULL,nsr_flush_thread,this) != 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "could not start flush thread");
+                /* TBD: treat this as a fatal error? */
+        }
+
+        // Start the recon process. Then start the leader thread.
+        /*
+         * REVIEW
+         * Logs belong in /var/log not /tmp.
+         */
+
+        ret = mkdir (NSR_LOG_DIR, 0777);
+        if (ret != 0) {
+                if (errno != EEXIST) {
+                        gf_log (this->name, GF_LOG_ERROR, "Couldn't create"
+                                " nsr log directory (%s)", strerror (errno));
+                        goto err;
+                }
+        }
+
+        recon_log_dir = GF_CALLOC (1, strlen (NSR_LOG_DIR) + strlen(morph_name)
+                                   + 2, gf_mt_nsr_private_t);
+        if (!recon_log_dir) {
+                gf_log (this->name, GF_LOG_ERROR, "Couldn't allocate recon log "
+                        "dir name");
+                goto err;
+        }
+        sprintf (recon_log_dir, "%s/%s", NSR_LOG_DIR, morph_name);
+        ret = mkdir (recon_log_dir, 0777);
+
+        if (ret != 0){
+                if (errno != EEXIST) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Couldn't create brick log dir (%s)",
+                                strerror (errno));
+                        goto err;
+                }
+        }
+
+        recon_log = GF_CALLOC (1, strlen (recon_log_dir)+
+                               strlen ("reconciliation.log") + 2,
+                               gf_mt_nsr_private_t);
+        if (!recon_log) {
+                gf_log (this->name, GF_LOG_ERROR, "Couldn't allocate recon log"
+                        " file name");
+                goto err;
+        }
+        sprintf (recon_log, "%s/reconciliation.log", recon_log_dir);
+
+        if (!stat(priv->vol_file, &buf)) {
+
+                runinit (&runner);
+                runner_add_args(&runner, SBIN_DIR "/glusterfs",
+                        "-f", recon_file,
+                        "-p", recon_pid_file,
+                        "-l", recon_log,
+                        NULL);
+                ret = runner_run (&runner);
+                if (ret != 0)  {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "could not exec reconciliation process %s",
+                                SBIN_DIR "/glusterfs");
+                        goto err;
+                }
+
+                // TBD - convert this to make sure recon process runs
+                sleep(2);
+                priv->nsr_recon_start = _gf_true;
+        }
+
+
+        (void)pthread_create(&kid,NULL,nsr_recon_notify_thread,this);
+        while (priv->recon_notify_inited == 0) {
+                  sleep(1);
+        }
+
+        if (pthread_create(&kid,NULL,nsr_leader_thread,this) != 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to start leader thread");
+        }
+        while (priv->leader_inited == 0) {
+                  sleep(1);
+        }
+
+
+        /*
+         * Calling glfs_new changes old->ctx, even if THIS still points
+         * to global_xlator.  That causes problems later in the main
+         * thread, when gf_log_dump_graph tries to use the FILE after
+         * we've mucked with it and gets a segfault in __fprintf_chk.
+         * We can avoid all that by undoing the damage before we
+         * continue.
+         */
+        global_xlator.ctx = oldctx;
+
+	return 0;
+
+err:
+        nsr_deallocate_priv(priv);
+        return -1;
+}
+
+
+void
+nsr_fini (xlator_t *this)
+{
+        nsr_deallocate_priv(this->private);
+}
+
+class_methods_t class_methods = {
+        .init           = nsr_init,
+        .fini           = nsr_fini,
+        .reconfigure    = nsr_reconfigure,
+        .notify         = nsr_notify,
+};
+
+struct volume_options options[] = {
+        { .key = {"leader"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "false",
+          .description = "Start in the leader role.  This is only for "
+                         "bootstrapping the code, and should go away when we "
+                         "have real leader election."
+        },
+        { .key ={"vol-name"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "volume name"
+        },
+        { .key = {"my-name"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "brick name in form of host:/path"
+        },
+        { .key = {"etcd-servers"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "list of comma seperated etc servers"
+        },
+        { .key = {"subvol-uuid"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "UUID for this NSR (sub)volume"
+        },
+        { .key = {"quorum-percent"},
+          .type = GF_OPTION_TYPE_PERCENT,
+          .default_value = "50.0",
+          .description = "percentage of rep_count-1 that must be up"
+        },
+	{ .key = {NULL} },
+};
diff --git a/xlators/cluster/nsr-server/src/recon_notify.c b/xlators/cluster/nsr-server/src/recon_notify.c
new file mode 100644
index 000000000..1c50de234
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/recon_notify.c
@@ -0,0 +1,389 @@
+/*
+   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 <string.h>
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "call-stub.h"
+#include "defaults.h"
+#include "xlator.h"
+#include "api/src/glfs.h"
+#include "api/src/glfs-internal.h"
+#include "etcd-api.h"
+#include "nsr-internal.h"
+#include "../../nsr-recon/src/recon_driver.h"
+#include "../../nsr-recon/src/recon_xlator.h"
+
+
+
+typedef struct _nsr_recon_notify_ctx_t {
+        nsr_recon_notify_ev_t   recon_head;
+        pthread_mutex_t         recon_mutex;
+        pthread_cond_t          recon_cv;
+        char                    **hosts; // list of hosts ordered depending on child indices
+        uint32_t                current_term;
+        uint32_t                last_reconciled_term;
+        glfs_t                  *fs;
+        glfs_fd_t               *fd;
+} nsr_recon_notify_ctx_t;
+
+static int
+xlator_get_option (xlator_t *xl, char *key, char **value)
+{
+        GF_ASSERT (xl);
+        return dict_get_str (xl->options, key, value);
+}
+
+void nsr_recon_notify_event_set_leader(nsr_private_t *priv)
+{
+        nsr_recon_notify_ev_t *ev;
+        nsr_recon_notify_ctx_t *ctx = (nsr_recon_notify_ctx_t *)priv->recon_ctx;
+
+        ev = GF_CALLOC (1, sizeof (nsr_recon_notify_ev_t), 0);
+        ev->id = NSR_RECON_SET_LEADER;
+        INIT_LIST_HEAD(&(ev->list));
+        pthread_mutex_lock(&ctx->recon_mutex);
+        list_add_tail(&ev->list, &ctx->recon_head.list);
+        pthread_cond_signal(&ctx->recon_cv);
+        pthread_mutex_unlock(&ctx->recon_mutex);
+}
+
+void nsr_recon_notify_event_add_child(nsr_private_t *priv, uint32_t index) 
+{
+        nsr_recon_notify_ev_t *ev;
+        nsr_recon_notify_ctx_t *ctx = (nsr_recon_notify_ctx_t *)priv->recon_ctx;
+
+        ev = GF_CALLOC (1, sizeof (nsr_recon_notify_ev_t), 0);
+        ev->id = NSR_RECON_ADD_CHILD;
+        ev->index = index;
+        INIT_LIST_HEAD(&(ev->list));
+        pthread_mutex_lock(&ctx->recon_mutex);
+        list_add_tail(&ev->list, &ctx->recon_head.list);
+        pthread_cond_signal(&ctx->recon_cv);
+        pthread_mutex_unlock(&ctx->recon_mutex);
+}
+
+
+static void
+nsr_recon_set_leader (xlator_t *this)
+{
+
+        nsr_private_t   *priv   = this->private;
+        nsr_recon_notify_ctx_t *ctx = (nsr_recon_notify_ctx_t *)priv->recon_ctx;
+        nsr_recon_role_t role;
+        xlator_t        *old            = this;
+        uint32_t i=0;
+
+        if (priv->leader != _gf_true)
+                return;
+
+        if (ctx->last_reconciled_term == priv->current_term)
+                return;
+
+        /*
+         * Quorum for reconciliation is not the same as quorum for I/O.  Here,
+         * we require a true majority.  The +1 is because we don't count
+         * ourselves as part of n_children or up_children.
+         *
+         * TBD: re-evaluate when to reconcile (including partial)
+         */
+        if (priv->up_children <= (priv->n_children / 2))
+                return;
+
+	gf_log (this->name, GF_LOG_INFO,
+                "Sending message to do recon with %d nodes\n",
+                priv->up_children);
+
+	role.num = 0;
+        role.role = leader;
+        for (i = 0; i < priv->n_children; ++i) {
+                if (priv->kid_state & (1 << i)) {
+		        gf_log (this->name, GF_LOG_INFO,
+                                "Recon using host %s",
+                                 ctx->hosts[i]);
+                        strcpy(role.info[role.num].name, ctx->hosts[i]);
+                        (role.num)++;
+                }
+        }
+
+	gf_log (this->name, GF_LOG_INFO,
+                "setting current term as %d", priv->current_term);
+        role.current_term = priv->current_term;
+        ENDIAN_CONVERSION_RR(role, _gf_false);
+
+        // inform the reconciliator that this is leader
+        // in the callback (once reconciliation is done),
+        // we will unfence the IOs.
+        // TBD - error handling later.
+        if (glfs_lseek(ctx->fd, nsr_recon_xlator_sector_1, SEEK_SET) == -1) {
+		gf_log (this->name, GF_LOG_ERROR,
+                        "doing lseek failed\n");
+		return;
+	}
+
+        glusterfs_this_set(old);
+        gf_log (this->name, GF_LOG_INFO,
+                "Writing to local node to set leader");
+	do {
+		 if (priv->leader != _gf_true) {
+			glusterfs_this_set(old);
+			gf_log (this->name, GF_LOG_ERROR, "no longer leader\n");
+			return;
+		 }
+		 if (glfs_write(ctx->fd, &role, sizeof(role), 0) == -1) {
+			if (errno == EAGAIN) {
+				// Wait for old reconciliation to bail out.
+				glusterfs_this_set(old);
+				gf_log (this->name, GF_LOG_ERROR,
+                        		"write failed with retry. retrying after some time\n");
+				sleep(5);
+				continue;
+			}
+			else{
+				glusterfs_this_set(old);
+				gf_log (this->name, GF_LOG_ERROR,
+                        		"doing write failed\n");
+				// This is because reconciliation has returned with error
+				// because some node has died in between.
+				// What should be done? Either we retry being leader
+				// or hook to CHILD_DOWN notification.
+				// Put that logic later. As of now we will just retry.
+				// This is easier.
+				sleep(5);
+				continue;
+			}
+		} else {
+			glusterfs_this_set(old);
+			gf_log (this->name, GF_LOG_INFO, "doing write with success\n");
+			break;
+		}
+	} while(1);
+        glusterfs_this_set(old);
+	gf_log (this->name, GF_LOG_INFO,
+                "glfs_write returned. unfencing IO\n");
+
+        // TBD - error handling
+
+        ctx->last_reconciled_term = priv->current_term;
+	priv->index = 0; // reset changelog index
+        atomic_fetch_and(&(priv->fence_io), 0);
+
+        return;
+}
+
+static void
+nsr_recon_add_child (xlator_t *this, uint32_t index)
+{
+        nsr_private_t   *priv   = this->private;
+        nsr_recon_notify_ctx_t *ctx = (nsr_recon_notify_ctx_t *)priv->recon_ctx;
+        nsr_recon_role_t role;
+        xlator_t        *old            = this;
+
+        if (priv->leader != _gf_true)
+                return;
+
+        // reconciliation still pending.
+        // Check if we have majority
+        if (ctx->last_reconciled_term != priv->current_term) {
+                nsr_recon_set_leader(this);
+        } else {
+                // Reconciliation done.
+                // new child joining the majority/
+                // Do reconciliation only fot this child but after fencing new IO and draining old IO
+                role.num = 1;
+                role.role = joiner;
+
+                atomic_fetch_or(&(priv->fence_io), 1);
+                while(priv->ops_in_flight) {
+                        sleep(1);
+                }
+                        
+                strcpy(role.info[0].name, ctx->hosts[index]);
+                role.current_term = priv->current_term;
+                ENDIAN_CONVERSION_RR(role, _gf_false);
+                glfs_lseek(ctx->fd, nsr_recon_xlator_sector_1, SEEK_SET);
+                glusterfs_this_set(old);
+                gf_log (this->name, GF_LOG_INFO,
+                        "Writing to local node to join %s\n", role.info[0].name);
+                glfs_write(ctx->fd, &role,
+                           sizeof(role), 0);
+                glusterfs_this_set(old);
+                gf_log (this->name, GF_LOG_INFO,
+                        "Write to local node to set joiner returned\n");
+
+                // TBD - error handling
+                atomic_fetch_and(&(priv->fence_io), 0);
+        }
+
+        return;
+}
+
+static uint32_t
+nsr_setup_recon (xlator_t *this)
+{
+        nsr_private_t   *priv           = this->private;
+        xlator_t        *old            = this;
+        uint32_t         ret             = 0;
+        nsr_recon_notify_ctx_t *ctx = (nsr_recon_notify_ctx_t *)priv->recon_ctx;
+
+        if (priv->nsr_recon_start == _gf_false)
+                return 0;
+
+        ctx->fs = glfs_new(priv->subvol_uuid);
+        if (!ctx->fs) {
+                ret = 1;
+                gf_log (this->name, GF_LOG_ERROR, "failed to initialise glfs \n");
+                goto done;
+        }
+
+        glusterfs_this_set(old);
+        ret = glfs_set_volfile(ctx->fs, priv->vol_file);
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_ERROR, "failed to set volfile \n");
+                goto done;
+        }
+
+        glusterfs_this_set(old);
+        /*
+         * REVIEW
+         * Logs belong in /var/log not /tmp.
+         */
+        glfs_set_logging (ctx->fs,"/tmp/glfs-log", 7);
+        if (glfs_init(ctx->fs) < 0) {
+                gf_log (this->name, GF_LOG_ERROR, "failed to init volfile \n");
+                ret = 1;
+                goto done;
+        }
+
+        glusterfs_this_set(old);
+        ctx->fd = glfs_open (ctx->fs, "/", O_RDWR);
+        if (ctx->fd == NULL) {
+                ret = 1;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to open fd to communicate with recon process \n");
+                goto done;
+        }
+
+
+done:
+        glusterfs_this_set(old);
+        return ret;
+}
+
+
+static void
+nsr_setup_hosts(xlator_t *this)
+{
+        xlator_list_t   *trav;
+        nsr_private_t   *priv   = this->private;
+        uint32_t        i       = 0;
+        nsr_recon_notify_ctx_t *ctx = (nsr_recon_notify_ctx_t *)priv->recon_ctx;
+
+        ctx->hosts = GF_CALLOC(sizeof(char *), priv->n_children, gf_mt_nsr_private_t);
+        // Iterate thru all the children
+        for (trav = this->children; trav; trav = trav->next) {
+                char            *hostname = NULL, *vol = NULL;
+                int             ret1 = 0, ret2 = 0, ret = 0;
+                xlator_t *xl = trav->xlator;
+                // If the child type is that of protocol/client
+                if (!strcmp(trav->xlator->type, "protocol/client")) {
+                        ret1 = xlator_get_option (xl, "remote-host", &hostname);
+                        ret2 = xlator_get_option (xl, "remote-subvolume", &vol);
+                        if (!ret1 && !ret2) {
+                                // add the name of that host to the hosts
+                                ctx->hosts[i] = GF_CALLOC(sizeof(char), strlen(hostname) + strlen(vol) + 2, 0);
+                                strcpy(ctx->hosts[i], hostname);
+                                strcat(ctx->hosts[i], ":");
+                                strcat(ctx->hosts[i], vol);
+                                gf_log (this->name, GF_LOG_INFO,
+                                        "adding hosts %s to recon notfiy list", ctx->hosts[i]);
+                        } else {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "CANNOT FIND HOSTNAME FOR A CHILD");
+                                GF_ASSERT(0);
+                        }
+                //  local brick
+                } else {
+                        ret = xlator_get_option (this, "my-name", &hostname);
+                        if (!ret) {
+                                uint32_t len = strlen(hostname);
+                                ctx->hosts[i] = GF_CALLOC(sizeof(char),
+                                                           len+1,
+                                                           gf_mt_nsr_private_t);
+                                strcpy(ctx->hosts[i], hostname);
+                                gf_log (this->name, GF_LOG_INFO,
+                                        "adding  my host %s to recon notfiy list", ctx->hosts[i]);
+                        } else {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "CANNOT FIND MY HOSTNAME");
+                                GF_ASSERT(0);
+                        }
+                }
+                i++;
+        }
+}
+
+void *
+nsr_recon_notify_thread (void *arg)
+{
+        xlator_t *this = (xlator_t *)arg;
+        nsr_private_t   *priv   = this->private;
+        nsr_recon_notify_ev_t *ev;
+        nsr_recon_notify_ctx_t *ctx;
+
+        priv->recon_ctx = GF_CALLOC(1, sizeof(nsr_recon_notify_ctx_t), gf_mt_nsr_private_t);
+        if (!priv->recon_ctx) {
+                gf_log (this->name, GF_LOG_ERROR, "calloc error");
+                return NULL;
+        }
+        ctx = priv->recon_ctx;
+
+        pthread_mutex_init(&(ctx->recon_mutex), NULL);
+        pthread_cond_init(&(ctx->recon_cv), NULL);
+        INIT_LIST_HEAD(&(ctx->recon_head.list));
+
+        nsr_setup_hosts(this);
+
+        if (nsr_setup_recon(this)) {
+                gf_log (this->name, GF_LOG_ERROR, "recon notify thread : initing glfs error");
+                return NULL;
+        }
+
+        priv->recon_notify_inited = 1;
+
+        while(1) {
+                pthread_mutex_lock(&ctx->recon_mutex);
+                while (list_empty(&(ctx->recon_head.list))) {
+                        pthread_cond_wait(&ctx->recon_cv, &ctx->recon_mutex);
+                }
+                pthread_mutex_unlock(&ctx->recon_mutex);
+
+                list_for_each_entry(ev, &(ctx->recon_head.list), list) {
+
+                        if (ev->id == NSR_RECON_SET_LEADER) {
+                                gf_log (this->name, GF_LOG_INFO,
+                                        "got add leader notfiy event");
+                                nsr_recon_set_leader(this);
+                        } else if (ev->id == NSR_RECON_ADD_CHILD) {
+                                gf_log (this->name, GF_LOG_INFO,
+                                        "got add child notify event");
+                                nsr_recon_add_child(this, ev->index);
+                        }
+                }
+                list_del_init (&ev->list);
+        }
+
+        return NULL;
+}
+
diff --git a/xlators/cluster/nsr-server/src/yajl.c b/xlators/cluster/nsr-server/src/yajl.c
new file mode 100644
index 000000000..54e6474fc
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl.c
@@ -0,0 +1,175 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "yajl/yajl_parse.h"
+#include "yajl_lex.h"
+#include "yajl_parser.h"
+#include "yajl_alloc.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <stdarg.h>
+#include <assert.h>
+
+const char *
+yajl_status_to_string(yajl_status stat)
+{
+    const char * statStr = "unknown";
+    switch (stat) {
+        case yajl_status_ok:
+            statStr = "ok, no error";
+            break;
+        case yajl_status_client_canceled:
+            statStr = "client canceled parse";
+            break;
+        case yajl_status_error:
+            statStr = "parse error";
+            break;
+    }
+    return statStr;
+}
+
+yajl_handle
+yajl_alloc(const yajl_callbacks * callbacks,
+           yajl_alloc_funcs * afs,
+           void * ctx)
+{
+    yajl_handle hand = NULL;
+    yajl_alloc_funcs afsBuffer;
+
+    /* first order of business is to set up memory allocation routines */
+    if (afs != NULL) {
+        if (afs->malloc == NULL || afs->realloc == NULL || afs->free == NULL)
+        {
+            return NULL;
+        }
+    } else {
+        yajl_set_default_alloc_funcs(&afsBuffer);
+        afs = &afsBuffer;
+    }
+
+    hand = (yajl_handle) YA_MALLOC(afs, sizeof(struct yajl_handle_t));
+
+    /* copy in pointers to allocation routines */
+    memcpy((void *) &(hand->alloc), (void *) afs, sizeof(yajl_alloc_funcs));
+
+    hand->callbacks = callbacks;
+    hand->ctx = ctx;
+    hand->lexer = NULL; 
+    hand->bytesConsumed = 0;
+    hand->decodeBuf = yajl_buf_alloc(&(hand->alloc));
+    hand->flags	    = 0;
+    yajl_bs_init(hand->stateStack, &(hand->alloc));
+    yajl_bs_push(hand->stateStack, yajl_state_start);
+
+    return hand;
+}
+
+int
+yajl_config(yajl_handle h, yajl_option opt, ...)
+{
+    int rv = 1;
+    va_list ap;
+    va_start(ap, opt);
+
+    switch(opt) {
+        case yajl_allow_comments:
+        case yajl_dont_validate_strings:
+        case yajl_allow_trailing_garbage:
+        case yajl_allow_multiple_values:
+        case yajl_allow_partial_values:
+            if (va_arg(ap, int)) h->flags |= opt;
+            else h->flags &= ~opt;
+            break;
+        default:
+            rv = 0;
+    }
+    va_end(ap);
+
+    return rv;
+}
+
+void
+yajl_free(yajl_handle handle)
+{
+    yajl_bs_free(handle->stateStack);
+    yajl_buf_free(handle->decodeBuf);
+    if (handle->lexer) {
+        yajl_lex_free(handle->lexer);
+        handle->lexer = NULL;
+    }
+    YA_FREE(&(handle->alloc), handle);
+}
+
+yajl_status
+yajl_parse(yajl_handle hand, const unsigned char * jsonText,
+           size_t jsonTextLen)
+{
+    yajl_status status;
+
+    /* lazy allocation of the lexer */
+    if (hand->lexer == NULL) {
+        hand->lexer = yajl_lex_alloc(&(hand->alloc),
+                                     hand->flags & yajl_allow_comments,
+                                     !(hand->flags & yajl_dont_validate_strings));
+    }
+
+    status = yajl_do_parse(hand, jsonText, jsonTextLen);
+    return status;
+}
+
+
+yajl_status
+yajl_complete_parse(yajl_handle hand)
+{
+    /* The lexer is lazy allocated in the first call to parse.  if parse is
+     * never called, then no data was provided to parse at all.  This is a
+     * "premature EOF" error unless yajl_allow_partial_values is specified.
+     * allocating the lexer now is the simplest possible way to handle this
+     * case while preserving all the other semantics of the parser
+     * (multiple values, partial values, etc). */
+    if (hand->lexer == NULL) {
+        hand->lexer = yajl_lex_alloc(&(hand->alloc),
+                                     hand->flags & yajl_allow_comments,
+                                     !(hand->flags & yajl_dont_validate_strings));
+    }
+
+    return yajl_do_finish(hand);
+}
+
+unsigned char *
+yajl_get_error(yajl_handle hand, int verbose,
+               const unsigned char * jsonText, size_t jsonTextLen)
+{
+    return yajl_render_error_string(hand, jsonText, jsonTextLen, verbose);
+}
+
+size_t
+yajl_get_bytes_consumed(yajl_handle hand)
+{
+    if (!hand) return 0;
+    else return hand->bytesConsumed;
+}
+
+
+void
+yajl_free_error(yajl_handle hand, unsigned char * str)
+{
+    /* use memory allocation functions if set */
+    YA_FREE(&(hand->alloc), str);
+}
+
+/* XXX: add utility routines to parse from file */
diff --git a/xlators/cluster/nsr-server/src/yajl/yajl_common.h b/xlators/cluster/nsr-server/src/yajl/yajl_common.h
new file mode 100644
index 000000000..49ca3a5cb
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl/yajl_common.h
@@ -0,0 +1,75 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#ifndef __YAJL_COMMON_H__
+#define __YAJL_COMMON_H__
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif    
+
+#define YAJL_MAX_DEPTH 128
+
+/* msft dll export gunk.  To build a DLL on windows, you
+ * must define WIN32, YAJL_SHARED, and YAJL_BUILD.  To use a shared
+ * DLL, you must define YAJL_SHARED and WIN32 */
+#if defined(WIN32) && defined(YAJL_SHARED)
+#  ifdef YAJL_BUILD
+#    define YAJL_API __declspec(dllexport)
+#  else
+#    define YAJL_API __declspec(dllimport)
+#  endif
+#else
+#  if defined(__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__) >= 303
+#    define YAJL_API __attribute__ ((visibility("default")))
+#  else
+#    define YAJL_API
+#  endif
+#endif 
+
+/** pointer to a malloc function, supporting client overriding memory
+ *  allocation routines */
+typedef void * (*yajl_malloc_func)(void *ctx, size_t sz);
+
+/** pointer to a free function, supporting client overriding memory
+ *  allocation routines */
+typedef void (*yajl_free_func)(void *ctx, void * ptr);
+
+/** pointer to a realloc function which can resize an allocation. */
+typedef void * (*yajl_realloc_func)(void *ctx, void * ptr, size_t sz);
+
+/** A structure which can be passed to yajl_*_alloc routines to allow the
+ *  client to specify memory allocation functions to be used. */
+typedef struct
+{
+    /** pointer to a function that can allocate uninitialized memory */
+    yajl_malloc_func malloc;
+    /** pointer to a function that can resize memory allocations */
+    yajl_realloc_func realloc;
+    /** pointer to a function that can free memory allocated using
+     *  reallocFunction or mallocFunction */
+    yajl_free_func free;
+    /** a context pointer that will be passed to above allocation routines */
+    void * ctx;
+} yajl_alloc_funcs;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/xlators/cluster/nsr-server/src/yajl/yajl_gen.h b/xlators/cluster/nsr-server/src/yajl/yajl_gen.h
new file mode 100644
index 000000000..52fa99fc2
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl/yajl_gen.h
@@ -0,0 +1,157 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+/**
+ * \file yajl_gen.h
+ * Interface to YAJL's JSON generation facilities.
+ */
+
+#include <yajl/yajl_common.h>
+
+#ifndef __YAJL_GEN_H__
+#define __YAJL_GEN_H__
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+    /** generator status codes */
+    typedef enum {
+        /** no error */
+        yajl_gen_status_ok = 0,
+        /** at a point where a map key is generated, a function other than
+         *  yajl_gen_string was called */
+        yajl_gen_keys_must_be_strings,
+        /** YAJL's maximum generation depth was exceeded.  see
+         *  YAJL_MAX_DEPTH */
+        yajl_max_depth_exceeded,
+        /** A generator function (yajl_gen_XXX) was called while in an error
+         *  state */
+        yajl_gen_in_error_state,
+        /** A complete JSON document has been generated */
+        yajl_gen_generation_complete,                
+        /** yajl_gen_double was passed an invalid floating point value
+         *  (infinity or NaN). */
+        yajl_gen_invalid_number,
+        /** A print callback was passed in, so there is no internal
+         * buffer to get from */
+        yajl_gen_no_buf,
+        /** returned from yajl_gen_string() when the yajl_gen_validate_utf8
+         *  option is enabled and an invalid was passed by client code.
+         */
+        yajl_gen_invalid_string
+    } yajl_gen_status;
+
+    /** an opaque handle to a generator */
+    typedef struct yajl_gen_t * yajl_gen;
+
+    /** a callback used for "printing" the results. */
+    typedef void (*yajl_print_t)(void * ctx,
+                                 const char * str,
+                                 size_t len);
+
+    /** configuration parameters for the parser, these may be passed to
+     *  yajl_gen_config() along with option specific argument(s).  In general,
+     *  all configuration parameters default to *off*. */
+    typedef enum {
+        /** generate indented (beautiful) output */
+        yajl_gen_beautify = 0x01,
+        /**
+         * Set an indent string which is used when yajl_gen_beautify
+         * is enabled.  Maybe something like \\t or some number of
+         * spaces.  The default is four spaces ' '.
+         */
+        yajl_gen_indent_string = 0x02,
+        /**
+         * Set a function and context argument that should be used to
+         * output generated json.  the function should conform to the
+         * yajl_print_t prototype while the context argument is a
+         * void * of your choosing.
+         *
+         * example:
+         *   yajl_gen_config(g, yajl_gen_print_callback, myFunc, myVoidPtr);
+         */
+        yajl_gen_print_callback = 0x04,
+        /**
+         * Normally the generator does not validate that strings you
+         * pass to it via yajl_gen_string() are valid UTF8.  Enabling
+         * this option will cause it to do so.
+         */
+        yajl_gen_validate_utf8 = 0x08,
+        /**
+         * the forward solidus (slash or '/' in human) is not required to be
+         * escaped in json text.  By default, YAJL will not escape it in the
+         * iterest of saving bytes.  Setting this flag will cause YAJL to
+         * always escape '/' in generated JSON strings.
+         */
+        yajl_gen_escape_solidus = 0x10
+    } yajl_gen_option;
+
+    /** allow the modification of generator options subsequent to handle
+     *  allocation (via yajl_alloc)
+     *  \returns zero in case of errors, non-zero otherwise
+     */
+    YAJL_API int yajl_gen_config(yajl_gen g, yajl_gen_option opt, ...);
+
+    /** allocate a generator handle
+     *  \param allocFuncs an optional pointer to a structure which allows
+     *                    the client to overide the memory allocation
+     *                    used by yajl.  May be NULL, in which case
+     *                    malloc/free/realloc will be used.
+     *
+     *  \returns an allocated handle on success, NULL on failure (bad params)
+     */
+    YAJL_API yajl_gen yajl_gen_alloc(const yajl_alloc_funcs * allocFuncs);
+
+    /** free a generator handle */
+    YAJL_API void yajl_gen_free(yajl_gen handle);
+
+    YAJL_API yajl_gen_status yajl_gen_integer(yajl_gen hand, long long int number);
+    /** generate a floating point number.  number may not be infinity or
+     *  NaN, as these have no representation in JSON.  In these cases the
+     *  generator will return 'yajl_gen_invalid_number' */
+    YAJL_API yajl_gen_status yajl_gen_double(yajl_gen hand, double number);
+    YAJL_API yajl_gen_status yajl_gen_number(yajl_gen hand,
+                                             const char * num,
+                                             size_t len);
+    YAJL_API yajl_gen_status yajl_gen_string(yajl_gen hand,
+                                             const unsigned char * str,
+                                             size_t len);
+    YAJL_API yajl_gen_status yajl_gen_null(yajl_gen hand);
+    YAJL_API yajl_gen_status yajl_gen_bool(yajl_gen hand, int boolean);
+    YAJL_API yajl_gen_status yajl_gen_map_open(yajl_gen hand);
+    YAJL_API yajl_gen_status yajl_gen_map_close(yajl_gen hand);
+    YAJL_API yajl_gen_status yajl_gen_array_open(yajl_gen hand);
+    YAJL_API yajl_gen_status yajl_gen_array_close(yajl_gen hand);
+
+    /** access the null terminated generator buffer.  If incrementally
+     *  outputing JSON, one should call yajl_gen_clear to clear the
+     *  buffer.  This allows stream generation. */
+    YAJL_API yajl_gen_status yajl_gen_get_buf(yajl_gen hand,
+                                              const unsigned char ** buf,
+                                              size_t * len);
+
+    /** clear yajl's output buffer, but maintain all internal generation
+     *  state.  This function will not "reset" the generator state, and is
+     *  intended to enable incremental JSON outputing. */
+    YAJL_API void yajl_gen_clear(yajl_gen hand);
+
+#ifdef __cplusplus
+}
+#endif    
+
+#endif
diff --git a/xlators/cluster/nsr-server/src/yajl/yajl_parse.h b/xlators/cluster/nsr-server/src/yajl/yajl_parse.h
new file mode 100644
index 000000000..55c831101
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl/yajl_parse.h
@@ -0,0 +1,226 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+/**
+ * \file yajl_parse.h
+ * Interface to YAJL's JSON stream parsing facilities.
+ */
+
+#include <yajl/yajl_common.h>
+
+#ifndef __YAJL_PARSE_H__
+#define __YAJL_PARSE_H__
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+    /** error codes returned from this interface */
+    typedef enum {
+        /** no error was encountered */
+        yajl_status_ok,
+        /** a client callback returned zero, stopping the parse */
+        yajl_status_client_canceled,
+        /** An error occured during the parse.  Call yajl_get_error for
+         *  more information about the encountered error */
+        yajl_status_error
+    } yajl_status;
+
+    /** attain a human readable, english, string for an error */
+    YAJL_API const char * yajl_status_to_string(yajl_status code);
+
+    /** an opaque handle to a parser */
+    typedef struct yajl_handle_t * yajl_handle;
+
+    /** yajl is an event driven parser.  this means as json elements are
+     *  parsed, you are called back to do something with the data.  The
+     *  functions in this table indicate the various events for which
+     *  you will be called back.  Each callback accepts a "context"
+     *  pointer, this is a void * that is passed into the yajl_parse
+     *  function which the client code may use to pass around context.
+     *
+     *  All callbacks return an integer.  If non-zero, the parse will
+     *  continue.  If zero, the parse will be canceled and
+     *  yajl_status_client_canceled will be returned from the parse.
+     *
+     *  \attention {
+     *    A note about the handling of numbers:
+     *
+     *    yajl will only convert numbers that can be represented in a
+     *    double or a 64 bit (long long) int.  All other numbers will
+     *    be passed to the client in string form using the yajl_number
+     *    callback.  Furthermore, if yajl_number is not NULL, it will
+     *    always be used to return numbers, that is yajl_integer and
+     *    yajl_double will be ignored.  If yajl_number is NULL but one
+     *    of yajl_integer or yajl_double are defined, parsing of a
+     *    number larger than is representable in a double or 64 bit
+     *    integer will result in a parse error.
+     *  }
+     */
+    typedef struct {
+        int (* yajl_null)(void * ctx);
+        int (* yajl_boolean)(void * ctx, int boolVal);
+        int (* yajl_integer)(void * ctx, long long integerVal);
+        int (* yajl_double)(void * ctx, double doubleVal);
+        /** A callback which passes the string representation of the number
+         *  back to the client.  Will be used for all numbers when present */
+        int (* yajl_number)(void * ctx, const char * numberVal,
+                            size_t numberLen);
+
+        /** strings are returned as pointers into the JSON text when,
+         * possible, as a result, they are _not_ null padded */
+        int (* yajl_string)(void * ctx, const unsigned char * stringVal,
+                            size_t stringLen);
+
+        int (* yajl_start_map)(void * ctx);
+        int (* yajl_map_key)(void * ctx, const unsigned char * key,
+                             size_t stringLen);
+        int (* yajl_end_map)(void * ctx);
+
+        int (* yajl_start_array)(void * ctx);
+        int (* yajl_end_array)(void * ctx);
+    } yajl_callbacks;
+
+    /** allocate a parser handle
+     *  \param callbacks  a yajl callbacks structure specifying the
+     *                    functions to call when different JSON entities
+     *                    are encountered in the input text.  May be NULL,
+     *                    which is only useful for validation.
+     *  \param afs        memory allocation functions, may be NULL for to use
+     *                    C runtime library routines (malloc and friends) 
+     *  \param ctx        a context pointer that will be passed to callbacks.
+     */
+    YAJL_API yajl_handle yajl_alloc(const yajl_callbacks * callbacks,
+                                    yajl_alloc_funcs * afs,
+                                    void * ctx);
+
+
+    /** configuration parameters for the parser, these may be passed to
+     *  yajl_config() along with option specific argument(s).  In general,
+     *  all configuration parameters default to *off*. */
+    typedef enum {
+        /** Ignore javascript style comments present in
+         *  JSON input.  Non-standard, but rather fun
+         *  arguments: toggled off with integer zero, on otherwise.
+         *
+         *  example:
+         *    yajl_config(h, yajl_allow_comments, 1); // turn comment support on
+         */
+        yajl_allow_comments = 0x01,
+        /**
+         * When set the parser will verify that all strings in JSON input are
+         * valid UTF8 and will emit a parse error if this is not so.  When set,
+         * this option makes parsing slightly more expensive (~7% depending
+         * on processor and compiler in use)
+         *
+         * example:
+         *   yajl_config(h, yajl_dont_validate_strings, 1); // disable utf8 checking
+         */
+        yajl_dont_validate_strings     = 0x02,
+        /**
+         * By default, upon calls to yajl_complete_parse(), yajl will
+         * ensure the entire input text was consumed and will raise an error
+         * otherwise.  Enabling this flag will cause yajl to disable this
+         * check.  This can be useful when parsing json out of a that contains more
+         * than a single JSON document.
+         */
+        yajl_allow_trailing_garbage = 0x04,
+        /**
+         * Allow multiple values to be parsed by a single handle.  The
+         * entire text must be valid JSON, and values can be seperated
+         * by any kind of whitespace.  This flag will change the
+         * behavior of the parser, and cause it continue parsing after
+         * a value is parsed, rather than transitioning into a
+         * complete state.  This option can be useful when parsing multiple
+         * values from an input stream.
+         */
+        yajl_allow_multiple_values = 0x08,
+        /**
+         * When yajl_complete_parse() is called the parser will
+         * check that the top level value was completely consumed.  I.E.,
+         * if called whilst in the middle of parsing a value
+         * yajl will enter an error state (premature EOF).  Setting this
+         * flag suppresses that check and the corresponding error.
+         */
+        yajl_allow_partial_values = 0x10
+    } yajl_option;
+
+    /** allow the modification of parser options subsequent to handle
+     *  allocation (via yajl_alloc)
+     *  \returns zero in case of errors, non-zero otherwise
+     */
+    YAJL_API int yajl_config(yajl_handle h, yajl_option opt, ...);
+
+    /** free a parser handle */
+    YAJL_API void yajl_free(yajl_handle handle);
+
+    /** Parse some json!
+     *  \param hand - a handle to the json parser allocated with yajl_alloc
+     *  \param jsonText - a pointer to the UTF8 json text to be parsed
+     *  \param jsonTextLength - the length, in bytes, of input text
+     */
+    YAJL_API yajl_status yajl_parse(yajl_handle hand,
+                                    const unsigned char * jsonText,
+                                    size_t jsonTextLength);
+
+    /** Parse any remaining buffered json.
+     *  Since yajl is a stream-based parser, without an explicit end of
+     *  input, yajl sometimes can't decide if content at the end of the
+     *  stream is valid or not.  For example, if "1" has been fed in,
+     *  yajl can't know whether another digit is next or some character
+     *  that would terminate the integer token.
+     *
+     *  \param hand - a handle to the json parser allocated with yajl_alloc
+     */
+    YAJL_API yajl_status yajl_complete_parse(yajl_handle hand);
+
+    /** get an error string describing the state of the
+     *  parse.
+     *
+     *  If verbose is non-zero, the message will include the JSON
+     *  text where the error occured, along with an arrow pointing to
+     *  the specific char.
+     *
+     *  \returns A dynamically allocated string will be returned which should
+     *  be freed with yajl_free_error
+     */
+    YAJL_API unsigned char * yajl_get_error(yajl_handle hand, int verbose,
+                                            const unsigned char * jsonText,
+                                            size_t jsonTextLength);
+
+    /**
+     * get the amount of data consumed from the last chunk passed to YAJL.
+     *
+     * In the case of a successful parse this can help you understand if
+     * the entire buffer was consumed (which will allow you to handle
+     * "junk at end of input").
+     *
+     * In the event an error is encountered during parsing, this function
+     * affords the client a way to get the offset into the most recent
+     * chunk where the error occured.  0 will be returned if no error
+     * was encountered.
+     */
+    YAJL_API size_t yajl_get_bytes_consumed(yajl_handle hand);
+
+    /** free an error returned from yajl_get_error */
+    YAJL_API void yajl_free_error(yajl_handle hand, unsigned char * str);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/xlators/cluster/nsr-server/src/yajl/yajl_tree.h b/xlators/cluster/nsr-server/src/yajl/yajl_tree.h
new file mode 100644
index 000000000..8b377f636
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl/yajl_tree.h
@@ -0,0 +1,177 @@
+/*
+ * Copyright (c) 2010-2011  Florian Forster  <ff at octo.it>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+/**
+ * \file yajl_tree.h
+ *
+ * Parses JSON data and returns the data in tree form.
+ *
+ * \author Florian Forster
+ * \date August 2010
+ *
+ * This interface makes quick parsing and extraction of
+ * smallish JSON docs trivial:
+ *
+ * \include example/parse_config.c
+ */
+
+#ifndef YAJL_TREE_H
+#define YAJL_TREE_H 1
+
+#include <yajl/yajl_common.h>
+
+/** possible data types that a yajl_val_s can hold */
+typedef enum {
+    yajl_t_string = 1,
+    yajl_t_number = 2,
+    yajl_t_object = 3,
+    yajl_t_array = 4,
+    yajl_t_true = 5,
+    yajl_t_false = 6,
+    yajl_t_null = 7,
+    /** The any type isn't valid for yajl_val_s.type, but can be
+     *  used as an argument to routines like yajl_tree_get().
+     */
+    yajl_t_any = 8
+} yajl_type;
+
+#define YAJL_NUMBER_INT_VALID    0x01
+#define YAJL_NUMBER_DOUBLE_VALID 0x02
+
+/** A pointer to a node in the parse tree */
+typedef struct yajl_val_s * yajl_val;
+
+/**
+ * A JSON value representation capable of holding one of the seven
+ * types above. For "string", "number", "object", and "array"
+ * additional data is available in the union.  The "YAJL_IS_*"
+ * and "YAJL_GET_*" macros below allow type checking and convenient
+ * value extraction.
+ */
+struct yajl_val_s
+{
+    /** Type of the value contained. Use the "YAJL_IS_*" macors to check for a
+     * specific type. */
+    yajl_type type;
+    /** Type-specific data. You may use the "YAJL_GET_*" macros to access these
+     * members. */
+    union
+    {
+        char * string;
+        struct {
+            long long i; /*< integer value, if representable. */
+            double  d;   /*< double value, if representable. */
+            /** Signals whether the \em i and \em d members are
+             * valid. See \c YAJL_NUMBER_INT_VALID and
+             * \c YAJL_NUMBER_DOUBLE_VALID. */
+            char   *r;   /*< unparsed number in string form. */
+            unsigned int flags;
+        } number;
+        struct {
+            const char **keys; /*< Array of keys */
+            yajl_val *values; /*< Array of values. */
+            size_t len; /*< Number of key-value-pairs. */
+        } object;
+        struct {
+            yajl_val *values; /*< Array of elements. */
+            size_t len; /*< Number of elements. */
+        } array;
+    } u;
+};
+
+/**
+ * Parse a string.
+ *
+ * Parses an null-terminated string containing JSON data and returns a pointer
+ * to the top-level value (root of the parse tree).
+ *
+ * \param input              Pointer to a null-terminated utf8 string containing
+ *                           JSON data.
+ * \param error_buffer       Pointer to a buffer in which an error message will
+ *                           be stored if \em yajl_tree_parse fails, or
+ *                           \c NULL. The buffer will be initialized before
+ *                           parsing, so its content will be destroyed even if
+ *                           \em yajl_tree_parse succeeds.
+ * \param error_buffer_size  Size of the memory area pointed to by
+ *                           \em error_buffer_size. If \em error_buffer_size is
+ *                           \c NULL, this argument is ignored.
+ *
+ * \returns Pointer to the top-level value or \c NULL on error. The memory
+ * pointed to must be freed using \em yajl_tree_free. In case of an error, a
+ * null terminated message describing the error in more detail is stored in
+ * \em error_buffer if it is not \c NULL.
+ */
+YAJL_API yajl_val yajl_tree_parse (const char *input,
+                                   char *error_buffer, size_t error_buffer_size);
+
+/**
+ * Free a parse tree returned by "yajl_tree_parse".
+ *
+ * \param v Pointer to a JSON value returned by "yajl_tree_parse". Passing NULL
+ * is valid and results in a no-op.
+ */
+YAJL_API void yajl_tree_free (yajl_val v);
+
+/**
+ * Access a nested value inside a tree.
+ *
+ * \param parent the node under which you'd like to extract values.
+ * \param path A null terminated array of strings, each the name of an object key
+ * \param type the yajl_type of the object you seek, or yajl_t_any if any will do.
+ *
+ * \returns a pointer to the found value, or NULL if we came up empty.
+ * 
+ * Future Ideas:  it'd be nice to move path to a string and implement support for
+ * a teeny tiny micro language here, so you can extract array elements, do things
+ * like .first and .last, even .length.  Inspiration from JSONPath and css selectors?
+ * No it wouldn't be fast, but that's not what this API is about.
+ */
+YAJL_API yajl_val yajl_tree_get(yajl_val parent, const char ** path, yajl_type type);
+
+/* Various convenience macros to check the type of a `yajl_val` */
+#define YAJL_IS_STRING(v) (((v) != NULL) && ((v)->type == yajl_t_string))
+#define YAJL_IS_NUMBER(v) (((v) != NULL) && ((v)->type == yajl_t_number))
+#define YAJL_IS_INTEGER(v) (YAJL_IS_NUMBER(v) && ((v)->u.flags & YAJL_NUMBER_INT_VALID))
+#define YAJL_IS_DOUBLE(v) (YAJL_IS_NUMBER(v) && ((v)->u.flags & YAJL_NUMBER_DOUBLE_VALID))
+#define YAJL_IS_OBJECT(v) (((v) != NULL) && ((v)->type == yajl_t_object))
+#define YAJL_IS_ARRAY(v)  (((v) != NULL) && ((v)->type == yajl_t_array ))
+#define YAJL_IS_TRUE(v)   (((v) != NULL) && ((v)->type == yajl_t_true  ))
+#define YAJL_IS_FALSE(v)  (((v) != NULL) && ((v)->type == yajl_t_false ))
+#define YAJL_IS_NULL(v)   (((v) != NULL) && ((v)->type == yajl_t_null  ))
+
+/** Given a yajl_val_string return a ptr to the bare string it contains,
+ *  or NULL if the value is not a string. */
+#define YAJL_GET_STRING(v) (YAJL_IS_STRING(v) ? (v)->u.string : NULL)
+
+/** Get the string representation of a number.  You should check type first,
+ *  perhaps using YAJL_IS_NUMBER */
+#define YAJL_GET_NUMBER(v) ((v)->u.number.r)
+
+/** Get the double representation of a number.  You should check type first,
+ *  perhaps using YAJL_IS_DOUBLE */
+#define YAJL_GET_DOUBLE(v) ((v)->u.number.d)
+
+/** Get the 64bit (long long) integer representation of a number.  You should
+ *  check type first, perhaps using YAJL_IS_INTEGER */
+#define YAJL_GET_INTEGER(v) ((v)->u.number.i)
+
+/** Get a pointer to a yajl_val_object or NULL if the value is not an object. */
+#define YAJL_GET_OBJECT(v) (YAJL_IS_OBJECT(v) ? &(v)->u.object : NULL)
+
+/** Get a pointer to a yajl_val_array or NULL if the value is not an object. */
+#define YAJL_GET_ARRAY(v)  (YAJL_IS_ARRAY(v)  ? &(v)->u.array  : NULL)
+
+#endif /* YAJL_TREE_H */
diff --git a/xlators/cluster/nsr-server/src/yajl/yajl_version.h b/xlators/cluster/nsr-server/src/yajl/yajl_version.h
new file mode 100644
index 000000000..0fba9b8fc
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl/yajl_version.h
@@ -0,0 +1,23 @@
+#ifndef YAJL_VERSION_H_
+#define YAJL_VERSION_H_
+
+#include <yajl/yajl_common.h>
+
+#define YAJL_MAJOR 2
+#define YAJL_MINOR 0
+#define YAJL_MICRO 1
+
+#define YAJL_VERSION ((YAJL_MAJOR * 10000) + (YAJL_MINOR * 100) + YAJL_MICRO)
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+extern int YAJL_API yajl_version(void);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* YAJL_VERSION_H_ */
+
diff --git a/xlators/cluster/nsr-server/src/yajl_alloc.c b/xlators/cluster/nsr-server/src/yajl_alloc.c
new file mode 100644
index 000000000..276315af7
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_alloc.c
@@ -0,0 +1,49 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+/**
+ * \file yajl_alloc.h
+ * default memory allocation routines for yajl which use malloc/realloc and
+ * free
+ */
+
+#include "yajl_alloc.h"
+#include <stdlib.h>
+
+static void * yajl_internal_malloc(void *ctx, size_t sz)
+{
+    return malloc(sz);
+}
+
+static void * yajl_internal_realloc(void *ctx, void * previous,
+                                    size_t sz)
+{
+    return realloc(previous, sz);
+}
+
+static void yajl_internal_free(void *ctx, void * ptr)
+{
+    free(ptr);
+}
+
+void yajl_set_default_alloc_funcs(yajl_alloc_funcs * yaf)
+{
+    yaf->malloc = yajl_internal_malloc;
+    yaf->free = yajl_internal_free;
+    yaf->realloc = yajl_internal_realloc;
+    yaf->ctx = NULL;
+}
+
diff --git a/xlators/cluster/nsr-server/src/yajl_alloc.h b/xlators/cluster/nsr-server/src/yajl_alloc.h
new file mode 100644
index 000000000..a8a9e45e6
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_alloc.h
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+/**
+ * \file yajl_alloc.h
+ * default memory allocation routines for yajl which use malloc/realloc and
+ * free
+ */
+
+#ifndef __YAJL_ALLOC_H__
+#define __YAJL_ALLOC_H__
+
+#include "yajl/yajl_common.h"
+
+#define YA_MALLOC(afs, sz) (afs)->malloc((afs)->ctx, (sz))
+#define YA_FREE(afs, ptr) (afs)->free((afs)->ctx, (ptr))
+#define YA_REALLOC(afs, ptr, sz) (afs)->realloc((afs)->ctx, (ptr), (sz))
+
+void yajl_set_default_alloc_funcs(yajl_alloc_funcs * yaf);
+
+#endif
diff --git a/xlators/cluster/nsr-server/src/yajl_buf.c b/xlators/cluster/nsr-server/src/yajl_buf.c
new file mode 100644
index 000000000..0d249d364
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_buf.c
@@ -0,0 +1,103 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "yajl_buf.h"
+
+#include <assert.h>
+#include <stdlib.h>
+#include <string.h>
+
+#define YAJL_BUF_INIT_SIZE 2048
+
+struct yajl_buf_t {
+    size_t len;
+    size_t used;
+    unsigned char * data;
+    yajl_alloc_funcs * alloc;
+};
+
+static
+void yajl_buf_ensure_available(yajl_buf buf, size_t want)
+{
+    size_t need;
+    
+    assert(buf != NULL);
+
+    /* first call */
+    if (buf->data == NULL) {
+        buf->len = YAJL_BUF_INIT_SIZE;
+        buf->data = (unsigned char *) YA_MALLOC(buf->alloc, buf->len);
+        buf->data[0] = 0;
+    }
+
+    need = buf->len;
+
+    while (want >= (need - buf->used)) need <<= 1;
+
+    if (need != buf->len) {
+        buf->data = (unsigned char *) YA_REALLOC(buf->alloc, buf->data, need);
+        buf->len = need;
+    }
+}
+
+yajl_buf yajl_buf_alloc(yajl_alloc_funcs * alloc)
+{
+    yajl_buf b = YA_MALLOC(alloc, sizeof(struct yajl_buf_t));
+    memset((void *) b, 0, sizeof(struct yajl_buf_t));
+    b->alloc = alloc;
+    return b;
+}
+
+void yajl_buf_free(yajl_buf buf)
+{
+    assert(buf != NULL);
+    if (buf->data) YA_FREE(buf->alloc, buf->data);
+    YA_FREE(buf->alloc, buf);
+}
+
+void yajl_buf_append(yajl_buf buf, const void * data, size_t len)
+{
+    yajl_buf_ensure_available(buf, len);
+    if (len > 0) {
+        assert(data != NULL);
+        memcpy(buf->data + buf->used, data, len);
+        buf->used += len;
+        buf->data[buf->used] = 0;
+    }
+}
+
+void yajl_buf_clear(yajl_buf buf)
+{
+    buf->used = 0;
+    if (buf->data) buf->data[buf->used] = 0;
+}
+
+const unsigned char * yajl_buf_data(yajl_buf buf)
+{
+    return buf->data;
+}
+
+size_t yajl_buf_len(yajl_buf buf)
+{
+    return buf->used;
+}
+
+void
+yajl_buf_truncate(yajl_buf buf, size_t len)
+{
+    assert(len <= buf->used);
+    buf->used = len;
+}
diff --git a/xlators/cluster/nsr-server/src/yajl_buf.h b/xlators/cluster/nsr-server/src/yajl_buf.h
new file mode 100644
index 000000000..94929a519
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_buf.h
@@ -0,0 +1,57 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#ifndef __YAJL_BUF_H__
+#define __YAJL_BUF_H__
+
+#include "yajl/yajl_common.h"
+#include "yajl_alloc.h"
+
+/*
+ * Implementation/performance notes.  If this were moved to a header
+ * only implementation using #define's where possible we might be 
+ * able to sqeeze a little performance out of the guy by killing function
+ * call overhead.  YMMV.
+ */
+
+/**
+ * yajl_buf is a buffer with exponential growth.  the buffer ensures that
+ * you are always null padded.
+ */
+typedef struct yajl_buf_t * yajl_buf;
+
+/* allocate a new buffer */
+yajl_buf yajl_buf_alloc(yajl_alloc_funcs * alloc);
+
+/* free the buffer */
+void yajl_buf_free(yajl_buf buf);
+
+/* append a number of bytes to the buffer */
+void yajl_buf_append(yajl_buf buf, const void * data, size_t len);
+
+/* empty the buffer */
+void yajl_buf_clear(yajl_buf buf);
+
+/* get a pointer to the beginning of the buffer */
+const unsigned char * yajl_buf_data(yajl_buf buf);
+
+/* get the length of the buffer */
+size_t yajl_buf_len(yajl_buf buf);
+
+/* truncate the buffer */
+void yajl_buf_truncate(yajl_buf buf, size_t len);
+
+#endif
diff --git a/xlators/cluster/nsr-server/src/yajl_bytestack.h b/xlators/cluster/nsr-server/src/yajl_bytestack.h
new file mode 100644
index 000000000..1fc50c470
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_bytestack.h
@@ -0,0 +1,69 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+/*
+ * A header only implementation of a simple stack of bytes, used in YAJL
+ * to maintain parse state.
+ */
+
+#ifndef __YAJL_BYTESTACK_H__
+#define __YAJL_BYTESTACK_H__
+
+#include "yajl/yajl_common.h"
+
+#define YAJL_BS_INC 128
+
+typedef struct yajl_bytestack_t
+{
+    unsigned char * stack;
+    size_t size;
+    size_t used;
+    yajl_alloc_funcs * yaf;
+} yajl_bytestack;
+
+/* initialize a bytestack */
+#define yajl_bs_init(obs, _yaf) {               \
+        (obs).stack = NULL;                     \
+        (obs).size = 0;                         \
+        (obs).used = 0;                         \
+        (obs).yaf = (_yaf);                     \
+    }                                           \
+
+
+/* initialize a bytestack */
+#define yajl_bs_free(obs)                 \
+    if ((obs).stack) (obs).yaf->free((obs).yaf->ctx, (obs).stack);
+
+#define yajl_bs_current(obs)               \
+    (assert((obs).used > 0), (obs).stack[(obs).used - 1])
+
+#define yajl_bs_push(obs, byte) {                       \
+    if (((obs).size - (obs).used) == 0) {               \
+        (obs).size += YAJL_BS_INC;                      \
+        (obs).stack = (obs).yaf->realloc((obs).yaf->ctx,\
+                                         (void *) (obs).stack, (obs).size);\
+    }                                                   \
+    (obs).stack[((obs).used)++] = (byte);               \
+}
+
+/* removes the top item of the stack, returns nothing */
+#define yajl_bs_pop(obs) { ((obs).used)--; }
+
+#define yajl_bs_set(obs, byte)                          \
+    (obs).stack[((obs).used) - 1] = (byte);
+
+
+#endif
diff --git a/xlators/cluster/nsr-server/src/yajl_encode.c b/xlators/cluster/nsr-server/src/yajl_encode.c
new file mode 100644
index 000000000..9dc9a3e81
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_encode.c
@@ -0,0 +1,220 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "yajl_encode.h"
+
+#include <assert.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+
+static void CharToHex(unsigned char c, char * hexBuf)
+{
+    const char * hexchar = "0123456789ABCDEF";
+    hexBuf[0] = hexchar[c >> 4];
+    hexBuf[1] = hexchar[c & 0x0F];
+}
+
+void
+yajl_string_encode(const yajl_print_t print,
+                   void * ctx,
+                   const unsigned char * str,
+                   size_t len,
+                   int escape_solidus)
+{
+    size_t beg = 0;
+    size_t end = 0;
+    char hexBuf[7];
+    hexBuf[0] = '\\'; hexBuf[1] = 'u'; hexBuf[2] = '0'; hexBuf[3] = '0';
+    hexBuf[6] = 0;
+
+    while (end < len) {
+        const char * escaped = NULL;
+        switch (str[end]) {
+            case '\r': escaped = "\\r"; break;
+            case '\n': escaped = "\\n"; break;
+            case '\\': escaped = "\\\\"; break;
+            /* it is not required to escape a solidus in JSON:
+             * read sec. 2.5: http://www.ietf.org/rfc/rfc4627.txt
+             * specifically, this production from the grammar:
+             *   unescaped = %x20-21 / %x23-5B / %x5D-10FFFF
+             */
+            case '/': if (escape_solidus) escaped = "\\/"; break;
+            case '"': escaped = "\\\""; break;
+            case '\f': escaped = "\\f"; break;
+            case '\b': escaped = "\\b"; break;
+            case '\t': escaped = "\\t"; break;
+            default:
+                if ((unsigned char) str[end] < 32) {
+                    CharToHex(str[end], hexBuf + 4);
+                    escaped = hexBuf;
+                }
+                break;
+        }
+        if (escaped != NULL) {
+            print(ctx, (const char *) (str + beg), end - beg);
+            print(ctx, escaped, (unsigned int)strlen(escaped));
+            beg = ++end;
+        } else {
+            ++end;
+        }
+    }
+    print(ctx, (const char *) (str + beg), end - beg);
+}
+
+static void hexToDigit(unsigned int * val, const unsigned char * hex)
+{
+    unsigned int i;
+    for (i=0;i<4;i++) {
+        unsigned char c = hex[i];
+        if (c >= 'A') c = (c & ~0x20) - 7;
+        c -= '0';
+        assert(!(c & 0xF0));
+        *val = (*val << 4) | c;
+    }
+}
+
+static void Utf32toUtf8(unsigned int codepoint, char * utf8Buf) 
+{
+    if (codepoint < 0x80) {
+        utf8Buf[0] = (char) codepoint;
+        utf8Buf[1] = 0;
+    } else if (codepoint < 0x0800) {
+        utf8Buf[0] = (char) ((codepoint >> 6) | 0xC0);
+        utf8Buf[1] = (char) ((codepoint & 0x3F) | 0x80);
+        utf8Buf[2] = 0;
+    } else if (codepoint < 0x10000) {
+        utf8Buf[0] = (char) ((codepoint >> 12) | 0xE0);
+        utf8Buf[1] = (char) (((codepoint >> 6) & 0x3F) | 0x80);
+        utf8Buf[2] = (char) ((codepoint & 0x3F) | 0x80);
+        utf8Buf[3] = 0;
+    } else if (codepoint < 0x200000) {
+        utf8Buf[0] =(char)((codepoint >> 18) | 0xF0);
+        utf8Buf[1] =(char)(((codepoint >> 12) & 0x3F) | 0x80);
+        utf8Buf[2] =(char)(((codepoint >> 6) & 0x3F) | 0x80);
+        utf8Buf[3] =(char)((codepoint & 0x3F) | 0x80);
+        utf8Buf[4] = 0;
+    } else {
+        utf8Buf[0] = '?';
+        utf8Buf[1] = 0;
+    }
+}
+
+void yajl_string_decode(yajl_buf buf, const unsigned char * str,
+                        size_t len)
+{
+    size_t beg = 0;
+    size_t end = 0;    
+
+    while (end < len) {
+        if (str[end] == '\\') {
+            char utf8Buf[5];
+            const char * unescaped = "?";
+            yajl_buf_append(buf, str + beg, end - beg);
+            switch (str[++end]) {
+                case 'r': unescaped = "\r"; break;
+                case 'n': unescaped = "\n"; break;
+                case '\\': unescaped = "\\"; break;
+                case '/': unescaped = "/"; break;
+                case '"': unescaped = "\""; break;
+                case 'f': unescaped = "\f"; break;
+                case 'b': unescaped = "\b"; break;
+                case 't': unescaped = "\t"; break;
+                case 'u': {
+                    unsigned int codepoint = 0;
+                    hexToDigit(&codepoint, str + ++end);
+                    end+=3;
+                    /* check if this is a surrogate */
+                    if ((codepoint & 0xFC00) == 0xD800) {
+                        end++;
+                        if (str[end] == '\\' && str[end + 1] == 'u') {
+                            unsigned int surrogate = 0;
+                            hexToDigit(&surrogate, str + end + 2);
+                            codepoint =
+                                (((codepoint & 0x3F) << 10) | 
+                                 ((((codepoint >> 6) & 0xF) + 1) << 16) | 
+                                 (surrogate & 0x3FF));
+                            end += 5;
+                        } else {
+                            unescaped = "?";
+                            break;
+                        }
+                    }
+                    
+                    Utf32toUtf8(codepoint, utf8Buf);
+                    unescaped = utf8Buf;
+
+                    if (codepoint == 0) {
+                        yajl_buf_append(buf, unescaped, 1);
+                        beg = ++end;
+                        continue;
+                    }
+
+                    break;
+                }
+                default:
+                    assert("this should never happen" == NULL);
+            }
+            yajl_buf_append(buf, unescaped, (unsigned int)strlen(unescaped));
+            beg = ++end;
+        } else {
+            end++;
+        }
+    }
+    yajl_buf_append(buf, str + beg, end - beg);
+}
+
+#define ADV_PTR s++; if (!(len--)) return 0;
+
+int yajl_string_validate_utf8(const unsigned char * s, size_t len)
+{
+    if (!len) return 1;
+    if (!s) return 0;
+    
+    while (len--) {
+        /* single byte */
+        if (*s <= 0x7f) {
+            /* noop */
+        }
+        /* two byte */ 
+        else if ((*s >> 5) == 0x6) {
+            ADV_PTR;
+            if (!((*s >> 6) == 0x2)) return 0;
+        }
+        /* three byte */
+        else if ((*s >> 4) == 0x0e) {
+            ADV_PTR;
+            if (!((*s >> 6) == 0x2)) return 0;
+            ADV_PTR;
+            if (!((*s >> 6) == 0x2)) return 0;
+        }
+        /* four byte */        
+        else if ((*s >> 3) == 0x1e) {
+            ADV_PTR;
+            if (!((*s >> 6) == 0x2)) return 0;
+            ADV_PTR;
+            if (!((*s >> 6) == 0x2)) return 0;
+            ADV_PTR;
+            if (!((*s >> 6) == 0x2)) return 0;
+        } else {
+            return 0;
+        }
+        
+        s++;
+    }
+    
+    return 1;
+}
diff --git a/xlators/cluster/nsr-server/src/yajl_encode.h b/xlators/cluster/nsr-server/src/yajl_encode.h
new file mode 100644
index 000000000..af1e8bbde
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_encode.h
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#ifndef __YAJL_ENCODE_H__
+#define __YAJL_ENCODE_H__
+
+#include "yajl_buf.h"
+#include "yajl/yajl_gen.h"
+
+void yajl_string_encode(const yajl_print_t printer,
+                        void * ctx,
+                        const unsigned char * str,
+                        size_t length,
+                        int escape_solidus);
+
+void yajl_string_decode(yajl_buf buf, const unsigned char * str,
+                        size_t length);
+
+int yajl_string_validate_utf8(const unsigned char * s, size_t len);
+
+#endif
diff --git a/xlators/cluster/nsr-server/src/yajl_gen.c b/xlators/cluster/nsr-server/src/yajl_gen.c
new file mode 100644
index 000000000..73763a9e0
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_gen.c
@@ -0,0 +1,350 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "yajl/yajl_gen.h"
+#include "yajl_buf.h"
+#include "yajl_encode.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <math.h>
+#include <stdarg.h>
+
+typedef enum {
+    yajl_gen_start,
+    yajl_gen_map_start,
+    yajl_gen_map_key,
+    yajl_gen_map_val,
+    yajl_gen_array_start,
+    yajl_gen_in_array,
+    yajl_gen_complete,
+    yajl_gen_error
+} yajl_gen_state;
+
+struct yajl_gen_t
+{
+    unsigned int flags;
+    unsigned int depth;
+    const char * indentString;
+    yajl_gen_state state[YAJL_MAX_DEPTH];
+    yajl_print_t print;
+    void * ctx; /* yajl_buf */
+    /* memory allocation routines */
+    yajl_alloc_funcs alloc;
+};
+
+int
+yajl_gen_config(yajl_gen g, yajl_gen_option opt, ...)
+{
+    int rv = 1;
+    va_list ap;
+    va_start(ap, opt);
+
+    switch(opt) {
+        case yajl_gen_beautify:
+        case yajl_gen_validate_utf8:
+            if (va_arg(ap, int)) g->flags |= opt;
+            else g->flags &= ~opt;
+            break;
+        case yajl_gen_indent_string: {
+            const char *indent = va_arg(ap, const char *);
+            g->indentString = indent;
+            for (; *indent; indent++) {
+                if (*indent != '\n'
+                    && *indent != '\v'
+                    && *indent != '\f'
+                    && *indent != '\t'
+                    && *indent != '\r'
+                    && *indent != ' ')
+                {
+                    g->indentString = NULL;
+                    rv = 0;
+                }
+            }
+            break;
+        }
+        case yajl_gen_print_callback:
+            yajl_buf_free(g->ctx);
+            g->print = va_arg(ap, const yajl_print_t);
+            g->ctx = va_arg(ap, void *);
+            break;
+        default:
+            rv = 0;
+    }
+
+    va_end(ap);
+
+    return rv;
+}
+
+
+
+yajl_gen
+yajl_gen_alloc(const yajl_alloc_funcs * afs)
+{
+    yajl_gen g = NULL;
+    yajl_alloc_funcs afsBuffer;
+
+    /* first order of business is to set up memory allocation routines */
+    if (afs != NULL) {
+        if (afs->malloc == NULL || afs->realloc == NULL || afs->free == NULL)
+        {
+            return NULL;
+        }
+    } else {
+        yajl_set_default_alloc_funcs(&afsBuffer);
+        afs = &afsBuffer;
+    }
+
+    g = (yajl_gen) YA_MALLOC(afs, sizeof(struct yajl_gen_t));
+    if (!g) return NULL;
+
+    memset((void *) g, 0, sizeof(struct yajl_gen_t));
+    /* copy in pointers to allocation routines */
+    memcpy((void *) &(g->alloc), (void *) afs, sizeof(yajl_alloc_funcs));
+
+    g->print = (yajl_print_t)&yajl_buf_append;
+    g->ctx = yajl_buf_alloc(&(g->alloc));
+    g->indentString = "    ";
+
+    return g;
+}
+
+void
+yajl_gen_free(yajl_gen g)
+{
+    if (g->print == (yajl_print_t)&yajl_buf_append) yajl_buf_free((yajl_buf)g->ctx);
+    YA_FREE(&(g->alloc), g);
+}
+
+#define INSERT_SEP \
+    if (g->state[g->depth] == yajl_gen_map_key ||               \
+        g->state[g->depth] == yajl_gen_in_array) {              \
+        g->print(g->ctx, ",", 1);                               \
+        if ((g->flags & yajl_gen_beautify)) g->print(g->ctx, "\n", 1);               \
+    } else if (g->state[g->depth] == yajl_gen_map_val) {        \
+        g->print(g->ctx, ":", 1);                               \
+        if ((g->flags & yajl_gen_beautify)) g->print(g->ctx, " ", 1);                \
+   } 
+
+#define INSERT_WHITESPACE                                               \
+    if ((g->flags & yajl_gen_beautify)) {                                                    \
+        if (g->state[g->depth] != yajl_gen_map_val) {                   \
+            unsigned int _i;                                            \
+            for (_i=0;_i<g->depth;_i++)                                 \
+                g->print(g->ctx,                                        \
+                         g->indentString,                               \
+                         (unsigned int)strlen(g->indentString));        \
+        }                                                               \
+    }
+
+#define ENSURE_NOT_KEY \
+    if (g->state[g->depth] == yajl_gen_map_key ||       \
+        g->state[g->depth] == yajl_gen_map_start)  {    \
+        return yajl_gen_keys_must_be_strings;           \
+    }                                                   \
+
+/* check that we're not complete, or in error state.  in a valid state
+ * to be generating */
+#define ENSURE_VALID_STATE \
+    if (g->state[g->depth] == yajl_gen_error) {   \
+        return yajl_gen_in_error_state;\
+    } else if (g->state[g->depth] == yajl_gen_complete) {   \
+        return yajl_gen_generation_complete;                \
+    }
+
+#define INCREMENT_DEPTH \
+    if (++(g->depth) >= YAJL_MAX_DEPTH) return yajl_max_depth_exceeded;
+
+#define DECREMENT_DEPTH \
+  if (--(g->depth) >= YAJL_MAX_DEPTH) return yajl_gen_error;
+
+#define APPENDED_ATOM \
+    switch (g->state[g->depth]) {                   \
+        case yajl_gen_start:                        \
+            g->state[g->depth] = yajl_gen_complete; \
+            break;                                  \
+        case yajl_gen_map_start:                    \
+        case yajl_gen_map_key:                      \
+            g->state[g->depth] = yajl_gen_map_val;  \
+            break;                                  \
+        case yajl_gen_array_start:                  \
+            g->state[g->depth] = yajl_gen_in_array; \
+            break;                                  \
+        case yajl_gen_map_val:                      \
+            g->state[g->depth] = yajl_gen_map_key;  \
+            break;                                  \
+        default:                                    \
+            break;                                  \
+    }                                               \
+
+#define FINAL_NEWLINE                                        \
+    if ((g->flags & yajl_gen_beautify) && g->state[g->depth] == yajl_gen_complete) \
+        g->print(g->ctx, "\n", 1);
+
+yajl_gen_status
+yajl_gen_integer(yajl_gen g, long long int number)
+{
+    char i[32];
+    ENSURE_VALID_STATE; ENSURE_NOT_KEY; INSERT_SEP; INSERT_WHITESPACE;
+    sprintf(i, "%lld", number);
+    g->print(g->ctx, i, (unsigned int)strlen(i));
+    APPENDED_ATOM;
+    FINAL_NEWLINE;
+    return yajl_gen_status_ok;
+}
+
+#ifdef WIN32
+#include <float.h>
+#define isnan _isnan
+#define isinf !_finite
+#endif
+
+yajl_gen_status
+yajl_gen_double(yajl_gen g, double number)
+{
+    char i[32];
+    ENSURE_VALID_STATE; ENSURE_NOT_KEY; 
+    if (isnan(number) || isinf(number)) return yajl_gen_invalid_number;
+    INSERT_SEP; INSERT_WHITESPACE;
+    sprintf(i, "%.20g", number);
+    g->print(g->ctx, i, (unsigned int)strlen(i));
+    APPENDED_ATOM;
+    FINAL_NEWLINE;
+    return yajl_gen_status_ok;
+}
+
+yajl_gen_status
+yajl_gen_number(yajl_gen g, const char * s, size_t l)
+{
+    ENSURE_VALID_STATE; ENSURE_NOT_KEY; INSERT_SEP; INSERT_WHITESPACE;
+    g->print(g->ctx, s, l);
+    APPENDED_ATOM;
+    FINAL_NEWLINE;
+    return yajl_gen_status_ok;
+}
+
+yajl_gen_status
+yajl_gen_string(yajl_gen g, const unsigned char * str,
+                size_t len)
+{
+    // if validation is enabled, check that the string is valid utf8
+    // XXX: This checking could be done a little faster, in the same pass as
+    // the string encoding
+    if (g->flags & yajl_gen_validate_utf8) {
+        if (!yajl_string_validate_utf8(str, len)) {
+            return yajl_gen_invalid_string;
+        }
+    }
+    ENSURE_VALID_STATE; INSERT_SEP; INSERT_WHITESPACE;
+    g->print(g->ctx, "\"", 1);
+    yajl_string_encode(g->print, g->ctx, str, len, g->flags & yajl_gen_escape_solidus);
+    g->print(g->ctx, "\"", 1);
+    APPENDED_ATOM;
+    FINAL_NEWLINE;
+    return yajl_gen_status_ok;
+}
+
+yajl_gen_status
+yajl_gen_null(yajl_gen g)
+{
+    ENSURE_VALID_STATE; ENSURE_NOT_KEY; INSERT_SEP; INSERT_WHITESPACE;
+    g->print(g->ctx, "null", strlen("null"));
+    APPENDED_ATOM;
+    FINAL_NEWLINE;
+    return yajl_gen_status_ok;
+}
+
+yajl_gen_status
+yajl_gen_bool(yajl_gen g, int boolean)
+{
+    const char * val = boolean ? "true" : "false";
+
+	ENSURE_VALID_STATE; ENSURE_NOT_KEY; INSERT_SEP; INSERT_WHITESPACE;
+    g->print(g->ctx, val, (unsigned int)strlen(val));
+    APPENDED_ATOM;
+    FINAL_NEWLINE;
+    return yajl_gen_status_ok;
+}
+
+yajl_gen_status
+yajl_gen_map_open(yajl_gen g)
+{
+    ENSURE_VALID_STATE; ENSURE_NOT_KEY; INSERT_SEP; INSERT_WHITESPACE;
+    INCREMENT_DEPTH; 
+    
+    g->state[g->depth] = yajl_gen_map_start;
+    g->print(g->ctx, "{", 1);
+    if ((g->flags & yajl_gen_beautify)) g->print(g->ctx, "\n", 1);
+    FINAL_NEWLINE;
+    return yajl_gen_status_ok;
+}
+
+yajl_gen_status
+yajl_gen_map_close(yajl_gen g)
+{
+    ENSURE_VALID_STATE; 
+    DECREMENT_DEPTH;
+    
+    if ((g->flags & yajl_gen_beautify)) g->print(g->ctx, "\n", 1);
+    APPENDED_ATOM;
+    INSERT_WHITESPACE;
+    g->print(g->ctx, "}", 1);
+    FINAL_NEWLINE;
+    return yajl_gen_status_ok;
+}
+
+yajl_gen_status
+yajl_gen_array_open(yajl_gen g)
+{
+    ENSURE_VALID_STATE; ENSURE_NOT_KEY; INSERT_SEP; INSERT_WHITESPACE;
+    INCREMENT_DEPTH; 
+    g->state[g->depth] = yajl_gen_array_start;
+    g->print(g->ctx, "[", 1);
+    if ((g->flags & yajl_gen_beautify)) g->print(g->ctx, "\n", 1);
+    FINAL_NEWLINE;
+    return yajl_gen_status_ok;
+}
+
+yajl_gen_status
+yajl_gen_array_close(yajl_gen g)
+{
+    ENSURE_VALID_STATE;
+    DECREMENT_DEPTH;
+    if ((g->flags & yajl_gen_beautify)) g->print(g->ctx, "\n", 1);
+    APPENDED_ATOM;
+    INSERT_WHITESPACE;
+    g->print(g->ctx, "]", 1);
+    FINAL_NEWLINE;
+    return yajl_gen_status_ok;
+}
+
+yajl_gen_status
+yajl_gen_get_buf(yajl_gen g, const unsigned char ** buf,
+                 size_t * len)
+{
+    if (g->print != (yajl_print_t)&yajl_buf_append) return yajl_gen_no_buf;
+    *buf = yajl_buf_data((yajl_buf)g->ctx);
+    *len = yajl_buf_len((yajl_buf)g->ctx);
+    return yajl_gen_status_ok;
+}
+
+void
+yajl_gen_clear(yajl_gen g)
+{
+    if (g->print == (yajl_print_t)&yajl_buf_append) yajl_buf_clear((yajl_buf)g->ctx);
+}
diff --git a/xlators/cluster/nsr-server/src/yajl_lex.c b/xlators/cluster/nsr-server/src/yajl_lex.c
new file mode 100644
index 000000000..b098e6a99
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_lex.c
@@ -0,0 +1,763 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "yajl_lex.h"
+#include "yajl_buf.h"
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <assert.h>
+#include <string.h>
+
+#ifdef YAJL_LEXER_DEBUG
+static const char *
+tokToStr(yajl_tok tok) 
+{
+    switch (tok) {
+        case yajl_tok_bool: return "bool";
+        case yajl_tok_colon: return "colon";
+        case yajl_tok_comma: return "comma";
+        case yajl_tok_eof: return "eof";
+        case yajl_tok_error: return "error";
+        case yajl_tok_left_brace: return "brace";
+        case yajl_tok_left_bracket: return "bracket";
+        case yajl_tok_null: return "null";
+        case yajl_tok_integer: return "integer";
+        case yajl_tok_double: return "double";
+        case yajl_tok_right_brace: return "brace";
+        case yajl_tok_right_bracket: return "bracket";
+        case yajl_tok_string: return "string";
+        case yajl_tok_string_with_escapes: return "string_with_escapes";
+    }
+    return "unknown";
+}
+#endif
+
+/* Impact of the stream parsing feature on the lexer:
+ *
+ * YAJL support stream parsing.  That is, the ability to parse the first
+ * bits of a chunk of JSON before the last bits are available (still on
+ * the network or disk).  This makes the lexer more complex.  The
+ * responsibility of the lexer is to handle transparently the case where
+ * a chunk boundary falls in the middle of a token.  This is
+ * accomplished is via a buffer and a character reading abstraction. 
+ *
+ * Overview of implementation
+ *
+ * When we lex to end of input string before end of token is hit, we
+ * copy all of the input text composing the token into our lexBuf.
+ * 
+ * Every time we read a character, we do so through the readChar function.
+ * readChar's responsibility is to handle pulling all chars from the buffer
+ * before pulling chars from input text
+ */
+
+struct yajl_lexer_t {
+    /* the overal line and char offset into the data */
+    size_t lineOff;
+    size_t charOff;
+
+    /* error */
+    yajl_lex_error error;
+
+    /* a input buffer to handle the case where a token is spread over
+     * multiple chunks */ 
+    yajl_buf buf;
+
+    /* in the case where we have data in the lexBuf, bufOff holds
+     * the current offset into the lexBuf. */
+    size_t bufOff;
+
+    /* are we using the lex buf? */
+    unsigned int bufInUse;
+
+    /* shall we allow comments? */
+    unsigned int allowComments;
+
+    /* shall we validate utf8 inside strings? */
+    unsigned int validateUTF8;
+
+    yajl_alloc_funcs * alloc;
+};
+
+#define readChar(lxr, txt, off)                      \
+    (((lxr)->bufInUse && yajl_buf_len((lxr)->buf) && lxr->bufOff < yajl_buf_len((lxr)->buf)) ? \
+     (*((const unsigned char *) yajl_buf_data((lxr)->buf) + ((lxr)->bufOff)++)) : \
+     ((txt)[(*(off))++]))
+
+#define unreadChar(lxr, off) ((*(off) > 0) ? (*(off))-- : ((lxr)->bufOff--))
+
+yajl_lexer
+yajl_lex_alloc(yajl_alloc_funcs * alloc,
+               unsigned int allowComments, unsigned int validateUTF8)
+{
+    yajl_lexer lxr = (yajl_lexer) YA_MALLOC(alloc, sizeof(struct yajl_lexer_t));
+    memset((void *) lxr, 0, sizeof(struct yajl_lexer_t));
+    lxr->buf = yajl_buf_alloc(alloc);
+    lxr->allowComments = allowComments;
+    lxr->validateUTF8 = validateUTF8;
+    lxr->alloc = alloc;
+    return lxr;
+}
+
+void
+yajl_lex_free(yajl_lexer lxr)
+{
+    yajl_buf_free(lxr->buf);
+    YA_FREE(lxr->alloc, lxr);
+    return;
+}
+
+/* a lookup table which lets us quickly determine three things:
+ * VEC - valid escaped control char
+ * note.  the solidus '/' may be escaped or not.
+ * IJC - invalid json char
+ * VHC - valid hex char
+ * NFP - needs further processing (from a string scanning perspective)
+ * NUC - needs utf8 checking when enabled (from a string scanning perspective)
+ */
+#define VEC 0x01
+#define IJC 0x02
+#define VHC 0x04
+#define NFP 0x08
+#define NUC 0x10
+
+static const char charLookupTable[256] =
+{
+/*00*/ IJC    , IJC    , IJC    , IJC    , IJC    , IJC    , IJC    , IJC    ,
+/*08*/ IJC    , IJC    , IJC    , IJC    , IJC    , IJC    , IJC    , IJC    ,
+/*10*/ IJC    , IJC    , IJC    , IJC    , IJC    , IJC    , IJC    , IJC    ,
+/*18*/ IJC    , IJC    , IJC    , IJC    , IJC    , IJC    , IJC    , IJC    ,
+
+/*20*/ 0      , 0      , NFP|VEC|IJC, 0      , 0      , 0      , 0      , 0      ,
+/*28*/ 0      , 0      , 0      , 0      , 0      , 0      , 0      , VEC    ,
+/*30*/ VHC    , VHC    , VHC    , VHC    , VHC    , VHC    , VHC    , VHC    ,
+/*38*/ VHC    , VHC    , 0      , 0      , 0      , 0      , 0      , 0      ,
+
+/*40*/ 0      , VHC    , VHC    , VHC    , VHC    , VHC    , VHC    , 0      ,
+/*48*/ 0      , 0      , 0      , 0      , 0      , 0      , 0      , 0      ,
+/*50*/ 0      , 0      , 0      , 0      , 0      , 0      , 0      , 0      ,
+/*58*/ 0      , 0      , 0      , 0      , NFP|VEC|IJC, 0      , 0      , 0      ,
+
+/*60*/ 0      , VHC    , VEC|VHC, VHC    , VHC    , VHC    , VEC|VHC, 0      ,
+/*68*/ 0      , 0      , 0      , 0      , 0      , 0      , VEC    , 0      ,
+/*70*/ 0      , 0      , VEC    , 0      , VEC    , 0      , 0      , 0      ,
+/*78*/ 0      , 0      , 0      , 0      , 0      , 0      , 0      , 0      ,
+
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    ,
+       NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC    , NUC
+};
+
+/** process a variable length utf8 encoded codepoint.
+ *
+ *  returns:
+ *    yajl_tok_string - if valid utf8 char was parsed and offset was
+ *                      advanced
+ *    yajl_tok_eof - if end of input was hit before validation could
+ *                   complete
+ *    yajl_tok_error - if invalid utf8 was encountered
+ * 
+ *  NOTE: on error the offset will point to the first char of the
+ *  invalid utf8 */
+#define UTF8_CHECK_EOF if (*offset >= jsonTextLen) { return yajl_tok_eof; }
+
+static yajl_tok
+yajl_lex_utf8_char(yajl_lexer lexer, const unsigned char * jsonText,
+                   size_t jsonTextLen, size_t * offset,
+                   unsigned char curChar)
+{
+    if (curChar <= 0x7f) {
+        /* single byte */
+        return yajl_tok_string;
+    } else if ((curChar >> 5) == 0x6) {
+        /* two byte */ 
+        UTF8_CHECK_EOF;
+        curChar = readChar(lexer, jsonText, offset);
+        if ((curChar >> 6) == 0x2) return yajl_tok_string;
+    } else if ((curChar >> 4) == 0x0e) {
+        /* three byte */
+        UTF8_CHECK_EOF;
+        curChar = readChar(lexer, jsonText, offset);
+        if ((curChar >> 6) == 0x2) {
+            UTF8_CHECK_EOF;
+            curChar = readChar(lexer, jsonText, offset);
+            if ((curChar >> 6) == 0x2) return yajl_tok_string;
+        }
+    } else if ((curChar >> 3) == 0x1e) {
+        /* four byte */
+        UTF8_CHECK_EOF;
+        curChar = readChar(lexer, jsonText, offset);
+        if ((curChar >> 6) == 0x2) {
+            UTF8_CHECK_EOF;
+            curChar = readChar(lexer, jsonText, offset);
+            if ((curChar >> 6) == 0x2) {
+                UTF8_CHECK_EOF;
+                curChar = readChar(lexer, jsonText, offset);
+                if ((curChar >> 6) == 0x2) return yajl_tok_string;
+            }
+        }
+    } 
+
+    return yajl_tok_error;
+}
+
+/* lex a string.  input is the lexer, pointer to beginning of
+ * json text, and start of string (offset).
+ * a token is returned which has the following meanings:
+ * yajl_tok_string: lex of string was successful.  offset points to
+ *                  terminating '"'.
+ * yajl_tok_eof: end of text was encountered before we could complete
+ *               the lex.
+ * yajl_tok_error: embedded in the string were unallowable chars.  offset
+ *               points to the offending char
+ */
+#define STR_CHECK_EOF \
+if (*offset >= jsonTextLen) { \
+   tok = yajl_tok_eof; \
+   goto finish_string_lex; \
+}
+
+/** scan a string for interesting characters that might need further
+ *  review.  return the number of chars that are uninteresting and can
+ *  be skipped.
+ * (lth) hi world, any thoughts on how to make this routine faster? */
+static size_t
+yajl_string_scan(const unsigned char * buf, size_t len, int utf8check)
+{
+    unsigned char mask = IJC|NFP|(utf8check ? NUC : 0);
+    size_t skip = 0;
+    while (skip < len && !(charLookupTable[*buf] & mask))
+    {
+        skip++;
+        buf++;
+    }
+    return skip;
+}
+
+static yajl_tok
+yajl_lex_string(yajl_lexer lexer, const unsigned char * jsonText,
+                size_t jsonTextLen, size_t * offset)
+{
+    yajl_tok tok = yajl_tok_error;
+    int hasEscapes = 0;
+
+    for (;;) {
+        unsigned char curChar;
+
+        /* now jump into a faster scanning routine to skip as much
+         * of the buffers as possible */
+        {
+            const unsigned char * p;
+            size_t len;
+            
+            if ((lexer->bufInUse && yajl_buf_len(lexer->buf) &&
+                 lexer->bufOff < yajl_buf_len(lexer->buf)))
+            {
+                p = ((const unsigned char *) yajl_buf_data(lexer->buf) +
+                     (lexer->bufOff));
+                len = yajl_buf_len(lexer->buf) - lexer->bufOff;
+                lexer->bufOff += yajl_string_scan(p, len, lexer->validateUTF8);
+            }                
+            else if (*offset < jsonTextLen) 
+            {
+                p = jsonText + *offset;
+                len = jsonTextLen - *offset;
+                *offset += yajl_string_scan(p, len, lexer->validateUTF8);
+            }
+        }
+
+        STR_CHECK_EOF;
+
+        curChar = readChar(lexer, jsonText, offset);
+
+        /* quote terminates */
+        if (curChar == '"') {
+            tok = yajl_tok_string;
+            break;
+        }
+        /* backslash escapes a set of control chars, */
+        else if (curChar == '\\') {
+            hasEscapes = 1;
+            STR_CHECK_EOF;
+
+            /* special case \u */
+            curChar = readChar(lexer, jsonText, offset);
+            if (curChar == 'u') {
+                unsigned int i = 0;
+
+                for (i=0;i<4;i++) {
+                    STR_CHECK_EOF;                
+                    curChar = readChar(lexer, jsonText, offset);                
+                    if (!(charLookupTable[curChar] & VHC)) {
+                        /* back up to offending char */
+                        unreadChar(lexer, offset);
+                        lexer->error = yajl_lex_string_invalid_hex_char;
+                        goto finish_string_lex;
+                    }
+                }
+            } else if (!(charLookupTable[curChar] & VEC)) {
+                /* back up to offending char */
+                unreadChar(lexer, offset);
+                lexer->error = yajl_lex_string_invalid_escaped_char;
+                goto finish_string_lex;                
+            } 
+        }
+        /* when not validating UTF8 it's a simple table lookup to determine
+         * if the present character is invalid */
+        else if(charLookupTable[curChar] & IJC) {
+            /* back up to offending char */
+            unreadChar(lexer, offset);
+            lexer->error = yajl_lex_string_invalid_json_char;
+            goto finish_string_lex;                
+        }
+        /* when in validate UTF8 mode we need to do some extra work */
+        else if (lexer->validateUTF8) {
+            yajl_tok t = yajl_lex_utf8_char(lexer, jsonText, jsonTextLen,
+                                            offset, curChar);
+            
+            if (t == yajl_tok_eof) {
+                tok = yajl_tok_eof;
+                goto finish_string_lex;
+            } else if (t == yajl_tok_error) {
+                lexer->error = yajl_lex_string_invalid_utf8;
+                goto finish_string_lex;
+            } 
+        }
+        /* accept it, and move on */ 
+    }
+  finish_string_lex:
+    /* tell our buddy, the parser, wether he needs to process this string
+     * again */
+    if (hasEscapes && tok == yajl_tok_string) {
+        tok = yajl_tok_string_with_escapes;
+    } 
+
+    return tok;
+}
+
+#define RETURN_IF_EOF if (*offset >= jsonTextLen) return yajl_tok_eof;
+
+static yajl_tok
+yajl_lex_number(yajl_lexer lexer, const unsigned char * jsonText,
+                size_t jsonTextLen, size_t * offset)
+{
+    /** XXX: numbers are the only entities in json that we must lex
+     *       _beyond_ in order to know that they are complete.  There
+     *       is an ambiguous case for integers at EOF. */
+
+    unsigned char c;
+
+    yajl_tok tok = yajl_tok_integer;
+
+    RETURN_IF_EOF;    
+    c = readChar(lexer, jsonText, offset);
+
+    /* optional leading minus */
+    if (c == '-') {
+        RETURN_IF_EOF;    
+        c = readChar(lexer, jsonText, offset); 
+    }
+
+    /* a single zero, or a series of integers */
+    if (c == '0') {
+        RETURN_IF_EOF;    
+        c = readChar(lexer, jsonText, offset); 
+    } else if (c >= '1' && c <= '9') {
+        do {
+            RETURN_IF_EOF;    
+            c = readChar(lexer, jsonText, offset); 
+        } while (c >= '0' && c <= '9');
+    } else {
+        unreadChar(lexer, offset);
+        lexer->error = yajl_lex_missing_integer_after_minus;
+        return yajl_tok_error;
+    }
+
+    /* optional fraction (indicates this is floating point) */
+    if (c == '.') {
+        int numRd = 0;
+        
+        RETURN_IF_EOF;
+        c = readChar(lexer, jsonText, offset); 
+
+        while (c >= '0' && c <= '9') {
+            numRd++;
+            RETURN_IF_EOF;
+            c = readChar(lexer, jsonText, offset); 
+        } 
+
+        if (!numRd) {
+            unreadChar(lexer, offset);
+            lexer->error = yajl_lex_missing_integer_after_decimal;
+            return yajl_tok_error;
+        }
+        tok = yajl_tok_double;
+    }
+
+    /* optional exponent (indicates this is floating point) */
+    if (c == 'e' || c == 'E') {
+        RETURN_IF_EOF;
+        c = readChar(lexer, jsonText, offset); 
+
+        /* optional sign */
+        if (c == '+' || c == '-') {
+            RETURN_IF_EOF;
+            c = readChar(lexer, jsonText, offset); 
+        }
+
+        if (c >= '0' && c <= '9') {
+            do {
+                RETURN_IF_EOF;
+                c = readChar(lexer, jsonText, offset); 
+            } while (c >= '0' && c <= '9');
+        } else {
+            unreadChar(lexer, offset);
+            lexer->error = yajl_lex_missing_integer_after_exponent;
+            return yajl_tok_error;
+        }
+        tok = yajl_tok_double;
+    }
+    
+    /* we always go "one too far" */
+    unreadChar(lexer, offset);
+    
+    return tok;
+}
+
+static yajl_tok
+yajl_lex_comment(yajl_lexer lexer, const unsigned char * jsonText,
+                 size_t jsonTextLen, size_t * offset)
+{
+    unsigned char c;
+
+    yajl_tok tok = yajl_tok_comment;
+
+    RETURN_IF_EOF;    
+    c = readChar(lexer, jsonText, offset);
+
+    /* either slash or star expected */
+    if (c == '/') {
+        /* now we throw away until end of line */
+        do {
+            RETURN_IF_EOF;    
+            c = readChar(lexer, jsonText, offset); 
+        } while (c != '\n');
+    } else if (c == '*') {
+        /* now we throw away until end of comment */        
+        for (;;) {
+            RETURN_IF_EOF;    
+            c = readChar(lexer, jsonText, offset); 
+            if (c == '*') {
+                RETURN_IF_EOF;    
+                c = readChar(lexer, jsonText, offset);                 
+                if (c == '/') {
+                    break;
+                } else {
+                    unreadChar(lexer, offset);
+                }
+            }
+        }
+    } else {
+        lexer->error = yajl_lex_invalid_char;
+        tok = yajl_tok_error;
+    }
+    
+    return tok;
+}
+
+yajl_tok
+yajl_lex_lex(yajl_lexer lexer, const unsigned char * jsonText,
+             size_t jsonTextLen, size_t * offset,
+             const unsigned char ** outBuf, size_t * outLen)
+{
+    yajl_tok tok = yajl_tok_error;
+    unsigned char c;
+    size_t startOffset = *offset;
+
+    *outBuf = NULL;
+    *outLen = 0;
+
+    for (;;) {
+        assert(*offset <= jsonTextLen);
+
+        if (*offset >= jsonTextLen) {
+            tok = yajl_tok_eof;
+            goto lexed;
+        }
+
+        c = readChar(lexer, jsonText, offset);
+
+        switch (c) {
+            case '{':
+                tok = yajl_tok_left_bracket;
+                goto lexed;
+            case '}':
+                tok = yajl_tok_right_bracket;
+                goto lexed;
+            case '[':
+                tok = yajl_tok_left_brace;
+                goto lexed;
+            case ']':
+                tok = yajl_tok_right_brace;
+                goto lexed;
+            case ',':
+                tok = yajl_tok_comma;
+                goto lexed;
+            case ':':
+                tok = yajl_tok_colon;
+                goto lexed;
+            case '\t': case '\n': case '\v': case '\f': case '\r': case ' ':
+                startOffset++;
+                break;
+            case 't': {
+                const char * want = "rue";
+                do {
+                    if (*offset >= jsonTextLen) {
+                        tok = yajl_tok_eof;
+                        goto lexed;
+                    }
+                    c = readChar(lexer, jsonText, offset);
+                    if (c != *want) {
+                        unreadChar(lexer, offset);
+                        lexer->error = yajl_lex_invalid_string;
+                        tok = yajl_tok_error;
+                        goto lexed;
+                    }
+                } while (*(++want));
+                tok = yajl_tok_bool;
+                goto lexed;
+            }
+            case 'f': {
+                const char * want = "alse";
+                do {
+                    if (*offset >= jsonTextLen) {
+                        tok = yajl_tok_eof;
+                        goto lexed;
+                    }
+                    c = readChar(lexer, jsonText, offset);
+                    if (c != *want) {
+                        unreadChar(lexer, offset);
+                        lexer->error = yajl_lex_invalid_string;
+                        tok = yajl_tok_error;
+                        goto lexed;
+                    }
+                } while (*(++want));
+                tok = yajl_tok_bool;
+                goto lexed;
+            }
+            case 'n': {
+                const char * want = "ull";
+                do {
+                    if (*offset >= jsonTextLen) {
+                        tok = yajl_tok_eof;
+                        goto lexed;
+                    }
+                    c = readChar(lexer, jsonText, offset);
+                    if (c != *want) {
+                        unreadChar(lexer, offset);
+                        lexer->error = yajl_lex_invalid_string;
+                        tok = yajl_tok_error;
+                        goto lexed;
+                    }
+                } while (*(++want));
+                tok = yajl_tok_null;
+                goto lexed;
+            }
+            case '"': {
+                tok = yajl_lex_string(lexer, (const unsigned char *) jsonText,
+                                      jsonTextLen, offset);
+                goto lexed;
+            }
+            case '-':
+            case '0': case '1': case '2': case '3': case '4': 
+            case '5': case '6': case '7': case '8': case '9': {
+                /* integer parsing wants to start from the beginning */
+                unreadChar(lexer, offset);
+                tok = yajl_lex_number(lexer, (const unsigned char *) jsonText,
+                                      jsonTextLen, offset);
+                goto lexed;
+            }
+            case '/':
+                /* hey, look, a probable comment!  If comments are disabled
+                 * it's an error. */
+                if (!lexer->allowComments) {
+                    unreadChar(lexer, offset);
+                    lexer->error = yajl_lex_unallowed_comment;
+                    tok = yajl_tok_error;
+                    goto lexed;
+                }
+                /* if comments are enabled, then we should try to lex
+                 * the thing.  possible outcomes are
+                 * - successful lex (tok_comment, which means continue),
+                 * - malformed comment opening (slash not followed by
+                 *   '*' or '/') (tok_error)
+                 * - eof hit. (tok_eof) */
+                tok = yajl_lex_comment(lexer, (const unsigned char *) jsonText,
+                                       jsonTextLen, offset);
+                if (tok == yajl_tok_comment) {
+                    /* "error" is silly, but that's the initial
+                     * state of tok.  guilty until proven innocent. */  
+                    tok = yajl_tok_error;
+                    yajl_buf_clear(lexer->buf);
+                    lexer->bufInUse = 0;
+                    startOffset = *offset; 
+                    break;
+                }
+                /* hit error or eof, bail */
+                goto lexed;
+            default:
+                lexer->error = yajl_lex_invalid_char;
+                tok = yajl_tok_error;
+                goto lexed;
+        }
+    }
+
+
+  lexed:
+    /* need to append to buffer if the buffer is in use or
+     * if it's an EOF token */
+    if (tok == yajl_tok_eof || lexer->bufInUse) {
+        if (!lexer->bufInUse) yajl_buf_clear(lexer->buf);
+        lexer->bufInUse = 1;
+        yajl_buf_append(lexer->buf, jsonText + startOffset, *offset - startOffset);
+        lexer->bufOff = 0;
+        
+        if (tok != yajl_tok_eof) {
+            *outBuf = yajl_buf_data(lexer->buf);
+            *outLen = yajl_buf_len(lexer->buf);
+            lexer->bufInUse = 0;
+        }
+    } else if (tok != yajl_tok_error) {
+        *outBuf = jsonText + startOffset;
+        *outLen = *offset - startOffset;
+    }
+
+    /* special case for strings. skip the quotes. */
+    if (tok == yajl_tok_string || tok == yajl_tok_string_with_escapes)
+    {
+        assert(*outLen >= 2);
+        (*outBuf)++;
+        *outLen -= 2; 
+    }
+
+
+#ifdef YAJL_LEXER_DEBUG
+    if (tok == yajl_tok_error) {
+        printf("lexical error: %s\n",
+               yajl_lex_error_to_string(yajl_lex_get_error(lexer)));
+    } else if (tok == yajl_tok_eof) {
+        printf("EOF hit\n");
+    } else {
+        printf("lexed %s: '", tokToStr(tok));
+        fwrite(*outBuf, 1, *outLen, stdout);
+        printf("'\n");
+    }
+#endif
+
+    return tok;
+}
+
+const char *
+yajl_lex_error_to_string(yajl_lex_error error)
+{
+    switch (error) {
+        case yajl_lex_e_ok:
+            return "ok, no error";
+        case yajl_lex_string_invalid_utf8:
+            return "invalid bytes in UTF8 string.";
+        case yajl_lex_string_invalid_escaped_char:
+            return "inside a string, '\\' occurs before a character "
+                   "which it may not.";
+        case yajl_lex_string_invalid_json_char:            
+            return "invalid character inside string.";
+        case yajl_lex_string_invalid_hex_char:
+            return "invalid (non-hex) character occurs after '\\u' inside "
+                   "string.";
+        case yajl_lex_invalid_char:
+            return "invalid char in json text.";
+        case yajl_lex_invalid_string:
+            return "invalid string in json text.";
+        case yajl_lex_missing_integer_after_exponent:
+            return "malformed number, a digit is required after the exponent.";
+        case yajl_lex_missing_integer_after_decimal:
+            return "malformed number, a digit is required after the "
+                   "decimal point.";
+        case yajl_lex_missing_integer_after_minus:
+            return "malformed number, a digit is required after the "
+                   "minus sign.";
+        case yajl_lex_unallowed_comment:
+            return "probable comment found in input text, comments are "
+                   "not enabled.";
+    }
+    return "unknown error code";
+}
+
+
+/** allows access to more specific information about the lexical
+ *  error when yajl_lex_lex returns yajl_tok_error. */
+yajl_lex_error
+yajl_lex_get_error(yajl_lexer lexer)
+{
+    if (lexer == NULL) return (yajl_lex_error) -1;
+    return lexer->error;
+}
+
+size_t yajl_lex_current_line(yajl_lexer lexer)
+{
+    return lexer->lineOff;
+}
+
+size_t yajl_lex_current_char(yajl_lexer lexer)
+{
+    return lexer->charOff;
+}
+
+yajl_tok yajl_lex_peek(yajl_lexer lexer, const unsigned char * jsonText,
+                       size_t jsonTextLen, size_t offset)
+{
+    const unsigned char * outBuf;
+    size_t outLen;
+    size_t bufLen = yajl_buf_len(lexer->buf);
+    size_t bufOff = lexer->bufOff;
+    unsigned int bufInUse = lexer->bufInUse;
+    yajl_tok tok;
+    
+    tok = yajl_lex_lex(lexer, jsonText, jsonTextLen, &offset,
+                       &outBuf, &outLen);
+
+    lexer->bufOff = bufOff;
+    lexer->bufInUse = bufInUse;
+    yajl_buf_truncate(lexer->buf, bufLen);
+    
+    return tok;
+}
diff --git a/xlators/cluster/nsr-server/src/yajl_lex.h b/xlators/cluster/nsr-server/src/yajl_lex.h
new file mode 100644
index 000000000..cbaae0c13
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_lex.h
@@ -0,0 +1,117 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#ifndef __YAJL_LEX_H__
+#define __YAJL_LEX_H__
+
+#include "yajl/yajl_common.h"
+
+typedef enum {
+    yajl_tok_bool,         
+    yajl_tok_colon,
+    yajl_tok_comma,     
+    yajl_tok_eof,
+    yajl_tok_error,
+    yajl_tok_left_brace,     
+    yajl_tok_left_bracket,
+    yajl_tok_null,         
+    yajl_tok_right_brace,     
+    yajl_tok_right_bracket,
+
+    /* we differentiate between integers and doubles to allow the
+     * parser to interpret the number without re-scanning */
+    yajl_tok_integer, 
+    yajl_tok_double, 
+
+    /* we differentiate between strings which require further processing,
+     * and strings that do not */
+    yajl_tok_string,
+    yajl_tok_string_with_escapes,
+
+    /* comment tokens are not currently returned to the parser, ever */
+    yajl_tok_comment
+} yajl_tok;
+
+typedef struct yajl_lexer_t * yajl_lexer;
+
+yajl_lexer yajl_lex_alloc(yajl_alloc_funcs * alloc,
+                          unsigned int allowComments,
+                          unsigned int validateUTF8);
+
+void yajl_lex_free(yajl_lexer lexer);
+
+/**
+ * run/continue a lex. "offset" is an input/output parameter.
+ * It should be initialized to zero for a
+ * new chunk of target text, and upon subsetquent calls with the same
+ * target text should passed with the value of the previous invocation.
+ *
+ * the client may be interested in the value of offset when an error is
+ * returned from the lexer.  This allows the client to render useful
+n * error messages.
+ *
+ * When you pass the next chunk of data, context should be reinitialized
+ * to zero.
+ * 
+ * Finally, the output buffer is usually just a pointer into the jsonText,
+ * however in cases where the entity being lexed spans multiple chunks,
+ * the lexer will buffer the entity and the data returned will be
+ * a pointer into that buffer.
+ *
+ * This behavior is abstracted from client code except for the performance
+ * implications which require that the client choose a reasonable chunk
+ * size to get adequate performance.
+ */
+yajl_tok yajl_lex_lex(yajl_lexer lexer, const unsigned char * jsonText,
+                      size_t jsonTextLen, size_t * offset,
+                      const unsigned char ** outBuf, size_t * outLen);
+
+/** have a peek at the next token, but don't move the lexer forward */
+yajl_tok yajl_lex_peek(yajl_lexer lexer, const unsigned char * jsonText,
+                       size_t jsonTextLen, size_t offset);
+
+
+typedef enum {
+    yajl_lex_e_ok = 0,
+    yajl_lex_string_invalid_utf8,
+    yajl_lex_string_invalid_escaped_char,
+    yajl_lex_string_invalid_json_char,
+    yajl_lex_string_invalid_hex_char,
+    yajl_lex_invalid_char,
+    yajl_lex_invalid_string,
+    yajl_lex_missing_integer_after_decimal,
+    yajl_lex_missing_integer_after_exponent,
+    yajl_lex_missing_integer_after_minus,
+    yajl_lex_unallowed_comment
+} yajl_lex_error;
+
+const char * yajl_lex_error_to_string(yajl_lex_error error);
+
+/** allows access to more specific information about the lexical
+ *  error when yajl_lex_lex returns yajl_tok_error. */
+yajl_lex_error yajl_lex_get_error(yajl_lexer lexer);
+
+/** get the current offset into the most recently lexed json string. */
+size_t yajl_lex_current_offset(yajl_lexer lexer);
+
+/** get the number of lines lexed by this lexer instance */
+size_t yajl_lex_current_line(yajl_lexer lexer);
+
+/** get the number of chars lexed by this lexer instance since the last
+ *  \n or \r */
+size_t yajl_lex_current_char(yajl_lexer lexer);
+
+#endif
diff --git a/xlators/cluster/nsr-server/src/yajl_parser.c b/xlators/cluster/nsr-server/src/yajl_parser.c
new file mode 100644
index 000000000..bf9ef24ef
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_parser.c
@@ -0,0 +1,492 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "yajl/yajl_parse.h"
+#include "yajl_lex.h"
+#include "yajl_parser.h"
+#include "yajl_encode.h"
+#include "yajl_bytestack.h"
+
+#include <stdlib.h>
+#include <limits.h>
+#include <errno.h>
+#include <stdio.h>
+#include <string.h>
+#include <ctype.h>
+#include <assert.h>
+#include <math.h>
+
+#define MAX_VALUE_TO_MULTIPLY ((LLONG_MAX / 10) + (LLONG_MAX % 10))
+
+ /* same semantics as strtol */
+long long
+yajl_parse_integer(const unsigned char *number, unsigned int length)
+{
+    long long ret  = 0;
+    long sign = 1;
+    const unsigned char *pos = number;
+    if (*pos == '-') { pos++; sign = -1; }
+    if (*pos == '+') { pos++; }
+
+    while (pos < number + length) {
+        if ( ret > MAX_VALUE_TO_MULTIPLY ) {
+            errno = ERANGE;
+            return sign == 1 ? LLONG_MAX : LLONG_MIN;
+        }
+        ret *= 10;
+        if (LLONG_MAX - ret < (*pos - '0')) {
+            errno = ERANGE;
+            return sign == 1 ? LLONG_MAX : LLONG_MIN;
+        }
+        ret += (*pos++ - '0');
+    }
+
+    return sign * ret;
+}
+
+unsigned char *
+yajl_render_error_string(yajl_handle hand, const unsigned char * jsonText,
+                         size_t jsonTextLen, int verbose)
+{
+    size_t offset = hand->bytesConsumed;
+    unsigned char * str;
+    const char * errorType = NULL;
+    const char * errorText = NULL;
+    char text[72];
+    const char * arrow = "                     (right here) ------^\n";
+
+    if (yajl_bs_current(hand->stateStack) == yajl_state_parse_error) {
+        errorType = "parse";
+        errorText = hand->parseError;
+    } else if (yajl_bs_current(hand->stateStack) == yajl_state_lexical_error) {
+        errorType = "lexical";
+        errorText = yajl_lex_error_to_string(yajl_lex_get_error(hand->lexer));
+    } else {
+        errorType = "unknown";
+    }
+
+    {
+        size_t memneeded = 0;
+        memneeded += strlen(errorType);
+        memneeded += strlen(" error");
+        if (errorText != NULL) {
+            memneeded += strlen(": ");
+            memneeded += strlen(errorText);
+        }
+        str = (unsigned char *) YA_MALLOC(&(hand->alloc), memneeded + 2);
+        if (!str) return NULL;
+        str[0] = 0;
+        strcat((char *) str, errorType);
+        strcat((char *) str, " error");
+        if (errorText != NULL) {
+            strcat((char *) str, ": ");
+            strcat((char *) str, errorText);
+        }
+        strcat((char *) str, "\n");
+    }
+
+    /* now we append as many spaces as needed to make sure the error
+     * falls at char 41, if verbose was specified */
+    if (verbose) {
+        size_t start, end, i;
+        size_t spacesNeeded;
+
+        spacesNeeded = (offset < 30 ? 40 - offset : 10);
+        start = (offset >= 30 ? offset - 30 : 0);
+        end = (offset + 30 > jsonTextLen ? jsonTextLen : offset + 30);
+
+        for (i=0;i<spacesNeeded;i++) text[i] = ' ';
+
+        for (;start < end;start++, i++) {
+            if (jsonText[start] != '\n' && jsonText[start] != '\r')
+            {
+                text[i] = jsonText[start];
+            }
+            else
+            {
+                text[i] = ' ';
+            }
+        }
+        assert(i <= 71);
+        text[i++] = '\n';
+        text[i] = 0;
+        {
+            char * newStr = (char *)
+                YA_MALLOC(&(hand->alloc), (unsigned int)(strlen((char *) str) +
+                                                         strlen((char *) text) +
+                                                         strlen(arrow) + 1));
+            if (newStr) {
+                newStr[0] = 0;
+                strcat((char *) newStr, (char *) str);
+                strcat((char *) newStr, text);
+                strcat((char *) newStr, arrow);
+            }
+            YA_FREE(&(hand->alloc), str);
+            str = (unsigned char *) newStr;
+        }
+    }
+    return str;
+}
+
+/* check for client cancelation */
+#define _CC_CHK(x)                                                \
+    if (!(x)) {                                                   \
+        yajl_bs_set(hand->stateStack, yajl_state_parse_error);    \
+        hand->parseError =                                        \
+            "client cancelled parse via callback return value";   \
+        return yajl_status_client_canceled;                       \
+    }
+
+
+yajl_status
+yajl_do_finish(yajl_handle hand)
+{
+    yajl_status stat;
+    stat = yajl_do_parse(hand,(const unsigned char *) " ",1);
+
+    if (stat != yajl_status_ok) return stat;
+
+    switch(yajl_bs_current(hand->stateStack))
+    {
+        case yajl_state_parse_error:
+        case yajl_state_lexical_error:
+            return yajl_status_error;
+        case yajl_state_got_value:
+        case yajl_state_parse_complete:
+            return yajl_status_ok;
+        default:
+            if (!(hand->flags & yajl_allow_partial_values))
+            {
+                yajl_bs_set(hand->stateStack, yajl_state_parse_error);
+                hand->parseError = "premature EOF";
+                return yajl_status_error;
+            }
+            return yajl_status_ok;
+    }
+}
+
+yajl_status
+yajl_do_parse(yajl_handle hand, const unsigned char * jsonText,
+              size_t jsonTextLen)
+{
+    yajl_tok tok;
+    const unsigned char * buf;
+    size_t bufLen;
+    size_t * offset = &(hand->bytesConsumed);
+
+    *offset = 0;
+
+  around_again:
+    switch (yajl_bs_current(hand->stateStack)) {
+        case yajl_state_parse_complete:
+            if (hand->flags & yajl_allow_multiple_values) {
+                yajl_bs_set(hand->stateStack, yajl_state_got_value);
+                goto around_again;
+            }
+            if (!(hand->flags & yajl_allow_trailing_garbage)) {
+                if (*offset != jsonTextLen) {
+                    tok = yajl_lex_lex(hand->lexer, jsonText, jsonTextLen,
+                                       offset, &buf, &bufLen);
+                    if (tok != yajl_tok_eof) {
+                        yajl_bs_set(hand->stateStack, yajl_state_parse_error);
+                        hand->parseError = "trailing garbage";
+                    }
+                    goto around_again;
+                }
+            }
+            return yajl_status_ok;
+        case yajl_state_lexical_error:
+        case yajl_state_parse_error:
+            return yajl_status_error;
+        case yajl_state_start:
+        case yajl_state_got_value:
+        case yajl_state_map_need_val:
+        case yajl_state_array_need_val:
+        case yajl_state_array_start:  {
+            /* for arrays and maps, we advance the state for this
+             * depth, then push the state of the next depth.
+             * If an error occurs during the parsing of the nesting
+             * enitity, the state at this level will not matter.
+             * a state that needs pushing will be anything other
+             * than state_start */
+
+            yajl_state stateToPush = yajl_state_start;
+
+            tok = yajl_lex_lex(hand->lexer, jsonText, jsonTextLen,
+                               offset, &buf, &bufLen);
+
+            switch (tok) {
+                case yajl_tok_eof:
+                    return yajl_status_ok;
+                case yajl_tok_error:
+                    yajl_bs_set(hand->stateStack, yajl_state_lexical_error);
+                    goto around_again;
+                case yajl_tok_string:
+                    if (hand->callbacks && hand->callbacks->yajl_string) {
+                        _CC_CHK(hand->callbacks->yajl_string(hand->ctx,
+                                                             buf, bufLen));
+                    }
+                    break;
+                case yajl_tok_string_with_escapes:
+                    if (hand->callbacks && hand->callbacks->yajl_string) {
+                        yajl_buf_clear(hand->decodeBuf);
+                        yajl_string_decode(hand->decodeBuf, buf, bufLen);
+                        _CC_CHK(hand->callbacks->yajl_string(
+                                    hand->ctx, yajl_buf_data(hand->decodeBuf),
+                                    yajl_buf_len(hand->decodeBuf)));
+                    }
+                    break;
+                case yajl_tok_bool:
+                    if (hand->callbacks && hand->callbacks->yajl_boolean) {
+                        _CC_CHK(hand->callbacks->yajl_boolean(hand->ctx,
+                                                              *buf == 't'));
+                    }
+                    break;
+                case yajl_tok_null:
+                    if (hand->callbacks && hand->callbacks->yajl_null) {
+                        _CC_CHK(hand->callbacks->yajl_null(hand->ctx));
+                    }
+                    break;
+                case yajl_tok_left_bracket:
+                    if (hand->callbacks && hand->callbacks->yajl_start_map) {
+                        _CC_CHK(hand->callbacks->yajl_start_map(hand->ctx));
+                    }
+                    stateToPush = yajl_state_map_start;
+                    break;
+                case yajl_tok_left_brace:
+                    if (hand->callbacks && hand->callbacks->yajl_start_array) {
+                        _CC_CHK(hand->callbacks->yajl_start_array(hand->ctx));
+                    }
+                    stateToPush = yajl_state_array_start;
+                    break;
+                case yajl_tok_integer:
+                    if (hand->callbacks) {
+                        if (hand->callbacks->yajl_number) {
+                            _CC_CHK(hand->callbacks->yajl_number(
+                                        hand->ctx,(const char *) buf, bufLen));
+                        } else if (hand->callbacks->yajl_integer) {
+                            long long int i = 0;
+                            i = yajl_parse_integer(buf, bufLen);
+                            if ((i == LLONG_MIN || i == LLONG_MAX) &&
+                                errno == ERANGE)
+                            {
+                                yajl_bs_set(hand->stateStack,
+                                            yajl_state_parse_error);
+                                hand->parseError = "integer overflow" ;
+                                /* try to restore error offset */
+                                if (*offset >= bufLen) *offset -= bufLen;
+                                else *offset = 0;
+                                goto around_again;
+                            }
+                            _CC_CHK(hand->callbacks->yajl_integer(hand->ctx,
+                                                                  i));
+                        }
+                    }
+                    break;
+                case yajl_tok_double:
+                    if (hand->callbacks) {
+                        if (hand->callbacks->yajl_number) {
+                            _CC_CHK(hand->callbacks->yajl_number(
+                                        hand->ctx, (const char *) buf, bufLen));
+                        } else if (hand->callbacks->yajl_double) {
+                            double d = 0.0;
+                            yajl_buf_clear(hand->decodeBuf);
+                            yajl_buf_append(hand->decodeBuf, buf, bufLen);
+                            buf = yajl_buf_data(hand->decodeBuf);
+                            d = strtod((char *) buf, NULL);
+                            if ((d == HUGE_VAL || d == -HUGE_VAL) &&
+                                errno == ERANGE)
+                            {
+                                yajl_bs_set(hand->stateStack,
+                                            yajl_state_parse_error);
+                                hand->parseError = "numeric (floating point) "
+                                    "overflow";
+                                /* try to restore error offset */
+                                if (*offset >= bufLen) *offset -= bufLen;
+                                else *offset = 0;
+                                goto around_again;
+                            }
+                            _CC_CHK(hand->callbacks->yajl_double(hand->ctx,
+                                                                 d));
+                        }
+                    }
+                    break;
+                case yajl_tok_right_brace: {
+                    if (yajl_bs_current(hand->stateStack) ==
+                        yajl_state_array_start)
+                    {
+                        if (hand->callbacks &&
+                            hand->callbacks->yajl_end_array)
+                        {
+                            _CC_CHK(hand->callbacks->yajl_end_array(hand->ctx));
+                        }
+                        yajl_bs_pop(hand->stateStack);
+                        goto around_again;
+                    }
+                    /* intentional fall-through */
+                }
+                case yajl_tok_colon:
+                case yajl_tok_comma:
+                case yajl_tok_right_bracket:
+                    yajl_bs_set(hand->stateStack, yajl_state_parse_error);
+                    hand->parseError =
+                        "unallowed token at this point in JSON text";
+                    goto around_again;
+                default:
+                    yajl_bs_set(hand->stateStack, yajl_state_parse_error);
+                    hand->parseError = "invalid token, internal error";
+                    goto around_again;
+            }
+            /* got a value.  transition depends on the state we're in. */
+            {
+                yajl_state s = yajl_bs_current(hand->stateStack);
+                if (s == yajl_state_start || s == yajl_state_got_value) {
+                    yajl_bs_set(hand->stateStack, yajl_state_parse_complete);
+                } else if (s == yajl_state_map_need_val) {
+                    yajl_bs_set(hand->stateStack, yajl_state_map_got_val);
+                } else {
+                    yajl_bs_set(hand->stateStack, yajl_state_array_got_val);
+                }
+            }
+            if (stateToPush != yajl_state_start) {
+                yajl_bs_push(hand->stateStack, stateToPush);
+            }
+
+            goto around_again;
+        }
+        case yajl_state_map_start:
+        case yajl_state_map_need_key: {
+            /* only difference between these two states is that in
+             * start '}' is valid, whereas in need_key, we've parsed
+             * a comma, and a string key _must_ follow */
+            tok = yajl_lex_lex(hand->lexer, jsonText, jsonTextLen,
+                               offset, &buf, &bufLen);
+            switch (tok) {
+                case yajl_tok_eof:
+                    return yajl_status_ok;
+                case yajl_tok_error:
+                    yajl_bs_set(hand->stateStack, yajl_state_lexical_error);
+                    goto around_again;
+                case yajl_tok_string_with_escapes:
+                    if (hand->callbacks && hand->callbacks->yajl_map_key) {
+                        yajl_buf_clear(hand->decodeBuf);
+                        yajl_string_decode(hand->decodeBuf, buf, bufLen);
+                        buf = yajl_buf_data(hand->decodeBuf);
+                        bufLen = yajl_buf_len(hand->decodeBuf);
+                    }
+                    /* intentional fall-through */
+                case yajl_tok_string:
+                    if (hand->callbacks && hand->callbacks->yajl_map_key) {
+                        _CC_CHK(hand->callbacks->yajl_map_key(hand->ctx, buf,
+                                                              bufLen));
+                    }
+                    yajl_bs_set(hand->stateStack, yajl_state_map_sep);
+                    goto around_again;
+                case yajl_tok_right_bracket:
+                    if (yajl_bs_current(hand->stateStack) ==
+                        yajl_state_map_start)
+                    {
+                        if (hand->callbacks && hand->callbacks->yajl_end_map) {
+                            _CC_CHK(hand->callbacks->yajl_end_map(hand->ctx));
+                        }
+                        yajl_bs_pop(hand->stateStack);
+                        goto around_again;
+                    }
+                default:
+                    yajl_bs_set(hand->stateStack, yajl_state_parse_error);
+                    hand->parseError =
+                        "invalid object key (must be a string)"; 
+                    goto around_again;
+            }
+        }
+        case yajl_state_map_sep: {
+            tok = yajl_lex_lex(hand->lexer, jsonText, jsonTextLen,
+                               offset, &buf, &bufLen);
+            switch (tok) {
+                case yajl_tok_colon:
+                    yajl_bs_set(hand->stateStack, yajl_state_map_need_val);
+                    goto around_again;
+                case yajl_tok_eof:
+                    return yajl_status_ok;
+                case yajl_tok_error:
+                    yajl_bs_set(hand->stateStack, yajl_state_lexical_error);
+                    goto around_again;
+                default:
+                    yajl_bs_set(hand->stateStack, yajl_state_parse_error);
+                    hand->parseError = "object key and value must "
+                        "be separated by a colon (':')";
+                    goto around_again;
+            }
+        }
+        case yajl_state_map_got_val: {
+            tok = yajl_lex_lex(hand->lexer, jsonText, jsonTextLen,
+                               offset, &buf, &bufLen);
+            switch (tok) {
+                case yajl_tok_right_bracket:
+                    if (hand->callbacks && hand->callbacks->yajl_end_map) {
+                        _CC_CHK(hand->callbacks->yajl_end_map(hand->ctx));
+                    }
+                    yajl_bs_pop(hand->stateStack);
+                    goto around_again;
+                case yajl_tok_comma:
+                    yajl_bs_set(hand->stateStack, yajl_state_map_need_key);
+                    goto around_again;
+                case yajl_tok_eof:
+                    return yajl_status_ok;
+                case yajl_tok_error:
+                    yajl_bs_set(hand->stateStack, yajl_state_lexical_error);
+                    goto around_again;
+                default:
+                    yajl_bs_set(hand->stateStack, yajl_state_parse_error);
+                    hand->parseError = "after key and value, inside map, "
+                                       "I expect ',' or '}'";
+                    /* try to restore error offset */
+                    if (*offset >= bufLen) *offset -= bufLen;
+                    else *offset = 0;
+                    goto around_again;
+            }
+        }
+        case yajl_state_array_got_val: {
+            tok = yajl_lex_lex(hand->lexer, jsonText, jsonTextLen,
+                               offset, &buf, &bufLen);
+            switch (tok) {
+                case yajl_tok_right_brace:
+                    if (hand->callbacks && hand->callbacks->yajl_end_array) {
+                        _CC_CHK(hand->callbacks->yajl_end_array(hand->ctx));
+                    }
+                    yajl_bs_pop(hand->stateStack);
+                    goto around_again;
+                case yajl_tok_comma:
+                    yajl_bs_set(hand->stateStack, yajl_state_array_need_val);
+                    goto around_again;
+                case yajl_tok_eof:
+                    return yajl_status_ok;
+                case yajl_tok_error:
+                    yajl_bs_set(hand->stateStack, yajl_state_lexical_error);
+                    goto around_again;
+                default:
+                    yajl_bs_set(hand->stateStack, yajl_state_parse_error);
+                    hand->parseError =
+                        "after array element, I expect ',' or ']'";
+                    goto around_again;
+            }
+        }
+    }
+
+    abort();
+    return yajl_status_error;
+}
+
diff --git a/xlators/cluster/nsr-server/src/yajl_parser.h b/xlators/cluster/nsr-server/src/yajl_parser.h
new file mode 100644
index 000000000..53409731a
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_parser.h
@@ -0,0 +1,78 @@
+/*
+ * Copyright (c) 2007-2011, Lloyd Hilaiel <lloyd@hilaiel.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#ifndef __YAJL_PARSER_H__
+#define __YAJL_PARSER_H__
+
+#include "yajl/yajl_parse.h"
+#include "yajl_bytestack.h"
+#include "yajl_buf.h"
+#include "yajl_lex.h"
+
+
+typedef enum {
+    yajl_state_start = 0,
+    yajl_state_parse_complete,
+    yajl_state_parse_error,
+    yajl_state_lexical_error,
+    yajl_state_map_start,
+    yajl_state_map_sep,
+    yajl_state_map_need_val,
+    yajl_state_map_got_val,
+    yajl_state_map_need_key,
+    yajl_state_array_start,
+    yajl_state_array_got_val,
+    yajl_state_array_need_val,
+    yajl_state_got_value,
+} yajl_state;
+
+struct yajl_handle_t {
+    const yajl_callbacks * callbacks;
+    void * ctx;
+    yajl_lexer lexer;
+    const char * parseError;
+    /* the number of bytes consumed from the last client buffer,
+     * in the case of an error this will be an error offset, in the
+     * case of an error this can be used as the error offset */
+    size_t bytesConsumed;
+    /* temporary storage for decoded strings */
+    yajl_buf decodeBuf;
+    /* a stack of states.  access with yajl_state_XXX routines */
+    yajl_bytestack stateStack;
+    /* memory allocation routines */
+    yajl_alloc_funcs alloc;
+    /* bitfield */
+    unsigned int flags;
+};
+
+yajl_status
+yajl_do_parse(yajl_handle handle, const unsigned char * jsonText,
+              size_t jsonTextLen);
+
+yajl_status
+yajl_do_finish(yajl_handle handle);
+
+unsigned char *
+yajl_render_error_string(yajl_handle hand, const unsigned char * jsonText,
+                         size_t jsonTextLen, int verbose);
+
+/* A little built in integer parsing routine with the same semantics as strtol
+ * that's unaffected by LOCALE. */
+long long
+yajl_parse_integer(const unsigned char *number, unsigned int length);
+
+
+#endif
diff --git a/xlators/cluster/nsr-server/src/yajl_tree.c b/xlators/cluster/nsr-server/src/yajl_tree.c
new file mode 100644
index 000000000..1a69134e7
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_tree.c
@@ -0,0 +1,501 @@
+/*
+ * Copyright (c) 2010-2011  Florian Forster  <ff at octo.it>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <errno.h>
+#include <assert.h>
+
+#include "yajl/yajl_tree.h"
+#include "yajl/yajl_parse.h"
+
+#include "yajl_parser.h"
+
+#ifdef WIN32
+#define snprintf sprintf_s
+#endif
+
+#define STATUS_CONTINUE 1
+#define STATUS_ABORT    0
+
+struct stack_elem_s;
+typedef struct stack_elem_s stack_elem_t;
+struct stack_elem_s
+{
+    char * key;
+    yajl_val value;
+    stack_elem_t *next;
+};
+
+struct context_s
+{
+    stack_elem_t *stack;
+    yajl_val root;
+    char *errbuf;
+    size_t errbuf_size;
+};
+typedef struct context_s context_t;
+
+#define RETURN_ERROR(ctx,retval,...) {                                  \
+        if ((ctx)->errbuf != NULL)                                      \
+            snprintf ((ctx)->errbuf, (ctx)->errbuf_size, __VA_ARGS__);  \
+        return (retval);                                                \
+    }
+
+static yajl_val value_alloc (yajl_type type)
+{
+    yajl_val v;
+
+    v = malloc (sizeof (*v));
+    if (v == NULL) return (NULL);
+    memset (v, 0, sizeof (*v));
+    v->type = type;
+
+    return (v);
+}
+
+static void yajl_object_free (yajl_val v)
+{
+    size_t i;
+
+    if (!YAJL_IS_OBJECT(v)) return;
+
+    for (i = 0; i < v->u.object.len; i++)
+    {
+        free((char *) v->u.object.keys[i]);
+        v->u.object.keys[i] = NULL;
+        yajl_tree_free (v->u.object.values[i]);
+        v->u.object.values[i] = NULL;
+    }
+
+    free((void*) v->u.object.keys);
+    free(v->u.object.values);
+    free(v);
+}
+
+static void yajl_array_free (yajl_val v)
+{
+    size_t i;
+
+    if (!YAJL_IS_ARRAY(v)) return;
+
+    for (i = 0; i < v->u.array.len; i++)
+    {
+        yajl_tree_free (v->u.array.values[i]);
+        v->u.array.values[i] = NULL;
+    }
+
+    free(v->u.array.values);
+    free(v);
+}
+
+/*
+ * Parsing nested objects and arrays is implemented using a stack. When a new
+ * object or array starts (a curly or a square opening bracket is read), an
+ * appropriate value is pushed on the stack. When the end of the object is
+ * reached (an appropriate closing bracket has been read), the value is popped
+ * off the stack and added to the enclosing object using "context_add_value".
+ */
+static int context_push(context_t *ctx, yajl_val v)
+{
+    stack_elem_t *stack;
+
+    stack = malloc (sizeof (*stack));
+    if (stack == NULL)
+        RETURN_ERROR (ctx, ENOMEM, "Out of memory");
+    memset (stack, 0, sizeof (*stack));
+
+    assert ((ctx->stack == NULL)
+            || YAJL_IS_OBJECT (v)
+            || YAJL_IS_ARRAY (v));
+
+    stack->value = v;
+    stack->next = ctx->stack;
+    ctx->stack = stack;
+
+    return (0);
+}
+
+static yajl_val context_pop(context_t *ctx)
+{
+    stack_elem_t *stack;
+    yajl_val v;
+
+    if (ctx->stack == NULL)
+        RETURN_ERROR (ctx, NULL, "context_pop: "
+                      "Bottom of stack reached prematurely");
+
+    stack = ctx->stack;
+    ctx->stack = stack->next;
+
+    v = stack->value;
+
+    free (stack);
+
+    return (v);
+}
+
+static int object_add_keyval(context_t *ctx,
+                             yajl_val obj, char *key, yajl_val value)
+{
+    const char **tmpk;
+    yajl_val *tmpv;
+
+    /* We're checking for NULL in "context_add_value" or its callers. */
+    assert (ctx != NULL);
+    assert (obj != NULL);
+    assert (key != NULL);
+    assert (value != NULL);
+
+    /* We're assuring that "obj" is an object in "context_add_value". */
+    assert(YAJL_IS_OBJECT(obj));
+
+    tmpk = realloc((void *) obj->u.object.keys, sizeof(*(obj->u.object.keys)) * (obj->u.object.len + 1));
+    if (tmpk == NULL)
+        RETURN_ERROR(ctx, ENOMEM, "Out of memory");
+    obj->u.object.keys = tmpk;
+
+    tmpv = realloc(obj->u.object.values, sizeof (*obj->u.object.values) * (obj->u.object.len + 1));
+    if (tmpv == NULL)
+        RETURN_ERROR(ctx, ENOMEM, "Out of memory");
+    obj->u.object.values = tmpv;
+
+    obj->u.object.keys[obj->u.object.len] = key;
+    obj->u.object.values[obj->u.object.len] = value;
+    obj->u.object.len++;
+
+    return (0);
+}
+
+static int array_add_value (context_t *ctx,
+                            yajl_val array, yajl_val value)
+{
+    yajl_val *tmp;
+
+    /* We're checking for NULL pointers in "context_add_value" or its
+     * callers. */
+    assert (ctx != NULL);
+    assert (array != NULL);
+    assert (value != NULL);
+
+    /* "context_add_value" will only call us with array values. */
+    assert(YAJL_IS_ARRAY(array));
+    
+    tmp = realloc(array->u.array.values,
+                  sizeof(*(array->u.array.values)) * (array->u.array.len + 1));
+    if (tmp == NULL)
+        RETURN_ERROR(ctx, ENOMEM, "Out of memory");
+    array->u.array.values = tmp;
+    array->u.array.values[array->u.array.len] = value;
+    array->u.array.len++;
+
+    return 0;
+}
+
+/*
+ * Add a value to the value on top of the stack or the "root" member in the
+ * context if the end of the parsing process is reached.
+ */
+static int context_add_value (context_t *ctx, yajl_val v)
+{
+    /* We're checking for NULL values in all the calling functions. */
+    assert (ctx != NULL);
+    assert (v != NULL);
+
+    /*
+     * There are three valid states in which this function may be called:
+     *   - There is no value on the stack => This is the only value. This is the
+     *     last step done when parsing a document. We assign the value to the
+     *     "root" member and return.
+     *   - The value on the stack is an object. In this case store the key on the
+     *     stack or, if the key has already been read, add key and value to the
+     *     object.
+     *   - The value on the stack is an array. In this case simply add the value
+     *     and return.
+     */
+    if (ctx->stack == NULL)
+    {
+        assert (ctx->root == NULL);
+        ctx->root = v;
+        return (0);
+    }
+    else if (YAJL_IS_OBJECT (ctx->stack->value))
+    {
+        if (ctx->stack->key == NULL)
+        {
+            if (!YAJL_IS_STRING (v))
+                RETURN_ERROR (ctx, EINVAL, "context_add_value: "
+                              "Object key is not a string (%#04x)",
+                              v->type);
+
+            ctx->stack->key = v->u.string;
+            v->u.string = NULL;
+            free(v);
+            return (0);
+        }
+        else /* if (ctx->key != NULL) */
+        {
+            char * key;
+
+            key = ctx->stack->key;
+            ctx->stack->key = NULL;
+            return (object_add_keyval (ctx, ctx->stack->value, key, v));
+        }
+    }
+    else if (YAJL_IS_ARRAY (ctx->stack->value))
+    {
+        return (array_add_value (ctx, ctx->stack->value, v));
+    }
+    else
+    {
+        RETURN_ERROR (ctx, EINVAL, "context_add_value: Cannot add value to "
+                      "a value of type %#04x (not a composite type)",
+                      ctx->stack->value->type);
+    }
+}
+
+static int handle_string (void *ctx,
+                          const unsigned char *string, size_t string_length)
+{
+    yajl_val v;
+
+    v = value_alloc (yajl_t_string);
+    if (v == NULL)
+        RETURN_ERROR ((context_t *) ctx, STATUS_ABORT, "Out of memory");
+
+    v->u.string = malloc (string_length + 1);
+    if (v->u.string == NULL)
+    {
+        free (v);
+        RETURN_ERROR ((context_t *) ctx, STATUS_ABORT, "Out of memory");
+    }
+    memcpy(v->u.string, string, string_length);
+    v->u.string[string_length] = 0;
+
+    return ((context_add_value (ctx, v) == 0) ? STATUS_CONTINUE : STATUS_ABORT);
+}
+
+static int handle_number (void *ctx, const char *string, size_t string_length)
+{
+    yajl_val v;
+    char *endptr;
+
+    v = value_alloc(yajl_t_number);
+    if (v == NULL)
+        RETURN_ERROR((context_t *) ctx, STATUS_ABORT, "Out of memory");
+
+    v->u.number.r = malloc(string_length + 1);
+    if (v->u.number.r == NULL)
+    {
+        free(v);
+        RETURN_ERROR((context_t *) ctx, STATUS_ABORT, "Out of memory");
+    }
+    memcpy(v->u.number.r, string, string_length);
+    v->u.number.r[string_length] = 0;
+
+    v->u.number.flags = 0;
+
+    endptr = NULL;
+    errno = 0;
+    v->u.number.i = yajl_parse_integer((const unsigned char *) v->u.number.r,
+                                       strlen(v->u.number.r));
+    if ((errno == 0) && (endptr != NULL) && (*endptr == 0))
+        v->u.number.flags |= YAJL_NUMBER_INT_VALID;
+
+    endptr = NULL;
+    errno = 0;
+    v->u.number.d = strtod(v->u.number.r, &endptr);
+    if ((errno == 0) && (endptr != NULL) && (*endptr == 0))
+        v->u.number.flags |= YAJL_NUMBER_DOUBLE_VALID;
+
+    return ((context_add_value(ctx, v) == 0) ? STATUS_CONTINUE : STATUS_ABORT);
+}
+
+static int handle_start_map (void *ctx)
+{
+    yajl_val v;
+
+    v = value_alloc(yajl_t_object);
+    if (v == NULL)
+        RETURN_ERROR ((context_t *) ctx, STATUS_ABORT, "Out of memory");
+
+    v->u.object.keys = NULL;
+    v->u.object.values = NULL;
+    v->u.object.len = 0;
+
+    return ((context_push (ctx, v) == 0) ? STATUS_CONTINUE : STATUS_ABORT);
+}
+
+static int handle_end_map (void *ctx)
+{
+    yajl_val v;
+
+    v = context_pop (ctx);
+    if (v == NULL)
+        return (STATUS_ABORT);
+
+    return ((context_add_value (ctx, v) == 0) ? STATUS_CONTINUE : STATUS_ABORT);
+}
+
+static int handle_start_array (void *ctx)
+{
+    yajl_val v;
+
+    v = value_alloc(yajl_t_array);
+    if (v == NULL)
+        RETURN_ERROR ((context_t *) ctx, STATUS_ABORT, "Out of memory");
+
+    v->u.array.values = NULL;
+    v->u.array.len = 0;
+
+    return ((context_push (ctx, v) == 0) ? STATUS_CONTINUE : STATUS_ABORT);
+}
+
+static int handle_end_array (void *ctx)
+{
+    yajl_val v;
+
+    v = context_pop (ctx);
+    if (v == NULL)
+        return (STATUS_ABORT);
+
+    return ((context_add_value (ctx, v) == 0) ? STATUS_CONTINUE : STATUS_ABORT);
+}
+
+static int handle_boolean (void *ctx, int boolean_value)
+{
+    yajl_val v;
+
+    v = value_alloc (boolean_value ? yajl_t_true : yajl_t_false);
+    if (v == NULL)
+        RETURN_ERROR ((context_t *) ctx, STATUS_ABORT, "Out of memory");
+
+    return ((context_add_value (ctx, v) == 0) ? STATUS_CONTINUE : STATUS_ABORT);
+}
+
+static int handle_null (void *ctx)
+{
+    yajl_val v;
+
+    v = value_alloc (yajl_t_null);
+    if (v == NULL)
+        RETURN_ERROR ((context_t *) ctx, STATUS_ABORT, "Out of memory");
+
+    return ((context_add_value (ctx, v) == 0) ? STATUS_CONTINUE : STATUS_ABORT);
+}
+
+/*
+ * Public functions
+ */
+yajl_val yajl_tree_parse (const char *input,
+                          char *error_buffer, size_t error_buffer_size)
+{
+    static const yajl_callbacks callbacks =
+        {
+            /* null        = */ handle_null,
+            /* boolean     = */ handle_boolean,
+            /* integer     = */ NULL,
+            /* double      = */ NULL,
+            /* number      = */ handle_number,
+            /* string      = */ handle_string,
+            /* start map   = */ handle_start_map,
+            /* map key     = */ handle_string,
+            /* end map     = */ handle_end_map,
+            /* start array = */ handle_start_array,
+            /* end array   = */ handle_end_array
+        };
+
+    yajl_handle handle;
+    yajl_status status;
+	context_t ctx = { NULL, NULL, NULL, 0 };
+
+	ctx.errbuf = error_buffer;
+	ctx.errbuf_size = error_buffer_size;
+
+    if (error_buffer != NULL)
+        memset (error_buffer, 0, error_buffer_size);
+
+    handle = yajl_alloc (&callbacks, NULL, &ctx);
+    yajl_config(handle, yajl_allow_comments, 1);
+
+    status = yajl_parse(handle,
+                        (unsigned char *) input,
+                        strlen (input));
+    status = yajl_complete_parse (handle);
+    if (status != yajl_status_ok) {
+        if (error_buffer != NULL && error_buffer_size > 0) {
+            snprintf(
+                error_buffer, error_buffer_size, "%s",
+                (char *) yajl_get_error(handle, 1,
+                                        (const unsigned char *) input,
+                                        strlen(input)));
+        }
+        yajl_free (handle);
+        return NULL;
+    }
+
+    yajl_free (handle);
+    return (ctx.root);
+}
+
+yajl_val yajl_tree_get(yajl_val n, const char ** path, yajl_type type)
+{
+    if (!path) return NULL;
+    while (n && *path) {
+        unsigned int i;
+
+        if (n->type != yajl_t_object) return NULL;
+        for (i = 0; i < n->u.object.len; i++) {
+            if (!strcmp(*path, n->u.object.keys[i])) {
+                n = n->u.object.values[i];
+                break;
+            }
+        }
+        if (i == n->u.object.len) return NULL;
+        path++;
+    }
+    if (n && type != yajl_t_any && type != n->type) n = NULL;
+    return n;
+}
+
+void yajl_tree_free (yajl_val v)
+{
+    if (v == NULL) return;
+
+    if (YAJL_IS_STRING(v))
+    {
+        free(v->u.string);
+        free(v);
+    }
+    else if (YAJL_IS_NUMBER(v))
+    {
+        free(v->u.number.r);
+        free(v);
+    }
+    else if (YAJL_GET_OBJECT(v))
+    {
+        yajl_object_free(v);
+    }
+    else if (YAJL_GET_ARRAY(v))
+    {
+        yajl_array_free(v);
+    }
+    else /* if (yajl_t_true or yajl_t_false or yajl_t_null) */
+    {
+        free(v);
+    }
+}
diff --git a/xlators/cluster/nsr-server/src/yajl_version.c b/xlators/cluster/nsr-server/src/yajl_version.c
new file mode 100644
index 000000000..0671da722
--- /dev/null
+++ b/xlators/cluster/nsr-server/src/yajl_version.c
@@ -0,0 +1,7 @@
+#include <yajl/yajl_version.h>
+
+int yajl_version(void)
+{
+	return YAJL_VERSION;
+}
+
diff --git a/xlators/cluster/stripe/src/Makefile.am b/xlators/cluster/stripe/src/Makefile.am
index 0db3c9eeb..4268d6f03 100644
--- a/xlators/cluster/stripe/src/Makefile.am
+++ b/xlators/cluster/stripe/src/Makefile.am
@@ -1,17 +1,19 @@
-
 xlator_LTLIBRARIES = stripe.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
 
-stripe_la_LDFLAGS = -module -avoidversion
+stripe_la_LDFLAGS = -module -avoid-version
+
+stripe_la_SOURCES = stripe.c stripe-helpers.c \
+	$(top_builddir)/xlators/lib/src/libxlator.c
 
-stripe_la_SOURCES = stripe.c $(top_builddir)/xlators/lib/src/libxlator.c
 stripe_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
 noinst_HEADERS = stripe.h stripe-mem-types.h $(top_builddir)/xlators/lib/src/libxlator.h
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS) \
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
 	-I$(top_srcdir)/xlators/lib/src
 
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
 CLEANFILES = 
 
diff --git a/xlators/cluster/stripe/src/stripe-helpers.c b/xlators/cluster/stripe/src/stripe-helpers.c
new file mode 100644
index 000000000..3c12809d6
--- /dev/null
+++ b/xlators/cluster/stripe/src/stripe-helpers.c
@@ -0,0 +1,677 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <fnmatch.h>
+
+#include "stripe.h"
+#include "byte-order.h"
+#include "mem-types.h"
+#include "logging.h"
+
+void
+stripe_local_wipe (stripe_local_t *local)
+{
+        if (!local)
+                goto out;
+
+        loc_wipe (&local->loc);
+        loc_wipe (&local->loc2);
+
+        if (local->fd)
+                fd_unref (local->fd);
+
+        if (local->inode)
+                inode_unref (local->inode);
+
+        if (local->xattr)
+                dict_unref (local->xattr);
+
+        if (local->xdata)
+                dict_unref (local->xdata);
+
+out:
+        return;
+}
+
+
+
+int
+stripe_aggregate (dict_t *this, char *key, data_t *value, void *data)
+{
+        dict_t  *dst  = NULL;
+        int64_t *ptr  = 0, *size = NULL;
+        int32_t  ret  = -1;
+
+        dst = data;
+
+        if (strcmp (key, GF_XATTR_QUOTA_SIZE_KEY) == 0) {
+                ret = dict_get_bin (dst, key, (void **)&size);
+                if (ret < 0) {
+                        size = GF_CALLOC (1, sizeof (int64_t),
+                                          gf_common_mt_char);
+                        if (size == NULL) {
+                                gf_log ("stripe", GF_LOG_WARNING,
+                                        "memory allocation failed");
+                                goto out;
+                        }
+                        ret = dict_set_bin (dst, key, size, sizeof (int64_t));
+                        if (ret < 0) {
+                                gf_log ("stripe", GF_LOG_WARNING,
+                                        "stripe aggregate dict set failed");
+                                GF_FREE (size);
+                                goto out;
+                        }
+                }
+
+                ptr = data_to_bin (value);
+                if (ptr == NULL) {
+                        gf_log ("stripe", GF_LOG_WARNING, "data to bin failed");
+                        goto out;
+                }
+
+                *size = hton64 (ntoh64 (*size) + ntoh64 (*ptr));
+        } else if (strcmp (key, GF_CONTENT_KEY)) {
+                /* No need to aggregate 'CONTENT' data */
+                ret = dict_set (dst, key, value);
+                if (ret)
+                        gf_log ("stripe", GF_LOG_WARNING, "xattr dict set failed");
+        }
+
+out:
+        return 0;
+}
+
+
+void
+stripe_aggregate_xattr (dict_t *dst, dict_t *src)
+{
+        if ((dst == NULL) || (src == NULL)) {
+                goto out;
+        }
+
+        dict_foreach (src, stripe_aggregate, dst);
+out:
+        return;
+}
+
+
+int32_t
+stripe_xattr_aggregate (char *buffer, stripe_local_t *local, int32_t *total)
+{
+        int32_t              i     = 0;
+        int32_t              ret   = -1;
+        int32_t              len   = 0;
+        char                *sbuf  = NULL;
+        stripe_xattr_sort_t *xattr = NULL;
+
+        if (!buffer || !local || !local->xattr_list)
+                goto out;
+
+        sbuf = buffer;
+
+        for (i = 0; i < local->nallocs; i++) {
+                xattr = local->xattr_list + i;
+                len = xattr->xattr_len;
+
+                if (len && xattr && xattr->xattr_value) {
+                        memcpy (buffer, xattr->xattr_value, len);
+                        buffer += len;
+                        *buffer++ = ' ';
+                }
+        }
+
+        *--buffer = '\0';
+        if (total)
+                *total = buffer - sbuf;
+        ret = 0;
+
+ out:
+        return ret;
+}
+
+int32_t
+stripe_free_xattr_str (stripe_local_t *local)
+{
+        int32_t              i     = 0;
+        int32_t              ret   = -1;
+        stripe_xattr_sort_t *xattr = NULL;
+
+        if (!local || !local->xattr_list)
+                goto out;
+
+        for (i = 0; i < local->nallocs; i++) {
+                xattr = local->xattr_list + i;
+
+                if (xattr && xattr->xattr_value)
+                        GF_FREE (xattr->xattr_value);
+        }
+
+        ret = 0;
+ out:
+        return ret;
+}
+
+
+int32_t
+stripe_fill_lockinfo_xattr (xlator_t *this, stripe_local_t *local,
+                            void **xattr_serz)
+{
+        int32_t              ret   = -1, i = 0, len = 0;
+        dict_t              *tmp1  = NULL, *tmp2 = NULL;
+        char                *buf   = NULL;
+        stripe_xattr_sort_t *xattr = NULL;
+
+        if (xattr_serz == NULL) {
+                goto out;
+        }
+
+        tmp2 = dict_new ();
+
+        if (tmp2 == NULL) {
+                goto out;
+        }
+
+        for (i = 0; i < local->nallocs; i++) {
+                xattr = local->xattr_list + i;
+                len = xattr->xattr_len;
+
+                if (len && xattr && xattr->xattr_value) {
+                        ret = dict_reset (tmp2);
+                        if (ret < 0) {
+                                gf_log (this->name, GF_LOG_DEBUG,
+                                        "dict_reset failed (%s)",
+                                        strerror (-ret));
+                        }
+
+                        ret = dict_unserialize (xattr->xattr_value,
+                                                xattr->xattr_len,
+                                                &tmp2);
+                        if (ret < 0) {
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "dict_unserialize failed (%s)",
+                                        strerror (-ret));
+                                ret = -1;
+                                goto out;
+                        }
+
+                        tmp1 = dict_copy (tmp2, tmp1);
+                        if (tmp1 == NULL) {
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "dict_copy failed (%s)",
+                                        strerror (-ret));
+                                ret = -1;
+                                goto out;
+                        }
+                }
+        }
+
+        len = dict_serialized_length (tmp1);
+        if (len > 0) {
+                buf = GF_CALLOC (1, len, gf_common_mt_dict_t);
+                if (buf == NULL) {
+                        ret = -1;
+                        goto out;
+                }
+
+                ret = dict_serialize (tmp1, buf);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "dict_serialize failed (%s)", strerror (-ret));
+                        GF_FREE(buf);
+                        ret = -1;
+                        goto out;
+                }
+
+                *xattr_serz = buf;
+        }
+
+        ret = 0;
+out:
+        if (tmp1 != NULL) {
+                dict_unref (tmp1);
+        }
+
+        if (tmp2 != NULL) {
+                dict_unref (tmp2);
+        }
+
+        return ret;
+}
+
+
+int32_t
+stripe_fill_pathinfo_xattr (xlator_t *this, stripe_local_t *local,
+                            char **xattr_serz)
+{
+        int      ret             = -1;
+        int32_t  padding         = 0;
+        int32_t  tlen            = 0;
+        char stripe_size_str[20] = {0,};
+        char    *pathinfo_serz   = NULL;
+
+        if (!local) {
+                gf_log (this->name, GF_LOG_ERROR, "Possible NULL deref");
+                goto out;
+        }
+
+        (void) snprintf (stripe_size_str, 20, "%"PRId64,
+                         (long long) (local->fctx) ? local->fctx->stripe_size : 0);
+
+        /* extra bytes for decorations (brackets and <>'s) */
+        padding = strlen (this->name) + strlen (STRIPE_PATHINFO_HEADER)
+                + strlen (stripe_size_str) + 7;
+        local->xattr_total_len += (padding + 2);
+
+        pathinfo_serz = GF_CALLOC (local->xattr_total_len, sizeof (char),
+                                   gf_common_mt_char);
+        if (!pathinfo_serz)
+                goto out;
+
+        /* xlator info */
+        (void) sprintf (pathinfo_serz, "(<"STRIPE_PATHINFO_HEADER"%s:[%s]> ",
+                        this->name, stripe_size_str);
+
+        ret = stripe_xattr_aggregate (pathinfo_serz + padding, local, &tlen);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Cannot aggregate pathinfo list");
+                GF_FREE(pathinfo_serz);
+                goto out;
+        }
+
+        *(pathinfo_serz + padding + tlen) = ')';
+        *(pathinfo_serz + padding + tlen + 1) = '\0';
+
+        *xattr_serz = pathinfo_serz;
+
+        ret = 0;
+ out:
+        return ret;
+}
+
+/**
+ * stripe_get_matching_bs - Get the matching block size for the given path.
+ */
+int32_t
+stripe_get_matching_bs (const char *path, stripe_private_t *priv)
+{
+        struct stripe_options *trav       = NULL;
+        uint64_t               block_size = 0;
+
+        GF_VALIDATE_OR_GOTO ("stripe", priv, out);
+        GF_VALIDATE_OR_GOTO ("stripe", path, out);
+
+        LOCK (&priv->lock);
+        {
+                block_size = priv->block_size;
+                trav = priv->pattern;
+                while (trav) {
+                        if (!fnmatch (trav->path_pattern, path, FNM_NOESCAPE)) {
+                                block_size = trav->block_size;
+                                break;
+                        }
+                        trav = trav->next;
+                }
+        }
+        UNLOCK (&priv->lock);
+
+out:
+        return block_size;
+}
+
+int32_t
+stripe_ctx_handle (xlator_t *this, call_frame_t *prev, stripe_local_t *local,
+                   dict_t *dict)
+{
+        char            key[256]       = {0,};
+        data_t         *data            = NULL;
+        int32_t         index           = 0;
+        stripe_private_t *priv          = NULL;
+
+        priv = this->private;
+
+
+        if (!local->fctx) {
+                local->fctx =  GF_CALLOC (1, sizeof (stripe_fd_ctx_t),
+                                         gf_stripe_mt_stripe_fd_ctx_t);
+                if (!local->fctx) {
+                        local->op_errno = ENOMEM;
+                        local->op_ret = -1;
+                        goto out;
+                }
+
+                local->fctx->static_array = 0;
+        }
+        /* Stripe block size */
+        sprintf (key, "trusted.%s.stripe-size", this->name);
+        data = dict_get (dict, key);
+        if (!data) {
+                local->xattr_self_heal_needed = 1;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Failed to get stripe-size");
+                goto out;
+        } else {
+                if (!local->fctx->stripe_size) {
+                        local->fctx->stripe_size =
+                                     data_to_int64 (data);
+                }
+
+                if (local->fctx->stripe_size != data_to_int64 (data)) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "stripe-size mismatch in blocks");
+                        local->xattr_self_heal_needed = 1;
+                }
+        }
+
+        /* Stripe count */
+        sprintf (key, "trusted.%s.stripe-count", this->name);
+        data = dict_get (dict, key);
+
+        if (!data) {
+                local->xattr_self_heal_needed = 1;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Failed to get stripe-count");
+                goto out;
+        }
+        if (!local->fctx->xl_array) {
+                local->fctx->stripe_count = data_to_int32 (data);
+                if (!local->fctx->stripe_count) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "error with stripe-count xattr");
+                        local->op_ret   = -1;
+                        local->op_errno = EIO;
+                        goto out;
+                }
+
+                local->fctx->xl_array = GF_CALLOC (local->fctx->stripe_count,
+                                                   sizeof (xlator_t *),
+                                                   gf_stripe_mt_xlator_t);
+
+                if (!local->fctx->xl_array) {
+                        local->op_errno = ENOMEM;
+                        local->op_ret   = -1;
+                        goto out;
+                }
+        }
+        if (local->fctx->stripe_count != data_to_int32 (data)) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "error with stripe-count xattr (%d != %d)",
+                        local->fctx->stripe_count, data_to_int32 (data));
+                local->op_ret   = -1;
+                local->op_errno = EIO;
+                goto out;
+        }
+
+        /* index */
+        sprintf (key, "trusted.%s.stripe-index", this->name);
+        data = dict_get (dict, key);
+        if (!data) {
+                local->xattr_self_heal_needed = 1;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Failed to get stripe-index");
+                goto out;
+        }
+        index = data_to_int32 (data);
+        if (index > priv->child_count) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "error with stripe-index xattr (%d)", index);
+                local->op_ret   = -1;
+                local->op_errno = EIO;
+                goto out;
+        }
+        if (local->fctx->xl_array) {
+                if (!local->fctx->xl_array[index])
+                        local->fctx->xl_array[index] = prev->this;
+        }
+
+	sprintf(key, "trusted.%s.stripe-coalesce", this->name);
+	data = dict_get(dict, key);
+	if (!data) {
+		/*
+		 * The file was probably created prior to coalesce support.
+		 * Assume non-coalesce mode for this file to maintain backwards
+		 * compatibility.
+		 */
+		gf_log(this->name, GF_LOG_DEBUG, "missing stripe-coalesce "
+			"attr, assume non-coalesce mode");
+		local->fctx->stripe_coalesce = 0;
+	} else {
+		local->fctx->stripe_coalesce = data_to_int32(data);
+	}
+
+
+out:
+        return 0;
+}
+
+int32_t
+stripe_xattr_request_build (xlator_t *this, dict_t *dict, uint64_t stripe_size,
+                            uint32_t stripe_count, uint32_t stripe_index,
+			    uint32_t stripe_coalesce)
+{
+        char            key[256]       = {0,};
+        int32_t         ret             = -1;
+
+        sprintf (key, "trusted.%s.stripe-size", this->name);
+        ret = dict_set_int64 (dict, key, stripe_size);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "failed to set %s in xattr_req dict", key);
+                goto out;
+        }
+
+        sprintf (key, "trusted.%s.stripe-count", this->name);
+        ret = dict_set_int32 (dict, key, stripe_count);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "failed to set %s in xattr_req dict", key);
+                goto out;
+        }
+
+        sprintf (key, "trusted.%s.stripe-index", this->name);
+        ret = dict_set_int32 (dict, key, stripe_index);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "failed to set %s in xattr_req dict", key);
+                goto out;
+        }
+
+	sprintf(key, "trusted.%s.stripe-coalesce", this->name);
+	ret = dict_set_int32(dict, key, stripe_coalesce);
+	if (ret) {
+		gf_log(this->name, GF_LOG_WARNING,
+			"failed to set %s in xattr_req_dict", key);
+		goto out;
+	}
+out:
+        return ret;
+}
+
+
+static int
+set_default_block_size (stripe_private_t *priv, char *num)
+{
+
+        int             ret = -1;
+        GF_VALIDATE_OR_GOTO ("stripe", THIS, out);
+        GF_VALIDATE_OR_GOTO (THIS->name, priv, out);
+        GF_VALIDATE_OR_GOTO (THIS->name, num, out);
+
+
+        if (gf_string2bytesize_uint64 (num, &priv->block_size) != 0) {
+                gf_log (THIS->name, GF_LOG_ERROR,
+                        "invalid number format \"%s\"", num);
+                goto out;
+        }
+
+        ret = 0;
+
+ out:
+        return ret;
+
+}
+
+
+int
+set_stripe_block_size (xlator_t *this, stripe_private_t *priv, char *data)
+{
+        int                    ret = -1;
+        char                  *tmp_str = NULL;
+        char                  *tmp_str1 = NULL;
+        char                  *dup_str = NULL;
+        char                  *stripe_str = NULL;
+        char                  *pattern = NULL;
+        char                  *num = NULL;
+        struct stripe_options *temp_stripeopt = NULL;
+        struct stripe_options *stripe_opt = NULL;
+
+        if (!this || !priv || !data)
+                goto out;
+
+        /* Get the pattern for striping.
+           "option block-size *avi:10MB" etc */
+        stripe_str = strtok_r (data, ",", &tmp_str);
+        while (stripe_str) {
+                dup_str = gf_strdup (stripe_str);
+                stripe_opt = GF_CALLOC (1, sizeof (struct stripe_options),
+                                        gf_stripe_mt_stripe_options);
+                if (!stripe_opt) {
+                        goto out;
+                }
+
+                pattern = strtok_r (dup_str, ":", &tmp_str1);
+                num = strtok_r (NULL, ":", &tmp_str1);
+                if (!num) {
+                        num = pattern;
+                        pattern = "*";
+                        ret = set_default_block_size (priv, num);
+                        if (ret)
+                                goto out;
+                }
+                if (gf_string2bytesize_uint64 (num, &stripe_opt->block_size) != 0) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "invalid number format \"%s\"", num);
+                        goto out;
+                }
+
+                if (stripe_opt->block_size < STRIPE_MIN_BLOCK_SIZE) {
+                        gf_log (this->name, GF_LOG_ERROR, "Invalid Block-size: "
+                                "%s. Should be atleast %llu bytes", num,
+                                STRIPE_MIN_BLOCK_SIZE);
+                        goto out;
+                }
+                if (stripe_opt->block_size % 512) {
+                        gf_log (this->name, GF_LOG_ERROR, "Block-size: %s should"
+                                " be a multiple of 512 bytes", num);
+                        goto out;
+                }
+
+                memcpy (stripe_opt->path_pattern, pattern, strlen (pattern));
+
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "block-size : pattern %s : size %"PRId64,
+                        stripe_opt->path_pattern, stripe_opt->block_size);
+
+                if (priv->pattern)
+                        temp_stripeopt = NULL;
+                else
+                        temp_stripeopt = priv->pattern;
+
+                stripe_opt->next = temp_stripeopt;
+
+                priv->pattern = stripe_opt;
+                stripe_opt = NULL;
+
+                GF_FREE (dup_str);
+                dup_str = NULL;
+
+                stripe_str = strtok_r (NULL, ",", &tmp_str);
+        }
+
+        ret = 0;
+out:
+
+        GF_FREE (dup_str);
+
+        GF_FREE (stripe_opt);
+
+        return ret;
+}
+
+int32_t
+stripe_iatt_merge (struct iatt *from, struct iatt *to)
+{
+        if (to->ia_size < from->ia_size)
+                to->ia_size = from->ia_size;
+        if (to->ia_mtime < from->ia_mtime)
+                to->ia_mtime = from->ia_mtime;
+        if (to->ia_ctime < from->ia_ctime)
+                to->ia_ctime = from->ia_ctime;
+        if (to->ia_atime < from->ia_atime)
+                to->ia_atime = from->ia_atime;
+        return 0;
+}
+
+off_t
+coalesced_offset(off_t offset, uint64_t stripe_size, int stripe_count)
+{
+	size_t line_size = 0;
+	uint64_t stripe_num = 0;
+	off_t coalesced_offset = 0;
+
+	line_size = stripe_size * stripe_count;
+	stripe_num = offset / line_size;
+
+	coalesced_offset = (stripe_num * stripe_size) +
+		(offset % stripe_size);
+
+	return coalesced_offset;
+}
+
+off_t
+uncoalesced_size(off_t size, uint64_t stripe_size, int stripe_count,
+		int stripe_index)
+{
+        uint64_t nr_full_stripe_chunks = 0, mod = 0;
+
+        if (!size)
+                return size;
+
+	/*
+	 * Estimate the number of fully written stripes from the
+	 * local file size. Each stripe_size chunk corresponds to
+	 * a stripe.
+	 */
+        nr_full_stripe_chunks = (size / stripe_size) * stripe_count;
+        mod = size % stripe_size;
+
+        if (!mod) {
+                /*
+		 * There is no remainder, thus we could have overestimated
+		 * the size of the file in terms of chunks. Trim the number
+		 * of chunks by the following stripe members and leave it
+		 * up to those nodes to respond with a larger size (if
+		 * necessary).
+		 */
+                nr_full_stripe_chunks -= stripe_count -
+                        (stripe_index + 1);
+                size = nr_full_stripe_chunks * stripe_size;
+        } else {
+		/*
+		 * There is a remainder and thus we own the last chunk of the
+		 * file. Add the preceding stripe members of the final stripe
+		 * along with the remainder to calculate the exact size.
+		 */
+                nr_full_stripe_chunks += stripe_index;
+                size = nr_full_stripe_chunks * stripe_size + mod;
+        }
+
+        return size;
+}
diff --git a/xlators/cluster/stripe/src/stripe-mem-types.h b/xlators/cluster/stripe/src/stripe-mem-types.h
index c376a1fc9..e9ac9cf46 100644
--- a/xlators/cluster/stripe/src/stripe-mem-types.h
+++ b/xlators/cluster/stripe/src/stripe-mem-types.h
@@ -1,21 +1,11 @@
-
 /*
-   Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -25,12 +15,12 @@
 #include "mem-types.h"
 
 enum gf_stripe_mem_types_ {
-        gf_stripe_mt_stripe_local_t = gf_common_mt_end + 1,
-        gf_stripe_mt_iovec,
-        gf_stripe_mt_readv_replies,
+        gf_stripe_mt_iovec = gf_common_mt_end + 1,
+        gf_stripe_mt_stripe_replies,
         gf_stripe_mt_stripe_fd_ctx_t,
         gf_stripe_mt_char,
         gf_stripe_mt_int8_t,
+        gf_stripe_mt_int32_t,
         gf_stripe_mt_xlator_t,
         gf_stripe_mt_stripe_private_t,
         gf_stripe_mt_stripe_options,
diff --git a/xlators/cluster/stripe/src/stripe.c b/xlators/cluster/stripe/src/stripe.c
index bf54bdea2..0ebea8168 100644
--- a/xlators/cluster/stripe/src/stripe.c
+++ b/xlators/cluster/stripe/src/stripe.c
@@ -1,25 +1,16 @@
 /*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 /**
  * xlators/cluster/stripe:
- *    Stripe translator, stripes the data accross its child nodes,
+ *    Stripe translator, stripes the data across its child nodes,
  *    as per the options given in the volfile. The striping works
  *    fairly simple. It writes files at different offset as per
  *    calculation. So, 'ls -l' output at the real posix level will
@@ -32,6 +23,7 @@
  *  very much necessary, or else, use it in combination with AFR, to have a
  *  backup copy.
  */
+#include <fnmatch.h>
 
 #include "stripe.h"
 #include "libxlator.h"
@@ -40,73 +32,10 @@
 
 struct volume_options options[];
 
-void
-stripe_local_wipe (stripe_local_t *local)
-{
-        if (!local)
-                goto out;
-
-        loc_wipe (&local->loc);
-        loc_wipe (&local->loc2);
-
-        if (local->fd)
-                fd_unref (local->fd);
-
-        if (local->inode)
-                inode_unref (local->inode);
-
-        if (local->xattr)
-                dict_unref (local->xattr);
-
-        if (local->dict)
-                dict_unref (local->dict);
-
-out:
-        return;
-}
-
-/**
- * stripe_get_matching_bs - Get the matching block size for the given path.
- */
-int32_t
-stripe_get_matching_bs (const char *path, struct stripe_options *opts,
-                        uint64_t default_bs)
-{
-        struct stripe_options *trav       = NULL;
-        char                  *pathname   = NULL;
-        uint64_t               block_size = 0;
-
-        block_size = default_bs;
-
-        if (!path || !opts)
-                goto out;
-
-        /* FIXME: is a strdup really necessary? */
-        pathname = gf_strdup (path);
-        if (!pathname)
-                goto out;
-
-        trav = opts;
-        while (trav) {
-                if (!fnmatch (trav->path_pattern, pathname, FNM_NOESCAPE)) {
-                        block_size = trav->block_size;
-                        break;
-                }
-                trav = trav->next;
-        }
-
-        GF_FREE (pathname);
-
-out:
-        return block_size;
-}
-
-
-
 int32_t
 stripe_sh_chown_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                      int32_t op_ret, int32_t op_errno,
-                     struct iatt *preop, struct iatt *postop)
+                     struct iatt *preop, struct iatt *postop, dict_t *xdata)
 {
         int             callcnt = -1;
         stripe_local_t *local   = NULL;
@@ -135,7 +64,7 @@ int32_t
 stripe_sh_make_entry_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                           int32_t op_ret, int32_t op_errno, inode_t *inode,
                           struct iatt *buf, struct iatt *preparent,
-                          struct iatt *postparent)
+                          struct iatt *postparent, dict_t *xdata)
 {
         stripe_local_t *local = NULL;
         call_frame_t    *prev = NULL;
@@ -150,7 +79,7 @@ stripe_sh_make_entry_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         STACK_WIND (frame, stripe_sh_chown_cbk, prev->this,
                     prev->this->fops->setattr, &local->loc,
-                    &local->stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID));
+                    &local->stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID), NULL);
 
 out:
         return 0;
@@ -164,7 +93,7 @@ stripe_entry_self_heal (call_frame_t *frame, xlator_t *this,
         call_frame_t     *rframe = NULL;
         stripe_local_t   *rlocal = NULL;
         stripe_private_t *priv   = NULL;
-        dict_t           *dict   = NULL;
+        dict_t           *xdata  = NULL;
         int               ret    = 0;
 
         if (!local || !this || !frame) {
@@ -182,8 +111,7 @@ stripe_entry_self_heal (call_frame_t *frame, xlator_t *this,
         if (!rframe) {
                 goto out;
         }
-        rlocal = GF_CALLOC (1, sizeof (stripe_local_t),
-                            gf_stripe_mt_stripe_local_t);
+        rlocal = mem_get0 (this->local_pool);
         if (!rlocal) {
                 goto out;
         }
@@ -192,14 +120,14 @@ stripe_entry_self_heal (call_frame_t *frame, xlator_t *this,
         loc_copy (&rlocal->loc, &local->loc);
         memcpy (&rlocal->stbuf, &local->stbuf, sizeof (struct iatt));
 
-        dict = dict_new ();
-        if (!dict)
+        xdata = dict_new ();
+        if (!xdata)
                 goto out;
 
-        ret = dict_set_static_bin (dict, "gfid-req", local->stbuf.ia_gfid, 16);
+        ret = dict_set_static_bin (xdata, "gfid-req", local->stbuf.ia_gfid, 16);
         if (ret)
                 gf_log (this->name, GF_LOG_WARNING,
-                        "failed to set gfid-req");
+                        "%s: failed to set gfid-req", local->loc.path);
 
         while (trav) {
                 if (IA_ISREG (local->stbuf.ia_type)) {
@@ -207,101 +135,43 @@ stripe_entry_self_heal (call_frame_t *frame, xlator_t *this,
                                     trav->xlator, trav->xlator->fops->mknod,
                                     &local->loc,
                                     st_mode_from_ia (local->stbuf.ia_prot,
-                                                     local->stbuf.ia_type), 0,
-                                    dict);
+                                                     local->stbuf.ia_type),
+                                    0, 0, xdata);
                 }
                 if (IA_ISDIR (local->stbuf.ia_type)) {
                         STACK_WIND (rframe, stripe_sh_make_entry_cbk,
                                     trav->xlator, trav->xlator->fops->mkdir,
-                                    &local->loc, st_mode_from_ia (local->stbuf.ia_prot,
-                                                                  local->stbuf.ia_type),
-                                    dict);
+                                    &local->loc,
+                                    st_mode_from_ia (local->stbuf.ia_prot,
+                                                     local->stbuf.ia_type),
+                                    0, xdata);
                 }
                 trav = trav->next;
         }
 
-        if (dict)
-                dict_unref (dict);
+        if (xdata)
+                dict_unref (xdata);
         return 0;
 
 out:
         if (rframe)
                 STRIPE_STACK_DESTROY (rframe);
-        if (dict)
-                dict_unref (dict);
+        if (xdata)
+                dict_unref (xdata);
 
         return 0;
 }
 
 
-void
-stripe_aggregate (dict_t *this, char *key, data_t *value, void *data)
-{
-        dict_t  *dst  = NULL;
-        int64_t *ptr  = 0, *size = NULL;
-        int32_t  ret  = -1;
-
-        dst = data;
-
-        if (strcmp (key, GF_XATTR_QUOTA_SIZE_KEY) == 0) {
-                ret = dict_get_bin (dst, key, (void **)&size);
-                if (ret < 0) {
-                        size = GF_CALLOC (1, sizeof (int64_t),
-                                          gf_common_mt_char);
-                        if (size == NULL) {
-                                gf_log ("stripe", GF_LOG_WARNING,
-                                        "memory allocation failed");
-                                goto out;
-                        }
-                        ret = dict_set_bin (dst, key, size, sizeof (int64_t));
-                        if (ret < 0) {
-                                gf_log ("stripe", GF_LOG_WARNING,
-                                        "stripe aggregate dict set failed");
-                                GF_FREE (size);
-                                goto out;
-                        }
-                }
-
-                ptr = data_to_bin (value);
-                if (ptr == NULL) {
-                        gf_log ("stripe", GF_LOG_WARNING, "data to bin failed");
-                        goto out;
-                }
-
-                *size = hton64 (ntoh64 (*size) + ntoh64 (*ptr));
-        } else if (strcmp (key, GF_CONTENT_KEY)) {
-                /* No need to aggregate 'CONTENT' data */
-                ret = dict_set (dst, key, value);
-                if (ret)
-                        gf_log ("stripe", GF_LOG_WARNING, "xattr dict set failed");
-        }
-
-out:
-        return;
-}
-
-
-void
-stripe_aggregate_xattr (dict_t *dst, dict_t *src)
-{
-        if ((dst == NULL) || (src == NULL)) {
-                goto out;
-        }
-
-        dict_foreach (src, stripe_aggregate, dst);
-out:
-        return;
-}
-
-
 int32_t
 stripe_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                    int32_t op_ret, int32_t op_errno, inode_t *inode,
-                   struct iatt *buf, dict_t *dict, struct iatt *postparent)
+                   struct iatt *buf, dict_t *xdata, struct iatt *postparent)
 {
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
+        int32_t         callcnt     = 0;
+        stripe_local_t *local       = NULL;
+        call_frame_t   *prev        = NULL;
+        int             ret         = 0;
 
         if (!this || !frame || !frame->local || !cookie) {
                 gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
@@ -332,30 +202,42 @@ stripe_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
                 if (op_ret >= 0) {
                         local->op_ret = 0;
+                        if (IA_ISREG (buf->ia_type)) {
+                                ret = stripe_ctx_handle (this, prev, local,
+                                                         xdata);
+                                if (ret)
+                                        gf_log (this->name, GF_LOG_ERROR,
+                                                 "Error getting fctx info from"
+                                                 " dict");
+                        }
 
                         if (FIRST_CHILD(this) == prev->this) {
                                 local->stbuf      = *buf;
                                 local->postparent = *postparent;
                                 local->inode = inode_ref (inode);
-                                local->dict = dict_ref (dict);
+                                if (xdata)
+                                        local->xdata = dict_ref (xdata);
                                 if (local->xattr) {
-                                        stripe_aggregate_xattr (local->dict,
+                                        stripe_aggregate_xattr (local->xdata,
                                                                 local->xattr);
                                         dict_unref (local->xattr);
                                         local->xattr = NULL;
                                 }
                         }
-                        if (!local->dict && !local->xattr) {
-                                local->xattr = dict_ref (dict);
-                        } else if (local->dict) {
-                                stripe_aggregate_xattr (local->dict, dict);
+
+                        if (!local->xdata && !local->xattr) {
+                                local->xattr = dict_ref (xdata);
+                        } else if (local->xdata) {
+                                stripe_aggregate_xattr (local->xdata, xdata);
                         } else if (local->xattr) {
-                                stripe_aggregate_xattr (local->xattr, dict);
+                                stripe_aggregate_xattr (local->xattr, xdata);
                         }
 
                         local->stbuf_blocks      += buf->ia_blocks;
                         local->postparent_blocks += postparent->ia_blocks;
 
+			correct_file_size(buf, local->fctx, prev);
+
                         if (local->stbuf_size < buf->ia_size)
                                 local->stbuf_size = buf->ia_size;
                         if (local->postparent_size < postparent->ia_size)
@@ -375,7 +257,8 @@ stripe_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         UNLOCK (&frame->lock);
 
         if (!callcnt) {
-                if (local->op_ret == 0 && local->entry_self_heal_needed)
+                if (local->op_ret == 0 && local->entry_self_heal_needed &&
+                    !uuid_is_null (local->loc.inode->gfid))
                         stripe_entry_self_heal (frame, this, local);
 
                 if (local->failed)
@@ -386,11 +269,13 @@ stripe_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->stbuf.ia_size        = local->stbuf_size;
                         local->postparent.ia_blocks = local->postparent_blocks;
                         local->postparent.ia_size   = local->postparent_size;
+                        inode_ctx_put (local->inode, this,
+                                       (uint64_t) (long)local->fctx);
                 }
 
                 STRIPE_STACK_UNWIND (lookup, frame, local->op_ret,
                                      local->op_errno, local->inode,
-                                     &local->stbuf, local->dict,
+                                     &local->stbuf, local->xdata,
                                      &local->postparent);
         }
 out:
@@ -399,14 +284,15 @@ out:
 
 int32_t
 stripe_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               dict_t *xattr_req)
+               dict_t *xdata)
 {
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        stripe_private_t *priv = NULL;
+        stripe_local_t   *local    = NULL;
+        xlator_list_t    *trav     = NULL;
+        stripe_private_t *priv     = NULL;
         int32_t           op_errno = EINVAL;
         int64_t           filesize = 0;
-        int               ret = 0;
+        int               ret      = 0;
+        uint64_t          tmpctx   = 0;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -418,8 +304,7 @@ stripe_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
         trav = this->children;
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -428,10 +313,37 @@ stripe_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
         frame->local = local;
         loc_copy (&local->loc, loc);
 
-        if (xattr_req && dict_get (xattr_req, GF_CONTENT_KEY)) {
-                ret = dict_get_int64 (xattr_req, GF_CONTENT_KEY, &filesize);
+        inode_ctx_get (local->inode, this, &tmpctx);
+        if (tmpctx)
+                local->fctx = (stripe_fd_ctx_t*) (long)tmpctx;
+
+        /* quick-read friendly changes */
+        if (xdata && dict_get (xdata, GF_CONTENT_KEY)) {
+                ret = dict_get_int64 (xdata, GF_CONTENT_KEY, &filesize);
                 if (!ret && (filesize > priv->block_size))
-                        dict_del (xattr_req, GF_CONTENT_KEY);
+                        dict_del (xdata, GF_CONTENT_KEY);
+        }
+
+        /* get stripe-size xattr on lookup. This would be required for
+         * open/read/write/pathinfo calls. Hence we send down the request
+         * even when type == IA_INVAL */
+
+	/*
+	 * We aren't guaranteed to have xdata here. We need the format info for
+	 * the file, so allocate xdata if necessary.
+	 */
+	if (!xdata)
+		xdata = dict_new();
+	else
+		xdata = dict_ref(xdata);
+
+        if (xdata && (IA_ISREG (loc->inode->ia_type) ||
+            (loc->inode->ia_type == IA_INVAL))) {
+                ret = stripe_xattr_request_build (this, xdata, 8, 4, 4, 0);
+                if (ret)
+                        gf_log (this->name , GF_LOG_ERROR, "Failed to build"
+                                " xattr request for %s", loc->path);
+
         }
 
         /* Everytime in stripe lookup, all child nodes
@@ -439,11 +351,12 @@ stripe_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
         local->call_count = priv->child_count;
         while (trav) {
                 STACK_WIND (frame, stripe_lookup_cbk, trav->xlator,
-                            trav->xlator->fops->lookup,
-                            loc, xattr_req);
+                            trav->xlator->fops->lookup, loc, xdata);
                 trav = trav->next;
         }
 
+	dict_unref(xdata);
+
         return 0;
 err:
         STRIPE_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
@@ -453,7 +366,7 @@ err:
 
 int32_t
 stripe_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, struct iatt *buf)
+                 int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
@@ -488,6 +401,9 @@ stripe_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         }
 
                         local->stbuf_blocks += buf->ia_blocks;
+
+			correct_file_size(buf, local->fctx, prev);
+
                         if (local->stbuf_size < buf->ia_size)
                                 local->stbuf_size = buf->ia_size;
                 }
@@ -504,18 +420,19 @@ stripe_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 }
 
                 STRIPE_STACK_UNWIND (stat, frame, local->op_ret,
-                                     local->op_errno, &local->stbuf);
+                                     local->op_errno, &local->stbuf, NULL);
         }
 out:
         return 0;
 }
 
 int32_t
-stripe_stat (call_frame_t *frame, xlator_t *this, loc_t *loc)
+stripe_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
         xlator_list_t    *trav = NULL;
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
+	stripe_fd_ctx_t  *fctx = NULL;
         int32_t           op_errno = EINVAL;
 
         VALIDATE_OR_GOTO (frame, err);
@@ -533,8 +450,7 @@ stripe_stat (call_frame_t *frame, xlator_t *this, loc_t *loc)
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -543,23 +459,30 @@ stripe_stat (call_frame_t *frame, xlator_t *this, loc_t *loc)
         frame->local = local;
         local->call_count = priv->child_count;
 
+	if (IA_ISREG(loc->inode->ia_type)) {
+		inode_ctx_get(loc->inode, this, (uint64_t *) &fctx);
+		if (!fctx)
+			goto err;
+		local->fctx = fctx;
+	}
+
         while (trav) {
                 STACK_WIND (frame, stripe_stat_cbk, trav->xlator,
-                            trav->xlator->fops->stat, loc);
+                            trav->xlator->fops->stat, loc, NULL);
                 trav = trav->next;
         }
 
         return 0;
 
 err:
-        STRIPE_STACK_UNWIND (stat, frame, -1, op_errno, NULL);
+        STRIPE_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
         return 0;
 }
 
 
 int32_t
 stripe_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct statvfs *stbuf)
+                   int32_t op_ret, int32_t op_errno, struct statvfs *stbuf, dict_t *xdata)
 {
         stripe_local_t *local = NULL;
         int32_t         callcnt = 0;
@@ -597,14 +520,14 @@ stripe_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         if (!callcnt) {
                 STRIPE_STACK_UNWIND (statfs, frame, local->op_ret,
-                                     local->op_errno, &local->statvfs_buf);
+                                     local->op_errno, &local->statvfs_buf, NULL);
         }
 out:
         return 0;
 }
 
 int32_t
-stripe_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
+stripe_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
         stripe_local_t   *local = NULL;
         xlator_list_t    *trav = NULL;
@@ -619,8 +542,7 @@ stripe_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
         priv = this->private;
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -632,13 +554,13 @@ stripe_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
         local->call_count = priv->child_count;
         while (trav) {
                 STACK_WIND (frame, stripe_statfs_cbk, trav->xlator,
-                            trav->xlator->fops->statfs, loc);
+                            trav->xlator->fops->statfs, loc, NULL);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (statfs, frame, -1, op_errno, NULL);
+        STRIPE_STACK_UNWIND (statfs, frame, -1, op_errno, NULL, NULL);
         return 0;
 }
 
@@ -647,7 +569,7 @@ err:
 int32_t
 stripe_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                     struct iatt *postbuf)
+                     struct iatt *postbuf, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
@@ -685,6 +607,9 @@ stripe_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->prebuf_blocks  += prebuf->ia_blocks;
                         local->postbuf_blocks += postbuf->ia_blocks;
 
+			correct_file_size(prebuf, local->fctx, prev);
+			correct_file_size(postbuf, local->fctx, prev);
+
                         if (local->prebuf_size < prebuf->ia_size)
                                 local->prebuf_size = prebuf->ia_size;
 
@@ -707,19 +632,21 @@ stripe_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
                 STRIPE_STACK_UNWIND (truncate, frame, local->op_ret,
                                      local->op_errno, &local->pre_buf,
-                                     &local->post_buf);
+                                     &local->post_buf, NULL);
         }
 out:
         return 0;
 }
 
 int32_t
-stripe_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset)
+stripe_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset, dict_t *xdata)
 {
-        xlator_list_t    *trav = NULL;
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
+	stripe_fd_ctx_t  *fctx = NULL;
         int32_t           op_errno = EINVAL;
+	int		  i, eof_idx;
+	off_t		  dest_offset, tmp_offset;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -728,7 +655,6 @@ stripe_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset)
         VALIDATE_OR_GOTO (loc->inode, err);
 
         priv = this->private;
-        trav = this->children;
 
         if (priv->first_child_down) {
                 op_errno = ENOTCONN;
@@ -736,8 +662,7 @@ stripe_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset)
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -746,15 +671,55 @@ stripe_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset)
         frame->local = local;
         local->call_count = priv->child_count;
 
-        while (trav) {
-                STACK_WIND (frame, stripe_truncate_cbk, trav->xlator,
-                            trav->xlator->fops->truncate, loc, offset);
-                trav = trav->next;
-        }
+	inode_ctx_get(loc->inode, this, (uint64_t *) &fctx);
+	if (!fctx) {
+		gf_log(this->name, GF_LOG_ERROR, "no stripe context");
+		op_errno = EINVAL;
+		goto err;
+	}
+
+	local->fctx = fctx;
+	eof_idx = (offset / fctx->stripe_size) % fctx->stripe_count;
+
+	for (i = 0; i < fctx->stripe_count; i++) {
+		if (!fctx->xl_array[i]) {
+			gf_log(this->name, GF_LOG_ERROR,
+				"no xlator at index %d", i);
+			op_errno = EINVAL;
+			goto err;
+		}
+
+		if (fctx->stripe_coalesce) {
+			/*
+	 		 * The node that owns EOF is truncated to the exact
+	 		 * coalesced offset. Nodes prior to this index should
+	 		 * be rounded up to the size of the complete stripe,
+	 		 * while nodes after this index should be rounded down
+			 * to the size of the previous stripe.
+			 */
+			if (i < eof_idx)
+				tmp_offset = roof(offset, fctx->stripe_size *
+						fctx->stripe_count);
+			else if (i > eof_idx)
+				tmp_offset = floor(offset, fctx->stripe_size *
+						fctx->stripe_count);
+			else
+				tmp_offset = offset;
+
+			dest_offset = coalesced_offset(tmp_offset,
+					fctx->stripe_size, fctx->stripe_count);
+		} else {
+			dest_offset = offset;
+		}
+
+		STACK_WIND(frame, stripe_truncate_cbk, fctx->xl_array[i],
+			fctx->xl_array[i]->fops->truncate, loc, dest_offset,
+			NULL);
+	}
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL);
+        STRIPE_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
@@ -762,7 +727,7 @@ err:
 int32_t
 stripe_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                     int32_t op_ret, int32_t op_errno,
-                    struct iatt *preop, struct iatt *postop)
+                    struct iatt *preop, struct iatt *postop, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
@@ -801,6 +766,9 @@ stripe_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->prebuf_blocks  += preop->ia_blocks;
                         local->postbuf_blocks += postop->ia_blocks;
 
+			correct_file_size(preop, local->fctx, prev);
+			correct_file_size(postop, local->fctx, prev);
+
                         if (local->prebuf_size < preop->ia_size)
                                 local->prebuf_size = preop->ia_size;
                         if (local->postbuf_size < postop->ia_size)
@@ -822,7 +790,7 @@ stripe_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
                 STRIPE_STACK_UNWIND (setattr, frame, local->op_ret,
                                      local->op_errno, &local->pre_buf,
-                                     &local->post_buf);
+                                     &local->post_buf, NULL);
         }
 out:
         return 0;
@@ -831,11 +799,12 @@ out:
 
 int32_t
 stripe_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                struct iatt *stbuf, int32_t valid)
+                struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
         xlator_list_t    *trav = NULL;
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
+	stripe_fd_ctx_t	 *fctx = NULL;
         int32_t           op_errno = EINVAL;
 
         VALIDATE_OR_GOTO (frame, err);
@@ -853,33 +822,47 @@ stripe_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
         local->op_ret = -1;
         frame->local = local;
-        local->call_count = priv->child_count;
+        if (!IA_ISDIR (loc->inode->ia_type) &&
+            !IA_ISREG (loc->inode->ia_type)) {
+                local->call_count = 1;
+                STACK_WIND (frame, stripe_setattr_cbk, FIRST_CHILD (this),
+                            FIRST_CHILD (this)->fops->setattr,
+                            loc, stbuf, valid, NULL);
+                return 0;
+        }
+
+	if (IA_ISREG(loc->inode->ia_type)) {
+		inode_ctx_get(loc->inode, this, (uint64_t *) &fctx);
+		if (!fctx)
+			goto err;
+		local->fctx = fctx;
+	}
 
+        local->call_count = priv->child_count;
         while (trav) {
                 STACK_WIND (frame, stripe_setattr_cbk,
                             trav->xlator, trav->xlator->fops->setattr,
-                            loc, stbuf, valid);
+                            loc, stbuf, valid, NULL);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL);
+        STRIPE_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
 
 int32_t
 stripe_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                 struct iatt *stbuf, int32_t valid)
+                 struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
@@ -895,8 +878,7 @@ stripe_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
         trav = this->children;
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -907,13 +889,13 @@ stripe_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
 
         while (trav) {
                 STACK_WIND (frame, stripe_setattr_cbk, trav->xlator,
-                            trav->xlator->fops->fsetattr, fd, stbuf, valid);
+                            trav->xlator->fops->fsetattr, fd, stbuf, valid, NULL);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL);
+        STRIPE_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
@@ -921,7 +903,8 @@ int32_t
 stripe_stack_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                          int32_t op_ret, int32_t op_errno, struct iatt *buf,
                          struct iatt *preoldparent, struct iatt *postoldparent,
-                         struct iatt *prenewparent, struct iatt *postnewparent)
+                         struct iatt *prenewparent, struct iatt *postnewparent,
+                         dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
@@ -958,6 +941,8 @@ stripe_stack_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->pre_buf.ia_blocks    += prenewparent->ia_blocks;
                         local->post_buf.ia_blocks   += postnewparent->ia_blocks;
 
+			correct_file_size(buf, local->fctx, prev);
+
                         if (local->stbuf.ia_size < buf->ia_size)
                                 local->stbuf.ia_size =  buf->ia_size;
 
@@ -983,7 +968,7 @@ stripe_stack_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 STRIPE_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
                                      &local->stbuf, &local->preparent,
                                      &local->postparent,  &local->pre_buf,
-                                     &local->post_buf);
+                                     &local->post_buf, NULL);
         }
 out:
         return 0;
@@ -993,7 +978,8 @@ int32_t
 stripe_first_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                          int32_t op_ret, int32_t op_errno, struct iatt *buf,
                          struct iatt *preoldparent, struct iatt *postoldparent,
-                         struct iatt *prenewparent, struct iatt *postnewparent)
+                         struct iatt *prenewparent, struct iatt *postnewparent,
+                         dict_t *xdata)
 {
         stripe_local_t *local = NULL;
         xlator_list_t  *trav = NULL;
@@ -1024,24 +1010,25 @@ stripe_first_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         while (trav) {
                 STACK_WIND (frame, stripe_stack_rename_cbk,
                             trav->xlator, trav->xlator->fops->rename,
-                            &local->loc, &local->loc2);
+                            &local->loc, &local->loc2, NULL);
                 trav = trav->next;
         }
         return 0;
 
 unwind:
         STRIPE_STACK_UNWIND (rename, frame, -1, op_errno, buf, preoldparent,
-                             postoldparent, prenewparent, postnewparent);
+                             postoldparent, prenewparent, postnewparent, NULL);
         return 0;
 }
 
 int32_t
 stripe_rename (call_frame_t *frame, xlator_t *this, loc_t *oldloc,
-               loc_t *newloc)
+               loc_t *newloc, dict_t *xdata)
 {
         stripe_private_t *priv = NULL;
         stripe_local_t   *local = NULL;
         xlator_list_t    *trav = NULL;
+	stripe_fd_ctx_t	 *fctx = NULL;
         int32_t           op_errno = EINVAL;
 
         VALIDATE_OR_GOTO (frame, err);
@@ -1061,36 +1048,79 @@ stripe_rename (call_frame_t *frame, xlator_t *this, loc_t *oldloc,
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
+
+        frame->local = local;
+
         local->op_ret = -1;
         loc_copy (&local->loc, oldloc);
         loc_copy (&local->loc2, newloc);
 
         local->call_count = priv->child_count;
 
-        frame->local = local;
+	if (IA_ISREG(oldloc->inode->ia_type)) {
+		inode_ctx_get(oldloc->inode, this, (uint64_t *) &fctx);
+		if (!fctx)
+			goto err;
+		local->fctx = fctx;
+	}
 
         STACK_WIND (frame, stripe_first_rename_cbk, trav->xlator,
-                    trav->xlator->fops->rename, oldloc, newloc);
+                    trav->xlator->fops->rename, oldloc, newloc, NULL);
 
         return 0;
 err:
         STRIPE_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL,
-                             NULL, NULL);
+                             NULL, NULL, NULL);
+        return 0;
+}
+int32_t
+stripe_first_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                         struct iatt *postparent, dict_t *xdata)
+{
+        stripe_local_t *local   = NULL;
+        call_frame_t   *prev = NULL;
+
+        if (!this || !frame || !frame->local || !cookie) {
+                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
+                goto out;
+        }
+
+        prev  = cookie;
+        local = frame->local;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_DEBUG, "%s returned %s",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+        local->op_ret = 0;
+        local->preparent  = *preparent;
+        local->postparent = *postparent;
+        local->preparent_blocks  += preparent->ia_blocks;
+        local->postparent_blocks += postparent->ia_blocks;
+
+        STRIPE_STACK_UNWIND(unlink, frame, local->op_ret, local->op_errno,
+                            &local->preparent, &local->postparent, xdata);
+        return 0;
+out:
+        STRIPE_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL);
+
         return 0;
 }
 
 
 
+
 int32_t
 stripe_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                    int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                   struct iatt *postparent)
+                   struct iatt *postparent, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local   = NULL;
@@ -1112,49 +1142,33 @@ stripe_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         gf_log (this->name, GF_LOG_DEBUG, "%s returned %s",
                                 prev->this->name, strerror (op_errno));
                         local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
+                        if (op_errno != ENOENT) {
                                 local->failed = 1;
-                }
-                if (op_ret >= 0) {
-                        local->op_ret = op_ret;
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->preparent  = *preparent;
-                                local->postparent = *postparent;
+                                local->op_ret = op_ret;
                         }
-                        local->preparent_blocks  += preparent->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-                        if (local->preparent_size < preparent->ia_size)
-                                local->preparent_size = preparent->ia_size;
-
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
                 }
         }
         UNLOCK (&frame->lock);
 
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->preparent.ia_blocks  = local->preparent_blocks;
-                        local->preparent.ia_size    = local->preparent_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
+        if (callcnt == 1) {
+                if (local->failed) {
+                        op_errno = local->op_errno;
+                        goto out;
                 }
-
-                STRIPE_STACK_UNWIND (unlink, frame, local->op_ret,
-                                     local->op_errno, &local->preparent,
-                                     &local->postparent);
+                STACK_WIND(frame, stripe_first_unlink_cbk, FIRST_CHILD (this),
+                           FIRST_CHILD (this)->fops->unlink, &local->loc,
+                           local->xflag, local->xdata);
         }
+        return 0;
 out:
+        STRIPE_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL);
+
         return 0;
 }
 
 int32_t
-stripe_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
+stripe_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc,
+               int xflag, dict_t *xdata)
 {
         xlator_list_t    *trav = NULL;
         stripe_local_t   *local = NULL;
@@ -1182,26 +1196,32 @@ stripe_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
         local->op_ret = -1;
+        loc_copy (&local->loc, loc);
+        local->xflag = xflag;
+
+        if (xdata)
+            local->xdata = dict_ref (xdata);
+
         frame->local = local;
         local->call_count = priv->child_count;
+        trav = trav->next; /* Skip the first child */
 
         while (trav) {
                 STACK_WIND (frame, stripe_unlink_cbk,
                             trav->xlator, trav->xlator->fops->unlink,
-                            loc);
+                            loc, xflag, xdata);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL);
+        STRIPE_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
@@ -1209,10 +1229,8 @@ err:
 int32_t
 stripe_first_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         int32_t op_ret, int32_t op_errno,struct iatt *preparent,
-                        struct iatt *postparent)
-
+                        struct iatt *postparent, dict_t *xdata)
 {
-        xlator_list_t *trav = NULL;
         stripe_local_t *local = NULL;
 
         if (!this || !frame || !frame->local) {
@@ -1225,11 +1243,10 @@ stripe_first_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 goto err;
         }
 
-        trav = this->children;
         local = frame->local;
+        local->op_ret = 0;
 
         local->call_count--; /* First child successful */
-        trav = trav->next; /* Skip first child */
 
         local->preparent  = *preparent;
         local->postparent = *postparent;
@@ -1238,22 +1255,60 @@ stripe_first_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         local->preparent_blocks  += preparent->ia_blocks;
         local->postparent_blocks += postparent->ia_blocks;
 
-        while (trav) {
-                STACK_WIND (frame, stripe_unlink_cbk, trav->xlator,
-                            trav->xlator->fops->rmdir, &local->loc,
-                            local->flags);
-                trav = trav->next;
-        }
-
+        STRIPE_STACK_UNWIND (rmdir, frame, local->op_ret, local->op_errno,
+                             &local->preparent, &local->postparent, xdata);
         return 0;
 err:
-        STRIPE_STACK_UNWIND (rmdir, frame, op_ret, op_errno, NULL, NULL);
+        STRIPE_STACK_UNWIND (rmdir, frame, op_ret, op_errno, NULL, NULL, NULL);
         return 0;
 
 }
 
 int32_t
-stripe_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
+stripe_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+        int32_t         callcnt = 0;
+        stripe_local_t *local   = NULL;
+        call_frame_t   *prev = NULL;
+
+        if (!this || !frame || !frame->local || !cookie) {
+                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
+                goto out;
+        }
+
+        prev  = cookie;
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                callcnt = --local->call_count;
+
+                if (op_ret == -1) {
+                        gf_log (this->name, GF_LOG_DEBUG, "%s returned %s",
+                                prev->this->name, strerror (op_errno));
+                        if (op_errno != ENOENT)
+                                local->failed = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        if (callcnt == 1) {
+                if (local->failed)
+                        goto out;
+                STACK_WIND (frame, stripe_first_rmdir_cbk, FIRST_CHILD (this),
+                            FIRST_CHILD (this)->fops->rmdir, &local->loc,
+                            local->flags, NULL);
+        }
+        return 0;
+out:
+        STRIPE_STACK_UNWIND (rmdir, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
+int32_t
+stripe_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags, dict_t *xdata)
 {
         xlator_list_t    *trav = NULL;
         stripe_local_t   *local = NULL;
@@ -1276,8 +1331,7 @@ stripe_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -1287,13 +1341,17 @@ stripe_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
         loc_copy (&local->loc, loc);
         local->flags = flags;
         local->call_count = priv->child_count;
+        trav = trav->next; /* skip the first child */
 
-        STACK_WIND (frame, stripe_first_rmdir_cbk,  trav->xlator,
-                    trav->xlator->fops->rmdir, loc, flags);
+        while (trav) {
+                STACK_WIND (frame, stripe_rmdir_cbk,  trav->xlator,
+                            trav->xlator->fops->rmdir, loc, flags, NULL);
+                trav = trav->next;
+        }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (rmdir, frame, -1, op_errno, NULL, NULL);
+        STRIPE_STACK_UNWIND (rmdir, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
@@ -1302,7 +1360,7 @@ int32_t
 stripe_mknod_ifreg_fail_unlink_cbk (call_frame_t *frame, void *cookie,
                                     xlator_t *this, int32_t op_ret,
                                     int32_t op_errno, struct iatt *preparent,
-                                    struct iatt *postparent)
+                                    struct iatt *postparent, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
@@ -1323,7 +1381,7 @@ stripe_mknod_ifreg_fail_unlink_cbk (call_frame_t *frame, void *cookie,
         if (!callcnt) {
                 STRIPE_STACK_UNWIND (mknod, frame, local->op_ret, local->op_errno,
                                      local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
+                                     &local->preparent, &local->postparent, NULL);
         }
 out:
         return 0;
@@ -1335,7 +1393,7 @@ out:
 int32_t
 stripe_mknod_ifreg_setxattr_cbk (call_frame_t *frame, void *cookie,
                                  xlator_t *this, int32_t op_ret,
-                                 int32_t op_errno)
+                                 int32_t op_errno, dict_t *xdata)
 {
         int32_t           callcnt = 0;
         stripe_local_t   *local = NULL;
@@ -1374,7 +1432,7 @@ stripe_mknod_ifreg_setxattr_cbk (call_frame_t *frame, void *cookie,
                                             stripe_mknod_ifreg_fail_unlink_cbk,
                                             trav->xlator,
                                             trav->xlator->fops->unlink,
-                                            &local->loc);
+                                            &local->loc, 0, NULL);
                                 trav = trav->next;
                         }
                         return 0;
@@ -1382,7 +1440,7 @@ stripe_mknod_ifreg_setxattr_cbk (call_frame_t *frame, void *cookie,
 
                 STRIPE_STACK_UNWIND (mknod, frame, local->op_ret, local->op_errno,
                                      local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
+                                     &local->preparent, &local->postparent, NULL);
         }
 out:
         return 0;
@@ -1392,7 +1450,7 @@ int32_t
 stripe_mknod_ifreg_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         int32_t op_ret, int32_t op_errno, inode_t *inode,
                         struct iatt *buf, struct iatt *preparent,
-                        struct iatt *postparent)
+                        struct iatt *postparent, dict_t *xdata)
 {
         int32_t           callcnt = 0;
         stripe_local_t   *local = NULL;
@@ -1406,7 +1464,7 @@ stripe_mknod_ifreg_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         }
 
         prev  = cookie;
-	priv  = this->private;
+        priv  = this->private;
         local = frame->local;
 
         LOCK (&frame->lock);
@@ -1422,25 +1480,24 @@ stripe_mknod_ifreg_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                 local->failed = 1;
                         local->op_errno = op_errno;
                 }
-
                 if (op_ret >= 0) {
                         local->op_ret = op_ret;
 
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->stbuf      = *buf;
-                                local->preparent  = *preparent;
-                                local->postparent = *postparent;
-                        }
-
                         /* Can be used as a mechanism to understand if mknod
                            was successful in at least one place */
                         if (uuid_is_null (local->ia_gfid))
                                 uuid_copy (local->ia_gfid, buf->ia_gfid);
 
+			if (stripe_ctx_handle(this, prev, local, xdata))
+				gf_log(this->name, GF_LOG_ERROR,
+					"Error getting fctx info from dict");
+
                         local->stbuf_blocks += buf->ia_blocks;
                         local->preparent_blocks  += preparent->ia_blocks;
                         local->postparent_blocks += postparent->ia_blocks;
 
+			correct_file_size(buf, local->fctx, prev);
+
                         if (local->stbuf_size < buf->ia_size)
                                 local->stbuf_size = buf->ia_size;
                         if (local->preparent_size < preparent->ia_size)
@@ -1465,7 +1522,7 @@ stripe_mknod_ifreg_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                             stripe_mknod_ifreg_fail_unlink_cbk,
                                             trav->xlator,
                                             trav->xlator->fops->unlink,
-                                            &local->loc);
+                                            &local->loc, 0, NULL);
                                 trav = trav->next;
                         }
                         return 0;
@@ -1479,13 +1536,13 @@ stripe_mknod_ifreg_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->postparent.ia_size   = local->postparent_size;
                         local->stbuf.ia_size        = local->stbuf_size;
                         local->stbuf.ia_blocks      = local->stbuf_blocks;
-                }
+                        inode_ctx_put (local->inode, this,
+                                       (uint64_t)(long) local->fctx);
 
-                /* Create itself has failed.. so return
-                   without setxattring */
+                }
                 STRIPE_STACK_UNWIND (mknod, frame, local->op_ret, local->op_errno,
                                      local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
+                                     &local->preparent, &local->postparent, NULL);
         }
 out:
         return 0;
@@ -1493,29 +1550,118 @@ out:
 
 
 int32_t
+stripe_mknod_first_ifreg_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, inode_t *inode,
+                        struct iatt *buf, struct iatt *preparent,
+                        struct iatt *postparent, dict_t *xdata)
+{
+        stripe_local_t   *local = NULL;
+        stripe_private_t *priv = NULL;
+        call_frame_t     *prev = NULL;
+        xlator_list_t    *trav = NULL;
+        int               i    = 1;
+        dict_t           *dict           = NULL;
+        int               ret            = 0;
+        int               need_unref     = 0;
+
+        if (!this || !frame || !frame->local || !cookie) {
+                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
+                goto out;
+        }
+
+        prev  = cookie;
+        priv  = this->private;
+        local = frame->local;
+        trav = this->children;
+
+        local->call_count--;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_DEBUG, "%s returned error %s",
+                        prev->this->name, strerror (op_errno));
+                local->failed = 1;
+                local->op_errno = op_errno;
+                goto out;
+        }
+
+        local->op_ret = op_ret;
+
+        local->stbuf      = *buf;
+        local->preparent  = *preparent;
+        local->postparent = *postparent;
+
+        if (uuid_is_null (local->ia_gfid))
+                uuid_copy (local->ia_gfid, buf->ia_gfid);
+        local->preparent.ia_blocks  = local->preparent_blocks;
+        local->preparent.ia_size    = local->preparent_size;
+        local->postparent.ia_blocks = local->postparent_blocks;
+        local->postparent.ia_size   = local->postparent_size;
+        local->stbuf.ia_size        = local->stbuf_size;
+        local->stbuf.ia_blocks      = local->stbuf_blocks;
+
+        trav = trav->next;
+        while (trav) {
+                if (priv->xattr_supported) {
+                        dict = dict_new ();
+                        if (!dict) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "failed to allocate dict %s", local->loc.path);
+                        }
+                        need_unref = 1;
+
+                        dict_copy (local->xattr, dict);
+
+                        ret = stripe_xattr_request_build (this, dict,
+                                                          local->stripe_size,
+                                                          priv->child_count, i,
+							  priv->coalesce);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "Failed to build xattr request");
+
+                } else {
+                        dict = local->xattr;
+                }
+
+                STACK_WIND (frame, stripe_mknod_ifreg_cbk,
+                            trav->xlator, trav->xlator->fops->mknod,
+                            &local->loc, local->mode, local->rdev, 0, dict);
+                trav = trav->next;
+                i++;
+
+                if (dict && need_unref)
+                        dict_unref (dict);
+        }
+
+        return 0;
+
+out:
+
+       STRIPE_STACK_UNWIND (mknod, frame, op_ret, op_errno, NULL, NULL, NULL, NULL, NULL);
+       return 0;
+}
+
+
+int32_t
 stripe_single_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                          int32_t op_ret, int32_t op_errno, inode_t *inode,
                          struct iatt *buf, struct iatt *preparent,
-                         struct iatt *postparent)
+                         struct iatt *postparent, dict_t *xdata)
 {
         STRIPE_STACK_UNWIND (mknod, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
+                             preparent, postparent, xdata);
         return 0;
 }
 
 
 int
 stripe_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-              dev_t rdev, dict_t *params)
+              dev_t rdev, mode_t umask, dict_t *xdata)
 {
         stripe_private_t *priv           = NULL;
         stripe_local_t   *local          = NULL;
-        xlator_list_t    *trav           = NULL;
         int32_t           op_errno       = EINVAL;
         int32_t           i              = 0;
-        char              size_key[256]  = {0,};
-        char              index_key[256] = {0,};
-        char              count_key[256] = {0,};
         dict_t           *dict           = NULL;
         int               ret            = 0;
         int               need_unref     = 0;
@@ -1527,7 +1673,6 @@ stripe_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
         VALIDATE_OR_GOTO (loc->inode, err);
 
         priv = this->private;
-        trav = this->children;
 
         if (priv->first_child_down) {
                 op_errno = ENOTCONN;
@@ -1546,86 +1691,63 @@ stripe_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
                 }
 
                 /* Initialization */
-                local = GF_CALLOC (1, sizeof (stripe_local_t),
-                                   gf_stripe_mt_stripe_local_t);
+                local = mem_get0 (this->local_pool);
                 if (!local) {
                         op_errno = ENOMEM;
                         goto err;
                 }
                 local->op_ret = -1;
                 local->op_errno = ENOTCONN;
-                local->stripe_size = stripe_get_matching_bs (loc->path,
-                                                             priv->pattern,
-                                                             priv->block_size);
+                local->stripe_size = stripe_get_matching_bs (loc->path, priv);
                 frame->local = local;
                 local->inode = inode_ref (loc->inode);
                 loc_copy (&local->loc, loc);
+                local->xattr = dict_copy_with_ref (xdata, NULL);
+                local->mode = mode;
+                local->umask = umask;
+                local->rdev = rdev;
 
                 /* Everytime in stripe lookup, all child nodes should
                    be looked up */
                 local->call_count = priv->child_count;
 
-                /* Send a setxattr request to nodes where the
-                   files are created */
-                sprintf (size_key,
-                         "trusted.%s.stripe-size", this->name);
-                sprintf (count_key,
-                         "trusted.%s.stripe-count", this->name);
-                sprintf (index_key,
-                         "trusted.%s.stripe-index", this->name);
-
-                while (trav) {
-                        if (priv->xattr_supported) {
-                                dict = dict_new ();
-                                if (!dict) {
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "failed to allocate dict %s", loc->path);
-                                }
-                                need_unref = 1;
+                if (priv->xattr_supported) {
+                        dict = dict_new ();
+                        if (!dict) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "failed to allocate dict %s", loc->path);
+                        }
+                        need_unref = 1;
 
-                                dict_copy (params, dict);
+                        dict_copy (xdata, dict);
 
-                                ret = dict_set_int64 (dict, size_key,
-                                                      local->stripe_size);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s: set stripe-size failed", loc->path);
-                                ret = dict_set_int32 (dict, count_key,
-                                                      priv->child_count);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s: set child_count failed",
-                                                loc->path);
-                                ret = dict_set_int32 (dict, index_key, i);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s: set stripe-index failed",
-                                                loc->path);
-                        } else {
-                                dict = params;
-                        }
+                        ret = stripe_xattr_request_build (this, dict,
+                                                          local->stripe_size,
+                                                          priv->child_count,
+                                                          i, priv->coalesce);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "failed to build xattr request");
+                } else {
+                        dict = xdata;
+                }
 
-                        STACK_WIND (frame, stripe_mknod_ifreg_cbk,
-                                    trav->xlator, trav->xlator->fops->mknod,
-                                    loc, mode, rdev, dict);
-                        trav = trav->next;
-                        i++;
+                STACK_WIND (frame, stripe_mknod_first_ifreg_cbk,
+                            FIRST_CHILD (this), FIRST_CHILD (this)->fops->mknod,
+                            loc, mode, rdev, umask, dict);
 
                         if (dict && need_unref)
                                 dict_unref (dict);
-                }
-
-                /* This case is handled, no need to continue further. */
                 return 0;
         }
 
         STACK_WIND (frame, stripe_single_mknod_cbk,
                     FIRST_CHILD(this), FIRST_CHILD(this)->fops->mknod,
-                    loc, mode, rdev, params);
+                    loc, mode, rdev, umask, xdata);
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+        STRIPE_STACK_UNWIND (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
         return 0;
 }
 
@@ -1634,7 +1756,7 @@ int32_t
 stripe_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                   int32_t op_ret, int32_t op_errno, inode_t *inode,
                   struct iatt *buf, struct iatt *preparent,
-                  struct iatt *postparent)
+                  struct iatt *postparent, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t  *local   = NULL;
@@ -1665,12 +1787,6 @@ stripe_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 if (op_ret >= 0) {
                         local->op_ret = 0;
 
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->inode      = inode_ref (inode);
-                                local->stbuf      = *buf;
-                                local->postparent = *postparent;
-                                local->preparent  = *preparent;
-                        }
                         local->stbuf_blocks      += buf->ia_blocks;
                         local->preparent_blocks  += preparent->ia_blocks;
                         local->postparent_blocks += postparent->ia_blocks;
@@ -1686,10 +1802,7 @@ stripe_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         UNLOCK (&frame->lock);
 
         if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
+                if (local->failed != -1) {
                         local->preparent.ia_blocks  = local->preparent_blocks;
                         local->preparent.ia_size    = local->preparent_size;
                         local->postparent.ia_blocks = local->postparent_blocks;
@@ -1700,16 +1813,76 @@ stripe_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 STRIPE_STACK_UNWIND (mkdir, frame, local->op_ret,
                                      local->op_errno, local->inode,
                                      &local->stbuf, &local->preparent,
-                                     &local->postparent);
+                                     &local->postparent, NULL);
         }
 out:
         return 0;
 }
 
 
+int32_t
+stripe_first_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, inode_t *inode,
+                        struct iatt *buf, struct iatt *preparent,
+                        struct iatt *postparent, dict_t *xdata)
+{
+        stripe_local_t  *local   = NULL;
+        call_frame_t    *prev = NULL;
+        xlator_list_t        *trav = NULL;
+
+        if (!this || !frame || !frame->local || !cookie) {
+                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
+                goto out;
+        }
+
+        prev  = cookie;
+        local = frame->local;
+        trav = this->children;
+
+        local->call_count--; /* first child is successful */
+        trav = trav->next;   /* skip first child */
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_DEBUG, "%s returned error %s",
+                        prev->this->name, strerror (op_errno));
+                local->op_errno = op_errno;
+                goto out;
+        }
+
+        local->op_ret = 0;
+
+        local->inode      = inode_ref (inode);
+        local->stbuf      = *buf;
+        local->postparent = *postparent;
+        local->preparent  = *preparent;
+
+        local->stbuf_blocks      += buf->ia_blocks;
+        local->preparent_blocks  += preparent->ia_blocks;
+        local->postparent_blocks += postparent->ia_blocks;
+
+        local->stbuf_size = buf->ia_size;
+        local->preparent_size = preparent->ia_size;
+        local->postparent_size = postparent->ia_size;
+
+        while (trav) {
+                STACK_WIND (frame, stripe_mkdir_cbk, trav->xlator,
+                            trav->xlator->fops->mkdir, &local->loc, local->mode,
+                            local->umask, local->xdata);
+                trav = trav->next;
+        }
+        return 0;
+out:
+        STRIPE_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL,
+                             NULL, NULL);
+
+        return 0;
+
+}
+
+
 int
 stripe_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-              dict_t *params)
+              mode_t umask, dict_t *xdata)
 {
         stripe_private_t *priv = NULL;
         stripe_local_t   *local = NULL;
@@ -1731,27 +1904,27 @@ stripe_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
         local->op_ret = -1;
         local->call_count = priv->child_count;
+        if (xdata)
+                local->xdata = dict_ref (xdata);
+        local->mode  = mode;
+        local->umask = umask;
+        loc_copy (&local->loc, loc);
         frame->local = local;
 
         /* Everytime in stripe lookup, all child nodes should be looked up */
-        while (trav) {
-                STACK_WIND (frame, stripe_mkdir_cbk,
-                            trav->xlator, trav->xlator->fops->mkdir,
-                            loc, mode, params);
-                trav = trav->next;
-        }
+        STACK_WIND (frame, stripe_first_mkdir_cbk, trav->xlator,
+                    trav->xlator->fops->mkdir, loc, mode, umask, xdata);
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+        STRIPE_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
         return 0;
 }
 
@@ -1760,11 +1933,12 @@ int32_t
 stripe_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                  int32_t op_ret, int32_t op_errno, inode_t *inode,
                  struct iatt *buf, struct iatt *preparent,
-                 struct iatt *postparent)
+                 struct iatt *postparent, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t  *local   = NULL;
         call_frame_t    *prev = NULL;
+	stripe_fd_ctx_t *fctx = NULL;
 
         if (!this || !frame || !frame->local || !cookie) {
                 gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
@@ -1791,6 +1965,16 @@ stripe_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 if (op_ret >= 0) {
                         local->op_ret = 0;
 
+			if (IA_ISREG(inode->ia_type)) {
+				inode_ctx_get(inode, this, (uint64_t *) &fctx);
+				if (!fctx) {
+					gf_log(this->name, GF_LOG_ERROR,
+						"failed to get stripe context");
+					op_ret = -1;
+					op_errno = EINVAL;
+				}
+			}
+
                         if (FIRST_CHILD(this) == prev->this) {
                                 local->inode      = inode_ref (inode);
                                 local->stbuf      = *buf;
@@ -1801,6 +1985,8 @@ stripe_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->preparent_blocks  += preparent->ia_blocks;
                         local->postparent_blocks += postparent->ia_blocks;
 
+			correct_file_size(buf, fctx, prev);
+
                         if (local->stbuf_size < buf->ia_size)
                                 local->stbuf_size = buf->ia_size;
                         if (local->preparent_size < preparent->ia_size)
@@ -1826,14 +2012,14 @@ stripe_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 STRIPE_STACK_UNWIND (link, frame, local->op_ret,
                                      local->op_errno, local->inode,
                                      &local->stbuf, &local->preparent,
-                                     &local->postparent);
+                                     &local->postparent, NULL);
         }
 out:
         return 0;
 }
 
 int32_t
-stripe_link (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc)
+stripe_link (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc, dict_t *xdata)
 {
         xlator_list_t    *trav = NULL;
         stripe_local_t   *local = NULL;
@@ -1856,8 +2042,7 @@ stripe_link (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc)
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -1871,13 +2056,13 @@ stripe_link (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc)
         while (trav) {
                 STACK_WIND (frame, stripe_link_cbk,
                             trav->xlator, trav->xlator->fops->link,
-                            oldloc, newloc);
+                            oldloc, newloc, NULL);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (link, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+        STRIPE_STACK_UNWIND (link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
         return 0;
 }
 
@@ -1885,7 +2070,7 @@ int32_t
 stripe_create_fail_unlink_cbk (call_frame_t *frame, void *cookie,
                                xlator_t *this, int32_t op_ret,
                                int32_t op_errno, struct iatt *preparent,
-                               struct iatt *postparent)
+                               struct iatt *postparent, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
@@ -1906,7 +2091,7 @@ stripe_create_fail_unlink_cbk (call_frame_t *frame, void *cookie,
         if (!callcnt) {
                 STRIPE_STACK_UNWIND (create, frame, local->op_ret, local->op_errno,
                                      local->fd, local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
+                                     &local->preparent, &local->postparent, NULL);
         }
 out:
         return 0;
@@ -1917,12 +2102,11 @@ int32_t
 stripe_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                    int32_t op_ret, int32_t op_errno, fd_t *fd,
                    inode_t *inode, struct iatt *buf, struct iatt *preparent,
-                   struct iatt *postparent)
+                   struct iatt *postparent, dict_t *xdata)
 {
         int32_t           callcnt = 0;
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
-        stripe_fd_ctx_t  *fctx = NULL;
         call_frame_t     *prev = NULL;
         xlator_list_t    *trav = NULL;
 
@@ -1943,26 +2127,26 @@ stripe_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "%s returned error %s",
                                 prev->this->name, strerror (op_errno));
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
+                        local->failed = 1;
                         local->op_errno = op_errno;
                 }
 
                 if (op_ret >= 0) {
+			if (IA_ISREG(buf->ia_type)) {
+				if (stripe_ctx_handle(this, prev, local, xdata))
+					gf_log(this->name, GF_LOG_ERROR,
+						"Error getting fctx info from "
+						"dict");
+			}
+
                         local->op_ret = op_ret;
-                        /* Get the mapping in inode private */
-                        /* Get the stat buf right */
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->stbuf      = *buf;
-                                local->preparent  = *preparent;
-                                local->postparent = *postparent;
-                        }
 
                         local->stbuf_blocks += buf->ia_blocks;
                         local->preparent_blocks  += preparent->ia_blocks;
                         local->postparent_blocks += postparent->ia_blocks;
 
+			correct_file_size(buf, local->fctx, prev);
+
                         if (local->stbuf_size < buf->ia_size)
                                 local->stbuf_size = buf->ia_size;
                         if (local->preparent_size < preparent->ia_size)
@@ -1985,7 +2169,7 @@ stripe_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                             stripe_create_fail_unlink_cbk,
                                             trav->xlator,
                                             trav->xlator->fops->unlink,
-                                            &local->loc);
+                                            &local->loc, 0, NULL);
                                 trav = trav->next;
                         }
 
@@ -2000,29 +2184,19 @@ stripe_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->stbuf.ia_size        = local->stbuf_size;
                         local->stbuf.ia_blocks      = local->stbuf_blocks;
 
-                        fctx = GF_CALLOC (1, sizeof (stripe_fd_ctx_t),
-                                          gf_stripe_mt_stripe_fd_ctx_t);
-                        if (!fctx) {
-                                local->op_ret = -1;
-                                local->op_errno = ENOMEM;
-                                goto unwind;
-                        }
-
-                        fctx->stripe_size  = local->stripe_size;
-                        fctx->stripe_count = priv->child_count;
-                        fctx->static_array = 1;
-                        fctx->xl_array = priv->xl_array;
-                        fd_ctx_set (local->fd, this,
-                                    (uint64_t)(long)fctx);
+			stripe_copy_xl_array(local->fctx->xl_array,
+				             priv->xl_array,
+					     local->fctx->stripe_count);
+			inode_ctx_put(local->inode, this,
+					(uint64_t) local->fctx);
                 }
 
-        unwind:
                 /* Create itself has failed.. so return
                    without setxattring */
                 STRIPE_STACK_UNWIND (create, frame, local->op_ret,
                                      local->op_errno, local->fd,
                                      local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
+                                     &local->preparent, &local->postparent, NULL);
         }
 
 out:
@@ -2030,6 +2204,122 @@ out:
 }
 
 
+
+int32_t
+stripe_first_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, fd_t *fd,
+                   inode_t *inode, struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+        stripe_local_t   *local = NULL;
+        stripe_private_t *priv = NULL;
+        call_frame_t     *prev = NULL;
+        xlator_list_t    *trav = NULL;
+        int               i    = 1;
+        dict_t           *dict = NULL;
+        loc_t            *loc  = NULL;
+        int32_t           need_unref = 0;
+        int32_t           ret  = -1;
+
+        if (!this || !frame || !frame->local || !cookie) {
+                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
+                goto out;
+        }
+
+        prev  = cookie;
+        priv  = this->private;
+        local = frame->local;
+        trav = this->children;
+        loc = &local->loc;
+
+        --local->call_count;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_DEBUG, "%s returned error %s",
+                        prev->this->name, strerror (op_errno));
+                 local->failed = 1;
+                 local->op_errno = op_errno;
+        }
+
+        local->op_ret = 0;
+        /* Get the mapping in inode private */
+        /* Get the stat buf right */
+        local->stbuf      = *buf;
+        local->preparent  = *preparent;
+        local->postparent = *postparent;
+
+        local->stbuf_blocks += buf->ia_blocks;
+        local->preparent_blocks  += preparent->ia_blocks;
+        local->postparent_blocks += postparent->ia_blocks;
+
+        if (local->stbuf_size < buf->ia_size)
+              local->stbuf_size = buf->ia_size;
+        if (local->preparent_size < preparent->ia_size)
+              local->preparent_size = preparent->ia_size;
+        if (local->postparent_size < postparent->ia_size)
+              local->postparent_size = postparent->ia_size;
+
+        if (local->failed)
+                local->op_ret = -1;
+
+        if (local->op_ret == -1) {
+                local->call_count = 1;
+                STACK_WIND (frame, stripe_create_fail_unlink_cbk,
+                            FIRST_CHILD (this), FIRST_CHILD (this)->fops->unlink,
+                            &local->loc, 0, NULL);
+                return 0;
+        }
+
+        if (local->op_ret >= 0) {
+                local->preparent.ia_blocks  = local->preparent_blocks;
+                local->preparent.ia_size    = local->preparent_size;
+                local->postparent.ia_blocks = local->postparent_blocks;
+                local->postparent.ia_size   = local->postparent_size;
+                local->stbuf.ia_size        = local->stbuf_size;
+                local->stbuf.ia_blocks      = local->stbuf_blocks;
+        }
+
+        /* Send a setxattr request to nodes where the
+           files are created */
+        trav = trav->next;
+        while (trav) {
+                if (priv->xattr_supported) {
+                        dict = dict_new ();
+                        if (!dict) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "failed to allocate dict %s", loc->path);
+                        }
+                        need_unref = 1;
+
+                        dict_copy (local->xattr, dict);
+
+                        ret = stripe_xattr_request_build (this, dict,
+                                                          local->stripe_size,
+                                                          priv->child_count,
+                                                          i, priv->coalesce);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "failed to build xattr request");
+                } else {
+                        dict = local->xattr;
+                }
+
+                STACK_WIND (frame, stripe_create_cbk, trav->xlator,
+                            trav->xlator->fops->create, &local->loc,
+                            local->flags, local->mode, local->umask, local->fd,
+                            dict);
+                trav = trav->next;
+                if (need_unref && dict)
+                        dict_unref (dict);
+                i++;
+        }
+
+out:
+        return 0;
+}
+
+
+
 /**
  * stripe_create - If a block-size is specified for the 'name', create the
  *    file in all the child nodes. If not, create it in only first child.
@@ -2038,18 +2328,14 @@ out:
  */
 int32_t
 stripe_create (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               int32_t flags, mode_t mode, fd_t *fd, dict_t *params)
+               int32_t flags, mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
 {
         stripe_private_t *priv = NULL;
         stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
         int32_t           op_errno = EINVAL;
         int               ret            = 0;
         int               need_unref     = 0;
         int               i              = 0;
-        char              size_key[256]  = {0,};
-        char              index_key[256] = {0,};
-        char              count_key[256] = {0,};
         dict_t           *dict           = NULL;
 
         VALIDATE_OR_GOTO (frame, err);
@@ -2071,82 +2357,68 @@ stripe_create (call_frame_t *frame, xlator_t *this, loc_t *loc,
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
         local->op_ret = -1;
         local->op_errno = ENOTCONN;
-        local->stripe_size = stripe_get_matching_bs (loc->path,
-                                                     priv->pattern,
-                                                     priv->block_size);
+        local->stripe_size = stripe_get_matching_bs (loc->path, priv);
         frame->local = local;
         local->inode = inode_ref (loc->inode);
         loc_copy (&local->loc, loc);
         local->fd = fd_ref (fd);
+        local->flags = flags;
+        local->mode = mode;
+        local->umask = umask;
+        if (xdata)
+                local->xattr = dict_ref (xdata);
 
         local->call_count = priv->child_count;
         /* Send a setxattr request to nodes where the
            files are created */
-        sprintf (size_key, "trusted.%s.stripe-size", this->name);
-        sprintf (count_key, "trusted.%s.stripe-count", this->name);
-        sprintf (index_key, "trusted.%s.stripe-index", this->name);
 
-        trav = this->children;
-        while (trav) {
-                if (priv->xattr_supported) {
-                        dict = dict_new ();
-                        if (!dict) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "failed to allocate dict %s", loc->path);
-                        }
-                        need_unref = 1;
+        if (priv->xattr_supported) {
+                dict = dict_new ();
+                if (!dict) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "failed to allocate dict %s", loc->path);
+                }
+                need_unref = 1;
 
-                        dict_copy (params, dict);
+                dict_copy (xdata, dict);
 
-                        ret = dict_set_int64 (dict, size_key,
-                                              local->stripe_size);
-                        if (ret)
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "%s: set stripe-size failed", loc->path);
-                        ret = dict_set_int32 (dict, count_key,
-                                              priv->child_count);
-                        if (ret)
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "%s: set child_count failed",
-                                        loc->path);
-                        ret = dict_set_int32 (dict, index_key, i);
-                        if (ret)
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "%s: set stripe-index failed",
-                                        loc->path);
-                } else {
-                        dict = params;
-                }
+                ret = stripe_xattr_request_build (this, dict,
+                                                  local->stripe_size,
+                                                  priv->child_count,
+                                                  i, priv->coalesce);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "failed to build xattr request");
+        } else {
+                        dict = xdata;
+        }
 
-                STACK_WIND (frame, stripe_create_cbk, trav->xlator,
-                            trav->xlator->fops->create, loc, flags,
-                            mode, fd, dict);
-                trav = trav->next;
-                i++;
 
-                if (need_unref && dict)
-                        dict_unref (dict);
+        STACK_WIND (frame, stripe_first_create_cbk, FIRST_CHILD (this),
+                    FIRST_CHILD (this)->fops->create, loc, flags, mode,
+                    umask, fd, dict);
+
+        if (need_unref && dict)
+                dict_unref (dict);
 
-        }
 
         return 0;
 err:
         STRIPE_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL,
-                             NULL, NULL);
+                             NULL, NULL, xdata);
         return 0;
 }
 
 int32_t
 stripe_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, fd_t *fd)
+                 int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
@@ -2184,224 +2456,25 @@ stripe_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 if (local->failed)
                         local->op_ret = -1;
 
-                if (local->op_ret == -1) {
-                        if (local->fctx) {
-                                if (!local->fctx->static_array)
-                                        GF_FREE (local->fctx->xl_array);
-                                GF_FREE (local->fctx);
-                        }
-                } else {
-                        fd_ctx_set (local->fd, this,
-                                    (uint64_t)(long)local->fctx);
-                }
-
                 STRIPE_STACK_UNWIND (open, frame, local->op_ret,
-                                     local->op_errno, local->fd);
+                                     local->op_errno, local->fd, xdata);
         }
 out:
         return 0;
 }
 
-int32_t
-stripe_open_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, inode_t *inode,
-                        struct iatt *buf, dict_t *dict, struct iatt *postparent)
-{
-        int32_t           index = 0;
-        int32_t           callcnt = 0;
-        char              key[256] = {0,};
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        stripe_private_t *priv = NULL;
-        data_t           *data = NULL;
-        call_frame_t     *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = (call_frame_t *)cookie;
-        priv  = this->private;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_ret = -1;
-                        if (local->op_errno != EIO)
-                                local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                        goto unlock;
-                }
-
-                if (!dict)
-                        goto unlock;
-
-                if (!local->fctx) {
-                        local->fctx =  GF_CALLOC (1, sizeof (stripe_fd_ctx_t),
-                                                  gf_stripe_mt_stripe_fd_ctx_t);
-                        if (!local->fctx) {
-                                local->op_errno = ENOMEM;
-                                local->op_ret = -1;
-                                goto unlock;
-                        }
-
-                        local->fctx->static_array = 0;
-                }
-                /* Stripe block size */
-                sprintf (key, "trusted.%s.stripe-size", this->name);
-                data = dict_get (dict, key);
-                if (!data) {
-                        local->xattr_self_heal_needed = 1;
-                } else {
-                        if (!local->fctx->stripe_size) {
-                                local->fctx->stripe_size =
-                                        data_to_int64 (data);
-                        }
-
-                        if (local->fctx->stripe_size != data_to_int64 (data)) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "stripe-size mismatch in blocks");
-                                local->xattr_self_heal_needed = 1;
-                        }
-                }
-                /* Stripe count */
-                sprintf (key, "trusted.%s.stripe-count", this->name);
-                data = dict_get (dict, key);
-                if (!data) {
-                        local->xattr_self_heal_needed = 1;
-                        goto unlock;
-                }
-                if (!local->fctx->xl_array) {
-                        local->fctx->stripe_count = data_to_int32 (data);
-                        if (!local->fctx->stripe_count) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "error with stripe-count xattr");
-                                local->op_ret   = -1;
-                                local->op_errno = EIO;
-                                goto unlock;
-                        }
-
-                        local->fctx->xl_array =
-                                GF_CALLOC (local->fctx->stripe_count,
-                                           sizeof (xlator_t *),
-                                           gf_stripe_mt_xlator_t);
-                        if (!local->fctx->xl_array) {
-                                local->op_errno = ENOMEM;
-                                local->op_ret   = -1;
-                                goto unlock;
-                        }
-                }
-                if (local->fctx->stripe_count != data_to_int32 (data)) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "error with stripe-count xattr (%d != %d)",
-                                local->fctx->stripe_count, data_to_int32 (data));
-                        local->op_ret   = -1;
-                        local->op_errno = EIO;
-                        goto unlock;
-                }
-
-                /* index */
-                sprintf (key, "trusted.%s.stripe-index", this->name);
-                data = dict_get (dict, key);
-                if (!data) {
-                        local->xattr_self_heal_needed = 1;
-                        goto unlock;
-                }
-                index = data_to_int32 (data);
-                if (index > priv->child_count) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "error with stripe-index xattr (%d)", index);
-                        local->op_ret   = -1;
-                        local->op_errno = EIO;
-                        goto unlock;
-                }
-                if (local->fctx->xl_array) {
-                        if (local->fctx->xl_array[index]) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "duplicate entry @ index (%d)", index);
-                                local->op_ret   = -1;
-                                local->op_errno = EIO;
-                                goto unlock;
-                        }
-                        local->fctx->xl_array[index] = prev->this;
-                }
-                local->entry_count++;
-                local->op_ret = 0;
-        }
-unlock:
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                /* TODO: if self-heal flag is set, do it */
-                if (local->xattr_self_heal_needed) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s: stripe info need to be healed",
-                                local->loc.path);
-                }
-
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret)
-                        goto err;
-
-                if (local->entry_count != local->fctx->stripe_count) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "entry-count (%d) != stripe-count (%d)",
-                                local->entry_count, local->fctx->stripe_count);
-                        local->op_ret = -1;
-                        local->op_errno = EIO;
-                        goto err;
-                }
-                if (!local->fctx->stripe_size) {
-                        gf_log (this->name, GF_LOG_ERROR, "stripe size not set");
-                        local->op_ret = -1;
-                        local->op_errno = EIO;
-                        goto err;
-                }
-
-                local->call_count = local->fctx->stripe_count;
-
-                trav = this->children;
-                while (trav) {
-                        STACK_WIND (frame, stripe_open_cbk, trav->xlator,
-                                    trav->xlator->fops->open, &local->loc,
-                                    local->flags, local->fd, 0);
-                        trav = trav->next;
-                }
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (open, frame, local->op_ret, local->op_errno,
-                             local->fd);
-out:
-        return 0;
-}
 
 /**
  * stripe_open -
  */
 int32_t
 stripe_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             int32_t flags, fd_t *fd, int32_t wbflags)
+             int32_t flags, fd_t *fd, dict_t *xdata)
 {
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
         xlator_list_t    *trav = NULL;
         int32_t           op_errno = 1;
-        dict_t           *dict = NULL;
-        int               ret = 0;
-        char              key[256] = {0,};
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -2418,8 +2491,7 @@ stripe_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -2435,73 +2507,25 @@ stripe_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
         /* Striped files */
         local->flags = flags;
         local->call_count = priv->child_count;
-        local->stripe_size = stripe_get_matching_bs (loc->path,
-                                                     priv->pattern,
-                                                     priv->block_size);
-
-        if (priv->xattr_supported) {
-                dict = dict_new ();
-                if (!dict)
-                        goto err;
-
-                sprintf (key, "trusted.%s.stripe-size", this->name);
-                ret = dict_set_int64 (dict, key, 8);
-                if (ret)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set %s in xattr_req dict", key);
-
-                sprintf (key, "trusted.%s.stripe-count", this->name);
-                ret = dict_set_int32 (dict, key, 4);
-                if (ret)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set %s in xattr_req dict", key);
-
-                sprintf (key, "trusted.%s.stripe-index", this->name);
-                ret = dict_set_int32 (dict, key, 4);
-                if (ret)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set %s in xattr_req dict", key);
-
-                while (trav) {
-                        STACK_WIND (frame, stripe_open_lookup_cbk,
-                                    trav->xlator, trav->xlator->fops->lookup,
-                                    loc, dict);
-                        trav = trav->next;
-                }
-                if (dict)
-                        dict_unref (dict);
-
-                return 0;
-        }
-        local->fctx =  GF_CALLOC (1, sizeof (stripe_fd_ctx_t),
-                                  gf_stripe_mt_stripe_fd_ctx_t);
-        if (!local->fctx) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        local->fctx->static_array = 1;
-        local->fctx->stripe_size  = local->stripe_size;
-        local->fctx->stripe_count = priv->child_count;
-        local->fctx->xl_array     = priv->xl_array;
+        local->stripe_size = stripe_get_matching_bs (loc->path, priv);
 
         while (trav) {
                 STACK_WIND (frame, stripe_open_cbk, trav->xlator,
                             trav->xlator->fops->open,
                             &local->loc, local->flags, local->fd,
-                            wbflags);
+                            xdata);
                 trav = trav->next;
         }
         return 0;
 err:
-        STRIPE_STACK_UNWIND (open, frame, -1, op_errno, NULL);
+        STRIPE_STACK_UNWIND (open, frame, -1, op_errno, NULL, NULL);
         return 0;
 }
 
 
 int32_t
 stripe_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, fd_t *fd)
+                    int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
@@ -2534,7 +2558,7 @@ stripe_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         if (!callcnt) {
                 STRIPE_STACK_UNWIND (opendir, frame, local->op_ret,
-                                     local->op_errno, local->fd);
+                                     local->op_errno, local->fd, NULL);
         }
 out:
         return 0;
@@ -2542,7 +2566,7 @@ out:
 
 
 int32_t
-stripe_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
+stripe_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd, dict_t *xdata)
 {
         xlator_list_t    *trav = NULL;
         stripe_local_t   *local = NULL;
@@ -2564,8 +2588,7 @@ stripe_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -2576,19 +2599,19 @@ stripe_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
 
         while (trav) {
                 STACK_WIND (frame, stripe_opendir_cbk, trav->xlator,
-                            trav->xlator->fops->opendir, loc, fd);
+                            trav->xlator->fops->opendir, loc, fd, NULL);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (opendir, frame, -1, op_errno, NULL);
+        STRIPE_STACK_UNWIND (opendir, frame, -1, op_errno, NULL, NULL);
         return 0;
 }
 
 int32_t
 stripe_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int32_t op_ret, int32_t op_errno, struct gf_flock *lock)
+               int32_t op_ret, int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
@@ -2628,7 +2651,7 @@ stripe_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 if (local->failed)
                         local->op_ret = -1;
                 STRIPE_STACK_UNWIND (lk, frame, local->op_ret,
-                                     local->op_errno, &local->lock);
+                                     local->op_errno, &local->lock, NULL);
         }
 out:
         return 0;
@@ -2636,7 +2659,7 @@ out:
 
 int32_t
 stripe_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
-           struct gf_flock *lock)
+           struct gf_flock *lock, dict_t *xdata)
 {
         stripe_local_t   *local = NULL;
         xlator_list_t    *trav = NULL;
@@ -2652,8 +2675,7 @@ stripe_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
         priv = this->private;
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -2664,20 +2686,20 @@ stripe_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
 
         while (trav) {
                 STACK_WIND (frame, stripe_lk_cbk, trav->xlator,
-                            trav->xlator->fops->lk, fd, cmd, lock);
+                            trav->xlator->fops->lk, fd, cmd, lock, NULL);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (lk, frame, -1, op_errno, NULL);
+        STRIPE_STACK_UNWIND (lk, frame, -1, op_errno, NULL, NULL);
         return 0;
 }
 
 
 int32_t
 stripe_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno)
+                  int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local   = NULL;
@@ -2714,14 +2736,14 @@ stripe_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->op_ret = -1;
 
                 STRIPE_STACK_UNWIND (flush, frame, local->op_ret,
-                                     local->op_errno);
+                                     local->op_errno, NULL);
         }
 out:
         return 0;
 }
 
 int32_t
-stripe_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
+stripe_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
@@ -2741,8 +2763,7 @@ stripe_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
                 goto err;
         }
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -2753,13 +2774,13 @@ stripe_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
 
         while (trav) {
                 STACK_WIND (frame, stripe_flush_cbk,  trav->xlator,
-                            trav->xlator->fops->flush, fd);
+                            trav->xlator->fops->flush, fd, NULL);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (flush, frame, -1, op_errno);
+        STRIPE_STACK_UNWIND (flush, frame, -1, op_errno, NULL);
         return 0;
 }
 
@@ -2768,7 +2789,7 @@ err:
 int32_t
 stripe_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                  struct iatt *postbuf)
+                  struct iatt *postbuf, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local   = NULL;
@@ -2804,6 +2825,9 @@ stripe_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->prebuf_blocks  += prebuf->ia_blocks;
                         local->postbuf_blocks += postbuf->ia_blocks;
 
+			correct_file_size(prebuf, local->fctx, prev);
+			correct_file_size(postbuf, local->fctx, prev);
+
                         if (local->prebuf_size < prebuf->ia_size)
                                 local->prebuf_size = prebuf->ia_size;
 
@@ -2826,18 +2850,19 @@ stripe_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
                 STRIPE_STACK_UNWIND (fsync, frame, local->op_ret,
                                      local->op_errno, &local->pre_buf,
-                                     &local->post_buf);
+                                     &local->post_buf, NULL);
         }
 out:
         return 0;
 }
 
 int32_t
-stripe_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
+stripe_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags, dict_t *xdata)
 {
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
         xlator_list_t    *trav = NULL;
+	stripe_fd_ctx_t  *fctx = NULL;
         int32_t           op_errno = 1;
 
         VALIDATE_OR_GOTO (frame, err);
@@ -2849,31 +2874,38 @@ stripe_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
         trav = this->children;
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
-        local->op_ret = -1;
+
         frame->local = local;
+
+	inode_ctx_get(fd->inode, this, (uint64_t *) &fctx);
+	if (!fctx) {
+		op_errno = EINVAL;
+		goto err;
+	}
+	local->fctx = fctx;
+        local->op_ret = -1;
         local->call_count = priv->child_count;
 
         while (trav) {
                 STACK_WIND (frame, stripe_fsync_cbk, trav->xlator,
-                            trav->xlator->fops->fsync, fd, flags);
+                            trav->xlator->fops->fsync, fd, flags, NULL);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL);
+        STRIPE_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
 int32_t
 stripe_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct iatt *buf)
+                  int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
@@ -2908,6 +2940,9 @@ stripe_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                 local->stbuf = *buf;
 
                         local->stbuf_blocks += buf->ia_blocks;
+
+			correct_file_size(buf, local->fctx, prev);
+
                         if (local->stbuf_size < buf->ia_size)
                                 local->stbuf_size = buf->ia_size;
                 }
@@ -2924,7 +2959,7 @@ stripe_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 }
 
                 STRIPE_STACK_UNWIND (fstat, frame, local->op_ret,
-                                     local->op_errno, &local->stbuf);
+                                     local->op_errno, &local->stbuf, NULL);
         }
 
 out:
@@ -2934,11 +2969,12 @@ out:
 int32_t
 stripe_fstat (call_frame_t *frame,
               xlator_t *this,
-              fd_t *fd)
+              fd_t *fd, dict_t *xdata)
 {
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
         xlator_list_t    *trav = NULL;
+	stripe_fd_ctx_t  *fctx = NULL;
         int32_t           op_errno = 1;
 
         VALIDATE_OR_GOTO (frame, err);
@@ -2950,8 +2986,7 @@ stripe_fstat (call_frame_t *frame,
         trav = this->children;
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -2960,26 +2995,35 @@ stripe_fstat (call_frame_t *frame,
         frame->local = local;
         local->call_count = priv->child_count;
 
+	if (IA_ISREG(fd->inode->ia_type)) {
+		inode_ctx_get(fd->inode, this, (uint64_t *) &fctx);
+		if (!fctx)
+			goto err;
+		local->fctx = fctx;
+	}
+
         while (trav) {
                 STACK_WIND (frame, stripe_fstat_cbk, trav->xlator,
-                            trav->xlator->fops->fstat, fd);
+                            trav->xlator->fops->fstat, fd, NULL);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (fstat, frame, -1, op_errno, NULL);
+        STRIPE_STACK_UNWIND (fstat, frame, -1, op_errno, NULL, NULL);
         return 0;
 }
 
 
 int32_t
-stripe_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset)
+stripe_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, dict_t *xdata)
 {
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
+	stripe_fd_ctx_t  *fctx = NULL;
+	int		  i, eof_idx;
+	off_t		  dest_offset, tmp_offset;
+        int32_t		  op_errno = 1;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -2987,11 +3031,9 @@ stripe_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset)
         VALIDATE_OR_GOTO (fd->inode, err);
 
         priv = this->private;
-        trav = this->children;
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -3000,22 +3042,60 @@ stripe_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset)
         frame->local = local;
         local->call_count = priv->child_count;
 
-        while (trav) {
-                STACK_WIND (frame, stripe_truncate_cbk, trav->xlator,
-                            trav->xlator->fops->ftruncate, fd, offset);
-                trav = trav->next;
-        }
+	inode_ctx_get(fd->inode, this, (uint64_t *) &fctx);
+	if (!fctx) {
+		gf_log(this->name, GF_LOG_ERROR, "no stripe context");
+		op_errno = EINVAL;
+		goto err;
+	}
+	if (!fctx->stripe_count) {
+		gf_log(this->name, GF_LOG_ERROR, "no stripe count");
+		op_errno = EINVAL;
+		goto err;
+	}
+
+	local->fctx = fctx;
+	eof_idx = (offset / fctx->stripe_size) % fctx->stripe_count;
+
+	for (i = 0; i < fctx->stripe_count; i++) {
+		if (!fctx->xl_array[i]) {
+			gf_log(this->name, GF_LOG_ERROR, "no xlator at index "
+				"%d", i);
+			op_errno = EINVAL;
+			goto err;
+		}
+
+		if (fctx->stripe_coalesce) {
+			if (i < eof_idx)
+				tmp_offset = roof(offset, fctx->stripe_size *
+						fctx->stripe_count);
+			else if (i > eof_idx)
+				tmp_offset = floor(offset, fctx->stripe_size *
+						fctx->stripe_count);
+			else
+				tmp_offset = offset;
+
+			dest_offset = coalesced_offset(tmp_offset,
+				fctx->stripe_size, fctx->stripe_count);
+		} else {
+			dest_offset = offset;
+		}
+
+		STACK_WIND(frame, stripe_truncate_cbk, fctx->xl_array[i],
+			fctx->xl_array[i]->fops->ftruncate, fd, dest_offset,
+			NULL);
+	}
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL);
+        STRIPE_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
 
 int32_t
 stripe_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno)
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local   = NULL;
@@ -3052,14 +3132,14 @@ stripe_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->op_ret = -1;
 
                 STRIPE_STACK_UNWIND (fsyncdir, frame, local->op_ret,
-                                     local->op_errno);
+                                     local->op_errno, NULL);
         }
 out:
         return 0;
 }
 
 int32_t
-stripe_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
+stripe_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags, dict_t *xdata)
 {
         stripe_local_t   *local = NULL;
         stripe_private_t *priv = NULL;
@@ -3075,8 +3155,7 @@ stripe_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
         trav = this->children;
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -3087,20 +3166,20 @@ stripe_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
 
         while (trav) {
                 STACK_WIND (frame, stripe_fsyncdir_cbk, trav->xlator,
-                            trav->xlator->fops->fsyncdir, fd, flags);
+                            trav->xlator->fops->fsyncdir, fd, flags, NULL);
                 trav = trav->next;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (fsyncdir, frame, -1, op_errno);
+        STRIPE_STACK_UNWIND (fsyncdir, frame, -1, op_errno, NULL);
         return 0;
 }
 
 
 int32_t
 stripe_readv_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, struct iatt *buf)
+                        int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
 {
         int32_t         i = 0;
         int32_t         callcnt = 0;
@@ -3110,6 +3189,7 @@ stripe_readv_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         struct iatt     tmp_stbuf = {0,};
         struct iobref  *tmp_iobref = NULL;
         struct iobuf   *iobuf = NULL;
+	call_frame_t   *prev = NULL;
 
         if (!this || !frame || !frame->local) {
                 gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
@@ -3117,13 +3197,16 @@ stripe_readv_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         }
 
         local = frame->local;
+	prev = cookie;
 
         LOCK (&frame->lock);
         {
                 callcnt = --local->call_count;
-                if (op_ret != -1)
+                if (op_ret != -1) {
+			correct_file_size(buf, local->fctx, prev);
                         if (local->stbuf_size < buf->ia_size)
                                 local->stbuf_size = buf->ia_size;
+		}
         }
         UNLOCK (&frame->lock);
 
@@ -3152,7 +3235,8 @@ stripe_readv_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                 vec[count].iov_len  =
                                         (local->replies[i].requested_size -
                                          local->replies[i].op_ret);
-                                iobuf = iobuf_get (this->ctx->iobuf_pool);
+                                iobuf = iobuf_get2 (this->ctx->iobuf_pool,
+                                                    vec[count].iov_len);
                                 if (!iobuf) {
                                         gf_log (this->name, GF_LOG_ERROR,
                                                 "Out of memory.");
@@ -3161,9 +3245,11 @@ stripe_readv_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                         goto done;
                                 }
                                 memset (iobuf->ptr, 0, vec[count].iov_len);
-                                iobref_add (local->iobref, iobuf);
                                 vec[count].iov_base = iobuf->ptr;
 
+                                iobref_add (local->iobref, iobuf);
+                                iobuf_unref(iobuf);
+
                                 op_ret += vec[count].iov_len;
                                 count++;
                         }
@@ -3181,11 +3267,10 @@ stripe_readv_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 GF_FREE (local->replies);
                 tmp_iobref = local->iobref;
                 STRIPE_STACK_UNWIND (readv, frame, op_ret, op_errno, vec,
-                                     count, &tmp_stbuf, tmp_iobref);
+                                     count, &tmp_stbuf, tmp_iobref, NULL);
 
                 iobref_unref (tmp_iobref);
-                if (vec)
-                        GF_FREE (vec);
+                GF_FREE (vec);
         }
 out:
         return 0;
@@ -3198,7 +3283,7 @@ out:
 int32_t
 stripe_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                   int32_t op_ret, int32_t op_errno, struct iovec *vector,
-                  int32_t count, struct iatt *stbuf, struct iobref *iobref)
+                  int32_t count, struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
 {
         int32_t         index = 0;
         int32_t         callcnt = 0;
@@ -3209,8 +3294,10 @@ stripe_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         stripe_local_t *local = NULL;
         struct iovec   *final_vec = NULL;
         struct iatt     tmp_stbuf = {0,};
+        struct iatt    *tmp_stbuf_p = NULL; //need it for a warning
         struct iobref  *tmp_iobref = NULL;
         stripe_fd_ctx_t  *fctx = NULL;
+	call_frame_t	*prev = NULL;
 
         if (!this || !frame || !frame->local || !cookie) {
                 gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
@@ -3219,6 +3306,7 @@ stripe_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         local  = frame->local;
         index  = local->node_index;
+	prev = cookie;
         mframe = local->orig_frame;
         if (!mframe)
                 goto out;
@@ -3239,6 +3327,12 @@ stripe_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         mlocal->replies[index].count  = count;
                         mlocal->replies[index].vector = iov_dup (vector, count);
 
+			correct_file_size(stbuf, fctx, prev);
+
+                        if (local->stbuf_size < stbuf->ia_size)
+                                local->stbuf_size = stbuf->ia_size;
+                        local->stbuf_blocks += stbuf->ia_blocks;
+
                         if (!mlocal->iobref)
                                 mlocal->iobref = iobref_new ();
                         iobref_merge (mlocal->iobref, iobref);
@@ -3295,17 +3389,21 @@ stripe_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                  * cause any bugs at higher levels */
                 memcpy (&tmp_stbuf, &mlocal->replies[0].stbuf,
                         sizeof (struct iatt));
+                tmp_stbuf.ia_size = local->stbuf_size;
+                tmp_stbuf.ia_blocks = local->stbuf_blocks;
 
         done:
                 /* */
                 GF_FREE (mlocal->replies);
                 tmp_iobref = mlocal->iobref;
+                /* work around for nfs truncated read. Bug 3774 */
+                tmp_stbuf_p = &tmp_stbuf;
+                WIPE (tmp_stbuf_p);
                 STRIPE_STACK_UNWIND (readv, mframe, op_ret, op_errno, final_vec,
-                                     final_count, &tmp_stbuf, tmp_iobref);
+                                     final_count, &tmp_stbuf, tmp_iobref, NULL);
 
                 iobref_unref (tmp_iobref);
-                if (final_vec)
-                        GF_FREE (final_vec);
+                GF_FREE (final_vec);
         }
 
         goto out;
@@ -3317,7 +3415,7 @@ check_size:
                 STACK_WIND (mframe, stripe_readv_fstat_cbk,
                             (fctx->xl_array[index]),
                             (fctx->xl_array[index])->fops->fstat,
-                            mlocal->fd);
+                            mlocal->fd, NULL);
         }
 
 out:
@@ -3329,7 +3427,7 @@ end:
 
 int32_t
 stripe_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              size_t size, off_t offset)
+              size_t size, off_t offset, uint32_t flags, dict_t *xdata)
 {
         int32_t           op_errno = EINVAL;
         int32_t           idx = 0;
@@ -3342,6 +3440,7 @@ stripe_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
         uint64_t          stripe_size = 0;
         off_t             rounded_start = 0;
         off_t             frame_offset = offset;
+	off_t		  dest_offset = 0;
         stripe_local_t   *local = NULL;
         call_frame_t     *rframe = NULL;
         stripe_local_t   *rlocal = NULL;
@@ -3352,7 +3451,7 @@ stripe_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
         VALIDATE_OR_GOTO (fd, err);
         VALIDATE_OR_GOTO (fd->inode, err);
 
-        fd_ctx_get (fd, this, &tmp_fctx);
+        inode_ctx_get (fd->inode, this, &tmp_fctx);
         if (!tmp_fctx) {
                 op_errno = EBADFD;
                 goto err;
@@ -3360,6 +3459,8 @@ stripe_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
         fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
         stripe_size = fctx->stripe_size;
 
+        STRIPE_VALIDATE_FCTX (fctx, err);
+
         if (!stripe_size) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "Wrong stripe size for the file");
@@ -3374,8 +3475,7 @@ stripe_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
         rounded_end = roof (offset+size, stripe_size);
         num_stripe = (rounded_end- rounded_start)/stripe_size;
 
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -3383,8 +3483,8 @@ stripe_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
         frame->local = local;
 
         /* This is where all the vectors should be copied. */
-        local->replies = GF_CALLOC (num_stripe, sizeof (struct readv_replies),
-                                    gf_stripe_mt_readv_replies);
+        local->replies = GF_CALLOC (num_stripe, sizeof (struct stripe_replies),
+                                    gf_stripe_mt_stripe_replies);
         if (!local->replies) {
                 op_errno = ENOMEM;
                 goto err;
@@ -3399,8 +3499,7 @@ stripe_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
 
         for (index = off_index; index < (num_stripe + off_index); index++) {
                 rframe = copy_frame (frame);
-                rlocal = GF_CALLOC (1, sizeof (stripe_local_t),
-                                    gf_stripe_mt_stripe_local_t);
+                rlocal = mem_get0 (this->local_pool);
                 if (!rlocal) {
                         op_errno = ENOMEM;
                         goto err;
@@ -3414,9 +3513,16 @@ stripe_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
                 rlocal->readv_size = frame_size;
                 rframe->local = rlocal;
                 idx = (index % fctx->stripe_count);
+
+		if (fctx->stripe_coalesce)
+			dest_offset = coalesced_offset(frame_offset,
+				stripe_size, fctx->stripe_count);
+		else
+			dest_offset = frame_offset;
+
                 STACK_WIND (rframe, stripe_readv_cbk, fctx->xl_array[idx],
                             fctx->xl_array[idx]->fops->readv,
-                            fd, frame_size, frame_offset);
+                            fd, frame_size, dest_offset, flags, xdata);
 
                 frame_offset += frame_size;
         }
@@ -3426,7 +3532,7 @@ err:
         if (rframe)
                 STRIPE_STACK_DESTROY (rframe);
 
-        STRIPE_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL);
+        STRIPE_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
         return 0;
 }
 
@@ -3434,11 +3540,15 @@ err:
 int32_t
 stripe_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                   struct iatt *postbuf)
+                   struct iatt *postbuf, dict_t *xdata)
 {
         int32_t         callcnt = 0;
         stripe_local_t *local = NULL;
+	stripe_local_t *mlocal = NULL;
         call_frame_t   *prev = NULL;
+	call_frame_t   *mframe = NULL;
+	struct stripe_replies *reply = NULL;
+	int32_t		i = 0;
 
         if (!this || !frame || !frame->local || !cookie) {
                 gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
@@ -3447,39 +3557,82 @@ stripe_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         prev  = cookie;
         local = frame->local;
+	mframe = local->orig_frame;
+	mlocal = mframe->local;
 
         LOCK(&frame->lock);
         {
-                callcnt = ++local->call_count;
+                callcnt = ++mlocal->call_count;
+
+		mlocal->replies[local->node_index].op_ret = op_ret;
+		mlocal->replies[local->node_index].op_errno = op_errno;
 
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        local->op_ret = -1;
-                }
                 if (op_ret >= 0) {
-                        local->op_ret += op_ret;
-                        local->post_buf = *postbuf;
-                        local->pre_buf = *prebuf;
+                        mlocal->post_buf = *postbuf;
+                        mlocal->pre_buf = *prebuf;
+
+			mlocal->prebuf_blocks  += prebuf->ia_blocks;
+			mlocal->postbuf_blocks += postbuf->ia_blocks;
+
+			correct_file_size(prebuf, mlocal->fctx, prev);
+			correct_file_size(postbuf, mlocal->fctx, prev);
+
+			if (mlocal->prebuf_size < prebuf->ia_size)
+				mlocal->prebuf_size = prebuf->ia_size;
+			if (mlocal->postbuf_size < postbuf->ia_size)
+				mlocal->postbuf_size = postbuf->ia_size;
                 }
         }
         UNLOCK (&frame->lock);
 
-        if ((callcnt == local->wind_count) && local->unwind) {
-                STRIPE_STACK_UNWIND (writev, frame, local->op_ret,
-                                     local->op_errno, &local->pre_buf,
-                                     &local->post_buf);
+        if ((callcnt == mlocal->wind_count) && mlocal->unwind) {
+		mlocal->pre_buf.ia_size = mlocal->prebuf_size;
+		mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks;
+		mlocal->post_buf.ia_size = mlocal->postbuf_size;
+		mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks;
+
+		/*
+		 * Only return the number of consecutively written bytes up until
+		 * the first error. Only return an error if it occurs first.
+		 *
+		 * When a short write occurs, the application should retry at the
+		 * appropriate offset, at which point we'll potentially pass back
+		 * the error.
+		 */
+		for (i = 0, reply = mlocal->replies; i < mlocal->wind_count;
+			i++, reply++) {
+			if (reply->op_ret == -1) {
+				gf_log(this->name, GF_LOG_DEBUG, "reply %d "
+					"returned error %s", i,
+					strerror(reply->op_errno));
+				if (!mlocal->op_ret) {
+					mlocal->op_ret = -1;
+					mlocal->op_errno = reply->op_errno;
+				}
+				break;
+			}
+
+			mlocal->op_ret += reply->op_ret;
+
+			if (reply->op_ret < reply->requested_size)
+				break;
+		}
+
+		GF_FREE(mlocal->replies);
+
+                STRIPE_STACK_UNWIND (writev, mframe, mlocal->op_ret,
+                                     mlocal->op_errno, &mlocal->pre_buf,
+                                     &mlocal->post_buf, NULL);
         }
 out:
+	STRIPE_STACK_DESTROY(frame);
         return 0;
 }
 
 int32_t
 stripe_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
                struct iovec *vector, int32_t count, off_t offset,
-               struct iobref *iobref)
+               uint32_t flags, struct iobref *iobref, dict_t *xdata)
 {
         struct iovec     *tmp_vec = NULL;
         stripe_local_t   *local = NULL;
@@ -3493,13 +3646,19 @@ stripe_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
         off_t             fill_size = 0;
         uint64_t          stripe_size = 0;
         uint64_t          tmp_fctx = 0;
+	off_t		  dest_offset = 0;
+	off_t		  rounded_start = 0;
+	off_t		  rounded_end = 0;
+	int32_t		  total_chunks = 0;
+	call_frame_t	  *wframe = NULL;
+	stripe_local_t	  *wlocal = NULL;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
         VALIDATE_OR_GOTO (fd->inode, err);
 
-        fd_ctx_get (fd, this, &tmp_fctx);
+        inode_ctx_get (fd->inode, this, &tmp_fctx);
         if (!tmp_fctx) {
                 op_errno = EINVAL;
                 goto err;
@@ -3507,22 +3666,51 @@ stripe_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
         fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
         stripe_size = fctx->stripe_size;
 
+        STRIPE_VALIDATE_FCTX (fctx, err);
+
         /* File has to be stripped across the child nodes */
         for (idx = 0; idx< count; idx ++) {
                 total_size += vector[idx].iov_len;
         }
         remaining_size = total_size;
 
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
         }
         frame->local = local;
         local->stripe_size = stripe_size;
+	local->fctx = fctx;
 
+        if (!stripe_size) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Wrong stripe size for the file");
+                op_errno = EINVAL;
+                goto err;
+        }
+
+	rounded_start = floor(offset, stripe_size);
+	rounded_end = roof(offset + total_size, stripe_size);
+	total_chunks = (rounded_end - rounded_start) / stripe_size;
+	local->replies = GF_CALLOC(total_chunks, sizeof(struct stripe_replies),
+				gf_stripe_mt_stripe_replies);
+	if (!local->replies) {
+		op_errno = ENOMEM;
+		goto err;
+	}
+
+	total_chunks = 0;
         while (1) {
+		wframe = copy_frame(frame);
+		wlocal = mem_get0(this->local_pool);
+		if (!wlocal) {
+			op_errno = ENOMEM;
+			goto err;
+		}
+		wlocal->orig_frame = frame;
+		wframe->local = wlocal;
+
                 /* Send striped chunk of the vector to child
                    nodes appropriately. */
                 idx = (((offset + offset_offset) /
@@ -3550,32 +3738,554 @@ stripe_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
                 if (remaining_size == 0)
                         local->unwind = 1;
 
-                STACK_WIND (frame, stripe_writev_cbk, fctx->xl_array[idx],
+		/*
+		 * Store off the request index (with respect to the chunk of the
+		 * initial offset) and the size of the request. This is required
+		 * in the callback to calculate an appropriate return value in
+		 * the event of a write failure in one or more requests.
+		 */
+		wlocal->node_index = total_chunks;
+		local->replies[total_chunks].requested_size = fill_size;
+
+		dest_offset = offset + offset_offset;
+		if (fctx->stripe_coalesce)
+			dest_offset = coalesced_offset(dest_offset,
+					local->stripe_size, fctx->stripe_count);
+
+                STACK_WIND (wframe, stripe_writev_cbk, fctx->xl_array[idx],
                             fctx->xl_array[idx]->fops->writev, fd, tmp_vec,
-                            tmp_count, offset + offset_offset, iobref);
+                            tmp_count, dest_offset, flags, iobref,
+                            xdata);
+
                 GF_FREE (tmp_vec);
                 offset_offset += fill_size;
+		total_chunks++;
                 if (remaining_size == 0)
                         break;
         }
 
         return 0;
 err:
-        STRIPE_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL);
+	if (wframe)
+		STRIPE_STACK_DESTROY(wframe);
+
+        STRIPE_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL, NULL);
         return 0;
 }
 
 
 int32_t
-stripe_release (xlator_t *this, fd_t *fd)
+stripe_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+        int32_t         callcnt = 0;
+        stripe_local_t *local = NULL;
+	stripe_local_t *mlocal = NULL;
+        call_frame_t   *prev = NULL;
+	call_frame_t   *mframe = NULL;
+
+        if (!this || !frame || !frame->local || !cookie) {
+                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
+                goto out;
+        }
+
+        prev  = cookie;
+        local = frame->local;
+	mframe = local->orig_frame;
+	mlocal = mframe->local;
+
+        LOCK(&frame->lock);
+        {
+                callcnt = ++mlocal->call_count;
+
+                if (op_ret == 0) {
+                        mlocal->post_buf = *postbuf;
+                        mlocal->pre_buf = *prebuf;
+
+			mlocal->prebuf_blocks  += prebuf->ia_blocks;
+			mlocal->postbuf_blocks += postbuf->ia_blocks;
+
+			correct_file_size(prebuf, mlocal->fctx, prev);
+			correct_file_size(postbuf, mlocal->fctx, prev);
+
+			if (mlocal->prebuf_size < prebuf->ia_size)
+				mlocal->prebuf_size = prebuf->ia_size;
+			if (mlocal->postbuf_size < postbuf->ia_size)
+				mlocal->postbuf_size = postbuf->ia_size;
+                }
+
+		/* return the first failure */
+		if (mlocal->op_ret == 0) {
+			mlocal->op_ret = op_ret;
+			mlocal->op_errno = op_errno;
+		}
+        }
+        UNLOCK (&frame->lock);
+
+        if ((callcnt == mlocal->wind_count) && mlocal->unwind) {
+		mlocal->pre_buf.ia_size = mlocal->prebuf_size;
+		mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks;
+		mlocal->post_buf.ia_size = mlocal->postbuf_size;
+		mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks;
+
+                STRIPE_STACK_UNWIND (fallocate, mframe, mlocal->op_ret,
+                                     mlocal->op_errno, &mlocal->pre_buf,
+                                     &mlocal->post_buf, NULL);
+        }
+out:
+	STRIPE_STACK_DESTROY(frame);
+        return 0;
+}
+
+int32_t
+stripe_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+		 off_t offset, size_t len, dict_t *xdata)
 {
+        stripe_local_t   *local = NULL;
+        stripe_fd_ctx_t  *fctx = NULL;
+        int32_t           op_errno = 1;
+        int32_t           idx = 0;
+        int32_t           offset_offset = 0;
+        int32_t           remaining_size = 0;
+        off_t             fill_size = 0;
+        uint64_t          stripe_size = 0;
         uint64_t          tmp_fctx = 0;
+	off_t		  dest_offset = 0;
+	call_frame_t	  *fframe = NULL;
+	stripe_local_t	  *flocal = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (fd->inode, err);
+
+        inode_ctx_get (fd->inode, this, &tmp_fctx);
+        if (!tmp_fctx) {
+                op_errno = EINVAL;
+                goto err;
+        }
+        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
+        stripe_size = fctx->stripe_size;
+
+        STRIPE_VALIDATE_FCTX (fctx, err);
+
+        remaining_size = len;
+
+        local = mem_get0 (this->local_pool);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+        frame->local = local;
+        local->stripe_size = stripe_size;
+	local->fctx = fctx;
+
+        if (!stripe_size) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Wrong stripe size for the file");
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        while (1) {
+		fframe = copy_frame(frame);
+		flocal = mem_get0(this->local_pool);
+		if (!flocal) {
+			op_errno = ENOMEM;
+			goto err;
+		}
+		flocal->orig_frame = frame;
+		fframe->local = flocal;
+
+		/* send fallocate request to the associated child node */
+                idx = (((offset + offset_offset) /
+                        local->stripe_size) % fctx->stripe_count);
+
+                fill_size = (local->stripe_size -
+                             ((offset + offset_offset) % local->stripe_size));
+                if (fill_size > remaining_size)
+                        fill_size = remaining_size;
+
+                remaining_size -= fill_size;
+
+                local->wind_count++;
+                if (remaining_size == 0)
+                        local->unwind = 1;
+
+		dest_offset = offset + offset_offset;
+		if (fctx->stripe_coalesce)
+			dest_offset = coalesced_offset(dest_offset,
+					local->stripe_size, fctx->stripe_count);
+
+		/*
+		 * TODO: Create a separate handler for coalesce mode that sends a
+		 * single fallocate per-child (since the ranges are linear).
+		 */
+		STACK_WIND(fframe, stripe_fallocate_cbk, fctx->xl_array[idx],
+			   fctx->xl_array[idx]->fops->fallocate, fd, mode,
+			   dest_offset, fill_size, xdata);
+
+                offset_offset += fill_size;
+                if (remaining_size == 0)
+                        break;
+        }
+
+        return 0;
+err:
+	if (fframe)
+		STRIPE_STACK_DESTROY(fframe);
+
+        STRIPE_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
+
+int32_t
+stripe_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+        int32_t         callcnt = 0;
+        stripe_local_t *local = NULL;
+	stripe_local_t *mlocal = NULL;
+        call_frame_t   *prev = NULL;
+	call_frame_t   *mframe = NULL;
+
+        if (!this || !frame || !frame->local || !cookie) {
+                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
+                goto out;
+        }
+
+        prev  = cookie;
+        local = frame->local;
+	mframe = local->orig_frame;
+	mlocal = mframe->local;
+
+        LOCK(&frame->lock);
+        {
+                callcnt = ++mlocal->call_count;
+
+                if (op_ret == 0) {
+                        mlocal->post_buf = *postbuf;
+                        mlocal->pre_buf = *prebuf;
+
+			mlocal->prebuf_blocks  += prebuf->ia_blocks;
+			mlocal->postbuf_blocks += postbuf->ia_blocks;
+
+			correct_file_size(prebuf, mlocal->fctx, prev);
+			correct_file_size(postbuf, mlocal->fctx, prev);
+
+			if (mlocal->prebuf_size < prebuf->ia_size)
+				mlocal->prebuf_size = prebuf->ia_size;
+			if (mlocal->postbuf_size < postbuf->ia_size)
+				mlocal->postbuf_size = postbuf->ia_size;
+                }
+
+		/* return the first failure */
+		if (mlocal->op_ret == 0) {
+			mlocal->op_ret = op_ret;
+			mlocal->op_errno = op_errno;
+		}
+        }
+        UNLOCK (&frame->lock);
+
+        if ((callcnt == mlocal->wind_count) && mlocal->unwind) {
+		mlocal->pre_buf.ia_size = mlocal->prebuf_size;
+		mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks;
+		mlocal->post_buf.ia_size = mlocal->postbuf_size;
+		mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks;
+
+                STRIPE_STACK_UNWIND (discard, mframe, mlocal->op_ret,
+                                     mlocal->op_errno, &mlocal->pre_buf,
+                                     &mlocal->post_buf, NULL);
+        }
+out:
+	STRIPE_STACK_DESTROY(frame);
+        return 0;
+}
+
+int32_t
+stripe_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+	       size_t len, dict_t *xdata)
+{
+        stripe_local_t   *local = NULL;
         stripe_fd_ctx_t  *fctx = NULL;
+        int32_t           op_errno = 1;
+        int32_t           idx = 0;
+        int32_t           offset_offset = 0;
+        int32_t           remaining_size = 0;
+        off_t             fill_size = 0;
+        uint64_t          stripe_size = 0;
+        uint64_t          tmp_fctx = 0;
+	off_t		  dest_offset = 0;
+	call_frame_t	  *fframe = NULL;
+	stripe_local_t	  *flocal = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (fd->inode, err);
+
+        inode_ctx_get (fd->inode, this, &tmp_fctx);
+        if (!tmp_fctx) {
+                op_errno = EINVAL;
+                goto err;
+        }
+        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
+        stripe_size = fctx->stripe_size;
+
+        STRIPE_VALIDATE_FCTX (fctx, err);
+
+        remaining_size = len;
+
+        local = mem_get0 (this->local_pool);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+        frame->local = local;
+        local->stripe_size = stripe_size;
+	local->fctx = fctx;
+
+        if (!stripe_size) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Wrong stripe size for the file");
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        while (1) {
+		fframe = copy_frame(frame);
+		flocal = mem_get0(this->local_pool);
+		if (!flocal) {
+			op_errno = ENOMEM;
+			goto err;
+		}
+		flocal->orig_frame = frame;
+		fframe->local = flocal;
+
+		/* send discard request to the associated child node */
+                idx = (((offset + offset_offset) /
+                        local->stripe_size) % fctx->stripe_count);
+
+                fill_size = (local->stripe_size -
+                             ((offset + offset_offset) % local->stripe_size));
+                if (fill_size > remaining_size)
+                        fill_size = remaining_size;
+
+                remaining_size -= fill_size;
+
+                local->wind_count++;
+                if (remaining_size == 0)
+                        local->unwind = 1;
+
+		dest_offset = offset + offset_offset;
+		if (fctx->stripe_coalesce)
+			dest_offset = coalesced_offset(dest_offset,
+					local->stripe_size, fctx->stripe_count);
+
+		/*
+		 * TODO: Create a separate handler for coalesce mode that sends a
+		 * single discard per-child (since the ranges are linear).
+		 */
+		STACK_WIND(fframe, stripe_discard_cbk, fctx->xl_array[idx],
+			   fctx->xl_array[idx]->fops->discard, fd, dest_offset,
+			   fill_size, xdata);
+
+                offset_offset += fill_size;
+                if (remaining_size == 0)
+                        break;
+        }
+
+        return 0;
+err:
+	if (fframe)
+		STRIPE_STACK_DESTROY(fframe);
+
+        STRIPE_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
+int32_t
+stripe_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+        int32_t         callcnt    = 0;
+        stripe_local_t *local      = NULL;
+        stripe_local_t *mlocal     = NULL;
+        call_frame_t   *prev       = NULL;
+        call_frame_t   *mframe     = NULL;
+
+        GF_ASSERT (frame);
+
+        if (!this || !frame->local || !cookie) {
+                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
+                goto out;
+        }
+
+        prev  = cookie;
+        local = frame->local;
+        mframe = local->orig_frame;
+        mlocal = mframe->local;
+
+        LOCK(&frame->lock);
+        {
+                callcnt = ++mlocal->call_count;
+
+                if (op_ret == 0) {
+                        mlocal->post_buf = *postbuf;
+                        mlocal->pre_buf = *prebuf;
+
+                        mlocal->prebuf_blocks  += prebuf->ia_blocks;
+                        mlocal->postbuf_blocks += postbuf->ia_blocks;
+
+                        correct_file_size(prebuf, mlocal->fctx, prev);
+                        correct_file_size(postbuf, mlocal->fctx, prev);
+
+                        if (mlocal->prebuf_size < prebuf->ia_size)
+                                mlocal->prebuf_size = prebuf->ia_size;
+                        if (mlocal->postbuf_size < postbuf->ia_size)
+                                mlocal->postbuf_size = postbuf->ia_size;
+                }
+
+                /* return the first failure */
+                if (mlocal->op_ret == 0) {
+                        mlocal->op_ret = op_ret;
+                        mlocal->op_errno = op_errno;
+                }
+        }
+        UNLOCK (&frame->lock);
 
+        if ((callcnt == mlocal->wind_count) && mlocal->unwind) {
+                mlocal->pre_buf.ia_size = mlocal->prebuf_size;
+                mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks;
+                mlocal->post_buf.ia_size = mlocal->postbuf_size;
+                mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks;
+
+                STRIPE_STACK_UNWIND (zerofill, mframe, mlocal->op_ret,
+                                     mlocal->op_errno, &mlocal->pre_buf,
+                                     &mlocal->post_buf, NULL);
+        }
+out:
+        STRIPE_STACK_DESTROY(frame);
+        return 0;
+}
+
+int32_t
+stripe_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               off_t len, dict_t *xdata)
+{
+        stripe_local_t   *local            = NULL;
+        stripe_fd_ctx_t  *fctx             = NULL;
+        int32_t           op_errno         = 1;
+        int32_t           idx              = 0;
+        int32_t           offset_offset    = 0;
+        int32_t           remaining_size   = 0;
+        off_t             fill_size        = 0;
+        uint64_t          stripe_size      = 0;
+        uint64_t          tmp_fctx         = 0;
+        off_t             dest_offset      = 0;
+        call_frame_t     *fframe           = NULL;
+        stripe_local_t   *flocal           = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (fd->inode, err);
 
-        fd_ctx_del (fd, this, &tmp_fctx);
+        inode_ctx_get (fd->inode, this, &tmp_fctx);
+        if (!tmp_fctx) {
+                op_errno = EINVAL;
+                goto err;
+        }
+        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
+        stripe_size = fctx->stripe_size;
+
+        STRIPE_VALIDATE_FCTX (fctx, err);
+
+        remaining_size = len;
+
+        local = mem_get0 (this->local_pool);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+        frame->local = local;
+        local->stripe_size = stripe_size;
+        local->fctx = fctx;
+
+        if (!stripe_size) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Wrong stripe size for the file");
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        while (1) {
+                fframe = copy_frame(frame);
+                flocal = mem_get0(this->local_pool);
+                if (!flocal) {
+                        op_errno = ENOMEM;
+                        goto err;
+                }
+                flocal->orig_frame = frame;
+                fframe->local = flocal;
+
+                idx = (((offset + offset_offset) /
+                        local->stripe_size) % fctx->stripe_count);
+
+                fill_size = (local->stripe_size -
+                             ((offset + offset_offset) % local->stripe_size));
+                if (fill_size > remaining_size)
+                        fill_size = remaining_size;
+
+                remaining_size -= fill_size;
+
+                local->wind_count++;
+                if (remaining_size == 0)
+                        local->unwind = 1;
+
+                dest_offset = offset + offset_offset;
+                if (fctx->stripe_coalesce)
+                        dest_offset = coalesced_offset(dest_offset,
+                                                       local->stripe_size,
+                                                       fctx->stripe_count);
+
+                STACK_WIND(fframe, stripe_zerofill_cbk, fctx->xl_array[idx],
+                           fctx->xl_array[idx]->fops->zerofill, fd,
+                           dest_offset, fill_size, xdata);
+                offset_offset += fill_size;
+                if (remaining_size == 0)
+                        break;
+        }
+
+        return 0;
+err:
+        if (fframe)
+                STRIPE_STACK_DESTROY(fframe);
+
+        STRIPE_STACK_UNWIND (zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
+int32_t
+stripe_release (xlator_t *this, fd_t *fd)
+{
+	return 0;
+}
+
+int
+stripe_forget (xlator_t *this, inode_t *inode)
+{
+        uint64_t          tmp_fctx = 0;
+        stripe_fd_ctx_t  *fctx = NULL;
+
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (inode, err);
+
+        (void) inode_ctx_del (inode, this, &tmp_fctx);
         if (!tmp_fctx) {
                 goto err;
         }
@@ -3586,18 +4296,17 @@ stripe_release (xlator_t *this, fd_t *fd)
                 GF_FREE (fctx->xl_array);
 
         GF_FREE (fctx);
-
 err:
-	return 0;
+        return 0;
 }
 
-
 int32_t
 notify (xlator_t *this, int32_t event, void *data, ...)
 {
         stripe_private_t *priv = NULL;
         int               down_client = 0;
         int               i = 0;
+        gf_boolean_t      heard_from_all_children = _gf_false;
 
         if (!this)
                 return 0;
@@ -3609,30 +4318,34 @@ notify (xlator_t *this, int32_t event, void *data, ...)
         switch (event)
         {
         case GF_EVENT_CHILD_UP:
-        case GF_EVENT_CHILD_CONNECTING:
         {
                 /* get an index number to set */
                 for (i = 0; i < priv->child_count; i++) {
                         if (data == priv->xl_array[i])
                                 break;
                 }
-                priv->state[i] = 1;
-                for (i = 0; i < priv->child_count; i++) {
-                        if (!priv->state[i])
-                                down_client++;
+
+                if (priv->child_count == i) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "got GF_EVENT_CHILD_UP bad subvolume %s",
+                                data? ((xlator_t *)data)->name: NULL);
+                        break;
                 }
 
                 LOCK (&priv->lock);
                 {
-                        priv->nodes_down = down_client;
                         if (data == FIRST_CHILD (this))
                                 priv->first_child_down = 0;
-                        if (!priv->nodes_down)
-                                default_notify (this, event, data);
+                        priv->last_event[i] = event;
                 }
                 UNLOCK (&priv->lock);
         }
         break;
+        case GF_EVENT_CHILD_CONNECTING:
+        {
+                // 'CONNECTING' doesn't ensure its CHILD_UP, so do nothing
+                goto out;
+        }
         case GF_EVENT_CHILD_DOWN:
         {
                 /* get an index number to set */
@@ -3640,20 +4353,19 @@ notify (xlator_t *this, int32_t event, void *data, ...)
                         if (data == priv->xl_array[i])
                                 break;
                 }
-                priv->state[i] = 0;
-                for (i = 0; i < priv->child_count; i++) {
-                        if (!priv->state[i])
-                                down_client++;
+
+                if (priv->child_count == i) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "got GF_EVENT_CHILD_DOWN bad subvolume %s",
+                                data? ((xlator_t *)data)->name: NULL);
+                        break;
                 }
 
                 LOCK (&priv->lock);
                 {
-                        priv->nodes_down = down_client;
-
                         if (data == FIRST_CHILD (this))
                                 priv->first_child_down = 1;
-                        if (priv->nodes_down)
-                                default_notify (this, event, data);
+                        priv->last_event[i] = event;
                 }
                 UNLOCK (&priv->lock);
         }
@@ -3663,79 +4375,252 @@ notify (xlator_t *this, int32_t event, void *data, ...)
         {
                 /* */
                 default_notify (this, event, data);
+                goto out;
         }
         break;
         }
 
+        // Consider child as down if it's last_event is not CHILD_UP
+        for (i = 0, down_client = 0; i < priv->child_count; i++)
+                if (priv->last_event[i] != GF_EVENT_CHILD_UP)
+                        down_client++;
+
+        LOCK (&priv->lock);
+        {
+                priv->nodes_down = down_client;
+        }
+        UNLOCK (&priv->lock);
+
+        heard_from_all_children = _gf_true;
+        for (i = 0; i < priv->child_count; i++)
+                if (!priv->last_event[i])
+                        heard_from_all_children = _gf_false;
+
+        if (heard_from_all_children)
+                default_notify (this, event, data);
+out:
         return 0;
 }
 
 int
-set_stripe_block_size (xlator_t *this, stripe_private_t *priv, char *data)
+stripe_setxattr_cbk (call_frame_t *frame, void *cookie,
+                     xlator_t *this, int op_ret, int op_errno, dict_t *xdata)
 {
-        int                    ret = -1;
-        char                  *tmp_str = NULL;
-        char                  *tmp_str1 = NULL;
-        char                  *dup_str = NULL;
-        char                  *stripe_str = NULL;
-        char                  *pattern = NULL;
-        char                  *num = NULL;
-        struct stripe_options *temp_stripeopt = NULL;
-        struct stripe_options *stripe_opt = NULL;
-
-        if (!this || !priv || !data)
-                goto out;
+        int ret = -1;
+        int call_cnt = 0;
+        stripe_local_t *local = NULL;
 
-        /* Get the pattern for striping.
-           "option block-size *avi:10MB" etc */
-        stripe_str = strtok_r (data, ",", &tmp_str);
-        while (stripe_str) {
-                dup_str = gf_strdup (stripe_str);
-                stripe_opt = CALLOC (1, sizeof (struct stripe_options));
-                if (!stripe_opt) {
-                        GF_FREE (dup_str);
-                        goto out;
-                }
+        if (!frame || !frame->local || !this) {
+                gf_log ("", GF_LOG_ERROR, "Possible NULL deref");
+                return ret;
+        }
 
-                pattern = strtok_r (dup_str, ":", &tmp_str1);
-                num = strtok_r (NULL, ":", &tmp_str1);
-                if (!num) {
-                        num = pattern;
-                        pattern = "*";
-                }
-                if (gf_string2bytesize (num, &stripe_opt->block_size) != 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "invalid number format \"%s\"", num);
-                        goto out;
-                }
+        local = frame->local;
 
-                if (stripe_opt->block_size < 512) {
-                        gf_log (this->name, GF_LOG_ERROR, "Invalid Block-size: "
-                                "%s. Should be atleast 512 bytes", num);
-                        goto out;
+        LOCK (&frame->lock);
+        {
+                call_cnt = --local->wind_count;
+
+                /**
+                 * We overwrite ->op_* values here for subsequent faliure
+                 * conditions, hence we propogate the last errno down the
+                 * stack.
+                 */
+                if (op_ret < 0) {
+                        local->op_ret = op_ret;
+                        local->op_errno = op_errno;
+                        goto unlock;
                 }
-                if (stripe_opt->block_size % 512) {
-                        gf_log (this->name, GF_LOG_ERROR, "Block-size: %s should"
-                                " be a multiple of 512 bytes", num);
-                        goto out;
+        }
+
+ unlock:
+        UNLOCK (&frame->lock);
+
+        if (!call_cnt) {
+                STRIPE_STACK_UNWIND (setxattr, frame, local->op_ret,
+                                     local->op_errno, xdata);
+        }
+
+        return 0;
+}
+
+#ifdef HAVE_BD_XLATOR
+int
+stripe_is_bd (dict_t *this, char *key, data_t *value, void *data)
+{
+        gf_boolean_t *is_bd = data;
+
+        if (data == NULL)
+                return 0;
+
+        if (XATTR_IS_BD (key))
+            *is_bd = _gf_true;
+
+        return 0;
+}
+
+static inline gf_boolean_t
+stripe_setxattr_is_bd (dict_t *dict)
+{
+        gf_boolean_t is_bd = _gf_false;
+
+        if (dict == NULL)
+                goto out;
+
+        dict_foreach (dict, stripe_is_bd, &is_bd);
+out:
+        return is_bd;
+}
+#else
+#define stripe_setxattr_is_bd(dict) _gf_false
+#endif
+
+int
+stripe_setxattr (call_frame_t *frame, xlator_t *this,
+                 loc_t *loc, dict_t *dict, int flags, dict_t *xdata)
+{
+        int32_t           op_errno = EINVAL;
+        xlator_list_t    *trav     = NULL;
+        stripe_private_t *priv     = NULL;
+        stripe_local_t   *local    = NULL;
+        int               i        = 0;
+        gf_boolean_t      is_bd    = _gf_false;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.*stripe*", dict,
+                                   op_errno, err);
+
+        priv = this->private;
+        trav = this->children;
+
+        local = mem_get0 (this->local_pool);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        frame->local = local;
+        local->wind_count = priv->child_count;
+        local->op_ret = local->op_errno = 0;
+
+        is_bd = stripe_setxattr_is_bd (dict);
+
+        /**
+         * Set xattrs for directories on all subvolumes. Additionally
+         * this power is only given to a special client. Bd xlator
+         * also needs xattrs for regular files (ie LVs)
+         */
+        if (((frame->root->pid == GF_CLIENT_PID_GSYNCD) &&
+             IA_ISDIR (loc->inode->ia_type)) || is_bd) {
+                for (i = 0; i < priv->child_count; i++, trav = trav->next) {
+                        STACK_WIND (frame, stripe_setxattr_cbk,
+                                    trav->xlator, trav->xlator->fops->setxattr,
+                                    loc, dict, flags, xdata);
                 }
+        } else {
+                local->wind_count = 1;
+                STACK_WIND (frame, stripe_setxattr_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->setxattr,
+                            loc, dict, flags, xdata);
+        }
 
-                memcpy (stripe_opt->path_pattern, pattern, strlen (pattern));
+        return 0;
+err:
+        STRIPE_STACK_UNWIND (setxattr, frame, -1,  op_errno, NULL);
+        return 0;
+}
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "block-size : pattern %s : size %"PRId64,
-                        stripe_opt->path_pattern, stripe_opt->block_size);
 
-                if (!priv->pattern) {
-                        priv->pattern = stripe_opt;
-                } else {
-                        temp_stripeopt = priv->pattern;
-                        while (temp_stripeopt->next)
-                                temp_stripeopt = temp_stripeopt->next;
-                        temp_stripeopt->next = stripe_opt;
+int
+stripe_fsetxattr_cbk (call_frame_t *frame, void *cookie,
+                      xlator_t *this, int op_ret, int op_errno, dict_t *xdata)
+{
+        STRIPE_STACK_UNWIND (fsetxattr, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+
+int
+stripe_is_special_key (dict_t *this,
+                       char *key,
+                       data_t *value,
+                       void *data)
+{
+        gf_boolean_t *is_special = NULL;
+
+        if (data == NULL) {
+                goto out;
+        }
+
+        is_special = data;
+
+        if (XATTR_IS_LOCKINFO (key) || XATTR_IS_BD (key))
+                *is_special = _gf_true;
+
+out:
+        return 0;
+}
+
+int32_t
+stripe_fsetxattr_everyone_cbk (call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               dict_t *xdata)
+{
+        int             call_count = 0;
+        stripe_local_t *local      = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                call_count = --local->wind_count;
+
+                if (op_ret < 0) {
+                        local->op_ret = op_ret;
+                        local->op_errno = op_errno;
                 }
-                stripe_str = strtok_r (NULL, ",", &tmp_str);
-                GF_FREE (dup_str);
+        }
+        UNLOCK (&frame->lock);
+
+        if (call_count == 0) {
+                STRIPE_STACK_UNWIND (fsetxattr, frame, local->op_ret,
+                                     local->op_errno, NULL);
+        }
+        return 0;
+}
+
+int
+stripe_fsetxattr_to_everyone (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                              dict_t *dict, int flags, dict_t *xdata)
+{
+        xlator_list_t    *trav  = NULL;
+        stripe_private_t *priv  = NULL;
+        int               ret   = -1;
+        stripe_local_t   *local = NULL;
+
+        priv = this->private;
+
+        local = mem_get0 (this->local_pool);
+        if (local == NULL) {
+                goto out;
+        }
+
+        frame->local = local;
+
+        local->wind_count = priv->child_count;
+
+        trav = this->children;
+
+        while (trav) {
+                STACK_WIND (frame, stripe_fsetxattr_everyone_cbk,
+                            trav->xlator, trav->xlator->fops->fsetxattr,
+                            fd, dict, flags, xdata);
+                trav = trav->next;
         }
 
         ret = 0;
@@ -3743,80 +4628,220 @@ out:
         return ret;
 }
 
-int32_t
-stripe_iatt_merge (struct iatt *from, struct iatt *to)
+static inline gf_boolean_t
+stripe_fsetxattr_is_special (dict_t *dict)
 {
-        if (to->ia_size < from->ia_size)
-                to->ia_size = from->ia_size;
-        if (to->ia_mtime < from->ia_mtime)
-                to->ia_mtime = from->ia_mtime;
-        if (to->ia_ctime < from->ia_ctime)
-                to->ia_ctime = from->ia_ctime;
-        if (to->ia_atime < from->ia_atime)
-                to->ia_atime = from->ia_atime;
-        return 0;
+        gf_boolean_t is_spl = _gf_false;
+
+        if (dict == NULL) {
+                goto out;
+        }
+
+        dict_foreach (dict, stripe_is_special_key, &is_spl);
+
+out:
+        return is_spl;
 }
 
-int32_t
-stripe_readdirp_entry_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                                int32_t op_ret, int32_t op_errno, struct iatt *buf)
+int
+stripe_fsetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  dict_t *dict, int flags, dict_t *xdata)
 {
-        gf_dirent_t    *entry = NULL;
-        stripe_local_t *local = NULL;
-        int32_t        done = 0;
+        int32_t      op_ret = -1, ret = -1, op_errno = EINVAL;
+        gf_boolean_t is_spl = _gf_false;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.*stripe*", dict,
+                                   op_errno, err);
+
+        is_spl = stripe_fsetxattr_is_special (dict);
+        if (is_spl) {
+                ret = stripe_fsetxattr_to_everyone (frame, this, fd, dict,
+                                                    flags, xdata);
+                if (ret < 0) {
+                        op_errno = ENOMEM;
+                        goto err;
+                }
 
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log (this->name, GF_LOG_DEBUG, "possible NULL deref");
                 goto out;
         }
-        entry = cookie;
+
+        STACK_WIND (frame, stripe_fsetxattr_cbk,
+                    FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fsetxattr,
+                    fd, dict, flags, xdata);
+out:
+        return 0;
+err:
+        STRIPE_STACK_UNWIND (fsetxattr, frame, op_ret, op_errno, NULL);
+        return 0;
+}
+
+int
+stripe_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        STRIPE_STACK_UNWIND (removexattr, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+int
+stripe_removexattr (call_frame_t *frame, xlator_t *this,
+                    loc_t *loc, const char *name, dict_t *xdata)
+{
+        int32_t         op_errno = EINVAL;
+
+        VALIDATE_OR_GOTO (this, err);
+
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.*stripe*",
+                                 name, op_errno, err);
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (loc, err);
+
+        STACK_WIND (frame, stripe_removexattr_cbk,
+                    FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->removexattr,
+                    loc, name, xdata);
+        return 0;
+err:
+        STRIPE_STACK_UNWIND (removexattr, frame, -1,  op_errno, NULL);
+        return 0;
+}
+
+
+int
+stripe_fremovexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        STRIPE_STACK_UNWIND (fremovexattr, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+int
+stripe_fremovexattr (call_frame_t *frame, xlator_t *this,
+                     fd_t *fd, const char *name, dict_t *xdata)
+{
+        int32_t         op_ret   = -1;
+        int32_t         op_errno = EINVAL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.*stripe*",
+                                 name, op_errno, err);
+
+        STACK_WIND (frame, stripe_fremovexattr_cbk,
+                    FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fremovexattr,
+                    fd, name, xdata);
+        return 0;
+ err:
+        STRIPE_STACK_UNWIND (fremovexattr, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+int32_t
+stripe_readdirp_lookup_cbk (call_frame_t *frame, void *cookie,
+                            xlator_t *this, int op_ret, int op_errno,
+                            inode_t *inode, struct iatt *stbuf,
+                            dict_t *xattr, struct iatt *parent)
+{
+        stripe_local_t          *local          = NULL;
+        call_frame_t            *main_frame     = NULL;
+        stripe_local_t          *main_local     = NULL;
+        gf_dirent_t             *entry          = NULL;
+        call_frame_t            *prev           = NULL;
+        int                      done           = 0;
+
         local = frame->local;
+        prev = cookie;
+
+        entry = local->dirent;
+
+        main_frame = local->orig_frame;
+        main_local = main_frame->local;
         LOCK (&frame->lock);
         {
 
-                local->wind_count--;
-                if (!local->wind_count)
+                local->call_count--;
+                if (!local->call_count)
                         done = 1;
                 if (op_ret == -1) {
                         local->op_errno = op_errno;
                         local->op_ret = op_ret;
                         goto unlock;
                 }
-                stripe_iatt_merge (buf, &entry->d_stat);
+
+		if (stripe_ctx_handle(this, prev, local, xattr))
+			gf_log(this->name, GF_LOG_ERROR,
+				"Error getting fctx info from dict.");
+
+		correct_file_size(stbuf, local->fctx, prev);
+
+                stripe_iatt_merge (stbuf, &entry->d_stat);
+                local->stbuf_blocks += stbuf->ia_blocks;
         }
 unlock:
         UNLOCK(&frame->lock);
 
         if (done) {
-                frame->local = NULL;
-                STRIPE_STACK_UNWIND (readdir, frame, local->op_ret,
-                                     local->op_errno, &local->entries);
+                inode_ctx_put (entry->inode, this,
+                               (uint64_t) (long)local->fctx);
 
-                gf_dirent_free (&local->entries);
+                done = 0;
+                LOCK (&main_frame->lock);
+                {
+                        main_local->wind_count--;
+                        if (!main_local->wind_count)
+                                done = 1;
+                        if (local->op_ret == -1) {
+                                main_local->op_errno = local->op_errno;
+                                main_local->op_ret = local->op_ret;
+                        }
+                        entry->d_stat.ia_blocks = local->stbuf_blocks;
+                }
+                UNLOCK (&main_frame->lock);
+                if (done) {
+                        main_frame->local = NULL;
+                        STRIPE_STACK_UNWIND (readdir, main_frame,
+                                             main_local->op_ret,
+                                             main_local->op_errno,
+                                             &main_local->entries, NULL);
+                        gf_dirent_free (&main_local->entries);
+                        stripe_local_wipe (main_local);
+                        mem_put (main_local);
+                }
+                frame->local = NULL;
                 stripe_local_wipe (local);
-                GF_FREE (local);
+                mem_put (local);
+                STRIPE_STACK_DESTROY (frame);
         }
-out:
-        return 0;
 
+        return 0;
 }
 
 int32_t
 stripe_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, gf_dirent_t *orig_entries)
+                     int32_t op_ret, int32_t op_errno,
+                     gf_dirent_t *orig_entries, dict_t *xdata)
 {
         stripe_local_t *local = NULL;
         call_frame_t   *prev = NULL;
         gf_dirent_t    *local_entry = NULL;
-        int32_t        ret = -1;
         gf_dirent_t    *tmp_entry = NULL;
         xlator_list_t  *trav = NULL;
         loc_t          loc = {0, };
-        inode_t        *inode = NULL;
-        char           *path;
         int32_t        count = 0;
         stripe_private_t *priv = NULL;
         int32_t        subvols = 0;
+        dict_t         *xattrs = NULL;
+        call_frame_t   *local_frame = NULL;
+        stripe_local_t *local_ent = NULL;
 
         if (!this || !frame || !frame->local || !cookie) {
                 gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
@@ -3842,8 +4867,9 @@ stripe_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->op_ret = op_ret;
                         list_splice_init (&orig_entries->list,
                                           &local->entries.list);
-                        local->wind_count = op_ret * subvols;
+                        local->wind_count = op_ret;
                 }
+
         }
 unlock:
         UNLOCK (&frame->lock);
@@ -3851,77 +4877,85 @@ unlock:
         if (op_ret == -1)
                 goto out;
 
+        xattrs = dict_new ();
+        if (xattrs)
+                (void) stripe_xattr_request_build (this, xattrs, 0, 0, 0, 0);
         count = op_ret;
-        ret = 0;
         list_for_each_entry_safe (local_entry, tmp_entry,
                                   (&local->entries.list), list) {
 
                 if (!local_entry)
                         break;
-                if (!IA_ISREG (local_entry->d_stat.ia_type)) {
+                if (!IA_ISREG (local_entry->d_stat.ia_type) || !local_entry->inode) {
                         LOCK (&frame->lock);
                         {
-                                local->wind_count -= subvols;
+                                local->wind_count--;
                                 count = local->wind_count;
                         }
                         UNLOCK (&frame->lock);
                         continue;
                 }
 
-                inode = inode_new (local->fd->inode->table);
-                if (!inode)
+                local_frame = copy_frame (frame);
+
+                if (!local_frame) {
+                        op_errno = ENOMEM;
+                        op_ret = -1;
                         goto out;
+                }
 
-                loc.ino = inode->ino = local_entry->d_ino;
-                loc.inode = inode;
-                loc.parent = local->fd->inode;
-                ret = inode_path (local->fd->inode, local_entry->d_name, &path);
-                if (ret != -1) {
-                        loc.path = path;
-                } else  if (inode) {
-                        ret = inode_path (inode, NULL, &path);
-                        if (ret != -1) {
-                                loc.path = path;
-                        } else {
-                                goto out;
-                        }
+                local_ent = mem_get0 (this->local_pool);
+                if (!local_ent) {
+                        op_errno = ENOMEM;
+                        op_ret = -1;
+                        goto out;
                 }
 
-                loc.name = strrchr (loc.path, '/');
-                loc.name++;
+                loc.inode = inode_ref (local_entry->inode);
+
+                uuid_copy (loc.gfid, local_entry->d_stat.ia_gfid);
+
+                local_ent->orig_frame = frame;
+
+                local_ent->call_count = subvols;
+
+                local_ent->dirent = local_entry;
+
+                local_frame->local = local_ent;
+
                 trav = this->children;
                 while (trav) {
-                        STACK_WIND_COOKIE (frame, stripe_readdirp_entry_stat_cbk,
-                                           local_entry, trav->xlator,
-                                           trav->xlator->fops->stat, &loc);
+                        STACK_WIND (local_frame, stripe_readdirp_lookup_cbk,
+                                    trav->xlator, trav->xlator->fops->lookup,
+                                    &loc, xattrs);
                         trav = trav->next;
                 }
-                inode_unref (loc.inode);
+                loc_wipe (&loc);
         }
 out:
         if (!count) {
                 /* all entries are directories */
                 frame->local = NULL;
                 STRIPE_STACK_UNWIND (readdir, frame, local->op_ret,
-                                     local->op_errno, &local->entries);
+                                     local->op_errno, &local->entries, NULL);
                 gf_dirent_free (&local->entries);
                 stripe_local_wipe (local);
-                GF_FREE (local);
+                mem_put (local);
         }
-
+        if (xattrs)
+                dict_unref (xattrs);
         return 0;
 
 }
 int32_t
 stripe_readdirp (call_frame_t *frame, xlator_t *this,
-                 fd_t *fd, size_t size, off_t off)
+                 fd_t *fd, size_t size, off_t off, dict_t *xdata)
 {
         stripe_local_t  *local  = NULL;
         stripe_private_t *priv = NULL;
         xlator_list_t   *trav = NULL;
         int             op_errno = -1;
 
-
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
@@ -3935,8 +4969,7 @@ stripe_readdirp (call_frame_t *frame, xlator_t *this,
         }
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -3956,15 +4989,16 @@ stripe_readdirp (call_frame_t *frame, xlator_t *this,
                 goto err;
 
         STACK_WIND (frame, stripe_readdirp_cbk, trav->xlator,
-                    trav->xlator->fops->readdirp, fd, size, off);
+                    trav->xlator->fops->readdirp, fd, size, off, xdata);
         return 0;
 err:
         op_errno = (op_errno == -1) ? errno : op_errno;
-        STRIPE_STACK_UNWIND (readdir, frame, -1, op_errno, NULL);
+        STRIPE_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
 
         return 0;
 
 }
+
 int32_t
 mem_acct_init (xlator_t *this)
 {
@@ -3984,65 +5018,87 @@ mem_acct_init (xlator_t *this)
 out:
         return ret;
 }
-int
-validate_options (xlator_t *this, char **op_errstr)
-{
-        int                 ret = 0;
-        volume_opt_list_t  *vol_opt = NULL;
-        volume_opt_list_t  *tmp;
 
-        if (!this) {
-                gf_log (this->name, GF_LOG_DEBUG, "'this' not a valid ptr");
-                ret =-1;
-                goto out;
-        }
+static int
+clear_pattern_list (stripe_private_t *priv)
+{
+        struct stripe_options *prev = NULL;
+        struct stripe_options *trav = NULL;
+        int                    ret = -1;
 
-        if (list_empty (&this->volume_options))
-                goto out;
+        GF_VALIDATE_OR_GOTO ("stripe", priv, out);
 
-        vol_opt = list_entry (this->volume_options.next,
-                              volume_opt_list_t, list);
-        list_for_each_entry_safe (vol_opt, tmp, &this->volume_options, list) {
-                ret = validate_xlator_volume_options_attacherr (this,
-                                                                vol_opt->given_opt,
-                                                                op_errstr);
+        trav = priv->pattern;
+        priv->pattern = NULL;
+        while (trav) {
+                prev = trav;
+                trav = trav->next;
+                GF_FREE (prev);
         }
 
-out:
-
+        ret = 0;
+ out:
         return ret;
+
+
 }
 
+
 int
 reconfigure (xlator_t *this, dict_t *options)
 {
 
-	stripe_private_t *priv = NULL;
-	data_t           *data = NULL;
-	int		  ret = 0;
+        stripe_private_t *priv = NULL;
+        data_t           *data = NULL;
+        int               ret = -1;
+        volume_option_t  *opt = NULL;
 
-	priv = this->private;
+        GF_ASSERT (this);
+        GF_ASSERT (this->private);
 
-	data = dict_get (options, "block-size");
-        if (data) {
-		gf_log (this->name, GF_LOG_TRACE,"Reconfiguring Stripe"
-			" Block-size");
-                ret = set_stripe_block_size (this, priv, data->data);
-                if (ret) {
-			gf_log (this->name, GF_LOG_ERROR,
-                                "Reconfigue: Block-Size reconfiguration failed");
-                        ret = -1;
-			goto out;
-		}
-		gf_log (this->name, GF_LOG_TRACE,
-                        "Reconfigue: Block-Size reconfigured Successfully");
-	}
-        else {
-                priv->block_size = (128 * GF_UNIT_KB);
+        priv = this->private;
+
+
+        ret = 0;
+        LOCK (&priv->lock);
+        {
+                ret = clear_pattern_list (priv);
+                if (ret)
+                         goto unlock;
+
+                data = dict_get (options, "block-size");
+                if (data) {
+                        ret = set_stripe_block_size (this, priv, data->data);
+                        if (ret)
+                                goto unlock;
+                } else {
+                        opt = xlator_volume_option_get (this, "block-size");
+                        if (!opt) {
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "option 'block-size' not found");
+                                ret = -1;
+                                goto unlock;
+                        }
+
+                        if (gf_string2bytesize_uint64 (opt->default_value, &priv->block_size)){
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "Unable to set default block-size ");
+                                ret = -1;
+                                goto unlock;
+                        }
+                }
+
+		GF_OPTION_RECONF("coalesce", priv->coalesce, options, bool,
+				unlock);
         }
+ unlock:
+        UNLOCK (&priv->lock);
+        if (ret)
+                goto out;
 
-out:
-	return ret;
+        ret = 0;
+ out:
+        return ret;
 
 }
 
@@ -4055,9 +5111,9 @@ int32_t
 init (xlator_t *this)
 {
         stripe_private_t *priv = NULL;
+        volume_option_t  *opt = NULL;
         xlator_list_t    *trav = NULL;
         data_t           *data = NULL;
-        char             *def_blk_size = NULL;
         int32_t           count = 0;
         int               ret = -1;
 
@@ -4099,9 +5155,9 @@ init (xlator_t *this)
         if (!priv->xl_array)
                 goto out;
 
-        priv->state = GF_CALLOC (count, sizeof (int8_t),
-                                 gf_stripe_mt_int8_t);
-        if (!priv->state)
+        priv->last_event = GF_CALLOC (count, sizeof (int),
+                                      gf_stripe_mt_int32_t);
+        if (!priv->last_event)
                 goto out;
 
         priv->child_count = count;
@@ -4121,77 +5177,56 @@ init (xlator_t *this)
                 goto out;
         }
 
-        if (xlator_get_volopt_info (&this->volume_options, "block-size",
-                                    &def_blk_size, NULL)) {
-                gf_log (this->name, GF_LOG_ERROR, "Default value of stripe "
-                         "block-size corrupt");
-                ret = -1;
-                goto out;
-        } else {
-                if (gf_string2bytesize (def_blk_size, &priv->block_size)) {
-                        gf_log (this->name, GF_LOG_ERROR, "Default value of "
-                                 "stripe block-size corrupt");
+        ret = 0;
+        LOCK (&priv->lock);
+        {
+                opt = xlator_volume_option_get (this, "block-size");
+                if (!opt) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "option 'block-size' not found");
                         ret = -1;
-                        goto out;
+                        goto unlock;
+                }
+                if (gf_string2bytesize_uint64 (opt->default_value, &priv->block_size)){
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Unable to set default block-size ");
+                        ret = -1;
+                        goto unlock;
+                }
+                /* option stripe-pattern *avi:1GB,*pdf:16K */
+                data = dict_get (this->options, "block-size");
+                if (data) {
+                        ret = set_stripe_block_size (this, priv, data->data);
+                        if (ret)
+                                goto unlock;
                 }
         }
+ unlock:
+        UNLOCK (&priv->lock);
+        if (ret)
+                goto out;
 
+        GF_OPTION_INIT ("use-xattr", priv->xattr_supported, bool, out);
+        /* notify related */
+        priv->nodes_down = priv->child_count;
 
-        /* option stripe-pattern *avi:1GB,*pdf:4096 */
-        data = dict_get (this->options, "block-size");
-        if (!data) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "No \"option block-size <x>\" given, defaulting "
-                        "to %s", def_blk_size);
-        } else {
-                ret = set_stripe_block_size (this, priv, data->data);
-                if (ret)
-                        goto out;
-        }
+	GF_OPTION_INIT("coalesce", priv->coalesce, bool, out);
 
-        if (xlator_get_volopt_info (&this->volume_options, "use-xattr",
-                                    &def_blk_size, NULL)) {
+        this->local_pool = mem_pool_new (stripe_local_t, 128);
+        if (!this->local_pool) {
                 ret = -1;
                 gf_log (this->name, GF_LOG_ERROR,
-                        "error setting(default) hard check for extended"
-                        " attribute");
+                        "failed to create local_t's memory pool");
                 goto out;
-
-        }
-        else {
-                if (gf_string2boolean (def_blk_size,
-                                       &priv->xattr_supported)) {
-                        ret = -1;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "error setting(default) hard check for extended"
-                                " attribute");
-                        goto out;
-                }
-        }
-
-
-        data = dict_get (this->options, "use-xattr");
-        if (data) {
-                if (gf_string2boolean (data->data,
-                                       &priv->xattr_supported) == -1) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "error setting hard check for extended "
-                                "attribute");
-                        //return -1;
-                }
         }
 
-        /* notify related */
-        priv->nodes_down = priv->child_count;
         this->private = priv;
 
-
         ret = 0;
 out:
         if (ret) {
                 if (priv) {
-                        if (priv->xl_array)
-                                GF_FREE (priv->xl_array);
+                        GF_FREE (priv->xl_array);
                         GF_FREE (priv);
                 }
         }
@@ -4215,15 +5250,15 @@ fini (xlator_t *this)
         priv = this->private;
         if (priv) {
                 this->private = NULL;
-                if (priv->xl_array)
-                        GF_FREE (priv->xl_array);
+                GF_FREE (priv->xl_array);
 
                 trav = priv->pattern;
                 while (trav) {
                         prev = trav;
                         trav = trav->next;
-                        FREE (prev);
+                        GF_FREE (prev);
                 }
+                GF_FREE (priv->last_event);
                 LOCK_DESTROY (&priv->lock);
                 GF_FREE (priv);
         }
@@ -4234,17 +5269,50 @@ out:
 
 int32_t
 stripe_getxattr_unwind (call_frame_t *frame,
-                        int op_ret, int op_errno, dict_t *dict)
+                        int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
 
 {
-        STRIPE_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
+        STRIPE_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, xdata);
         return 0;
 }
 
+int
+stripe_internal_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                              int op_ret, int op_errno, dict_t *xattr,
+                              dict_t *xdata)
+{
+
+        char        size_key[256]  = {0,};
+        char        index_key[256] = {0,};
+        char        count_key[256] = {0,};
+        char        coalesce_key[256] = {0,};
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (frame->local, out);
+
+        if (!xattr || (op_ret == -1))
+            goto out;
+
+        sprintf (size_key, "trusted.%s.stripe-size", this->name);
+        sprintf (count_key, "trusted.%s.stripe-count", this->name);
+        sprintf (index_key, "trusted.%s.stripe-index", this->name);
+	sprintf (coalesce_key, "trusted.%s.stripe-coalesce", this->name);
+
+        dict_del (xattr, size_key);
+        dict_del (xattr, count_key);
+        dict_del (xattr, index_key);
+        dict_del (xattr, coalesce_key);
+
+out:
+        STRIPE_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr, xdata);
+
+        return 0;
+
+}
 
 int
 stripe_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int op_ret, int op_errno, dict_t *xattr)
+                     int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
 {
         int                     call_cnt = 0;
         stripe_local_t         *local = NULL;
@@ -4274,92 +5342,39 @@ stripe_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 out:
         if (!call_cnt) {
                 STRIPE_STACK_UNWIND (getxattr, frame, local->op_ret, op_errno,
-                                     local->xattr);
+                                     local->xattr, xdata);
         }
 
         return 0;
 }
 
 int32_t
-stripe_pathinfo_aggregate (char *buffer, stripe_local_t *local, int32_t *total)
-{
-        int32_t              i     = 0;
-        int32_t              ret   = -1;
-        int32_t              len   = 0;
-        char                *sbuf  = NULL;
-        stripe_xattr_sort_t *xattr = NULL;
-
-        if (!buffer || !local || !local->xattr_list)
-                goto out;
-
-        sbuf = buffer;
-
-        for (i = 0; i < local->nallocs; i++) {
-                xattr = local->xattr_list + i;
-                len = xattr->pathinfo_len;
-
-                if (len && xattr && xattr->pathinfo) {
-                        memcpy (buffer, xattr->pathinfo, len);
-                        buffer += len;
-                        *buffer++ = ' ';
-                }
-        }
-
-        *--buffer = '\0';
-        if (total)
-                *total = buffer - sbuf;
-        ret = 0;
-
- out:
-        return ret;
-}
-
-int32_t
-stripe_free_pathinfo_str (stripe_local_t *local)
+stripe_vgetxattr_cbk (call_frame_t *frame, void *cookie,
+                      xlator_t *this, int32_t op_ret, int32_t op_errno,
+                      dict_t *dict, dict_t *xdata)
 {
-        int32_t              i     = 0;
-        int32_t              ret   = -1;
-        stripe_xattr_sort_t *xattr = NULL;
-
-        if (!local || !local->xattr_list)
-                goto out;
-
-        for (i = 0; i < local->nallocs; i++) {
-                xattr = local->xattr_list + i;
-
-                if (xattr && xattr->pathinfo)
-                        GF_FREE (xattr->pathinfo);
-        }
-
-        ret = 0;
- out:
-        return ret;
-}
-
-int32_t
-stripe_getxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
-                             xlator_t *this, int32_t op_ret, int32_t op_errno,
-                             dict_t *dict) {
         stripe_local_t      *local         = NULL;
         int32_t              callcnt       = 0;
         int32_t              ret           = -1;
         long                 cky           = 0;
-        char                *pathinfo      = NULL;
-        char                *pathinfo_serz = NULL;
-        int32_t              padding       = 0;
-        int32_t              tlen          = 0;
-        char stripe_size_str[20]           = {0,};
+        void                *xattr_val     = NULL;
+        void                *xattr_serz    = NULL;
         stripe_xattr_sort_t *xattr         = NULL;
         dict_t              *stripe_xattr  = NULL;
 
         if (!frame || !frame->local || !this) {
-                gf_log (this->name, GF_LOG_ERROR, "Possible NULL deref");
+                gf_log ("", GF_LOG_ERROR, "Possible NULL deref");
                 return ret;
         }
 
         local = frame->local;
         cky = (long) cookie;
 
+        if (local->xsel[0] == '\0') {
+                gf_log (this->name, GF_LOG_ERROR, "Empty xattr in cbk");
+                return ret;
+        }
+
         LOCK (&frame->lock);
         {
                 callcnt = --local->wind_count;
@@ -4368,23 +5383,26 @@ stripe_getxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
                         goto out;
 
                 if (!local->xattr_list)
-                        local->xattr_list = (stripe_xattr_sort_t *) GF_CALLOC (local->nallocs,
-                                                                               sizeof (stripe_xattr_sort_t),
-                                                                               gf_stripe_mt_xattr_sort_t);
+                        local->xattr_list = (stripe_xattr_sort_t *)
+                                GF_CALLOC (local->nallocs,
+                                           sizeof (stripe_xattr_sort_t),
+                                           gf_stripe_mt_xattr_sort_t);
 
                 if (local->xattr_list) {
-                        ret = dict_get_str (dict, GF_XATTR_PATHINFO_KEY, &pathinfo);
-                        if (ret)
-                                goto out;
-
                         xattr = local->xattr_list + (int32_t) cky;
 
-                        pathinfo = gf_strdup (pathinfo);
+                        ret = dict_get_ptr_and_len (dict, local->xsel,
+                                                    &xattr_val,
+                                                    &xattr->xattr_len);
+                        if (xattr->xattr_len == 0)
+                                goto out;
+
                         xattr->pos = cky;
-                        xattr->pathinfo = pathinfo;
-                        xattr->pathinfo_len = strlen (pathinfo);
+                        xattr->xattr_value = gf_memdup (xattr_val,
+                                                        xattr->xattr_len);
 
-                        local->xattr_total_len += strlen (pathinfo) + 1;
+                        if (xattr->xattr_value != NULL)
+                                local->xattr_total_len += xattr->xattr_len + 1;
                 }
         }
  out:
@@ -4398,41 +5416,36 @@ stripe_getxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
                 if (!stripe_xattr)
                         goto unwind;
 
-                snprintf (stripe_size_str, 20, "%ld", local->stripe_size);
-
-                /* extra bytes for decorations (brackets and <>'s) */
-                padding = strlen (this->name) + strlen (STRIPE_PATHINFO_HEADER)
-                        + strlen (stripe_size_str) + 7;
-                local->xattr_total_len += (padding + 2);
-
-                pathinfo_serz = GF_CALLOC (local->xattr_total_len, sizeof (char),
-                                           gf_common_mt_char);
-                if (!pathinfo_serz)
-                        goto unwind;
-
-                /* xlator info */
-                sprintf (pathinfo_serz, "(<"STRIPE_PATHINFO_HEADER"%s:[%s]> ", this->name, stripe_size_str);
-
-                ret = stripe_pathinfo_aggregate (pathinfo_serz + padding, local, &tlen);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "Cannot aggregate pathinfo list");
+                /* select filler based on ->xsel */
+                if (XATTR_IS_PATHINFO (local->xsel))
+                        ret = stripe_fill_pathinfo_xattr (this, local,
+                                                          (char **)&xattr_serz);
+                else if (XATTR_IS_LOCKINFO (local->xsel)) {
+                        ret = stripe_fill_lockinfo_xattr (this, local,
+                                                          &xattr_serz);
+                } else {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "Unknown xattr in xattr request");
                         goto unwind;
                 }
 
-                *(pathinfo_serz + padding + tlen) = ')';
-                *(pathinfo_serz + padding + tlen + 1) = '\0';
-
-                ret = dict_set_dynstr (stripe_xattr, GF_XATTR_PATHINFO_KEY, pathinfo_serz);
-                if (ret)
-                        gf_log (this->name, GF_LOG_ERROR, "Cannot set pathinfo key in dict");
+                if (!ret) {
+                        ret = dict_set_dynptr (stripe_xattr, local->xsel,
+                                               xattr_serz,
+                                               local->xattr_total_len);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "Can't set %s key in dict",
+                                        local->xsel);
+                }
 
         unwind:
-                STRIPE_STACK_UNWIND (getxattr, frame, op_ret, op_errno, stripe_xattr);
+                STRIPE_STACK_UNWIND (getxattr, frame, op_ret, op_errno,
+                                     stripe_xattr, NULL);
 
-                ret = stripe_free_pathinfo_str (local);
+                ret = stripe_free_xattr_str (local);
 
-                if (local->xattr_list)
-                        GF_FREE (local->xattr_list);
+                GF_FREE (local->xattr_list);
 
                 if (stripe_xattr)
                         dict_unref (stripe_xattr);
@@ -4443,14 +5456,15 @@ stripe_getxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
 
 int32_t
 stripe_getxattr (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, const char *name)
+                 loc_t *loc, const char *name, dict_t *xdata)
 {
-        stripe_local_t     *local = NULL;
-        xlator_list_t      *trav = NULL;
-        stripe_private_t   *priv = NULL;
-        int32_t             op_errno = EINVAL;
-        int                 i = 0;
-        xlator_t          **sub_volumes;
+        stripe_local_t    *local    = NULL;
+        xlator_list_t     *trav     = NULL;
+        stripe_private_t  *priv     = NULL;
+        int32_t            op_errno = EINVAL;
+        int                i        = 0;
+        xlator_t         **sub_volumes;
+        int                ret      = 0;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -4462,8 +5476,7 @@ stripe_getxattr (call_frame_t *frame, xlator_t *this,
         trav = this->children;
 
         /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
+        local = mem_get0 (this->local_pool);
         if (!local) {
                 op_errno = ENOMEM;
                 goto err;
@@ -4474,7 +5487,7 @@ stripe_getxattr (call_frame_t *frame, xlator_t *this,
 
 
         if (name && (strcmp (GF_XATTR_MARKER_KEY, name) == 0)
-            && (-1 == frame->root->pid)) {
+            && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
                 local->marker.call_count = priv->child_count;
 
                 sub_volumes = alloca ( priv->child_count *
@@ -4489,7 +5502,8 @@ stripe_getxattr (call_frame_t *frame, xlator_t *this,
                 if (cluster_getmarkerattr (frame, this, loc, name,
                                            local, stripe_getxattr_unwind,
                                            sub_volumes, priv->child_count,
-                                           MARKER_UUID_TYPE, priv->vol_uuid)) {
+                                           MARKER_UUID_TYPE, marker_uuid_default_gauge,
+                                           priv->vol_uuid)) {
                         op_errno = EINVAL;
                         goto err;
                 }
@@ -4505,25 +5519,37 @@ stripe_getxattr (call_frame_t *frame, xlator_t *this,
                              trav = trav->next) {
                         STACK_WIND (frame, stripe_getxattr_cbk,
                                     trav->xlator, trav->xlator->fops->getxattr,
-                                    loc, name);
+                                    loc, name, xdata);
                 }
 
                 return 0;
         }
 
-        if (name && (strncmp (name, GF_XATTR_PATHINFO_KEY,
-                              strlen (GF_XATTR_PATHINFO_KEY)) == 0)) {
-                local->stripe_size = stripe_get_matching_bs (loc->path,
-                                                             priv->pattern,
-                                                             priv->block_size);
+        if (name && (XATTR_IS_PATHINFO (name))) {
+                if (IA_ISREG (loc->inode->ia_type)) {
+                        ret = inode_ctx_get (loc->inode, this,
+                                             (uint64_t *) &local->fctx);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "stripe size unavailable from fctx"
+                                        " relying on pathinfo could lead to"
+                                        " wrong results");
+                }
+
                 local->nallocs = local->wind_count = priv->child_count;
+                (void) strncpy (local->xsel, name, strlen (name));
 
+                /**
+                 * for xattrs that need info from all childs, fill ->xsel
+                 * as above and call the filler function in cbk based on
+                 * it
+                 */
                 for (i = 0, trav = this->children; i < priv->child_count; i++,
                      trav = trav->next) {
-                        STACK_WIND_COOKIE (frame, stripe_getxattr_pathinfo_cbk,
+                        STACK_WIND_COOKIE (frame, stripe_vgetxattr_cbk,
                                            (void *) (long) i, trav->xlator,
                                            trav->xlator->fops->getxattr,
-                                           loc, name);
+                                           loc, name, xdata);
                 }
 
                 return 0;
@@ -4531,46 +5557,127 @@ stripe_getxattr (call_frame_t *frame, xlator_t *this,
 
         if (name &&(*priv->vol_uuid)) {
                 if ((match_uuid_local (name, priv->vol_uuid) == 0)
-                    && (-1 == frame->root->pid)) {
-                        local->marker.call_count = priv->child_count;
+                    && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
+
+                        if (!IA_FILE_OR_DIR (loc->inode->ia_type))
+                                local->marker.call_count = 1;
+                        else
+                                local->marker.call_count = priv->child_count;
 
-                        sub_volumes = alloca ( priv->child_count *
-                                               sizeof (xlator_t *));
-                        for (i = 0, trav = this->children; trav ;
-                             trav = trav->next, i++) {
+                        sub_volumes = alloca (local->marker.call_count *
+                                              sizeof (xlator_t *));
 
+                        for (i = 0, trav = this->children;
+                             i < local->marker.call_count;
+                             i++, trav = trav->next) {
                                 *(sub_volumes + i) = trav->xlator;
 
                         }
 
                         if (cluster_getmarkerattr (frame, this, loc, name,
-                                                   local, stripe_getxattr_unwind,
+                                                   local,
+                                                   stripe_getxattr_unwind,
                                                    sub_volumes,
-                                                   priv->child_count,
+                                                   local->marker.call_count,
                                                    MARKER_XTIME_TYPE,
+                                                   marker_xtime_default_gauge,
                                                    priv->vol_uuid)) {
                                 op_errno = EINVAL;
                                 goto err;
                         }
+
                         return 0;
                 }
         }
 
 
-        STACK_WIND (frame, default_getxattr_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->getxattr, loc, name);
+        STACK_WIND (frame, stripe_internal_getxattr_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+
+        return 0;
+
+err:
+        STRIPE_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
+
+static inline gf_boolean_t
+stripe_is_special_xattr (const char *name)
+{
+        gf_boolean_t    is_spl = _gf_false;
+
+        if (!name) {
+                goto out;
+        }
+
+        if (!strncmp (name, GF_XATTR_LOCKINFO_KEY,
+                      strlen (GF_XATTR_LOCKINFO_KEY))
+            || XATTR_IS_PATHINFO (name))
+                is_spl = _gf_true;
+out:
+        return is_spl;
+}
+
+int32_t
+stripe_fgetxattr_from_everyone (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                                const char *name, dict_t *xdata)
+{
+        stripe_local_t   *local = NULL;
+        stripe_private_t *priv  = NULL;
+        int32_t           ret   = -1, op_errno = 0;
+        int               i     = 0;
+        xlator_list_t    *trav  = NULL;
+
+        priv = this->private;
+
+        local = mem_get0 (this->local_pool);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->op_ret = -1;
+        frame->local = local;
+
+        strncpy (local->xsel, name, strlen (name));
+        local->nallocs = local->wind_count = priv->child_count;
+
+        for (i = 0, trav = this->children; i < priv->child_count; i++,
+                     trav = trav->next) {
+                STACK_WIND_COOKIE (frame, stripe_vgetxattr_cbk,
+                                   (void *) (long) i, trav->xlator,
+                                   trav->xlator->fops->fgetxattr,
+                                   fd, name, xdata);
+        }
 
         return 0;
 
 err:
-        STRIPE_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL);
+        STACK_UNWIND_STRICT (fgetxattr, frame, -1, op_errno, NULL, NULL);
+        return ret;
+}
+
+int32_t
+stripe_fgetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  const char *name, dict_t *xdata)
+{
+        if (stripe_is_special_xattr (name)) {
+                stripe_fgetxattr_from_everyone (frame, this, fd, name, xdata);
+                goto out;
+        }
+
+        STACK_WIND (frame, stripe_internal_getxattr_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+
+out:
         return 0;
 }
 
+
+
 int32_t
 stripe_priv_dump (xlator_t *this)
 {
-        char                    key_prefix[GF_DUMP_MAX_BUF_LEN];
         char                    key[GF_DUMP_MAX_BUF_LEN];
         int                     i = 0;
         stripe_private_t       *priv = NULL;
@@ -4588,39 +5695,26 @@ stripe_priv_dump (xlator_t *this)
                 goto out;
 
         gf_proc_dump_add_section("xlator.cluster.stripe.%s.priv", this->name);
-        gf_proc_dump_build_key(key_prefix,"xlator.cluster.stripe","%s.priv",
-                               this->name);
-        gf_proc_dump_build_key(key, key_prefix, "child_count");
-        gf_proc_dump_write(key,"%d", priv->child_count);
+        gf_proc_dump_write("child_count","%d", priv->child_count);
 
         for (i = 0; i < priv->child_count; i++) {
-                gf_proc_dump_build_key (key, key_prefix, "subvolumes[%d]", i);
+                sprintf (key, "subvolumes[%d]", i);
                 gf_proc_dump_write (key, "%s.%s", priv->xl_array[i]->type,
                                     priv->xl_array[i]->name);
         }
 
         options = priv->pattern;
         while (options != NULL) {
-                gf_proc_dump_build_key (key, key_prefix, "path_pattern");
-                gf_proc_dump_write (key, "%s", priv->pattern->path_pattern);
-
-                gf_proc_dump_build_key (key, key_prefix, "options_block_size");
-                gf_proc_dump_write (key, "%ul", options->block_size);
+                gf_proc_dump_write ("path_pattern", "%s", priv->pattern->path_pattern);
+                gf_proc_dump_write ("options_block_size", "%ul", options->block_size);
 
                 options = options->next;
         }
 
-        gf_proc_dump_build_key (key, key_prefix, "block_size");
-        gf_proc_dump_write (key, "%ul", priv->block_size);
-
-        gf_proc_dump_build_key (key, key_prefix, "nodes_down");
-        gf_proc_dump_write (key, "%d", priv->nodes_down);
-
-        gf_proc_dump_build_key (key, key_prefix, "first_child_down");
-        gf_proc_dump_write (key, "%d", priv->first_child_down);
-
-        gf_proc_dump_build_key (key, key_prefix, "xatter_supported");
-        gf_proc_dump_write (key, "%d", priv->xattr_supported);
+        gf_proc_dump_write ("block_size", "%ul", priv->block_size);
+        gf_proc_dump_write ("nodes-down", "%d", priv->nodes_down);
+        gf_proc_dump_write ("first-child_down", "%d", priv->first_child_down);
+        gf_proc_dump_write ("xattr_supported", "%d", priv->xattr_supported);
 
         UNLOCK (&priv->lock);
 
@@ -4629,36 +5723,44 @@ out:
 }
 
 struct xlator_fops fops = {
-        .stat        = stripe_stat,
-        .unlink      = stripe_unlink,
-        .rename      = stripe_rename,
-        .link        = stripe_link,
-        .truncate    = stripe_truncate,
-        .create      = stripe_create,
-        .open        = stripe_open,
-        .readv       = stripe_readv,
-        .writev      = stripe_writev,
-        .statfs      = stripe_statfs,
-        .flush       = stripe_flush,
-        .fsync       = stripe_fsync,
-        .ftruncate   = stripe_ftruncate,
-        .fstat       = stripe_fstat,
-        .mkdir       = stripe_mkdir,
-        .rmdir       = stripe_rmdir,
-        .lk          = stripe_lk,
-        .opendir     = stripe_opendir,
-        .fsyncdir    = stripe_fsyncdir,
-        .setattr     = stripe_setattr,
-        .fsetattr    = stripe_fsetattr,
-        .lookup      = stripe_lookup,
-        .mknod       = stripe_mknod,
-
-        .getxattr    = stripe_getxattr,
-        .readdirp    = stripe_readdirp,
+        .stat           = stripe_stat,
+        .unlink         = stripe_unlink,
+        .rename         = stripe_rename,
+        .link           = stripe_link,
+        .truncate       = stripe_truncate,
+        .create         = stripe_create,
+        .open           = stripe_open,
+        .readv          = stripe_readv,
+        .writev         = stripe_writev,
+        .statfs         = stripe_statfs,
+        .flush          = stripe_flush,
+        .fsync          = stripe_fsync,
+        .ftruncate      = stripe_ftruncate,
+        .fstat          = stripe_fstat,
+        .mkdir          = stripe_mkdir,
+        .rmdir          = stripe_rmdir,
+        .lk             = stripe_lk,
+        .opendir        = stripe_opendir,
+        .fsyncdir       = stripe_fsyncdir,
+        .setattr        = stripe_setattr,
+        .fsetattr       = stripe_fsetattr,
+        .lookup         = stripe_lookup,
+        .mknod          = stripe_mknod,
+        .setxattr       = stripe_setxattr,
+        .fsetxattr      = stripe_fsetxattr,
+        .getxattr       = stripe_getxattr,
+        .fgetxattr      = stripe_fgetxattr,
+        .removexattr    = stripe_removexattr,
+        .fremovexattr   = stripe_fremovexattr,
+        .readdirp       = stripe_readdirp,
+	.fallocate	= stripe_fallocate,
+	.discard	= stripe_discard,
+        .zerofill       = stripe_zerofill,
 };
 
 struct xlator_cbks cbks = {
         .release = stripe_release,
+        .forget  = stripe_forget,
 };
 
 struct xlator_dumpops dumpops = {
@@ -4667,8 +5769,9 @@ struct xlator_dumpops dumpops = {
 
 struct volume_options options[] = {
         { .key  = {"block-size"},
-          .type = GF_OPTION_TYPE_ANY,
+          .type = GF_OPTION_TYPE_SIZE_LIST,
           .default_value = "128KB",
+          .min = STRIPE_MIN_BLOCK_SIZE,
           .description = "Size of the stripe unit that would be read "
                          "from or written to the striped servers."
         },
@@ -4676,5 +5779,12 @@ struct volume_options options[] = {
           .type = GF_OPTION_TYPE_BOOL,
           .default_value = "true"
         },
+	{ .key = {"coalesce"},
+	  .type = GF_OPTION_TYPE_BOOL,
+	  .default_value = "true",
+	  .description = "Enable/Disable coalesce mode to flatten striped "
+			 "files as stored on the server (i.e., eliminate holes "
+			 "caused by the traditional format)."
+	},
         { .key  = {NULL} },
 };
diff --git a/xlators/cluster/stripe/src/stripe.h b/xlators/cluster/stripe/src/stripe.h
index a6b9375b4..5673d18f3 100644
--- a/xlators/cluster/stripe/src/stripe.h
+++ b/xlators/cluster/stripe/src/stripe.h
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  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.
 */
 
 
@@ -38,36 +29,53 @@
 #include <signal.h>
 
 #define STRIPE_PATHINFO_HEADER "STRIPE:"
-
+#define STRIPE_MIN_BLOCK_SIZE  (16*GF_UNIT_KB)
 
 #define STRIPE_STACK_UNWIND(fop, frame, params ...) do {           \
                 stripe_local_t *__local = NULL;                    \
-                if (frame) {                                    \
-                        __local = frame->local;                 \
-                        frame->local = NULL;                    \
-                }                                               \
-                STACK_UNWIND_STRICT (fop, frame, params);       \
+                if (frame) {                                       \
+                        __local = frame->local;                    \
+                        frame->local = NULL;                       \
+                }                                                  \
+                STACK_UNWIND_STRICT (fop, frame, params);          \
+                if (__local) {                                     \
+                        stripe_local_wipe(__local);                \
+                        mem_put (__local);       \
+                }                                                  \
+        } while (0)
+
+#define STRIPE_STACK_DESTROY(frame) do {                        \
+                stripe_local_t *__local = NULL;                 \
+                __local = frame->local;                         \
+                frame->local = NULL;                            \
+                STACK_DESTROY (frame->root);                    \
                 if (__local) {                                  \
-                        stripe_local_wipe(__local);             \
-                        GF_FREE (__local);                      \
+                        stripe_local_wipe (__local);            \
+                        mem_put (__local);    \
                 }                                               \
         } while (0)
 
-#define STRIPE_STACK_DESTROY(frame) do {                  \
-                stripe_local_t *__local = NULL;           \
-                __local = frame->local;                   \
-                frame->local = NULL;                      \
-                STACK_DESTROY (frame->root);              \
-                if (__local) {                            \
-                        stripe_local_wipe (__local);      \
-                        GF_FREE (__local);                \
-                }                                         \
-        } while (0)
+#define STRIPE_VALIDATE_FCTX(fctx, label) do {                  \
+        int     idx = 0;                                        \
+        if (!fctx) {                                            \
+                op_errno = EINVAL;                              \
+                goto label;                                     \
+        }                                                       \
+        for (idx = 0; idx < fctx->stripe_count; idx++) {        \
+                if (!fctx->xl_array[idx]) {                     \
+                        gf_log (this->name, GF_LOG_ERROR,       \
+                                "fctx->xl_array[%d] is NULL",   \
+                                idx);                           \
+                        op_errno = ESTALE;                      \
+                        goto label;                             \
+                }                                               \
+        }                                                       \
+       } while (0)
 
 typedef struct stripe_xattr_sort {
-        int32_t  pos;
-        int32_t  pathinfo_len;
-        char    *pathinfo;
+        int   pos;
+        int   xattr_len;
+        char *xattr_value;
 } stripe_xattr_sort_t;
 
 /**
@@ -90,16 +98,17 @@ struct stripe_private {
         gf_lock_t               lock;
         uint8_t                 nodes_down;
         int8_t                  first_child_down;
+        int                    *last_event;
         int8_t                  child_count;
-        int8_t                 *state; /* Current state of child node */
         gf_boolean_t            xattr_supported;  /* default yes */
+	gf_boolean_t		coalesce;
         char                    vol_uuid[UUID_SIZE + 1];
 };
 
 /**
- * Used to keep info about the replies received from fops->readv calls
+ * Used to keep info about the replies received from readv/writev calls
  */
-struct readv_replies {
+struct stripe_replies {
         struct iovec *vector;
         int32_t       count;    //count of vector
         int32_t       op_ret;   //op_ret of readv
@@ -111,6 +120,7 @@ struct readv_replies {
 typedef struct _stripe_fd_ctx {
         off_t      stripe_size;
         int        stripe_count;
+	int	   stripe_coalesce;
         int        static_array;
         xlator_t **xl_array;
 } stripe_fd_ctx_t;
@@ -146,7 +156,7 @@ struct stripe_local {
         blkcnt_t             preparent_blocks;
         blkcnt_t             postparent_blocks;
 
-        struct readv_replies *replies;
+        struct stripe_replies *replies;
         struct statvfs        statvfs_buf;
         dir_entry_t          *entry;
 
@@ -170,12 +180,15 @@ struct stripe_local {
         loc_t                loc;
         loc_t                loc2;
 
+        mode_t               mode;
+        dev_t                rdev;
         /* For File I/O fops */
-        dict_t              *dict;
+        dict_t              *xdata;
 
         stripe_xattr_sort_t *xattr_list;
         int32_t              xattr_total_len;
         int32_t              nallocs;
+        char xsel[256];
 
         struct marker_str    marker;
 
@@ -192,12 +205,84 @@ struct stripe_local {
         void                *value;
         struct iobref       *iobref;
         gf_dirent_t          entries;
+        gf_dirent_t         *dirent;
         dict_t              *xattr;
         uuid_t               ia_gfid;
+
+        int                  xflag;
+        mode_t               umask;
 };
 
 typedef struct stripe_local   stripe_local_t;
 typedef struct stripe_private stripe_private_t;
 
+/*
+ * Determine the stripe index of a particular frame based on the translator.
+ */
+static inline int32_t stripe_get_frame_index(stripe_fd_ctx_t *fctx,
+					     call_frame_t *prev)
+{
+	int32_t i, idx = -1;
+
+	for (i = 0; i < fctx->stripe_count; i++) {
+		if (fctx->xl_array[i] == prev->this) {
+			idx = i;
+			break;
+		}
+	}
+
+	return idx;
+}
+
+static inline void stripe_copy_xl_array(xlator_t **dst, xlator_t **src,
+					int count)
+{
+	int i;
+
+	for (i = 0; i < count; i++)
+		dst[i] = src[i];
+}
+
+void stripe_local_wipe (stripe_local_t *local);
+int32_t stripe_ctx_handle (xlator_t *this, call_frame_t *prev,
+                           stripe_local_t *local, dict_t *dict);
+void stripe_aggregate_xattr (dict_t *dst, dict_t *src);
+int32_t stripe_xattr_request_build (xlator_t *this, dict_t *dict,
+                                    uint64_t stripe_size, uint32_t stripe_count,
+                                    uint32_t stripe_index,
+				    uint32_t stripe_coalesce);
+int32_t stripe_get_matching_bs (const char *path, stripe_private_t *priv);
+int set_stripe_block_size (xlator_t *this, stripe_private_t *priv, char *data);
+int32_t stripe_iatt_merge (struct iatt *from, struct iatt *to);
+int32_t stripe_fill_pathinfo_xattr (xlator_t *this, stripe_local_t *local,
+                                    char **xattr_serz);
+int32_t stripe_free_xattr_str (stripe_local_t *local);
+int32_t stripe_xattr_aggregate (char *buffer, stripe_local_t *local,
+                                int32_t *total);
+off_t coalesced_offset(off_t offset, uint64_t stripe_size, int stripe_count);
+off_t uncoalesced_size(off_t size, uint64_t stripe_size, int stripe_count,
+			int stripe_index);
+int32_t
+stripe_fill_lockinfo_xattr (xlator_t *this, stripe_local_t *local,
+                            void **xattr_serz);
+
+/*
+ * Adjust the size attribute for files if coalesce is enabled.
+ */
+static inline void correct_file_size(struct iatt *buf, stripe_fd_ctx_t *fctx,
+	call_frame_t *prev)
+{
+	int index;
+
+	if (!IA_ISREG(buf->ia_type))
+		return;
+
+	if (!fctx || !fctx->stripe_coalesce)
+		return;
+
+	index = stripe_get_frame_index(fctx, prev);
+	buf->ia_size = uncoalesced_size(buf->ia_size, fctx->stripe_size,
+		fctx->stripe_count, index);
+}
 
 #endif /* _STRIPE_H_ */
diff --git a/xlators/cluster/unify/src/Makefile.am b/xlators/cluster/unify/src/Makefile.am
deleted file mode 100644
index 2a1fe8372..000000000
--- a/xlators/cluster/unify/src/Makefile.am
+++ /dev/null
@@ -1,16 +0,0 @@
-
-xlator_LTLIBRARIES = unify.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/legacy/cluster
-
-unify_la_LDFLAGS = -module -avoidversion
-
-unify_la_SOURCES = unify.c unify-self-heal.c
-unify_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-
-noinst_HEADERS = unify.h
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/cluster/unify/src/unify-mem-types.h b/xlators/cluster/unify/src/unify-mem-types.h
deleted file mode 100644
index dcf964779..000000000
--- a/xlators/cluster/unify/src/unify-mem-types.h
+++ /dev/null
@@ -1,41 +0,0 @@
-
-/*
-   Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-
-#ifndef __UNIFY_MEM_TYPES_H__
-#define __UNIFY_MEM_TYPES_H__
-
-#include "mem-types.h"
-
-enum gf_unify_mem_types_ {
-        gf_unify_mt_char = gf_common_mt_end + 1,
-        gf_unify_mt_int16_t,
-        gf_unify_mt_xlator_t,
-        gf_unify_mt_unify_private_t,
-        gf_unify_mt_xlator_list_t,
-        gf_unify_mt_dir_entry_t,
-        gf_unify_mt_off_t,
-        gf_unify_mt_int,
-        gf_unify_mt_unify_self_heal_struct,
-        gf_unify_mt_unify_local_t,
-        gf_unify_mt_end
-};
-#endif
-
diff --git a/xlators/cluster/unify/src/unify-self-heal.c b/xlators/cluster/unify/src/unify-self-heal.c
deleted file mode 100644
index 725523a2e..000000000
--- a/xlators/cluster/unify/src/unify-self-heal.c
+++ /dev/null
@@ -1,1239 +0,0 @@
-/*
-  Copyright (c) 2007-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-/**
- * unify-self-heal.c : 
- *   This file implements few functions which enables 'unify' translator 
- *  to be consistent in its behaviour when 
- *     > a node fails, 
- *     > a node gets added, 
- *     > a failed node comes back
- *     > a new namespace server is added (ie, an fresh namespace server).
- * 
- *  This functionality of 'unify' will enable glusterfs to support storage
- *  system failure, and maintain consistancy. This works both ways, ie, when
- *  an entry (either file or directory) is found on namespace server, and not
- *  on storage nodes, its created in storage nodes and vica-versa.
- * 
- *  The two fops, where it can be implemented are 'getdents ()' and 'lookup ()'
- *
- */
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "unify.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "common-utils.h"
-
-int32_t
-unify_sh_getdents_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       dir_entry_t *entry,
-		       int32_t count);
-
-int32_t
-unify_sh_ns_getdents_cbk (call_frame_t *frame,
-			  void *cookie,
-			  xlator_t *this,
-			  int32_t op_ret,
-			  int32_t op_errno,
-			  dir_entry_t *entry,
-			  int32_t count);
-
-int32_t
-unify_bgsh_getdents_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-			 dir_entry_t *entry,
-			 int32_t count);
-
-int32_t
-unify_bgsh_ns_getdents_cbk (call_frame_t *frame,
-			    void *cookie,
-			    xlator_t *this,
-			    int32_t op_ret,
-			    int32_t op_errno,
-			    dir_entry_t *entry,
-			    int32_t count);
-
-/**
- * unify_local_wipe - free all the extra allocation of local->* here.
- */
-static void 
-unify_local_wipe (unify_local_t *local)
-{
-	/* Free the strdup'd variables in the local structure */
-	if (local->name) {
-		GF_FREE (local->name);
-	}
-
-	if (local->sh_struct) {
-		if (local->sh_struct->offset_list)
-			GF_FREE (local->sh_struct->offset_list);
-
-		if (local->sh_struct->entry_list)
-			GF_FREE (local->sh_struct->entry_list);
-
-		if (local->sh_struct->count_list)
-			GF_FREE (local->sh_struct->count_list);
-
-		GF_FREE (local->sh_struct);
-	}
-
-	loc_wipe (&local->loc1);
-	loc_wipe (&local->loc2);
-}
-
-int32_t 
-unify_sh_setdents_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	inode_t *inode = NULL;
-	dict_t *tmp_dict = NULL;
-	dir_entry_t *prev, *entry, *trav;
-
-	LOCK (&frame->lock);
-	{
-		/* if local->call_count == 0, that means, setdents on 
-		 * storagenodes is still pending.
-		 */
-		if (local->call_count)
-			callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-
-	if (callcnt == 0) {
-		if (local->sh_struct->entry_list[0]) {
-			prev = entry = local->sh_struct->entry_list[0];
-			if (!entry)
-				return 0;
-			trav = entry->next;
-			while (trav) {
-				prev->next = trav->next;
-				GF_FREE (trav->name);
-				if (IA_ISLNK (trav->buf.ia_type))
-					GF_FREE (trav->link);
-				GF_FREE (trav);
-				trav = prev->next;
-			}
-			GF_FREE (entry);
-		}
-
-		if (!local->flags) {
-			if (local->sh_struct->count_list[0] >= 
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-				/* count == size, that means, there are more entries
-				   to read from */
-				//local->call_count = 0;
-				local->sh_struct->offset_list[0] += 
-					UNIFY_SELF_HEAL_GETDENTS_COUNT;
-				STACK_WIND (frame,
-					    unify_sh_ns_getdents_cbk,
-					    NS(this),
-					    NS(this)->fops->getdents,
-					    local->fd,
-					    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-					    local->sh_struct->offset_list[0],
-					    GF_GET_DIR_ONLY);
-			}		
-		} else {
-			inode = local->loc1.inode;
-			fd_unref (local->fd);
-			tmp_dict = local->dict;
-
-			unify_local_wipe (local);
-			
-			STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-				      inode, &local->stbuf, local->dict,
-                                      &local->oldpostparent);
-			if (tmp_dict)
-				dict_unref (tmp_dict);
-		}
-	}
-  
-	return 0;
-}
-
-
-int32_t
-unify_sh_ns_getdents_cbk (call_frame_t *frame,
-			  void *cookie,
-			  xlator_t *this,
-			  int32_t op_ret,
-			  int32_t op_errno,
-			  dir_entry_t *entry,
-			  int32_t count)
-{
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = 0;
-	unsigned long final = 0;
-	dir_entry_t *tmp = GF_CALLOC (1, sizeof (dir_entry_t),
-                                      gf_unify_mt_dir_entry_t);
-	
-	local->sh_struct->entry_list[0] = tmp;
-	local->sh_struct->count_list[0] = count;
-	if (entry) {
-		tmp->next = entry->next;
-		entry->next = NULL;
-	}
-
-	if ((count < UNIFY_SELF_HEAL_GETDENTS_COUNT) || !entry) {
-		final = 1;
-	}
-
-	LOCK (&frame->lock);
-	{
-		/* local->call_count will be '0' till now. make it 1 so, it 
-		   can be UNWIND'ed for the last call. */
-		local->call_count = priv->child_count;
-		if (final)
-			local->flags = 1;
-	}
-	UNLOCK (&frame->lock);
-
-	for (index = 0; index < priv->child_count; index++) 
-	{
-		STACK_WIND_COOKIE (frame,
-				   unify_sh_setdents_cbk, 
-				   (void *)index,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->setdents,
-				   local->fd, GF_SET_DIR_ONLY,
-				   local->sh_struct->entry_list[0], count);
-	}
-
-	return 0;
-}
-
-int32_t 
-unify_sh_ns_setdents_cbk (call_frame_t *frame,
-			  void *cookie,
-			  xlator_t *this,
-			  int32_t op_ret,
-			  int32_t op_errno)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = (long)cookie;
-	dir_entry_t *prev, *entry, *trav;
-
-	LOCK (&frame->lock);
-	{
-		if (local->sh_struct->entry_list[index]) {
-			prev = entry = local->sh_struct->entry_list[index];
-			trav = entry->next;
-			while (trav) {
-				prev->next = trav->next;
-				GF_FREE (trav->name);
-				if (IA_ISLNK (trav->buf.ia_type))
-					GF_FREE (trav->link);
-				GF_FREE (trav);
-				trav = prev->next;
-			}
-			GF_FREE (entry);
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	if (local->sh_struct->count_list[index] < 
-	    UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-		LOCK (&frame->lock);
-		{
-			callcnt = --local->call_count;
-		}
-		UNLOCK (&frame->lock);
-	} else {
-		/* count == size, that means, there are more entries 
-		   to read from */
-		local->sh_struct->offset_list[index] += 
-			UNIFY_SELF_HEAL_GETDENTS_COUNT;
-		STACK_WIND_COOKIE (frame,
-				   unify_sh_getdents_cbk,
-				   cookie,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->getdents,
-				   local->fd,
-				   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-				   local->sh_struct->offset_list[index],
-				   GF_GET_ALL);
-    
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"readdir on (%s) with offset %"PRId64"", 
-			priv->xl_array[index]->name, 
-			local->sh_struct->offset_list[index]);
-	}
-
-	if (!callcnt) {
-		/* All storage nodes have done unified setdents on NS node.
-		 * Now, do getdents from NS and do setdents on storage nodes.
-		 */
-    
-		/* sh_struct->offset_list is no longer required for
-		   storage nodes now */
-		local->sh_struct->offset_list[0] = 0; /* reset */
-
-		STACK_WIND (frame,
-			    unify_sh_ns_getdents_cbk,
-			    NS(this),
-			    NS(this)->fops->getdents,
-			    local->fd,
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-			    0, /* In this call, do send '0' as offset */
-			    GF_GET_DIR_ONLY);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_sh_getdents_cbk -
- */
-int32_t
-unify_sh_getdents_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       dir_entry_t *entry,
-		       int32_t count)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = (long)cookie;
-	dir_entry_t *tmp = NULL; 
-
-	if (op_ret >= 0 && count > 0) {
-		/* There is some dentry found, just send the dentry to NS */
-		tmp = GF_CALLOC (1, sizeof (dir_entry_t),
-                                 gf_unify_mt_dir_entry_t);
-		local->sh_struct->entry_list[index] = tmp;
-		local->sh_struct->count_list[index] = count;
-		if (entry) {
-			tmp->next = entry->next;
-			entry->next = NULL;
-		}
-		STACK_WIND_COOKIE (frame,
-				   unify_sh_ns_setdents_cbk,
-				   cookie,
-				   NS(this),
-				   NS(this)->fops->setdents,
-				   local->fd,
-				   GF_SET_IF_NOT_PRESENT,
-				   local->sh_struct->entry_list[index],
-				   count);
-		return 0;
-	}
-  
-	if (count < UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-		LOCK (&frame->lock);
-		{
-			callcnt = --local->call_count;
-		}
-		UNLOCK (&frame->lock);
-	} else {
-		/* count == size, that means, there are more entries 
-		   to read from */
-		local->sh_struct->offset_list[index] += 
-			UNIFY_SELF_HEAL_GETDENTS_COUNT;
-		STACK_WIND_COOKIE (frame,
-				   unify_sh_getdents_cbk,
-				   cookie,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->getdents,
-				   local->fd,
-				   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-				   local->sh_struct->offset_list[index],
-				   GF_GET_ALL);
-    
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"readdir on (%s) with offset %"PRId64"", 
-			priv->xl_array[index]->name, 
-			local->sh_struct->offset_list[index]);
-	}
-
-	if (!callcnt) {
-		/* All storage nodes have done unified setdents on NS node.
-		 * Now, do getdents from NS and do setdents on storage nodes.
-		 */
-    
-		/* sh_struct->offset_list is no longer required for
-		   storage nodes now */
-		local->sh_struct->offset_list[0] = 0; /* reset */
-
-		STACK_WIND (frame,
-			    unify_sh_ns_getdents_cbk,
-			    NS(this),
-			    NS(this)->fops->getdents,
-			    local->fd,
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-			    0, /* In this call, do send '0' as offset */
-			    GF_GET_DIR_ONLY);
-	}
-
-	return 0;
-}
-
-/**
- * unify_sh_opendir_cbk -
- *
- * @cookie: 
- */
-int32_t 
-unify_sh_opendir_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      fd_t *fd)
-{
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t index = 0;
-	inode_t *inode = NULL;
-	dict_t *tmp_dict = NULL;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret >= 0) {
-			local->op_ret = op_ret;
-		} else {
-			gf_log (this->name, GF_LOG_WARNING, "failed");
-			local->failed = 1;
-		}
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		local->call_count = priv->child_count + 1;
-    
-		if (!local->failed) {
-			/* send getdents() namespace after finishing
-			   storage nodes */
-			local->call_count--; 
-      
-			fd_bind (fd);
-
-			if (local->call_count) {
-				/* Used as the offset index. This list keeps
-				 * track of offset sent to each node during
-				 * STACK_WIND.
-				 */
-				local->sh_struct->offset_list = 
-					GF_CALLOC (priv->child_count, 
-					           sizeof (off_t),
-                                                   gf_unify_mt_off_t);
-				ERR_ABORT (local->sh_struct->offset_list);
-	
-				local->sh_struct->entry_list = 
-					GF_CALLOC (priv->child_count, 
-						   sizeof (dir_entry_t *),
-                                                   gf_unify_mt_dir_entry_t);
-				ERR_ABORT (local->sh_struct->entry_list);
-
-				local->sh_struct->count_list = 
-					GF_CALLOC (priv->child_count, 
-						   sizeof (int),
-                                                   gf_unify_mt_int);
-				ERR_ABORT (local->sh_struct->count_list);
-
-				/* Send getdents on all the fds */
-				for (index = 0; 
-				     index < priv->child_count; index++) {
-					STACK_WIND_COOKIE (frame,
-							   unify_sh_getdents_cbk,
-							   (void *)(long)index,
-							   priv->xl_array[index],
-							   priv->xl_array[index]->fops->getdents,
-							   local->fd,
-							   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-							   0, /* In this call, do send '0' as offset */
-							   GF_GET_ALL);
-				}
-
-				/* did stack wind, so no need to unwind here */
-				return 0;
-			} /* (local->call_count) */
-		} /* (!local->failed) */
-
-		/* Opendir failed on one node. */ 
-		inode = local->loc1.inode;
-		fd_unref (local->fd);
-		tmp_dict = local->dict;
-
-		unify_local_wipe (local);
-		/* Only 'self-heal' failed, lookup() was successful. */
-		local->op_ret = 0;
-
-		/* This is lookup_cbk ()'s UNWIND. */
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, inode,
-			      &local->stbuf, local->dict, &local->oldpostparent);
-		if (tmp_dict)
-			dict_unref (tmp_dict);
-	}
-
-	return 0;
-}
-
-/**
- * gf_sh_checksum_cbk - 
- * 
- * @frame: frame used in lookup. get a copy of it, and use that copy.
- * @this: pointer to unify xlator.
- * @inode: pointer to inode, for which the consistency check is required.
- *
- */
-int32_t 
-unify_sh_checksum_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       uint8_t *file_checksum,
-		       uint8_t *dir_checksum)
-{
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t index = 0;
-	int32_t callcnt = 0;
-	inode_t *inode = NULL;
-	dict_t *tmp_dict = NULL;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret >= 0) {
-			if (NS(this) == (xlator_t *)cookie) {
-				memcpy (local->sh_struct->ns_file_checksum, 
-					file_checksum, NAME_MAX);
-				memcpy (local->sh_struct->ns_dir_checksum, 
-					dir_checksum, NAME_MAX);
-			} else {
-				if (local->entry_count == 0) {
-					/* Initialize the dir_checksum to be 
-					 * used for comparision with other
-					 * storage nodes. Should be done for
-					 * the first successful call *only*. 
-					 */
-                                        /* Using 'entry_count' as a flag */
-					local->entry_count = 1;
-					memcpy (local->sh_struct->dir_checksum,
-						dir_checksum, NAME_MAX);
-				}
-
-				/* Reply from the storage nodes */
-				for (index = 0; 
-				     index < NAME_MAX; index++) {
-					/* Files should be present in
-					   only one node */
-					local->sh_struct->file_checksum[index] ^= file_checksum[index];
-	  
-					/* directory structure should be
-					   same accross */
-					if (local->sh_struct->dir_checksum[index] != dir_checksum[index])
-						local->failed = 1;
-				}
-			}
-		} 
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		for (index = 0; index < NAME_MAX ; index++) {
-			if (local->sh_struct->file_checksum[index] != 
-			    local->sh_struct->ns_file_checksum[index]) {
-				local->failed = 1;
-				break;
-			}
-			if (local->sh_struct->dir_checksum[index] != 
-			    local->sh_struct->ns_dir_checksum[index]) {
-				local->failed = 1;
-				break;
-			}
-		}
-	
-		if (local->failed) {
-			/* Log it, it should be a rare event */
-			gf_log (this->name, GF_LOG_WARNING, 
-				"Self-heal triggered on directory %s", 
-				local->loc1.path);
-
-			/* Any self heal will be done at directory level */
-			local->call_count = 0;
-			local->op_ret = -1;
-			local->failed = 0;
-      
-			local->fd = fd_create (local->loc1.inode, 
-					       frame->root->pid);
-
-			local->call_count = priv->child_count + 1;
-	
-			for (index = 0; 
-			     index < (priv->child_count + 1); index++) {
-				STACK_WIND_COOKIE (frame,
-						   unify_sh_opendir_cbk,
-						   priv->xl_array[index]->name,
-						   priv->xl_array[index],
-						   priv->xl_array[index]->fops->opendir,
-						   &local->loc1,
-						   local->fd);
-			}
-			/* opendir can be done on the directory */
-			return 0;
-		}
-
-		/* no mismatch */
-		inode = local->loc1.inode;
-		tmp_dict = local->dict;
-
-		unify_local_wipe (local);
-
-		/* This is lookup_cbk ()'s UNWIND. */
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      inode,
-			      &local->stbuf,
-			      local->dict, &local->oldpostparent);
-		if (tmp_dict)
-			dict_unref (tmp_dict);
-	}
-
-	return 0;
-}
-
-/* Foreground self-heal part over */
-
-/* Background self-heal part */
-
-int32_t 
-unify_bgsh_setdents_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	dir_entry_t *prev, *entry, *trav;
-
-	LOCK (&frame->lock);
-	{
-		/* if local->call_count == 0, that means, setdents 
-		   on storagenodes is still pending. */
-		if (local->call_count)
-			callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-
-
-	if (callcnt == 0) {
-		if (local->sh_struct->entry_list[0]) {
-			prev = entry = local->sh_struct->entry_list[0];
-			trav = entry->next;
-			while (trav) {
-				prev->next = trav->next;
-				GF_FREE (trav->name);
-				if (IA_ISLNK (trav->buf.ia_type))
-					GF_FREE (trav->link);
-				GF_FREE (trav);
-				trav = prev->next;
-			}
-			GF_FREE (entry);
-		}
-
-		if (!local->flags) {
-			if (local->sh_struct->count_list[0] >= 
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-				/* count == size, that means, there are more
-				   entries to read from */
-				//local->call_count = 0;
-				local->sh_struct->offset_list[0] += 
-					UNIFY_SELF_HEAL_GETDENTS_COUNT;
-				STACK_WIND (frame,
-					    unify_bgsh_ns_getdents_cbk,
-					    NS(this),
-					    NS(this)->fops->getdents,
-					    local->fd,
-					    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-					    local->sh_struct->offset_list[0],
-					    GF_GET_DIR_ONLY);
-			}		
-		} else {
-			fd_unref (local->fd);
-			unify_local_wipe (local);
-			STACK_DESTROY (frame->root);
-		}
-	}
-
-	return 0;
-}
-
-
-int32_t
-unify_bgsh_ns_getdents_cbk (call_frame_t *frame,
-			    void *cookie,
-			    xlator_t *this,
-			    int32_t op_ret,
-			    int32_t op_errno,
-			    dir_entry_t *entry,
-			    int32_t count)
-{
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = 0;
-	unsigned long final = 0;
-	dir_entry_t *tmp = GF_CALLOC (1, sizeof (dir_entry_t),
-                                      gf_unify_mt_dir_entry_t);
-
-	local->sh_struct->entry_list[0] = tmp;
-	local->sh_struct->count_list[0] = count;
-	if (entry) {
-		tmp->next = entry->next;
-		entry->next = NULL;
-	}
-
-	if ((count < UNIFY_SELF_HEAL_GETDENTS_COUNT) || !entry) {
-		final = 1;
-	}
-
-	LOCK (&frame->lock);
-	{
-		/* local->call_count will be '0' till now. make it 1 so, 
-		   it can be UNWIND'ed for the last call. */
-		local->call_count = priv->child_count;
-		if (final)
-			local->flags = 1;
-	}
-	UNLOCK (&frame->lock);
-
-	for (index = 0; index < priv->child_count; index++) 
-	{
-		STACK_WIND_COOKIE (frame,
-				   unify_bgsh_setdents_cbk, 
-				   (void *)index,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->setdents,
-				   local->fd, GF_SET_DIR_ONLY,
-				   local->sh_struct->entry_list[0], count);
-	}
-
-	return 0;
-}
-
-int32_t 
-unify_bgsh_ns_setdents_cbk (call_frame_t *frame,
-			    void *cookie,
-			    xlator_t *this,
-			    int32_t op_ret,
-			    int32_t op_errno)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = (long)cookie;
-	dir_entry_t *prev, *entry, *trav;
-
-	if (local->sh_struct->entry_list[index]) {
-		prev = entry = local->sh_struct->entry_list[index];
-		if (!entry)
-			return 0;
-		trav = entry->next;
-		while (trav) {
-			prev->next = trav->next;
-			GF_FREE (trav->name);
-			if (IA_ISLNK (trav->buf.ia_type))
-				GF_FREE (trav->link);
-			GF_FREE (trav);
-			trav = prev->next;
-		}
-		GF_FREE (entry);
-	}
-
-	if (local->sh_struct->count_list[index] < 
-	    UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-		LOCK (&frame->lock);
-		{
-			callcnt = --local->call_count;
-		}
-		UNLOCK (&frame->lock);
-	} else {
-		/* count == size, that means, there are more entries 
-		   to read from */
-		local->sh_struct->offset_list[index] += 
-			UNIFY_SELF_HEAL_GETDENTS_COUNT;
-		STACK_WIND_COOKIE (frame,
-				   unify_bgsh_getdents_cbk,
-				   cookie,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->getdents,
-				   local->fd,
-				   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-				   local->sh_struct->offset_list[index],
-				   GF_GET_ALL);
-    
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"readdir on (%s) with offset %"PRId64"", 
-			priv->xl_array[index]->name, 
-			local->sh_struct->offset_list[index]);
-	}
-
-	if (!callcnt) {
-		/* All storage nodes have done unified setdents on NS node.
-		 * Now, do getdents from NS and do setdents on storage nodes.
-		 */
-    
-		/* sh_struct->offset_list is no longer required for
-		   storage nodes now */
-		local->sh_struct->offset_list[0] = 0; /* reset */
-
-		STACK_WIND (frame,
-			    unify_bgsh_ns_getdents_cbk,
-			    NS(this),
-			    NS(this)->fops->getdents,
-			    local->fd,
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-			    0, /* In this call, do send '0' as offset */
-			    GF_GET_DIR_ONLY);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_bgsh_getdents_cbk -
- */
-int32_t
-unify_bgsh_getdents_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-			 dir_entry_t *entry,
-			 int32_t count)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = (long)cookie;
-	dir_entry_t *tmp = NULL; 
-
-	if (op_ret >= 0 && count > 0) {
-		/* There is some dentry found, just send the dentry to NS */
-		tmp = GF_CALLOC (1, sizeof (dir_entry_t),
-                                 gf_unify_mt_dir_entry_t);
-		local->sh_struct->entry_list[index] = tmp;
-		local->sh_struct->count_list[index] = count;
-		if (entry) {
-			tmp->next = entry->next;
-			entry->next = NULL;
-		}
-		STACK_WIND_COOKIE (frame,
-				   unify_bgsh_ns_setdents_cbk,
-				   cookie,
-				   NS(this),
-				   NS(this)->fops->setdents,
-				   local->fd,
-				   GF_SET_IF_NOT_PRESENT,
-				   local->sh_struct->entry_list[index],
-				   count);
-		return 0;
-	}
-  
-	if (count < UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-		LOCK (&frame->lock);
-		{
-			callcnt = --local->call_count;
-		}
-		UNLOCK (&frame->lock);
-	} else {
-		/* count == size, that means, there are more entries to read from */
-		local->sh_struct->offset_list[index] += 
-			UNIFY_SELF_HEAL_GETDENTS_COUNT;
-
-		STACK_WIND_COOKIE (frame,
-				   unify_bgsh_getdents_cbk,
-				   cookie,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->getdents,
-				   local->fd,
-				   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-				   local->sh_struct->offset_list[index],
-				   GF_GET_ALL);
-    
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"readdir on (%s) with offset %"PRId64"", 
-			priv->xl_array[index]->name, 
-			local->sh_struct->offset_list[index]);
-	}
-
-	if (!callcnt) {
-		/* All storage nodes have done unified setdents on NS node.
-		 * Now, do getdents from NS and do setdents on storage nodes.
-		 */
-    
-		/* sh_struct->offset_list is no longer required for 
-		   storage nodes now */
-		local->sh_struct->offset_list[0] = 0; /* reset */
-
-		STACK_WIND (frame,
-			    unify_bgsh_ns_getdents_cbk,
-			    NS(this),
-			    NS(this)->fops->getdents,
-			    local->fd,
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-			    0, /* In this call, do send '0' as offset */
-			    GF_GET_DIR_ONLY);
-	}
-
-	return 0;
-}
-
-/**
- * unify_bgsh_opendir_cbk -
- *
- * @cookie: 
- */
-int32_t 
-unify_bgsh_opendir_cbk (call_frame_t *frame,
-			void *cookie,
-			xlator_t *this,
-			int32_t op_ret,
-			int32_t op_errno,
-			fd_t *fd)
-{
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int32_t callcnt = 0;
-	int16_t index = 0;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret >= 0) {
-			local->op_ret = op_ret;
-		} else {
-			local->failed = 1;
-		}
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		local->call_count = priv->child_count + 1;
-    
-		if (!local->failed) {
-			/* send getdents() namespace after finishing 
-			   storage nodes */
-			local->call_count--; 
-			callcnt = local->call_count;
-      
-			fd_bind (fd);
-
-			if (local->call_count) {
-				/* Used as the offset index. This list keeps 
-				   track of offset sent to each node during 
-				   STACK_WIND. */
-				local->sh_struct->offset_list = 
-					GF_CALLOC (priv->child_count, 
-					           sizeof (off_t),
-                                                   gf_unify_mt_off_t);
-				ERR_ABORT (local->sh_struct->offset_list);
-	
-				local->sh_struct->entry_list = 
-					GF_CALLOC (priv->child_count, 
-						   sizeof (dir_entry_t *),
-                                                   gf_unify_mt_dir_entry_t);
-				ERR_ABORT (local->sh_struct->entry_list);
-
-				local->sh_struct->count_list = 
-					GF_CALLOC (priv->child_count, 
-						   sizeof (int),
-                                                   gf_unify_mt_int);
-				ERR_ABORT (local->sh_struct->count_list);
-
-				/* Send getdents on all the fds */
-				for (index = 0; 
-				     index < priv->child_count; index++) {
-					STACK_WIND_COOKIE (frame,
-							   unify_bgsh_getdents_cbk,
-							   (void *)(long)index,
-							   priv->xl_array[index],
-							   priv->xl_array[index]->fops->getdents,
-							   local->fd,
-							   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-							   0, /* In this call, do send '0' as offset */
-							   GF_GET_ALL);
-				}
-				/* did a stack wind, so no need to unwind here */
-				return 0;
-			} /* (local->call_count) */
-		} /* (!local->failed) */
-
-		/* Opendir failed on one node. 	 */
-		fd_unref (local->fd);
-		
-		unify_local_wipe (local);
-		STACK_DESTROY (frame->root);
-	}
-
-	return 0;
-}
-
-/**
- * gf_bgsh_checksum_cbk - 
- * 
- * @frame: frame used in lookup. get a copy of it, and use that copy.
- * @this: pointer to unify xlator.
- * @inode: pointer to inode, for which the consistency check is required.
- *
- */
-int32_t 
-unify_bgsh_checksum_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-			 uint8_t *file_checksum,
-			 uint8_t *dir_checksum)
-{
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t index = 0;
-	int32_t callcnt = 0;
-  
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret >= 0) {
-			if (NS(this) == (xlator_t *)cookie) {
-				memcpy (local->sh_struct->ns_file_checksum, 
-					file_checksum, NAME_MAX);
-				memcpy (local->sh_struct->ns_dir_checksum, 
-					dir_checksum, NAME_MAX);
-			} else {
-				if (local->entry_count == 0) {
-					/* Initialize the dir_checksum to be 
-					 * used for comparision with other 
-					 * storage nodes. Should be done for
-					 * the first successful call *only*. 
-					 */
-					/* Using 'entry_count' as a flag */
-					local->entry_count = 1; 
-					memcpy (local->sh_struct->dir_checksum,
-						dir_checksum, NAME_MAX);
-				}
-
-				/* Reply from the storage nodes */
-				for (index = 0; 
-				     index < NAME_MAX; index++) {
-					/* Files should be present in only 
-					   one node */
-					local->sh_struct->file_checksum[index] ^= file_checksum[index];
-	  
-					/* directory structure should be same 
-					   accross */
-					if (local->sh_struct->dir_checksum[index] != dir_checksum[index])
-						local->failed = 1;
-				}
-			}
-		} 
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		for (index = 0; index < NAME_MAX ; index++) {
-			if (local->sh_struct->file_checksum[index] != 
-			    local->sh_struct->ns_file_checksum[index]) {
-				local->failed = 1;
-				break;
-			}
-			if (local->sh_struct->dir_checksum[index] != 
-			    local->sh_struct->ns_dir_checksum[index]) {
-				local->failed = 1;
-				break;
-			}
-		}
-	
-		if (local->failed) {
-			/* Log it, it should be a rare event */
-			gf_log (this->name, GF_LOG_WARNING, 
-				"Self-heal triggered on directory %s", 
-				local->loc1.path);
-
-			/* Any self heal will be done at the directory level */
-			local->op_ret = -1;
-			local->failed = 0;
-      
-			local->fd = fd_create (local->loc1.inode, 
-					       frame->root->pid);
-			local->call_count = priv->child_count + 1;
-	
-			for (index = 0; 
-			     index < (priv->child_count + 1); index++) {
-				STACK_WIND_COOKIE (frame,
-						   unify_bgsh_opendir_cbk,
-						   priv->xl_array[index]->name,
-						   priv->xl_array[index],
-						   priv->xl_array[index]->fops->opendir,
-						   &local->loc1,
-						   local->fd);
-			}
-      
-			/* opendir can be done on the directory */
-			return 0;
-		}
-
-		/* no mismatch */
-		unify_local_wipe (local);
-		STACK_DESTROY (frame->root);
-	}
-
-	return 0;
-}
-
-/* Background self-heal part over */
-
-
-
-
-/**
- * zr_unify_self_heal - 
- * 
- * @frame: frame used in lookup. get a copy of it, and use that copy.
- * @this: pointer to unify xlator.
- * @inode: pointer to inode, for which the consistency check is required.
- *
- */
-int32_t 
-zr_unify_self_heal (call_frame_t *frame,
-		    xlator_t *this,
-		    unify_local_t *local)
-{
-	unify_private_t *priv = this->private;
-	call_frame_t *bg_frame = NULL;
-	unify_local_t *bg_local = NULL;
-	inode_t *tmp_inode = NULL;
-	dict_t *tmp_dict = NULL;
-	int16_t index = 0;
-  
-	if (local->inode_generation < priv->inode_generation) {
-		/* Any self heal will be done at the directory level */
-		/* Update the inode's generation to the current generation
-		   value. */
-		local->inode_generation = priv->inode_generation;
-		inode_ctx_put (local->loc1.inode, this, 
-			  (uint64_t)(long)local->inode_generation);
-
-		if (priv->self_heal == ZR_UNIFY_FG_SELF_HEAL) {
-			local->op_ret = 0;
-			local->failed = 0;
-			local->call_count = priv->child_count + 1;
-			local->sh_struct = 
-	         	     GF_CALLOC (1, sizeof (struct unify_self_heal_struct),
-                                        gf_unify_mt_unify_self_heal_struct);
-      
-			/* +1 is for NS */
-			for (index = 0; 
-			     index < (priv->child_count + 1); index++) {
-				STACK_WIND_COOKIE (frame,
-						   unify_sh_checksum_cbk,
-						   priv->xl_array[index],
-						   priv->xl_array[index],
-						   priv->xl_array[index]->fops->checksum,
-						   &local->loc1,
-						   0);
-			}
-
-			/* Self-heal in foreground, hence no need 
-			   to UNWIND here */
-			return 0;
-		}
-
-		/* Self Heal done in background */
-		bg_frame = copy_frame (frame);
-		INIT_LOCAL (bg_frame, bg_local);
-		loc_copy (&bg_local->loc1, &local->loc1);
-		bg_local->op_ret = 0;
-		bg_local->failed = 0;
-		bg_local->call_count = priv->child_count + 1;
-		bg_local->sh_struct = 
-			GF_CALLOC (1, sizeof (struct unify_self_heal_struct),
-                                   gf_unify_mt_unify_self_heal_struct);
-    
-		/* +1 is for NS */
-		for (index = 0; index < (priv->child_count + 1); index++) {
-			STACK_WIND_COOKIE (bg_frame,
-					   unify_bgsh_checksum_cbk,
-					   priv->xl_array[index],
-					   priv->xl_array[index],
-					   priv->xl_array[index]->fops->checksum,
-					   &bg_local->loc1,
-					   0);
-		}
-	}
-
-	/* generation number matches, self heal already done or
-	 * self heal done in background: just do STACK_UNWIND 
-	 */
-	tmp_inode = local->loc1.inode;
-	tmp_dict = local->dict;
-
-	unify_local_wipe (local);
-
-	/* This is lookup_cbk ()'s UNWIND. */
-	STACK_UNWIND (frame,
-		      local->op_ret,
-		      local->op_errno,
-		      tmp_inode,
-		      &local->stbuf,
-		      local->dict,
-                      &local->oldpostparent);
-
-	if (tmp_dict)
-		dict_unref (tmp_dict);
-
-	return 0;
-}
-
diff --git a/xlators/cluster/unify/src/unify.c b/xlators/cluster/unify/src/unify.c
deleted file mode 100644
index 422c8d6d8..000000000
--- a/xlators/cluster/unify/src/unify.c
+++ /dev/null
@@ -1,4589 +0,0 @@
-/*
-  Copyright (c) 2006-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-/**
- * xlators/cluster/unify:
- *     - This xlator is one of the main translator in GlusterFS, which
- *   actually does the clustering work of the file system. One need to 
- *   understand that, unify assumes file to be existing in only one of 
- *   the child node, and directories to be present on all the nodes. 
- *
- * NOTE:
- *   Now, unify has support for global namespace, which is used to keep a 
- * global view of fs's namespace tree. The stat for directories are taken
- * just from the namespace, where as for files, just 'ia_ino' is taken from
- * Namespace node, and other stat info is taken from the actual storage node.
- * Also Namespace node helps to keep consistant inode for files across 
- * glusterfs (re-)mounts.
- */
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "unify.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include <signal.h>
-#include <libgen.h>
-#include "compat-errno.h"
-#include "compat.h"
-
-#define UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR(_loc) do { \
-  if (!(_loc && _loc->inode)) {                            \
-    STACK_UNWIND (frame, -1, EINVAL, NULL, NULL, NULL);    \
-    return 0;                                              \
-  }                                                        \
-} while(0)
-
-
-#define UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR(_fd) do { \
-  if (!(_fd && !fd_ctx_get (_fd, this, NULL))) {       \
-    STACK_UNWIND (frame, -1, EBADFD, NULL, NULL);      \
-    return 0;                                          \
-  }                                                    \
-} while(0)
-
-#define UNIFY_CHECK_FD_AND_UNWIND_ON_ERR(_fd) do { \
-  if (!_fd) {                                      \
-    STACK_UNWIND (frame, -1, EBADFD, NULL, NULL);  \
-    return 0;                                      \
-  }                                                \
-} while(0)
-
-/**
- * unify_local_wipe - free all the extra allocation of local->* here.
- */
-static void 
-unify_local_wipe (unify_local_t *local)
-{
-	/* Free the strdup'd variables in the local structure */
-	if (local->name) {
-		GF_FREE (local->name);
-	}
-	loc_wipe (&local->loc1);
-	loc_wipe (&local->loc2);
-}
-
-
-
-/*
- * unify_normalize_stats -
- */
-void
-unify_normalize_stats (struct statvfs *buf,
-		       unsigned long bsize,
-		       unsigned long frsize)
-{
-	double factor;
-
-	if (buf->f_bsize != bsize) {
-		factor = ((double) buf->f_bsize) / bsize;
-		buf->f_bsize  = bsize;
-		buf->f_bfree  = (fsblkcnt_t) (factor * buf->f_bfree);
-		buf->f_bavail = (fsblkcnt_t) (factor * buf->f_bavail);
-	}
-  
-	if (buf->f_frsize != frsize) {
-		factor = ((double) buf->f_frsize) / frsize;
-		buf->f_frsize = frsize;
-		buf->f_blocks = (fsblkcnt_t) (factor * buf->f_blocks);
-	}
-}
-
-
-xlator_t *
-unify_loc_subvol (loc_t *loc, xlator_t *this)
-{
-	unify_private_t *priv = NULL;
-	xlator_t        *subvol = NULL;
-	int16_t         *list = NULL;
-	long             index = 0;
-	xlator_t        *subvol_i = NULL;
-	int              ret = 0;
-	uint64_t         tmp_list = 0;
-
-	priv   = this->private;
-	subvol = NS (this);
-
-	if (!IA_ISDIR (loc->inode->ia_type)) {
-		ret = inode_ctx_get (loc->inode, this, &tmp_list);
-		list = (int16_t *)(long)tmp_list;
-		if (!list)
-			goto out;
-
-		for (index = 0; list[index] != -1; index++) {
-			subvol_i = priv->xl_array[list[index]];
-			if (subvol_i != NS (this)) {
-				subvol = subvol_i;
-				break;
-			}
-		}
-	}
-out:
-	return subvol;
-}
-
-
-
-/**
- * unify_statfs_cbk -
- */
-int32_t
-unify_statfs_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct statvfs *stbuf)
-{
-	int32_t callcnt = 0;
-	struct statvfs *dict_buf = NULL;
-	unsigned long bsize;
-	unsigned long frsize;
-	unify_local_t *local = (unify_local_t *)frame->local;
-	call_frame_t *prev_frame = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret >= 0) {
-			/* when a call is successfull, add it to local->dict */
-			dict_buf = &local->statvfs_buf;
-
-			if (dict_buf->f_bsize != 0) {
-				bsize  = max (dict_buf->f_bsize, 
-					      stbuf->f_bsize);
-
-				frsize = max (dict_buf->f_frsize, 
-					      stbuf->f_frsize);
-				unify_normalize_stats(dict_buf, bsize, frsize);
-				unify_normalize_stats(stbuf, bsize, frsize);
-			} else {
-				dict_buf->f_bsize   = stbuf->f_bsize;
-				dict_buf->f_frsize  = stbuf->f_frsize;
-			}
-      
-			dict_buf->f_blocks += stbuf->f_blocks;
-			dict_buf->f_bfree  += stbuf->f_bfree;
-			dict_buf->f_bavail += stbuf->f_bavail;
-			dict_buf->f_files  += stbuf->f_files;
-			dict_buf->f_ffree  += stbuf->f_ffree;
-			dict_buf->f_favail += stbuf->f_favail;
-			dict_buf->f_fsid    = stbuf->f_fsid;
-			dict_buf->f_flag    = stbuf->f_flag;
-			dict_buf->f_namemax = stbuf->f_namemax;
-			local->op_ret = op_ret;
-		} else {
-			/* fop on storage node has failed due to some error */
-			if (op_errno != ENOTCONN) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"child(%s): %s", 
-					prev_frame->this->name, 
-					strerror (op_errno));
-			}
-			local->op_errno = op_errno;
-		}
-		callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      &local->statvfs_buf);
-	}
-
-	return 0;
-}
-
-/**
- * unify_statfs -
- */
-int32_t
-unify_statfs (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc)
-{
-	unify_local_t *local = NULL;
-	xlator_list_t *trav = this->children;
-
-	INIT_LOCAL (frame, local);
-	local->call_count = ((unify_private_t *)this->private)->child_count;
-
-	while(trav) {
-		STACK_WIND (frame,
-			    unify_statfs_cbk,
-			    trav->xlator,
-			    trav->xlator->fops->statfs,
-			    loc);
-		trav = trav->next;
-	}
-
-	return 0;
-}
-
-/**
- * unify_buf_cbk - 
- */
-int32_t
-unify_buf_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       struct iatt *buf)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"%s(): child(%s): path(%s): %s", 
-				gf_fop_list[frame->root->op],
-				prev_frame->this->name, 
-				(local->loc1.path)?local->loc1.path:"", 
-				strerror (op_errno));
-
-			local->op_errno = op_errno;
-			if ((op_errno == ENOENT) && priv->optimist) 
-				local->op_ret = 0;
-		}
-
-		if (op_ret >= 0) {
-			local->op_ret = 0;
-
-			if (NS (this) == prev_frame->this) {
-				local->ia_ino = buf->ia_ino;
-				/* If the entry is directory, get the stat
-				   from NS node */
-				if (IA_ISDIR (buf->ia_type) || 
-				    !local->stbuf.ia_blksize) {
-					local->stbuf = *buf;
-				}
-			}
-
-			if ((!IA_ISDIR (buf->ia_type)) && 
-			    (NS (this) != prev_frame->this)) {
-				/* If file, take the stat info from Storage 
-				   node. */
-				local->stbuf = *buf;
-			}
-		}
-	}
-	UNLOCK (&frame->lock);
-    
-	if (!callcnt) {
-		/* If the inode number is not filled, operation should
-		   fail */
-		if (!local->ia_ino)
-			local->op_ret = -1;
-
-		local->stbuf.ia_ino = local->ia_ino;
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      &local->stbuf);
-	}
-
-	return 0;
-}
-
-#define check_if_dht_linkfile(s) \
-        ((st_mode_from_ia (s->ia_prot, s->ia_type) & ~S_IFMT) == S_ISVTX)
-
-/**
- * unify_lookup_cbk - 
- */
-int32_t 
-unify_lookup_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  inode_t *inode,
-		  struct iatt *buf,
-		  dict_t *dict,
-                  struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	inode_t *tmp_inode = NULL;
-	dict_t *local_dict = NULL;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
- 
-		if (op_ret == -1) {
-                        if (local->revalidate &&
-                            (op_errno == ESTALE)) {
-                                /* ESTALE takes priority */
-                                local->op_errno = op_errno;
-                                local->failed = 1;
-                        }
-
-			if ((op_errno != ENOTCONN) 
-                            && (op_errno != ENOENT)
-                            && (local->op_errno != ESTALE)) {
-                                /* if local->op_errno is already ESTALE, then
-                                 * ESTALE has to propogated to the parent first.
-                                 * do not enter here.
-                                 */
-				gf_log (this->name, GF_LOG_ERROR,
-					"child(%s): path(%s): %s", 
-					priv->xl_array[(long)cookie]->name, 
-					local->loc1.path, strerror (op_errno));
-                                local->op_errno = op_errno;
-				local->failed = 1;
-
-			} else if (local->revalidate && 
-                                   (local->op_errno != ESTALE) &&
-				   !(priv->optimist && (op_errno == ENOENT))) {
-
-				gf_log (this->name, 
-					(op_errno == ENOTCONN) ? 
-					GF_LOG_DEBUG:GF_LOG_ERROR,
-					"child(%s): path(%s): %s", 
-					priv->xl_array[(long)cookie]->name, 
-					local->loc1.path, strerror (op_errno));
-				local->op_errno = op_errno;
-				local->failed = 1;
-			}
-		}
-
-		if (op_ret == 0) {
-			local->op_ret = 0; 
-			
-			if (check_if_dht_linkfile(buf)) {
-				gf_log (this->name, GF_LOG_CRITICAL,
-					"file %s may be DHT link file on %s, "
-					"make sure the backend is not shared "
-					"between unify and DHT", 
-					local->loc1.path, 
-					priv->xl_array[(long)cookie]->name);
-			}
-
-			if (local->stbuf.ia_type && local->stbuf.ia_blksize) {
-				/* make sure we already have a stbuf
-				   stored in local->stbuf */
-				if (IA_ISDIR (local->stbuf.ia_type) && 
-				    !IA_ISDIR (buf->ia_type)) {
-					gf_log (this->name, GF_LOG_CRITICAL, 
-						"[CRITICAL] '%s' is directory "
-						"on namespace, non-directory "
-						"on node '%s', returning EIO",
-						local->loc1.path, 
-						priv->xl_array[(long)cookie]->name);
-					local->return_eio = 1;
-				}
-				if (!IA_ISDIR (local->stbuf.ia_type) && 
-				    IA_ISDIR (buf->ia_type)) {
-					gf_log (this->name, GF_LOG_CRITICAL, 
-						"[CRITICAL] '%s' is directory "
-						"on node '%s', non-directory "
-						"on namespace, returning EIO",
-						local->loc1.path, 
-						priv->xl_array[(long)cookie]->name);
-					local->return_eio = 1;
-				}
-			}
-	
-			if (!local->revalidate && !IA_ISDIR (buf->ia_type)) {
-				/* This is the first time lookup on file*/
-				if (!local->list) {
-					/* list is not allocated, allocate 
-					   the max possible range */
-					local->list = GF_CALLOC (1, 2 * (priv->child_count + 2),
-                                                                 gf_unify_mt_int16_t);
-					if (!local->list) {
-						gf_log (this->name, 
-							GF_LOG_CRITICAL, 
-							"Not enough memory");
-						STACK_UNWIND (frame, -1, 
-							      ENOMEM, inode, 
-							      NULL, NULL, NULL);
-						return 0;
-					}
-				}
-				/* update the index of the list */
-				local->list [local->index++] = 
-					(int16_t)(long)cookie;
-			}
-                        
-                        if (!local->revalidate && IA_ISDIR (buf->ia_type)) {
-                                /* fresh lookup of a directory */
-                                inode_ctx_put (local->loc1.inode, this, 
-                                               priv->inode_generation);
-                        }
-
-			if ((!local->dict) && dict &&
-			    (priv->xl_array[(long)cookie] != NS(this)))	{
-				local->dict = dict_ref (dict);
-			}
-
-			/* index of NS node is == total child count */
-			if (priv->child_count == (int16_t)(long)cookie) {
-				/* Take the inode number from namespace */
-				local->ia_ino = buf->ia_ino;
-				if (IA_ISDIR (buf->ia_type) || 
-				    !(local->stbuf.ia_blksize)) {
-					local->stbuf = *buf;
-                                        local->oldpostparent = *postparent;
-				}
-			} else if (!IA_ISDIR (buf->ia_type)) {
-				/* If file, then get the stat from 
-				   storage node */
-				local->stbuf = *buf;
-			}
-
-			if (local->ia_nlink < buf->ia_nlink) {
-				local->ia_nlink = buf->ia_nlink;
-			}
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		local_dict = local->dict;
-		if (local->return_eio) {
-			gf_log (this->name, GF_LOG_CRITICAL, 
-				"[CRITICAL] Unable to fix the path (%s) with "
-				"self-heal, try manual verification. "
-				"returning EIO.", local->loc1.path);
-			unify_local_wipe (local);
-			STACK_UNWIND (frame, -1, EIO, inode, NULL, NULL);
-			if (local_dict)	{
-				dict_unref (local_dict);
-			}
-			return 0;
-		}
-
-		if (!local->stbuf.ia_blksize) {
-			/* Inode not present */
-			local->op_ret = -1;
-		} else {
-			if (!local->revalidate && 
-			    !IA_ISDIR (local->stbuf.ia_type)) { 
-				/* If its a file, big array is useless, 
-				   allocate the smaller one */
-				int16_t *list = NULL;
-				list = GF_CALLOC (1, 2 * (local->index + 1),
-                                                  gf_unify_mt_int16_t);
-				ERR_ABORT (list);
-				memcpy (list, local->list, 2 * local->index);
-				/* Make the end of the list as -1 */
-				GF_FREE (local->list);
-				local->list = list;
-				local->list [local->index] = -1;
-				/* Update the inode's ctx with proper array */
-				/* TODO: log on failure */
-				inode_ctx_put (local->loc1.inode, this, 
-					       (uint64_t)(long)local->list);
-			}
-
-			if (IA_ISDIR(local->loc1.inode->ia_type)) {
-				/* lookup is done for directory */
-				if (local->failed && priv->self_heal) {
-					/* Triggering self-heal */
-                                        /* means, self-heal required for this 
-					   inode */
-					local->inode_generation = 0; 
-					priv->inode_generation++;
-				}
-			} else {
-				local->stbuf.ia_ino = local->ia_ino;
-			}
-	  
-			local->stbuf.ia_nlink = local->ia_nlink;
-		}
-		if (local->op_ret == -1) {
-			if (!local->revalidate && local->list)
-				GF_FREE (local->list);
-		}
-
-		if ((local->op_ret >= 0) && local->failed && 
-		    local->revalidate) {
-			/* Done revalidate, but it failed */
-			if ((op_errno != ENOTCONN) 
-                            && (local->op_errno != ESTALE)) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"Revalidate failed for path(%s): %s", 
-					local->loc1.path, strerror (op_errno));
-			}
-			local->op_ret = -1;
-		}
-
-		if ((priv->self_heal && !priv->optimist) && 
-		    (!local->revalidate && (local->op_ret == 0) && 
-		     IA_ISDIR(local->stbuf.ia_type))) {
-			/* Let the self heal be done here */
-			zr_unify_self_heal (frame, this, local);
-			local_dict = NULL;
-		} else {
-                        if (local->failed) {
-                                /* NOTE: directory lookup is sent to all 
-                                 * subvolumes and success from a subvolume
-                                 * might set local->op_ret to 0 (zero) */
-                                local->op_ret = -1;
-                        }
-
-			/* either no self heal, or op_ret == -1 (failure) */
-			tmp_inode = local->loc1.inode;
-			unify_local_wipe (local);
-			STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-				      tmp_inode, &local->stbuf, local->dict,
-                                      &local->oldpostparent);
-		}
-		if (local_dict) {
-			dict_unref (local_dict);
-		}
-	}
-  
-	return 0;
-}
-
-/**
- * unify_lookup - 
- */
-int32_t 
-unify_lookup (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      dict_t *xattr_req)
-{
-	unify_local_t *local = NULL;
-	unify_private_t *priv = this->private;
-	int16_t *list = NULL;
-	long index = 0;
-
-	if (!(loc && loc->inode)) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"%s: Argument not right", loc?loc->path:"(null)");
-		STACK_UNWIND (frame, -1, EINVAL, NULL, NULL, NULL, NULL);
-		return 0;
-	}
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, loc->inode, NULL, NULL, NULL);
-		return 0;
-	}
-        
-        if (inode_ctx_get (loc->inode, this, NULL)
-            && IA_ISDIR (loc->inode->ia_type)) {
-                local->revalidate = 1;
-        }
-
-	if (!inode_ctx_get (loc->inode, this, NULL) && 
-	    loc->inode->ia_type && 
-	    !IA_ISDIR (loc->inode->ia_type)) {
-		uint64_t tmp_list = 0;
-		/* check if revalidate or fresh lookup */
-		inode_ctx_get (loc->inode, this, &tmp_list);
-		local->list = (int16_t *)(long)tmp_list;
-	}
-
-	if (local->list) {
-		list = local->list;
-		for (index = 0; list[index] != -1; index++);
-		if (index != 2) {
-			if (index < 2) {
-				gf_log (this->name, GF_LOG_ERROR,
-					"returning ESTALE for %s: file "
-					"count is %ld", loc->path, index);
-				/* Print where all the file is present */
-				for (index = 0; 
-				     local->list[index] != -1; index++) {
-					gf_log (this->name, GF_LOG_ERROR, 
-						"%s: found on %s", loc->path, 
-						priv->xl_array[list[index]]->name);
-				}
-				unify_local_wipe (local);
-				STACK_UNWIND (frame, -1, ESTALE, 
-					      NULL, NULL, NULL, NULL);
-				return 0;  
-			} else {
-				/* There are more than 2 presences */
-				/* Just log and continue */
-				gf_log (this->name, GF_LOG_ERROR,
-					"%s: file count is %ld", 
-					loc->path, index);
-				/* Print where all the file is present */
-				for (index = 0; 
-				     local->list[index] != -1; index++) {
-					gf_log (this->name, GF_LOG_ERROR, 
-						"%s: found on %s", loc->path, 
-						priv->xl_array[list[index]]->name);
-				}
-			}
-		}
-      
-		/* is revalidate */
-		local->revalidate = 1;
-      
-		for (index = 0; list[index] != -1; index++)
-			local->call_count++;
-      
-		for (index = 0; list[index] != -1; index++) {
-			char need_break = (list[index+1] == -1);
-			STACK_WIND_COOKIE (frame,
-					   unify_lookup_cbk,
-					   (void *)(long)list[index], //cookie
-					   priv->xl_array [list[index]],
-					   priv->xl_array [list[index]]->fops->lookup,
-					   loc,
-					   xattr_req);
-			if (need_break)
-				break;
-		}
-	} else {
-		if (loc->inode->ia_type) {
-			if (inode_ctx_get (loc->inode, this, NULL)) {
-				inode_ctx_get (loc->inode, this, 
-					       &local->inode_generation);
-			}
-		}
-		/* This is first call, there is no list */
-		/* call count should be all child + 1 namespace */
-		local->call_count = priv->child_count + 1;
-      
-		for (index = 0; index <= priv->child_count; index++) {
-			STACK_WIND_COOKIE (frame,
-					   unify_lookup_cbk,
-					   (void *)index, //cookie
-					   priv->xl_array[index],
-					   priv->xl_array[index]->fops->lookup,
-					   loc,
-					   xattr_req);
-		}
-	}
-
-	return 0;
-}
-
-/**
- * unify_stat - if directory, get the stat directly from NameSpace child.
- *     if file, check for a hint and send it only there (also to NS).
- *     if its a fresh stat, then do it on all the nodes.
- *
- * NOTE: for all the call, sending cookie as xlator pointer, which will be 
- *       used in cbk.
- */
-int32_t
-unify_stat (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc)
-{
-	unify_local_t *local = NULL;
-	unify_private_t *priv = this->private;
-	int16_t index = 0;
-	int16_t *list = NULL;
-	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, NULL);
-		return 0;
-	}
-	local->ia_ino = loc->inode->ino;
-	if (IA_ISDIR (loc->inode->ia_type)) {
-		/* Directory */
-		local->call_count = 1;
-		STACK_WIND (frame, unify_buf_cbk, NS(this),
-			    NS(this)->fops->stat, loc);
-	} else {
-		/* File */
-		inode_ctx_get (loc->inode, this, &tmp_list);
-    		list = (int16_t *)(long)tmp_list;
-
-		for (index = 0; list[index] != -1; index++)
-			local->call_count++;
-    
-		for (index = 0; list[index] != -1; index++) {
-			char need_break = (list[index+1] == -1);
-			STACK_WIND (frame,
-				    unify_buf_cbk,
-				    priv->xl_array[list[index]],
-				    priv->xl_array[list[index]]->fops->stat,
-				    loc);
-			if (need_break)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-/**
- * unify_access_cbk -
- */
-int32_t
-unify_access_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-/**
- * unify_access - Send request to only namespace, which has all the 
- *      attributes set for the file.
- */
-int32_t
-unify_access (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      int32_t mask)
-{
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	STACK_WIND (frame,
-		    unify_access_cbk,
-		    NS(this),
-		    NS(this)->fops->access,
-		    loc,
-		    mask);
-
-	return 0;
-}
-
-int32_t
-unify_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)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	inode_t *tmp_inode = NULL;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-  
-		if ((op_ret == -1) && !(priv->optimist && 
-					(op_errno == ENOENT || 
-					 op_errno == EEXIST))) {
-			/* TODO: Decrement the inode_generation of 
-			 * this->inode's parent inode, hence the missing 
-			 * directory is created properly by self-heal. 
-			 * Currently, there is no way to get the parent 
-			 * inode directly.
-			 */
-			gf_log (this->name, GF_LOG_ERROR,
-				"child(%s): path(%s): %s", 
-				priv->xl_array[(long)cookie]->name, 
-				local->loc1.path, strerror (op_errno));
-			if (op_errno != EEXIST)
-				local->failed = 1;
-			local->op_errno = op_errno;
-		}
-  
-		if (op_ret >= 0)
-			local->op_ret = 0;
-
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		if (!local->failed) {
-			inode_ctx_put (local->loc1.inode, this, 
-				       priv->inode_generation);
-		}
-		
-		tmp_inode = local->loc1.inode;
-		unify_local_wipe (local);
-
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      tmp_inode, &local->stbuf,
-                              &local->oldpreparent, &local->oldpostparent);
-	}
-
-	return 0;
-}
-
-/**
- * unify_ns_mkdir_cbk -
- */
-int32_t
-unify_ns_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)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	long index = 0;
-
-	if (op_ret == -1) {
-		/* No need to send mkdir request to other servers, 
-		 * as namespace action failed 
-		 */
-		gf_log (this->name, GF_LOG_ERROR,
-			"namespace: path(%s): %s", 
-			local->name, strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, inode, NULL,
-                              NULL, NULL);
-		return 0;
-	}
-  
-	/* Create one inode for this entry */
-	local->op_ret = 0;
-	local->stbuf = *buf;
-
-        local->oldpreparent = *preparent;
-        local->oldpostparent = *postparent;
-
-	local->call_count = priv->child_count;
-
-	/* Send mkdir request to all the nodes now */
-	for (index = 0; index < priv->child_count; index++) {
-		STACK_WIND_COOKIE (frame,
-				   unify_mkdir_cbk,
-				   (void *)index, //cookie
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->mkdir,
-				   &local->loc1,
-				   local->mode);
-	}
-  
-	return 0;
-}
-
-
-/**
- * unify_mkdir -
- */
-int32_t
-unify_mkdir (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     mode_t mode)
-{
-	unify_local_t *local = NULL;
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	local->mode = mode;
-
-	loc_copy (&local->loc1, loc);
-
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_ns_mkdir_cbk,
-		    NS(this),
-		    NS(this)->fops->mkdir,
-		    loc,
-		    mode);
-	return 0;
-}
-
-/**
- * unify_rmdir_cbk -
- */
-int32_t
-unify_rmdir_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-                 struct iatt *preparent,
-                 struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret == 0 || (priv->optimist && (op_errno == ENOENT)))
-			local->op_ret = 0;
-		if (op_ret == -1)
-			local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                              &local->oldpreparent, &local->oldpostparent);
-	}
-
-	return 0;
-}
-
-/**
- * unify_ns_rmdir_cbk -
- */
-int32_t
-unify_ns_rmdir_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-                    struct iatt *preparent,
-                    struct iatt *postparent)
-{
-	int16_t index = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-  
-	if (op_ret == -1) {
-		/* No need to send rmdir request to other servers, 
-		 * as namespace action failed 
-		 */
-		gf_log (this->name, 
-			((op_errno != ENOTEMPTY) ? 
-			 GF_LOG_ERROR : GF_LOG_DEBUG),
-			"namespace: path(%s): %s", 
-			local->loc1.path, strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, NULL, NULL);
-		return 0;
-	}
-
-	local->call_count = priv->child_count;
-
-        local->oldpreparent = *preparent;
-        local->oldpostparent = *postparent;
-
-	for (index = 0; index < priv->child_count; index++) {
-		STACK_WIND (frame,
-			    unify_rmdir_cbk,
-			    priv->xl_array[index],
-			    priv->xl_array[index]->fops->rmdir,
-			    &local->loc1);
-	}
-
-	return 0;
-}
-
-/**
- * unify_rmdir -
- */
-int32_t
-unify_rmdir (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc)
-{
-	unify_local_t *local = NULL;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-
-	loc_copy (&local->loc1, loc);
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_ns_rmdir_cbk,
-		    NS(this),
-		    NS(this)->fops->rmdir,
-		    loc);
-
-	return 0;
-}
-
-/**
- * unify_open_cbk -
- */
-int32_t
-unify_open_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		fd_t *fd)
-{
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret >= 0) {
-			local->op_ret = op_ret;
-			if (NS(this) != (xlator_t *)cookie) {
-				/* Store child node's ptr, used in 
-				   all the f*** / FileIO calls */
-				fd_ctx_set (fd, this, (uint64_t)(long)cookie);
-			}
-		}
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			local->failed = 1;
-		}
-		callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		if ((local->failed == 1) && (local->op_ret >= 0)) {
-			local->call_count = 1;
-			/* return -1 to user */
-			local->op_ret = -1;
-			//local->op_errno = EIO; 
-      
-			if (!fd_ctx_get (local->fd, this, NULL)) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"Open success on child node, "
-					"failed on namespace");
-			} else {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"Open success on namespace, "
-					"failed on child node");
-			}
-		}
-
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, 
-			      local->op_errno, local->fd);
-	}
-
-	return 0;
-}
-
-#ifdef GF_DARWIN_HOST_OS
-/**
- * unify_create_lookup_cbk - 
- */
-int32_t 
-unify_open_lookup_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       inode_t *inode,
-		       struct iatt *buf,
-		       dict_t *dict,
-		       struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	int16_t index = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if ((op_ret == -1) && (op_errno != ENOENT)) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"child(%s): path(%s): %s", 
-				priv->xl_array[(long)cookie]->name, 
-				local->loc1.path, strerror (op_errno));
-			local->op_errno = op_errno;
-		}
-    
-		if (op_ret >= 0) {
-			local->op_ret = op_ret; 
-			local->index++;
-			if (NS(this) == priv->xl_array[(long)cookie]) {
-				local->list[0] = (int16_t)(long)cookie;
-			} else {
-				local->list[1] = (int16_t)(long)cookie;
-			}
-			if (IA_ISDIR (buf->ia_type))
-				local->failed = 1;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		int16_t file_list[3] = {0,};
-		local->op_ret = -1;
-
-		file_list[0] = local->list[0];
-		file_list[1] = local->list[1];
-		file_list[2] = -1;
-
-		if (local->index != 2) {
-			/* Lookup failed, can't do open */
-			gf_log (this->name, GF_LOG_ERROR,
-				"%s: present on %d nodes", 
-				local->name, local->index);
-
-			if (local->index < 2) {
-				unify_local_wipe (local);
-				gf_log (this->name, GF_LOG_ERROR,
-					"returning as file found on less "
-					"than 2 nodes");
-				STACK_UNWIND (frame, local->op_ret, 
-					      local->op_errno, local->fd);
-				return 0;
-			}
-		}
-
-		if (local->failed) {
-			/* Open on directory, return EISDIR */
-			unify_local_wipe (local);
-			STACK_UNWIND (frame, -1, EISDIR, local->fd);
-			return 0;
-		}
-
-		/* Everything is perfect :) */    
-		local->call_count = 2;
-    
-		for (index = 0; file_list[index] != -1; index++) {
-			char need_break = (file_list[index+1] == -1);
-			STACK_WIND_COOKIE (frame,
-					   unify_open_cbk,
-					   priv->xl_array[file_list[index]],
-					   priv->xl_array[file_list[index]],
-					   priv->xl_array[file_list[index]]->fops->open,
-					   &local->loc1,
-					   local->flags,
-					   local->fd, local->wbflags);
-			if (need_break)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-
-int32_t
-unify_open_readlink_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-			 const char *path,
-			 struct iatt *sbuf)
-{
-	int16_t index = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1) {
-		STACK_UNWIND (frame, -1, ENOENT);
-		return 0;
-	}
-
-	if (path[0] == '/') {
-		local->name = gf_strdup (path);
-		ERR_ABORT (local->name);
-	} else {
-		char *tmp_str = gf_strdup (local->loc1.path);
-		char *tmp_base = dirname (tmp_str);
-		local->name = GF_CALLOC (1, ZR_PATH_MAX, gf_unify_mt_char);
-		strcpy (local->name, tmp_base);
-		strncat (local->name, "/", 1);
-		strcat (local->name, path);
-		GF_FREE (tmp_str);
-	}
-  
-	local->list = GF_CALLOC (1, sizeof (int16_t) * 3,
-                                 gf_unify_mt_int16_t);
-	ERR_ABORT (local->list);
-	local->call_count = priv->child_count + 1;
-	local->op_ret = -1;
-	for (index = 0; index <= priv->child_count; index++) {
-		/* Send the lookup to all the nodes including namespace */
-		STACK_WIND_COOKIE (frame,
-				   unify_open_lookup_cbk,
-				   (void *)(long)index,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->lookup,
-				   &local->loc1,
-				   NULL);
-	}
-
-	return 0;
-}
-#endif /* GF_DARWIN_HOST_OS */
-
-/**
- * unify_open - 
- */
-int32_t
-unify_open (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    int32_t flags,
-	    fd_t *fd,
-            int32_t wbflags)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = NULL;
-	int16_t *list = NULL;
-	int16_t index = 0;
-	int16_t file_list[3] = {0,};
-	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Init */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-	local->fd    = fd;
-	local->flags = flags;
-        local->wbflags = wbflags;
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	local->list = list;
-	file_list[0] = priv->child_count; /* Thats namespace */
-	file_list[2] = -1;
-	for (index = 0; list[index] != -1; index++) {
-		local->call_count++;
-		if (list[index] != priv->child_count)
-			file_list[1] = list[index];
-	}
-
-	if (local->call_count != 2) {
-		/* If the lookup was done for file */
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: entry_count is %d",
-			loc->path, local->call_count);
-		for (index = 0; local->list[index] != -1; index++)
-			gf_log (this->name, GF_LOG_ERROR, "%s: found on %s",
-				loc->path, priv->xl_array[list[index]]->name);
-
-		if (local->call_count < 2) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"returning EIO as file found on onlyone node");
-			STACK_UNWIND (frame, -1, EIO, fd);
-			return 0;
-		}
-	}
-
-#ifdef GF_DARWIN_HOST_OS
-	/* Handle symlink here */
-	if (IA_ISLNK (loc->inode->ia_type)) {
-		/* Callcount doesn't matter here */
-		STACK_WIND (frame,
-			    unify_open_readlink_cbk,
-			    NS(this),
-			    NS(this)->fops->readlink,
-			    loc, ZR_PATH_MAX);
-		return 0;
-	}
-#endif /* GF_DARWIN_HOST_OS */
-
-	local->call_count = 2;
-	for (index = 0; file_list[index] != -1; index++) {
-		char need_break = (file_list[index+1] == -1);
-		STACK_WIND_COOKIE (frame,
-				   unify_open_cbk,
-				   priv->xl_array[file_list[index]], //cookie
-				   priv->xl_array[file_list[index]],
-				   priv->xl_array[file_list[index]]->fops->open,
-				   loc,
-				   flags,
-				   fd, wbflags);
-		if (need_break)
-			break;
-	}
-
-	return 0;
-}
-
-
-int32_t 
-unify_create_unlink_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-                         struct iatt *preparent,
-                         struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-	inode_t *inode = local->loc1.inode;
-
-	unify_local_wipe (local);
-
-	STACK_UNWIND (frame, local->op_ret, local->op_errno, local->fd, 
-		      inode, &local->stbuf,
-                      &local->oldpreparent, &local->oldpostparent);
-  
-	return 0;
-}
-
-/**
- * unify_create_open_cbk -
- */
-int32_t
-unify_create_open_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       fd_t *fd)
-{
-	int ret = 0;
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-	inode_t *inode = NULL;
-	xlator_t *child = NULL;
-	uint64_t tmp_value = 0;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret >= 0) {
-			local->op_ret = op_ret;
-			if (NS(this) != (xlator_t *)cookie) {
-				/* Store child node's ptr, used in all 
-				   the f*** / FileIO calls */
-				/* TODO: log on failure */
-				ret = fd_ctx_get (fd, this, &tmp_value);
-				cookie = (void *)(long)tmp_value;
-			} else {
-				/* NOTE: open successful on namespace.
-				 *       fd's ctx can be used to identify open 
-				 *       failure on storage subvolume. cool 
-				 *       ide ;) */
-				local->failed = 0;
-			}
-		} else {
-			gf_log (this->name, GF_LOG_ERROR,
-				"child(%s): path(%s): %s", 
-				((xlator_t *)cookie)->name,
-				local->loc1.path, strerror (op_errno));
-			local->op_errno = op_errno;
-			local->failed = 1;
-		}
-		callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		if (local->failed == 1 && (local->op_ret >= 0)) {
-			local->call_count = 1;
-			/* return -1 to user */
-			local->op_ret = -1;
-			local->op_errno = EIO;
-			local->fd = fd;
-			local->call_count = 1;
-
-			if (!fd_ctx_get (local->fd, this, &tmp_value)) {
-				child = (xlator_t *)(long)tmp_value;
-
-				gf_log (this->name, GF_LOG_ERROR, 
-					"Create success on child node, "
-					"failed on namespace");
-
-				STACK_WIND (frame,
-					    unify_create_unlink_cbk,
-					    child,
-					    child->fops->unlink,
-					    &local->loc1);
-			} else {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"Create success on namespace, "
-					"failed on child node");
-
-				STACK_WIND (frame,
-					    unify_create_unlink_cbk,
-					    NS(this),
-					    NS(this)->fops->unlink,
-					    &local->loc1);
-			}
-			return 0;
-		}
-		inode = local->loc1.inode;
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, fd,
-			      inode, &local->stbuf,
-                              &local->oldpreparent, &local->oldpostparent);
-	}
-	return 0;
-}
-
-/**
- * unify_create_lookup_cbk - 
- */
-int32_t 
-unify_create_lookup_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-			 inode_t *inode,
-			 struct iatt *buf,
-			 dict_t *dict,
-                         struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	int16_t index = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"child(%s): path(%s): %s", 
-				priv->xl_array[(long)cookie]->name, 
-				local->loc1.path, strerror (op_errno));
-			local->op_errno = op_errno;
-			local->failed = 1;
-		}
-
-		if (op_ret >= 0) {
-			local->op_ret = op_ret; 
-			local->list[local->index++] = (int16_t)(long)cookie;
-			if (NS(this) == priv->xl_array[(long)cookie]) {
-				local->ia_ino = buf->ia_ino;
-			} else {
-				local->stbuf = *buf;
-			}
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		int16_t *list = local->list;
-		int16_t file_list[3] = {0,};
-		local->op_ret = -1;
-
-		local->list [local->index] = -1;
-		file_list[0] = list[0];
-		file_list[1] = list[1];
-		file_list[2] = -1;
-
-		local->stbuf.ia_ino = local->ia_ino;
-		/* TODO: log on failure */
-		inode_ctx_put (local->loc1.inode, this, 
-			       (uint64_t)(long)local->list);
-
-		if (local->index != 2) {
-			/* Lookup failed, can't do open */
-			gf_log (this->name, GF_LOG_ERROR,
-				"%s: present on %d nodes", 
-				local->loc1.path, local->index);
-			file_list[0] = priv->child_count;
-			for (index = 0; list[index] != -1; index++) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"%s: found on %s", local->loc1.path, 
-					priv->xl_array[list[index]]->name);
-				if (list[index] != priv->child_count)
-					file_list[1] = list[index];
-			}
-
-			if (local->index < 2) {
-				unify_local_wipe (local);
-				gf_log (this->name, GF_LOG_ERROR,
-					"returning EIO as file found on "
-					"only one node");
-				STACK_UNWIND (frame, -1, EIO, 
-					      local->fd, inode, NULL,
-                                              NULL, NULL);
-				return 0;
-			}
-		}
-		/* Everything is perfect :) */    
-		local->call_count = 2;
-    
-		for (index = 0; file_list[index] != -1; index++) {
-			char need_break = (file_list[index+1] == -1);
-			STACK_WIND_COOKIE (frame,
-					   unify_create_open_cbk,
-					   priv->xl_array[file_list[index]],
-					   priv->xl_array[file_list[index]],
-					   priv->xl_array[file_list[index]]->fops->open,
-					   &local->loc1,
-					   local->flags,
-					   local->fd, 0);
-			if (need_break)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_create_cbk -
- */
-int32_t
-unify_create_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  fd_t *fd,
-		  inode_t *inode,
-		  struct iatt *buf,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-	int ret = 0;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-	inode_t *tmp_inode = NULL;
-
-	if (op_ret == -1) {
-		/* send unlink () on Namespace */
-		local->op_errno = op_errno;
-		local->op_ret = -1;
-		local->call_count = 1;
-		gf_log (this->name, GF_LOG_ERROR,
-			"create failed on %s (file %s, error %s), "
-			"sending unlink to namespace", 
-			prev_frame->this->name, 
-			local->loc1.path, strerror (op_errno));
-
-		STACK_WIND (frame,
-			    unify_create_unlink_cbk,
-			    NS(this),
-			    NS(this)->fops->unlink,
-			    &local->loc1);
-
-		return 0;
-	}
-
-	if (op_ret >= 0) {
-		local->op_ret = op_ret;
-		local->stbuf = *buf;
-		/* Just inode number should be from NS node */
-		local->stbuf.ia_ino = local->ia_ino;
-
-		/* TODO: log on failure */
-		ret = fd_ctx_set (fd, this, (uint64_t)(long)prev_frame->this);
-	}
-  
-	tmp_inode = local->loc1.inode;
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, local->op_ret, local->op_errno, local->fd, 
-		      tmp_inode, &local->stbuf,
-                      &local->oldpreparent, &local->oldpostparent);
-
-	return 0;
-}
-
-/**
- * unify_ns_create_cbk -
- * 
- */
-int32_t
-unify_ns_create_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     fd_t *fd,
-		     inode_t *inode,
-		     struct iatt *buf,
-                     struct iatt *preparent,
-                     struct iatt *postparent)
-{
-	struct sched_ops *sched_ops = NULL;
-	xlator_t *sched_xl = NULL;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t *list = NULL;
-	int16_t index = 0;
-
-	if (op_ret == -1) {
-		/* No need to send create request to other servers, as 
-		   namespace action failed. Handle exclusive create here. */
-		if ((op_errno != EEXIST) || 
-		    ((op_errno == EEXIST) && 
-		     ((local->flags & O_EXCL) == O_EXCL))) {
-			/* If its just a create call without O_EXCL, 
-			   don't do this */
-			gf_log (this->name, GF_LOG_ERROR,
-				"namespace: path(%s): %s", 
-				local->loc1.path, strerror (op_errno));
-			unify_local_wipe (local);
-			STACK_UNWIND (frame, op_ret, op_errno, fd, inode, buf,
-                                      preparent, postparent);
-			return 0;
-		}
-	}
-  
-	if (op_ret >= 0) {
-		/* Get the inode number from the NS node */
-		local->ia_ino = buf->ia_ino;
-  
-                local->oldpreparent = *preparent;
-                local->oldpostparent = *postparent;
-
-		local->op_ret = -1;
-
-		/* Start the mapping list */
-		list = GF_CALLOC (1, sizeof (int16_t) * 3,
-                                  gf_unify_mt_int16_t);
-		ERR_ABORT (list);
-		inode_ctx_put (inode, this, (uint64_t)(long)list);
-		list[0] = priv->child_count;
-		list[2] = -1;
-
-		/* This means, file doesn't exist anywhere in the Filesystem */
-		sched_ops = priv->sched_ops;
-
-		/* Send create request to the scheduled node now */
-		sched_xl = sched_ops->schedule (this, local->loc1.path);
-		if (sched_xl == NULL)
-		{
-			/* send unlink () on Namespace */
-			local->op_errno = ENOTCONN;
-			local->op_ret = -1;
-			local->call_count = 1;
-			gf_log (this->name, GF_LOG_ERROR,
-				"no node online to schedule create:(file %s) "
-				"sending unlink to namespace", 
-				(local->loc1.path)?local->loc1.path:"");
-
-			STACK_WIND (frame,
-				    unify_create_unlink_cbk,
-				    NS(this),
-				    NS(this)->fops->unlink,
-				    &local->loc1);
-	
-			return 0;
-		}
-
-		for (index = 0; index < priv->child_count; index++)
-			if (sched_xl == priv->xl_array[index])
-				break;
-		list[1] = index;
-
-		STACK_WIND (frame, unify_create_cbk,
-			    sched_xl, sched_xl->fops->create,
-			    &local->loc1, local->flags, local->mode, fd);
-	} else {
-		/* File already exists, and there is no O_EXCL flag */
-
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"File(%s) already exists on namespace, sending "
-			"open instead", local->loc1.path);
-
-		local->list = GF_CALLOC (1, sizeof (int16_t) * 3,
-                                         gf_unify_mt_int16_t);
-		ERR_ABORT (local->list);
-		local->call_count = priv->child_count + 1;
-		local->op_ret = -1;
-		for (index = 0; index <= priv->child_count; index++) {
-			/* Send lookup() to all nodes including namespace */
-			STACK_WIND_COOKIE (frame,
-					   unify_create_lookup_cbk,
-					   (void *)(long)index,
-					   priv->xl_array[index],
-					   priv->xl_array[index]->fops->lookup,
-					   &local->loc1,
-					   NULL);
-		}
-	}
-	return 0;
-}
-
-/**
- * unify_create - create a file in global namespace first, so other 
- *    clients can see them. Create the file in storage nodes in background.
- */
-int32_t
-unify_create (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      int32_t flags,
-	      mode_t mode,
-	      fd_t *fd)
-{
-	unify_local_t *local = NULL;
-  
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	local->mode = mode;
-	local->flags = flags;
-	local->fd = fd;
-
-	loc_copy (&local->loc1, loc);
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, fd, loc->inode, NULL,
-                              NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_ns_create_cbk,
-		    NS(this),
-		    NS(this)->fops->create,
-		    loc,
-		    flags | O_EXCL,
-		    mode,
-		    fd);
-  
-	return 0;
-}
-
-
-/**
- * unify_opendir_cbk - 
- */
-int32_t
-unify_opendir_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   fd_t *fd)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fd);
-
-	return 0;
-}
-
-/** 
- * unify_opendir -
- */
-int32_t
-unify_opendir (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *loc,
-	       fd_t *fd)
-{
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	STACK_WIND (frame, unify_opendir_cbk,
-		    NS(this), NS(this)->fops->opendir, loc, fd);
-
-	return 0;
-}
-
-
-int32_t
-unify_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)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"%s(): child(%s): path(%s): %s", 
-				gf_fop_list[frame->root->op],
-				prev_frame->this->name, 
-				(local->loc1.path)?local->loc1.path:"", 
-				strerror (op_errno));
-
-			local->op_errno = op_errno;
-			if ((op_errno == ENOENT) && priv->optimist) 
-				local->op_ret = 0;
-		}
-
-		if (op_ret >= 0) {
-			local->op_ret = 0;
-
-			if (NS (this) == prev_frame->this) {
-				local->ia_ino = statpost->ia_ino;
-				/* If the entry is directory, get the stat
-				   from NS node */
-				if (IA_ISDIR (statpost->ia_type) || 
-				    !local->stpost.ia_blksize) {
-					local->stpre = *statpre;
-                                        local->stpost = *statpost;
-				}
-			}
-
-			if ((!IA_ISDIR (statpost->ia_type)) && 
-			    (NS (this) != prev_frame->this)) {
-				/* If file, take the stat info from Storage 
-				   node. */
-				local->stpre = *statpre;
-                                local->stpost = *statpost;
-			}
-		}
-	}
-	UNLOCK (&frame->lock);
-    
-	if (!callcnt) {
-		/* If the inode number is not filled, operation should
-		   fail */
-		if (!local->ia_ino)
-			local->op_ret = -1;
-
-                local->stpre.ia_ino = local->ia_ino;
-		local->stpost.ia_ino = local->ia_ino;
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      &local->stpre, &local->stpost);
-	}
-
-	return 0;
-}
-
-
-int32_t
-unify_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               struct iatt *stbuf, int32_t valid)
-{
-	unify_local_t *local = NULL;
-	unify_private_t *priv = this->private;
-	int32_t index = 0;
-	int32_t callcnt = 0;
-  	uint64_t tmp_list = 0;
-
-        if (!(loc && loc->inode)) {                     
-                STACK_UNWIND (frame, -1, EINVAL, NULL, NULL);
-                return 0;
-        }                                                    
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-
-	if (IA_ISDIR (loc->inode->ia_type)) {
-		local->call_count = 1;
-      
-                STACK_WIND (frame,
-                            unify_setattr_cbk,
-                            NS (this),
-                            NS (this)->fops->setattr,
-                            loc, stbuf, valid);
-        } else {
-		inode_ctx_get (loc->inode, this, &tmp_list);
-		local->list = (int16_t *)(long)tmp_list;
-
-		for (index = 0; local->list[index] != -1; index++) {
-			local->call_count++;
-			callcnt++;
-		}
-      
-		for (index = 0; local->list[index] != -1; index++) {
-			STACK_WIND (frame,
-				    unify_setattr_cbk,
-				    priv->xl_array[local->list[index]],
-				    priv->xl_array[local->list[index]]->fops->setattr,
-				    loc, stbuf, valid);
-
-			if (!--callcnt)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-
-int32_t
-unify_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                struct iatt *stbuf, int32_t valid)
-{
-	unify_local_t *local = NULL;
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR(fd);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-
-	if (!fd_ctx_get (fd, this, &tmp_child)) {
-		/* If its set, then its file */
-		child = (xlator_t *)(long)tmp_child;		     
-
-		local->call_count = 2;
-
-		STACK_WIND (frame, unify_setattr_cbk, child,
-			    child->fops->fsetattr, fd, stbuf, valid);
-
-		STACK_WIND (frame, unify_setattr_cbk, NS(this),
-			    NS(this)->fops->fsetattr, fd, stbuf, valid);
-	} else {
-		local->call_count = 1;
-    
-		STACK_WIND (frame, unify_setattr_cbk,
-			    NS(this), NS(this)->fops->fsetattr,
-			    fd, stbuf, valid);
-	}
-  
-	return 0;
-}
-
-
-/**
- * unify_truncate_cbk - 
- */
-int32_t
-unify_truncate_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    struct iatt *prebuf,
-                    struct iatt *postbuf)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"child(%s): path(%s): %s", 
-				prev_frame->this->name, 
-				(local->loc1.path)?local->loc1.path:"", 
-				strerror (op_errno));
-			local->op_errno = op_errno;
-			if (!((op_errno == ENOENT) && priv->optimist))
-				local->op_ret = -1;
-		}
-
-		if (op_ret >= 0) {
-			if (NS (this) == prev_frame->this) {
-				local->ia_ino = postbuf->ia_ino;
-				/* If the entry is directory, get the 
-				   stat from NS node */
-				if (IA_ISDIR (postbuf->ia_type) ||
-				    !local->stbuf.ia_blksize) {
-					local->stbuf = *prebuf;
-                                        local->poststbuf = *postbuf;
-				}
-			}
-
-			if ((!IA_ISDIR (postbuf->ia_type)) &&
-			    (NS (this) != prev_frame->this)) {
-				/* If file, take the stat info from 
-				   Storage node. */
-				local->stbuf = *prebuf;
-                                local->poststbuf = *postbuf;
-			}
-		}
-	}
-	UNLOCK (&frame->lock);
-    
-	if (!callcnt) {
-		if (local->ia_ino) {
-			local->stbuf.ia_ino = local->ia_ino;
-                        local->poststbuf.ia_ino = local->ia_ino;
-                } else {
-			local->op_ret = -1;
-                }
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      &local->stbuf, &local->poststbuf);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_truncate - 
- */
-int32_t
-unify_truncate (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		off_t offset)
-{
-	unify_local_t *local = NULL;
-	unify_private_t *priv = this->private;
-	int32_t index = 0;
-	int32_t callcnt = 0;
-  	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-	local->ia_ino = loc->inode->ino;
-
-	if (IA_ISDIR (loc->inode->ia_type)) {
-		local->call_count = 1;
-      
-		STACK_WIND (frame,
-			    unify_truncate_cbk,
-			    NS(this),
-			    NS(this)->fops->truncate,
-			    loc,
-                            0);
-	} else {
-		local->op_ret = 0;
-		inode_ctx_get (loc->inode, this, &tmp_list);
-		local->list = (int16_t *)(long)tmp_list;
-
-		for (index = 0; local->list[index] != -1; index++) {
-			local->call_count++;
-			callcnt++;
-		}
-      
-		/* Don't send offset to NS truncate */
-		STACK_WIND (frame, unify_truncate_cbk, NS(this),
-			    NS(this)->fops->truncate, loc, 0);
-		callcnt--;
-
-		for (index = 0; local->list[index] != -1; index++) {
-			if (NS(this) != priv->xl_array[local->list[index]]) {
-				STACK_WIND (frame,
-					    unify_truncate_cbk,
-					    priv->xl_array[local->list[index]],
-					    priv->xl_array[local->list[index]]->fops->truncate,
-					    loc,
-					    offset);
-				if (!--callcnt)
-					break;
-			}
-		}
-	}
-
-	return 0;
-}
-
-/**
- * unify_readlink_cbk - 
- */
-int32_t
-unify_readlink_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    const char *path,
-                    struct iatt *sbuf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, path, sbuf);
-	return 0;
-}
-
-/**
- * unify_readlink - Read the link only from the storage node.
- */
-int32_t
-unify_readlink (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		size_t size)
-{
-	unify_private_t *priv = this->private;
-	int32_t entry_count = 0;
-	int16_t *list = NULL;
-	int16_t index = 0;
-  	uint64_t tmp_list = 0;
-  
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-	
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	for (index = 0; list[index] != -1; index++)
-		entry_count++;
-
-	if (entry_count >= 2) {
-		for (index = 0; list[index] != -1; index++) {
-			if (priv->xl_array[list[index]] != NS(this)) {
-				STACK_WIND (frame,
-					    unify_readlink_cbk,
-					    priv->xl_array[list[index]],
-					    priv->xl_array[list[index]]->fops->readlink,
-					    loc,
-					    size);
-				break;
-			}
-		}
-	} else {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"returning ENOENT, no softlink files found "
-			"on storage node");
-		STACK_UNWIND (frame, -1, ENOENT, NULL);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_unlink_cbk - 
- */
-int32_t
-unify_unlink_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret == 0  || ((op_errno == ENOENT) && priv->optimist))
-			local->op_ret = 0;
-		if (op_ret == -1)
-			local->op_errno = op_errno;
-
-                if (((call_frame_t *)cookie)->this == NS(this)) {
-                        local->oldpreparent = *preparent;
-                        local->oldpostparent = *postparent;
-                }
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                              &local->oldpreparent, &local->oldpostparent);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_unlink - 
- */
-int32_t
-unify_unlink (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = NULL;
-	int16_t *list = NULL;
-	int16_t index = 0;
-  	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	for (index = 0; list[index] != -1; index++)
-		local->call_count++;
-
-	if (local->call_count) {
-		for (index = 0; list[index] != -1; index++) {
-			char need_break = (list[index+1] == -1);
-			STACK_WIND (frame,
-				    unify_unlink_cbk,
-				    priv->xl_array[list[index]],
-				    priv->xl_array[list[index]]->fops->unlink,
-				    loc);
-			if (need_break)
-				break;
-		}
-	} else {
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: returning ENOENT", loc->path);
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_readv_cbk - 
- */
-int32_t
-unify_readv_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct iovec *vector,
-		 int32_t count,
-		 struct iatt *stbuf,
-                 struct iobref *iobref)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, vector, count, stbuf, iobref);
-	return 0;
-}
-
-/**
- * unify_readv - 
- */
-int32_t
-unify_readv (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     size_t size,
-	     off_t offset)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame,
-		    unify_readv_cbk,
-		    child,
-		    child->fops->readv,
-		    fd,
-		    size,
-		    offset);
-
-
-	return 0;
-}
-
-/**
- * unify_writev_cbk - 
- */
-int32_t
-unify_writev_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-                  struct iatt *prebuf,
-		  struct iatt *postbuf)
-{
-	unify_local_t *local = NULL;
-
-        local = frame->local;
-
-        local->stbuf = *prebuf;
-        local->stbuf.ia_ino = local->ia_ino;
-
-        local->poststbuf = *postbuf;
-        local->poststbuf.ia_ino = local->ia_ino;
-
-	STACK_UNWIND (frame, op_ret, op_errno,
-                      &local->stbuf, &local->poststbuf);
-	return 0;
-}
-
-/**
- * unify_writev - 
- */
-int32_t
-unify_writev (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      struct iovec *vector,
-	      int32_t count,
-	      off_t off,
-              struct iobref *iobref)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-	unify_local_t *local = NULL;
-
-	INIT_LOCAL (frame, local);
-        local->ia_ino = fd->inode->ino;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame,
-		    unify_writev_cbk,
-		    child,
-		    child->fops->writev,
-		    fd,
-		    vector,
-		    count,
-		    off,
-                    iobref);
-
-	return 0;
-}
-
-/**
- * unify_ftruncate -
- */
-int32_t
-unify_ftruncate (call_frame_t *frame,
-		 xlator_t *this,
-		 fd_t *fd,
-		 off_t offset)
-{
-	xlator_t *child = NULL;
-	unify_local_t *local = NULL;
-	uint64_t tmp_child = 0;
-
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR(fd);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	local->op_ret = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	local->call_count = 2;
-  
-	STACK_WIND (frame, unify_truncate_cbk,
-		    child, child->fops->ftruncate,
-		    fd, offset);
-  
-	STACK_WIND (frame, unify_truncate_cbk,
-		    NS(this), NS(this)->fops->ftruncate,
-		    fd, 0);
-  
-	return 0;
-}
-
-
-/**
- * unify_flush_cbk - 
- */
-int32_t
-unify_flush_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-/**
- * unify_flush -
- */
-int32_t
-unify_flush (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_flush_cbk, child, 
-		    child->fops->flush, fd);
-
-	return 0;
-}
-
-
-/**
- * unify_fsync_cbk - 
- */
-int32_t
-unify_fsync_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-                 struct iatt *prebuf,
-                 struct iatt *postbuf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-/**
- * unify_fsync - 
- */
-int32_t
-unify_fsync (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     int32_t flags)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_fsync_cbk, child,
-		    child->fops->fsync, fd, flags);
-
-	return 0;
-}
-
-/**
- * unify_fstat - Send fstat FOP to Namespace only if its directory, and to 
- *     both namespace and the storage node if its a file.
- */
-int32_t
-unify_fstat (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd)
-{
-	unify_local_t *local = NULL;
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR(fd);
-
-	INIT_LOCAL (frame, local);
-	local->ia_ino = fd->inode->ino;
-
-	if (!fd_ctx_get (fd, this, &tmp_child)) {
-		/* If its set, then its file */
-		child = (xlator_t *)(long)tmp_child;		     
-		local->call_count = 2;
-
-		STACK_WIND (frame, unify_buf_cbk, child,
-			    child->fops->fstat, fd);
-
-		STACK_WIND (frame, unify_buf_cbk, NS(this),
-			    NS(this)->fops->fstat, fd);
-
-	} else {
-		/* this is an directory */
-		local->call_count = 1;
-		STACK_WIND (frame, unify_buf_cbk, NS(this),
-			    NS(this)->fops->fstat, fd);
-	}
-
-	return 0;
-}
-
-/**
- * unify_getdents_cbk - 
- */
-int32_t
-unify_getdents_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    dir_entry_t *entry,
-		    int32_t count)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, entry, count);
-	return 0;
-}
-
-/**
- * unify_getdents - send the FOP request to all the nodes.
- */
-int32_t
-unify_getdents (call_frame_t *frame,
-		xlator_t *this,
-		fd_t *fd,
-		size_t size,
-		off_t offset,
-		int32_t flag)
-{
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR (fd);
-
-	STACK_WIND (frame, unify_getdents_cbk, NS(this),
-		    NS(this)->fops->getdents, fd, size, offset, flag);
-
-	return 0;
-}
-
-
-/**
- * unify_readdir_cbk - 
- */
-int32_t
-unify_readdir_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   gf_dirent_t *buf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-
-	return 0;
-}
-
-/**
- * unify_readdir - send the FOP request to all the nodes.
- */
-int32_t
-unify_readdir (call_frame_t *frame,
-	       xlator_t *this,
-	       fd_t *fd,
-	       size_t size,
-	       off_t offset)
-{
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR (fd);
-
-	STACK_WIND (frame, unify_readdir_cbk, NS(this),
-		    NS(this)->fops->readdir, fd, size, offset);
-
-	return 0;
-}
-
-
-int32_t
-unify_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, gf_dirent_t *buf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-
-	return 0;
-}
-
-
-int32_t
-unify_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                off_t offset)
-{
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR (fd);
-
-	STACK_WIND (frame, unify_readdirp_cbk, NS(this),
-		    NS(this)->fops->readdirp, fd, size, offset);
-
-	return 0;
-}
-
-
-/**
- * unify_fsyncdir_cbk - 
- */
-int32_t
-unify_fsyncdir_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-
-	return 0;
-}
-
-/**
- * unify_fsyncdir -
- */
-int32_t
-unify_fsyncdir (call_frame_t *frame,
-		xlator_t *this,
-		fd_t *fd,
-		int32_t flags)
-{
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR (fd);
-
-	STACK_WIND (frame, unify_fsyncdir_cbk,
-		    NS(this), NS(this)->fops->fsyncdir, fd, flags);
-
-	return 0;
-}
-
-/**
- * unify_lk_cbk - UNWIND frame with the proper return arguments.
- */
-int32_t
-unify_lk_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      struct gf_flock *lock)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, lock);
-	return 0;
-}
-
-/**
- * unify_lk - Send it to all the storage nodes, (should be 1) which has file.
- */
-int32_t
-unify_lk (call_frame_t *frame,
-	  xlator_t *this,
-	  fd_t *fd,
-	  int32_t cmd,
-	  struct gf_flock *lock)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_lk_cbk, child,
-		    child->fops->lk, fd, cmd, lock);
-
-	return 0;
-}
-
-
-int32_t
-unify_setxattr_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno);
-
-static int32_t
-unify_setxattr_file_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno)
-{
-	unify_private_t *private = this->private;
-	unify_local_t *local = frame->local;
-	xlator_t *sched_xl = NULL;
-	struct sched_ops *sched_ops = NULL;
-
-	if (op_ret == -1) {
-		if (!ENOTSUP)
-			gf_log (this->name, GF_LOG_ERROR,
-				"setxattr with XATTR_CREATE on ns: "
-				"path(%s) key(%s): %s",
-				local->loc1.path, local->name, 
-				strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno);
-		return 0;
-	} 
-
-	LOCK (&frame->lock);
-	{
-		local->failed = 0;
-		local->op_ret = 0;
-		local->op_errno = 0;
-		local->call_count = 1;
-	}
-	UNLOCK (&frame->lock);
-
-	/* schedule XATTR_CREATE on one of the child node */
-	sched_ops = private->sched_ops;
-    
-	/* Send create request to the scheduled node now */
-	sched_xl = sched_ops->schedule (this, local->name); 
-	if (!sched_xl) {
-		STACK_UNWIND (frame, -1, ENOTCONN);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_setxattr_cbk,
-		    sched_xl,
-		    sched_xl->fops->setxattr,
-		    &local->loc1,
-		    local->dict,
-		    local->flags);
-	return 0;
-}
-
-/**
- * unify_setxattr_cbk - When all the child nodes return, UNWIND frame.
- */
-int32_t
-unify_setxattr_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno)
-{
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-	dict_t *dict = NULL;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret == -1) {
-			gf_log (this->name, (((op_errno == ENOENT) || 
-					      (op_errno == ENOTSUP))? 
-					     GF_LOG_DEBUG : GF_LOG_ERROR), 
-				"child(%s): path(%s): %s", 
-				prev_frame->this->name, 
-				(local->loc1.path)?local->loc1.path:"", 
-				strerror (op_errno));
-			if (local->failed == -1) {
-				local->failed = 1;
-			}
-			local->op_errno = op_errno;
-		} else {
-			local->failed = 0;
-			local->op_ret = op_ret;
-		}
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		if (local->failed && local->name && 
-		    ZR_FILE_CONTENT_REQUEST(local->name)) {      
-			dict = get_new_dict ();
-			dict_set (dict, local->dict->members_list->key, 
-				  data_from_dynptr(NULL, 0));
-			dict_ref (dict);
-
-			local->call_count = 1;
-
-			STACK_WIND (frame,
-				    unify_setxattr_file_cbk,
-				    NS(this),
-				    NS(this)->fops->setxattr,
-				    &local->loc1,
-				    dict,
-				    XATTR_CREATE);
-
-			dict_unref (dict);
-			return 0;
-		}
-    
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno);
-	}
-
-	return 0;
-}
-
-/**
- * unify_sexattr - This function should be sent to all the storage nodes, 
- *       which contains the file, (excluding namespace).
- */
-int32_t
-unify_setxattr (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		dict_t *dict,
-		int32_t flags)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = NULL;
-	int16_t *list = NULL;
-	int16_t index = 0;
-	int32_t call_count = 0;
-  	uint64_t tmp_list = 0;
-	data_pair_t *trav = dict->members_list;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	local->failed = -1;
-	loc_copy (&local->loc1, loc);
-
-	if (IA_ISDIR (loc->inode->ia_type)) {
-
-		if (trav && trav->key && ZR_FILE_CONTENT_REQUEST(trav->key)) {
-			/* direct the storage xlators to change file 
-			   content only if file exists */
-			local->flags = flags;
-			local->dict = dict;
-			local->name = gf_strdup (trav->key);
-			flags |= XATTR_REPLACE;
-		}
-
-		local->call_count = priv->child_count;
-		for (index = 0; index < priv->child_count; index++) {
-			STACK_WIND (frame,
-				    unify_setxattr_cbk,
-				    priv->xl_array[index],
-				    priv->xl_array[index]->fops->setxattr,
-				    loc, dict, flags);
-		}
-		return 0;
-	}
-
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	for (index = 0; list[index] != -1; index++) {
-		if (NS(this) != priv->xl_array[list[index]]) {
-			local->call_count++;
-			call_count++;
-		}
-	}
-  
-	if (local->call_count) {
-		for (index = 0; list[index] != -1; index++) {
-			if (priv->xl_array[list[index]] != NS(this)) {
-				STACK_WIND (frame,
-					    unify_setxattr_cbk,
-					    priv->xl_array[list[index]],
-					    priv->xl_array[list[index]]->fops->setxattr,
-					    loc,
-					    dict,
-					    flags);
-				if (!--call_count)
-					break;
-			}
-		}
-		return 0;
-	}
-
-	/* No entry in storage nodes */
-	gf_log (this->name, GF_LOG_DEBUG, 
-		"returning ENOENT, file not found on storage node.");
-	STACK_UNWIND (frame, -1, ENOENT);
-
-	return 0;
-}
-
-
-/**
- * unify_getxattr_cbk - This function is called from only one child, so, no
- *     need of any lock or anything else, just send it to above layer 
- */
-int32_t
-unify_getxattr_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    dict_t *value)
-{
-	int32_t callcnt = 0;
-	dict_t *local_value = NULL;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-  
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, 
-				(((op_errno == ENOENT) || 
-				  (op_errno == ENODATA) || 
-				  (op_errno == ENOTSUP)) ? 
-				 GF_LOG_DEBUG : GF_LOG_ERROR), 
-				"child(%s): path(%s): %s", 
-				prev_frame->this->name, 
-				(local->loc1.path)?local->loc1.path:"", 
-				strerror (op_errno));
-		} else {
-			if (!local->dict)
-				local->dict = dict_ref (value);
-			local->op_ret = op_ret;
-		}
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		local_value = local->dict;
-		local->dict = NULL;
-      
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      local_value);
-      
-		if (local_value)
-			dict_unref (local_value);
-	} 
-
-	return 0;
-}
-
-
-/** 
- * unify_getxattr - This FOP is sent to only the storage node.
- */
-int32_t
-unify_getxattr (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		const char *name)
-{
-	unify_private_t *priv = this->private;
-	int16_t *list = NULL;
-	int16_t index = 0;
-	int16_t count = 0;
-	unify_local_t *local = NULL;
-  	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-	INIT_LOCAL (frame, local);
-
-	if (IA_ISDIR (loc->inode->ia_type)) {
-		local->call_count = priv->child_count;
-		for (index = 0; index < priv->child_count; index++)
-			STACK_WIND (frame,
-				    unify_getxattr_cbk,
-				    priv->xl_array[index],
-				    priv->xl_array[index]->fops->getxattr,
-				    loc,
-				    name);
-		return 0;
-	}
-
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	for (index = 0; list[index] != -1; index++) {
-		if (NS(this) != priv->xl_array[list[index]]) {
-			local->call_count++;
-			count++;
-		}
-	}
-
-	if (count) {
-		for (index = 0; list[index] != -1; index++) {
-			if (priv->xl_array[list[index]] != NS(this)) {
-				STACK_WIND (frame,
-					    unify_getxattr_cbk,
-					    priv->xl_array[list[index]],
-					    priv->xl_array[list[index]]->fops->getxattr,
-					    loc,
-					    name);
-				if (!--count)
-					break;
-			}
-		}
-	} else {
-		dict_t *tmp_dict = get_new_dict ();
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"%s: returning ENODATA, no file found on storage node",
-			loc->path);
-		STACK_UNWIND (frame, -1, ENODATA, tmp_dict);
-		dict_destroy (tmp_dict);
-	}
-
-	return 0;
-}
-
-/**
- * unify_removexattr_cbk - Wait till all the child node returns the call
- *      and then UNWIND to above layer.
- */
-int32_t
-unify_removexattr_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno)
-{
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-
-	LOCK (&frame->lock);
-	{ 
-		callcnt = --local->call_count;
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			if (op_errno != ENOTSUP)
-				gf_log (this->name, GF_LOG_ERROR, 
-					"child(%s): path(%s): %s", 
-					prev_frame->this->name, 
-					local->loc1.path, strerror (op_errno));
-		} else {
-			local->op_ret = op_ret;
-		}
-	}
-	UNLOCK (&frame->lock);  
-
-	if (!callcnt) {
-		STACK_UNWIND (frame, local->op_ret, local->op_errno);
-	}
-
-	return 0;
-}
-
-/**
- * unify_removexattr - Send it to all the child nodes which has the files.
- */
-int32_t
-unify_removexattr (call_frame_t *frame,
-		   xlator_t *this,
-		   loc_t *loc,
-		   const char *name)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = NULL;
-	int16_t *list = NULL;
-	int16_t index = 0;
-	int32_t call_count = 0;
-  	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-
-	if (IA_ISDIR (loc->inode->ia_type)) {
-		local->call_count = priv->child_count;
-		for (index = 0; index < priv->child_count; index++)
-			STACK_WIND (frame, 		    
-				    unify_removexattr_cbk,
-				    priv->xl_array[index],
-				    priv->xl_array[index]->fops->removexattr,
-				    loc,
-				    name);
-
-		return 0;
-	}
-
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	for (index = 0; list[index] != -1; index++) {
-		if (NS(this) != priv->xl_array[list[index]]) {
-			local->call_count++;
-			call_count++;
-		}
-	}
-
-	if (local->call_count) {
-		for (index = 0; list[index] != -1; index++) {
-			if (priv->xl_array[list[index]] != NS(this)) {
-				STACK_WIND (frame,
-					    unify_removexattr_cbk,
-					    priv->xl_array[list[index]],
-					    priv->xl_array[list[index]]->fops->removexattr,
-					    loc,
-					    name);
-				if (!--call_count)
-					break;
-			}
-		}
-		return 0;
-	} 
-
-	gf_log (this->name, GF_LOG_DEBUG, 
-		"%s: returning ENOENT, not found on storage node.", loc->path);
-	STACK_UNWIND (frame, -1, ENOENT);
-
-	return 0;
-}
-
-
-int32_t 
-unify_mknod_unlink_cbk (call_frame_t *frame,
-			void *cookie,
-			xlator_t *this,
-			int32_t op_ret,
-			int32_t op_errno,
-                        struct iatt *preparent,
-                        struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1)
-		gf_log (this->name, GF_LOG_ERROR, 
-			"%s: %s", local->loc1.path, strerror (op_errno));
-  
-	unify_local_wipe (local);
-	/* No log required here as this -1 is for mknod call */
-	STACK_UNWIND (frame, -1, local->op_errno, NULL, NULL);
-	return 0;
-}
-
-/**
- * unify_mknod_cbk - 
- */
-int32_t
-unify_mknod_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 inode_t *inode,
-                 struct iatt *buf,
-                 struct iatt *preparent,
-                 struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"mknod failed on storage node, sending unlink to "
-			"namespace");
-		local->op_errno = op_errno;
-		STACK_WIND (frame,
-			    unify_mknod_unlink_cbk,
-			    NS(this),
-			    NS(this)->fops->unlink,
-			    &local->loc1);
-		return 0;
-	}
-  
-	local->stbuf = *buf;
-	local->stbuf.ia_ino = local->ia_ino;
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, op_ret, op_errno, inode, &local->stbuf,
-                      &local->oldpreparent, &local->oldpostparent);
-	return 0;
-}
-
-/**
- * unify_ns_mknod_cbk - 
- */
-int32_t
-unify_ns_mknod_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    inode_t *inode,
-                    struct iatt *buf,
-                    struct iatt *preparent,
-                    struct iatt *postparent)
-{
-	struct sched_ops *sched_ops = NULL;
-	xlator_t *sched_xl = NULL;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t *list = NULL;
-	int16_t index = 0;
-	call_frame_t *prev_frame = cookie;
-
-	if (op_ret == -1) {
-		/* No need to send mknod request to other servers, 
-		 * as namespace action failed 
-		 */
-		gf_log (this->name, GF_LOG_ERROR, 
-			"child(%s): path(%s): %s", 
-			prev_frame->this->name, local->loc1.path, 
-			strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, inode, buf,
-                              preparent, postparent);
-		return 0;
-	}
-  
-	/* Create one inode for this entry */
-	local->op_ret = 0;
-	local->stbuf = *buf;
-	local->ia_ino = buf->ia_ino;
-
-        local->oldpreparent = *preparent;
-        local->oldpostparent = *postparent;
-
-	list = GF_CALLOC (1, sizeof (int16_t) * 3, gf_unify_mt_int16_t);
-	ERR_ABORT (list);
-	list[0] = priv->child_count;
-	list[2] = -1;
-	inode_ctx_put (inode, this, (uint64_t)(long)list);
-
-	sched_ops = priv->sched_ops;
-
-	/* Send mknod request to scheduled node now */
-	sched_xl = sched_ops->schedule (this, local->loc1.path); 
-	if (!sched_xl) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"mknod failed on storage node, no node online "
-			"at the moment, sending unlink to NS");
-		local->op_errno = ENOTCONN;
-		STACK_WIND (frame,
-			    unify_mknod_unlink_cbk,
-			    NS(this),
-			    NS(this)->fops->unlink,
-			    &local->loc1);
-      
-		return 0;
-	}
-
-	for (index = 0; index < priv->child_count; index++)
-		if (sched_xl == priv->xl_array[index])
-			break;
-	list[1] = index;
-  
-	STACK_WIND (frame,  unify_mknod_cbk,
-		    sched_xl,  sched_xl->fops->mknod,
-		    &local->loc1, local->mode, local->dev);
-
-	return 0;
-}
-
-/**
- * unify_mknod - Create a device on namespace first, and later create on 
- *       the storage node.
- */
-int32_t
-unify_mknod (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     mode_t mode,
-	     dev_t rdev)
-{
-	unify_local_t *local = NULL;
-  
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	local->mode = mode;
-	local->dev = rdev;
-	loc_copy (&local->loc1, loc);
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, loc->inode, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_ns_mknod_cbk,
-		    NS(this),
-		    NS(this)->fops->mknod,
-		    loc,
-		    mode,
-		    rdev);
-
-	return 0;
-}
-
-int32_t 
-unify_symlink_unlink_cbk (call_frame_t *frame,
-			  void *cookie,
-			  xlator_t *this,
-			  int32_t op_ret,
-			  int32_t op_errno,
-                          struct iatt *preparent,
-                          struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-	if (op_ret == -1)
-		gf_log (this->name, GF_LOG_ERROR, 
-			"%s: %s", local->loc1.path, strerror (op_errno));
-
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, -1, local->op_errno, NULL, NULL);
-	return 0;
-}
-
-/**
- * unify_symlink_cbk - 
- */
-int32_t
-unify_symlink_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   inode_t *inode,
-                   struct iatt *buf,
-                   struct iatt *preparent,
-                   struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1) {
-		/* Symlink on storage node failed, hence send unlink 
-		   to the NS node */
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, 
-			"symlink on storage node failed, sending unlink "
-			"to namespace");
-
-		STACK_WIND (frame,
-			    unify_symlink_unlink_cbk,
-			    NS(this),
-			    NS(this)->fops->unlink,
-			    &local->loc1);
-    
-		return 0;
-	}
-  
-	local->stbuf = *buf;
-	local->stbuf.ia_ino = local->ia_ino;
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, op_ret, op_errno, inode, &local->stbuf,
-                      &local->oldpreparent, &local->oldpostparent);
-
-	return 0;
-}
-
-/**
- * unify_ns_symlink_cbk - 
- */
-int32_t
-unify_ns_symlink_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      inode_t *inode,
-                      struct iatt *buf,
-                      struct iatt *preparent,
-                      struct iatt *postparent)
-{
-
-	struct sched_ops *sched_ops = NULL;
-	xlator_t *sched_xl = NULL;
-	int16_t *list = NULL;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t index = 0;
-
-	if (op_ret == -1) {
-		/* No need to send symlink request to other servers, 
-		 * as namespace action failed 
-		 */
-		gf_log (this->name, GF_LOG_ERROR, 
-			"namespace: path(%s): %s", 
-			local->loc1.path, strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, NULL, buf,
-                              preparent, postparent);
-		return 0;
-	}
-  
-	/* Create one inode for this entry */
-	local->op_ret = 0;
-	local->ia_ino = buf->ia_ino;
-  
-        local->oldpreparent = *preparent;
-        local->oldpostparent = *postparent;
-
-	/* Start the mapping list */
-
-	list = GF_CALLOC (1, sizeof (int16_t) * 3, gf_unify_mt_int16_t);
-	ERR_ABORT (list);
-	list[0] = priv->child_count; //namespace's index
-	list[2] = -1;
-	inode_ctx_put (inode, this, (uint64_t)(long)list);
-
-	sched_ops = priv->sched_ops;
-
-	/* Send symlink request to all the nodes now */
-	sched_xl = sched_ops->schedule (this, local->loc1.path); 
-	if (!sched_xl) {
-		/* Symlink on storage node failed, hence send unlink 
-		   to the NS node */
-		local->op_errno = ENOTCONN;
-		gf_log (this->name, GF_LOG_ERROR, 
-			"symlink on storage node failed, no node online, "
-			"sending unlink to namespace");
-      
-		STACK_WIND (frame,
-			    unify_symlink_unlink_cbk,
-			    NS(this),
-			    NS(this)->fops->unlink,
-			    &local->loc1);
-      
-		return 0;
-	}
-
-	for (index = 0; index < priv->child_count; index++)
-		if (sched_xl == priv->xl_array[index])
-			break;
-	list[1] = index;
-	
-	STACK_WIND (frame,
-		    unify_symlink_cbk,
-		    sched_xl,
-		    sched_xl->fops->symlink,
-		    local->name,
-		    &local->loc1);
-
-	return 0;
-}
-
-/**
- * unify_symlink - 
- */
-int32_t
-unify_symlink (call_frame_t *frame,
-	       xlator_t *this,
-	       const char *linkpath,
-	       loc_t *loc)
-{
-	unify_local_t *local = NULL;
-  
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-	local->name = gf_strdup (linkpath);
-
-	if ((local->name == NULL) || 
-	    (local->loc1.path == NULL)) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, loc->inode, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_ns_symlink_cbk,
-		    NS(this),
-		    NS(this)->fops->symlink,
-		    linkpath,
-		    loc);
-
-	return 0;
-}
-
-
-int32_t 
-unify_rename_unlink_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-                         struct iatt *preparent,
-                         struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-  
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"child(%s): path(%s -> %s): %s", 
-			prev_frame->this->name, 
-			local->loc1.path, local->loc2.path, 
-			strerror (op_errno));
-      
-	}
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		local->stbuf.ia_ino = local->ia_ino;
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      &local->stbuf);
-	}
-	return 0;
-}
-
-int32_t 
-unify_ns_rename_undo_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)
-{
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"namespace: path(%s -> %s): %s", 
-			local->loc1.path, local->loc2.path, 
-			strerror (op_errno));
-	}
-
-	local->stbuf.ia_ino = local->ia_ino;
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, local->op_ret, local->op_errno, &local->stbuf);
-	return 0;
-}
-
-int32_t 
-unify_rename_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct iatt *buf,
-                  struct iatt *preoldparent,
-                  struct iatt *postoldparent,
-                  struct iatt *prenewparent,
-                  struct iatt *postnewparent)
-{
-	int32_t index = 0;
-	int32_t callcnt = 0;
-	int16_t *list = NULL;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-  
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret >= 0) {
-			if (!IA_ISDIR (buf->ia_type))
-				local->stbuf = *buf;
-			local->op_ret = op_ret;
-		} else {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"child(%s): path(%s -> %s): %s", 
-				prev_frame->this->name, 
-				local->loc1.path, local->loc2.path, 
-				strerror (op_errno));
-			local->op_errno = op_errno;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		local->stbuf.ia_ino = local->ia_ino;
-		if (IA_ISDIR (local->loc1.inode->ia_type)) {
-			unify_local_wipe (local);
-			STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                                      &local->stbuf, &local->oldpreparent,
-                                      &local->oldpostparent, &local->newpreparent,
-                                      &local->newpostparent);
-			return 0;
-		}
-
-		if (local->op_ret == -1) {
-			/* TODO: check this logic */
-
-			/* Rename failed in storage node, successful on NS, 
-			 * hence, rename back the entries in NS */
-			/* NOTE: this will be done only if the destination 
-			 * doesn't exists, if  the destination exists, the 
-			 * job of correcting NS is left to self-heal
-			 */
-			if (!local->index) {
-				loc_t tmp_oldloc = {
-                                        /* its actual 'newloc->path' */
-					.path = local->loc2.path, 
-					.inode = local->loc1.inode,
-					.parent = local->loc2.parent
-				};
-	
-				loc_t tmp_newloc = {
-					/* Actual 'oldloc->path' */
-					.path = local->loc1.path,
-					.parent = local->loc1.parent
-				};
-
-				gf_log (this->name, GF_LOG_ERROR, 
-					"rename succussful on namespace, on "
-					"stroage node failed, reverting back");
-
-				STACK_WIND (frame,
-					    unify_ns_rename_undo_cbk,
-					    NS(this),
-					    NS(this)->fops->rename,
-					    &tmp_oldloc,
-					    &tmp_newloc);
-				return 0;
-			}
-		} else {
-			/* Rename successful on storage nodes */
-
-			int32_t idx = 0;
-			int16_t *tmp_list = NULL;
-			uint64_t tmp_list_int64 = 0;
-			if (local->loc2.inode) {
-				inode_ctx_get (local->loc2.inode, 
-					       this, &tmp_list_int64);
-				list = (int16_t *)(long)tmp_list_int64;
-
-			}
-
-			if (list) {				
-				for (index = 0; list[index] != -1; index++);
-				tmp_list = GF_CALLOC (1, index * 2, 
-                                                      gf_unify_mt_int16_t);
-				memcpy (tmp_list, list, index * 2);
-
-				for (index = 0; list[index] != -1; index++) {
-					/* TODO: Check this logic. */
-					/* If the destination file exists in 
-					 * the same storage node where we sent
-					 * 'rename' call, no need to send 
-					 * unlink 
-					 */
-					for (idx = 0; 
-					     local->list[idx] != -1; idx++) {
-						if (tmp_list[index] == local->list[idx]) {
-							tmp_list[index] = priv->child_count;
-							continue;
-						}
-					}
-	  
-					if (NS(this) != priv->xl_array[tmp_list[index]]) {
-						local->call_count++;
-						callcnt++;
-					}
-				}
-
-				if (local->call_count) {
-					if (callcnt > 1)
-						gf_log (this->name, 
-							GF_LOG_ERROR, 
-							"%s->%s: more (%d) "
-							"subvolumes have the "
-							"newloc entry", 
-							local->loc1.path, 
-							local->loc2.path, 
-							callcnt);
-
-					for (index=0; 
-					     tmp_list[index] != -1; index++) {
-						if (NS(this) != priv->xl_array[tmp_list[index]]) {		    
-							STACK_WIND (frame,
-								    unify_rename_unlink_cbk,
-								    priv->xl_array[tmp_list[index]],
-								    priv->xl_array[tmp_list[index]]->fops->unlink,
-								    &local->loc2);
-							if (!--callcnt)
-								break;
-						}
-					}
-
-					GF_FREE (tmp_list);
-					return 0;
-				}
-				if (tmp_list)
-					GF_FREE (tmp_list);
-			}
-		}
-    
-		/* Need not send 'unlink' to storage node */
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, 
-			      local->op_errno, &local->stbuf,
-                              &local->oldpreparent, &local->oldpostparent,
-                              &local->newpreparent, &local->newpostparent);
-	}
-
-	return 0;
-}
-
-int32_t 
-unify_ns_rename_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     struct iatt *buf,
-                     struct iatt *preoldparent,
-                     struct iatt *postoldparent,
-                     struct iatt *prenewparent,
-                     struct iatt *postnewparent)
-{
-	int32_t index = 0;
-	int32_t callcnt = 0;
-	int16_t *list = NULL;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1) {
-		/* Free local->new_inode */
-		gf_log (this->name, GF_LOG_ERROR, 
-			"namespace: path(%s -> %s): %s", 
-			local->loc1.path, local->loc2.path, 
-			strerror (op_errno));
-
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, buf,
-                              preoldparent, postoldparent,
-                              prenewparent, postnewparent);
-		return 0;
-	}
-
-	local->stbuf = *buf;
-	local->ia_ino = buf->ia_ino;
-
-        local->oldpreparent = *preoldparent;
-        local->oldpostparent = *postoldparent;
-        local->newpreparent = *prenewparent;
-        local->newpostparent = *postnewparent;
-
-	/* Everything is fine. */
-	if (IA_ISDIR (buf->ia_type)) {
-		local->call_count = priv->child_count;
-		for (index=0; index < priv->child_count; index++) {
-			STACK_WIND (frame,
-				    unify_rename_cbk,
-				    priv->xl_array[index],
-				    priv->xl_array[index]->fops->rename,
-				    &local->loc1,
-				    &local->loc2);
-		}
-
-		return 0;
-	}
-
-	local->call_count = 0;  
-	/* send rename */
-	list = local->list;
-	for (index=0; list[index] != -1; index++) {
-		if (NS(this) != priv->xl_array[list[index]]) {
-			local->call_count++;
-			callcnt++;
-		}
-	}
-
-	if (local->call_count) {
-		for (index=0; list[index] != -1; index++) {
-			if (NS(this) != priv->xl_array[list[index]]) {
-				STACK_WIND (frame,
-					    unify_rename_cbk,
-					    priv->xl_array[list[index]],
-					    priv->xl_array[list[index]]->fops->rename,
-					    &local->loc1,
-					    &local->loc2);
-				if (!--callcnt)
-					break;
-			}
-		}
-	} else {
-		/* file doesn't seem to be present in storage nodes */
-		gf_log (this->name, GF_LOG_CRITICAL,
-			"CRITICAL: source file not in storage node, "
-			"rename successful on namespace :O");
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, -1, EIO, NULL,
-                              NULL, NULL, /* preoldparent, postoldparent */
-                              NULL, NULL); /* prenewparent, postnewparent */
-	}
-	return 0;
-}
-
-
-/**
- * unify_rename - One of the tricky function. The deadliest of all :O
- */
-int32_t
-unify_rename (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *oldloc,
-	      loc_t *newloc)
-{
-	unify_local_t *local = NULL;
-  	uint64_t tmp_list = 0;
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, oldloc);
-	loc_copy (&local->loc2, newloc);
-
-	if ((local->loc1.path == NULL) || 
-	    (local->loc2.path == NULL)) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, NULL,
-                              NULL, NULL, /* preoldparent, postoldparent */
-                              NULL, NULL); /* prenewparent, postnewparent */
-		return 0;
-	}
-  
-	inode_ctx_get (oldloc->inode, this, &tmp_list);
-	local->list = (int16_t *)(long)tmp_list;
-
-	STACK_WIND (frame,
-		    unify_ns_rename_cbk,
-		    NS(this),
-		    NS(this)->fops->rename,
-		    oldloc,
-		    newloc);
-	return 0;
-}
-
-/**
- * unify_link_cbk -
- */
-int32_t
-unify_link_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		inode_t *inode,
-                struct iatt *buf,
-                struct iatt *preparent,
-                struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-
-	if (op_ret >= 0) 
-		local->stbuf = *buf;
-	local->stbuf.ia_ino = local->ia_ino;
-
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, op_ret, op_errno, inode, &local->stbuf,
-                      &local->oldpreparent, &local->oldpostparent);
-
-	return 0;
-}
-
-/**
- * unify_ns_link_cbk - 
- */
-int32_t
-unify_ns_link_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   inode_t *inode,
-                   struct iatt *buf,
-                   struct iatt *preparent,
-                   struct iatt *postparent)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	int16_t *list = local->list;
-	int16_t index = 0;
-
-	if (op_ret == -1) {
-		/* No need to send link request to other servers, 
-		 * as namespace action failed 
-		 */
-		gf_log (this->name, GF_LOG_ERROR, 
-			"namespace: path(%s -> %s): %s", 
-			local->loc1.path, local->loc2.path, 
-			strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, inode, buf,
-                              preparent, postparent);
-		return 0;
-	}
-
-	/* Update inode for this entry */
-	local->op_ret = 0;
-	local->ia_ino = buf->ia_ino;
-
-        local->oldpreparent = *preparent;
-        local->oldpostparent = *postparent;
-
-	/* Send link request to the node now */
-	for (index = 0; list[index] != -1; index++) {
-		char need_break = (list[index+1] == -1);
-		if (priv->xl_array[list[index]] != NS (this)) {
-			STACK_WIND (frame,
-				    unify_link_cbk,
-				    priv->xl_array[list[index]],
-				    priv->xl_array[list[index]]->fops->link,
-				    &local->loc1,
-				    &local->loc2);
-                        break;
-		}
-		if (need_break)
-			break;
-	}
-
-	return 0;
-}
-
-/**
- * unify_link - 
- */
-int32_t
-unify_link (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *oldloc,
-	    loc_t *newloc)
-{
-	unify_local_t *local = NULL;
-  	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (oldloc);
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (newloc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-
-	loc_copy (&local->loc1, oldloc);
-	loc_copy (&local->loc2, newloc);
-
-	inode_ctx_get (oldloc->inode, this, &tmp_list);
-	local->list = (int16_t *)(long)tmp_list;
-
-	STACK_WIND (frame,
-		    unify_ns_link_cbk,
-		    NS(this),
-		    NS(this)->fops->link,
-		    oldloc,
-		    newloc);
-
-	return 0;
-}
-
-
-/**
- * unify_checksum_cbk - 
- */
-int32_t
-unify_checksum_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    uint8_t *fchecksum,
-		    uint8_t *dchecksum)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fchecksum, dchecksum);
-
-	return 0;
-}
-
-/**
- * unify_checksum - 
- */
-int32_t
-unify_checksum (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		int32_t flag)
-{
-	STACK_WIND (frame,
-		    unify_checksum_cbk,
-		    NS(this),
-		    NS(this)->fops->checksum,
-		    loc,
-		    flag);
-
-	return 0;
-}
-
-
-/**
- * unify_finodelk_cbk - 
- */
-int
-unify_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-/**
- * unify_finodelk
- */
-int
-unify_finodelk (call_frame_t *frame, xlator_t *this,
-		const char *volume, fd_t *fd, int cmd, struct gf_flock *flock)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_finodelk_cbk,
-		    child, child->fops->finodelk,
-		    volume, fd, cmd, flock);
-
-	return 0;
-}
-
-
-
-/**
- * unify_fentrylk_cbk - 
- */
-int
-unify_fentrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-/**
- * unify_fentrylk
- */
-int
-unify_fentrylk (call_frame_t *frame, xlator_t *this,
-		const char *volume, fd_t *fd, const char *basename,
-		entrylk_cmd cmd, entrylk_type type)
-		
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_fentrylk_cbk,
-		    child, child->fops->fentrylk,
-		    volume, fd, basename, cmd, type);
-
-	return 0;
-}
-
-
-
-/**
- * unify_fxattrop_cbk - 
- */
-int
-unify_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, dict_t *xattr)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, xattr);
-	return 0;
-}
-
-/**
- * unify_fxattrop
- */
-int
-unify_fxattrop (call_frame_t *frame, xlator_t *this,
-		fd_t *fd, gf_xattrop_flags_t optype, dict_t *xattr)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_fxattrop_cbk,
-		    child, child->fops->fxattrop,
-		    fd, optype, xattr);
-
-	return 0;
-}
-
-
-/**
- * unify_inodelk_cbk - 
- */
-int
-unify_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		   int32_t op_ret, int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-/**
- * unify_inodelk
- */
-int
-unify_inodelk (call_frame_t *frame, xlator_t *this,
-	       const char *volume, loc_t *loc, int cmd, struct gf_flock *flock)
-{
-	xlator_t *child = NULL;
-
-	child = unify_loc_subvol (loc, this);
-
-	STACK_WIND (frame, unify_inodelk_cbk,
-		    child, child->fops->inodelk,
-		    volume, loc, cmd, flock);
-
-	return 0;
-}
-
-
-
-/**
- * unify_entrylk_cbk - 
- */
-int
-unify_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		   int32_t op_ret, int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-/**
- * unify_entrylk
- */
-int
-unify_entrylk (call_frame_t *frame, xlator_t *this,
-	       const char *volume, loc_t *loc, const char *basename,
-	       entrylk_cmd cmd, entrylk_type type)
-		
-{
-	xlator_t *child = NULL;
-
-	child = unify_loc_subvol (loc, this);
-
-	STACK_WIND (frame, unify_entrylk_cbk,
-		    child, child->fops->entrylk,
-		    volume, loc, basename, cmd, type);
-
-	return 0;
-}
-
-
-
-/**
- * unify_xattrop_cbk - 
- */
-int
-unify_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		   int32_t op_ret, int32_t op_errno, dict_t *xattr)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, xattr);
-	return 0;
-}
-
-/**
- * unify_xattrop
- */
-int
-unify_xattrop (call_frame_t *frame, xlator_t *this,
-		loc_t *loc, gf_xattrop_flags_t optype, dict_t *xattr)
-{
-	xlator_t *child = NULL;
-
-	child = unify_loc_subvol (loc, this);
-
-	STACK_WIND (frame, unify_xattrop_cbk,
-		    child, child->fops->xattrop,
-		    loc, optype, xattr);
-
-	return 0;
-}
-
-int
-unify_forget (xlator_t *this,
-              inode_t *inode)
-{
-        int16_t *list = NULL;
-        uint64_t tmp_list = 0;
-
-        if (inode->ia_type && (!IA_ISDIR(inode->ia_type))) {
-                inode_ctx_get (inode, this, &tmp_list);
-                if (tmp_list) {
-                        list = (int16_t *)(long)tmp_list;
-                        GF_FREE (list);
-                }
-        }
-
-        return 0;
-}
-
-/**
- * notify
- */
-int32_t
-notify (xlator_t *this,
-        int32_t event,
-        void *data,
-        ...)
-{
-	unify_private_t *priv = this->private;
-	struct sched_ops *sched = NULL;
-
-	if (!priv) {
-		return 0;
-	}
-
-	sched = priv->sched_ops;    
-	if (!sched) {
-		gf_log (this->name, GF_LOG_CRITICAL, "No scheduler :O");
-		raise (SIGTERM);
-		return 0;
-	}
-	if (priv->namespace == data) {
-		if (event == GF_EVENT_CHILD_UP) {
-			sched->notify (this, event, data);
-		}
-		return 0;
-	}
-
-	switch (event)
-	{
-	case GF_EVENT_CHILD_UP:
-	{
-		/* Call scheduler's update () to enable it for scheduling */
-		sched->notify (this, event, data);
-	
-		LOCK (&priv->lock);
-		{
-			/* Increment the inode's generation, which is 
-			   used for self_heal */
-			++priv->inode_generation;
-			++priv->num_child_up;
-		}
-		UNLOCK (&priv->lock);
-
-		if (!priv->is_up) {
-			default_notify (this, event, data);
-			priv->is_up = 1;
-		}
-	}
-	break;
-	case GF_EVENT_CHILD_DOWN:
-	{
-		/* Call scheduler's update () to disable the child node 
-		 * for scheduling
-		 */
-		sched->notify (this, event, data);
-		LOCK (&priv->lock);
-		{
-			--priv->num_child_up;
-		}
-		UNLOCK (&priv->lock);
-
-		if (priv->num_child_up == 0) {
-			/* Send CHILD_DOWN to upper layer */
-			default_notify (this, event, data);
-			priv->is_up = 0;
-		}
-	}
-	break;
-
-	default:
-	{
-		default_notify (this, event, data);
-	}
-	break;
-	}
-
-	return 0;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_unify_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
-
-        return ret;
-}
-
-/** 
- * init - This function is called first in the xlator, while initializing.
- *   All the config file options are checked and appropriate flags are set.
- *
- * @this - 
- */
-int32_t 
-init (xlator_t *this)
-{
-	int32_t          ret       = 0;
-	int32_t          count     = 0;
-	data_t          *scheduler = NULL;
-	data_t          *data      = NULL;
-	xlator_t        *ns_xl     = NULL;
-	xlator_list_t   *trav      = NULL;
-	xlator_list_t   *xlparent  = NULL;
-	xlator_list_t   *parent    = NULL;
-	unify_private_t *_private  = NULL; 
-
-
-	/* Check for number of child nodes, if there is no child nodes, exit */
-	if (!this->children) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"No child nodes specified. check \"subvolumes \" "
-			"option in volfile");
-		return -1;
-	}
-
-  	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-  
-	/* Check for 'scheduler' in volume */
-	scheduler = dict_get (this->options, "scheduler");
-	if (!scheduler) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"\"option scheduler <x>\" is missing in volfile");
-		return -1;
-	}
-
-	/* Setting "option namespace <node>" */
-	data = dict_get (this->options, "namespace");
-	if(!data) {
-		gf_log (this->name, GF_LOG_CRITICAL, 
-			"namespace option not specified, Exiting");
-		return -1;
-	}
-	/* Search namespace in the child node, if found, exit */
-	trav = this->children;
-	while (trav) {
-		if (strcmp (trav->xlator->name, data->data) == 0)
-			break;
-		trav = trav->next;
-	}
-	if (trav) {
-		gf_log (this->name, GF_LOG_CRITICAL, 
-			"namespace node used as a subvolume, Exiting");
-		return -1;
-	}
-	
-	/* Search for the namespace node, if found, continue */
-	ns_xl = this->next;
-	while (ns_xl) {
-		if (strcmp (ns_xl->name, data->data) == 0)
-			break;
-		ns_xl = ns_xl->next;
-	}
-	if (!ns_xl) {
-		gf_log (this->name, GF_LOG_CRITICAL, 
-			"namespace node not found in volfile, Exiting");
-		return -1;
-	}
-	
-	gf_log (this->name, GF_LOG_DEBUG, 
-		"namespace node specified as %s", data->data);
-	
-	_private = GF_CALLOC (1, sizeof (*_private), 
-                              gf_unify_mt_unify_private_t);
-	ERR_ABORT (_private);
-	_private->sched_ops = get_scheduler (this, scheduler->data);
-	if (!_private->sched_ops) {
-		gf_log (this->name, GF_LOG_CRITICAL, 
-			"Error while loading scheduler. Exiting");
-		GF_FREE (_private);
-		return -1;
-	}
-	
-	if (ns_xl->parents) {
-		gf_log (this->name, GF_LOG_CRITICAL,
-			"Namespace node should not be a child of any other node. Exiting");
-		GF_FREE (_private);
-		return -1;
-	}
-
-	_private->namespace = ns_xl;
-	
-	/* update _private structure */
-	{
-		count = 0;
-		trav = this->children;
-		/* Get the number of child count */
-		while (trav) {
-			count++;
-			trav = trav->next;
-		}
-		
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"Child node count is %d", count);    
-
-		_private->child_count = count;
-		if (count == 1) {
-			/* TODO: Should I error out here? */
-			gf_log (this->name, GF_LOG_CRITICAL, 
-				"WARNING: You have defined only one "
-				"\"subvolumes\" for unify volume. It may not "
-				"be the desired config, review your volume "
-				"volfile. If this is how you are testing it,"
-				" you may hit some performance penalty");
-		}
-		
-		_private->xl_array = GF_CALLOC (1, 
-					        sizeof (xlator_t) * (count + 1),
-                                                gf_unify_mt_xlator_t);
-		ERR_ABORT (_private->xl_array);
-		
-		count = 0;
-		trav = this->children;
-		while (trav) {
-			_private->xl_array[count++] = trav->xlator;
-			trav = trav->next;
-		}
-		_private->xl_array[count] = _private->namespace;
-		
-		/* self-heal part, start with generation '1' */
-		_private->inode_generation = 1; 
-                /* Because, Foreground part is tested well */
-		_private->self_heal = ZR_UNIFY_FG_SELF_HEAL; 
-		data = dict_get (this->options, "self-heal");
-		if (data) {
-			if (strcasecmp (data->data, "off") == 0) 
-				_private->self_heal = ZR_UNIFY_SELF_HEAL_OFF;
-
-			if (strcasecmp (data->data, "foreground") == 0)
-				_private->self_heal = ZR_UNIFY_FG_SELF_HEAL;
-
-			if (strcasecmp (data->data, "background") == 0)
-				_private->self_heal = ZR_UNIFY_BG_SELF_HEAL;
-		}
-    
-		/* optimist - ask bulde for more about it */
-		data = dict_get (this->options, "optimist");
-		if (data) {
-			if (gf_string2boolean (data->data, 
-					       &_private->optimist) == -1) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"optimist excepts only boolean "
-					"options");
-			}
-		}
-
-		LOCK_INIT (&_private->lock);
-	}
-
-	/* Now that everything is fine. */
-	this->private = (void *)_private;
-	{
-                ret = _private->sched_ops->mem_acct_init (this);
-
-                if (ret == -1) {
-                        return -1;
-                }
-
-		/* Initialize scheduler, if everything else is successful */
-		ret = _private->sched_ops->init (this); 
-		if (ret == -1) {
-			gf_log (this->name, GF_LOG_CRITICAL,
-				"Initializing scheduler failed, Exiting");
-			GF_FREE (_private);
-			return -1;
-		}
-
-
-		ret = 0;
-
-		/* This section is required because some fops may look 
-		 * for 'xl->parent' variable 
-		 */
-		xlparent = GF_CALLOC (1, sizeof (*xlparent),
-                                      gf_unify_mt_xlator_list_t);
-		xlparent->xlator = this;
-		if (!ns_xl->parents) {
-			ns_xl->parents = xlparent;
-		} else {
-			parent = ns_xl->parents;
-			while (parent->next)
-				parent = parent->next;
-			parent->next = xlparent;
-		}
-	}
-
-	/* Tell namespace node that init is done */
-	xlator_notify (ns_xl, GF_EVENT_PARENT_UP, this);
-
-	return 0;
-}
-
-/** 
- * fini  - Free all the allocated memory 
- */
-void
-fini (xlator_t *this)
-{
-	unify_private_t *priv = this->private;
-	priv->sched_ops->fini (this);
-	this->private = NULL;
-	LOCK_DESTROY (&priv->lock);
-	GF_FREE (priv->xl_array);
-	GF_FREE (priv);
-	return;
-}
-
-
-struct xlator_fops fops = {
-	.stat        = unify_stat,
-	.readlink    = unify_readlink,
-	.mknod       = unify_mknod,
-	.mkdir       = unify_mkdir,
-	.unlink      = unify_unlink,
-	.rmdir       = unify_rmdir,
-	.symlink     = unify_symlink,
-	.rename      = unify_rename,
-	.link        = unify_link,
-	.truncate    = unify_truncate,
-	.create      = unify_create,
-	.open        = unify_open,
-	.readv       = unify_readv,
-	.writev      = unify_writev,
-	.statfs      = unify_statfs,
-	.flush       = unify_flush,
-	.fsync       = unify_fsync,
-	.setxattr    = unify_setxattr,
-	.getxattr    = unify_getxattr,
-	.removexattr = unify_removexattr,
-	.opendir     = unify_opendir,
-	.readdir     = unify_readdir,
-	.readdirp    = unify_readdirp,
-	.fsyncdir    = unify_fsyncdir,
-	.access      = unify_access,
-	.ftruncate   = unify_ftruncate,
-	.fstat       = unify_fstat,
-	.lk          = unify_lk,
-	.lookup      = unify_lookup,
-	.getdents    = unify_getdents,
-	.checksum    = unify_checksum,
-	.inodelk     = unify_inodelk,
-	.finodelk    = unify_finodelk,
-	.entrylk     = unify_entrylk,
-	.fentrylk    = unify_fentrylk,
-	.xattrop     = unify_xattrop,
-	.fxattrop    = unify_fxattrop,
-        .setattr     = unify_setattr, 
-        .fsetattr    = unify_fsetattr,
-};
-
-
-struct xlator_cbks cbks = {
-        .forget  = unify_forget,
-};
-
-struct volume_options options[] = {
-	{ .key   = { "namespace" },  
-	  .type  = GF_OPTION_TYPE_XLATOR 
-	},
-	{ .key   = { "scheduler" },  
-	  .value = { "alu", "rr", "random", "nufa", "switch" },
-	  .type  = GF_OPTION_TYPE_STR
-	},
-	{ .key   = {"self-heal"},  
-	  .value = { "foreground", "background", "off" },
-	  .type  = GF_OPTION_TYPE_STR
-	},
-	/* TODO: remove it some time later */
-	{ .key   = {"optimist"},  
-	  .type  = GF_OPTION_TYPE_BOOL 
-	},
-
-	{ .key   = {NULL} },
-};
diff --git a/xlators/cluster/unify/src/unify.h b/xlators/cluster/unify/src/unify.h
deleted file mode 100644
index 3cfe725f4..000000000
--- a/xlators/cluster/unify/src/unify.h
+++ /dev/null
@@ -1,146 +0,0 @@
-/*
-  Copyright (c) 2006-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#ifndef _UNIFY_H
-#define _UNIFY_H
-
-#include "scheduler.h"
-#include "list.h"
-#include "unify-mem-types.h"
-
-#define MAX_DIR_ENTRY_STRING     (32 * 1024)
-
-#define ZR_UNIFY_SELF_HEAL_OFF 0
-#define ZR_UNIFY_FG_SELF_HEAL  1
-#define ZR_UNIFY_BG_SELF_HEAL  2
-
-/* Sometimes one should use completely random numbers.. its good :p */
-#define UNIFY_SELF_HEAL_GETDENTS_COUNT 512
-
-#define NS(xl)          (((unify_private_t *)xl->private)->namespace)
-
-/* This is used to allocate memory for local structure */
-#define INIT_LOCAL(fr, loc)                   \
-do {                                          \
-  loc = GF_CALLOC (1, sizeof (unify_local_t), gf_unify_mt_unify_local_t);   \
-  ERR_ABORT (loc);			      \
-  if (!loc) {                                 \
-    STACK_UNWIND (fr, -1, ENOMEM);            \
-    return 0;                                 \
-  }                                           \
-  fr->local = loc;                            \
-  loc->op_ret = -1;                           \
-  loc->op_errno = ENOENT;                     \
-} while (0)
-
-
-
-struct unify_private {
-	/* Update this structure depending on requirement */
-	void *scheduler;               /* THIS SHOULD BE THE FIRST VARIABLE, 
-					  if xlator is using scheduler */
-	struct sched_ops *sched_ops;   /* Scheduler options  */
-	xlator_t *namespace;           /* ptr to namespace xlator */
-	xlator_t **xl_array;
-	gf_boolean_t optimist;
-	int16_t child_count;
-	int16_t num_child_up;
-	uint8_t self_heal;
-	uint8_t is_up;
-	uint64_t inode_generation;
-	gf_lock_t lock;
-};
-typedef struct unify_private unify_private_t;
-
-struct unify_self_heal_struct {
-	uint8_t dir_checksum[NAME_MAX];
-	uint8_t ns_dir_checksum[NAME_MAX];
-	uint8_t file_checksum[NAME_MAX];
-	uint8_t ns_file_checksum[NAME_MAX];
-	off_t *offset_list;
-	int   *count_list;
-	dir_entry_t **entry_list;
-};
-
-
-struct _unify_local_t {
-	int32_t call_count;
-	int32_t op_ret;
-	int32_t op_errno;
-	mode_t mode;
-	off_t offset;
-	dev_t dev;
-	uid_t uid;
-	gid_t gid;
-	int32_t flags;
-	int32_t entry_count;
-	int32_t count;    // dir_entry_t count;
-	fd_t *fd;
-	struct iatt stbuf;
-        struct iatt stpre;
-        struct iatt stpost;
-	struct statvfs statvfs_buf;
-	struct timespec tv[2];
-	char *name;
-	int32_t revalidate;
-
-	ino_t ia_ino;
-	nlink_t ia_nlink;
-  
-	dict_t *dict;
-
-	int16_t *list;
-	int16_t *new_list; /* Used only in case of rename */
-	int16_t index;
-
-	int32_t failed;
-	int32_t return_eio;  /* Used in case of different st-mode 
-				present for a given path */
-
-	uint64_t inode_generation; /* used to store the per directory 
-				    * inode_generation. Got from inode's ctx 
-				    * of directory inodes
-				    */
-
-	struct unify_self_heal_struct *sh_struct;
-	loc_t loc1, loc2;
-
-        struct iatt poststbuf;
-        /* When not used for rename, old*
-         * are used as the attrs for the current
-         * parent directory.
-         */
-        struct iatt oldpreparent;
-        struct iatt oldpostparent;
-        struct iatt newpreparent;
-        struct iatt newpostparent;
-        int32_t wbflags;
-};
-typedef struct _unify_local_t unify_local_t;
-
-int32_t zr_unify_self_heal (call_frame_t *frame,
-			    xlator_t *this,
-			    unify_local_t *local);
-
-#endif /* _UNIFY_H */
diff --git a/xlators/debug/error-gen/src/Makefile.am b/xlators/debug/error-gen/src/Makefile.am
index df9080358..5075c59a8 100644
--- a/xlators/debug/error-gen/src/Makefile.am
+++ b/xlators/debug/error-gen/src/Makefile.am
@@ -2,15 +2,16 @@
 xlator_LTLIBRARIES = error-gen.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
 
-error_gen_la_LDFLAGS = -module -avoidversion
+error_gen_la_LDFLAGS = -module -avoid-version
 
 error_gen_la_SOURCES = error-gen.c
 error_gen_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
 noinst_HEADERS = error-gen.h error-gen-mem-types.h
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 CLEANFILES = 
 
diff --git a/xlators/debug/error-gen/src/error-gen-mem-types.h b/xlators/debug/error-gen/src/error-gen-mem-types.h
index 0fae10944..f02280535 100644
--- a/xlators/debug/error-gen/src/error-gen-mem-types.h
+++ b/xlators/debug/error-gen/src/error-gen-mem-types.h
@@ -1,23 +1,13 @@
 /*
-   Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   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 __ERROR_GEN_MEM_TYPES_H__
 #define __ERROR_GEN_MEM_TYPES_H__
 
diff --git a/xlators/debug/error-gen/src/error-gen.c b/xlators/debug/error-gen/src/error-gen.c
index 458db3003..ec0874b35 100644
--- a/xlators/debug/error-gen/src/error-gen.c
+++ b/xlators/debug/error-gen/src/error-gen.c
@@ -1,22 +1,12 @@
 /*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   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 _CONFIG_H
 #define _CONFIG_H
 #include "config.h"
@@ -24,6 +14,7 @@
 
 #include "xlator.h"
 #include "error-gen.h"
+#include "statedump.h"
 
 sys_error_t error_no_list[] = {
         [GF_FOP_LOOKUP]            = { .error_no_count = 4,
@@ -91,9 +82,10 @@ sys_error_t error_no_list[] = {
         [GF_FOP_READ]              = { .error_no_count = 5,
                                     .error_no = {EINVAL,EBADF,EFAULT,EISDIR,
                                                  ENAMETOOLONG}},
-        [GF_FOP_WRITE]             = { .error_no_count = 5,
+        [GF_FOP_WRITE]             = { .error_no_count = 7,
                                     .error_no = {EINVAL,EBADF,EFAULT,EISDIR,
-                                                 ENAMETOOLONG}},
+                                                 ENAMETOOLONG,ENOSPC,
+						 GF_ERROR_SHORT_WRITE}},
         [GF_FOP_STATFS]            = {.error_no_count = 10,
                                    .error_no = {EACCES,EBADF,EFAULT,EINTR,
                                                 EIO,ENAMETOOLONG,ENOENT,
@@ -113,6 +105,15 @@ sys_error_t error_no_list[] = {
         [GF_FOP_REMOVEXATTR]       = { .error_no_count = 4,
                                     .error_no = {EACCES,EBADF,ENAMETOOLONG,
                                                  EINTR}},
+        [GF_FOP_FSETXATTR]          = { .error_no_count = 4,
+                                        .error_no = {EACCES,EBADF,EINTR,
+                                                     ENAMETOOLONG}},
+        [GF_FOP_FGETXATTR]          = { .error_no_count = 4,
+                                        .error_no = {EACCES,EBADF,ENAMETOOLONG,
+                                                     EINTR}},
+        [GF_FOP_FREMOVEXATTR]       = { .error_no_count = 4,
+                                        .error_no = {EACCES,EBADF,ENAMETOOLONG,
+                                                     EINTR}},
         [GF_FOP_OPENDIR]           = { .error_no_count = 8,
                                     .error_no = {EACCES,EEXIST,EFAULT,
                                                  EISDIR,EMFILE,
@@ -237,6 +238,8 @@ conv_errno_to_int (char **error_no)
                 return EINTR;
         else if (!strcmp ((*error_no), "EFBIG"))
                 return EFBIG;
+	else if (!strcmp((*error_no), "GF_ERROR_SHORT_WRITE"))
+		return GF_ERROR_SHORT_WRITE;
         else
                 return EAGAIN;
 }
@@ -286,6 +289,12 @@ get_fop_int (char **op_no_str)
                 return GF_FOP_GETXATTR;
         else if (!strcmp ((*op_no_str), "removexattr"))
                 return GF_FOP_REMOVEXATTR;
+        else if (!strcmp ((*op_no_str), "fsetxattr"))
+                return GF_FOP_FSETXATTR;
+        else if (!strcmp ((*op_no_str), "fgetxattr"))
+                return GF_FOP_FGETXATTR;
+        else if (!strcmp ((*op_no_str), "fremovexattr"))
+                return GF_FOP_FREMOVEXATTR;
         else if (!strcmp ((*op_no_str), "opendir"))
                 return GF_FOP_OPENDIR;
         else if (!strcmp ((*op_no_str), "readdir"))
@@ -362,7 +371,8 @@ error_gen (xlator_t *this, int op_no)
                                 rand_no = 0;
                         ret = error_no_list[op_no].error_no[rand_no];
                 }
-                egp->failure_iter_no = 3 + (rand () % GF_UNIVERSAL_ANSWER);
+                if (egp->random_failure == _gf_true)
+                        egp->failure_iter_no = 3 + (rand () % GF_UNIVERSAL_ANSWER);
         }
         return ret;
 }
@@ -371,17 +381,17 @@ error_gen (xlator_t *this, int op_no)
 int
 error_gen_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		      int32_t op_ret, int32_t op_errno, inode_t *inode,
-		      struct iatt *buf, dict_t *dict, struct iatt *postparent)
+		      struct iatt *buf, dict_t *xdata, struct iatt *postparent)
 {
 	STACK_UNWIND_STRICT (lookup, frame, op_ret, op_errno, inode,
-                             buf, dict, postparent);
-	return 0;
+                             buf, xdata, postparent);
+        return 0;
 }
 
 
 int
 error_gen_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                  dict_t *xattr_req)
+                  dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -396,36 +406,28 @@ error_gen_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
 		STACK_UNWIND_STRICT (lookup, frame, -1, op_errno, NULL, NULL, NULL,
-                              NULL);
-		return 0;
+                                     NULL);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_lookup_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->lookup,
-		    loc, xattr_req);
-	return 0;
-}
-
-
-int
-error_gen_forget (xlator_t *this, inode_t *inode)
-{
-	return 0;
+		    loc, xdata);
+        return 0;
 }
 
 
 int
 error_gen_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, struct iatt *buf)
+		    int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (stat, frame, op_ret, op_errno, buf);
-
-	return 0;
+	STACK_UNWIND_STRICT (stat, frame, op_ret, op_errno, buf, xdata);
+        return 0;
 }
 
 int
-error_gen_stat (call_frame_t *frame, xlator_t *this, loc_t *loc)
+error_gen_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -439,32 +441,31 @@ error_gen_stat (call_frame_t *frame, xlator_t *this, loc_t *loc)
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (stat, frame, -1, op_errno, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (stat, frame, -1, op_errno, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_stat_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->stat,
-		    loc);
-	return 0;
+		    loc, xdata);
+        return 0;
 }
 
 
 int
 error_gen_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                        int32_t op_ret, int32_t op_errno,
-                       struct iatt *preop, struct iatt *postop)
+                       struct iatt *preop, struct iatt *postop, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (setattr, frame, op_ret, op_errno, preop, postop);
-
-	return 0;
+	STACK_UNWIND_STRICT (setattr, frame, op_ret, op_errno, preop, postop, xdata);
+        return 0;
 }
 
 
 int
 error_gen_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                   struct iatt *stbuf, int32_t valid)
+                   struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -478,21 +479,21 @@ error_gen_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (setattr, frame, -1, op_errno, NULL, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (setattr, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_setattr_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->setattr,
-		    loc, stbuf, valid);
-	return 0;
+		    loc, stbuf, valid, xdata);
+        return 0;
 }
 
 
 int
 error_gen_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                    struct iatt *stbuf, int32_t valid)
+                    struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
 	int             op_errno = 0;
         eg_t            *egp = NULL;
@@ -506,32 +507,32 @@ error_gen_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fsetattr, frame, -1, op_errno, NULL, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (fsetattr, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_setattr_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->fsetattr,
-		    fd, stbuf, valid);
-	return 0;
+		    fd, stbuf, valid, xdata);
+        return 0;
 }
 
 
 int
 error_gen_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 			int32_t op_ret, int32_t op_errno,
-			struct iatt *prebuf, struct iatt *postbuf)
+			struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno,
-                             prebuf, postbuf);
-	return 0;
+                             prebuf, postbuf, xdata);
+        return 0;
 }
 
 
 int
 error_gen_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                    off_t offset)
+                    off_t offset, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -546,32 +547,32 @@ error_gen_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc,
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
 		STACK_UNWIND_STRICT (truncate, frame, -1, op_errno,
-                                     NULL, NULL);
-		return 0;
+                                     NULL, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_truncate_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->truncate,
-		    loc, offset);
-	return 0;
+		    loc, offset, xdata);
+        return 0;
 }
 
 
 int
 error_gen_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 			 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                         struct iatt *postbuf)
+                         struct iatt *postbuf, dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (ftruncate, frame, op_ret, op_errno,
-                             prebuf, postbuf);
-	return 0;
+                             prebuf, postbuf, xdata);
+        return 0;
 }
 
 
 int
 error_gen_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd,
-		     off_t offset)
+		     off_t offset, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp =NULL;
@@ -586,31 +587,30 @@ error_gen_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd,
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
 		STACK_UNWIND_STRICT (ftruncate, frame, -1, op_errno,
-                                     NULL, NULL);
-		return 0;
+                                     NULL, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_ftruncate_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->ftruncate,
-		    fd, offset);
-	return 0;
+		    fd, offset, xdata);
+        return 0;
 }
 
 
 int
 error_gen_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno)
+		      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (access, frame, op_ret, op_errno);
-
-	return 0;
+	STACK_UNWIND_STRICT (access, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
 
 int
 error_gen_access (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		  int32_t mask)
+		  int32_t mask, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -624,31 +624,31 @@ error_gen_access (call_frame_t *frame, xlator_t *this, loc_t *loc,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (access, frame, -1, op_errno);
-		return 0;
+		STACK_UNWIND_STRICT (access, frame, -1, op_errno, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_access_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->access,
-		    loc, mask);
-	return 0;
+		    loc, mask, xdata);
+        return 0;
 }
 
 
 int
 error_gen_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 			int32_t op_ret, int32_t op_errno,
-			const char *path, struct iatt *sbuf)
+			const char *path, struct iatt *sbuf, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (readlink, frame, op_ret, op_errno, path, sbuf);
-	return 0;
+	STACK_UNWIND_STRICT (readlink, frame, op_ret, op_errno, path, sbuf, xdata);
+        return 0;
 }
 
 
 int
 error_gen_readlink (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                    size_t size)
+                    size_t size, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -662,15 +662,15 @@ error_gen_readlink (call_frame_t *frame, xlator_t *this, loc_t *loc,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (readlink, frame, -1, op_errno, NULL, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (readlink, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_readlink_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->readlink,
-		    loc, size);
-	return 0;
+		    loc, size, xdata);
+        return 0;
 }
 
 
@@ -678,18 +678,18 @@ int
 error_gen_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		     int32_t op_ret, int32_t op_errno, inode_t *inode,
                      struct iatt *buf, struct iatt *preparent,
-                     struct iatt *postparent)
+                     struct iatt *postparent, dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (mknod, frame, op_ret, op_errno,
                              inode, buf,
-                             preparent, postparent);
-	return 0;
+                             preparent, postparent, xdata);
+        return 0;
 }
 
 
 int
 error_gen_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		 mode_t mode, dev_t rdev, dict_t *params)
+		 mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -704,15 +704,15 @@ error_gen_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc,
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
 		STACK_UNWIND_STRICT (mknod, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
+                                     NULL, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_mknod_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->mknod,
-		    loc, mode, rdev, params);
-	return 0;
+		    loc, mode, rdev, umask, xdata);
+        return 0;
 }
 
 
@@ -720,17 +720,17 @@ int
 error_gen_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		     int32_t op_ret, int32_t op_errno, inode_t *inode,
                      struct iatt *buf, struct iatt *preparent,
-                     struct iatt *postparent)
+                     struct iatt *postparent, dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (mkdir, frame, op_ret, op_errno,
                              inode, buf,
-                             preparent, postparent);
-	return 0;
+                             preparent, postparent, xdata);
+        return 0;
 }
 
 int
 error_gen_mkdir (call_frame_t *frame, xlator_t *this,
-		 loc_t *loc, mode_t mode, dict_t *params)
+		 loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -745,31 +745,32 @@ error_gen_mkdir (call_frame_t *frame, xlator_t *this,
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
 		STACK_UNWIND_STRICT (mkdir, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
+                                     NULL, NULL, xdata);
+                return 0;
 	}
 
 	STACK_WIND (frame, error_gen_mkdir_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->mkdir,
-		    loc, mode, params);
-	return 0;
+		    loc, mode, umask, xdata);
+        return 0;
 }
 
 
 int
 error_gen_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		      int32_t op_ret, int32_t op_errno,
-                      struct iatt *preparent, struct iatt *postparent)
+                      struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (unlink, frame, op_ret, op_errno,
-                             preparent, postparent);
-	return 0;
+                             preparent, postparent, xdata);
+        return 0;
 }
 
 
 int
-error_gen_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
+error_gen_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+                  dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -783,31 +784,34 @@ error_gen_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (unlink, frame, -1, op_errno, NULL, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (unlink, frame, -1, op_errno, NULL, NULL,
+                                     xdata);
+                return 0;
 	}
 
 	STACK_WIND (frame, error_gen_unlink_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->unlink,
-		    loc);
-	return 0;
+		    loc, xflag, xdata);
+        return 0;
 }
 
 
 int
 error_gen_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		     int32_t op_ret, int32_t op_errno,
-                     struct iatt *preparent, struct iatt *postparent)
+                     struct iatt *preparent, struct iatt *postparent,
+                     dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (rmdir, frame, op_ret, op_errno,
-                             preparent, postparent);
-	return 0;
+                             preparent, postparent, xdata);
+        return 0;
 }
 
 
 int
-error_gen_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
+error_gen_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+                 dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -821,15 +825,15 @@ error_gen_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (rmdir, frame, -1, op_errno, NULL, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (rmdir, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_rmdir_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->rmdir,
-		    loc, flags);
-	return 0;
+		    loc, flags, xdata);
+        return 0;
 }
 
 
@@ -837,17 +841,17 @@ int
 error_gen_symlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		       int32_t op_ret, int32_t op_errno, inode_t *inode,
                        struct iatt *buf, struct iatt *preparent,
-                       struct iatt *postparent)
+                       struct iatt *postparent, dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (symlink, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-	return 0;
+                             preparent, postparent, xdata);
+        return 0;
 }
 
 
 int
 error_gen_symlink (call_frame_t *frame, xlator_t *this, const char *linkpath,
-		   loc_t *loc, dict_t *params)
+		   loc_t *loc, mode_t umask, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -862,15 +866,15 @@ error_gen_symlink (call_frame_t *frame, xlator_t *this, const char *linkpath,
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
 		STACK_UNWIND_STRICT (symlink, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL); /* pre & post parent attr */
+                                     NULL, NULL, NULL); /* pre & post parent attr */
 		return 0;
 	}
 
 	STACK_WIND (frame, error_gen_symlink_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->symlink,
-		    linkpath, loc, params);
-	return 0;
+		    linkpath, loc, umask, xdata);
+        return 0;
 }
 
 
@@ -878,18 +882,19 @@ int
 error_gen_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		      int32_t op_ret, int32_t op_errno, struct iatt *buf,
                       struct iatt *preoldparent, struct iatt *postoldparent,
-                      struct iatt *prenewparent, struct iatt *postnewparent)
+                      struct iatt *prenewparent, struct iatt *postnewparent,
+                      dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (rename, frame, op_ret, op_errno, buf,
                              preoldparent, postoldparent,
-                             prenewparent, postnewparent);
-	return 0;
+                             prenewparent, postnewparent, xdata);
+        return 0;
 }
 
 
 int
 error_gen_rename (call_frame_t *frame, xlator_t *this,
-		  loc_t *oldloc, loc_t *newloc)
+		  loc_t *oldloc, loc_t *newloc, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -904,15 +909,15 @@ error_gen_rename (call_frame_t *frame, xlator_t *this,
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
 		STACK_UNWIND_STRICT (rename, frame, -1, op_errno, NULL,
-                                     NULL, NULL, NULL, NULL); /* pre & post parent attr */
+                                     NULL, NULL, NULL, NULL, NULL);
 		return 0;
 	}
 
 	STACK_WIND (frame, error_gen_rename_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->rename,
-		    oldloc, newloc);
-	return 0;
+		    oldloc, newloc, xdata);
+        return 0;
 }
 
 
@@ -920,17 +925,17 @@ int
 error_gen_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		    int32_t op_ret, int32_t op_errno, inode_t *inode,
                     struct iatt *buf, struct iatt *preparent,
-                    struct iatt *postparent)
+                    struct iatt *postparent, dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (link, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-	return 0;
+                             preparent, postparent, xdata);
+        return 0;
 }
 
 
 int
 error_gen_link (call_frame_t *frame, xlator_t *this,
-		loc_t *oldloc, loc_t *newloc)
+		loc_t *oldloc, loc_t *newloc, dict_t *xdata)
 {
 	int             op_errno = 0;
         eg_t            *egp = NULL;
@@ -945,15 +950,15 @@ error_gen_link (call_frame_t *frame, xlator_t *this,
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
 		STACK_UNWIND_STRICT (link, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL); /* pre & post parent attr */
+                                     NULL, NULL, NULL);
 		return 0;
 	}
 
 	STACK_WIND (frame, error_gen_link_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->link,
-		    oldloc, newloc);
-	return 0;
+		    oldloc, newloc, xdata);
+        return 0;
 }
 
 
@@ -961,17 +966,18 @@ int
 error_gen_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		      int32_t op_ret, int32_t op_errno,
 		      fd_t *fd, inode_t *inode, struct iatt *buf,
-                      struct iatt *preparent, struct iatt *postparent)
+                      struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf,
-                             preparent, postparent);
-	return 0;
+                             preparent, postparent, xdata);
+        return 0;
 }
 
 
 int
 error_gen_create (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		  int32_t flags, mode_t mode, fd_t *fd, dict_t *params)
+		  int32_t flags, mode_t mode, mode_t umask, fd_t *fd,
+                  dict_t *xdata)
 {
 	int             op_errno = 0;
         eg_t            *egp = NULL;
@@ -986,30 +992,30 @@ error_gen_create (call_frame_t *frame, xlator_t *this, loc_t *loc,
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
 		STACK_UNWIND_STRICT (create, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL, NULL); /* pre & post attr */
+                                     NULL, NULL, NULL, NULL);
 		return 0;
 	}
 
 	STACK_WIND (frame, error_gen_create_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->create,
-		    loc, flags, mode, fd, params);
-	return 0;
+		    loc, flags, mode, umask, fd, xdata);
+        return 0;
 }
 
 
 int
 error_gen_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, fd_t *fd)
+		    int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd);
-	return 0;
+	STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd, xdata);
+        return 0;
 }
 
 
 int
 error_gen_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		int32_t flags, fd_t *fd, int32_t wbflags)
+		int32_t flags, fd_t *fd, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1023,15 +1029,15 @@ error_gen_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (open, frame, -1, op_errno, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (open, frame, -1, op_errno, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_open_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->open,
-		    loc, flags, fd, wbflags);
-	return 0;
+		    loc, flags, fd, xdata);
+        return 0;
 }
 
 
@@ -1039,17 +1045,17 @@ int
 error_gen_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		     int32_t op_ret, int32_t op_errno,
 		     struct iovec *vector, int32_t count,
-		     struct iatt *stbuf, struct iobref *iobref)
+		     struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
 {
 	STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno,
-                             vector, count, stbuf, iobref);
-	return 0;
+                             vector, count, stbuf, iobref, xdata);
+        return 0;
 }
 
 
 int
 error_gen_readv (call_frame_t *frame, xlator_t *this,
-		 fd_t *fd, size_t size, off_t offset)
+		 fd_t *fd, size_t size, off_t offset, uint32_t flags, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1064,33 +1070,33 @@ error_gen_readv (call_frame_t *frame, xlator_t *this,
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
 		STACK_UNWIND_STRICT (readv, frame, -1, op_errno, NULL, 0,
-                                     NULL, NULL);
-		return 0;
+                                     NULL, NULL, xdata);
+        return 0;
 	}
 
 
 	STACK_WIND (frame, error_gen_readv_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->readv,
-		    fd, size, offset);
-	return 0;
+		    fd, size, offset, flags, xdata);
+        return 0;
 }
 
 
 int
 error_gen_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		      int32_t op_ret, int32_t op_errno,
-                      struct iatt *prebuf, struct iatt *postbuf)
+                      struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
+	STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+        return 0;
 }
 
 
 int
 error_gen_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
 		  struct iovec *vector, int32_t count,
-		  off_t off, struct iobref *iobref)
+		  off_t off, uint32_t flags, struct iobref *iobref, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1102,31 +1108,47 @@ error_gen_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
         if (enable)
                 op_errno = error_gen (this, GF_FOP_WRITE);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (writev, frame, -1, op_errno, NULL, NULL);
+	if (op_errno == GF_ERROR_SHORT_WRITE) {
+		struct iovec *shortvec;
+
+		/*
+		 * A short write error returns some value less than what was
+		 * requested from a write. To simulate this, replace the vector
+		 * with one half the size;
+		 */
+		shortvec = iov_dup(vector, 1);
+		shortvec->iov_len /= 2;
+
+		STACK_WIND(frame, error_gen_writev_cbk, FIRST_CHILD(this),
+			   FIRST_CHILD(this)->fops->writev, fd, shortvec, count,
+			   off, flags, iobref, xdata);
+		GF_FREE(shortvec);
 		return 0;
+	} else if (op_errno) {
+		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
+		STACK_UNWIND_STRICT (writev, frame, -1, op_errno, NULL, NULL, xdata);
+        	return 0;
 	}
 
 	STACK_WIND (frame, error_gen_writev_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->writev,
-		    fd, vector, count, off, iobref);
-	return 0;
+		    fd, vector, count, off, flags, iobref, xdata);
+        return 0;
 }
 
 
 int
 error_gen_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno)
+		     int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno);
-	return 0;
+	STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
 
 int
-error_gen_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
+error_gen_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1140,15 +1162,15 @@ error_gen_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (flush, frame, -1, op_errno);
-		return 0;
+		STACK_UNWIND_STRICT (flush, frame, -1, op_errno, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_flush_cbk,
                     FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->flush,
-		    fd);
-	return 0;
+		    fd, xdata);
+        return 0;
 }
 
 
@@ -1156,15 +1178,15 @@ int
 error_gen_fsync_cbk (call_frame_t *frame, void *cookie,
 		     xlator_t *this, int32_t op_ret,
 		     int32_t op_errno, struct iatt *prebuf,
-                     struct iatt *postbuf)
+                     struct iatt *postbuf, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
+	STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+        return 0;
 }
 
 
 int
-error_gen_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
+error_gen_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1178,29 +1200,29 @@ error_gen_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fsync, frame, -1, op_errno, NULL, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (fsync, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_fsync_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->fsync,
-		    fd, flags);
-	return 0;
+		    fd, flags, xdata);
+        return 0;
 }
 
 
 int
 error_gen_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *buf)
+                     int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fstat, frame, op_ret, op_errno, buf);
-	return 0;
+	STACK_UNWIND_STRICT (fstat, frame, op_ret, op_errno, buf, xdata);
+        return 0;
 }
 
 
 int
-error_gen_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd)
+error_gen_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1214,29 +1236,29 @@ error_gen_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd)
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fstat, frame, -1, op_errno, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (fstat, frame, -1, op_errno, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_fstat_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->fstat,
-		    fd);
-	return 0;
+		    fd, xdata);
+        return 0;
 }
 
 
 int
 error_gen_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, fd_t *fd)
+		       int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (opendir, frame, op_ret, op_errno, fd);
-	return 0;
+	STACK_UNWIND_STRICT (opendir, frame, op_ret, op_errno, fd, xdata);
+        return 0;
 }
 
 
 int
-error_gen_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
+error_gen_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1250,29 +1272,29 @@ error_gen_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (opendir, frame, -1, op_errno, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (opendir, frame, -1, op_errno, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_opendir_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->opendir,
-		    loc, fd);
-	return 0;
+		    loc, fd, xdata);
+        return 0;
 }
 
 int
 error_gen_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno)
+			int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fsyncdir, frame, op_ret, op_errno);
-	return 0;
+	STACK_UNWIND_STRICT (fsyncdir, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
 
 int
 error_gen_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                    int32_t flags)
+                    int32_t flags, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1286,30 +1308,29 @@ error_gen_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fsyncdir, frame, -1, op_errno);
-		return 0;
+		STACK_UNWIND_STRICT (fsyncdir, frame, -1, op_errno, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_fsyncdir_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->fsyncdir,
-		    fd, flags);
-	return 0;
+		    fd, flags, xdata);
+        return 0;
 }
 
 
 int
 error_gen_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, struct statvfs *buf)
+		      int32_t op_ret, int32_t op_errno, struct statvfs *buf, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (statfs, frame, op_ret, op_errno, buf);
-
-	return 0;
+	STACK_UNWIND_STRICT (statfs, frame, op_ret, op_errno, buf, xdata);
+        return 0;
 }
 
 
 int
-error_gen_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
+error_gen_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1323,31 +1344,30 @@ error_gen_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (statfs, frame, -1, op_errno, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (statfs, frame, -1, op_errno, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_statfs_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->statfs,
-		    loc);
-	return 0;
+		    loc, xdata);
+        return 0;
 }
 
 
 int
 error_gen_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno)
+			int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (setxattr, frame, op_ret, op_errno);
-
-	return 0;
+	STACK_UNWIND_STRICT (setxattr, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
 
 int
 error_gen_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		    dict_t *dict, int32_t flags)
+		    dict_t *dict, int32_t flags, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1361,30 +1381,30 @@ error_gen_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (setxattr, frame, -1, op_errno);
-		return 0;
+		STACK_UNWIND_STRICT (setxattr, frame, -1, op_errno, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_setxattr_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->setxattr,
-		    loc, dict, flags);
-	return 0;
+		    loc, dict, flags, xdata);
+        return 0;
 }
 
 
 int
 error_gen_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno, dict_t *dict)
+			int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (getxattr, frame, op_ret, op_errno, dict);
-	return 0;
+	STACK_UNWIND_STRICT (getxattr, frame, op_ret, op_errno, dict, xdata);
+        return 0;
 }
 
 
 int
 error_gen_getxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		    const char *name)
+		    const char *name, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1398,31 +1418,103 @@ error_gen_getxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (getxattr, frame, -1, op_errno, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (getxattr, frame, -1, op_errno, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_getxattr_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->getxattr,
-		    loc, name);
-	return 0;
+		    loc, name, xdata);
+        return 0;
+}
+
+int
+error_gen_fsetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+	STACK_UNWIND_STRICT (fsetxattr, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
 
 int
-error_gen_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *dict)
+error_gen_fsetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                     dict_t *dict, int32_t flags, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (xattrop, frame, op_ret, op_errno, dict);
+	int              op_errno = 0;
+        eg_t            *egp = NULL;
+        int              enable = 1;
 
-	return 0;
+        egp = this->private;
+        enable = egp->enable[GF_FOP_FSETXATTR];
+
+        if (enable)
+                op_errno = error_gen (this, GF_FOP_FSETXATTR);
+
+	if (op_errno) {
+		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
+		STACK_UNWIND_STRICT (fsetxattr, frame, -1, op_errno, xdata);
+        return 0;
+	}
+
+	STACK_WIND (frame, error_gen_fsetxattr_cbk,
+		    FIRST_CHILD(this),
+		    FIRST_CHILD(this)->fops->fsetxattr,
+		    fd, dict, flags, xdata);
+        return 0;
+}
+
+
+int
+error_gen_fgetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+	STACK_UNWIND_STRICT (fgetxattr, frame, op_ret, op_errno, dict, xdata);
+        return 0;
+}
+
+
+int
+error_gen_fgetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                     const char *name, dict_t *xdata)
+{
+	int              op_errno = 0;
+        eg_t            *egp = NULL;
+        int              enable = 1;
+
+        egp = this->private;
+        enable = egp->enable[GF_FOP_FGETXATTR];
+
+        if (enable)
+                op_errno = error_gen (this, GF_FOP_FGETXATTR);
+
+	if (op_errno) {
+		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
+		STACK_UNWIND_STRICT (fgetxattr, frame, -1, op_errno, NULL, xdata);
+        return 0;
+	}
+
+	STACK_WIND (frame, error_gen_fgetxattr_cbk,
+		    FIRST_CHILD(this),
+		    FIRST_CHILD(this)->fops->fgetxattr,
+		    fd, name, xdata);
+        return 0;
+}
+
+
+int
+error_gen_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+	STACK_UNWIND_STRICT (xattrop, frame, op_ret, op_errno, dict, xdata);
+        return 0;
 }
 
 
 int
 error_gen_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		   gf_xattrop_flags_t flags, dict_t *dict)
+		   gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1436,31 +1528,30 @@ error_gen_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (xattrop, frame, -1, op_errno, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (xattrop, frame, -1, op_errno, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_xattrop_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->xattrop,
-		    loc, flags, dict);
-	return 0;
+		    loc, flags, dict, xdata);
+        return 0;
 }
 
 
 int
 error_gen_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno, dict_t *dict)
+			int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fxattrop, frame, op_ret, op_errno, dict);
-
-	return 0;
+	STACK_UNWIND_STRICT (fxattrop, frame, op_ret, op_errno, dict, xdata);
+        return 0;
 }
 
 
 int
 error_gen_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
-		    gf_xattrop_flags_t flags, dict_t *dict)
+		    gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1474,31 +1565,30 @@ error_gen_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fxattrop, frame, -1, op_errno, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (fxattrop, frame, -1, op_errno, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_fxattrop_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->fxattrop,
-		    fd, flags, dict);
-	return 0;
+		    fd, flags, dict, xdata);
+        return 0;
 }
 
 
 int
 error_gen_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno)
+			   int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (removexattr, frame, op_ret, op_errno);
-
-	return 0;
+	STACK_UNWIND_STRICT (removexattr, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
 
 int
 error_gen_removexattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		       const char *name)
+		       const char *name, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1512,30 +1602,66 @@ error_gen_removexattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (removexattr, frame, -1, op_errno);
-		return 0;
+		STACK_UNWIND_STRICT (removexattr, frame, -1, op_errno, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_removexattr_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->removexattr,
-		    loc, name);
-	return 0;
+		    loc, name, xdata);
+        return 0;
+}
+
+int
+error_gen_fremovexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+			   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+	STACK_UNWIND_STRICT (fremovexattr, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+
+int
+error_gen_fremovexattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                        const char *name, dict_t *xdata)
+{
+	int              op_errno = 0;
+        eg_t            *egp = NULL;
+        int              enable = 1;
+
+        egp = this->private;
+        enable = egp->enable[GF_FOP_FREMOVEXATTR];
+
+        if (enable)
+                op_errno = error_gen (this, GF_FOP_FREMOVEXATTR);
+
+	if (op_errno) {
+		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
+		STACK_UNWIND_STRICT (fremovexattr, frame, -1, op_errno, xdata);
+        return 0;
+	}
+
+	STACK_WIND (frame, error_gen_fremovexattr_cbk,
+		    FIRST_CHILD(this),
+		    FIRST_CHILD(this)->fops->fremovexattr,
+		    fd, name, xdata);
+        return 0;
 }
 
 
 int
 error_gen_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int32_t op_ret, int32_t op_errno, struct gf_flock *lock)
+		  int32_t op_ret, int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, lock);
-	return 0;
+	STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, lock, xdata);
+        return 0;
 }
 
 
 int
 error_gen_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
-	      struct gf_flock *lock)
+	      struct gf_flock *lock, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1549,32 +1675,31 @@ error_gen_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (lk, frame, -1, op_errno, NULL);
-		return 0;
+		STACK_UNWIND_STRICT (lk, frame, -1, op_errno, NULL, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_lk_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->lk,
-		    fd, cmd, lock);
-	return 0;
+		    fd, cmd, lock, xdata);
+        return 0;
 }
 
 
 int
-error_gen_inodelk_cbk (call_frame_t *frame, void *cookie,
-		       xlator_t *this, int32_t op_ret, int32_t op_errno)
-
+error_gen_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno);
-	return 0;
+	STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
 
 int
 error_gen_inodelk (call_frame_t *frame, xlator_t *this,
 		   const char *volume, loc_t *loc, int32_t cmd,
-                   struct gf_flock *lock)
+                   struct gf_flock *lock, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1588,32 +1713,31 @@ error_gen_inodelk (call_frame_t *frame, xlator_t *this,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (inodelk, frame, -1, op_errno);
-		return 0;
+		STACK_UNWIND_STRICT (inodelk, frame, -1, op_errno, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_inodelk_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->inodelk,
-		    volume, loc, cmd, lock);
-	return 0;
+		    volume, loc, cmd, lock, xdata);
+        return 0;
 }
 
 
 int
-error_gen_finodelk_cbk (call_frame_t *frame, void *cookie,
-			xlator_t *this, int32_t op_ret, int32_t op_errno)
-
+error_gen_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (finodelk, frame, op_ret, op_errno);
-	return 0;
+	STACK_UNWIND_STRICT (finodelk, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
 
 int
 error_gen_finodelk (call_frame_t *frame, xlator_t *this,
 		    const char *volume, fd_t *fd, int32_t cmd,
-                    struct gf_flock *lock)
+                    struct gf_flock *lock, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1627,32 +1751,31 @@ error_gen_finodelk (call_frame_t *frame, xlator_t *this,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (finodelk, frame, -1, op_errno);
-		return 0;
+		STACK_UNWIND_STRICT (finodelk, frame, -1, op_errno, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_finodelk_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->finodelk,
-		    volume, fd, cmd, lock);
-	return 0;
+		    volume, fd, cmd, lock, xdata);
+        return 0;
 }
 
 
 int
-error_gen_entrylk_cbk (call_frame_t *frame, void *cookie,
-		       xlator_t *this, int32_t op_ret, int32_t op_errno)
-
+error_gen_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno);
-	return 0;
+	STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
 
 int
 error_gen_entrylk (call_frame_t *frame, xlator_t *this,
 		   const char *volume, loc_t *loc, const char *basename,
-		   entrylk_cmd cmd, entrylk_type type)
+		   entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
 {
 	int              op_errno = 0;
         eg_t            *egp = NULL;
@@ -1666,32 +1789,31 @@ error_gen_entrylk (call_frame_t *frame, xlator_t *this,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (entrylk, frame, -1, op_errno);
-		return 0;
+		STACK_UNWIND_STRICT (entrylk, frame, -1, op_errno, xdata);
+        return 0;
 	}
 
 	STACK_WIND (frame, error_gen_entrylk_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->entrylk,
-		    volume, loc, basename, cmd, type);
-	return 0;
+		    volume, loc, basename, cmd, type, xdata);
+        return 0;
 }
 
 
 int
-error_gen_fentrylk_cbk (call_frame_t *frame, void *cookie,
-			xlator_t *this, int32_t op_ret, int32_t op_errno)
-
+error_gen_fentrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fentrylk, frame, op_ret, op_errno);
-	return 0;
+	STACK_UNWIND_STRICT (fentrylk, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
 
 int
 error_gen_fentrylk (call_frame_t *frame, xlator_t *this,
 		    const char *volume, fd_t *fd, const char *basename,
-		    entrylk_cmd cmd, entrylk_type type)
+		    entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
 {
 	int             op_errno = 0;
         eg_t            *egp = NULL;
@@ -1705,15 +1827,15 @@ error_gen_fentrylk (call_frame_t *frame, xlator_t *this,
 
 	if (op_errno) {
 		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fentrylk, frame, -1, op_errno);
-		return 0;
+		STACK_UNWIND_STRICT (fentrylk, frame, -1, op_errno, xdata);
+                return 0;
 	}
 
 	STACK_WIND (frame, error_gen_fentrylk_cbk,
 		    FIRST_CHILD(this),
 		    FIRST_CHILD(this)->fops->fentrylk,
-		    volume, fd, basename, cmd, type);
-	return 0;
+		    volume, fd, basename, cmd, type, xdata);
+        return 0;
 }
 
 
@@ -1725,8 +1847,7 @@ error_gen_getspec_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		       int32_t op_ret, int32_t op_errno, char *spec_data)
 {
 	STACK_UNWIND_STRICT (getspec, frame, op_ret, op_errno, spec_data);
-
-	return 0;
+        return 0;
 }
 
 
@@ -1760,16 +1881,17 @@ error_gen_getspec (call_frame_t *frame, xlator_t *this, const char *key,
 
 int
 error_gen_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, gf_dirent_t *entries)
+		       int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                       dict_t *xdata)</

path: root/cli/src/cli-cmd.c

/*
   Copyright (c) 2010-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.

   This file is licensed to you under your choice of the GNU Lesser
   General Public License, version 3 or any later version (LGPLv3 or
   later), or the GNU General Public License, version 2 (GPLv2), in all
   cases as published by the Free Software Foundation.
*/
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <stdint.h>
#include <pthread.h>

#include "cli.h"
#include "cli-cmd.h"
#include "cli-mem-types.h"
#include "protocol-common.h"

#include <fnmatch.h>

static int cmd_done;
static int cmd_sent;
static pthread_cond_t cond = PTHREAD_COND_INITIALIZER;
static pthread_mutex_t cond_mutex = PTHREAD_MUTEX_INITIALIZER;
static pthread_cond_t conn = PTHREAD_COND_INITIALIZER;
static pthread_mutex_t conn_mutex = PTHREAD_MUTEX_INITIALIZER;

int cli_op_ret = 0;
int connected = 0;

int
cli_cmd_log_help_cbk(struct cli_state *state, struct cli_cmd_word *in_word,
                     const char **words, int wordcount);

static unsigned
cli_cmd_needs_connection(struct cli_cmd_word *word)
{
    if (!strcasecmp("quit", word->word))
        return 0;

    if (!strcasecmp("help", word->word))
        return 0;

    if (!strcasecmp("getwd", word->word))
        return 1;

    if (!strcasecmp("exit", word->word))
        return 0;

    return cli_default_conn_timeout;
}

int
cli_cmd_status_reset(void)
{
    int ret = 0;

    ret = cli_cmd_lock();
    {
        if (ret == 0) {
            cmd_sent = 0;
            cmd_done = 0;
        }
    }
    ret = cli_cmd_unlock();
    return ret;
}

int
cli_cmd_sent_status_get(int *status)
{
    int ret = 0;
    GF_ASSERT(status);

    ret = cli_cmd_lock();
    {
        if (ret == 0)
            *status = cmd_sent;
    }
    ret = cli_cmd_unlock();
    return ret;
}

int
cli_cmd_process(struct cli_state *state, int argc, char **argv)
{
    int ret = 0;
    struct cli_cmd_word *word = NULL;
    struct cli_cmd_word *next = NULL;
    int i = 0;

    word = &state->tree.root;

    if (!argc)
        return 0;

    for (i = 0; i < argc; i++) {
        next = cli_cmd_nextword(word, argv[i]);

        word = next;
        if (!word)
            break;

        if (word->cbkfn)
            break;
    }

    if (!word) {
        cli_out("unrecognized word: %s (position %d)\n", argv[i], i);
        usage();
        return -1;
    }

    if (!word->cbkfn) {
        cli_out("unrecognized command\n");
        usage();
        return -1;
    }

    if (strcmp(word->word, "help") == 0)
        goto callback;

    state->await_connected = cli_cmd_needs_connection(word);

    ret = cli_cmd_await_connected(state->await_connected);
    if (ret) {
        cli_out(
            "Connection failed. Please check if gluster "
            "daemon is operational.");
        gf_log("", GF_LOG_INFO, "Exiting with: %d", ret);
        exit(ret);
    }

callback:
    ret = word->cbkfn(state, word, (const char **)argv, argc);
    (void)cli_cmd_status_reset();
    return ret;
}

int
cli_cmd_input_token_count(const char *text)
{
    int count = 0;
    const char *trav = NULL;
    int is_spc = 1;

    for (trav = text; *trav; trav++) {
        if (*trav == ' ') {
            is_spc = 1;
        } else {
            if (is_spc) {
                count++;
                is_spc = 0;
            }
        }
    }

    return count;
}

int
cli_cmd_process_line(struct cli_state *state, const char *text)
{
    int count = 0;
    char **tokens = NULL;
    char **tokenp = NULL;
    char *token = NULL;
    char *copy = NULL;
    char *saveptr = NULL;
    int i = 0;
    int ret = -1;

    count = cli_cmd_input_token_count(text);

    tokens = calloc(count + 1, sizeof(*tokens));
    if (!tokens)
        return -1;

    copy = strdup(text);
    if (!copy)
        goto out;

    tokenp = tokens;

    for (token = strtok_r(copy, " \t\r\n", &saveptr); token;
         token = strtok_r(NULL, " \t\r\n", &saveptr)) {
        *tokenp = strdup(token);

        if (!*tokenp)
            goto out;
        tokenp++;
        i++;
    }

    ret = cli_cmd_process(state, count, tokens);
out:
    free(copy);

    if (tokens)
        cli_cmd_tokens_destroy(tokens);

    return ret;
}

int
cli_cmds_register(struct cli_state *state)
{
    int ret = 0;

    ret = cli_cmd_volume_register(state);
    if (ret)
        goto out;

    ret = cli_cmd_probe_register(state);
    if (ret)
        goto out;

    ret = cli_cmd_system_register(state);
    if (ret)
        goto out;

    ret = cli_cmd_misc_register(state);
    if (ret)
        goto out;

    ret = cli_cmd_snapshot_register(state);
    if (ret)
        goto out;
    ret = cli_cmd_global_register(state);
    if (ret)
        goto out;
out:
    return ret;
}

int
cli_cmd_cond_init()
{
    pthread_mutex_init(&cond_mutex, NULL);
    pthread_cond_init(&cond, NULL);

    pthread_mutex_init(&conn_mutex, NULL);
    pthread_cond_init(&conn, NULL);

    return 0;
}

int
cli_cmd_lock()
{
    pthread_mutex_lock(&cond_mutex);
    return 0;
}

int
cli_cmd_unlock()
{
    pthread_mutex_unlock(&cond_mutex);
    return 0;
}

static void
seconds_from_now(unsigned secs, struct timespec *ts)
{
    struct timeval tv = {
        0,
    };

    gettimeofday(&tv, NULL);

    ts->tv_sec = tv.tv_sec + secs;
    ts->tv_nsec = tv.tv_usec * 1000;
}

int
cli_cmd_await_response(unsigned time)
{
    struct timespec ts = {
        0,
    };
    int ret = 0;

    cli_op_ret = -1;

    seconds_from_now(time, &ts);
    while (!cmd_done && !ret) {
        ret = pthread_cond_timedwait(&cond, &cond_mutex, &ts);
    }

    if (!cmd_done) {
        if (ret == ETIMEDOUT)
            cli_out("Error : Request timed out");
        else
            cli_out("Error : Command returned with error code:%d", ret);
    }
    cmd_done = 0;

    return cli_op_ret;
}

/* This function must be called _only_ after all actions associated with
 * command processing is complete. Otherwise, gluster process may exit before
 * reporting results to stdout/stderr. */
int
cli_cmd_broadcast_response(int32_t status)
{
    pthread_mutex_lock(&cond_mutex);
    {
        if (!cmd_sent)
            goto out;
        cmd_done = 1;
        cli_op_ret = status;
        pthread_cond_broadcast(&cond);
    }

out:
    pthread_mutex_unlock(&cond_mutex);
    return 0;
}

int32_t
cli_cmd_await_connected(unsigned conn_timo)
{
    int32_t ret = 0;
    struct timespec ts = {
        0,
    };

    if (!conn_timo)
        return 0;

    pthread_mutex_lock(&conn_mutex);
    {
        seconds_from_now(conn_timo, &ts);
        while (!connected && !ret) {
            ret = pthread_cond_timedwait(&conn, &conn_mutex, &ts);
        }
    }
    pthread_mutex_unlock(&conn_mutex);

    return ret;
}

int32_t
cli_cmd_broadcast_connected()
{
    pthread_mutex_lock(&conn_mutex);
    {
        connected = 1;
        pthread_cond_broadcast(&conn);
    }

    pthread_mutex_unlock(&conn_mutex);

    return 0;
}

int
cli_cmd_submit(struct rpc_clnt *rpc, void *req, call_frame_t *frame,
               rpc_clnt_prog_t *prog, int procnum, struct iobref *iobref,
               xlator_t *this, fop_cbk_fn_t cbkfn, xdrproc_t xdrproc)
{
    int ret = -1;
    unsigned timeout = 0;

    if ((GLUSTER_CLI_PROFILE_VOLUME == procnum) ||
        (GLUSTER_CLI_HEAL_VOLUME == procnum))
        timeout = cli_ten_minutes_timeout;
    else
        timeout = cli_default_conn_timeout;

    cli_cmd_lock();
    cmd_sent = 0;
    ret = cli_submit_request(rpc, req, frame, prog, procnum, NULL, this, cbkfn,
                             xdrproc);

    if (!ret) {
        cmd_sent = 1;
        ret = cli_cmd_await_response(timeout);
    }

    cli_cmd_unlock();

    gf_log("cli", GF_LOG_DEBUG, "Returning %d", ret);
    return ret;
}

int
cli_cmd_pattern_cmp(void *a, void *b)
{
    struct cli_cmd *ia = NULL;
    struct cli_cmd *ib = NULL;
    int ret = 0;

    ia = a;
    ib = b;
    if (strcmp(ia->pattern, ib->pattern) > 0)
        ret = 1;
    else if (strcmp(ia->pattern, ib->pattern) < 0)
        ret = -1;
    else
        ret = 0;
    return ret;
}

void
cli_cmd_sort(struct cli_cmd *cmd, int count)
{
    gf_array_insertionsort(cmd, 1, count - 2, sizeof(struct cli_cmd),
                           cli_cmd_pattern_cmp);
}