249 files changed, 56393 insertions, 14819 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/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index a9acb4094..af01f2ef2 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -51,7 +51,7 @@
 
 #define AFR_ICTX_OPENDIR_DONE_MASK     0x0000000100000000ULL
 #define AFR_ICTX_READ_CHILD_MASK       0x00000000FFFFFFFFULL
-
+#define AFR_STATISTICS_HISTORY_SIZE    50
 int
 afr_lookup_done_success_action (call_frame_t *frame, xlator_t *this,
                                 gf_boolean_t fail_conflict);
@@ -779,6 +779,12 @@ afr_local_sh_cleanup (afr_local_t *local, xlator_t *this)
         sh = &local->self_heal;
         priv = this->private;
 
+        if (sh->data_sh_info && strcmp (sh->data_sh_info, ""))
+                GF_FREE (sh->data_sh_info);
+
+        if (sh->metadata_sh_info && strcmp (sh->metadata_sh_info, ""))
+                GF_FREE (sh->metadata_sh_info);
+
         GF_FREE (sh->buf);
 
         GF_FREE (sh->parentbufs);
@@ -826,7 +832,8 @@ afr_local_sh_cleanup (afr_local_t *local, xlator_t *this)
 void
 afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this)
 {
-        afr_private_t * priv = NULL;
+        afr_private_t *priv    = NULL;
+        int           i        = 0;
 
         priv = this->private;
 
@@ -836,7 +843,9 @@ afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this)
 
         GF_FREE (local->internal_lock.locked_nodes);
 
-        GF_FREE (local->internal_lock.inode_locked_nodes);
+        for (i = 0; local->internal_lock.inodelk[i].domain; i++) {
+                GF_FREE (local->internal_lock.inodelk[i].locked_nodes);
+        }
 
         GF_FREE (local->internal_lock.lower_locked_nodes);
 
@@ -1071,6 +1080,88 @@ afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *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 (afr_local_t *local, xlator_t *this,
+                       dict_t *rsp_dict)
+{
+        int32_t       *sources       = NULL;
+        dict_t        *xattr         = NULL;
+        data_t        *max_data      = NULL;
+        int64_t       max_quota_size = -1;
+        data_t        *data          = NULL;
+        int64_t       *size          = NULL;
+        int64_t       quota_size     = -1;
+        afr_private_t *priv          = NULL;
+        int           i              = 0;
+        int           ret            = -1;
+        gf_boolean_t  source_present = _gf_false;
+
+        priv    = this->private;
+        sources = local->cont.lookup.sources;
+
+        if (rsp_dict == NULL) {
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "%s: Invalid "
+                                  "response dictionary", local->loc.path);
+                return;
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (sources[i]) {
+                        source_present = _gf_true;
+                        break;
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                /*
+                 * If there is at least one source lets check
+                 * for maximum quota sizes among sources, otherwise take the
+                 * maximum of the ones present to be on the safer side.
+                 */
+                if (source_present && !sources[i])
+                        continue;
+
+                xattr = local->cont.lookup.xattrs[i];
+                if (!xattr)
+                        continue;
+
+                data = dict_get (xattr, QUOTA_SIZE_KEY);
+                if (!data)
+                        continue;
+
+                size = (int64_t*)data->data;
+                quota_size = ntoh64(*size);
+                gf_log (this->name, GF_LOG_DEBUG, "%s: %d, size: %"PRId64,
+                        local->loc.path, i, quota_size);
+                if (quota_size > max_quota_size) {
+                        if (max_data)
+                                data_unref (max_data);
+
+                        max_quota_size = quota_size;
+                        max_data = data_ref (data);
+                }
+        }
+
+        if (max_data) {
+                ret = dict_set (rsp_dict, QUOTA_SIZE_KEY, max_data);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR, "%s: Failed to set "
+                                "quota size", local->loc.path);
+                }
+
+                data_unref (max_data);
+        }
+}
+
 int
 afr_lookup_build_response_params (afr_local_t *local, xlator_t *this)
 {
@@ -1115,13 +1206,18 @@ afr_lookup_build_response_params (afr_local_t *local, xlator_t *this)
                 ret = -1;
                 goto out;
         }
+
         gf_log (this->name, GF_LOG_DEBUG, "Building lookup response from %d",
                 read_child);
         if (!*xattr)
                 *xattr = dict_ref (local->cont.lookup.xattrs[read_child]);
+
         *buf = local->cont.lookup.bufs[read_child];
         *postparent = local->cont.lookup.postparents[read_child];
 
+        if (dict_get (local->xattr_req, QUOTA_SIZE_KEY))
+                afr_handle_quota_size (local, this, *xattr);
+
         if (IA_INVAL == local->cont.lookup.inode->ia_type) {
                 /* fix for RT #602 */
                 local->cont.lookup.inode->ia_type = buf->ia_type;
@@ -1270,7 +1366,7 @@ afr_detect_self_heal_by_iatt (afr_local_t *local, xlator_t *this,
 
         if (uuid_compare (buf->ia_gfid, lookup_buf->ia_gfid)) {
                 /* mismatching gfid */
-                gf_log (this->name, GF_LOG_WARNING,
+                gf_log (this->name, GF_LOG_DEBUG,
                         "%s: gfid different on subvolume", local->loc.path);
         }
 }
@@ -1494,7 +1590,7 @@ afr_conflicting_iattrs (struct iatt *bufs, int32_t *success_children,
 
                 child2 = &bufs[success_children[i-1]];
                 if (FILETYPE_DIFFERS (child1, child2)) {
-                        gf_log (xlator_name, GF_LOG_WARNING, "%s: filetype "
+                        gf_log (xlator_name, GF_LOG_DEBUG, "%s: filetype "
                                 "differs on subvolumes (%d, %d)", path,
                                 success_children[i-1], success_children[i]);
                         conflicting = _gf_true;
@@ -1503,7 +1599,7 @@ afr_conflicting_iattrs (struct iatt *bufs, int32_t *success_children,
                 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"
+                       gf_log (xlator_name, GF_LOG_DEBUG, "%s: gfid differs"
                                " on subvolume %d", path, success_children[i]);
                        conflicting = _gf_true;
                        goto out;
@@ -1632,7 +1728,6 @@ afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this,
         afr_local_t *local = NULL;
         int         ret    = -1;
         dict_t      *xattr = NULL;
-        int32_t     spb    = 0;
 
         local = frame->local;
 
@@ -1663,10 +1758,6 @@ afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this,
                                        local->loc.path,
                                        local->self_heal.actual_sh_started);
                 }
-
-                if (local->loc.inode)
-                        spb = afr_is_split_brain (this, local->loc.inode);
-                ret = dict_set_int32 (xattr, "split-brain", spb);
         }
 out:
         AFR_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
@@ -1741,7 +1832,8 @@ afr_lookup_perform_self_heal (call_frame_t *frame, xlator_t *this,
 
         afr_lookup_set_self_heal_params (local, this);
         if (afr_can_self_heal_proceed (&local->self_heal, priv)) {
-                if  (afr_is_transaction_running (local))
+                if  (afr_is_transaction_running (local) &&
+                     (!local->allow_sh_for_running_transaction))
                         goto out;
 
                 reason = "lookup detected pending operations";
@@ -2336,7 +2428,7 @@ afr_lookup (call_frame_t *frame, xlator_t *this,
         int               call_count     = 0;
         uint64_t          ctx            = 0;
         int32_t           op_errno       = 0;
-
+        int               allow_sh       = 0;
         priv = this->private;
 
         AFR_LOCAL_ALLOC_OR_GOTO (local, out);
@@ -2353,6 +2445,13 @@ afr_lookup (call_frame_t *frame, xlator_t *this,
                 goto out;
         }
 
+        if (local->loc.path &&
+            (strcmp (local->loc.path, "/" GF_REPLICATE_TRASH_DIR) == 0)) {
+                op_errno = EPERM;
+                ret = -1;
+                goto out;
+        }
+
         ret = inode_ctx_get (local->loc.inode, this, &ctx);
         if (ret == 0) {
                 /* lookup is a revalidate */
@@ -2401,6 +2500,11 @@ afr_lookup (call_frame_t *frame, xlator_t *this,
         /* By default assume ENOTCONN. On success it will be set to 0. */
         local->op_errno = ENOTCONN;
 
+        ret = dict_get_int32 (xattr_req, "allow-sh-for-running-transaction",
+                              &allow_sh);
+        dict_del (xattr_req, "allow-sh-for-running-transaction");
+        local->allow_sh_for_running_transaction = allow_sh;
+
         ret = afr_lookup_xattr_req_prepare (local, this, xattr_req, &local->loc,
                                             &gfid_req);
         if (ret) {
@@ -2578,15 +2682,42 @@ afr_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
         return 0;
 }
 
+static int
+afr_flush_wrapper (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+        int           i      = 0;
+        afr_local_t   *local = NULL;
+        afr_private_t *priv  = NULL;
+        int call_count       = -1;
+
+        priv = this->private;
+        local = frame->local;
+        call_count = local->call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND_COOKIE (frame, afr_flush_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->flush,
+                                           local->fd, NULL);
+                        if (!--call_count)
+                                break;
+
+                }
+        }
+
+        return 0;
+}
+
 int
 afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
         afr_private_t *priv  = NULL;
         afr_local_t   *local = NULL;
+        call_stub_t   *stub = NULL;
         int            ret        = -1;
         int            op_errno   = 0;
-	int	       call_count = -1;
-	int	       i = 0;
 
         VALIDATE_OR_GOTO (frame, out);
         VALIDATE_OR_GOTO (this, out);
@@ -2602,23 +2733,14 @@ afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 		goto out;
 
 	local->fd = fd_ref(fd);
-        call_count = local->call_count;
-
-	afr_delayed_changelog_wake_up (this, fd);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_flush_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->flush,
-                                           local->fd, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
+        stub = fop_flush_stub (frame, afr_flush_wrapper, fd, xdata);
+        if (!stub) {
+                ret = -1;
+                op_errno = ENOMEM;
+                goto out;
         }
 
+        afr_delayed_changelog_wake_resume (this, fd, stub);
 	ret = 0;
 
 out:
@@ -2694,6 +2816,16 @@ 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, dict_t *xdata)
@@ -2718,13 +2850,13 @@ afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         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;
+                                local->cont.inode_wfop.prebuf  = *prebuf;
+                                local->cont.inode_wfop.postbuf = *postbuf;
                         }
 
                         local->success_count++;
@@ -2746,10 +2878,11 @@ afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		   post-op. This guarantee is expected by FUSE graph switching
 		   for example.
 		*/
-		stub = fop_fsync_cbk_stub (frame, default_fsync_cbk,
+		stub = fop_fsync_cbk_stub (frame, afr_fsync_unwind_cbk,
                                            local->op_ret, local->op_errno,
-                                           &local->cont.fsync.prebuf,
-                                           &local->cont.fsync.postbuf, xdata);
+                                           &local->cont.inode_wfop.prebuf,
+                                           &local->cont.inode_wfop.postbuf,
+                                           xdata);
 		if (!stub) {
 			AFR_STACK_UNWIND (fsync, frame, -1, ENOMEM, 0, 0, 0);
 			return 0;
@@ -3966,6 +4099,8 @@ afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno)
                 goto out;
         }
 
+	local->append_write = _gf_false;
+
         ret = 0;
 out:
         return ret;
@@ -3977,11 +4112,6 @@ 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->locked_nodes = GF_CALLOC (sizeof (*lk->locked_nodes),
                                       child_count, gf_afr_mt_char);
         if (NULL == lk->locked_nodes)
@@ -4040,6 +4170,21 @@ out:
 }
 
 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, xlator_t *this)
 {
         int            child_up_count = 0;
@@ -4052,6 +4197,14 @@ afr_transaction_local_init (afr_local_t *local, xlator_t *this)
         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 (local->child_up,
                                                 priv->child_count);
@@ -4278,6 +4431,16 @@ afr_priv_destroy (afr_private_t *priv)
         if (priv->shd.split_brain)
                 eh_destroy (priv->shd.split_brain);
 
+        for (i = 0; i < priv->child_count; i++)
+        {
+                if (priv->shd.statistics[i])
+                        eh_destroy (priv->shd.statistics[i]);
+        }
+
+        GF_FREE (priv->shd.statistics);
+
+        GF_FREE (priv->shd.crawl_events);
+
         GF_FREE (priv->last_event);
         if (priv->pending_key) {
                 for (i = 0; i < priv->child_count; i++)
@@ -4356,3 +4519,73 @@ afr_is_fd_fixable (fd_t *fd)
 
         return _gf_true;
 }
+
+void
+afr_handle_open_fd_count (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        inode_t         *inode = NULL;
+        afr_inode_ctx_t *ctx   = NULL;
+
+        local = frame->local;
+
+        if (local->fd)
+                inode = local->fd->inode;
+        else
+                inode = local->loc.inode;
+
+        if (!inode)
+                return;
+
+        LOCK (&inode->lock);
+        {
+                ctx = __afr_inode_ctx_get (inode, this);
+                ctx->open_fd_count = local->open_fd_count;
+        }
+        UNLOCK (&inode->lock);
+}
+
+int
+afr_initialise_statistics (xlator_t *this)
+{
+        afr_private_t       *priv = NULL;
+        int                 ret = -1;
+        int                 i = 0;
+        int                 child_count = 0;
+        eh_t                *stats_per_brick = NULL;
+        shd_crawl_event_t   ***shd_crawl_events = NULL;
+        priv = this->private;
+
+        priv->shd.statistics = GF_CALLOC (sizeof(eh_t *), priv->child_count,
+                                          gf_common_mt_eh_t);
+        if (!priv->shd.statistics) {
+                ret = -1;
+                goto out;
+        }
+        child_count = priv->child_count;
+        for (i=0; i < child_count ; i++) {
+                stats_per_brick = eh_new (AFR_STATISTICS_HISTORY_SIZE,
+                                          _gf_false,
+                                          _destroy_crawl_event_data);
+                if (!stats_per_brick) {
+                        ret = -1;
+                        goto out;
+                }
+                priv->shd.statistics[i] = stats_per_brick;
+
+        }
+
+        shd_crawl_events = (shd_crawl_event_t***)(&priv->shd.crawl_events);
+        *shd_crawl_events  = GF_CALLOC (sizeof(shd_crawl_event_t*),
+                                        priv->child_count,
+                                        gf_afr_mt_shd_crawl_event_t);
+
+        if (!priv->shd.crawl_events) {
+                ret = -1;
+                goto out;
+        }
+        ret = 0;
+out:
+        return ret;
+
+}
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index 0e1718814..689dd84e6 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -253,7 +253,7 @@ unlock:
                         goto out;
 
                 if (!afr_is_opendir_done (this, local->fd->inode) &&
-                    up_children_count > 1) {
+                    up_children_count > 1 && priv->entry_self_heal) {
 
                         /*
                          * This is the first opendir on this inode. We need
@@ -383,14 +383,36 @@ afr_forget_entries (fd_t *fd)
         }
 }
 
+static void
+afr_readdir_filter_trash_dir (gf_dirent_t *entries, fd_t *fd)
+{
+        gf_dirent_t *   entry       = NULL;
+        gf_dirent_t *   tmp         = NULL;
+
+        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
+                if (__is_root_gfid (fd->inode->gfid) &&
+                    !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) {
+                        list_del_init (&entry->list);
+                        GF_FREE (entry);
+                }
+        }
+}
 
 int32_t
 afr_readdir_cbk (call_frame_t *frame, void *cookie,
                  xlator_t *this, int32_t op_ret, int32_t op_errno,
                  gf_dirent_t *entries, dict_t *xdata)
 {
-        AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries, NULL);
+        afr_local_t     *local = NULL;
 
+        if (op_ret == -1)
+                goto out;
+
+        local = frame->local;
+        afr_readdir_filter_trash_dir (entries, local->fd);
+
+out:
+        AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries, NULL);
         return 0;
 }
 
@@ -400,8 +422,16 @@ afr_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                   int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
                   dict_t *xdata)
 {
-        AFR_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries, NULL);
+        afr_local_t     *local = NULL;
+
+        if (op_ret == -1)
+                goto out;
+
+        local = frame->local;
+        afr_readdir_filter_trash_dir (entries, local->fd);
 
+out:
+        AFR_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries, NULL);
         return 0;
 }
 
diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c
index 5a22696ce..1943b719b 100644
--- a/xlators/cluster/afr/src/afr-dir-write.c
+++ b/xlators/cluster/afr/src/afr-dir-write.c
@@ -235,26 +235,6 @@ out:
 }
 
 void
-afr_dir_fop_handle_all_fop_failures (call_frame_t *frame)
-{
-        xlator_t        *this = NULL;
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv = NULL;
-
-        this = frame->this;
-        local = frame->local;
-        priv = this->private;
-
-        if (local->op_ret >= 0)
-                goto out;
-
-        __mark_all_success (local->pending, priv->child_count,
-                            local->transaction.type);
-out:
-        return;
-}
-
-void
 afr_dir_fop_done (call_frame_t *frame, xlator_t *this)
 {
         afr_local_t     *local          = NULL;
@@ -273,7 +253,6 @@ afr_dir_fop_done (call_frame_t *frame, xlator_t *this)
 done:
         local->transaction.unwind (frame, this);
         afr_dir_fop_mark_entry_pending_changelog (frame, this);
-        afr_dir_fop_handle_all_fop_failures (frame);
         local->transaction.resume (frame, this);
 }
 
diff --git a/xlators/cluster/afr/src/afr-inode-read.c b/xlators/cluster/afr/src/afr-inode-read.c
index 051b7b164..e06e3b2f2 100644
--- a/xlators/cluster/afr/src/afr-inode-read.c
+++ b/xlators/cluster/afr/src/afr-inode-read.c
@@ -117,6 +117,8 @@ afr_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
 
         children = priv->children;
 
+        AFR_SBRAIN_CHECK_LOC (loc, out);
+
         AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
         local = frame->local;
 
@@ -232,6 +234,8 @@ afr_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 
         children = priv->children;
 
+        AFR_SBRAIN_CHECK_LOC (loc, out);
+
         AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
         local = frame->local;
 
@@ -348,10 +352,7 @@ afr_fstat (call_frame_t *frame, xlator_t *this,
 
         VALIDATE_OR_GOTO (fd->inode, out);
 
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+        AFR_SBRAIN_CHECK_FD (fd, out);
 
         AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
         local = frame->local;
@@ -472,6 +473,8 @@ afr_readlink (call_frame_t *frame, xlator_t *this,
 
         children = priv->children;
 
+        AFR_SBRAIN_CHECK_LOC (loc, out);
+
         AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
         local = frame->local;
 
@@ -1323,6 +1326,62 @@ afr_getxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
         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_min_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)
@@ -1355,6 +1414,8 @@ afr_is_special_xattr (const char *name, fop_getxattr_cbk_t *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;
         }
@@ -1403,7 +1464,7 @@ afr_getxattr (call_frame_t *frame, xlator_t *this,
         int32_t                 read_child    = -1;
         int                     ret           = -1;
         fop_getxattr_cbk_t      cbk           = NULL;
-
+        int                     afr_xtime_gauge[MCNT_MAX] = {0,};
 
         VALIDATE_OR_GOTO (frame, out);
         VALIDATE_OR_GOTO (this, out);
@@ -1414,6 +1475,8 @@ afr_getxattr (call_frame_t *frame, xlator_t *this,
 
         children = priv->children;
 
+        AFR_SBRAIN_CHECK_LOC (loc, out);
+
         AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
         local = frame->local;
 
@@ -1452,6 +1515,7 @@ afr_getxattr (call_frame_t *frame, xlator_t *this,
                                            sub_volumes,
                                            priv->child_count,
                                            MARKER_UUID_TYPE,
+                                           marker_uuid_default_gauge,
                                            priv->vol_uuid)) {
 
                         gf_log (this->name, GF_LOG_INFO,
@@ -1498,12 +1562,20 @@ afr_getxattr (call_frame_t *frame, xlator_t *this,
 
                         }
 
+                        /* don't err out on getting ENOTCONN (brick down)
+                         * from a subset of the bricks
+                         */
+                        memcpy (afr_xtime_gauge, marker_xtime_default_gauge,
+                                sizeof (afr_xtime_gauge));
+                        afr_xtime_gauge[MCNT_NOTFOUND] = 0;
+                        afr_xtime_gauge[MCNT_ENOTCONN] = 0;
                         if (cluster_getmarkerattr (frame, this, loc,
                                                    name, local,
                                                    afr_getxattr_unwind,
                                                    sub_volumes,
                                                    priv->child_count,
                                                    MARKER_XTIME_TYPE,
+                                                   afr_xtime_gauge,
                                                    priv->vol_uuid)) {
                                 gf_log (this->name, GF_LOG_INFO,
                                         "%s: failed to get marker attr (%s)",
@@ -1660,10 +1732,8 @@ afr_fgetxattr (call_frame_t *frame, xlator_t *this,
 
         children = priv->children;
 
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+        AFR_SBRAIN_CHECK_FD (fd, out);
+
         AFR_LOCAL_ALLOC_OR_GOTO (local, out);
         frame->local = local;
 
@@ -1819,10 +1889,7 @@ afr_readv (call_frame_t *frame, xlator_t *this,
         priv     = this->private;
         children = priv->children;
 
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+        AFR_SBRAIN_CHECK_FD (fd, out);
 
         AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
         local = frame->local;
diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index ce4fbf226..c1ec69a55 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -39,6 +39,46 @@
 #include "afr-transaction.h"
 #include "afr-self-heal-common.h"
 
+void
+__inode_write_fop_cbk (call_frame_t *frame, int child_index, int read_child,
+                       xlator_t *this, int32_t *op_ret, int32_t *op_errno,
+                       struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t     *local = NULL;
+
+        local = frame->local;
+
+        if (afr_fop_failed (*op_ret, *op_errno)) {
+                local->child_errno[child_index] = *op_errno;
+
+                switch (local->op) {
+                case GF_FOP_TRUNCATE:
+                case GF_FOP_FTRUNCATE:
+                        if (*op_errno != EFBIG)
+                                afr_transaction_fop_failed (frame, this,
+                                                            child_index);
+                break;
+                default:
+                        afr_transaction_fop_failed (frame, this, child_index);
+                break;
+                }
+                local->op_errno = *op_errno;
+                goto out;
+        }
+
+        if ((local->success_count == 0) || (read_child == child_index)) {
+                local->op_ret              = *op_ret;
+                if (prebuf)
+                        local->cont.inode_wfop.prebuf  = *prebuf;
+                if (postbuf)
+                        local->cont.inode_wfop.postbuf = *postbuf;
+        }
+
+        local->success_count++;
+out:
+        return;
+}
+
 /* {{{ writev */
 
 void
@@ -52,8 +92,8 @@ afr_writev_copy_outvars (call_frame_t *src_frame, call_frame_t *dst_frame)
 
         dst_local->op_ret = src_local->op_ret;
         dst_local->op_errno = src_local->op_errno;
-        dst_local->cont.writev.prebuf = src_local->cont.writev.prebuf;
-        dst_local->cont.writev.postbuf = src_local->cont.writev.postbuf;
+        dst_local->cont.inode_wfop.prebuf = src_local->cont.inode_wfop.prebuf;
+        dst_local->cont.inode_wfop.postbuf = src_local->cont.inode_wfop.postbuf;
 }
 
 void
@@ -64,8 +104,8 @@ afr_writev_unwind (call_frame_t *frame, xlator_t *this)
 
         AFR_STACK_UNWIND (writev, frame,
                           local->op_ret, local->op_errno,
-                          &local->cont.writev.prebuf,
-                          &local->cont.writev.postbuf,
+                          &local->cont.inode_wfop.prebuf,
+                          &local->cont.inode_wfop.postbuf,
                           NULL);
 }
 
@@ -133,12 +173,17 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                      struct iatt *postbuf, dict_t *xdata)
 {
         afr_local_t *   local = NULL;
+        afr_private_t  *priv  = NULL;
         call_frame_t    *fop_frame = NULL;
         int child_index = (long) cookie;
         int call_count  = -1;
         int read_child  = 0;
+        int      ret = 0;
+        uint32_t open_fd_count = 0;
+        uint32_t write_is_append = 0;
 
         local = frame->local;
+        priv  = this->private;
 
         read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
 
@@ -148,12 +193,14 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->read_child_returned = _gf_true;
                 }
 
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, prebuf, postbuf,
+                                       xdata);
+
 		local->replies[child_index].valid = 1;
 		local->replies[child_index].op_ret = op_ret;
 		local->replies[child_index].op_errno = op_errno;
 
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
 
 		/* stage the best case return value for unwind */
                 if ((local->success_count == 0) || (op_ret > local->op_ret)) {
@@ -162,12 +209,24 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		}
 
 		if (op_ret != -1) {
-			if ((local->success_count == 0) ||
-			    (child_index == read_child)) {
-				local->cont.writev.prebuf  = *prebuf;
-				local->cont.writev.postbuf = *postbuf;
-			}
-			local->success_count++;
+                        if (xdata) {
+                                ret = dict_get_uint32 (xdata,
+                                                       GLUSTERFS_OPEN_FD_COUNT,
+                                                       &open_fd_count);
+                                if ((ret == 0) &&
+                                    (open_fd_count > local->open_fd_count)) {
+                                        local->open_fd_count = open_fd_count;
+                                        local->update_open_fd_count = _gf_true;
+                                }
+
+				write_is_append = 0;
+                                ret = dict_get_uint32 (xdata,
+                                                       GLUSTERFS_WRITE_IS_APPEND,
+                                                       &write_is_append);
+                                if (ret || !write_is_append)
+					local->append_write = _gf_false;
+                        }
+
 		}
         }
         UNLOCK (&frame->lock);
@@ -176,10 +235,23 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         if (call_count == 0) {
 
-		if (!local->stable_write)
-			afr_fd_report_unstable_write (this, local->fd);
+                if (local->update_open_fd_count)
+                        afr_handle_open_fd_count (frame, this);
+
+                if (!local->stable_write && !local->append_write)
+			/* An appended write removes the necessity to
+			   fsync() the file. This is because self-heal
+			   has the logic to check for larger file when
+			   the xattrs are not reliably pointing at
+			   a stale file.
+			*/
+                        afr_fd_report_unstable_write (this, local->fd);
 
                 afr_writev_handle_short_writes (frame, this);
+                if (afr_any_fops_failed (local, priv)) {
+                        //Don't unwind until post-op is complete
+                        local->transaction.resume (frame, this);
+                } else {
                 /*
                  * Generally inode-write fops do transaction.unwind then
                  * transaction.resume, but writev needs to make sure that
@@ -191,10 +263,11 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                  * 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);
+                        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;
 }
@@ -206,6 +279,8 @@ afr_writev_wind (call_frame_t *frame, xlator_t *this)
         afr_private_t *priv = NULL;
         int i = 0;
         int call_count = -1;
+        dict_t *xdata = NULL;
+        GF_UNUSED int     ret = 0;
 
         local = frame->local;
         priv = this->private;
@@ -229,6 +304,19 @@ afr_writev_wind (call_frame_t *frame, xlator_t *this)
 		return 0;
 	}
 
+        xdata = dict_new ();
+        if (xdata) {
+                ret = dict_set_uint32 (xdata, GLUSTERFS_OPEN_FD_COUNT,
+                                       sizeof (uint32_t));
+		ret = dict_set_uint32 (xdata, GLUSTERFS_WRITE_IS_APPEND,
+				       0);
+		/* Set append_write to be true speculatively. If on any
+		   server it turns not be true, we unset it in the
+		   callback.
+		*/
+		local->append_write = _gf_true;
+        }
+
         for (i = 0; i < priv->child_count; i++) {
                 if (local->transaction.pre_op[i]) {
                         STACK_WIND_COOKIE (frame, afr_writev_wind_cbk,
@@ -241,13 +329,16 @@ afr_writev_wind (call_frame_t *frame, xlator_t *this)
                                            local->cont.writev.offset,
                                            local->cont.writev.flags,
                                            local->cont.writev.iobref,
-                                           NULL);
+                                           xdata);
 
                         if (!--call_count)
                                 break;
                 }
         }
 
+        if (xdata)
+                dict_unref (xdata);
+
         return 0;
 }
 
@@ -515,8 +606,8 @@ afr_truncate_unwind (call_frame_t *frame, xlator_t *this)
         if (main_frame) {
                 AFR_STACK_UNWIND (truncate, main_frame, local->op_ret,
                                   local->op_errno,
-                                  &local->cont.truncate.prebuf,
-                                  &local->cont.truncate.postbuf,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
                                   NULL);
         }
 
@@ -530,14 +621,11 @@ afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                        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);
 
@@ -547,38 +635,22 @@ afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->read_child_returned = _gf_true;
                 }
 
-                if (afr_fop_failed (op_ret, op_errno) && op_errno != EFBIG)
-                        afr_transaction_fop_failed (frame, this, child_index);
-
                 if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.truncate.prebuf  = *prebuf;
-                                local->cont.truncate.postbuf = *postbuf;
-                        }
-
-                        if (child_index == read_child) {
-                                local->cont.truncate.prebuf  = *prebuf;
-                                local->cont.truncate.postbuf = *postbuf;
-                        }
-
-                        local->success_count++;
-
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
+			if (prebuf->ia_size != postbuf->ia_size)
+				local->stable_write = _gf_false;
                 }
-                local->op_errno = op_errno;
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, prebuf, postbuf,
+                                       xdata);
         }
         UNLOCK (&frame->lock);
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
-
         call_count = afr_frame_return (frame);
 
         if (call_count == 0) {
+		if (local->stable_write && afr_txn_nothing_failed (frame, this))
+			local->transaction.unwind (frame, this);
+
                 local->transaction.resume (frame, this);
         }
 
@@ -606,6 +678,7 @@ afr_truncate_wind (call_frame_t *frame, xlator_t *this)
         }
 
         local->call_count = call_count;
+	local->stable_write = _gf_true;
 
         for (i = 0; i < priv->child_count; i++) {
                 if (local->transaction.pre_op[i]) {
@@ -726,8 +799,8 @@ afr_ftruncate_unwind (call_frame_t *frame, xlator_t *this)
         if (main_frame) {
                 AFR_STACK_UNWIND (ftruncate, main_frame, local->op_ret,
                                   local->op_errno,
-                                  &local->cont.ftruncate.prebuf,
-                                  &local->cont.ftruncate.postbuf,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
                                   NULL);
         }
         return 0;
@@ -740,14 +813,11 @@ afr_ftruncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         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);
 
@@ -757,38 +827,22 @@ afr_ftruncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         local->read_child_returned = _gf_true;
                 }
 
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
-
                 if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.ftruncate.prebuf  = *prebuf;
-                                local->cont.ftruncate.postbuf = *postbuf;
-                        }
-
-                        if (child_index == read_child) {
-                                local->cont.ftruncate.prebuf  = *prebuf;
-                                local->cont.ftruncate.postbuf = *postbuf;
-                        }
-
-                        local->success_count++;
-
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
+			if (prebuf->ia_size != postbuf->ia_size)
+				local->stable_write = _gf_false;
                 }
-                local->op_errno = op_errno;
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, prebuf, postbuf,
+                                       xdata);
         }
         UNLOCK (&frame->lock);
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
-
         call_count = afr_frame_return (frame);
 
         if (call_count == 0) {
+		if (local->stable_write && afr_txn_nothing_failed (frame, this))
+			local->transaction.unwind (frame, this);
+
                 local->transaction.resume (frame, this);
         }
 
@@ -816,6 +870,7 @@ afr_ftruncate_wind (call_frame_t *frame, xlator_t *this)
         }
 
         local->call_count = call_count;
+	local->stable_write = _gf_true;
 
         for (i = 0; i < priv->child_count; i++) {
                 if (local->transaction.pre_op[i]) {
@@ -970,8 +1025,8 @@ afr_setattr_unwind (call_frame_t *frame, xlator_t *this)
         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,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
                                   NULL);
         }
 
@@ -1002,29 +1057,14 @@ afr_setattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         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.setattr.preop_buf  = *preop;
-                                local->cont.setattr.postop_buf = *postop;
-                        }
-
-                        if (child_index == read_child) {
-                                local->cont.setattr.preop_buf  = *preop;
-                                local->cont.setattr.postop_buf = *postop;
-                        }
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, preop, postop,
+                                       xdata);
 
-                        local->success_count++;
-
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
+                if ((local->success_count >= priv->wait_count)
+                    && local->read_child_returned) {
+                        need_unwind = 1;
                 }
-                local->op_errno = op_errno;
         }
         UNLOCK (&frame->lock);
 
@@ -1179,8 +1219,8 @@ afr_fsetattr_unwind (call_frame_t *frame, xlator_t *this)
         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,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
                                   NULL);
         }
 
@@ -1211,29 +1251,14 @@ afr_fsetattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         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.fsetattr.preop_buf  = *preop;
-                                local->cont.fsetattr.postop_buf = *postop;
-                        }
-
-                        if (child_index == read_child) {
-                                local->cont.fsetattr.preop_buf  = *preop;
-                                local->cont.fsetattr.postop_buf = *postop;
-                        }
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, preop, postop,
+                                       xdata);
 
-                        local->success_count++;
-
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
+                if ((local->success_count >= priv->wait_count)
+                    && local->read_child_returned) {
+                        need_unwind = 1;
                 }
-                local->op_errno = op_errno;
         }
         UNLOCK (&frame->lock);
 
@@ -1406,28 +1431,23 @@ int
 afr_setxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_local_t      *local         = NULL;
-        afr_private_t    *priv          = NULL;
-        int               call_count    = -1;
-        int               need_unwind   = 0;
+        afr_local_t   *local      = NULL;
+        afr_private_t *priv       = NULL;
+        int           call_count  = -1;
+        int           need_unwind = 0;
+        int           child_index = (long) cookie;
 
         local = frame->local;
         priv = this->private;
 
         LOCK (&frame->lock);
         {
-                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;
-                        }
+                __inode_write_fop_cbk (frame, child_index, -1, this,
+                                       &op_ret, &op_errno, NULL, NULL,
+                                       xdata);
+                if (local->success_count == priv->child_count) {
+                        need_unwind = 1;
                 }
-
-                local->op_errno = op_errno;
         }
         UNLOCK (&frame->lock);
 
@@ -1596,28 +1616,24 @@ int
 afr_fsetxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_local_t     *local       = NULL;
-        afr_private_t   *priv        = NULL;
-        int              call_count  = -1;
-        int              need_unwind = 0;
+        afr_local_t   *local      = NULL;
+        afr_private_t *priv       = NULL;
+        int           call_count  = -1;
+        int           need_unwind = 0;
+        int           child_index = (long) cookie;
 
         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;
-                        }
+                __inode_write_fop_cbk (frame, child_index, -1, this,
+                                       &op_ret, &op_errno, NULL, NULL,
+                                       xdata);
+                if (local->success_count == priv->child_count) {
+                        need_unwind = 1;
                 }
-
-                local->op_errno = op_errno;
         }
         UNLOCK (&frame->lock);
 
@@ -1796,28 +1812,23 @@ int
 afr_removexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                           int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int call_count  = -1;
-        int need_unwind = 0;
+        afr_local_t   *local      = NULL;
+        afr_private_t *priv       = NULL;
+        int           call_count  = -1;
+        int           need_unwind = 0;
+        int           child_index = (long) cookie;
 
         local = frame->local;
         priv = this->private;
 
         LOCK (&frame->lock);
         {
-                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;
-                        }
+                __inode_write_fop_cbk (frame, child_index, -1, this,
+                                       &op_ret, &op_errno, NULL, NULL,
+                                       xdata);
+                if (local->success_count == priv->wait_count) {
+                        need_unwind = 1;
                 }
-
-                local->op_errno = op_errno;
         }
         UNLOCK (&frame->lock);
 
@@ -1985,28 +1996,24 @@ int
 afr_fremovexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                           int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int call_count  = -1;
-        int need_unwind = 0;
+        afr_local_t *   local       = NULL;
+        afr_private_t * priv        = NULL;
+        int             call_count  = -1;
+        int             need_unwind = 0;
+        int             child_index = (long) cookie;
 
         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++;
+                __inode_write_fop_cbk (frame, child_index, -1, this,
+                                       &op_ret, &op_errno, NULL, NULL,
+                                       xdata);
 
-                        if (local->success_count == priv->wait_count) {
-                                need_unwind = 1;
-                        }
+                if (local->success_count == priv->wait_count) {
+                        need_unwind = 1;
                 }
-
-                local->op_errno = op_errno;
         }
         UNLOCK (&frame->lock);
 
@@ -2151,3 +2158,704 @@ out:
 
         return 0;
 }
+
+static int
+afr_fallocate_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 (fallocate, main_frame, local->op_ret,
+                                  local->op_errno,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
+                                  NULL);
+        }
+        return 0;
+}
+
+static int
+afr_fallocate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                        struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = NULL;
+        int child_index = (long) cookie;
+        int call_count  = -1;
+        int need_unwind = 0;
+        int read_child  = 0;
+
+        local = frame->local;
+        priv  = this->private;
+
+        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
+
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, prebuf, postbuf,
+                                       xdata);
+
+                if ((local->success_count >= priv->wait_count)
+                    && local->read_child_returned) {
+                        need_unwind = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        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_fallocate_wind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_fallocate_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->fallocate,
+                                           local->fd,
+                                           local->cont.fallocate.mode,
+                                           local->cont.fallocate.offset,
+                                           local->cont.fallocate.len,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
+
+static int
+afr_fallocate_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);
+
+        return 0;
+}
+
+static int
+afr_do_fallocate (call_frame_t *frame, xlator_t *this)
+{
+        call_frame_t * transaction_frame = NULL;
+        afr_local_t *  local             = NULL;
+        int op_ret   = -1;
+        int op_errno = 0;
+
+        local = frame->local;
+
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                goto out;
+        }
+
+        transaction_frame->local = local;
+        frame->local = NULL;
+
+        local->op = GF_FOP_FALLOCATE;
+
+        local->transaction.fop    = afr_fallocate_wind;
+        local->transaction.done   = afr_fallocate_done;
+        local->transaction.unwind = afr_fallocate_unwind;
+
+        local->transaction.main_frame = frame;
+
+        local->transaction.start   = local->cont.fallocate.offset;
+        local->transaction.len     = 0;
+
+        /* fallocate can modify the file size */
+        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (op_ret < 0) {
+            op_errno = -op_ret;
+            goto out;
+        }
+
+        op_ret = 0;
+out:
+        if (op_ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (fallocate, frame, op_ret, op_errno, NULL,
+                                  NULL, NULL);
+        }
+
+        return 0;
+}
+
+int
+afr_fallocate (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+               off_t offset, size_t len, dict_t *xdata)
+{
+        afr_private_t * priv  = NULL;
+        afr_local_t   * local = NULL;
+        call_frame_t   *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
+        QUORUM_CHECK(fallocate,out);
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        local->cont.fallocate.mode = mode;
+        local->cont.fallocate.offset  = offset;
+        local->cont.fallocate.len = len;
+
+        local->fd = fd_ref (fd);
+
+        afr_open_fd_fix (fd, this);
+
+        afr_do_fallocate (frame, this);
+
+        ret = 0;
+out:
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+        }
+
+        return 0;
+}
+
+/* }}} */
+
+/* {{{ discard */
+
+static int
+afr_discard_unwind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        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 (discard, main_frame, local->op_ret,
+                                  local->op_errno,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
+                                  NULL);
+        }
+        return 0;
+}
+
+static int
+afr_discard_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = NULL;
+        int child_index = (long) cookie;
+        int call_count  = -1;
+        int need_unwind = 0;
+        int read_child  = 0;
+
+        local = frame->local;
+        priv  = this->private;
+
+        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
+
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, prebuf, postbuf,
+                                       xdata);
+
+                if ((local->success_count >= priv->wait_count)
+                    && local->read_child_returned) {
+                        need_unwind = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        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_discard_wind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_discard_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->discard,
+                                           local->fd,
+                                           local->cont.discard.offset,
+                                           local->cont.discard.len,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
+
+static int
+afr_discard_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);
+
+        return 0;
+}
+
+static int
+afr_do_discard (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;
+
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                goto out;
+        }
+
+        transaction_frame->local = local;
+        frame->local = NULL;
+
+        local->op = GF_FOP_DISCARD;
+
+        local->transaction.fop    = afr_discard_wind;
+        local->transaction.done   = afr_discard_done;
+        local->transaction.unwind = afr_discard_unwind;
+
+        local->transaction.main_frame = frame;
+
+        local->transaction.start   = local->cont.discard.offset;
+        local->transaction.len     = 0;
+
+        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (op_ret < 0) {
+            op_errno = -op_ret;
+            goto out;
+        }
+
+        op_ret = 0;
+out:
+        if (op_ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (discard, frame, op_ret, op_errno, NULL,
+                                  NULL, NULL);
+        }
+
+        return 0;
+}
+
+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;
+        call_frame_t   *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
+        QUORUM_CHECK(discard, out);
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        local->cont.discard.offset  = offset;
+        local->cont.discard.len = len;
+
+        local->fd = fd_ref (fd);
+
+        afr_open_fd_fix (fd, this);
+
+        afr_do_discard(frame, this);
+
+        ret = 0;
+out:
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL);
+        }
+
+        return 0;
+}
+
+
+/* {{{ zerofill */
+
+static int
+afr_zerofill_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 (zerofill, main_frame, local->op_ret,
+                                  local->op_errno,
+                                  &local->cont.zerofill.prebuf,
+                                  &local->cont.zerofill.postbuf,
+                                  NULL);
+        }
+        return 0;
+}
+
+static int
+afr_zerofill_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t       *local             = NULL;
+        afr_private_t     *priv              = NULL;
+        int                child_index       = (long) cookie;
+        int                call_count        = -1;
+        int                need_unwind       = 0;
+        int                read_child        = 0;
+
+        local = frame->local;
+        priv  = this->private;
+
+        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
+
+                if (afr_fop_failed (op_ret, op_errno)) {
+                        afr_transaction_fop_failed (frame, this, child_index);
+                }
+
+                if (op_ret != -1) {
+                        if (local->success_count == 0) {
+                                local->op_ret = op_ret;
+                                local->cont.zerofill.prebuf  = *prebuf;
+                                local->cont.zerofill.postbuf = *postbuf;
+                        }
+
+                        if (child_index == read_child) {
+                                local->cont.zerofill.prebuf  = *prebuf;
+                                local->cont.zerofill.postbuf = *postbuf;
+                        }
+
+                        local->success_count++;
+
+                        if ((local->success_count >= priv->wait_count)
+                            && local->read_child_returned) {
+                                need_unwind = 1;
+                        }
+                }
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+        if (need_unwind) {
+                local->transaction.unwind (frame, this);
+        }
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+        }
+
+        return 0;
+}
+
+static int
+afr_zerofill_wind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t    *local         = NULL;
+        afr_private_t  *priv          = NULL;
+        int             call_count    = -1;
+        int             i             = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_zerofill_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->zerofill,
+                                           local->fd,
+                                           local->cont.zerofill.offset,
+                                           local->cont.zerofill.len,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
+
+static int
+afr_zerofill_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);
+
+        return 0;
+}
+
+static int
+afr_do_zerofill(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;
+
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                goto out;
+        }
+
+        transaction_frame->local = local;
+        frame->local = NULL;
+
+        local->op = GF_FOP_ZEROFILL;
+
+        local->transaction.fop    = afr_zerofill_wind;
+        local->transaction.done   = afr_zerofill_done;
+        local->transaction.unwind = afr_zerofill_unwind;
+
+        local->transaction.main_frame = frame;
+
+        local->transaction.start   = local->cont.zerofill.offset;
+        local->transaction.len     = 0;
+
+        op_ret = afr_transaction (transaction_frame, this,
+                                  AFR_DATA_TRANSACTION);
+        if (op_ret < 0) {
+                op_errno = -op_ret;
+                goto out;
+        }
+
+        op_ret = 0;
+out:
+        if (op_ret < 0) {
+                if (transaction_frame) {
+                        AFR_STACK_DESTROY (transaction_frame);
+                }
+                AFR_STACK_UNWIND (zerofill, frame, op_ret, op_errno, NULL,
+                                  NULL, NULL);
+        }
+
+        return 0;
+}
+
+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           = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
+        QUORUM_CHECK(zerofill, out);
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0) {
+                goto out;
+        }
+        local->cont.zerofill.offset  = offset;
+        local->cont.zerofill.len = len;
+
+        local->fd = fd_ref (fd);
+
+        afr_open_fd_fix (fd, this);
+
+        afr_do_zerofill(frame, this);
+
+        ret = 0;
+out:
+        if (ret < 0) {
+                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 ed11079fd..8e93ca44a 100644
--- a/xlators/cluster/afr/src/afr-inode-write.h
+++ b/xlators/cluster/afr/src/afr-inode-write.h
@@ -68,4 +68,15 @@ 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,
+             size_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 e091a7939..060d78f35 100644
--- a/xlators/cluster/afr/src/afr-lk-common.c
+++ b/xlators/cluster/afr/src/afr-lk-common.c
@@ -57,11 +57,15 @@
 int
 afr_entry_lockee_cmp (const void *l1, const void *l2)
 {
-       const afr_entry_lockee_t       *r1 = l1;
-       const afr_entry_lockee_t       *r2 = l2;
-       int                            ret = 0;
-
-        ret = uuid_compare (r1->loc.gfid, r2->loc.gfid);
+        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)
@@ -75,7 +79,6 @@ afr_entry_lockee_cmp (const void *l1, const void *l2)
                 return -1;
         else
                 return 1;
-
 }
 
 int afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index);
@@ -556,19 +559,23 @@ 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;
 }
@@ -649,7 +656,9 @@ afr_unlock_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 {
         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;
@@ -658,14 +667,18 @@ afr_unlock_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                AFR_UNLOCK_OP, NULL, op_ret,
                                op_errno, child_index);
 
+        priv = this->private;
+
         if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
-                gf_log (this->name, GF_LOG_INFO, "%s: unlock failed on %d "
-                        "unlock by %s", local->loc.path, child_index,
+                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));
         }
 
 
-        int_lock->inode_locked_nodes[child_index] &= LOCKED_NO;
+        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;
 
@@ -679,6 +692,7 @@ 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,};
@@ -694,12 +708,14 @@ afr_unlock_inodelk (call_frame_t *frame, xlator_t *this)
         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;
 
         full_flock.l_type = F_UNLCK;
-        call_count = afr_locked_nodes_count (int_lock->inode_locked_nodes,
+        call_count = afr_locked_nodes_count (inodelk->locked_nodes,
                                              priv->child_count);
 
         int_lock->lk_call_count = call_count;
@@ -715,8 +731,7 @@ afr_unlock_inodelk (call_frame_t *frame, xlator_t *this)
                 fd_ctx = afr_fd_ctx_get (local->fd, this);
 
         for (i = 0; i < priv->child_count; i++) {
-                if ((int_lock->inode_locked_nodes[i] & LOCKED_YES)
-                    != LOCKED_YES)
+                if ((inodelk->locked_nodes[i] & LOCKED_YES) != LOCKED_YES)
                         continue;
 
                 if (local->fd) {
@@ -757,7 +772,7 @@ afr_unlock_inodelk (call_frame_t *frame, xlator_t *this)
                                            (void *) (long)i,
                                            priv->children[i],
                                            priv->children[i]->fops->finodelk,
-                                           this->name, local->fd,
+                                           int_lock->domain, local->fd,
                                            F_SETLK, flock_use, NULL);
 
                         if (!--call_count)
@@ -772,7 +787,7 @@ afr_unlock_inodelk (call_frame_t *frame, xlator_t *this)
                                            (void *) (long)i,
                                            priv->children[i],
                                            priv->children[i]->fops->inodelk,
-                                           this->name, &local->loc,
+                                           int_lock->domain, &local->loc,
                                            F_SETLK, &flock, NULL);
 
                         if (!--call_count)
@@ -858,7 +873,7 @@ afr_unlock_entrylk (call_frame_t *frame, xlator_t *this)
                                            (void *) (long) i,
                                            priv->children[index],
                                            priv->children[index]->fops->entrylk,
-                                           this->name,
+                                           int_lock->domain,
                                            &int_lock->lockee[lockee_no].loc,
                                            int_lock->lockee[lockee_no].basename,
                                            ENTRYLK_UNLOCK, ENTRYLK_WRLCK, NULL);
@@ -961,6 +976,7 @@ 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;
 
@@ -971,10 +987,10 @@ 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:
@@ -1025,6 +1041,7 @@ int
 afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
 {
         afr_internal_lock_t *int_lock    = NULL;
+        afr_inodelk_t       *inodelk     = NULL;
         afr_local_t         *local       = NULL;
         afr_private_t       *priv        = NULL;
         struct gf_flock flock = {0,};
@@ -1039,10 +1056,15 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
         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);
@@ -1064,8 +1086,6 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
         }
 
         if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
-                is_entrylk = afr_is_entrylk (int_lock, local->transaction.type);
-
                 if ((is_entrylk && int_lock->entrylk_lock_count == 0) ||
                     (!is_entrylk && int_lock->lock_count == 0)) {
                         gf_log (this->name, GF_LOG_INFO,
@@ -1114,7 +1134,7 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
                                            (void *) (long) child_index,
                                            priv->children[child_index],
                                            priv->children[child_index]->fops->finodelk,
-                                           this->name, local->fd,
+                                           int_lock->domain, local->fd,
                                            F_SETLKW, &flock, NULL);
 
                 } else {
@@ -1127,7 +1147,7 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
                                            (void *) (long) child_index,
                                            priv->children[child_index],
                                            priv->children[child_index]->fops->inodelk,
-                                           this->name, &local->loc,
+                                           int_lock->domain, &local->loc,
                                            F_SETLKW, &flock, NULL);
                 }
 
@@ -1148,7 +1168,7 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
                                            (void *) (long) cookie,
                                            priv->children[child_index],
                                            priv->children[child_index]->fops->fentrylk,
-                                           this->name, local->fd,
+                                           int_lock->domain, local->fd,
                                            int_lock->lockee[lockee_no].basename,
                                            ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
                 } else {
@@ -1161,7 +1181,7 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
                                            (void *) (long) cookie,
                                            priv->children[child_index],
                                            priv->children[child_index]->fops->entrylk,
-                                           this->name,
+                                           int_lock->domain,
                                            &int_lock->lockee[lockee_no].loc,
                                            int_lock->lockee[lockee_no].basename,
                                            ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
@@ -1316,6 +1336,7 @@ 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;
                 }
 
@@ -1390,6 +1411,7 @@ afr_nonblocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                              int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
         afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
         afr_local_t         *local    = NULL;
         int call_count  = 0;
         int child_index = (long) cookie;
@@ -1398,6 +1420,7 @@ afr_nonblocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         local    = frame->local;
         int_lock = &local->internal_lock;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
 
         AFR_TRACE_INODELK_OUT (frame, this, AFR_INODELK_NB_TRANSACTION,
                                AFR_LOCK_OP, NULL, op_ret,
@@ -1423,9 +1446,8 @@ afr_nonblocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 			if (local->transaction.eager_lock)
 				local->transaction.eager_lock[child_index] = 0;
 		} else {
-			int_lock->inode_locked_nodes[child_index]
-				|= LOCKED_YES;
-			int_lock->inodelk_lock_count++;
+			inodelk->locked_nodes[child_index] |= LOCKED_YES;
+			inodelk->lock_count++;
 
 			if (local->transaction.eager_lock &&
 			    local->transaction.eager_lock[child_index] &&
@@ -1448,8 +1470,7 @@ afr_nonblocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 gf_log (this->name, GF_LOG_TRACE,
                         "Last inode locking reply received");
                 /* all locks successful. Proceed to call FOP */
-                if (int_lock->inodelk_lock_count ==
-                                int_lock->lk_expected_count) {
+                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;
@@ -1473,6 +1494,7 @@ 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;
@@ -1488,11 +1510,13 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this)
         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 = int_lock->lk_flock.l_type;
+        full_flock.l_type = inodelk->flock.l_type;
 
         initialize_inodelk_variables (frame, this);
 
@@ -1508,6 +1532,7 @@ 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;
                 }
@@ -1566,7 +1591,7 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this)
                                            (void *) (long) i,
                                            priv->children[i],
                                            priv->children[i]->fops->finodelk,
-                                           this->name, local->fd,
+                                           int_lock->domain, local->fd,
                                            F_SETLK, flock_use, NULL);
 
                         if (!--call_count)
@@ -1588,7 +1613,7 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this)
                                            (void *) (long) i,
                                            priv->children[i],
                                            priv->children[i]->fops->inodelk,
-                                           this->name, &local->loc,
+                                           int_lock->domain, &local->loc,
                                            F_SETLK, &flock, NULL);
 
                         if (!--call_count)
@@ -2112,29 +2137,38 @@ out:
         return ret;
 }
 
-void
-afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src,
+int
+afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src, char *dom,
                           unsigned int child_count)
 {
-        afr_local_t *dst_local = NULL;
-        afr_local_t *src_local = NULL;
-        afr_internal_lock_t *dst_lock = NULL;
-        afr_internal_lock_t *src_lock = NULL;
+        afr_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;
 
-        dst_local = dst->local;
-        dst_lock  = &dst_local->internal_lock;
         src_local = src->local;
         src_lock  = &src_local->internal_lock;
-        if (src_lock->inode_locked_nodes) {
-                memcpy (dst_lock->inode_locked_nodes,
-                        src_lock->inode_locked_nodes,
-                        sizeof (*dst_lock->inode_locked_nodes) * child_count);
-                memset (src_lock->inode_locked_nodes, 0,
-                        sizeof (*src_lock->inode_locked_nodes) * child_count);
+        src_inodelk = afr_get_inodelk (src_lock, dom);
+        dst_local = dst->local;
+        dst_lock  = &dst_local->internal_lock;
+        dst_inodelk = afr_get_inodelk (dst_lock, dom);
+        if (!dst_inodelk || !src_inodelk)
+                goto out;
+        if (src_inodelk->locked_nodes) {
+                memcpy (dst_inodelk->locked_nodes, src_inodelk->locked_nodes,
+                        sizeof (*dst_inodelk->locked_nodes) * child_count);
+                memset (src_inodelk->locked_nodes, 0,
+                        sizeof (*src_inodelk->locked_nodes) * child_count);
         }
 
         dst_lock->transaction_lk_type = src_lock->transaction_lk_type;
         dst_lock->selfheal_lk_type    = src_lock->selfheal_lk_type;
-        dst_lock->inodelk_lock_count = src_lock->inodelk_lock_count;
-        src_lock->inodelk_lock_count = 0;
+        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 e01ab366f..73594f265 100644
--- a/xlators/cluster/afr/src/afr-mem-types.h
+++ b/xlators/cluster/afr/src/afr-mem-types.h
@@ -41,7 +41,10 @@ enum gf_afr_mem_types_ {
         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_reply_t,
+        gf_afr_mt_stats_t,
+        gf_afr_mt_shd_crawl_event_t,
+        gf_afr_mt_uint64_t,
         gf_afr_mt_end
 };
 #endif
diff --git a/xlators/cluster/afr/src/afr-self-heal-algorithm.c b/xlators/cluster/afr/src/afr-self-heal-algorithm.c
index 1721fd270..83846f152 100644
--- a/xlators/cluster/afr/src/afr-self-heal-algorithm.c
+++ b/xlators/cluster/afr/src/afr-self-heal-algorithm.c
@@ -100,7 +100,7 @@ sh_loop_driver_done (call_frame_t *sh_frame, xlator_t *this,
         }
 
         sh_private_cleanup (sh_frame, this);
-        if (sh->op_failed) {
+        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
                 GF_ASSERT (!last_loop_frame);
                 //loop_finish should have happened and the old_loop should be NULL
                 gf_log (this->name, GF_LOG_DEBUG,
@@ -143,7 +143,7 @@ sh_loop_finish (call_frame_t *loop_frame, xlator_t *this)
         }
 
         if (loop_sh && loop_sh->data_lock_held) {
-                afr_sh_data_unlock (loop_frame, this,
+                afr_sh_data_unlock (loop_frame, this, this->name,
                                     sh_destroy_frame);
         } else {
                 sh_destroy_frame (loop_frame, this);
@@ -214,7 +214,7 @@ sh_loop_frame_create (call_frame_t *sh_frame, xlator_t *this,
                 goto out;
         //We want the frame to have same lk_owner as sh_frame
         //so that locks translator allows conflicting locks
-        new_loop_local = afr_local_copy (local, this);
+        new_loop_local = afr_self_heal_local_init (local, this);
         if (!new_loop_local)
                 goto out;
         new_loop_frame->local = new_loop_local;
@@ -273,10 +273,10 @@ sh_loop_start (call_frame_t *sh_frame, xlator_t *this, off_t offset,
         new_loop_sh->offset = offset;
         new_loop_sh->block_size = sh->block_size;
         afr_sh_data_lock (new_loop_frame, this, offset, new_loop_sh->block_size,
-			  _gf_true, sh_loop_lock_success, sh_loop_lock_failure);
+                          _gf_true, this->name, sh_loop_lock_success, sh_loop_lock_failure);
         return 0;
 out:
-        sh->op_failed = 1;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
         if (old_loop_frame)
                 sh_loop_finish (old_loop_frame, this);
         sh_loop_return (sh_frame, this, new_loop_frame, -1, ENOMEM);
@@ -307,8 +307,9 @@ sh_loop_driver (call_frame_t *sh_frame, xlator_t *this,
                         sh_priv->loops_running--;
                 offset = sh_priv->offset;
                 block_size = sh->block_size;
-                while ((!sh->eof_reached) && (0 == sh->op_failed) &&
-                       (sh_priv->loops_running < priv->data_self_heal_window_size)
+                while ((!sh->eof_reached) &&
+                       (!is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) &&
+                     (sh_priv->loops_running < priv->data_self_heal_window_size)
                        && (sh_priv->offset < sh->file_size)) {
 
                         loop++;
@@ -327,7 +328,8 @@ sh_loop_driver (call_frame_t *sh_frame, xlator_t *this,
         if (0 == loop) {
                 //loop finish does unlock, but the erasing of the pending
                 //xattrs needs to happen before that so do not finish the loop
-                if (is_driver_done && !sh->op_failed)
+                if (is_driver_done &&
+                    !is_self_heal_failed (sh, AFR_CHECK_SPECIFIC))
                         goto driver_done;
                 if (old_loop_frame) {
                         sh_loop_finish (old_loop_frame, this);
@@ -338,7 +340,7 @@ sh_loop_driver (call_frame_t *sh_frame, xlator_t *this,
         //If we have more loops to form we should finish previous loop after
         //the next loop lock
         while (loop--) {
-                if (sh->op_failed) {
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
                         // op failed in other loop, stop spawning more loops
                         if (old_loop_frame) {
                                 sh_loop_finish (old_loop_frame, this);
@@ -384,7 +386,7 @@ sh_loop_return (call_frame_t *sh_frame, xlator_t *this, call_frame_t *loop_frame
         }
 
         if (op_ret == -1) {
-                sh->op_failed = 1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 afr_sh_set_error (sh, op_errno);
                 if (loop_frame) {
                         sh_loop_finish (loop_frame, this);
@@ -432,16 +434,16 @@ sh_loop_write_cbk (call_frame_t *loop_frame, void *cookie, xlator_t *this,
                         priv->children[child_index]->name,
                         strerror (op_errno));
 
-                sh->op_failed = 1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 afr_sh_set_error (loop_sh, op_errno);
         } else if (op_ret < loop_local->cont.writev.vector->iov_len) {
-		gf_log(this->name, GF_LOG_ERROR,
-		       "incomplete write to %s on subvolume %s "
-		       "(expected %lu, returned %d)", sh_local->loc.path,
-		       priv->children[child_index]->name,
-		       loop_local->cont.writev.vector->iov_len, op_ret);
-		sh->op_failed = 1;
-	}
+                gf_log (this->name, GF_LOG_ERROR,
+                        "incomplete write to %s on subvolume %s "
+                        "(expected %lu, returned %d)", sh_local->loc.path,
+                        priv->children[child_index]->name,
+                        loop_local->cont.writev.vector->iov_len, op_ret);
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        }
 
         call_count = afr_frame_return (loop_frame);
 
@@ -514,7 +516,7 @@ sh_loop_read_cbk (call_frame_t *loop_frame, void *cookie,
 
         if (op_ret <= 0) {
                 if (op_ret < 0) {
-                        sh->op_failed = 1;
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                         gf_log (this->name, GF_LOG_ERROR, "read failed on %d "
                                 "for %s reason :%s", sh->source,
                                 sh_local->loc.path, strerror (errno));
@@ -624,7 +626,7 @@ sh_diff_checksum_cbk (call_frame_t *loop_frame, void *cookie, xlator_t *this,
                         "checksum on %s failed on subvolume %s (%s)",
                         sh_local->loc.path, priv->children[child_index]->name,
                         strerror (op_errno));
-                sh->op_failed = 1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
         } else {
                 memcpy (loop_sh->checksum + child_index * MD5_DIGEST_LENGTH,
                         strong_checksum, MD5_DIGEST_LENGTH);
@@ -662,7 +664,8 @@ sh_diff_checksum_cbk (call_frame_t *loop_frame, void *cookie, xlator_t *this,
                 }
                 UNLOCK (&sh_priv->lock);
 
-                if (write_needed && !sh->op_failed) {
+                if (write_needed &&
+                    !is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
                         sh_loop_read (loop_frame, this);
                 } else {
                         sh_loop_return (sh_frame, this, loop_frame,
@@ -751,14 +754,15 @@ out:
         return sh_priv;
 }
 
-void
-afr_sh_transfer_lock (call_frame_t *dst, call_frame_t *src,
+int
+afr_sh_transfer_lock (call_frame_t *dst, call_frame_t *src, char *dom,
                       unsigned int child_count)
 {
         afr_local_t             *dst_local   = NULL;
         afr_self_heal_t         *dst_sh      = NULL;
         afr_local_t             *src_local   = NULL;
         afr_self_heal_t         *src_sh      = NULL;
+        int                     ret          = -1;
 
         dst_local = dst->local;
         dst_sh = &dst_local->self_heal;
@@ -766,9 +770,12 @@ afr_sh_transfer_lock (call_frame_t *dst, call_frame_t *src,
         src_sh = &src_local->self_heal;
         GF_ASSERT (src_sh->data_lock_held);
         GF_ASSERT (!dst_sh->data_lock_held);
-        afr_lk_transfer_datalock (dst, src, child_count);
+        ret = afr_lk_transfer_datalock (dst, src, dom, child_count);
+        if (ret)
+                return ret;
         src_sh->data_lock_held = _gf_false;
         dst_sh->data_lock_held = _gf_true;
+        return 0;
 }
 
 int
@@ -790,7 +797,10 @@ afr_sh_start_loops (call_frame_t *sh_frame, xlator_t *this,
         ret = sh_loop_frame_create (sh_frame, this, NULL, &first_loop_frame);
         if (ret)
                 goto out;
-        afr_sh_transfer_lock (first_loop_frame, sh_frame, priv->child_count);
+        ret = afr_sh_transfer_lock (first_loop_frame, sh_frame, this->name,
+                                    priv->child_count);
+        if (ret)
+                goto out;
         sh->private = afr_sh_priv_init ();
         if (!sh->private) {
                 ret = -1;
@@ -800,7 +810,7 @@ afr_sh_start_loops (call_frame_t *sh_frame, xlator_t *this,
         ret = 0;
 out:
         if (ret) {
-                sh->op_failed = 1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 sh_loop_driver_done (sh_frame, this, NULL);
         }
         return 0;
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c
index 2538f4c8b..ef92b4205 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.c
+++ b/xlators/cluster/afr/src/afr-self-heal-common.c
@@ -18,6 +18,28 @@
 #include "afr-self-heal.h"
 #include "pump.h"
 
+#define ADD_FMT_STRING(msg, off, sh_str, status, print_log)                 \
+        do {                                                                \
+                if (AFR_SELF_HEAL_NOT_ATTEMPTED != status) {                \
+                        off += snprintf (msg + off, sizeof (msg) - off,     \
+                                         " "sh_str" self heal %s,",         \
+                                         get_sh_completion_status (status));\
+                        print_log = 1;                                      \
+                }                                                           \
+        } while (0)
+
+#define ADD_FMT_STRING_SYNC(msg, off, sh_str, status, print_log)            \
+        do {                                                                \
+                if (AFR_SELF_HEAL_SYNC_BEGIN == status ||                   \
+                    AFR_SELF_HEAL_FAILED == status)  {                      \
+                        off += snprintf (msg + off, sizeof (msg) - off,     \
+                                         " "sh_str" self heal %s,",         \
+                                         get_sh_completion_status (status));\
+                        print_log = 1;                                      \
+                }                                                           \
+        } while (0)
+
+
 void
 afr_sh_reset (call_frame_t *frame, xlator_t *this)
 {
@@ -141,9 +163,8 @@ afr_sh_print_pending_matrix (int32_t *pending_matrix[], xlator_t *this)
         GF_FREE (buf);
 }
 
-void
-afr_sh_print_split_brain_log (int32_t *pending_matrix[], xlator_t *this,
-                              const char *loc)
+char*
+afr_get_pending_matrix_str (int32_t *pending_matrix[], xlator_t *this)
 {
         afr_private_t *  priv = this->private;
         char            *buf  = NULL;
@@ -173,10 +194,8 @@ afr_sh_print_split_brain_log (int32_t *pending_matrix[], xlator_t *this,
                         + (child_count * child_count * pending_entry_strlen);
 
         buf = GF_CALLOC (1, 1 + strlen (msg) + string_length , gf_afr_mt_char);
-        if (!buf) {
-                buf = "";
+        if (!buf)
                 goto out;
-        }
 
         ptr = buf;
         ptr += sprintf (ptr, "%s", msg);
@@ -192,11 +211,27 @@ afr_sh_print_split_brain_log (int32_t *pending_matrix[], xlator_t *this,
         ptr += sprintf (ptr, "%s", matrix_end);
 
 out:
+        return buf;
+}
+
+void
+afr_sh_print_split_brain_log (int32_t *pending_matrix[], xlator_t *this,
+                              const char *loc)
+{
+        char *buf      = NULL;
+        char *free_ptr = NULL;
+
+        buf = afr_get_pending_matrix_str (pending_matrix, this);
+        if (buf)
+                free_ptr = buf;
+        else
+                buf = "";
+
+
         gf_log (this->name, GF_LOG_ERROR, "Unable to self-heal contents of '%s'"
                 " (possible split-brain). Please delete the file from all but "
                 "the preferred subvolume.%s", loc, buf);
-        if (buf)
-                GF_FREE (buf);
+        GF_FREE (free_ptr);
         return;
 }
 
@@ -466,6 +501,8 @@ afr_find_biggest_witness_among_fools (int32_t *witnesses,
 {
         int i               = 0;
         int biggest_witness = -1;
+        int biggest_witness_idx = -1;
+        int biggest_witness_cnt = -1;
 
         GF_ASSERT (witnesses);
         GF_ASSERT (characters);
@@ -475,10 +512,21 @@ afr_find_biggest_witness_among_fools (int32_t *witnesses,
                 if (characters[i].type != AFR_NODE_FOOL)
                         continue;
 
-                if (biggest_witness < witnesses[i])
+                if (biggest_witness < witnesses[i]) {
                         biggest_witness = witnesses[i];
+			biggest_witness_idx = i;
+			biggest_witness_cnt = 1;
+			continue;
+		}
+
+		if (biggest_witness == witnesses[i])
+			biggest_witness_cnt++;
         }
-        return biggest_witness;
+
+	if (biggest_witness_cnt != 1)
+		return -1;
+
+        return biggest_witness_idx;
 }
 
 int
@@ -506,10 +554,84 @@ afr_mark_fool_as_source_by_witness (int32_t *sources, int32_t *witnesses,
         return nsources;
 }
 
+
+int
+afr_mark_fool_as_source_by_idx (int32_t *sources, int child_count, int idx)
+{
+	if (idx >= 0 && idx < child_count) {
+		sources[idx] = 1;
+		return 1;
+	}
+	return 0;
+}
+
+
+static int
+afr_find_largest_file_size (struct iatt *bufs, int32_t *success_children,
+			    int child_count)
+{
+	int idx = -1;
+	int i = -1;
+	int child = -1;
+	uint64_t max_size = 0;
+        uint64_t min_size = 0;
+        int      num_children = 0;
+
+	for (i = 0; i < child_count; i++) {
+		if (success_children[i] == -1)
+			break;
+
+		child = success_children[i];
+		if (bufs[child].ia_size > max_size) {
+			max_size = bufs[child].ia_size;
+			idx = child;
+		}
+
+                if ((num_children == 0) || (bufs[child].ia_size < min_size)) {
+                        min_size = bufs[child].ia_size;
+                }
+
+                num_children++;
+	}
+
+        /* If sizes are same for all of them, finding sources will have to
+         * happen with pending changelog. So return -1
+         */
+        if ((num_children > 1) && (min_size == max_size))
+                return -1;
+	return idx;
+}
+
+
+static int
+afr_find_newest_file (struct iatt *bufs, int32_t *success_children,
+		      int child_count)
+{
+	int idx = -1;
+	int i = -1;
+	int child = -1;
+	uint64_t max_ctime = 0;
+
+	for (i = 0; i < child_count; i++) {
+		if (success_children[i] == -1)
+			break;
+
+		child = success_children[i];
+		if (bufs[child].ia_ctime > max_ctime) {
+			max_ctime = bufs[child].ia_ctime;
+			idx = child;
+		}
+	}
+
+	return idx;
+}
+
+
 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 *success_children,
+                                     int child_count, struct iatt *bufs)
 {
         int32_t       biggest_witness = 0;
         int           nsources        = 0;
@@ -517,6 +639,11 @@ afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix,
 
         GF_ASSERT (child_count > 0);
 
+	biggest_witness = afr_find_largest_file_size (bufs, success_children,
+						      child_count);
+	if (biggest_witness != -1)
+		goto found;
+
         witnesses = GF_CALLOC (child_count, sizeof (*witnesses),
                                gf_afr_mt_int32_t);
         if (NULL == witnesses) {
@@ -529,9 +656,15 @@ afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix,
         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);
+	if (biggest_witness != -1)
+		goto found;
+
+	biggest_witness = afr_find_newest_file (bufs, success_children,
+						child_count);
+
+found:
+	nsources = afr_mark_fool_as_source_by_idx (sources, child_count,
+						   biggest_witness);
 out:
         GF_FREE (witnesses);
         return nsources;
@@ -875,7 +1008,8 @@ afr_mark_sources (xlator_t *this, int32_t *sources, int32_t **pending_matrix,
                 nsources = afr_mark_biggest_of_fools_as_source (sources,
                                                                 pending_matrix,
                                                                 characters,
-                                                                child_count);
+								success_children,
+                                                                child_count, bufs);
         }
 
 out:
@@ -1012,14 +1146,13 @@ afr_sh_missing_entries_done (call_frame_t *frame, xlator_t *this)
 
         afr_sh_reset (frame, this);
 
-        if (local->govinda_gOvinda) {
+        if (local->unhealable) {
                 gf_log (this->name, GF_LOG_DEBUG,
                         "split brain found, aborting selfheal of %s",
                         local->loc.path);
-                sh->op_failed = 1;
         }
 
-        if (sh->op_failed) {
+        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
                 sh->completion_cbk (frame, this);
         } else {
                 gf_log (this->name, GF_LOG_TRACE,
@@ -1251,7 +1384,7 @@ out:
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR, "impunge of %s failed, "
                         "reason: %s", local->loc.path, strerror (-ret));
-                sh->op_failed = 1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
         }
         afr_sh_missing_entries_finish (frame, this);
 }
@@ -1266,7 +1399,7 @@ afr_sh_create_entry_cbk (call_frame_t *frame, xlator_t *this,
         local = frame->local;
         sh = &local->self_heal;
         if (op_ret < 0)
-                sh->op_failed = 1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
         afr_sh_missing_entries_finish (frame, this);
         return 0;
 }
@@ -1290,7 +1423,7 @@ sh_missing_entries_create (call_frame_t *frame, xlator_t *this)
         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_set_local_for_unhealable (local);
                 afr_sh_missing_entries_finish (frame, this);
                 goto out;
         }
@@ -1323,8 +1456,9 @@ afr_sh_missing_entries_lookup_done (call_frame_t *frame, xlator_t *this,
         loc = &local->loc;
 
         if (op_ret < 0) {
-                if (op_errno == EIO)
-                        local->govinda_gOvinda = 1;
+                if (op_errno == EIO) {
+                        afr_set_local_for_unhealable (local);
+                }
                 // EIO can happen if finding the fresh parent dir failed
                 goto out;
         }
@@ -1386,7 +1520,7 @@ afr_sh_missing_entries_lookup_done (call_frame_t *frame, xlator_t *this,
         }
         return;
 out:
-        sh->op_failed = 1;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
         afr_sh_set_error (sh, op_errno);
         afr_sh_missing_entries_finish (frame, this);
         return;
@@ -1470,7 +1604,7 @@ afr_sh_remove_entry_cbk (call_frame_t *frame, xlator_t *this, int child,
                 LOCK (&frame->lock);
                 {
                         afr_sh_set_error (sh, EIO);
-                        sh->op_failed = 1;
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 }
                 UNLOCK (&frame->lock);
         }
@@ -1552,7 +1686,7 @@ afr_sh_purge_stale_entries_done (call_frame_t *frame, xlator_t *this)
         sh       = &local->self_heal;
         priv     = this->private;
 
-        if (sh->op_failed) {
+        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
                 afr_sh_missing_entries_finish (frame, this);
         } else {
                 if (afr_gfid_missing_count (this->name, sh->fresh_children,
@@ -1766,7 +1900,7 @@ afr_sh_children_lookup_done (call_frame_t *frame, xlator_t *this,
                                                priv->child_count, ENOENT);
         if (fresh_child_enoents == fresh_parent_count) {
                 afr_sh_set_error (sh, ENOENT);
-                sh->op_failed = 1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 afr_sh_purge_entry (frame, this);
         } else if (!afr_conflicting_iattrs (sh->buf, sh->fresh_children,
                                             priv->child_count, local->loc.path,
@@ -1780,14 +1914,14 @@ afr_sh_children_lookup_done (call_frame_t *frame, xlator_t *this,
                 afr_sh_purge_stale_entry (frame, this);
         } else {
                 op_errno = EIO;
-                local->govinda_gOvinda = 1;
+                afr_set_local_for_unhealable (local);
                 goto fail;
         }
 
         return;
 
 fail:
-        sh->op_failed = 1;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
         afr_sh_set_error (sh, op_errno);
         afr_sh_missing_entries_finish (frame, this);
         return;
@@ -1858,8 +1992,8 @@ afr_sh_find_fresh_parents (call_frame_t *frame, xlator_t *this,
 
 out:
         afr_sh_set_error (sh, op_errno);
-        sh->op_failed = 1;
-        afr_sh_missing_entries_finish (frame, this);
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+	afr_sh_missing_entries_finish (frame, this);
         return;
 }
 
@@ -1962,7 +2096,7 @@ afr_sh_post_nb_entrylk_missing_entry_sh_cbk (call_frame_t *frame,
         if (int_lock->lock_op_ret < 0) {
                 gf_log (this->name, GF_LOG_INFO,
                         "Non blocking entrylks failed.");
-                sh->op_failed = -1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 afr_sh_missing_entries_done (frame, this);
         } else {
 
@@ -1997,6 +2131,7 @@ afr_sh_entrylk (call_frame_t *frame, xlator_t *this, loc_t *loc,
         int_lock->lk_basename = base_name;
         int_lock->lk_loc      = loc;
         int_lock->lock_cbk    = lock_cbk;
+        int_lock->domain      = this->name;
 
         int_lock->lockee_count = 0;
         afr_init_entry_lockee (&int_lock->lockee[0], local, loc,
@@ -2041,19 +2176,29 @@ out:
 static int
 afr_self_heal_missing_entries (call_frame_t *frame, xlator_t *this)
 {
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+
+        sh->sh_type_in_action  = AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY;
+
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
+
         afr_self_heal_parent_entrylk (frame, this,
-                                   afr_sh_post_nb_entrylk_missing_entry_sh_cbk);
+                                      afr_sh_post_nb_entrylk_missing_entry_sh_cbk);
         return 0;
 }
 
 afr_local_t*
-afr_local_copy (afr_local_t *l, xlator_t *this)
+afr_self_heal_local_init (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;
-        int              i   = 0;
+        afr_private_t   *priv  = NULL;
+        afr_local_t     *lc    = NULL;
+        afr_self_heal_t *sh    = NULL;
+        afr_self_heal_t *shc   = NULL;
+        int             ret    = 0;
 
         priv = this->private;
 
@@ -2076,13 +2221,23 @@ afr_local_copy (afr_local_t *l, xlator_t *this)
         shc->forced_merge = sh->forced_merge;
         shc->background = sh->background;
         shc->type = sh->type;
+        shc->data_sh_info = "";
+        shc->metadata_sh_info =  "";
 
         uuid_copy (shc->sh_gfid_req, sh->sh_gfid_req);
-        if (l->loc.path)
-                loc_copy (&lc->loc, &l->loc);
+        if (l->loc.path) {
+                ret = loc_copy (&lc->loc, &l->loc);
+                if (ret < 0)
+                        goto out;
+        }
 
         lc->child_up  = memdup (l->child_up,
                                 sizeof (*lc->child_up) * priv->child_count);
+        if (!lc->child_up) {
+                ret = -1;
+                goto out;
+        }
+
         if (l->xattr_req)
                 lc->xattr_req = dict_ref (l->xattr_req);
 
@@ -2090,59 +2245,25 @@ afr_local_copy (afr_local_t *l, xlator_t *this)
                 lc->cont.lookup.inode = inode_ref (l->cont.lookup.inode);
         if (l->cont.lookup.xattr)
                 lc->cont.lookup.xattr = dict_ref (l->cont.lookup.xattr);
-        if (l->internal_lock.inode_locked_nodes)
-                lc->internal_lock.inode_locked_nodes =
-                        memdup (l->internal_lock.inode_locked_nodes,
-                                sizeof (*lc->internal_lock.inode_locked_nodes) * priv->child_count);
-        else
-                lc->internal_lock.inode_locked_nodes =
-                        GF_CALLOC (sizeof (*l->internal_lock.inode_locked_nodes),
-                                   priv->child_count,
-                                   gf_afr_mt_char);
-
-        if (l->internal_lock.locked_nodes)
-                lc->internal_lock.locked_nodes =
-                        memdup (l->internal_lock.locked_nodes,
-                                sizeof (*lc->internal_lock.locked_nodes) * priv->child_count);
-        else
-                lc->internal_lock.locked_nodes =
-                        GF_CALLOC (sizeof (*l->internal_lock.locked_nodes),
-                                   priv->child_count,
-                                   gf_afr_mt_char);
-
-        for (i = 0; i < l->internal_lock.lockee_count; i++) {
-                loc_copy (&lc->internal_lock.lockee[i].loc,
-                          &l->internal_lock.lockee[i].loc);
-
-                lc->internal_lock.lockee[i].locked_count =
-                        l->internal_lock.lockee[i].locked_count;
-
-                if (l->internal_lock.lockee[i].basename)
-                        lc->internal_lock.lockee[i].basename =
-                                gf_strdup (l->internal_lock.lockee[i].basename);
-
-                if (l->internal_lock.lockee[i].locked_nodes) {
-                        lc->internal_lock.lockee[i].locked_nodes =
-                                memdup (l->internal_lock.lockee[i].locked_nodes,
-                                        sizeof (*lc->internal_lock.lockee[i].locked_nodes) *
-                                        priv->child_count);
-                } else {
-                        lc->internal_lock.lockee[i].locked_nodes =
-                                GF_CALLOC (priv->child_count,
-                                           sizeof (*lc->internal_lock.lockee[i].locked_nodes),
-                                           gf_afr_mt_char);
-                }
 
+        lc->internal_lock.locked_nodes =
+                             GF_CALLOC (sizeof (*l->internal_lock.locked_nodes),
+                                        priv->child_count, gf_afr_mt_char);
+        if (!lc->internal_lock.locked_nodes) {
+                ret = -1;
+                goto out;
         }
-        lc->internal_lock.lockee_count = l->internal_lock.lockee_count;
-
-        lc->internal_lock.inodelk_lock_count =
-                l->internal_lock.inodelk_lock_count;
-        lc->internal_lock.entrylk_lock_count =
-                l->internal_lock.entrylk_lock_count;
 
+        ret = afr_inodelk_init (&lc->internal_lock.inodelk[0],
+                                this->name, priv->child_count);
+        if (ret)
+                goto out;
 
 out:
+        if (ret) {
+                afr_local_cleanup (lc, this);
+                lc = NULL;
+        }
         return lc;
 }
 
@@ -2155,32 +2276,28 @@ afr_self_heal_completion_cbk (call_frame_t *bgsh_frame, xlator_t *this)
         afr_local_t *     orig_frame_local = NULL;
         afr_self_heal_t * orig_frame_sh = NULL;
         char              sh_type_str[256] = {0,};
+        gf_loglevel_t     loglevel = 0;
 
         priv  = this->private;
         local = bgsh_frame->local;
         sh    = &local->self_heal;
 
-        if (local->govinda_gOvinda) {
+        if (local->unhealable) {
                 afr_set_split_brain (this, sh->inode, SPB, SPB);
-                sh->op_failed = 1;
         }
 
         afr_self_heal_type_str_get (sh, sh_type_str,
                                     sizeof(sh_type_str));
-        if (sh->op_failed) {
-                gf_loglevel_t     loglevel = GF_LOG_ERROR;
-                if (priv->shd.iamshd)
-                        loglevel = GF_LOG_DEBUG;
-
-                gf_log (this->name, loglevel, "background %s self-heal "
-                        "failed on %s", sh_type_str, local->loc.path);
-
+        if (is_self_heal_failed (sh, AFR_CHECK_ALL) && !priv->shd.iamshd) {
+                loglevel = GF_LOG_ERROR;
+        } else if (!is_self_heal_failed (sh, AFR_CHECK_ALL)) {
+                loglevel = GF_LOG_INFO;
         } else {
-                gf_log (this->name, GF_LOG_DEBUG, "background %s self-heal "
-                        "completed on %s", sh_type_str, local->loc.path);
-
+                loglevel = GF_LOG_DEBUG;
         }
 
+        afr_log_self_heal_completion_status (local, loglevel);
+
         FRAME_SU_UNDO (bgsh_frame, afr_local_t);
 
         if (!sh->unwound && sh->unwind) {
@@ -2188,7 +2305,7 @@ afr_self_heal_completion_cbk (call_frame_t *bgsh_frame, xlator_t *this)
                 orig_frame_sh = &orig_frame_local->self_heal;
                 orig_frame_sh->actual_sh_started = _gf_true;
                 sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno,
-                            sh->op_failed);
+                            is_self_heal_failed (sh, AFR_CHECK_ALL));
         }
 
         if (sh->background) {
@@ -2237,7 +2354,7 @@ afr_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode)
         afr_set_lk_owner (sh_frame, this, sh_frame->root);
         afr_set_low_priority (sh_frame);
 
-        sh_local        = afr_local_copy (local, this);
+        sh_local        = afr_self_heal_local_init (local, this);
         if (!sh_local)
                 goto out;
         sh_frame->local = sh_local;
@@ -2302,6 +2419,8 @@ afr_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode)
                 sh->do_gfid_self_heal = _gf_false;
         }
 
+        sh->sh_type_in_action = AFR_SELF_HEAL_INVALID;
+
         FRAME_SU_DO (sh_frame, afr_local_t);
         if (sh->do_missing_entry_self_heal || sh->do_gfid_self_heal) {
                 afr_self_heal_missing_entries (sh_frame, this);
@@ -2511,9 +2630,183 @@ out:
         GF_FREE (erase_xattr);
 
         if (ret < 0) {
-                sh->op_failed = _gf_true;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 finish (frame, this);
         }
 
         return 0;
 }
+
+void
+afr_set_self_heal_status(afr_self_heal_t *sh, afr_self_heal_status status)
+{
+        xlator_t                *this = NULL;
+        afr_sh_status_for_all_type *sh_status = &(sh->afr_all_sh_status);
+        afr_self_heal_type  sh_type_in_action = sh->sh_type_in_action;
+        this = THIS;
+
+        if (!sh) {
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "Null self heal"
+                                  "Structure");
+                goto out;
+        }
+
+        switch (sh_type_in_action) {
+                case AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY:
+                       sh_status->gfid_or_missing_entry_self_heal = status;
+                        break;
+                case AFR_SELF_HEAL_METADATA:
+                        sh_status->metadata_self_heal = status;
+                        break;
+                case AFR_SELF_HEAL_DATA:
+                        sh_status->data_self_heal = status;
+                        break;
+                case AFR_SELF_HEAL_ENTRY:
+                        sh_status->entry_self_heal = status;
+                        break;
+                case AFR_SELF_HEAL_INVALID:
+                        gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid"
+                                          "self heal type in action");
+                        break;
+        }
+out:
+        return;
+}
+
+void
+afr_set_local_for_unhealable (afr_local_t *local)
+{
+        afr_self_heal_t  *sh = NULL;
+
+        sh = &local->self_heal;
+
+        local->unhealable = 1;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+}
+
+int
+is_self_heal_failed (afr_self_heal_t *sh, afr_sh_fail_check_type type)
+{
+        afr_sh_status_for_all_type      sh_status = sh->afr_all_sh_status;
+        afr_self_heal_type   sh_type_in_action =  AFR_SELF_HEAL_INVALID;
+        afr_self_heal_status    status = AFR_SELF_HEAL_FAILED;
+        xlator_t                *this = NULL;
+        int                     sh_failed = 0;
+
+        this = THIS;
+
+        if (!sh) {
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "Null self heal "
+                                  "structure");
+                sh_failed = 1;
+                goto out;
+        }
+
+        if (type == AFR_CHECK_ALL) {
+                if ((sh_status.gfid_or_missing_entry_self_heal == AFR_SELF_HEAL_FAILED)
+                    || (sh_status.metadata_self_heal == AFR_SELF_HEAL_FAILED)
+                    || (sh_status.data_self_heal == AFR_SELF_HEAL_FAILED)
+                    || (sh_status.entry_self_heal == AFR_SELF_HEAL_FAILED))
+                sh_failed = 1;
+        } else if (type == AFR_CHECK_SPECIFIC) {
+                sh_type_in_action = sh->sh_type_in_action;
+                switch (sh_type_in_action) {
+                        case AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY:
+                             status = sh_status.gfid_or_missing_entry_self_heal;
+                                break;
+                        case AFR_SELF_HEAL_METADATA:
+                                status = sh_status.metadata_self_heal;
+                                break;
+                        case AFR_SELF_HEAL_ENTRY:
+                                status = sh_status.entry_self_heal;
+                                break;
+                        case AFR_SELF_HEAL_DATA:
+                                status = sh_status.data_self_heal;
+                                break;
+                        case AFR_SELF_HEAL_INVALID:
+                                status = AFR_SELF_HEAL_NOT_ATTEMPTED;
+                                break;
+                }
+                if (status == AFR_SELF_HEAL_FAILED)
+                        sh_failed = 1;
+
+        }
+
+out:
+        return sh_failed;
+}
+
+char *
+get_sh_completion_status (afr_self_heal_status status)
+{
+
+        char *not_attempted       = " is not attempted";
+        char *failed              = " failed";
+        char *started             = " is started";
+        char *sync_begin          = " is successfully completed";
+        char *result              = " has unknown status";
+
+        switch (status)
+        {
+                case AFR_SELF_HEAL_NOT_ATTEMPTED:
+                        result = not_attempted;
+                        break;
+                case AFR_SELF_HEAL_FAILED:
+                        result = failed;
+                        break;
+                case AFR_SELF_HEAL_STARTED:
+                        result = started;
+                        break;
+                case AFR_SELF_HEAL_SYNC_BEGIN:
+                        result = sync_begin;
+                        break;
+        }
+
+        return result;
+
+}
+
+void
+afr_log_self_heal_completion_status (afr_local_t *local, gf_loglevel_t loglvl)
+{
+
+        char sh_log[4096]              = {0};
+        afr_self_heal_t *sh            = &local->self_heal;
+        afr_sh_status_for_all_type   all_status = sh->afr_all_sh_status;
+        xlator_t      *this            = NULL;
+        size_t        off              = 0;
+        int           data_sh          = 0;
+        int           metadata_sh      = 0;
+        int           print_log        = 0;
+
+        this = THIS;
+
+        ADD_FMT_STRING (sh_log, off, "gfid or missing entry",
+                        all_status.gfid_or_missing_entry_self_heal, print_log);
+        ADD_FMT_STRING_SYNC (sh_log, off, "metadata",
+                             all_status.metadata_self_heal, print_log);
+        if (sh->background) {
+                ADD_FMT_STRING_SYNC (sh_log, off, "backgroung data",
+                                all_status.data_self_heal, print_log);
+        } else {
+                ADD_FMT_STRING_SYNC (sh_log, off, "foreground data",
+                                all_status.data_self_heal, print_log);
+        }
+        ADD_FMT_STRING_SYNC (sh_log, off, "entry", all_status.entry_self_heal,
+                             print_log);
+
+        if (AFR_SELF_HEAL_SYNC_BEGIN == all_status.data_self_heal &&
+	    strcmp (sh->data_sh_info, "") && sh->data_sh_info )
+                data_sh = 1;
+        if (AFR_SELF_HEAL_SYNC_BEGIN == all_status.metadata_self_heal &&
+	    strcmp (sh->metadata_sh_info, "") && sh->metadata_sh_info)
+                metadata_sh = 1;
+
+        if (!print_log)
+                return;
+
+        gf_log (this->name, loglvl, "%s %s %s on %s", sh_log,
+                ((data_sh == 1) ? sh->data_sh_info : ""),
+                ((metadata_sh == 1) ? sh->metadata_sh_info : ""),
+                local->loc.path);
+}
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.h b/xlators/cluster/afr/src/afr-self-heal-common.h
index 035fce543..473264776 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.h
+++ b/xlators/cluster/afr/src/afr-self-heal-common.h
@@ -15,13 +15,6 @@
 #define AFR_SH_MIN_PARTICIPANTS 2
 
 typedef enum {
-        AFR_SELF_HEAL_ENTRY,
-        AFR_SELF_HEAL_METADATA,
-        AFR_SELF_HEAL_DATA,
-        AFR_SELF_HEAL_INVALID = -1,
-} afr_self_heal_type;
-
-typedef enum {
         AFR_LOOKUP_FAIL_CONFLICTS = 1,
         AFR_LOOKUP_FAIL_MISSING_GFIDS = 2,
 } afr_lookup_flags_t;
@@ -98,13 +91,13 @@ int
 afr_sh_entry_impunge_create (call_frame_t *impunge_frame, xlator_t *this,
                              int child_index);
 int
-afr_sh_data_unlock (call_frame_t *frame, xlator_t *this,
+afr_sh_data_unlock (call_frame_t *frame, xlator_t *this, char *dom,
                     afr_lock_cbk_t lock_cbk);
 afr_local_t *
-afr_local_copy (afr_local_t *l, xlator_t *this);
+afr_self_heal_local_init (afr_local_t *l, xlator_t *this);
 int
 afr_sh_data_lock (call_frame_t *frame, xlator_t *this,
-                  off_t start, off_t len, gf_boolean_t block,
+                  off_t start, off_t len, gf_boolean_t block, char *dom,
                   afr_lock_cbk_t success_handler,
                   afr_lock_cbk_t failure_handler);
 void
@@ -133,4 +126,19 @@ int
 afr_sh_erase_pending (call_frame_t *frame, xlator_t *this,
                       afr_transaction_type type, afr_fxattrop_cbk_t cbk,
                       int (*finish)(call_frame_t *frame, xlator_t *this));
+
+void
+afr_set_local_for_unhealable (afr_local_t *local);
+
+int
+is_self_heal_failed (afr_self_heal_t *sh, afr_sh_fail_check_type type);
+
+void
+afr_set_self_heal_status (afr_self_heal_t *sh, afr_self_heal_status status);
+
+void
+afr_log_self_heal_completion_status (afr_local_t *local, gf_loglevel_t  logl);
+
+char*
+afr_get_pending_matrix_str (int32_t *pending_matrix[], xlator_t *this);
 #endif /* __AFR_SELF_HEAL_COMMON_H__ */
diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c
index 2f63ed27d..9de26ee56 100644
--- a/xlators/cluster/afr/src/afr-self-heal-data.c
+++ b/xlators/cluster/afr/src/afr-self-heal-data.c
@@ -156,6 +156,25 @@ afr_sh_data_close (call_frame_t *frame, xlator_t *this)
 }
 
 int
+afr_sh_dom_unlock (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh    = NULL;
+        afr_private_t   *priv  = NULL;
+
+        local = frame->local;
+        sh    = &local->self_heal;
+        priv  = this->private;
+
+        if (sh->sh_dom_lock_held)
+                afr_sh_data_unlock (frame, this, priv->sh_domain,
+                                    afr_sh_data_close);
+        else
+                afr_sh_data_close (frame, this);
+        return 0;
+}
+
+int
 afr_sh_data_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                          int32_t op_ret, int32_t op_errno, struct iatt *statpre,
                          struct iatt *statpost, dict_t *xdata)
@@ -283,31 +302,45 @@ afr_sh_set_timestamps (call_frame_t *frame, xlator_t *this)
 
 //Fun fact, lock_cbk is being used for both lock & unlock
 int
-afr_sh_data_unlock (call_frame_t *frame, xlator_t *this,
+afr_sh_data_unlock (call_frame_t *frame, xlator_t *this, char *dom,
                     afr_lock_cbk_t lock_cbk)
 {
         afr_local_t         *local    = NULL;
         afr_internal_lock_t *int_lock = NULL;
         afr_self_heal_t     *sh       = NULL;
+        afr_private_t       *priv     = NULL;
+        int                 ret       = 0;
 
         local    = frame->local;
         int_lock = &local->internal_lock;
         sh       = &local->self_heal;
+        priv     = this->private;
 
-        GF_ASSERT (sh->data_lock_held);
-
-        sh->data_lock_held = _gf_false;
+        if (strcmp (dom, this->name) == 0) {
+                sh->data_lock_held = _gf_false;
+        } else if (strcmp (dom, priv->sh_domain) == 0) {
+                sh->sh_dom_lock_held = _gf_false;
+        } else {
+                ret = -1;
+                goto out;
+        }
         int_lock->lock_cbk = lock_cbk;
+        int_lock->domain = dom;
         afr_unlock (frame, this);
 
+out:
+        if (ret) {
+                int_lock->lock_op_ret = -1;
+                int_lock->lock_cbk (frame, this);
+        }
         return 0;
 }
 
 int
 afr_sh_data_finish (call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t   *local = NULL;
-        afr_self_heal_t *sh = NULL;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh    = NULL;
 
         local = frame->local;
         sh = &local->self_heal;
@@ -316,9 +349,9 @@ afr_sh_data_finish (call_frame_t *frame, xlator_t *this)
                 "finishing data selfheal of %s", local->loc.path);
 
         if (sh->data_lock_held)
-                afr_sh_data_unlock (frame, this, afr_sh_data_close);
+                afr_sh_data_unlock (frame, this, this->name, afr_sh_dom_unlock);
         else
-                afr_sh_data_close (frame, this);
+                afr_sh_dom_unlock (frame, this);
 
         return 0;
 }
@@ -335,11 +368,8 @@ afr_sh_data_fail (call_frame_t *frame, xlator_t *this)
         gf_log (this->name, GF_LOG_DEBUG,
                 "finishing failed data selfheal of %s", local->loc.path);
 
-        sh->op_failed = 1;
-        if (sh->data_lock_held)
-                afr_sh_data_unlock (frame, this, afr_sh_data_close);
-        else
-                afr_sh_data_close (frame, this);
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        afr_sh_data_finish (frame, this);
         return 0;
 }
 
@@ -362,13 +392,13 @@ afr_sh_data_erase_pending_cbk (call_frame_t *frame, void *cookie,
                         "log failed on %s for subvol %s, reason: %s",
                         local->loc.path, priv->children[child_index]->name,
                         strerror (op_errno));
-                sh->op_failed = 1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
         }
 
         call_count = afr_frame_return (frame);
 
         if (call_count == 0) {
-                if (sh->op_failed) {
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
                         if (sh->old_loop_frame)
                                 sh_loop_finish (sh->old_loop_frame, this);
                         sh->old_loop_frame = NULL;
@@ -380,7 +410,7 @@ afr_sh_data_erase_pending_cbk (call_frame_t *frame, void *cookie,
                         goto out;
                 }
                 GF_ASSERT (sh->old_loop_frame);
-                afr_sh_data_lock (frame, this, 0, 0, _gf_true,
+                afr_sh_data_lock (frame, this, 0, 0, _gf_true, this->name,
                                   afr_post_sh_big_lock_success,
                                   afr_post_sh_big_lock_failure);
         }
@@ -418,7 +448,7 @@ afr_sh_data_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         priv->children[child_index]->name, strerror (op_errno));
                 LOCK (&frame->lock);
                 {
-                        sh->op_failed = 1;
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 }
                 UNLOCK (&frame->lock);
                 if (sh->old_loop_frame)
@@ -428,7 +458,7 @@ afr_sh_data_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         call_count = afr_frame_return (frame);
         if (call_count == 0) {
-                if (sh->op_failed)
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC))
                         afr_sh_data_fail (frame, this);
                 else
                         afr_sh_data_erase_pending (frame, this);
@@ -604,7 +634,7 @@ afr_sh_data_trim_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                 local->loc.path,
                                 priv->children[child_index]->name,
                                 strerror (op_errno));
-                        sh->op_failed = 1;
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 } else {
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "ftruncate of %s on subvolume %s completed",
@@ -617,7 +647,7 @@ afr_sh_data_trim_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         call_count = afr_frame_return (frame);
 
         if (call_count == 0) {
-                if (sh->op_failed)
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC))
                         afr_sh_data_fail (frame, this);
                 else
                         afr_sh_data_sync_prepare (frame, this);
@@ -697,6 +727,199 @@ out:
         return ret;
 }
 
+char*
+afr_get_sizes_str (afr_local_t *local, struct iatt *bufs, xlator_t *this)
+{
+        afr_private_t *priv = NULL;
+        int           i     = 0;
+        char          num[1024] = {0};
+        size_t        len = 0;
+        char          *sizes_str = NULL;
+        size_t        off = 0;
+        char          *fmt_str = "%llu bytes on %s, ";
+        char          *child_down =  " %s,";
+        char          *child_unknown = " %s,";
+        int           down_child_present = 0;
+        int           down_count = 0;
+        int           unknown_count = 0;
+        int           unknown_child_present = 0;
+        char          *down_subvol_1 = " down subvolume is ";
+        char          *unknown_subvol_1 = " unknown subvolume is ";
+        char          *down_subvol_2 = " down subvolumes are ";
+        char          *unknown_subvol_2 = " unknown subvolumes are ";
+
+        priv = this->private;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == 1) {
+                        len += snprintf (num, sizeof (num), fmt_str,
+                                         (unsigned long long) bufs[i].ia_size,
+                                         priv->children[i]->name);
+                } else if (local->child_up[i] == 0) {
+                        len += snprintf (num, sizeof (num), child_down,
+                                         priv->children[i]->name);
+                        if (!down_child_present)
+                                down_child_present = 1;
+                        down_count ++;
+                } else if (local->child_up[i] == -1) {
+                        len += snprintf (num, sizeof (num), child_unknown,
+                                         priv->children[i]->name);
+                        if (!unknown_child_present)
+                                unknown_child_present = 1;
+                        unknown_count++;
+                }
+
+        }
+
+        if (down_child_present) {
+                if (down_count > 1)
+                        len += snprintf (num, sizeof (num), "%s",
+                                         down_subvol_2);
+                else
+                        len += snprintf (num, sizeof (num), "%s",
+                                        down_subvol_1);
+        }
+        if (unknown_child_present) {
+                if (unknown_count > 1)
+                        len += snprintf (num, sizeof (num), "%s",
+                                         unknown_subvol_2);
+                else
+                        len += snprintf (num, sizeof (num), "%s",
+                                         unknown_subvol_1);
+        }
+
+        len++;//for '\0'
+
+        sizes_str = GF_CALLOC (len, sizeof (char), gf_common_mt_char);
+
+        if (!sizes_str)
+                return NULL;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == 1) {
+                        off += snprintf (sizes_str + off, len - off, fmt_str,
+                                         (unsigned long long) bufs[i].ia_size,
+                                         priv->children[i]->name);
+                }
+        }
+
+        if (down_child_present) {
+                if (down_count > 1) {
+                        off += snprintf (sizes_str + off, len - off, "%s",
+                                         down_subvol_2);
+                } else {
+                        off += snprintf (sizes_str + off, len - off, "%s",
+                                         down_subvol_1);
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == 0) {
+                        off += snprintf (sizes_str + off, len - off, child_down,
+                                         priv->children[i]->name);
+                }
+        }
+
+        if (unknown_child_present) {
+                if (unknown_count > 1) {
+                        off += snprintf (sizes_str + off, len - off, "%s",
+                                        unknown_subvol_2);
+                } else {
+                        off += snprintf (sizes_str + off, len - off, "%s",
+                                         unknown_subvol_1);
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == -1) {
+                        off += snprintf (sizes_str + off, len - off,
+                                         child_unknown,
+                                         priv->children[i]->name);
+
+                }
+        }
+
+        return sizes_str;
+}
+
+char*
+afr_get_sinks_str (xlator_t *this, afr_local_t *local, afr_self_heal_t *sh)
+{
+        afr_private_t   *priv = NULL;
+        int             i = 0;
+        char            num[1024] = {0};
+        size_t          len = 0;
+        char            *sinks_str = NULL;
+        char            *temp_str = " to sinks ";
+        char            *str_format = " %s,";
+        char            off = 0;
+
+        priv = this->private;
+
+        len += snprintf (num, sizeof (num), "%s", temp_str);
+        for (i = 0; i < priv->child_count; i++) {
+                if ((sh->sources[i] == 0) && (local->child_up[i] == 1)) {
+                        len += snprintf (num, sizeof (num), str_format,
+                                         priv->children[i]->name);
+                }
+        }
+
+        len ++;
+
+        sinks_str = GF_CALLOC (len, sizeof (char), gf_common_mt_char);
+
+        if (!sinks_str)
+                return NULL;
+
+        off += snprintf (sinks_str + off, len - off, "%s", temp_str);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if ((sh->sources[i] == 0) && (local->child_up[i] == 1)) {
+                        off += snprintf (sinks_str + off, len - off,
+                                         str_format,
+                                         priv->children[i]->name);
+                }
+        }
+
+        return sinks_str;
+
+}
+
+
+void
+afr_set_data_sh_info_str (afr_local_t *local, afr_self_heal_t *sh, xlator_t *this)
+{
+        char            *pending_matrix_str = NULL;
+        char            *sizes_str = NULL;
+        char            *sinks_str = NULL;
+        afr_private_t   *priv = NULL;
+
+        priv = this->private;
+
+        pending_matrix_str = afr_get_pending_matrix_str (sh->pending_matrix,
+                                                         this);
+        if (!pending_matrix_str)
+                pending_matrix_str = "";
+
+        sizes_str = afr_get_sizes_str (local, sh->buf, this);
+        if (!sizes_str)
+                sizes_str = "";
+
+        sinks_str = afr_get_sinks_str (this, local, sh);
+        if (!sinks_str)
+                sinks_str = "";
+
+        gf_asprintf (&sh->data_sh_info, " data self heal from %s %s with "
+                     "%s data %s", priv->children[sh->source]->name, sinks_str,
+                     sizes_str, pending_matrix_str);
+
+        if (pending_matrix_str && strcmp (pending_matrix_str, ""))
+                GF_FREE (pending_matrix_str);
+
+        if (sizes_str && strcmp (sizes_str, ""))
+                GF_FREE (sizes_str);
+}
+
 void
 afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
 {
@@ -718,7 +941,7 @@ afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
 
         if (sh->background && sh->unwind && !sh->unwound) {
                 sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno,
-                            sh->op_failed);
+                            is_self_heal_failed (sh, AFR_CHECK_SPECIFIC));
                 sh->unwound = _gf_true;
         }
 
@@ -737,6 +960,7 @@ afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
                 sh->active_sinks);
 
         sh->actual_sh_started = _gf_true;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN);
         afr_sh_data_trim_sinks (frame, this);
 }
 
@@ -817,6 +1041,7 @@ afr_sh_data_fxattrop_fstat_done (call_frame_t *frame, xlator_t *this)
                 }
                 afr_sh_data_setattr (frame, this, &sh->buf[tstamp_source]);
         } else {
+                afr_set_data_sh_info_str (local, sh, this);
                 if (nsources == 0) {
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "No self-heal needed for %s",
@@ -1174,6 +1399,22 @@ afr_sh_data_big_lock_success (call_frame_t *frame, xlator_t *this)
 }
 
 int
+afr_sh_dom_lock_success (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+
+        sh->sh_dom_lock_held = _gf_true;
+        afr_sh_data_lock (frame, this, 0, 0, _gf_true, this->name,
+                          afr_sh_data_big_lock_success,
+                          afr_sh_data_fail);
+        return 0;
+}
+
+int
 afr_sh_data_post_blocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
 {
         afr_internal_lock_t *int_lock = NULL;
@@ -1237,9 +1478,11 @@ afr_sh_data_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
 }
 
 int
-afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this, off_t start, off_t len)
+afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this, char *dom,
+                      off_t start, off_t len)
 {
         afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
         afr_local_t         *local    = NULL;
 
         local    = frame->local;
@@ -1250,11 +1493,14 @@ afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this, off_t start, off_t le
 
         afr_set_lock_number (frame, this);
 
-        int_lock->lk_flock.l_start = start;
-        int_lock->lk_flock.l_len   = len;
-        int_lock->lk_flock.l_type  = F_WRLCK;
         int_lock->lock_cbk         = afr_sh_data_post_nonblocking_inodelk_cbk;
 
+        int_lock->domain = dom;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+        inodelk->flock.l_start = start;
+        inodelk->flock.l_len   = len;
+        inodelk->flock.l_type  = F_WRLCK;
+
         afr_nonblocking_inodelk (frame, this);
 
         return 0;
@@ -1298,7 +1544,7 @@ afr_post_sh_big_lock_failure (call_frame_t *frame, xlator_t *this)
 int
 afr_sh_data_lock (call_frame_t *frame, xlator_t *this,
                   off_t start, off_t len, gf_boolean_t block,
-                  afr_lock_cbk_t success_handler,
+                  char *dom, afr_lock_cbk_t success_handler,
                   afr_lock_cbk_t failure_handler)
 {
         afr_local_t *   local = NULL;
@@ -1310,7 +1556,7 @@ afr_sh_data_lock (call_frame_t *frame, xlator_t *this,
         sh->data_lock_success_handler = success_handler;
         sh->data_lock_failure_handler = failure_handler;
 	sh->data_lock_block = block;
-        return afr_sh_data_lock_rec (frame, this, start, len);
+        return afr_sh_data_lock_rec (frame, this, dom, start, len);
 }
 
 int
@@ -1322,7 +1568,6 @@ afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         afr_private_t   *priv = NULL;
         int              call_count = 0;
         int              child_index = 0;
-	gf_boolean_t	 block = _gf_true;
 
         local = frame->local;
         sh = &local->self_heal;
@@ -1342,7 +1587,7 @@ afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                 local->loc.path,
                                 priv->children[child_index]->name,
                                 strerror (op_errno));
-                        sh->op_failed = 1;
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 } else {
                         gf_log (this->name, GF_LOG_TRACE,
                                 "open of %s succeeded on child %s",
@@ -1355,7 +1600,7 @@ afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         call_count = afr_frame_return (frame);
 
         if (call_count == 0) {
-                if (sh->op_failed) {
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
                         afr_sh_data_fail (frame, this);
                         return 0;
                 }
@@ -1364,15 +1609,8 @@ afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         "fd for %s opened, commencing sync",
                         local->loc.path);
 
-		/*
-		 * The read and write self-heal trigger codepaths do not provide
-		 * an unwind callback. We run a trylock in these codepaths
-		 * because we are sensitive to locking latency.
-		 */
-		block = sh->unwind ? _gf_true : _gf_false;
-                afr_sh_data_lock (frame, this, 0, 0, block,
-                                  afr_sh_data_big_lock_success,
-                                  afr_sh_data_fail);
+                afr_sh_data_lock (frame, this, 0, 0, _gf_true, priv->sh_domain,
+                                  afr_sh_dom_lock_success, afr_sh_data_fail);
         }
 
         return 0;
@@ -1477,18 +1715,31 @@ afr_can_start_data_self_heal (afr_self_heal_t *sh, afr_private_t *priv)
 int
 afr_self_heal_data (call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t   *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t *priv = this->private;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh    = NULL;
+        afr_private_t   *priv  = this->private;
+        int             ret    = -1;
 
         local = frame->local;
         sh = &local->self_heal;
 
+        sh->sh_type_in_action = AFR_SELF_HEAL_DATA;
+
         if (afr_can_start_data_self_heal (sh, priv)) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
+                ret = afr_inodelk_init (&local->internal_lock.inodelk[1],
+                                        priv->sh_domain, priv->child_count);
+                if (ret < 0) {
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                        afr_sh_data_done (frame, this);
+                        return 0;
+                }
+
                 if (IA_ISREG (sh->type)) {
                         afr_sh_data_open (frame, this);
                 } else {
                         afr_sh_data_lock (frame, this, 0, 0, _gf_true,
+                                          this->name,
                                           afr_sh_non_reg_lock_success,
                                           afr_sh_data_fail);
                 }
diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c
index c3c9f9fca..53491a1d7 100644
--- a/xlators/cluster/afr/src/afr-self-heal-entry.c
+++ b/xlators/cluster/afr/src/afr-self-heal-entry.c
@@ -162,7 +162,7 @@ afr_sh_entry_erase_pending (call_frame_t *frame, xlator_t *this)
         sh = &local->self_heal;
 
         if (sh->entries_skipped) {
-                sh->op_failed = _gf_true;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 goto out;
         }
         afr_sh_erase_pending (frame, this, AFR_ENTRY_TRANSACTION,
@@ -613,6 +613,19 @@ out:
         return 0;
 }
 
+static gf_boolean_t
+can_skip_entry_self_heal (char *name, loc_t *parent_loc)
+{
+        if (strcmp (name, ".") == 0) {
+                return _gf_true;
+        } else if (strcmp (name, "..") == 0) {
+                return _gf_true;
+        } else if (loc_is_root (parent_loc) &&
+                   (strcmp (name, GF_REPLICATE_TRASH_DIR) == 0)) {
+                return _gf_true;
+        }
+        return _gf_false;
+}
 
 int
 afr_sh_entry_expunge_entry (call_frame_t *frame, xlator_t *this,
@@ -640,13 +653,7 @@ afr_sh_entry_expunge_entry (call_frame_t *frame, xlator_t *this,
         sh->expunge_done = afr_sh_entry_expunge_entry_done;
 
         name = entry->d_name;
-
-        if ((strcmp (name, ".") == 0)
-            || (strcmp (name, "..") == 0)) {
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "skipping inspection of %s under %s",
-                        name, local->loc.path);
+        if (can_skip_entry_self_heal (name, &local->loc)) {
                 op_ret = 0;
                 goto out;
         }
@@ -799,7 +806,7 @@ afr_sh_entry_expunge_all (call_frame_t *frame, xlator_t *this)
         active_src = next_active_sink (frame, this, sh->active_source);
         sh->active_source = active_src;
 
-        if (sh->op_failed) {
+        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
                 goto out;
         }
 
@@ -1256,6 +1263,35 @@ afr_sh_entry_impunge_mknod (call_frame_t *impunge_frame, xlator_t *this,
                 gf_log (this->name, GF_LOG_INFO, "%s: gfid set failed",
                         impunge_local->loc.path);
 
+        /*
+         * Reason for adding GLUSTERFS_INTERNAL_FOP_KEY :
+         *
+         * Problem:
+         * While a brick is down in a replica pair, lets say the user creates
+         * one file(file-A) and a hard link to that file(h-file-A). After the
+         * brick comes back up, entry self-heal is attempted on parent dir of
+         * these two files. As part of readdir in self-heal it reads both the
+         * entries file-A and h-file-A for both of them it does name less lookup
+         * to check if there are any hardlinks already present in the
+         * destination brick. It finds that there are no hard links already
+         * present for files file-A, h-file-A. Self-heal does mknods for both
+         * file-A and h-file-A. This leads to file-A and h-file-A not being
+         * hardlinks anymore.
+         *
+         * Fix: (More like shrinking of race-window, the race itself is still
+         * present in posix-mknod).
+         * If mknod comes with the presence of GLUSTERFS_INTERNAL_FOP_KEY then
+         * posix_mknod checks if there are already any gfid-links and does
+         * link() instead of mknod. There still can be a race where two
+         * posix_mknods same gfid see that
+         * gfid-link file is not present and proceeds with mknods and result in
+         * two different files with same gfid.
+         */
+        ret = dict_set_str (dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+        if (ret)
+                gf_log (this->name, GF_LOG_INFO, "%s: %s set failed",
+                        impunge_local->loc.path, GLUSTERFS_INTERNAL_FOP_KEY);
+
         STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
                            (void *) (long) child_index,
                            priv->children[child_index],
@@ -1872,12 +1908,7 @@ afr_sh_entry_impunge_entry (call_frame_t *frame, xlator_t *this,
         active_src = sh->active_source;
         sh->impunge_done = afr_sh_entry_impunge_entry_done;
 
-        if ((strcmp (entry->d_name, ".") == 0)
-            || (strcmp (entry->d_name, "..") == 0)) {
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "skipping inspection of %s under %s",
-                        entry->d_name, local->loc.path);
+        if (can_skip_entry_self_heal (entry->d_name, &local->loc)) {
                 op_ret = 0;
                 goto out;
         }
@@ -1946,7 +1977,7 @@ afr_sh_entry_impunge_readdir_cbk (call_frame_t *frame, void *cookie,
                                 local->loc.path,
                                 priv->children[active_src]->name,
                                 strerror (op_errno));
-                        sh->op_failed = 1;
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 } else {
                         gf_log (this->name, GF_LOG_TRACE,
                                 "readdir of %s on subvolume %s complete",
@@ -2019,7 +2050,7 @@ afr_sh_entry_impunge_all (call_frame_t *frame, xlator_t *this)
         active_src = next_active_source (frame, this, sh->active_source);
         sh->active_source = active_src;
 
-        if (sh->op_failed) {
+        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
                 afr_sh_entry_finish (frame, this);
                 return 0;
         }
@@ -2068,7 +2099,7 @@ afr_sh_entry_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                                 local->loc.path,
                                 priv->children[child_index]->name,
                                 strerror (op_errno));
-                        sh->op_failed = 1;
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 }
         }
         UNLOCK (&frame->lock);
@@ -2076,7 +2107,7 @@ afr_sh_entry_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         call_count = afr_frame_return (frame);
 
         if (call_count == 0) {
-                if (sh->op_failed) {
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
                         afr_sh_entry_finish (frame, this);
                         return 0;
                 }
@@ -2209,6 +2240,7 @@ afr_sh_entry_sync_prepare (call_frame_t *frame, xlator_t *this)
                         local->loc.path);
 
         sh->actual_sh_started = _gf_true;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN);
         afr_sh_entry_open (frame, this);
 
         return 0;
@@ -2231,7 +2263,7 @@ afr_sh_entry_fix (call_frame_t *frame, xlator_t *this,
         priv = this->private;
 
         if (op_ret < 0) {
-                sh->op_failed = 1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 afr_sh_set_error (sh, op_errno);
                 afr_sh_entry_finish (frame, this);
                 goto out;
@@ -2294,7 +2326,7 @@ afr_sh_post_nonblocking_entry_cbk (call_frame_t *frame, xlator_t *this)
         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_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 afr_sh_entry_done (frame, this);
         } else {
 
@@ -2313,14 +2345,18 @@ afr_sh_post_nonblocking_entry_cbk (call_frame_t *frame, xlator_t *this)
 int
 afr_self_heal_entry (call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t   *local = NULL;
+        afr_local_t     *local = NULL;
         afr_private_t   *priv = NULL;
-
+        afr_self_heal_t *sh = NULL;
 
         priv = this->private;
         local = frame->local;
+        sh = &local->self_heal;
+
+        sh->sh_type_in_action = AFR_SELF_HEAL_ENTRY;
 
         if (local->self_heal.do_entry_self_heal && priv->entry_self_heal) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
                 afr_sh_entrylk (frame, this, &local->loc, NULL,
                                 afr_sh_post_nonblocking_entry_cbk);
         } else {
diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c
index cc85d9b9f..fd5da6cfd 100644
--- a/xlators/cluster/afr/src/afr-self-heal-metadata.c
+++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c
@@ -97,7 +97,7 @@ afr_sh_metadata_fail (call_frame_t *frame, xlator_t *this)
         local    = frame->local;
         sh       = &local->self_heal;
 
-        sh->op_failed = 1;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
         afr_sh_metadata_finish (frame, this);
         return 0;
 }
@@ -406,6 +406,140 @@ afr_sh_metadata_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         return 0;
 }
 
+static void
+afr_set_metadata_sh_info_str (afr_local_t *local, afr_self_heal_t *sh,
+                              xlator_t *this)
+{
+        afr_private_t    *priv = NULL;
+        int              i = 0;
+        char             num[1024] = {0};
+        size_t           len = 0;
+        char             *string = NULL;
+        size_t           off = 0;
+        char             *source_child =  " from source %s to";
+        char             *format = " %s, ";
+        char             *string_msg = " metadata self heal";
+        char             *pending_matrix_str = NULL;
+        int              down_child_present = 0;
+        int              unknown_child_present = 0;
+        char             *down_subvol_1 = " down subvolume is ";
+        char             *unknown_subvol_1 = " unknown subvolume is";
+        char             *down_subvol_2 = " down subvolumes are ";
+        char             *unknown_subvol_2 = " unknown subvolumes are ";
+        int              down_count = 0;
+        int              unknown_count = 0;
+
+        priv = this->private;
+
+        pending_matrix_str = afr_get_pending_matrix_str (sh->pending_matrix,
+                                                         this);
+
+        if (!pending_matrix_str)
+                pending_matrix_str = "";
+
+        len += snprintf (num, sizeof (num), "%s", string_msg);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if ((sh->source == i) && (local->child_up[i] == 1)) {
+                        len += snprintf (num, sizeof (num), source_child,
+                                         priv->children[i]->name);
+                } else if ((local->child_up[i] == 1) && (sh->sources[i] == 0)) {
+                        len += snprintf (num, sizeof (num), format,
+                                         priv->children[i]->name);
+                } else if (local->child_up[i] == 0) {
+                        len += snprintf (num, sizeof (num), format,
+                                         priv->children[i]->name);
+                        if (!down_child_present)
+                                down_child_present = 1;
+                        down_count++;
+                } else if (local->child_up[i] == -1) {
+                        len += snprintf (num, sizeof (num), format,
+                                         priv->children[i]->name);
+                        if (!unknown_child_present)
+                                unknown_child_present = 1;
+                        unknown_count++;
+                }
+        }
+
+        if (down_child_present) {
+                if (down_count > 1) {
+                        len += snprintf (num, sizeof (num), "%s",
+                                         down_subvol_2);
+                } else {
+                        len += snprintf (num, sizeof (num), "%s",
+                                         down_subvol_1);
+                }
+        }
+        if (unknown_child_present) {
+                if (unknown_count > 1) {
+                        len += snprintf (num, sizeof (num), "%s",
+                                         unknown_subvol_2);
+                } else {
+                        len += snprintf (num, sizeof (num), "%s",
+                                         unknown_subvol_1);
+                }
+        }
+
+        len ++;
+
+        string = GF_CALLOC (len, sizeof (char), gf_common_mt_char);
+        if (!string)
+                return;
+
+        off += snprintf (string + off, len - off, "%s", string_msg);
+        for (i=0; i < priv->child_count; i++) {
+                if ((sh->source == i) && (local->child_up[i] == 1))
+                        off += snprintf (string + off, len - off, source_child,
+                                         priv->children[i]->name);
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if ((local->child_up[i] == 1)&& (sh->sources[i] == 0))
+                        off += snprintf (string + off, len - off, format,
+                                         priv->children[i]->name);
+        }
+
+        if (down_child_present) {
+                if (down_count > 1) {
+                        off += snprintf (string + off, len - off, "%s",
+                                         down_subvol_2);
+                } else {
+                        off += snprintf (string + off, len - off, "%s",
+                                         down_subvol_1);
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == 0)
+                        off += snprintf (string + off, len - off, format,
+                                         priv->children[i]->name);
+        }
+
+        if (unknown_child_present) {
+                if (unknown_count > 1) {
+                        off += snprintf (string + off, len - off, "%s",
+                                 unknown_subvol_2);
+                } else {
+                        off += snprintf (string + off, len - off, "%s",
+                                         unknown_subvol_1);
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == -1)
+                        off += snprintf (string + off, len - off, format,
+                                         priv->children[i]->name);
+        }
+
+        gf_asprintf (&sh->metadata_sh_info, "%s metadata %s,", string,
+                     pending_matrix_str);
+
+        if (pending_matrix_str && strcmp (pending_matrix_str, ""))
+                GF_FREE (pending_matrix_str);
+
+        if (string && strcmp (string, ""))
+                GF_FREE (string);
+}
 
 int
 afr_sh_metadata_sync_prepare (call_frame_t *frame, xlator_t *this)
@@ -436,6 +570,8 @@ afr_sh_metadata_sync_prepare (call_frame_t *frame, xlator_t *this)
                 sh->active_sinks);
 
         sh->actual_sh_started = _gf_true;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN);
+        afr_set_metadata_sh_info_str (local, sh, this);
         STACK_WIND (frame, afr_sh_metadata_getxattr_cbk,
                     priv->children[source],
                     priv->children[source]->fops->getxattr,
@@ -461,7 +597,7 @@ afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this,
         priv = this->private;
 
         if (op_ret < 0) {
-                sh->op_failed = 1;
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
                 afr_sh_set_error (sh, op_errno);
                 afr_sh_metadata_finish (frame, this);
                 goto out;
@@ -579,19 +715,22 @@ int
 afr_sh_metadata_lock (call_frame_t *frame, xlator_t *this)
 {
         afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
         afr_local_t         *local    = NULL;
 
         local    = frame->local;
         int_lock = &local->internal_lock;
 
+        int_lock->domain           = this->name;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
         int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
         int_lock->selfheal_lk_type    = AFR_METADATA_SELF_HEAL_LK;
 
         afr_set_lock_number (frame, this);
 
-        int_lock->lk_flock.l_start = LLONG_MAX - 1;
-        int_lock->lk_flock.l_len   = 0;
-        int_lock->lk_flock.l_type  = F_WRLCK;
+        inodelk->flock.l_start = LLONG_MAX - 1;
+        inodelk->flock.l_len   = 0;
+        inodelk->flock.l_type  = F_WRLCK;
         int_lock->lock_cbk         = afr_sh_metadata_post_nonblocking_inodelk_cbk;
 
         afr_nonblocking_inodelk (frame, this);
@@ -618,8 +757,10 @@ afr_self_heal_metadata (call_frame_t *frame, xlator_t *this)
 
         local = frame->local;
         sh = &local->self_heal;
+        sh->sh_type_in_action = AFR_SELF_HEAL_METADATA;
 
         if (afr_can_start_metadata_self_heal (sh, priv)) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
                 afr_sh_metadata_lock (frame, this);
         } else {
                 afr_sh_metadata_done (frame, this);
diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c
index f33b04eed..1b48a1bca 100644
--- a/xlators/cluster/afr/src/afr-self-heald.c
+++ b/xlators/cluster/afr/src/afr-self-heald.c
@@ -25,7 +25,8 @@ typedef enum {
 
 typedef enum {
         HEAL = 1,
-        INFO
+        INFO,
+        STATISTICS_TO_BE_HEALED,
 } shd_crawl_op;
 
 typedef struct shd_dump {
@@ -84,6 +85,33 @@ _loc_assign_gfid_path (loc_t *loc)
 }
 
 void
+_destroy_crawl_event_data (void *data)
+{
+        shd_crawl_event_t        *crawl_event = NULL;
+
+        if (!data)
+                goto out;
+
+        crawl_event = (shd_crawl_event_t *)data;
+        GF_FREE (crawl_event->start_time_str);
+        GF_FREE (crawl_event->end_time_str);
+
+out:
+        return;
+}
+
+void
+_destroy_shd_event_data (void *data)
+{
+        shd_event_t             *event = NULL;
+        if (!data)
+                goto out;
+        event = (shd_event_t*)data;
+        GF_FREE (event->path);
+out:
+        return;
+}
+void
 shd_cleanup_event (void *event)
 {
         shd_event_t *shd_event = event;
@@ -128,6 +156,123 @@ _build_index_loc (xlator_t *this, loc_t *loc, char *name, loc_t *parent)
 }
 
 int
+_add_crawl_stats_to_dict (xlator_t *this, dict_t *output, int child,
+                          shd_crawl_event_t *shd_event, struct timeval *tv)
+{
+        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 = NULL;
+        char            *end_time_str = NULL;
+        char            *crawl_type = NULL;
+        int             progress = -1;
+
+        healed_count = shd_event->healed_count;
+        split_brain_count = shd_event->split_brain_count;
+        heal_failed_count = shd_event->heal_failed_count;
+        start_time_str = shd_event->start_time_str;
+        end_time_str = shd_event->end_time_str;
+        crawl_type = shd_event->crawl_type;
+
+        if (!start_time_str) {
+                ret = -1;
+                goto out;
+        }
+
+
+        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_dynstr (output, key, gf_strdup (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, gf_strdup(start_time_str));
+
+         if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_"
+                        "crawl_start_time to outout");
+                goto out;
+        }
+
+        snprintf (key, sizeof (key), "statistics_end_time-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+
+        if (!end_time_str)
+                end_time_str = "Could not determine the end time";
+        ret = dict_set_dynstr (output, key, gf_strdup(end_time_str));
+         if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_"
+                        "crawl_end_time to outout");
+                goto out;
+        }
+        snprintf (key, sizeof (key), "statistics_inprogress-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+
+        if (shd_event->crawl_inprogress == _gf_true)
+                progress = 1;
+        else
+                progress = 0;
+
+        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:
+        return ret;
+}
+
+int
 _add_path_to_dict (xlator_t *this, dict_t *output, int child, char *path,
                    struct timeval *tv, gf_boolean_t dyn)
 {
@@ -233,6 +378,20 @@ out:
 }
 
 int
+_add_crawl_event_statistics_to_dict (circular_buffer_t *cb, void *data)
+{
+        int               ret = 0;
+        shd_dump_t        *dump_data = NULL;
+        shd_crawl_event_t *shd_event = NULL;
+
+        dump_data = data;
+        shd_event = cb->data;
+        ret = _add_crawl_stats_to_dict (dump_data->this, dump_data->dict,
+                                        dump_data->child, shd_event, &cb->tv);
+        return ret;
+}
+
+int
 _add_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict, int child)
 {
         shd_dump_t dump_data = {0};
@@ -244,6 +403,26 @@ _add_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict, int child)
         return 0;
 }
 
+
+int
+_add_statistics_to_dict (xlator_t *this, dict_t *dict, int child)
+{
+        shd_dump_t              dump_data = {0};
+        afr_private_t           *priv  = NULL;
+        afr_self_heald_t        *shd = NULL;
+
+        priv = this->private;
+        shd = &priv->shd;
+
+        dump_data.this = this;
+        dump_data.dict = dict;
+        dump_data.child = child;
+        eh_dump (shd->statistics[child], &dump_data,
+                 _add_crawl_event_statistics_to_dict);
+        return 0;
+
+}
+
 void
 _remove_stale_index (xlator_t *this, xlator_t *readdir_xl,
                      loc_t *parent, char *fname)
@@ -269,6 +448,46 @@ out:
 }
 
 int
+_count_hard_links_under_base_indices_dir (xlator_t *this,
+                                           afr_crawl_data_t *crawl_data,
+                                           gf_dirent_t *entry, loc_t *childloc,
+                                           loc_t *parentloc, struct iatt *iattr)
+{
+        xlator_t                *readdir_xl = crawl_data->readdir_xl;
+        struct iatt             parent = {0};
+        int                     ret = 0;
+        dict_t                  *output = NULL;
+        int                     xl_id =  0;
+        char                    key[256] = {0};
+        int                     child  = -1;
+        uint64_t                     hardlinks = 0;
+
+        output = crawl_data->op_data;
+        child = crawl_data->child;
+
+        ret = syncop_lookup (readdir_xl, childloc, NULL, iattr, NULL, &parent);
+        if (ret)
+                goto out;
+
+        ret = dict_get_int32 (output, this->name, &xl_id);
+        if (ret)
+                goto out;
+
+        snprintf (key, sizeof (key), "%d-%d-hardlinks", xl_id, child);
+        ret =  dict_get_uint64 (output, key, &hardlinks);
+
+        /*Removing the count of base_entry under indices/base_indicies and
+         * entry under indices/xattrop */
+        hardlinks = hardlinks + iattr->ia_nlink - 2;
+        ret = dict_set_uint64 (output, key, hardlinks);
+        if (ret)
+                goto out;
+
+out:
+        return ret;
+}
+
+int
 _add_summary_to_dict (xlator_t *this, afr_crawl_data_t *crawl_data,
                       gf_dirent_t *entry,
                       loc_t *childloc, loc_t *parentloc, struct iatt *iattr)
@@ -307,16 +526,18 @@ _crawl_post_sh_action (xlator_t *this, loc_t *parent, loc_t *child,
                        int32_t op_ret, int32_t op_errno, dict_t *xattr_rsp,
                        afr_crawl_data_t *crawl_data)
 {
-        int              ret = 0;
-        afr_private_t    *priv = NULL;
-        afr_self_heald_t *shd = NULL;
-        eh_t             *eh = NULL;
-        char             *path = NULL;
-        char             gfid_str[64] = {0};
-        shd_event_t      *event = NULL;
-        int32_t          sh_failed = 0;
-        gf_boolean_t     split_brain = 0;
-        int32_t          actual_sh_done = 0;
+        int                ret = 0;
+        afr_private_t      *priv = NULL;
+        afr_self_heald_t   *shd = NULL;
+        eh_t               *eh = NULL;
+        char               *path = NULL;
+        char               gfid_str[64] = {0};
+        shd_event_t        *event = NULL;
+        int32_t            sh_failed = 0;
+        gf_boolean_t       split_brain = 0;
+        int32_t            actual_sh_done = 0;
+        shd_crawl_event_t  **shd_crawl_event = NULL;
+
         priv = this->private;
         shd  = &priv->shd;
         if (crawl_data->crawl == INDEX) {
@@ -343,16 +564,19 @@ _crawl_post_sh_action (xlator_t *this, loc_t *parent, loc_t *child,
                 ret = dict_get_int32 (xattr_rsp, "actual-sh-done", &actual_sh_done);
         }
 
-        split_brain = afr_is_split_brain (this, child->inode);
+        shd_crawl_event = (shd_crawl_event_t**)(shd->crawl_events);
 
+        split_brain = afr_is_split_brain (this, child->inode);
         if ((op_ret < 0 && op_errno == EIO) || split_brain) {
                 eh = shd->split_brain;
+                shd_crawl_event[crawl_data->child]->split_brain_count += 1;
         } else if ((op_ret < 0) || sh_failed) {
                 eh = shd->heal_failed;
+                shd_crawl_event[crawl_data->child]->heal_failed_count += 1;
         } else if (actual_sh_done == 1) {
-                      eh = shd->healed;
+                eh = shd->healed;
+                shd_crawl_event[crawl_data->child]->healed_count += 1;
         }
-
         ret = -1;
 
         if (eh != NULL) {
@@ -408,10 +632,20 @@ _self_heal_entry (xlator_t *this, afr_crawl_data_t *crawl_data, gf_dirent_t *ent
         struct iatt      parentbuf = {0};
         int              ret = 0;
         dict_t           *xattr_rsp = NULL;
+        dict_t           *xattr_req = NULL;
+
+        xattr_req = dict_new ();
+        if (!xattr_req) {
+                errno = ENOMEM;
+                ret = -1;
+                goto out;
+        }
+
+        ret = dict_set_int32 (xattr_req, "allow-sh-for-running-transaction", 1);
 
         gf_log (this->name, GF_LOG_DEBUG, "lookup %s", child->path);
 
-        ret = syncop_lookup (this, child, NULL,
+        ret = syncop_lookup (this, child, xattr_req,
                              iattr, &xattr_rsp, &parentbuf);
         _crawl_post_sh_action (this, parent, child, ret, errno, xattr_rsp,
                                crawl_data);
@@ -420,6 +654,9 @@ _self_heal_entry (xlator_t *this, afr_crawl_data_t *crawl_data, gf_dirent_t *ent
         if (ret == 0)
                 ret = _link_inode_update_loc (this, child, iattr);
 
+out:
+        if (xattr_req)
+                dict_unref(xattr_req);
         return ret;
 }
 
@@ -528,12 +765,20 @@ _do_crawl_op_on_local_subvols (xlator_t *this, afr_crawl_type_t crawl,
                                         status = "Started self-heal";
                                         _do_self_heal_on_subvol (this, i,
                                                                  crawl);
-                                } else if (output) {
+                                } else if (output && (op == INFO)) {
                                         status = "";
                                         afr_start_crawl (this, i, INDEX,
                                                          _add_summary_to_dict,
                                                          output, _gf_false, 0,
                                                          NULL);
+                                } else if (output &&
+                                           (op == STATISTICS_TO_BE_HEALED)) {
+                                            status = "";
+                                            afr_start_crawl (this, i,
+                                                             INDEX_TO_BE_HEALED,
+                                       _count_hard_links_under_base_indices_dir,
+                                                             output, _gf_false,
+                                                             0, NULL);
                                 }
                         }
                         if (output) {
@@ -567,8 +812,103 @@ _get_index_summary_on_local_subvols (xlator_t *this, dict_t *output)
         return _do_crawl_op_on_local_subvols (this, INDEX, INFO, output);
 }
 
+void
+afr_fill_completed_crawl_statistics_to_dict (xlator_t *this, dict_t *dict)
+{
+        afr_private_t           *priv  = NULL;
+        afr_self_heald_t        *shd = NULL;
+        int                     i = 0;
+        priv = this->private;
+        shd= &priv->shd;
+        for (i = 0; i < priv->child_count; i++) {
+                if (shd->pos[i] != AFR_POS_LOCAL)
+                        continue;
+                _add_statistics_to_dict (this, dict, i);
+        }
+
+        return ;
+}
+
+static void
+reset_crawl_event (shd_crawl_event_t *crawl_event)
+{
+    crawl_event->healed_count = 0;
+    crawl_event->split_brain_count = 0;
+    crawl_event->heal_failed_count = 0;
+    GF_FREE (crawl_event->start_time_str);
+    crawl_event->start_time_str = NULL;
+    crawl_event->end_time_str = NULL;
+    crawl_event->crawl_type = NULL;
+    crawl_event->crawl_inprogress = _gf_false;
+    return;
+}
+
+static void
+afr_copy_crawl_event_struct (shd_crawl_event_t *src, shd_crawl_event_t *dst)
+{
+        dst->healed_count = src->healed_count;
+        dst->split_brain_count = src->split_brain_count;
+        dst->heal_failed_count = src->heal_failed_count;
+        dst->start_time_str = gf_strdup (src->start_time_str);
+        dst->end_time_str = "Crawl is already in progress";
+        dst->crawl_type = src->crawl_type;
+        dst->crawl_inprogress = _gf_true;
+        return;
+}
+
+static int
+afr_fill_crawl_statistics_of_running_crawl(xlator_t *this, dict_t *dict)
+{
+        shd_crawl_event_t       *evnt = NULL;
+        int                     ret = 0;
+        afr_private_t           *priv = NULL;
+        afr_self_heald_t        *shd = NULL;
+        int                     i = 0;
+        priv = this->private;
+        shd = &priv->shd;
+
+        evnt = GF_CALLOC (1, sizeof (shd_crawl_event_t),
+                          gf_afr_mt_shd_crawl_event_t);
+        if (!evnt) {
+                ret = -1;
+                goto out;
+        }
+        LOCK (&priv->lock);
+        {
+                for (i = 0; i < priv->child_count; i++) {
+                        if (shd->pos[i] != AFR_POS_LOCAL)
+                                continue;
+
+                        reset_crawl_event (evnt);
+
+                        if (!shd->crawl_events[i]) {
+                                continue;
+                        }
+
+                        afr_copy_crawl_event_struct (shd->crawl_events[i],
+                                                     evnt);
+                        _add_crawl_stats_to_dict (this, dict, i, evnt, NULL);
+
+                }
+        }
+        UNLOCK (&priv->lock);
+        reset_crawl_event (evnt);
+        GF_FREE (evnt);
+
+out:
+        return ret;
+}
+
+static int
+_add_local_subvols_crawl_statistics_to_dict (xlator_t *this, dict_t *dict)
+{
+        int ret = 0;
+        afr_fill_completed_crawl_statistics_to_dict (this, dict);
+        ret = afr_fill_crawl_statistics_of_running_crawl (this, dict);
+        return ret;
+}
 int
-_add_all_subvols_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict)
+_add_local_subvols_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict)
 {
         afr_private_t           *priv = NULL;
         afr_self_heald_t        *shd = NULL;
@@ -618,16 +958,25 @@ afr_xl_op (xlator_t *this, dict_t *input, dict_t *output)
                 ret = 0;
                 break;
         case GF_AFR_OP_HEALED_FILES:
-                ret = _add_all_subvols_eh_to_dict (this, shd->healed, output);
+                ret = _add_local_subvols_eh_to_dict (this, shd->healed, output);
                 break;
         case GF_AFR_OP_HEAL_FAILED_FILES:
-                ret = _add_all_subvols_eh_to_dict (this, shd->heal_failed,
+                ret = _add_local_subvols_eh_to_dict (this, shd->heal_failed,
                                                    output);
                 break;
         case GF_AFR_OP_SPLIT_BRAIN_FILES:
-                ret = _add_all_subvols_eh_to_dict (this, shd->split_brain,
+                ret = _add_local_subvols_eh_to_dict (this, shd->split_brain,
                                                    output);
                 break;
+        case GF_AFR_OP_STATISTICS:
+                ret = _add_local_subvols_crawl_statistics_to_dict (this, output);
+                break;
+        case GF_AFR_OP_STATISTICS_HEAL_COUNT:
+        case GF_AFR_OP_STATISTICS_HEAL_COUNT_PER_REPLICA:
+                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;
@@ -642,7 +991,7 @@ afr_poll_self_heal (void *data)
 {
         afr_private_t    *priv = NULL;
         afr_self_heald_t *shd = NULL;
-        struct timeval   timeout = {0};
+        struct timespec  timeout = {0};
         xlator_t         *this = NULL;
         long             child = (long)data;
         gf_timer_t       *old_timer = NULL;
@@ -666,7 +1015,7 @@ afr_poll_self_heal (void *data)
         if (shd->enabled && (shd->pos[child] == AFR_POS_LOCAL))
                 _do_self_heal_on_subvol (this, child, INDEX);
         timeout.tv_sec = shd->timeout;
-        timeout.tv_usec = 0;
+        timeout.tv_nsec = 0;
         //notify and previous timer should be synchronized.
         LOCK (&priv->lock);
         {
@@ -800,6 +1149,7 @@ afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data,
         afr_private_t *priv = NULL;
         dict_t        *xattr = NULL;
         void          *index_gfid = NULL;
+        void          *base_indices_holder_vgfid = NULL;
         loc_t         rootloc = {0};
         struct iatt   iattr = {0};
         struct iatt   parent = {0};
@@ -809,7 +1159,7 @@ afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data,
         priv = this->private;
         if (crawl_data->crawl == FULL) {
                 afr_build_root_loc (this, dirloc);
-        } else {
+        } else if (crawl_data->crawl == INDEX) {
                 afr_build_root_loc (this, &rootloc);
                 ret = syncop_getxattr (readdir_xl, &rootloc, &xattr,
                                        GF_XATTROP_INDEX_GFID);
@@ -843,6 +1193,47 @@ afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data,
                 ret = _link_inode_update_loc (this, dirloc, &iattr);
                 if (ret)
                         goto out;
+        } else if (crawl_data->crawl == INDEX_TO_BE_HEALED) {
+                afr_build_root_loc (this, &rootloc);
+                ret = syncop_getxattr (readdir_xl, &rootloc, &xattr,
+                                       GF_BASE_INDICES_HOLDER_GFID);
+                if (ret < 0)
+                        goto out;
+                ret = dict_get_ptr (xattr, GF_BASE_INDICES_HOLDER_GFID,
+                                    &base_indices_holder_vgfid);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR, "index gfid empty "
+                                "on %s", readdir_xl->name);
+                        ret = -1;
+                        goto out;
+                }
+                if (!base_indices_holder_vgfid) {
+                        gf_log (this->name, GF_LOG_ERROR, "Base indices holder"
+                                "virtual gfid is null on %s", readdir_xl->name);
+                        ret = -1;
+                        goto out;
+                }
+                uuid_copy (dirloc->gfid,  base_indices_holder_vgfid);
+                dirloc->path = "";
+                dirloc->inode = inode_new (priv->root_inode->table);
+                ret = syncop_lookup (readdir_xl, dirloc, NULL, &iattr, NULL,
+                                     &parent);
+                if (ret < 0) {
+                        if (errno != ENOENT) {
+                                gf_log (this->name, GF_LOG_ERROR, "lookup "
+                                        "failed for base_indices_holder dir"
+                                        " on %s - (%s)", readdir_xl->name,
+                                        strerror (errno));
+
+                        } else {
+                                gf_log (this->name, GF_LOG_ERROR, "base_indices"
+                                        "_holder is not yet created.");
+                        }
+                        goto out;
+                }
+                ret = _link_inode_update_loc (this, dirloc, &iattr);
+                if (ret)
+                        goto out;
         }
         ret = 0;
 out:
@@ -907,6 +1298,16 @@ afr_crawl_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent,
         priv = this->private;
         if (crawl_data->crawl == FULL) {
                 ret = afr_build_child_loc (this, child, parent, entry->d_name);
+        } else if (crawl_data->crawl == INDEX_TO_BE_HEALED) {
+                ret = _build_index_loc (this, child, entry->d_name, parent);
+                if (ret)
+                        goto out;
+                child->inode = inode_new (priv->root_inode->table);
+                if (!child->inode) {
+                        ret = -1;
+                        goto out;
+                }
+                child->path = NULL;
         } else {
                 child->inode = inode_new (priv->root_inode->table);
                 if (!child->inode)
@@ -956,10 +1357,14 @@ _process_entries (xlator_t *this, loc_t *parentloc, gf_dirent_t *entries,
                 ret = crawl_data->process_entry (this, crawl_data, entry,
                                                  &entry_loc, parentloc, &iattr);
 
-                if (ret)
+                if (crawl_data->crawl == INDEX_TO_BE_HEALED && ret) {
+                       goto out;
+                } else if (ret) {
                         continue;
+                }
 
-                if (crawl_data->crawl == INDEX)
+                if ((crawl_data->crawl == INDEX) ||
+                    (crawl_data->crawl == INDEX_TO_BE_HEALED))
                         continue;
 
                 if (!IA_ISDIR (iattr.ia_type))
@@ -974,6 +1379,10 @@ _process_entries (xlator_t *this, loc_t *parentloc, gf_dirent_t *entries,
         }
         ret = 0;
 out:
+        if ((crawl_data->crawl == INDEX_TO_BE_HEALED)  && ret) {
+                gf_log (this->name, GF_LOG_ERROR,"Failed to get the hardlink "
+                        "count");
+        }
         loc_wipe (&entry_loc);
         return ret;
 }
@@ -1021,6 +1430,9 @@ _crawl_directory (fd_t *fd, loc_t *loc, afr_crawl_data_t *crawl_data)
 
                 ret = _process_entries (this, loc, &entries, &offset,
                                         crawl_data);
+                if ((ret < 0) && (crawl_data->crawl == INDEX_TO_BE_HEALED)) {
+                        goto out;
+                }
                 gf_dirent_free (&entries);
                 free_entries = _gf_false;
         }
@@ -1126,8 +1538,13 @@ afr_dir_crawl (void *data)
                 goto out;
 
         ret = afr_crawl_opendir (this, crawl_data, &fd, &dirloc);
-        if (ret)
+        if (ret) {
+                if (crawl_data->crawl == INDEX_TO_BE_HEALED) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to open base_"
+                                "indices_holder");
+                }
                 goto out;
+        }
 
         ret = _crawl_directory (fd, &dirloc, crawl_data);
         if (ret)
@@ -1141,12 +1558,102 @@ afr_dir_crawl (void *data)
 out:
         if (fd)
                 fd_unref (fd);
-        if (crawl_data->crawl == INDEX)
+        if ((crawl_data->crawl == INDEX) ||
+            (crawl_data->crawl == INDEX_TO_BE_HEALED ))
                 dirloc.path = NULL;
         loc_wipe (&dirloc);
         return ret;
 }
 
+char *
+get_crawl_type_in_string (afr_crawl_type_t crawl)
+{
+        char    *index = "INDEX";
+        char    *full  = "FULL";
+        char    *crawl_type = NULL;
+
+        if (crawl == INDEX){
+                crawl_type = index;
+        } else if (crawl == FULL) {
+                crawl_type = full;
+        }
+
+        return  crawl_type;
+}
+
+static int
+afr_allocate_crawl_event (xlator_t *this, int child, afr_crawl_type_t crawl)
+{
+        afr_private_t           *priv = NULL;
+        afr_self_heald_t        *shd = NULL;
+        int                     ret = 0;
+        shd_crawl_event_t       *crawl_event = NULL;
+        time_t                  get_time = 0;
+
+        priv = this->private;
+        shd = &priv->shd;
+
+        crawl_event = GF_CALLOC (sizeof (shd_crawl_event_t), 1,
+                                 gf_afr_mt_shd_crawl_event_t);
+        if (!crawl_event) {
+                ret = -1;
+                goto out;
+        }
+
+        get_time =  time(NULL);
+        if (get_time == ((time_t)-1)) {
+                 ret = -1;
+                goto out;
+        }
+
+        crawl_event->start_time_str = gf_strdup (ctime(&get_time));
+
+        crawl_event->crawl_type = get_crawl_type_in_string (crawl);
+        if (!crawl_event->crawl_type) {
+                ret = -1;
+                goto out;
+        }
+        LOCK (&priv->lock);
+        {
+                shd->crawl_events[child] = crawl_event;
+        }
+        UNLOCK (&priv->lock);
+        ret = 0;
+out:
+        return ret;
+
+}
+
+static int
+afr_put_crawl_event_in_eh (xlator_t *this, int child)
+{
+        afr_private_t           *priv = NULL;
+        afr_self_heald_t        *shd = NULL;
+        int                     ret = 0;
+        time_t                  get_time = 0;
+        shd_crawl_event_t       **crawl_event = NULL;
+
+        priv = this->private;
+        shd = &priv->shd;
+
+        get_time = time(NULL);
+        if (get_time == ((time_t)-1)) {
+                ret = -1;
+                goto out;
+        }
+        crawl_event = (shd_crawl_event_t**)shd->crawl_events;
+        LOCK (&priv->lock);
+        {
+                crawl_event[child]->end_time_str = gf_strdup (ctime(&get_time));
+                ret = eh_save_history (shd->statistics[child],
+                                       crawl_event[child]);
+                crawl_event[child] = NULL;
+        }
+        UNLOCK (&priv->lock);
+out:
+        return ret;
+}
+
 static int
 afr_dir_exclusive_crawl (void *data)
 {
@@ -1182,7 +1689,15 @@ afr_dir_exclusive_crawl (void *data)
         }
 
         do {
+                ret = afr_allocate_crawl_event (this, child, crawl_data->crawl);
+                if (ret)
+                        goto out;
                 afr_dir_crawl (data);
+
+                ret = afr_put_crawl_event_in_eh (this, child);
+                if (ret < 0)
+                        goto out;
+
                 LOCK (&priv->lock);
                 {
                         if (shd->pending[child] != NONE) {
@@ -1270,4 +1785,3 @@ afr_set_root_gfid (dict_t *dict)
 
         return ret;
 }
-
diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h
index 32a8aaca5..e0c083754 100644
--- a/xlators/cluster/afr/src/afr-self-heald.h
+++ b/xlators/cluster/afr/src/afr-self-heald.h
@@ -29,6 +29,19 @@ typedef struct afr_crawl_data_ {
                               struct iatt *iattr);
 } afr_crawl_data_t;
 
+typedef struct crawl_event_stats_ {
+        uint64_t healed_count;
+        uint64_t split_brain_count;
+        uint64_t heal_failed_count;
+        char     *start_time_str;
+        char     *end_time_str;
+        char     *crawl_type;
+        gf_boolean_t crawl_inprogress;
+} shd_crawl_event_t;
+
+void _destroy_crawl_event_data (void *data);
+void _destroy_shd_event_data (void *data);
+
 typedef int (*process_entry_cbk_t) (xlator_t *this, afr_crawl_data_t *crawl_data,
                               gf_dirent_t *entry, loc_t *child, loc_t *parent,
                               struct iatt *iattr);
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
index 217ff8548..20306e469 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -79,7 +79,7 @@ afr_save_lk_owner (call_frame_t *frame)
 
         local = frame->local;
 
-	local->saved_lk_owner = frame->root->lk_owner;
+        local->saved_lk_owner = frame->root->lk_owner;
 }
 
 
@@ -90,10 +90,9 @@ afr_restore_lk_owner (call_frame_t *frame)
 
         local = frame->local;
 
-	frame->root->lk_owner = local->saved_lk_owner;
+        frame->root->lk_owner = local->saved_lk_owner;
 }
 
-
 static void
 __mark_all_pending (int32_t *pending[], int child_count,
                     afr_transaction_type type)
@@ -369,7 +368,7 @@ afr_changelog_post_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         if (call_count == 0) {
                 if (local->transaction.resume_stub) {
-			AFR_CALL_RESUME (local->transaction.resume_stub);
+			call_resume (local->transaction.resume_stub);
                         local->transaction.resume_stub = NULL;
                 }
 
@@ -443,14 +442,30 @@ out:
         return;
 }
 
+afr_inodelk_t*
+afr_get_inodelk (afr_internal_lock_t *int_lock, char *dom)
+{
+        afr_inodelk_t *inodelk = NULL;
+        int           i = 0;
+
+        for (i = 0; int_lock->inodelk[i].domain; i++) {
+                inodelk = &int_lock->inodelk[i];
+                if (strcmp (dom, inodelk->domain) == 0)
+                        return inodelk;
+        }
+        return NULL;
+}
+
 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:
-                locked_nodes = int_lock->inode_locked_nodes;
+                inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+                locked_nodes = inodelk->locked_nodes;
         break;
 
         case AFR_ENTRY_TRANSACTION:
@@ -553,24 +568,77 @@ afr_set_postop_dict (afr_local_t *local, xlator_t *this, dict_t *xattr,
 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 index = -1;
-	int i = 0;
+        afr_private_t *priv = NULL;
+        afr_local_t *local = NULL;
+        int index = -1;
+        int i = 0;
 
-	local = frame->local;
-	priv = this->private;
+        local = frame->local;
+        priv = this->private;
 
         index = afr_index_for_transaction_type (local->transaction.type);
 
-	for (i = 0; i < priv->child_count; i++) {
+        for (i = 0; i < priv->child_count; i++) {
                 if (local->pending[i][index] == 0)
-			return _gf_false;
+                        return _gf_false;
         }
 
-	return _gf_true;
+        return _gf_true;
 }
 
+static void
+afr_dir_fop_handle_all_fop_failures (call_frame_t *frame)
+{
+        xlator_t        *this = NULL;
+        afr_local_t     *local = NULL;
+        afr_private_t   *priv = NULL;
+
+        this = frame->this;
+        local = frame->local;
+        priv = this->private;
+
+        if ((local->transaction.type != AFR_ENTRY_TRANSACTION) &&
+            (local->transaction.type != AFR_ENTRY_RENAME_TRANSACTION))
+                return;
+
+        if (local->op_ret >= 0)
+                goto out;
+
+        __mark_all_success (local->pending, priv->child_count,
+                            local->transaction.type);
+out:
+        return;
+}
+
+static void
+afr_data_handle_quota_errors (call_frame_t *frame, xlator_t *this)
+{
+        int     i = 0;
+        afr_private_t *priv = NULL;
+        afr_local_t   *local = NULL;
+        gf_boolean_t  all_quota_failures = _gf_false;
+
+        local = frame->local;
+        priv  = this->private;
+        if (local->transaction.type != AFR_DATA_TRANSACTION)
+                return;
+        /*
+         * Idea is to not leave the file in FOOL-FOOL scenario in case on
+         * all the bricks data transaction failed with EDQUOT to avoid
+         * increasing un-necessary load of self-heals in the system.
+         */
+        all_quota_failures = _gf_true;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i] &&
+                    (local->child_errno[i] != EDQUOT)) {
+                        all_quota_failures = _gf_false;
+                        break;
+                }
+        }
+        if (all_quota_failures)
+                __mark_all_success (local->pending, priv->child_count,
+                                    local->transaction.type);
+}
 
 int
 afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this)
@@ -593,6 +661,9 @@ afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this)
                                          local->transaction.pre_op,
                                          local->transaction.type);
 
+        afr_data_handle_quota_errors (frame, this);
+        afr_dir_fop_handle_all_fop_failures (frame);
+
         if (local->fd)
                 afr_transaction_rm_stale_children (frame, this,
                                                    local->fd->inode,
@@ -619,7 +690,7 @@ afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this)
                 goto out;
         }
 
-	nothing_failed = afr_txn_nothing_failed (frame, this);
+        nothing_failed = afr_txn_nothing_failed (frame, this);
 
         afr_compute_txn_changelog (local , priv);
 
@@ -645,14 +716,14 @@ afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this)
                                 break;
                         }
 
-			/* local->transaction.postop_piggybacked[] was
-			   precomputed in is_piggyback_postop() when called from
-			   afr_changelog_post_op_safe()
-			*/
+                        /* local->transaction.postop_piggybacked[] was
+                           precomputed in is_piggyback_postop() when called from
+                           afr_changelog_post_op_safe()
+                        */
 
-			piggyback = 0;
-			if (local->transaction.postop_piggybacked[i])
-				piggyback = 1;
+                        piggyback = 0;
+                        if (local->transaction.postop_piggybacked[i])
+                                piggyback = 1;
 
                         afr_set_postop_dict (local, this, xattr[i],
                                              piggyback, i);
@@ -906,7 +977,7 @@ afr_changelog_pre_op (call_frame_t *frame, xlator_t *this)
                         }
                         UNLOCK (&local->fd->lock);
 
-			afr_set_delayed_post_op (frame, this);
+                        afr_set_delayed_post_op (frame, this);
 
                         if (piggyback)
                                 afr_changelog_pre_op_cbk (frame, (void *)(long)i,
@@ -1179,12 +1250,14 @@ 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;
 }
@@ -1199,6 +1272,7 @@ 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:
@@ -1259,69 +1333,115 @@ afr_internal_lock_finish (call_frame_t *frame, xlator_t *this)
 void
 afr_set_delayed_post_op (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t    *local = NULL;
-	afr_private_t  *priv = NULL;
+        afr_local_t    *local = NULL;
+        afr_private_t  *priv = NULL;
 
-	/* call this function from any of the related optimizations
-	   which benefit from delaying post op are enabled, namely:
+        /* call this function from any of the related optimizations
+           which benefit from delaying post op are enabled, namely:
 
-	   - changelog piggybacking
-	   - eager locking
-	*/
+           - changelog piggybacking
+           - eager locking
+        */
 
-	priv = this->private;
-	if (!priv)
-		return;
+        priv = this->private;
+        if (!priv)
+                return;
 
-	if (!priv->post_op_delay_secs)
-		return;
+        if (!priv->post_op_delay_secs)
+                return;
 
         local = frame->local;
         if (!local->transaction.eager_lock_on)
                 return;
 
-	if (!local)
-		return;
+        if (!local)
+                return;
 
-	if (!local->fd)
-		return;
+        if (!local->fd)
+                return;
 
-	if (local->op == GF_FOP_WRITE)
-		local->delayed_post_op = _gf_true;
+        if (local->op == GF_FOP_WRITE)
+                local->delayed_post_op = _gf_true;
 }
 
+gf_boolean_t
+afr_are_multiple_fds_opened (inode_t *inode, xlator_t *this)
+{
+        afr_inode_ctx_t *ictx = NULL;
+
+        if (!inode) {
+                /* If false is returned, it may keep on taking eager-lock
+                 * which may lead to starvation, so return true to avoid that.
+                 */
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid inode");
+                return _gf_true;
+        }
+        /* Lets say mount1 has eager-lock(full-lock) and after the eager-lock
+         * is taken mount2 opened the same file, it won't be able to
+         * perform any data operations until mount1 releases eager-lock.
+         * To avoid such scenario do not enable eager-lock for this transaction
+         * if open-fd-count is > 1
+         */
+
+        ictx = afr_inode_ctx_get (inode, this);
+        if (!ictx)
+                return _gf_true;
+
+        if (ictx->open_fd_count > 1)
+                return _gf_true;
+
+        return _gf_false;
+}
+
+gf_boolean_t
+afr_any_fops_failed (afr_local_t *local, afr_private_t *priv)
+{
+        if (local->success_count != priv->child_count)
+                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;
+        afr_local_t      *local = NULL;
+        gf_boolean_t      res = _gf_false;
+        afr_private_t    *priv  = NULL;
 
-	local = frame->local;
-	if (!local)
-		goto out;
+        priv  = this->private;
 
-	if (!local->delayed_post_op)
-		goto out;
+        local = frame->local;
+        if (!local)
+                goto out;
 
-	res = _gf_true;
+        if (!local->delayed_post_op)
+                goto out;
+
+        //Mark pending changelog ASAP
+        if (afr_any_fops_failed (local, priv))
+                goto out;
+
+        if (local->fd && afr_are_multiple_fds_opened (local->fd->inode, this))
+                goto out;
+
+        res = _gf_true;
 out:
-	return res;
+        return res;
 }
 
 
 void
 afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
-			       call_stub_t *stub);
+                               call_stub_t *stub);
 
 void
 afr_delayed_changelog_wake_up_cbk (void *data)
 {
-	fd_t           *fd = NULL;
+        fd_t           *fd = NULL;
 
-	fd = data;
+        fd = data;
 
-	afr_delayed_changelog_wake_up (THIS, fd);
+        afr_delayed_changelog_wake_up (THIS, fd);
 }
 
 
@@ -1332,39 +1452,39 @@ afr_delayed_changelog_wake_up_cbk (void *data)
 static gf_boolean_t
 is_piggyback_post_op (call_frame_t *frame, fd_t *fd)
 {
-	afr_fd_ctx_t *fdctx = NULL;
-	afr_local_t *local = NULL;
-	gf_boolean_t piggyback = _gf_true;
-	afr_private_t *priv = NULL;
-	int i = 0;
+        afr_fd_ctx_t *fdctx = NULL;
+        afr_local_t *local = NULL;
+        gf_boolean_t piggyback = _gf_true;
+        afr_private_t *priv = NULL;
+        int i = 0;
 
-	priv = frame->this->private;
-	local = frame->local;
-	fdctx = afr_fd_ctx_get (fd, frame->this);
+        priv = frame->this->private;
+        local = frame->local;
+        fdctx = afr_fd_ctx_get (fd, frame->this);
 
-	LOCK(&fd->lock);
-	{
-		piggyback = _gf_true;
-
-		for (i = 0; i < priv->child_count; i++) {
-			if (!local->transaction.pre_op[i])
-				continue;
-			if (fdctx->pre_op_piggyback[i]) {
-				fdctx->pre_op_piggyback[i]--;
-				local->transaction.postop_piggybacked[i] = 1;
-			} else {
-				/* For at least _one_ subvolume we cannot
-				   piggyback on the changelog, and have to
-				   perform a hard POST-OP and therefore fsync
-				   if necesssary
-				*/
-				piggyback = _gf_false;
+        LOCK(&fd->lock);
+        {
+                piggyback = _gf_true;
+
+                for (i = 0; i < priv->child_count; i++) {
+                        if (!local->transaction.pre_op[i])
+                                continue;
+                        if (fdctx->pre_op_piggyback[i]) {
+                                fdctx->pre_op_piggyback[i]--;
+                                local->transaction.postop_piggybacked[i] = 1;
+                        } else {
+                                /* For at least _one_ subvolume we cannot
+                                   piggyback on the changelog, and have to
+                                   perform a hard POST-OP and therefore fsync
+                                   if necesssary
+                                */
+                                piggyback = _gf_false;
                                 GF_ASSERT (fdctx->pre_op_done[i]);
                                 fdctx->pre_op_done[i]--;
-			}
-		}
-	}
-	UNLOCK(&fd->lock);
+                        }
+                }
+        }
+        UNLOCK(&fd->lock);
 
         if (!afr_txn_nothing_failed (frame, frame->this)) {
                 /* something failed in this transaction,
@@ -1381,186 +1501,205 @@ is_piggyback_post_op (call_frame_t *frame, fd_t *fd)
 int
 afr_fd_report_unstable_write (xlator_t *this, fd_t *fd)
 {
-	afr_fd_ctx_t *fdctx = NULL;
+        afr_fd_ctx_t *fdctx = NULL;
 
-	fdctx = afr_fd_ctx_get (fd, this);
+        fdctx = afr_fd_ctx_get (fd, this);
 
-	LOCK(&fd->lock);
-	{
-		fdctx->witnessed_unstable_write = _gf_true;
-	}
-	UNLOCK(&fd->lock);
+        LOCK(&fd->lock);
+        {
+                fdctx->witnessed_unstable_write = _gf_true;
+        }
+        UNLOCK(&fd->lock);
 
-	return 0;
+        return 0;
 }
 
 /* TEST and CLEAR operation */
 gf_boolean_t
 afr_fd_has_witnessed_unstable_write (xlator_t *this, fd_t *fd)
 {
-	afr_fd_ctx_t *fdctx = NULL;
-	gf_boolean_t witness = _gf_false;
+        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);
+        LOCK(&fd->lock);
+        {
+                if (fdctx->witnessed_unstable_write) {
+                        witness = _gf_true;
+                        fdctx->witnessed_unstable_write = _gf_false;
+                }
+        }
+        UNLOCK (&fd->lock);
 
-	return witness;
+        return witness;
 }
 
 
 int
 afr_changelog_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			 int op_ret, int op_errno, struct iatt *pre,
-			 struct iatt *post, dict_t *xdata)
+                         int op_ret, int op_errno, struct iatt *pre,
+                         struct iatt *post, dict_t *xdata)
 {
-	afr_private_t *priv = NULL;
+        afr_private_t *priv = NULL;
         int child_index = (long) cookie;
-	int call_count = -1;
-	afr_local_t *local = NULL;
+        int call_count = -1;
+        afr_local_t *local = NULL;
 
-	priv = this->private;
-	local = frame->local;
+        priv = this->private;
+        local = frame->local;
 
-	if (afr_fop_failed (op_ret, op_errno)) {
-		/* Failure of fsync() is as good as failure of previous
-		   write(). So treat it like one.
-		*/
-		gf_log (this->name, GF_LOG_WARNING,
-			"fsync(%s) failed on subvolume %s. Transaction was %s",
-			uuid_utoa (local->fd->inode->gfid),
-			priv->children[child_index]->name,
-			gf_fop_list[local->op]);
-
-		afr_transaction_fop_failed (frame, this, child_index);
-	}
+        if (afr_fop_failed (op_ret, op_errno)) {
+                /* Failure of fsync() is as good as failure of previous
+                   write(). So treat it like one.
+                */
+                gf_log (this->name, GF_LOG_WARNING,
+                        "fsync(%s) failed on subvolume %s. Transaction was %s",
+                        uuid_utoa (local->fd->inode->gfid),
+                        priv->children[child_index]->name,
+                        gf_fop_list[local->op]);
+
+                afr_transaction_fop_failed (frame, this, child_index);
+        }
 
         call_count = afr_frame_return (frame);
 
         if (call_count == 0)
-		afr_changelog_post_op_now (frame, this);
+                afr_changelog_post_op_now (frame, this);
 
-	return 0;
+        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;
+        afr_local_t *local = NULL;
+        int i = 0;
+        int call_count = 0;
+        afr_private_t *priv = NULL;
+ 	dict_t *xdata = NULL;
+ 	GF_UNUSED int ret = -1;
 
-	local = frame->local;
-	priv = this->private;
+        local = frame->local;
+        priv = this->private;
 
         call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
                                                      priv->child_count);
 
-	if (!call_count) {
-		/* will go straight to unlock */
-		afr_changelog_post_op_now (frame, this);
-		return 0;
-	}
+        if (!call_count) {
+                /* will go straight to unlock */
+                afr_changelog_post_op_now (frame, this);
+                return 0;
+        }
 
-	local->call_count = call_count;
+        local->call_count = call_count;
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (!local->transaction.pre_op[i])
-			continue;
+	xdata = dict_new();
+	if (xdata)
+		ret = dict_set_int32 (xdata, "batch-fsync", 1);
 
-		STACK_WIND_COOKIE (frame, afr_changelog_fsync_cbk,
-				   (void *) (long) i, priv->children[i],
-				   priv->children[i]->fops->fsync, local->fd,
-				   1, NULL);
-		if (!--call_count)
-			break;
-	}
+        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;
+        }
 
-	return 0;
+	if (xdata)
+		dict_unref (xdata);
+
+        return 0;
 }
 
 
-int
+        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 (is_piggyback_post_op (frame, local->fd)) {
-		/* just detected that this post-op is about to
-		   be optimized away as a new write() has
-		   already piggybacked on this frame's changelog.
-		*/
-		afr_changelog_post_op_now (frame, this);
-		return 0;
-	}
+        if (!local->fd || local->transaction.type != AFR_DATA_TRANSACTION) {
+                afr_changelog_post_op_now (frame, this);
+                return 0;
+        }
 
-	/* Calling afr_changelog_post_op_now() now will result in
-	   issuing ->[f]xattrop().
-
-	   Performing a hard POST-OP (->[f]xattrop() FOP) is a more
-	   responsible operation that what it might appear on the surface.
-
-	   The changelog of a file (in the xattr of the file on the server)
-	   stores information (pending count) about the state of the file
-	   on the OTHER server. This changelog is blindly trusted, and must
-	   therefore be updated in such a way it remains trustworthy. This
-	   implies that decrementing the pending count (essentially "clearing
-	   the dirty flag") must be done STRICTLY after we are sure that the
-	   operation on the other server has reached stable storage.
-
-	   While the backend filesystem on that server will eventually flush
-	   it to stable storage, we (being in userspace) have no mechanism
-	   to get notified when the write became "stable".
-
-	   This means we need take matter into our own hands and issue an
-	   fsync() EVEN IF THE APPLICATION WAS PERFORMING UNSTABLE WRITES,
-	   and get an acknowledgement for it. And we need to wait for the
-	   fsync() acknowledgement before initiating the hard POST-OP.
-
-	   However if the FD itself was opened in O_SYNC or O_DSYNC then
-	   we are already guaranteed that the writes were made stable as
-	   part of the FOP itself. The same holds true for NFS stable
-	   writes which happen on an anonymous FD with O_DSYNC or O_SYNC
-	   flag set in the writev() @flags param. For all other write types,
-	   mark a flag in the fdctx whenever an unstable write is witnessed.
-	*/
-
-	if (!afr_fd_has_witnessed_unstable_write (this, local->fd)) {
-		afr_changelog_post_op_now (frame, this);
-		return 0;
-	}
+        if (is_piggyback_post_op (frame, local->fd)) {
+                /* just detected that this post-op is about to
+                   be optimized away as a new write() has
+                   already piggybacked on this frame's changelog.
+                   */
+                afr_changelog_post_op_now (frame, this);
+                return 0;
+        }
 
-	/* Time to fsync() */
+        /* 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;
+        }
 
-	afr_changelog_fsync (frame, this);
+        /* 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;
+        return 0;
 }
 
 
 void
 afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
-			       call_stub_t *stub)
+                               call_stub_t *stub)
 {
 	afr_fd_ctx_t      *fd_ctx = NULL;
 	call_frame_t      *prev_frame = NULL;
-	struct timeval     delta = {0, };
+	struct timespec    delta = {0, };
 	afr_private_t     *priv = NULL;
 	afr_local_t       *local = NULL;
 
@@ -1568,10 +1707,10 @@ afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
 
 	fd_ctx = afr_fd_ctx_get (fd, this);
 	if (!fd_ctx)
-		return;
+                goto out;
 
 	delta.tv_sec = priv->post_op_delay_secs;
-	delta.tv_usec = 0;
+	delta.tv_nsec = 0;
 
 	pthread_mutex_lock (&fd_ctx->delay_lock);
 	{
@@ -1590,6 +1729,7 @@ afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
 unlock:
 	pthread_mutex_unlock (&fd_ctx->delay_lock);
 
+out:
 	if (prev_frame) {
 		local = prev_frame->local;
 		local->transaction.resume_stub = stub;
@@ -1603,14 +1743,14 @@ unlock:
 void
 afr_changelog_post_op (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t  *local = NULL;
+        afr_local_t  *local = NULL;
 
-	local = frame->local;
+        local = frame->local;
 
-	if (is_afr_delayed_changelog_post_op_needed (frame, this))
-		afr_delayed_changelog_post_op (this, frame, local->fd, NULL);
-	else
-		afr_changelog_post_op_safe (frame, this);
+        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);
 }
 
 
@@ -1621,22 +1761,22 @@ afr_changelog_post_op (call_frame_t *frame, xlator_t *this)
 
    The @stub gets saved in @local and gets resumed in
    afr_local_cleanup()
-*/
+   */
 void
 afr_delayed_changelog_wake_resume (xlator_t *this, fd_t *fd, call_stub_t *stub)
 {
-	afr_delayed_changelog_post_op (this, NULL, fd, stub);
+        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);
+        afr_delayed_changelog_post_op (this, NULL, fd, NULL);
 }
 
 
-int
+        int
 afr_transaction_resume (call_frame_t *frame, xlator_t *this)
 {
         afr_internal_lock_t *int_lock = NULL;
@@ -1647,18 +1787,18 @@ afr_transaction_resume (call_frame_t *frame, xlator_t *this)
         int_lock = &local->internal_lock;
         priv     = this->private;
 
-	if (local->transaction.eager_lock_on) {
-		/* We don't need to retain "local" in the
-		   fd list anymore, writes to all subvols
-		   are finished by now */
-		LOCK (&local->fd->lock);
-		{
-			list_del_init (&local->transaction.eager_locked);
-		}
-		UNLOCK (&local->fd->lock);
-	}
+        if (local->transaction.eager_lock_on) {
+                /* We don't need to retain "local" in the
+                   fd list anymore, writes to all subvols
+                   are finished by now */
+                LOCK (&local->fd->lock);
+                {
+                        list_del_init (&local->transaction.eager_locked);
+                }
+                UNLOCK (&local->fd->lock);
+        }
 
-	afr_restore_lk_owner (frame);
+        afr_restore_lk_owner (frame);
 
         if (__fop_changelog_needed (frame, this)) {
                 afr_changelog_post_op (frame, this);
@@ -1680,7 +1820,8 @@ 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;
@@ -1689,55 +1830,56 @@ afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this, int child_index
         priv  = this->private;
 
         __mark_child_dead (local->pending, priv->child_count,
-                           child_index, local->transaction.type);
+                        child_index, local->transaction.type);
 }
 
 
 
-static gf_boolean_t
+        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));
-}
+        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;
+        afr_private_t *priv = NULL;
+        afr_fd_ctx_t  *fdctx = NULL;
+        afr_local_t   *each = NULL;
 
-	priv = this->private;
+        priv = this->private;
 
-	if (!local->fd)
-		return;
+        if (!local->fd)
+                return;
 
-	if (local->transaction.type != AFR_DATA_TRANSACTION)
-		return;
+        if (local->transaction.type != AFR_DATA_TRANSACTION)
+                return;
 
-	if (!priv->eager_lock)
-		return;
+        if (!priv->eager_lock)
+                return;
 
-	fdctx = afr_fd_ctx_get (local->fd, this);
-	if (!fdctx)
-		return;
+        fdctx = afr_fd_ctx_get (local->fd, this);
+        if (!fdctx)
+                return;
 
+        if (afr_are_multiple_fds_opened (local->fd->inode, this))
+                return;
         /*
          * Once full file lock is acquired in eager-lock phase, overlapping
          * writes do not compete for inode-locks, instead are transferred to the
@@ -1755,22 +1897,22 @@ afr_transaction_eager_lock_init (afr_local_t *local, xlator_t *this)
          * 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);
-	}
+        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);
+        UNLOCK (&local->fd->lock);
 }
 
 
@@ -1789,11 +1931,10 @@ afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type)
         local->transaction.type   = type;
 
         ret = afr_transaction_local_init (local, this);
-        if (ret < 0) {
+        if (ret < 0)
             goto out;
-        }
 
-	afr_transaction_eager_lock_init (local, this);
+        afr_transaction_eager_lock_init (local, this);
 
         if (local->fd && local->transaction.eager_lock_on)
                 afr_set_lk_owner (frame, this, local->fd);
@@ -1802,6 +1943,9 @@ afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type)
 
         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);
diff --git a/xlators/cluster/afr/src/afr-transaction.h b/xlators/cluster/afr/src/afr-transaction.h
index 55e8bbcca..fa626fd0d 100644
--- a/xlators/cluster/afr/src/afr-transaction.h
+++ b/xlators/cluster/afr/src/afr-transaction.h
@@ -23,6 +23,9 @@ 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);
 
@@ -40,4 +43,9 @@ afr_delayed_changelog_wake_up (xlator_t *this, fd_t *fd);
 void
 __mark_all_success (int32_t *pending[], int child_count,
                     afr_transaction_type type);
+gf_boolean_t
+afr_any_fops_failed (afr_local_t *local, afr_private_t *priv);
+
+gf_boolean_t
+afr_txn_nothing_failed (call_frame_t *frame, xlator_t *this);
 #endif /* __TRANSACTION_H__ */
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index bee10fd01..c724eb2ae 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -191,6 +191,8 @@ reconfigure (xlator_t *this, dict_t *options)
         /* Reset this so we re-discover in case the topology changed.  */
         GF_OPTION_RECONF ("readdir-failover", priv->readdir_failover, options,
                           bool, out);
+        GF_OPTION_RECONF ("ensure-durability", priv->ensure_durability, options,
+                          bool, out);
         priv->did_discovery = _gf_false;
 
         ret = 0;
@@ -335,6 +337,8 @@ init (xlator_t *this)
 
 	GF_OPTION_INIT ("post-op-delay-secs", priv->post_op_delay_secs, uint32, out);
         GF_OPTION_INIT ("readdir-failover", priv->readdir_failover, bool, out);
+        GF_OPTION_INIT ("ensure-durability", priv->ensure_durability, bool,
+                        out);
 
         priv->wait_count = 1;
 
@@ -376,8 +380,6 @@ 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;
                 }
@@ -386,6 +388,13 @@ 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) {
@@ -430,15 +439,18 @@ init (xlator_t *this)
         if (!priv->shd.timer)
                 goto out;
 
-        priv->shd.healed = eh_new (AFR_EH_HEALED_LIMIT, _gf_false);
+        priv->shd.healed = eh_new (AFR_EH_HEALED_LIMIT, _gf_false,
+                                   _destroy_shd_event_data);
         if (!priv->shd.healed)
                 goto out;
 
-        priv->shd.heal_failed = eh_new (AFR_EH_HEAL_FAIL_LIMIT, _gf_false);
+        priv->shd.heal_failed = eh_new (AFR_EH_HEAL_FAIL_LIMIT, _gf_false,
+                                        _destroy_shd_event_data);
         if (!priv->shd.heal_failed)
                 goto out;
 
-        priv->shd.split_brain = eh_new (AFR_EH_SPLIT_BRAIN_LIMIT, _gf_false);
+        priv->shd.split_brain = eh_new (AFR_EH_SPLIT_BRAIN_LIMIT, _gf_false,
+                                        _destroy_shd_event_data);
         if (!priv->shd.split_brain)
                 goto out;
 
@@ -448,7 +460,9 @@ init (xlator_t *this)
         priv->root_inode = inode_ref (this->itable->root);
         GF_OPTION_INIT ("node-uuid", priv->shd.node_uuid, str, out);
         GF_OPTION_INIT ("heal-timeout", priv->shd.timeout, int32, out);
-
+        ret = afr_initialise_statistics (this);
+        if (ret)
+                goto out;
         ret = 0;
 out:
         return ret;
@@ -483,6 +497,9 @@ 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,
@@ -765,5 +782,12 @@ struct volume_options options[] = {
           .description = "readdir(p) will not failover if this option is off",
           .default_value = "on",
         },
+        { .key = {"ensure-durability"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .description = "Afr performs fsyncs for transactions if this "
+                         "option is on to make sure the changelogs/data is "
+                         "written to the disk",
+          .default_value = "on",
+        },
         { .key  = {NULL} },
 };
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 387ed12ec..21064db58 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -28,8 +28,10 @@
 #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_LOCKEE_COUNT_MAX    3
+#define AFR_DOM_COUNT_MAX    3
 
 struct _pump_private;
 
@@ -86,26 +88,30 @@ typedef struct afr_inode_ctx_ {
         int32_t  *fresh_children;//increasing order of latency
         afr_spb_state_t mdata_spb;
         afr_spb_state_t data_spb;
+        uint32_t        open_fd_count;
 } afr_inode_ctx_t;
 
 typedef enum {
         NONE,
         INDEX,
+        INDEX_TO_BE_HEALED,
         FULL,
 } afr_crawl_type_t;
 
 typedef struct afr_self_heald_ {
-        gf_boolean_t     enabled;
-        gf_boolean_t     iamshd;
-        afr_crawl_type_t *pending;
-        gf_boolean_t     *inprogress;
-        afr_child_pos_t  *pos;
-        gf_timer_t       **timer;
-        eh_t             *healed;
-        eh_t             *heal_failed;
-        eh_t             *split_brain;
-        char             *node_uuid;
-        int              timeout;
+        gf_boolean_t            enabled;
+        gf_boolean_t            iamshd;
+        afr_crawl_type_t        *pending;
+        gf_boolean_t            *inprogress;
+        afr_child_pos_t         *pos;
+        gf_timer_t              **timer;
+        eh_t                    *healed;
+        eh_t                    *heal_failed;
+        eh_t                    *split_brain;
+        eh_t                    **statistics;
+        void                    **crawl_events;
+        char                    *node_uuid;
+        int                     timeout;
 } afr_self_heald_t;
 
 typedef struct _afr_private {
@@ -170,9 +176,38 @@ typedef struct _afr_private {
         gf_boolean_t           did_discovery;
         gf_boolean_t           readdir_failover;
         uint64_t               sh_readdir_size;
+        gf_boolean_t           ensure_durability;
+        char                   *sh_domain;
 } afr_private_t;
 
+typedef enum {
+        AFR_SELF_HEAL_NOT_ATTEMPTED,
+        AFR_SELF_HEAL_STARTED,
+        AFR_SELF_HEAL_FAILED,
+        AFR_SELF_HEAL_SYNC_BEGIN,
+} afr_self_heal_status;
+
 typedef struct {
+        afr_self_heal_status gfid_or_missing_entry_self_heal;
+        afr_self_heal_status metadata_self_heal;
+        afr_self_heal_status data_self_heal;
+        afr_self_heal_status entry_self_heal;
+} afr_sh_status_for_all_type;
+
+typedef enum {
+        AFR_SELF_HEAL_ENTRY,
+        AFR_SELF_HEAL_METADATA,
+        AFR_SELF_HEAL_DATA,
+        AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY,
+        AFR_SELF_HEAL_INVALID = -1,
+} afr_self_heal_type;
+
+typedef enum {
+        AFR_CHECK_ALL,
+        AFR_CHECK_SPECIFIC,
+} afr_sh_fail_check_type;
+
+struct afr_self_heal_ {
         /* External interface: These are variables (some optional) that
            are set by whoever has triggered self-heal */
 
@@ -249,10 +284,10 @@ typedef struct {
         const char *linkname;
         gf_boolean_t entries_skipped;
 
-        int   op_failed;
         gf_boolean_t actual_sh_started;
         gf_boolean_t sync_done;
         gf_boolean_t data_lock_held;
+        gf_boolean_t sh_dom_lock_held;
         gf_boolean_t eof_reached;
         fd_t  *healing_fd;
         int   file_has_holes;
@@ -263,13 +298,17 @@ typedef struct {
         uint8_t *checksum;
         afr_post_remove_call_t post_remove_call;
 
-        loc_t parent_loc;
+        char    *data_sh_info;
+        char    *metadata_sh_info;
 
+        loc_t parent_loc;
         call_frame_t *orig_frame;
         call_frame_t *old_loop_frame;
         gf_boolean_t unwound;
 
         afr_sh_algo_private_t *private;
+        afr_sh_status_for_all_type  afr_all_sh_status;
+        afr_self_heal_type       sh_type_in_action;
 
         struct afr_sh_algorithm  *algo;
         afr_lock_cbk_t data_lock_success_handler;
@@ -282,7 +321,9 @@ typedef struct {
         void (*gfid_sh_success_cbk) (call_frame_t *sh_frame, xlator_t *this);
 
         call_frame_t *sh_frame;
-} afr_self_heal_t;
+};
+
+typedef struct afr_self_heal_ afr_self_heal_t;
 
 typedef enum {
         AFR_DATA_TRANSACTION,          /* truncate, write, ... */
@@ -356,12 +397,19 @@ 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;
@@ -370,13 +418,11 @@ typedef struct {
 
         unsigned char *locked_nodes;
         unsigned char *lower_locked_nodes;
-        unsigned char *inode_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;
@@ -387,6 +433,7 @@ typedef struct {
         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;
 
 typedef struct _afr_locked_fd {
@@ -406,9 +453,11 @@ typedef struct _afr_local {
         unsigned int call_count;
         unsigned int success_count;
         unsigned int enoent_count;
+        uint32_t     open_fd_count;
+        gf_boolean_t update_open_fd_count;
 
 
-        unsigned int govinda_gOvinda;
+        unsigned int unhealable;
 
         unsigned int read_child_index;
         unsigned char read_child_returned;
@@ -448,15 +497,23 @@ typedef struct _afr_local {
         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;
+        gf_boolean_t      stable_write;
 
-        /*
-          This struct contains the arguments for the "continuation"
-          (scheme-like) of fops
+        /* 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;
+
+        int allow_sh_for_running_transaction;
+
+
+        /* This struct contains the arguments for the "continuation"
+           (scheme-like) of fops
         */
 
         int   op;
@@ -552,7 +609,9 @@ typedef struct _afr_local {
                 struct {
                         struct iatt prebuf;
                         struct iatt postbuf;
+                } inode_wfop; //common structure for all inode-write-fops
 
+                struct {
                         int32_t op_ret;
 
                         struct iovec *vector;
@@ -563,34 +622,21 @@ typedef struct _afr_local {
                 } writev;
 
                 struct {
-                        struct iatt prebuf;
-                        struct iatt postbuf;
-                } fsync;
-
-                struct {
                         off_t offset;
-                        struct iatt prebuf;
-                        struct iatt postbuf;
                 } truncate;
 
                 struct {
                         off_t offset;
-                        struct iatt prebuf;
-                        struct iatt postbuf;
                 } ftruncate;
 
                 struct {
                         struct iatt in_buf;
                         int32_t valid;
-                        struct iatt preop_buf;
-                        struct iatt postop_buf;
                 } setattr;
 
                 struct {
                         struct iatt in_buf;
                         int32_t valid;
-                        struct iatt preop_buf;
-                        struct iatt postop_buf;
                 } fsetattr;
 
                 struct {
@@ -652,6 +698,26 @@ typedef struct _afr_local {
                         dict_t *params;
                         char *linkpath;
                 } symlink;
+
+		struct {
+			int32_t mode;
+			off_t offset;
+			size_t len;
+		} fallocate;
+
+		struct {
+			off_t offset;
+			size_t len;
+		} discard;
+
+                struct {
+                        off_t offset;
+                        size_t len;
+                        struct iatt prebuf;
+                        struct iatt postbuf;
+                } zerofill;
+
+
         } cont;
 
         struct {
@@ -824,8 +890,8 @@ afr_blocking_lock (call_frame_t *frame, xlator_t *this);
 int
 afr_internal_lock_finish (call_frame_t *frame, xlator_t *this);
 
-void
-afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src,
+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);
@@ -927,22 +993,6 @@ afr_launch_openfd_self_heal (call_frame_t *frame, xlator_t *this, fd_t *fd);
                 }                                               \
         } while (0);
 
-#define AFR_CALL_RESUME(stub)					\
-        do {                                                    \
-                afr_local_t *__local = NULL;                    \
-                xlator_t    *__this = NULL;                     \
-								\
-		__local = stub->frame->local;			\
-		__this = stub->frame->this;			\
-		stub->frame->local = NULL;			\
-								\
-		call_resume (stub);				\
-                if (__local) {                                  \
-                        afr_local_cleanup (__local, __this);    \
-                        mem_put (__local);                      \
-                }                                               \
-        } while (0)
-
 #define AFR_NUM_CHANGE_LOGS            3 /*data + metadata + entry*/
 /* allocate and return a string that is the basename of argument */
 static inline char *
@@ -1120,6 +1170,27 @@ afr_xattr_array_destroy (dict_t **xattr, unsigned int child_count);
 } while (0);
 
 
+#define AFR_SBRAIN_MSG "Failed on %s as split-brain is seen. Returning EIO."
+
+#define AFR_SBRAIN_CHECK_FD(fd, label) do {                              \
+        if (fd->inode && afr_is_split_brain (this, fd->inode)) {        \
+                op_errno = EIO;                                         \
+                gf_log (this->name, GF_LOG_WARNING,                     \
+                        AFR_SBRAIN_MSG ,uuid_utoa (fd->inode->gfid));   \
+                goto label;                                             \
+        }                                                               \
+} while (0)
+
+#define AFR_SBRAIN_CHECK_LOC(loc, label) do {                           \
+        if (loc->inode && afr_is_split_brain (this, loc->inode)) {      \
+                op_errno = EIO;                                         \
+                loc_path (loc, NULL);                                   \
+                gf_log (this->name, GF_LOG_WARNING,                     \
+                        AFR_SBRAIN_MSG , loc->path);                    \
+                goto label;                                             \
+        }                                                               \
+} while (0)
+
 int
 afr_fd_report_unstable_write (xlator_t *this, fd_t *fd);
 
@@ -1129,4 +1200,13 @@ 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_inodelk_init (afr_inodelk_t *lk, char *dom, size_t child_count);
+
+void
+afr_handle_open_fd_count (call_frame_t *frame, xlator_t *this);
+
+afr_inode_ctx_t*
+afr_inode_ctx_get (inode_t *inode, xlator_t *this);
+
 #endif /* __AFR_H__ */
diff --git a/xlators/cluster/afr/src/pump.c b/xlators/cluster/afr/src/pump.c
index db83410c7..a7f72fb30 100644
--- a/xlators/cluster/afr/src/pump.c
+++ b/xlators/cluster/afr/src/pump.c
@@ -2501,6 +2501,12 @@ init (xlator_t *this)
 		i++;
 	}
 
+        ret = gf_asprintf (&priv->sh_domain, "%s-self-heal", this->name);
+        if (-1 == ret) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
         priv->first_lookup = 1;
         priv->root_inode = NULL;
 
@@ -2532,7 +2538,7 @@ init (xlator_t *this)
                 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");
@@ -2573,9 +2579,6 @@ fini (xlator_t *this)
         if (!pump_priv)
                 goto afr_priv;
 
-        if (pump_priv->env)
-                syncenv_destroy (pump_priv->env);
-
         GF_FREE (pump_priv->resume_path);
         LOCK_DESTROY (&pump_priv->resume_path_lock);
         LOCK_DESTROY (&pump_priv->pump_state_lock);
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index 2dbcd756d..8f61339e6 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -22,6 +22,7 @@
 #include "dht-common.h"
 #include "defaults.h"
 #include "byte-order.h"
+#include "glusterfs-acl.h"
 
 #include <sys/time.h>
 #include <libgen.h>
@@ -62,6 +63,11 @@ dht_aggregate (dict_t *this, char *key, data_t *value, void *data)
                 }
 
                 *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."))) {
@@ -149,7 +155,6 @@ dht_discover_complete (xlator_t *this, call_frame_t *discover_frame)
         int              ret = -1;
         dht_layout_t    *layout = NULL;
         dht_conf_t      *conf = NULL;
-        uint32_t         missing = 0;
 
         local = discover_frame->local;
         layout = local->layout;
@@ -186,26 +191,20 @@ dht_discover_complete (xlator_t *this, call_frame_t *discover_frame)
                         goto out;
                 }
         } else {
-                ret = dht_layout_normalize (this, &local->loc, layout,
-                                            &missing);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "normalizing failed on %s (internal error)",
-                                local->loc.path);
-                        op_errno = EIO;
-                        goto out;
-                }
-                if (missing == conf->subvolume_cnt) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "normalizing failed on %s, ENOENT errors: %u)",
-                                local->loc.path, missing);
-                        op_errno = ENOENT;
-                        goto out;
-                }
-                if (ret != 0) {
+                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)", local->loc.path);
+                                "(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)
@@ -409,7 +408,6 @@ dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         dht_layout_t *layout                  = NULL;
         int           ret                     = -1;
         int           is_dir                  = 0;
-        uint32_t      missing                 = 0;
 
         GF_VALIDATE_OR_GOTO ("dht", frame, out);
         GF_VALIDATE_OR_GOTO ("dht", this, out);
@@ -443,7 +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,
@@ -490,16 +488,9 @@ unlock:
                 }
 
                 if (local->op_ret == 0) {
-                        ret = dht_layout_normalize (this, &local->loc, layout,
-                                                    &missing);
-
-                        /*
-                         * Arguably, we shouldn't do self-heal just because
-                         * bricks are missing as long as there are no other
-                         * anomalies.  For now, though, just preserve the
-                         * existing behavior.
-                         */
-                        if ((ret != 0) || (missing != 0)) {
+                        ret = dht_layout_normalize (this, &local->loc, layout);
+
+                        if (ret != 0) {
                                 gf_log (this->name, GF_LOG_DEBUG,
                                         "fixing assignment on %s",
                                         local->loc.path);
@@ -1424,7 +1415,6 @@ dht_lookup (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);
 
         conf = this->private;
         if (!conf)
@@ -1645,7 +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)",
@@ -1658,7 +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);
@@ -1797,6 +1788,7 @@ dht_vgetxattr_alloc_and_fill (dht_local_t *local, dict_t *xattr, xlator_t *this,
                 }
 
                 (void) strcat (local->xattr_val, value);
+                (void) strcat (local->xattr_val, " ");
                 local->op_ret = 0;
         }
 
@@ -1821,6 +1813,8 @@ dht_vgetxattr_fill_and_set (dht_local_t *local, dict_t **dict, xlator_t *this,
         if (!*dict)
                 goto out;
 
+        local->xattr_val[strlen (local->xattr_val) - 1] = '\0';
+
         /* we would need max this many bytes to create xattr string
          * extra 40 bytes is just an estimated amount of additional
          * space required as we include translator name and some
@@ -2053,10 +2047,8 @@ dht_getxattr_get_real_filename_cbk (call_frame_t *frame, void *cookie,
 {
         int             this_call_cnt = 0;
         dht_local_t     *local = NULL;
-        dht_conf_t      *conf = NULL;
 
 
-        conf = this->private;
         local = frame->local;
 
 	if (op_ret != -1) {
@@ -2083,7 +2075,6 @@ int
 dht_getxattr_get_real_filename (call_frame_t *frame, xlator_t *this,
 				loc_t *loc, const char *key, dict_t *xdata)
 {
-	dht_conf_t      *conf = NULL;
 	dht_local_t     *local = NULL;
 	int              i = 0;
 	dht_layout_t    *layout = NULL;
@@ -2091,7 +2082,6 @@ dht_getxattr_get_real_filename (call_frame_t *frame, xlator_t *this,
 	xlator_t        *subvol = NULL;
 
 
-        conf   = this->private;
 	local = frame->local;
 	layout = local->layout;
 
@@ -2132,7 +2122,6 @@ 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;
@@ -2177,8 +2166,9 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
          * NOTE: Don't trust inode here, as that may not be valid
          *       (until inode_link() happens)
          */
-        if (key && (strcmp (key, GF_XATTR_PATHINFO_KEY) == 0)
-            && DHT_IS_DIR(layout)) {
+        if (key && DHT_IS_DIR(layout) &&
+            ((strcmp (key, GF_XATTR_PATHINFO_KEY) == 0)
+             || (strcmp (key, GF_XATTR_NODE_UUID_KEY) == 0))) {
                 (void) strncpy (local->xsel, key, 256);
                 cnt = local->call_cnt = layout->cnt;
                 for (i = 0; i < cnt; i++) {
@@ -2249,7 +2239,8 @@ 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;
                 }
@@ -2273,6 +2264,7 @@ 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;
@@ -2488,7 +2480,6 @@ dht_setxattr (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);
 
         conf   = this->private;
 
@@ -2702,7 +2693,6 @@ dht_removexattr (call_frame_t *frame, xlator_t *this,
         VALIDATE_OR_GOTO (frame, 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_REMOVEXATTR);
         if (!local) {
@@ -2934,7 +2924,6 @@ dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
         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;
@@ -3055,7 +3044,7 @@ 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_dir (NULL, (&orig_entry->d_stat), NULL) &&
-                    (prev->this != dht_first_up_subvol (this))) {
+                    (prev->this != local->first_up_subvol)) {
                         continue;
                 }
                 if (check_is_linkfile (NULL, (&orig_entry->d_stat),
@@ -3137,13 +3126,16 @@ done:
                 }
 
 		if (conf->readdir_optimize == _gf_true) {
-                        if (next_subvol != dht_first_up_subvol (this)) {
+                        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,
@@ -3289,6 +3281,7 @@ dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
         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);
 
@@ -3307,13 +3300,16 @@ dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
                                         "failed to set '%s' key",
                                         conf->link_xattr_name);
 			if (conf->readdir_optimize == _gf_true) {
-                                if (xvol != dht_first_up_subvol (this)) {
+                                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);
                                 }
 			}
                 }
@@ -3571,7 +3567,9 @@ dht_mknod (call_frame_t *frame, xlator_t *this,
                                    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 */
@@ -3992,7 +3990,7 @@ dht_create (call_frame_t *frame, xlator_t *this,
         }
         /* Choose the minimum filled volume, and create the
            files there */
-        avail_subvol = dht_free_disk_available_subvol (this, subvol);
+        avail_subvol = dht_free_disk_available_subvol (this, subvol, local);
         if (avail_subvol != subvol) {
                 local->params = dict_ref (params);
                 local->flags = flags;
@@ -4868,7 +4866,6 @@ dht_entrylk (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);
 
         local = dht_local_init (frame, loc, NULL, GF_FOP_ENTRYLK);
         if (!local) {
@@ -5176,8 +5173,8 @@ unlock:
                  * not need to handle CHILD_DOWN event here.
                  */
                 if (conf->defrag) {
-                        ret = pthread_create (&conf->defrag->th, NULL,
-                                              gf_defrag_start, this);
+                        ret = gf_thread_create (&conf->defrag->th, NULL,
+						gf_defrag_start, this);
                         if (ret) {
                                 conf->defrag = NULL;
                                 GF_FREE (conf->defrag);
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index 9de861360..5ccd66799 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -183,6 +183,7 @@ struct dht_local {
         xlator_t        *link_subvol;
 
         struct dht_rebalance_ rebalance;
+        xlator_t        *first_up_subvol;
 
 };
 typedef struct dht_local dht_local_t;
@@ -211,6 +212,10 @@ enum gf_defrag_status_t {
         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;
 
@@ -227,6 +232,7 @@ struct gf_defrag_info_ {
         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;
@@ -394,26 +400,18 @@ typedef enum {
         } while (0)
 
 #define is_greater_time(a, an, b, bn) (((a) < (b)) || (((a) == (b)) && ((an) < (bn))))
-
-dht_layout_t    *dht_layout_new         (xlator_t *this, int cnt);
-dht_layout_t    *dht_layout_get         (xlator_t *this, inode_t *inode);
-dht_layout_t    *dht_layout_for_subvol  (xlator_t *this, xlator_t *subvol);
-xlator_t        *dht_layout_search      (xlator_t *this, dht_layout_t *layout,
-                                         const char *name);
-int             dht_layout_normalize    (xlator_t *this, loc_t *loc,
-                                         dht_layout_t *layout,
-                                         uint32_t *missing_p);
-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);
+dht_layout_t                            *dht_layout_new (xlator_t *this, int cnt);
+dht_layout_t                            *dht_layout_get (xlator_t *this, inode_t *inode);
+dht_layout_t                            *dht_layout_for_subvol (xlator_t *this, xlator_t *subvol);
+xlator_t *dht_layout_search (xlator_t   *this, dht_layout_t *layout,
+                             const char *name);
+int                                      dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout);
+int dht_layout_anomalies (xlator_t      *this, loc_t *loc, dht_layout_t *layout,
+                          uint32_t      *holes_p, uint32_t *overlaps_p,
+                          uint32_t      *missing_p, uint32_t *down_p,
+                          uint32_t      *misc_p, uint32_t *no_space_p);
+int dht_layout_dir_mismatch (xlator_t   *this, dht_layout_t *layout,
+                             xlator_t   *subvol, loc_t *loc, dict_t *xattr);
 
 xlator_t *dht_linkfile_subvol (xlator_t *this, inode_t *inode,
                                struct iatt *buf, dict_t *xattr);
@@ -445,6 +443,7 @@ int dht_iatt_merge (xlator_t                 *this, struct iatt *to, struct iatt
 xlator_t *dht_subvol_get_hashed (xlator_t *this, loc_t *loc);
 xlator_t *dht_subvol_get_cached (xlator_t *this, inode_t *inode);
 xlator_t *dht_subvol_next (xlator_t *this, xlator_t *prev);
+xlator_t *dht_subvol_next_available (xlator_t *this, xlator_t *prev);
 int       dht_subvol_cnt (xlator_t *this, xlator_t *subvol);
 
 int dht_hash_compute (xlator_t *this, int type, const char *name, uint32_t *hash_p);
@@ -469,7 +468,8 @@ dht_layout_sort_volname (dht_layout_t *layout);
 int dht_get_du_info (call_frame_t *frame, xlator_t *this, loc_t *loc);
 
 gf_boolean_t dht_is_subvol_filled (xlator_t *this, xlator_t *subvol);
-xlator_t *dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol);
+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);
@@ -691,6 +691,12 @@ 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, size_t len, dict_t *xdata);
 
 int32_t dht_init (xlator_t *this);
 void    dht_fini (xlator_t *this);
@@ -760,9 +766,11 @@ 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_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_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);
 
@@ -773,4 +781,7 @@ 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 0c87f4a64..fe3955ecb 100644
--- a/xlators/cluster/dht/src/dht-diskusage.c
+++ b/xlators/cluster/dht/src/dht-diskusage.c
@@ -251,25 +251,45 @@ dht_is_subvol_filled (xlator_t *this, xlator_t *subvol)
 
 /*Get the best subvolume to create the file in*/
 xlator_t *
-dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol)
+dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol,
+                                dht_local_t *local)
 {
 	xlator_t   *avail_subvol = NULL;
 	dht_conf_t *conf = NULL;
+        dht_layout_t *layout = NULL;
+        loc_t      *loc = NULL;
 
 	conf = this->private;
+        if (!local)
+                goto out;
+        loc = &local->loc;
+        if (!local->layout) {
+                layout = dht_layout_get (this, loc->parent);
+
+                if (!layout) {
+                        gf_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);
+        }
 
-	LOCK (&conf->subvolume_lock);
+        LOCK (&conf->subvolume_lock);
 	{
-                avail_subvol = dht_subvol_with_free_space_inodes(this, subvol);
+                avail_subvol = dht_subvol_with_free_space_inodes(this, subvol,
+                                                                 layout);
                 if(!avail_subvol)
                 {
                         avail_subvol = dht_subvol_maxspace_nonzeroinode(this,
-                                                                        subvol);
+                                                                        subvol,
+                                                                        layout);
                 }
 
 	}
 	UNLOCK (&conf->subvolume_lock);
-
+out:
 	if (!avail_subvol) {
 		gf_log (this->name,
                         GF_LOG_DEBUG,
@@ -278,17 +298,42 @@ dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol)
                 avail_subvol = subvol;
 	}
 
-
+        if (layout)
+                dht_layout_unref (this, layout);
 	return avail_subvol;
 }
 
+static inline
+int32_t dht_subvol_has_err (xlator_t *this, dht_layout_t *layout)
+{
+        int ret = -1;
+        int i   = 0;
+
+        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_subvol_with_free_space_inodes(xlator_t *this, xlator_t *subvol,
+                                  dht_layout_t *layout)
 {
         int i = 0;
         double max = 0;
         double max_inodes = 0;
+        int    ignore_subvol = 0;
 
         xlator_t *avail_subvol = NULL;
         dht_conf_t *conf = NULL;
@@ -296,6 +341,12 @@ dht_subvol_with_free_space_inodes(xlator_t *this, xlator_t *subvol)
         conf = this->private;
 
         for(i=0; i < conf->subvolume_cnt; i++) {
+                /* check if subvol has layout errors, before selecting it */
+                ignore_subvol = dht_subvol_has_err (conf->subvolumes[i],
+                                                    layout);
+                if (ignore_subvol)
+                        continue;
+
                 if ((conf->disk_unit == 'p') &&
                     (conf->du_stats[i].avail_percent > conf->min_free_disk) &&
                     (conf->du_stats[i].avail_inodes  > conf->min_free_inodes)) {
@@ -325,10 +376,12 @@ dht_subvol_with_free_space_inodes(xlator_t *this, xlator_t *subvol)
 
 /* Get subvol which has atleast one inode and maximum space */
 xlator_t *
-dht_subvol_maxspace_nonzeroinode (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;
+        int         ignore_subvol = 0;
 
         xlator_t   *avail_subvol = NULL;
         dht_conf_t *conf = NULL;
@@ -336,6 +389,12 @@ dht_subvol_maxspace_nonzeroinode (xlator_t *this, xlator_t *subvol)
         conf = this->private;
 
         for (i = 0; i < conf->subvolume_cnt; i++) {
+                /* check if subvol has layout errors, before selecting it */
+                ignore_subvol = dht_subvol_has_err (conf->subvolumes[i],
+                                                    layout);
+                if (ignore_subvol)
+                        continue;
+
                 if (conf->disk_unit == 'p') {
                         if ((conf->du_stats[i].avail_percent > max)
                             && (conf->du_stats[i].avail_inodes > 0 )) {
diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c
index 23fc1ff96..311a48112 100644
--- a/xlators/cluster/dht/src/dht-helper.c
+++ b/xlators/cluster/dht/src/dht-helper.c
@@ -18,6 +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)
@@ -340,20 +362,6 @@ out:
         return local;
 }
 
-
-char *
-basestr (const char *str)
-{
-        char *basestr = NULL;
-
-        basestr = strrchr (str, '/');
-        if (basestr)
-                basestr ++;
-
-        return basestr;
-}
-
-
 xlator_t *
 dht_first_up_subvol (xlator_t *this)
 {
@@ -505,7 +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)
 {
@@ -698,6 +735,10 @@ dht_migration_complete_check_task (void *data)
         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;
@@ -709,6 +750,8 @@ dht_migration_complete_check_task (void *data)
         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*/
 
@@ -771,10 +814,7 @@ dht_migration_complete_check_task (void *data)
         /* update inode ctx (the layout) */
         dht_layout_unref (this, local->layout);
 
-        if (!local->loc.inode)
-                ret = dht_layout_preset (this, dst_node, local->fd->inode);
-        else
-                ret = dht_layout_preset (this, dst_node, local->loc.inode);
+        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",
@@ -792,10 +832,7 @@ dht_migration_complete_check_task (void *data)
                 goto out;
         }
 
-        if (!local->loc.inode)
-                ret = dht_layout_set (this, local->fd->inode, layout);
-        else
-                ret = dht_layout_set (this, local->loc.inode, layout);
+        ret = dht_layout_set (this, inode, layout);
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR,
                         "%s: failed to set the new layout",
@@ -806,41 +843,46 @@ dht_migration_complete_check_task (void *data)
         local->cached_subvol = dst_node;
         ret = 0;
 
-        if (!local->fd)
+        /* 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;
-        /* once we detect the migration complete, the fd-ctx is no more
-           required.. delete the ctx */
-        ret = fd_ctx_del (local->fd, this, NULL);
-        if (!ret)
+
+        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);
-        /* if 'local->fd' (ie, fd based operation), send a 'open()' on
-           destination if not already done */
-        if (local->loc.inode) {
-                ret = syncop_open (dst_node, &local->loc,
-                                   local->fd->flags, local->fd);
-        } else {
-                tmp_loc.inode = local->fd->inode;
-                inode_path (local->fd->inode, NULL, &path);
-                if (path)
-                        tmp_loc.path = path;
-                ret = syncop_open (dst_node, &tmp_loc,
-                                   local->fd->flags, local->fd);
-                GF_FREE (path);
 
+        /* 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;
+
+                ret = syncop_open (dst_node, &tmp_loc,
+                                   iter_fd->flags, iter_fd);
+                if (ret == -1) {
+                        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;
+                }
         }
+        GF_FREE (path);
+
         SYNCTASK_SETID (frame->root->uid, frame->root->gid);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to send open() on target file at %s",
-                        local->loc.path, dst_node->name);
+        if (open_failed) {
+                ret = -1;
                 goto out;
         }
-
         ret = 0;
 out:
 
@@ -885,6 +927,9 @@ dht_rebalance_inprogress_task (void *data)
         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;
@@ -896,6 +941,8 @@ dht_rebalance_inprogress_task (void *data)
         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) {
@@ -947,35 +994,47 @@ dht_rebalance_inprogress_task (void *data)
         }
 
         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);
-        if (local->loc.inode) {
-                ret = syncop_open (dst_node, &local->loc,
-                                   local->fd->flags, local->fd);
-        } else {
-                tmp_loc.inode = local->fd->inode;
-                inode_path (local->fd->inode, NULL, &path);
-                if (path)
-                        tmp_loc.path = path;
+
+        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;
+
                 ret = syncop_open (dst_node, &tmp_loc,
-                                   local->fd->flags, local->fd);
-                GF_FREE (path);
+                                   iter_fd->flags, iter_fd);
+                if (ret == -1) {
+                        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);
+                        open_failed = 1;
+                }
         }
+        GF_FREE (path);
 
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to send open() on target file at %s",
-                        local->loc.path, dst_node->name);
+        SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+
+        if (open_failed) {
+                ret = -1;
                 goto out;
         }
 
-        SYNCTASK_SETID (frame->root->uid, frame->root->gid);
-        ret = fd_ctx_set (local->fd, this, (uint64_t)(long)dst_node);
+done:
+        ret = dht_inode_ctx_set1 (this, inode, dst_node);
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to set fd-ctx target file at %s",
+                        "%s: failed to set inode-ctx target file at %s",
                         local->loc.path, dst_node->name);
                 goto out;
         }
diff --git a/xlators/cluster/dht/src/dht-inode-read.c b/xlators/cluster/dht/src/dht-inode-read.c
index d9b311fe2..ece84151a 100644
--- a/xlators/cluster/dht/src/dht-inode-read.c
+++ b/xlators/cluster/dht/src/dht-inode-read.c
@@ -130,10 +130,11 @@ 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)
 {
-        uint64_t      tmp_subvol = 0;
+        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);
@@ -157,19 +158,20 @@ dht_file_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         local->op_errno = op_errno;
         /* Check if the rebalance phase2 is true */
         if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (stbuf)) {
-                if (local->fd)
-                        ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-                if (ret) {
+                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);
-                }
-                if (!ret)
                         return 0;
+                }
         }
 
 out:
@@ -382,6 +384,8 @@ dht_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 {
         dht_local_t *local      = NULL;
         int          ret        = 0;
+        inode_t     *inode      = NULL;
+        xlator_t    *subvol = 0;
 
         local = frame->local;
         if (!local) {
@@ -400,17 +404,18 @@ dht_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         local->op_errno = op_errno;
         if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (stbuf)) {
                 /* File would be migrated to other node */
-                ret = fd_ctx_get (local->fd, this, NULL);
-                if (ret) {
+                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);
-                }
-                if (!ret)
                         return 0;
+                }
         }
 
 out:
@@ -501,18 +506,27 @@ dht_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         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_first_up_subvol (this);
+                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;
@@ -607,8 +621,9 @@ 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;
-        int           ret = -1;
+        dht_local_t  *local  = NULL;
+        inode_t      *inode  = NULL;
+        xlator_t     *subvol = 0;
 
         local = frame->local;
 
@@ -618,8 +633,8 @@ dht_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 goto out;
 
         /* If context is set, then send flush() it to the destination */
-        ret = fd_ctx_get (local->fd, this, NULL);
-        if (!ret) {
+        dht_inode_ctx_get1 (this, inode, &subvol);
+        if (subvol) {
                 dht_flush2 (this, frame, 0);
                 return 0;
         }
@@ -635,14 +650,10 @@ dht_flush2 (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;
 
-        ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
 
         if (!subvol)
                 subvol = local->cached_subvol;
@@ -704,6 +715,8 @@ dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
         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;
@@ -725,8 +738,8 @@ dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
         }
 
         local->op_errno = op_errno;
-        ret = fd_ctx_get (local->fd, this, NULL);
-        if (ret) {
+        dht_inode_ctx_get1 (this, inode, &subvol);
+        if (!subvol) {
                 local->rebalance.target_op_fn = dht_fsync2;
 
                 /* Check if the rebalance phase1 is true */
@@ -741,11 +754,12 @@ dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
                 if (IS_DHT_MIGRATION_PHASE2 (postbuf)) {
                         ret = dht_rebalance_complete_check (this, frame);
                 }
+                if (!ret)
+                        return 0;
         } else {
                 dht_fsync2 (this, frame, 0);
-        }
-        if (!ret)
                 return 0;
+        }
 
 out:
         DHT_STRIP_PHASE1_FLAGS (postbuf);
@@ -761,15 +775,10 @@ dht_fsync2 (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;
 
-        ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
-
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
         if (!subvol)
                 subvol = local->cached_subvol;
 
diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c
index b87d2f732..4b3f3a049 100644
--- a/xlators/cluster/dht/src/dht-inode-write.c
+++ b/xlators/cluster/dht/src/dht-inode-write.c
@@ -19,6 +19,9 @@
 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,
@@ -27,6 +30,7 @@ dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 {
         dht_local_t *local = NULL;
         int          ret   = -1;
+        xlator_t    *subvol = NULL;
 
         if (op_ret == -1 && (op_errno != ENOENT)) {
                 goto out;
@@ -63,8 +67,8 @@ dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 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) {
+                ret = dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+                if (subvol) {
                         dht_writev2 (this, frame, 0);
                         return 0;
                 }
@@ -88,14 +92,10 @@ dht_writev2 (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;
 
-        ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
 
         if (!subvol)
                 subvol = local->cached_subvol;
@@ -170,6 +170,8 @@ dht_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         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);
@@ -211,8 +213,9 @@ dht_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
                 dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
                 dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
-                ret = fd_ctx_get (local->fd, this, NULL);
-                if (!ret || !local->fd) {
+                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;
                 }
@@ -236,16 +239,13 @@ dht_truncate2 (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;
+        inode_t      *inode = NULL;
 
         local = frame->local;
 
-        if (local->fd)
-                ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
+        inode = local->fd ? local->fd->inode : local->loc.inode;
 
+        dht_inode_ctx_get1 (this, inode, &subvol);
         if (!subvol)
                 subvol = local->cached_subvol;
 
@@ -348,6 +348,407 @@ err:
         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) && (op_errno != ENOENT)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+
+                goto out;
+        }
+
+        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) && (op_errno != ENOENT)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+
+                goto out;
+        }
+
+        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) && (op_errno != ENOENT)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        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,
+            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_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,
@@ -399,15 +800,13 @@ dht_setattr2 (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;
+        inode_t      *inode = NULL;
 
         local = frame->local;
 
-        if (local->fd)
-                ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
+
+        dht_inode_ctx_get1 (this, inode, &subvol);
 
         if (!subvol)
                 subvol = local->cached_subvol;
diff --git a/xlators/cluster/dht/src/dht-layout.c b/xlators/cluster/dht/src/dht-layout.c
index 07e8cbae4..38e9970a7 100644
--- a/xlators/cluster/dht/src/dht-layout.c
+++ b/xlators/cluster/dht/src/dht-layout.c
@@ -454,12 +454,19 @@ dht_layout_entry_cmp (dht_layout_t *layout, int i, int j)
 {
         int64_t diff = 0;
 
+        /* swap zero'ed out layouts to front, if needed */
+        if (!layout->list[j].start && !layout->list[j].stop) {
+                diff = (int64_t) layout->list[i].stop
+                       - (int64_t) layout->list[j].stop;
+                       goto out;
+        }
         if (layout->list[i].err || layout->list[j].err)
                 diff = layout->list[i].err - layout->list[j].err;
         else
                 diff = (int64_t) layout->list[i].start
                         - (int64_t) layout->list[j].start;
 
+out:
         return diff;
 }
 
@@ -534,13 +541,13 @@ dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
                 case -1:
                 case ENOENT:
                         missing++;
-                        break;
+                        continue;
                 case ENOTCONN:
                         down++;
-                        break;
+                        continue;
                 case ENOSPC:
                         no_space++;
-                        break;
+                        continue;
                 case 0:
                         /* if err == 0 and start == stop, then it is a non misc++;
                          * participating subvolume(spread-cnt). Then, do not
@@ -552,6 +559,7 @@ dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
                         break;
                 default:
                         misc++;
+                        continue;
                  }
 
                 is_virgin = 0;
@@ -593,8 +601,7 @@ dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
 
 
 int
-dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout,
-                      uint32_t *missing_p)
+dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout)
 {
         int          ret   = 0;
         int          i = 0;
@@ -606,7 +613,6 @@ dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout,
 
         ret = dht_layout_sort (layout);
         if (ret == -1) {
-                /* defensive coding; this can't happen currently */
                 gf_log (this->name, GF_LOG_WARNING,
                         "sort failed?! how the ....");
                 goto out;
@@ -616,26 +622,23 @@ dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout,
                                     &holes, &overlaps,
                                     &missing, &down, &misc, NULL);
         if (ret == -1) {
-                /* defensive coding; this can't happen currently */
                 gf_log (this->name, GF_LOG_WARNING,
                         "error while finding anomalies in %s -- not good news",
                         loc->path);
                 goto out;
         }
 
-        ret = holes + overlaps;
-        if (ret) {
+        if (holes || overlaps) {
                 if (missing == layout->cnt) {
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "directory %s looked up first time",
                                 loc->path);
                 } else {
                         gf_log (this->name, GF_LOG_INFO,
-                                "found anomalies in %s. holes=%d overlaps=%d"
-                                " missing=%d down=%d misc=%d",
-                                loc->path, holes, overlaps, missing, down,
-                                misc);
+                                "found anomalies in %s. holes=%d overlaps=%d",
+                                loc->path, holes, overlaps);
                 }
+                ret = -1;
         }
 
         for (i = 0; i < layout->cnt; i++) {
@@ -650,14 +653,14 @@ dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout,
                                           (layout->list[i].xlator ?
                                            layout->list[i].xlator->name
                                            : "<>"));
+                        if ((layout->list[i].err == ENOENT) && (ret >= 0)) {
+                                ret++;
+                        }
                 }
         }
 
 
 out:
-        if (missing_p) {
-                *missing_p = missing;
-        }
         return ret;
 }
 
diff --git a/xlators/cluster/dht/src/dht-linkfile.c b/xlators/cluster/dht/src/dht-linkfile.c
index 39d72ae63..dbc9d0b3c 100644
--- a/xlators/cluster/dht/src/dht-linkfile.c
+++ b/xlators/cluster/dht/src/dht-linkfile.c
@@ -297,11 +297,11 @@ dht_linkfile_attr_heal (call_frame_t *frame, xlator_t *this)
         GF_VALIDATE_OR_GOTO ("dht", local, out);
         GF_VALIDATE_OR_GOTO ("dht", local->link_subvol, out);
 
-        if ((local->stbuf.ia_type == IA_INVAL) ||
-            (is_equal (frame->root->uid, local->stbuf.ia_uid) &&
-             is_equal (frame->root->gid, local->stbuf.ia_gid)))
+        if (local->stbuf.ia_type == IA_INVAL)
                 return 0;
 
+        uuid_copy (local->loc.gfid, local->stbuf.ia_gfid);
+
         copy = copy_frame (frame);
 
         if (!copy)
diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c
index a9f913c2d..bcb19f23e 100644
--- a/xlators/cluster/dht/src/dht-rebalance.c
+++ b/xlators/cluster/dht/src/dht-rebalance.c
@@ -243,7 +243,7 @@ out:
 
 static inline int
 __dht_rebalance_create_dst_file (xlator_t *to, xlator_t *from, loc_t *loc, struct iatt *stbuf,
-                                 dict_t *dict, fd_t **dst_fd)
+                                 dict_t *dict, fd_t **dst_fd, dict_t *xattr)
 {
         xlator_t    *this = NULL;
         int          ret  = -1;
@@ -307,6 +307,12 @@ __dht_rebalance_create_dst_file (xlator_t *to, xlator_t *from, loc_t *loc, struc
                 goto out;
         }
 
+        ret = syncop_fsetxattr (to, fd, xattr, 0);
+        if (ret == -1)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to set xattr on %s (%s)",
+                        loc->path, to->name, strerror (errno));
+
         ret = syncop_ftruncate (to, fd, stbuf->ia_size);
         if (ret < 0)
                 gf_log (this->name, GF_LOG_ERROR,
@@ -340,6 +346,9 @@ __dht_check_free_space (xlator_t *to, xlator_t *from, loc_t *loc,
         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);
@@ -363,22 +372,34 @@ __dht_check_free_space (xlator_t *to, xlator_t *from, loc_t *loc,
         if (flag != GF_DHT_MIGRATE_DATA)
                 goto check_avail_space;
 
-        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);
-
-                /* this is not a 'failure', but we don't want to
-                   consider this as 'success' too :-/ */
-                ret = 1;
-                goto out;
+        /* 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) {
@@ -620,6 +641,15 @@ migrate_special_files (xlator_t *this, xlator_t *from, xlator_t *to, loc_t *loc,
         }
 
 done:
+        ret = syncop_setattr (to, loc, buf,
+                              (GF_SET_ATTR_UID | GF_SET_ATTR_GID |
+                               GF_SET_ATTR_MODE), NULL, NULL);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to perform setattr on %s (%s)",
+                        loc->path, to->name, strerror (errno));
+        }
+
         ret = syncop_unlink (from, loc);
         if (ret)
                 gf_log (this->name, GF_LOG_WARNING, "%s: unlink failed (%s)",
@@ -699,9 +729,16 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
                 goto out;
         }
 
+        /* TODO: move all xattr related operations to fd based operations */
+        ret = syncop_listxattr (from, loc, &xattr);
+        if (ret == -1)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to get xattr from %s (%s)",
+                        loc->path, from->name, strerror (errno));
+
         /* create the destination, with required modes/xattr */
         ret = __dht_rebalance_create_dst_file (to, from, loc, &stbuf,
-                                               dict, &dst_fd);
+                                               dict, &dst_fd, xattr);
         if (ret)
                 goto out;
 
@@ -718,6 +755,7 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
                 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)",
@@ -747,19 +785,6 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
                 goto out;
         }
 
-        /* TODO: move all xattr related operations to fd based operations */
-        ret = syncop_listxattr (from, loc, &xattr);
-        if (ret == -1)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to get xattr from %s (%s)",
-                        loc->path, from->name, strerror (errno));
-
-        ret = syncop_setxattr (to, loc, xattr, 0);
-        if (ret == -1)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to set xattr on %s (%s)",
-                        loc->path, to->name, strerror (errno));
-
         /* TODO: Sync the locks */
 
         ret = syncop_fsync (to, dst_fd, 0);
@@ -823,6 +848,23 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
                 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 (errno));
+        }
+
+        /* remove the 'linkto' xattr from the destination */
+        ret = syncop_fremovexattr (to, dst_fd, conf->link_xattr_name);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to perform removexattr on %s (%s)",
+                        loc->path, to->name, strerror (errno));
+        }
+
         /* Do a stat and check the gfid before unlink */
         ret = syncop_stat (from, loc, &empty_iatt);
         if (ret) {
@@ -843,23 +885,6 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
                 }
         }
 
-        /* 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 (errno));
-        }
-
-        /* remove the 'linkto' xattr from the destination */
-        ret = syncop_fremovexattr (to, dst_fd, conf->link_xattr_name);
-        if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to perform removexattr on %s (%s)",
-                        loc->path, to->name, strerror (errno));
-        }
-
         ret = syncop_lookup (this, loc, NULL, NULL, NULL, NULL);
         if (ret) {
                 gf_log (this->name, GF_LOG_DEBUG,
@@ -1101,6 +1126,7 @@ gf_defrag_migrate_data (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
         struct timeval           end            = {0,};
         double                   elapsed        = {0,};
         struct timeval           start          = {0,};
+        int32_t                  err            = 0;
 
         gf_log (this->name, GF_LOG_INFO, "migrate data called on %s",
                 loc->path);
@@ -1258,9 +1284,21 @@ gf_defrag_migrate_data (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
                         ret = syncop_setxattr (this, &entry_loc, migrate_data,
                                                0);
                         if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR, "migrate-data"
-                                        " failed for %s", entry_loc.path);
-                                defrag->total_failures +=1;
+                                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 == -1) {
@@ -1659,6 +1697,7 @@ gf_defrag_status_get (gf_defrag_info_t *defrag, dict_t *dict)
         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,};
@@ -1675,6 +1714,7 @@ gf_defrag_status_get (gf_defrag_info_t *defrag, dict_t *dict)
         size   = defrag->total_data;
         lookup = defrag->num_files_lookedup;
         failures = defrag->total_failures;
+        skipped = defrag->skipped;
 
         gettimeofday (&end, NULL);
 
@@ -1698,6 +1738,7 @@ gf_defrag_status_get (gf_defrag_info_t *defrag, dict_t *dict)
                 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,
@@ -1710,6 +1751,14 @@ gf_defrag_status_get (gf_defrag_info_t *defrag, dict_t *dict)
         }
 
         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:
@@ -1727,13 +1776,15 @@ log:
         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, files, size,
-                lookup, failures);
+                PRIu64", lookups: %"PRIu64", failures: %"PRIu64", skipped: "
+                "%"PRIu64, files, size, lookup, failures, skipped);
 
 
 out:
diff --git a/xlators/cluster/dht/src/dht-rename.c b/xlators/cluster/dht/src/dht-rename.c
index 08965976b..5d6f4f232 100644
--- a/xlators/cluster/dht/src/dht-rename.c
+++ b/xlators/cluster/dht/src/dht-rename.c
@@ -306,7 +306,19 @@ err:
                           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)
 int
 dht_rename_done (call_frame_t *frame, xlator_t *this)
 {
@@ -377,7 +389,7 @@ 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;
@@ -401,13 +413,15 @@ 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) {
                 gf_log (this->name, GF_LOG_TRACE,
                         "unlinking linkfile %s @ %s => %s",
                         local->loc.path, dst_hashed->name, src_cached->name);
                 STACK_WIND (frame, dht_rename_unlink_cbk,
                             dst_hashed, dst_hashed->fops->unlink,
-                            &local->loc, 0, NULL);
+                            &local->loc, 0, xattr);
         }
 
         if (src_cached != dst_hashed) {
@@ -416,9 +430,12 @@ dht_rename_cleanup (call_frame_t *frame)
                         local->loc2.path, src_cached->name);
                 STACK_WIND (frame, dht_rename_unlink_cbk,
                             src_cached, src_cached->fops->unlink,
-                            &local->loc2, 0, NULL);
+                            &local->loc2, 0, xattr);
         }
 
+        if (xattr)
+                dict_unref (xattr);
+
         return 0;
 
 nolinks:
@@ -482,6 +499,7 @@ dht_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         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;
@@ -565,6 +583,8 @@ err:
         if (local->call_cnt == 0)
                 goto unwind;
 
+        DHT_MARK_FOP_INTERNAL (xattr);
+
         if (src_cached != dst_hashed && src_cached != dst_cached) {
                 gf_log (this->name, GF_LOG_TRACE,
                         "deleting old src datafile %s @ %s",
@@ -572,7 +592,7 @@ err:
 
                 STACK_WIND (frame, dht_rename_unlink_cbk,
                             src_cached, src_cached->fops->unlink,
-                            &local->loc, 0, NULL);
+                            &local->loc, 0, xattr);
         }
 
         if (src_hashed != rename_subvol && src_hashed != src_cached) {
@@ -582,7 +602,7 @@ err:
 
                 STACK_WIND (frame, dht_rename_unlink_cbk,
                             src_hashed, src_hashed->fops->unlink,
-                            &local->loc, 0, NULL);
+                            &local->loc, 0, xattr);
         }
 
         if (dst_cached
@@ -594,8 +614,10 @@ err:
 
                 STACK_WIND (frame, dht_rename_unlink_cbk,
                             dst_cached, dst_cached->fops->unlink,
-                            &local->loc2, 0, NULL);
+                            &local->loc2, 0, xattr);
         }
+        if (xattr)
+                dict_unref (xattr);
         return 0;
 
 unwind:
@@ -603,12 +625,16 @@ unwind:
         WIPE (&local->postoldparent);
         WIPE (&local->preparent);
         WIPE (&local->postparent);
+        if (xattr)
+                dict_unref (xattr);
 
         dht_rename_done (frame, this);
 
         return 0;
 
 cleanup:
+        if (xattr)
+                dict_unref (xattr);
         dht_rename_cleanup (frame);
 
         return 0;
@@ -742,6 +768,7 @@ 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;
@@ -752,6 +779,7 @@ 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) {
                 if (dst_hashed == dst_cached)
@@ -763,7 +791,7 @@ dht_rename_create_links (call_frame_t *frame)
 
 		STACK_WIND (frame, dht_rename_unlink_links_cbk,
 			    dst_hashed, dst_hashed->fops->unlink,
-			    &local->loc2, 0, NULL);
+			    &local->loc2, 0, xattr);
                 return 0;
         }
 
@@ -790,7 +818,7 @@ dht_rename_create_links (call_frame_t *frame)
 			local->loc2.path, src_cached->name);
 		STACK_WIND (frame, dht_rename_links_cbk,
 			    src_cached, src_cached->fops->link,
-			    &local->loc, &local->loc2, NULL);
+			    &local->loc, &local->loc2, xattr);
 	}
 
 nolinks:
@@ -798,6 +826,8 @@ nolinks:
                 /* skip to next step */
                 dht_do_rename (frame);
         }
+        if (xattr)
+                dict_unref (xattr);
 
         return 0;
 }
diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c
index b220a0e25..3fe96b1c7 100644
--- a/xlators/cluster/dht/src/dht-selfheal.c
+++ b/xlators/cluster/dht/src/dht-selfheal.c
@@ -17,6 +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;                            \
@@ -564,9 +565,33 @@ dht_get_layout_count (xlator_t *this, dht_layout_t *layout, int new_layout)
 
         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;
                 }
         }
 
@@ -776,7 +801,7 @@ dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc,
         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) {
@@ -789,7 +814,7 @@ 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) {
diff --git a/xlators/cluster/dht/src/dht-shared.c b/xlators/cluster/dht/src/dht-shared.c
index 6b91c56a1..70aac7710 100644
--- a/xlators/cluster/dht/src/dht-shared.c
+++ b/xlators/cluster/dht/src/dht-shared.c
@@ -262,6 +262,28 @@ out:
         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)
@@ -358,6 +380,10 @@ dht_reconfigure (xlator_t *this, dict_t *options)
                 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",
diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c
index 814f0a8eb..fc0ca2f77 100644
--- a/xlators/cluster/dht/src/dht.c
+++ b/xlators/cluster/dht/src/dht.c
@@ -70,6 +70,9 @@ struct xlator_fops fops = {
         .fxattrop    = dht_fxattrop,
         .setattr     = dht_setattr,
         .fsetattr    = dht_fsetattr,
+	.fallocate   = dht_fallocate,
+	.discard     = dht_discard,
+        .zerofill    = dht_zerofill,
 };
 
 struct xlator_dumpops dumpops = {
diff --git a/xlators/cluster/dht/src/nufa.c b/xlators/cluster/dht/src/nufa.c
index 5e5c68058..e934acdf0 100644
--- a/xlators/cluster/dht/src/nufa.c
+++ b/xlators/cluster/dht/src/nufa.c
@@ -323,7 +323,8 @@ 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) {
@@ -427,7 +428,8 @@ 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) {
@@ -482,97 +484,141 @@ same_first_part (char *str1, char term1, char *str2, char term2)
         }
 }
 
+typedef struct nufa_args {
+        xlator_t    *this;
+        char        *volname;
+        gf_boolean_t addr_match;
+} nufa_args_t;
+
+static void
+nufa_find_local_brick (xlator_t *xl, void *data)
+{
+        nufa_args_t     *args = data;
+        xlator_t        *this = args->this;
+        char            *local_volname = args->volname;
+        gf_boolean_t    addr_match = args->addr_match;
+        char            *brick_host = NULL;
+        dht_conf_t      *conf = this->private;
+        int             ret = -1;
+
+        /*This means a local subvol was already found. We pick the first brick
+         * that is local*/
+        if (conf->private)
+                return;
+
+        if (strcmp (xl->name, local_volname) == 0) {
+                conf->private = xl;
+                gf_log (this->name, GF_LOG_INFO, "Using specified subvol %s",
+                        local_volname);
+                return;
+        }
+
+        if (!addr_match)
+                return;
+
+        ret = dict_get_str (xl->options, "remote-host", &brick_host);
+        if ((ret == 0) &&
+            (gf_is_same_address (local_volname, brick_host) ||
+             gf_is_local_addr (brick_host))) {
+                conf->private = xl;
+                gf_log (this->name, GF_LOG_INFO, "Using the first local "
+                        "subvol %s", xl->name);
+                return;
+        }
+
+}
+
+static void
+nufa_to_dht (xlator_t *this)
+{
+        GF_ASSERT (this);
+        GF_ASSERT (this->fops);
+
+        this->fops->lookup = dht_lookup;
+        this->fops->create = dht_create;
+        this->fops->mknod  = dht_mknod;
+}
+
+int
+nufa_find_local_subvol (xlator_t *this,
+                        void (*fn) (xlator_t *each, void* data), void *data)
+{
+        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;
+        }
+
+        candidate = conf->private;
+        trav = candidate->parents;
+        while (trav) {
+
+                parent = trav->xlator;
+                if (strcmp (parent->type, "cluster/nufa") == 0) {
+                        gf_log (this->name, GF_LOG_INFO, "Found local subvol, "
+                                "%s", candidate->name);
+                        ret = 0;
+                        conf->private = candidate;
+                        break;
+                }
+
+                candidate = parent;
+                trav = parent->parents;
+        }
+
+        return ret;
+}
+
 int
 nufa_init (xlator_t *this)
 {
-        dht_conf_t    *conf = NULL;
-        xlator_list_t *trav = NULL;
         data_t        *data = NULL;
         char          *local_volname = NULL;
         int            ret = -1;
         char           my_hostname[256];
-        xlator_t      *local_subvol = NULL;
-        char          *brick_host = NULL;
-        xlator_t      *kid = NULL;
+        gf_boolean_t   addr_match = _gf_false;
+        nufa_args_t    args = {0, };
 
         ret = dht_init(this);
         if (ret) {
                 return ret;
         }
-        conf = this->private;
 
-        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));
-        }
+        if ((data = dict_get (this->options, "local-volume-name"))) {
+                local_volname = data->data;
 
-        if (ret == 0)
-                local_volname = my_hostname;
+        } else {
+                addr_match = _gf_true;
+                local_volname = "localhost";
+                ret = gethostname (my_hostname, 256);
+                if (ret == 0)
+                        local_volname = my_hostname;
 
-        data = dict_get (this->options, "local-volume-name");
-        if (data) {
-                local_volname = data->data;
-        }
+                else
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "could not find hostname (%s)",
+                                strerror (errno));
 
-        for (trav = this->children; trav; trav = trav->next) {
-                if (strcmp (trav->xlator->name, local_volname) == 0)
-                        break;
-                if (local_subvol) {
-                        continue;
-                }
-                kid = trav->xlator;
-                for (;;) {
-                        if (dict_get_str(trav->xlator->options,"remote-host",
-                                         &brick_host) == 0) {
-                                /* Found it. */
-                                break;
-                        }
-                        if (!kid->children) {
-                                /* Nowhere further to look. */
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "could not get remote-host");
-                                goto err;
-                        }
-                        if (kid->children->next) {
-                                /* Multiple choices, can't/shouldn't decide. */
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "NUFA found fan-out (type %s) volume",
-                                        kid->type);
-                                goto err;
-                        }
-                        /* One-to-one xlators are OK, try the next one. */
-                        kid = kid->children->xlator;
-                }
-                if (same_first_part(my_hostname,'.',brick_host,'.')) {
-                        local_subvol = trav->xlator;
-                }
         }
 
-        if (trav) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Using specified subvol %s", local_volname);
-                conf->private = trav->xlator;
-        }
-        else if (local_subvol) {
+        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,
-                        "Using first local subvol %s", local_subvol->name);
-                conf->private = local_subvol;
+                        "Unable to find local subvolume, switching "
+                        "to dht mode");
+                nufa_to_dht (this);
         }
-        else {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not find specified or local subvol");
-                goto err;
-
-        }
-
         return 0;
-
-err:
-        dht_fini(this);
-        return -1;
 }
 
 
diff --git a/xlators/cluster/dht/src/switch.c b/xlators/cluster/dht/src/switch.c
index 861012247..d3ea90ba8 100644
--- a/xlators/cluster/dht/src/switch.c
+++ b/xlators/cluster/dht/src/switch.c
@@ -437,7 +437,8 @@ 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) {
@@ -536,7 +537,8 @@ switch_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
         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) {
diff --git a/xlators/cluster/stripe/src/stripe.c b/xlators/cluster/stripe/src/stripe.c
index dadd3fec5..69b510e23 100644
--- a/xlators/cluster/stripe/src/stripe.c
+++ b/xlators/cluster/stripe/src/stripe.c
@@ -3774,6 +3774,524 @@ err:
 
 
 int32_t
+stripe_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+        int32_t         callcnt = 0;
+        stripe_local_t *local = NULL;
+	stripe_local_t *mlocal = NULL;
+        call_frame_t   *prev = NULL;
+	call_frame_t   *mframe = NULL;
+
+        if (!this || !frame || !frame->local || !cookie) {
+                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
+                goto out;
+        }
+
+        prev  = cookie;
+        local = frame->local;
+	mframe = local->orig_frame;
+	mlocal = mframe->local;
+
+        LOCK(&frame->lock);
+        {
+                callcnt = ++mlocal->call_count;
+
+                if (op_ret == 0) {
+                        mlocal->post_buf = *postbuf;
+                        mlocal->pre_buf = *prebuf;
+
+			mlocal->prebuf_blocks  += prebuf->ia_blocks;
+			mlocal->postbuf_blocks += postbuf->ia_blocks;
+
+			correct_file_size(prebuf, mlocal->fctx, prev);
+			correct_file_size(postbuf, mlocal->fctx, prev);
+
+			if (mlocal->prebuf_size < prebuf->ia_size)
+				mlocal->prebuf_size = prebuf->ia_size;
+			if (mlocal->postbuf_size < postbuf->ia_size)
+				mlocal->postbuf_size = postbuf->ia_size;
+                }
+
+		/* return the first failure */
+		if (mlocal->op_ret == 0) {
+			mlocal->op_ret = op_ret;
+			mlocal->op_errno = op_errno;
+		}
+        }
+        UNLOCK (&frame->lock);
+
+        if ((callcnt == mlocal->wind_count) && mlocal->unwind) {
+		mlocal->pre_buf.ia_size = mlocal->prebuf_size;
+		mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks;
+		mlocal->post_buf.ia_size = mlocal->postbuf_size;
+		mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks;
+
+                STRIPE_STACK_UNWIND (fallocate, mframe, mlocal->op_ret,
+                                     mlocal->op_errno, &mlocal->pre_buf,
+                                     &mlocal->post_buf, NULL);
+        }
+out:
+	STRIPE_STACK_DESTROY(frame);
+        return 0;
+}
+
+int32_t
+stripe_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+		 off_t offset, size_t len, dict_t *xdata)
+{
+        stripe_local_t   *local = NULL;
+        stripe_fd_ctx_t  *fctx = NULL;
+        int32_t           op_errno = 1;
+        int32_t           idx = 0;
+        int32_t           offset_offset = 0;
+        int32_t           remaining_size = 0;
+        off_t             fill_size = 0;
+        uint64_t          stripe_size = 0;
+        uint64_t          tmp_fctx = 0;
+	off_t		  dest_offset = 0;
+	call_frame_t	  *fframe = NULL;
+	stripe_local_t	  *flocal = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (fd->inode, err);
+
+        inode_ctx_get (fd->inode, this, &tmp_fctx);
+        if (!tmp_fctx) {
+                op_errno = EINVAL;
+                goto err;
+        }
+        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
+        stripe_size = fctx->stripe_size;
+
+        STRIPE_VALIDATE_FCTX (fctx, err);
+
+        remaining_size = len;
+
+        local = mem_get0 (this->local_pool);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+        frame->local = local;
+        local->stripe_size = stripe_size;
+	local->fctx = fctx;
+
+        if (!stripe_size) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Wrong stripe size for the file");
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        while (1) {
+		fframe = copy_frame(frame);
+		flocal = mem_get0(this->local_pool);
+		if (!flocal) {
+			op_errno = ENOMEM;
+			goto err;
+		}
+		flocal->orig_frame = frame;
+		fframe->local = flocal;
+
+		/* send fallocate request to the associated child node */
+                idx = (((offset + offset_offset) /
+                        local->stripe_size) % fctx->stripe_count);
+
+                fill_size = (local->stripe_size -
+                             ((offset + offset_offset) % local->stripe_size));
+                if (fill_size > remaining_size)
+                        fill_size = remaining_size;
+
+                remaining_size -= fill_size;
+
+                local->wind_count++;
+                if (remaining_size == 0)
+                        local->unwind = 1;
+
+		dest_offset = offset + offset_offset;
+		if (fctx->stripe_coalesce)
+			dest_offset = coalesced_offset(dest_offset,
+					local->stripe_size, fctx->stripe_count);
+
+		/*
+		 * TODO: Create a separate handler for coalesce mode that sends a
+		 * single fallocate per-child (since the ranges are linear).
+		 */
+		STACK_WIND(fframe, stripe_fallocate_cbk, fctx->xl_array[idx],
+			   fctx->xl_array[idx]->fops->fallocate, fd, mode,
+			   dest_offset, fill_size, xdata);
+
+                offset_offset += fill_size;
+                if (remaining_size == 0)
+                        break;
+        }
+
+        return 0;
+err:
+	if (fframe)
+		STRIPE_STACK_DESTROY(fframe);
+
+        STRIPE_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
+
+int32_t
+stripe_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+        int32_t         callcnt = 0;
+        stripe_local_t *local = NULL;
+	stripe_local_t *mlocal = NULL;
+        call_frame_t   *prev = NULL;
+	call_frame_t   *mframe = NULL;
+
+        if (!this || !frame || !frame->local || !cookie) {
+                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
+                goto out;
+        }
+
+        prev  = cookie;
+        local = frame->local;
+	mframe = local->orig_frame;
+	mlocal = mframe->local;
+
+        LOCK(&frame->lock);
+        {
+                callcnt = ++mlocal->call_count;
+
+                if (op_ret == 0) {
+                        mlocal->post_buf = *postbuf;
+                        mlocal->pre_buf = *prebuf;
+
+			mlocal->prebuf_blocks  += prebuf->ia_blocks;
+			mlocal->postbuf_blocks += postbuf->ia_blocks;
+
+			correct_file_size(prebuf, mlocal->fctx, prev);
+			correct_file_size(postbuf, mlocal->fctx, prev);
+
+			if (mlocal->prebuf_size < prebuf->ia_size)
+				mlocal->prebuf_size = prebuf->ia_size;
+			if (mlocal->postbuf_size < postbuf->ia_size)
+				mlocal->postbuf_size = postbuf->ia_size;
+                }
+
+		/* return the first failure */
+		if (mlocal->op_ret == 0) {
+			mlocal->op_ret = op_ret;
+			mlocal->op_errno = op_errno;
+		}
+        }
+        UNLOCK (&frame->lock);
+
+        if ((callcnt == mlocal->wind_count) && mlocal->unwind) {
+		mlocal->pre_buf.ia_size = mlocal->prebuf_size;
+		mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks;
+		mlocal->post_buf.ia_size = mlocal->postbuf_size;
+		mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks;
+
+                STRIPE_STACK_UNWIND (discard, mframe, mlocal->op_ret,
+                                     mlocal->op_errno, &mlocal->pre_buf,
+                                     &mlocal->post_buf, NULL);
+        }
+out:
+	STRIPE_STACK_DESTROY(frame);
+        return 0;
+}
+
+int32_t
+stripe_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+	       size_t len, dict_t *xdata)
+{
+        stripe_local_t   *local = NULL;
+        stripe_fd_ctx_t  *fctx = NULL;
+        int32_t           op_errno = 1;
+        int32_t           idx = 0;
+        int32_t           offset_offset = 0;
+        int32_t           remaining_size = 0;
+        off_t             fill_size = 0;
+        uint64_t          stripe_size = 0;
+        uint64_t          tmp_fctx = 0;
+	off_t		  dest_offset = 0;
+	call_frame_t	  *fframe = NULL;
+	stripe_local_t	  *flocal = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (fd->inode, err);
+
+        inode_ctx_get (fd->inode, this, &tmp_fctx);
+        if (!tmp_fctx) {
+                op_errno = EINVAL;
+                goto err;
+        }
+        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
+        stripe_size = fctx->stripe_size;
+
+        STRIPE_VALIDATE_FCTX (fctx, err);
+
+        remaining_size = len;
+
+        local = mem_get0 (this->local_pool);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+        frame->local = local;
+        local->stripe_size = stripe_size;
+	local->fctx = fctx;
+
+        if (!stripe_size) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Wrong stripe size for the file");
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        while (1) {
+		fframe = copy_frame(frame);
+		flocal = mem_get0(this->local_pool);
+		if (!flocal) {
+			op_errno = ENOMEM;
+			goto err;
+		}
+		flocal->orig_frame = frame;
+		fframe->local = flocal;
+
+		/* send discard request to the associated child node */
+                idx = (((offset + offset_offset) /
+                        local->stripe_size) % fctx->stripe_count);
+
+                fill_size = (local->stripe_size -
+                             ((offset + offset_offset) % local->stripe_size));
+                if (fill_size > remaining_size)
+                        fill_size = remaining_size;
+
+                remaining_size -= fill_size;
+
+                local->wind_count++;
+                if (remaining_size == 0)
+                        local->unwind = 1;
+
+		dest_offset = offset + offset_offset;
+		if (fctx->stripe_coalesce)
+			dest_offset = coalesced_offset(dest_offset,
+					local->stripe_size, fctx->stripe_count);
+
+		/*
+		 * TODO: Create a separate handler for coalesce mode that sends a
+		 * single discard per-child (since the ranges are linear).
+		 */
+		STACK_WIND(fframe, stripe_discard_cbk, fctx->xl_array[idx],
+			   fctx->xl_array[idx]->fops->discard, fd, dest_offset,
+			   fill_size, xdata);
+
+                offset_offset += fill_size;
+                if (remaining_size == 0)
+                        break;
+        }
+
+        return 0;
+err:
+	if (fframe)
+		STRIPE_STACK_DESTROY(fframe);
+
+        STRIPE_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
+int32_t
+stripe_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;
+
+        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 (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,
+               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;
+        }
+        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;
+        }
+        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;
@@ -3947,6 +4465,37 @@ stripe_setxattr_cbk (call_frame_t *frame, void *cookie,
         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;
+}
+
+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)
@@ -3956,6 +4505,7 @@ stripe_setxattr (call_frame_t *frame, xlator_t *this,
         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);
@@ -3978,11 +4528,15 @@ stripe_setxattr (call_frame_t *frame, xlator_t *this,
         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.
+         * 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)) {
+        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,
@@ -4013,21 +4567,21 @@ stripe_fsetxattr_cbk (call_frame_t *frame, void *cookie,
 
 
 int
-stripe_is_lockinfo (dict_t *this,
-                    char *key,
-                    data_t *value,
-                    void *data)
+stripe_is_special_key (dict_t *this,
+                       char *key,
+                       data_t *value,
+                       void *data)
 {
-        gf_boolean_t *is_lockinfo = NULL;
+        gf_boolean_t *is_special = NULL;
 
         if (data == NULL) {
                 goto out;
         }
 
-        is_lockinfo = data;
+        is_special = data;
 
-        if (XATTR_IS_LOCKINFO (key))
-                *is_lockinfo = _gf_true;
+        if (XATTR_IS_LOCKINFO (key) || XATTR_IS_BD (key))
+                *is_special = _gf_true;
 
 out:
         return 0;
@@ -4104,7 +4658,7 @@ stripe_fsetxattr_is_special (dict_t *dict)
                 goto out;
         }
 
-        dict_foreach (dict, stripe_is_lockinfo, &is_spl);
+        dict_foreach (dict, stripe_is_special_key, &is_spl);
 
 out:
         return is_spl;
@@ -4866,10 +5420,10 @@ stripe_vgetxattr_cbk (call_frame_t *frame, void *cookie,
 
                         xattr->pos = cky;
                         xattr->xattr_value = gf_memdup (xattr_val,
-                                                        xattr->xattr_value,
                                                         xattr->xattr_len);
 
-                        local->xattr_total_len += xattr->xattr_len + 1;
+                        if (xattr->xattr_value != NULL)
+                                local->xattr_total_len += xattr->xattr_len + 1;
                 }
         }
  out:
@@ -4969,7 +5523,8 @@ 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;
                 }
@@ -5048,6 +5603,7 @@ stripe_getxattr (call_frame_t *frame, xlator_t *this,
                                                    sub_volumes,
                                                    local->marker.call_count,
                                                    MARKER_XTIME_TYPE,
+                                                   marker_xtime_default_gauge,
                                                    priv->vol_uuid)) {
                                 op_errno = EINVAL;
                                 goto err;
@@ -5221,6 +5777,9 @@ struct xlator_fops fops = {
         .removexattr    = stripe_removexattr,
         .fremovexattr   = stripe_fremovexattr,
         .readdirp       = stripe_readdirp,
+	.fallocate	= stripe_fallocate,
+	.discard	= stripe_discard,
+        .zerofill       = stripe_zerofill,
 };
 
 struct xlator_cbks cbks = {
@@ -5246,10 +5805,10 @@ struct volume_options options[] = {
         },
 	{ .key = {"coalesce"},
 	  .type = GF_OPTION_TYPE_BOOL,
-	  .default_value = "false",
-	  .description = "Enable coalesce mode to flatten striped files as "
-			 "stored on the server (i.e., eliminate holes caused "
-			 "by the traditional format)."
+	  .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/debug/io-stats/src/io-stats.c b/xlators/debug/io-stats/src/io-stats.c
index 63bb8fa90..7fb697ae4 100644
--- a/xlators/debug/io-stats/src/io-stats.c
+++ b/xlators/debug/io-stats/src/io-stats.c
@@ -1724,6 +1724,40 @@ io_stats_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 
 int
+io_stats_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+		       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+		       struct iatt *postbuf, dict_t *xdata)
+{
+	UPDATE_PROFILE_STATS(frame, FALLOCATE);
+	STACK_UNWIND_STRICT(fallocate, frame, op_ret, op_errno, prebuf, postbuf,
+			    xdata);
+	return 0;
+}
+
+
+int
+io_stats_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+		     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+		     struct iatt *postbuf, dict_t *xdata)
+{
+	UPDATE_PROFILE_STATS(frame, DISCARD);
+	STACK_UNWIND_STRICT(discard, frame, op_ret, op_errno, prebuf, postbuf,
+			    xdata);
+	return 0;
+}
+
+int
+io_stats_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+        UPDATE_PROFILE_STATS(frame, ZEROFILL);
+        STACK_UNWIND_STRICT(zerofill, frame, op_ret, op_errno, prebuf, postbuf,
+                            xdata);
+        return 0;
+}
+
+int
 io_stats_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                  int32_t op_ret, int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
 {
@@ -2392,6 +2426,45 @@ io_stats_fstat (call_frame_t *frame, xlator_t *this,
 
 
 int
+io_stats_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+		   off_t offset, size_t len, dict_t *xdata)
+{
+	START_FOP_LATENCY(frame);
+
+	STACK_WIND(frame, io_stats_fallocate_cbk, FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+		   xdata);
+
+	return 0;
+}
+
+
+int
+io_stats_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+		 size_t len, dict_t *xdata)
+{
+	START_FOP_LATENCY(frame);
+
+	STACK_WIND(frame, io_stats_discard_cbk, FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+
+	return 0;
+}
+
+int
+io_stats_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                 size_t len, dict_t *xdata)
+{
+        START_FOP_LATENCY(frame);
+
+        STACK_WIND(frame, io_stats_zerofill_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+
+        return 0;
+}
+
+
+int
 io_stats_lk (call_frame_t *frame, xlator_t *this,
              fd_t *fd, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
 {
@@ -2817,6 +2890,9 @@ struct xlator_fops fops = {
         .fxattrop    = io_stats_fxattrop,
         .setattr     = io_stats_setattr,
         .fsetattr    = io_stats_fsetattr,
+	.fallocate   = io_stats_fallocate,
+	.discard     = io_stats_discard,
+        .zerofill    = io_stats_zerofill,
 };
 
 struct xlator_cbks cbks = {
diff --git a/xlators/debug/trace/src/trace.c b/xlators/debug/trace/src/trace.c
index a9c11babe..c9d839356 100644
--- a/xlators/debug/trace/src/trace.c
+++ b/xlators/debug/trace/src/trace.c
@@ -3134,7 +3134,7 @@ init (xlator_t *this)
         gf_log (this->name, GF_LOG_DEBUG, "logging to history %s",
                 (conf->log_history == _gf_true)?"enabled":"disabled");
 
-        history = eh_new (history_size, _gf_false);
+        history = eh_new (history_size, _gf_false, NULL);
         if (!history) {
                 gf_log (this->name, GF_LOG_ERROR, "event history cannot be "
                         "initialized");
diff --git a/xlators/encryption/Makefile.am b/xlators/encryption/Makefile.am
index 2cbde680f..36efc6698 100644
--- a/xlators/encryption/Makefile.am
+++ b/xlators/encryption/Makefile.am
@@ -1,3 +1,3 @@
-SUBDIRS = rot-13
+SUBDIRS = rot-13 crypt
 
 CLEANFILES = 
diff --git a/xlators/encryption/crypt/Makefile.am b/xlators/encryption/crypt/Makefile.am
new file mode 100644
index 000000000..d471a3f92
--- /dev/null
+++ b/xlators/encryption/crypt/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES = 
diff --git a/xlators/encryption/crypt/src/Makefile.am b/xlators/encryption/crypt/src/Makefile.am
new file mode 100644
index 000000000..faadd117f
--- /dev/null
+++ b/xlators/encryption/crypt/src/Makefile.am
@@ -0,0 +1,24 @@
+if ENABLE_CRYPT_XLATOR
+
+xlator_LTLIBRARIES = crypt.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/encryption
+
+crypt_la_LDFLAGS = -module -avoidversion -lssl -lcrypto
+
+crypt_la_SOURCES = keys.c data.c metadata.c atom.c crypt.c
+crypt_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = crypt-common.h crypt-mem-types.h crypt.h metadata.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
+
+else
+
+noinst_DIST = keys.c data.c metadata.c atom.c crypt.c
+noinst_HEADERS = crypt-common.h crypt-mem-types.h crypt.h metadata.h
+
+endif
\ No newline at end of file
diff --git a/xlators/encryption/crypt/src/atom.c b/xlators/encryption/crypt/src/atom.c
new file mode 100644
index 000000000..1ec41495c
--- /dev/null
+++ b/xlators/encryption/crypt/src/atom.c
@@ -0,0 +1,962 @@
+/*
+  Copyright (c) 2008-2013 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 "defaults.h"
+#include "crypt-common.h"
+#include "crypt.h"
+
+/*
+ *                              Glossary
+ *
+ *
+ * cblock            (or cipher block). A logical unit in a file.
+ *                   cblock size is defined as the number of bits
+ *                   in an input (or output) block of the block
+ *                   cipher (*). Cipher block size is a property of
+ *                   cipher algorithm. E.g. cblock size is 64 bits
+ *                   for DES, 128 bits for AES, etc.
+ *
+ * atomic cipher     A cipher algorithm, which requires some chunks of
+ * algorithm         text to be padded at left and(or) right sides before
+ *                   cipher transaform.
+ *
+ *
+ * block (atom)      Minimal chunk of file's data, which doesn't require
+ *                   padding. We'll consider logical units in a file of
+ *                   block size (atom size).
+ *
+ * cipher algorithm  Atomic cipher algorithm, which requires the last
+ * with EOF issue    incomplete cblock in a file to be padded with some
+ *                   data (usually zeros).
+ *
+ *
+ * operation, which  reading/writing from offset, which is not aligned to
+ * forms a gap at    to atom size
+ * the beginning
+ *
+ *
+ * operation, which  reading/writing count bytes starting from offset off,
+ * forms a gap at    so that off+count is not aligned to atom_size
+ * the end
+ *
+ * head block        the first atom affected by an operation, which forms
+ *                   a gap at the beginning, or(and) at the end.
+ *                   Сomment. Head block has at least one gap (either at
+ *                   the beginning, or at the end)
+ *
+ *
+ * tail block        the last atom different from head, affected by an
+ *                   operation, which forms a gap at the end.
+ *                   Сomment: Tail block has exactly one gap (at the end).
+ *
+ *
+ * partial block     head or tail block
+ *
+ *
+ * full block        block without gaps.
+ *
+ *
+ * (*) Recommendation for Block Cipher Modes of Operation
+ *     Methods and Techniques
+ *     NIST Special Publication 800-38A Edition 2001
+ */
+
+/*
+ * atom->offset_at()
+ */
+static off_t offset_at_head(struct avec_config *conf)
+{
+	return conf->aligned_offset;
+}
+
+static off_t offset_at_hole_head(call_frame_t *frame,
+					 struct object_cipher_info *object)
+{
+	return offset_at_head(get_hole_conf(frame));
+}
+
+static off_t offset_at_data_head(call_frame_t *frame,
+					 struct object_cipher_info *object)
+{
+	return offset_at_head(get_data_conf(frame));
+}
+
+
+static off_t offset_at_tail(struct avec_config *conf,
+				    struct object_cipher_info *object)
+{
+	return conf->aligned_offset +
+		(conf->off_in_head ? get_atom_size(object) : 0) +
+		(conf->nr_full_blocks << get_atom_bits(object));
+}
+
+static off_t offset_at_hole_tail(call_frame_t *frame,
+					 struct object_cipher_info *object)
+{
+	return offset_at_tail(get_hole_conf(frame), object);
+}
+
+
+static off_t offset_at_data_tail(call_frame_t *frame,
+					 struct object_cipher_info *object)
+{
+	return offset_at_tail(get_data_conf(frame), object);
+}
+
+static off_t offset_at_full(struct avec_config *conf,
+				    struct object_cipher_info *object)
+{
+	return conf->aligned_offset +
+		(conf->off_in_head ? get_atom_size(object) : 0);
+}
+
+static off_t offset_at_data_full(call_frame_t *frame,
+					 struct object_cipher_info *object)
+{
+	return offset_at_full(get_data_conf(frame), object);
+}
+
+static off_t offset_at_hole_full(call_frame_t *frame,
+					 struct object_cipher_info *object)
+{
+	return offset_at_full(get_hole_conf(frame), object);
+}
+
+/*
+ * atom->io_size_nopad()
+ */
+
+static uint32_t io_size_nopad_head(struct avec_config *conf,
+				   struct object_cipher_info *object)
+{
+	uint32_t gap_at_beg;
+	uint32_t gap_at_end;
+
+	check_head_block(conf);
+
+	gap_at_beg = conf->off_in_head;
+
+	if (has_tail_block(conf) || has_full_blocks(conf) || conf->off_in_tail == 0 )
+		gap_at_end = 0;
+	else
+		gap_at_end = get_atom_size(object) - conf->off_in_tail;
+
+	return get_atom_size(object) - (gap_at_beg + gap_at_end);
+}
+
+static uint32_t io_size_nopad_tail(struct avec_config *conf,
+				   struct object_cipher_info *object)
+{
+	check_tail_block(conf);
+	return conf->off_in_tail;
+}
+
+static uint32_t io_size_nopad_full(struct avec_config *conf,
+				   struct object_cipher_info *object)
+{	
+	check_full_block(conf);
+	return get_atom_size(object);
+}
+
+static uint32_t io_size_nopad_data_head(call_frame_t *frame,
+					struct object_cipher_info *object)
+{
+	return io_size_nopad_head(get_data_conf(frame), object);
+}
+
+static uint32_t io_size_nopad_hole_head(call_frame_t *frame,
+					struct object_cipher_info *object)
+{
+	return io_size_nopad_head(get_hole_conf(frame), object);
+}
+
+static uint32_t io_size_nopad_data_tail(call_frame_t *frame,
+					struct object_cipher_info *object)
+{
+	return io_size_nopad_tail(get_data_conf(frame), object);
+}
+
+static uint32_t io_size_nopad_hole_tail(call_frame_t *frame,
+					struct object_cipher_info *object)
+{
+	return io_size_nopad_tail(get_hole_conf(frame), object);
+}
+
+static uint32_t io_size_nopad_data_full(call_frame_t *frame,
+					struct object_cipher_info *object)
+{
+	return io_size_nopad_full(get_data_conf(frame), object);
+}
+
+static uint32_t io_size_nopad_hole_full(call_frame_t *frame,
+					struct object_cipher_info *object)
+{
+	return io_size_nopad_full(get_hole_conf(frame), object);
+}
+
+static uint32_t offset_in_head(struct avec_config *conf)
+{
+	check_cursor_head(conf);
+
+	return conf->off_in_head;
+}
+
+static uint32_t offset_in_tail(call_frame_t *frame,
+			       struct object_cipher_info *object)
+{
+	return 0;
+}
+
+static uint32_t offset_in_full(struct avec_config *conf,
+			       struct object_cipher_info *object)
+{
+	check_cursor_full(conf);
+
+	if (has_head_block(conf))
+		return (conf->cursor - 1) << get_atom_bits(object);
+	else
+		return conf->cursor << get_atom_bits(object);
+}
+
+static uint32_t offset_in_data_head(call_frame_t *frame,
+				    struct object_cipher_info *object)
+{
+	return offset_in_head(get_data_conf(frame));
+}
+
+static uint32_t offset_in_hole_head(call_frame_t *frame,
+				    struct object_cipher_info *object)
+{
+	return offset_in_head(get_hole_conf(frame));
+}
+
+static uint32_t offset_in_data_full(call_frame_t *frame,
+				    struct object_cipher_info *object)
+{
+	return offset_in_full(get_data_conf(frame), object);
+}
+
+static uint32_t offset_in_hole_full(call_frame_t *frame,
+				    struct object_cipher_info *object)
+{
+	return offset_in_full(get_hole_conf(frame), object);
+}
+
+/*
+ * atom->rmw()
+ */
+/*
+ * Pre-conditions:
+ * @vec contains plain text of the latest
+ * version.
+ *
+ * Uptodate gaps of the @partial block with
+ * this plain text, encrypt the whole block
+ * and write the result to disk.
+ */
+static int32_t rmw_partial_block(call_frame_t *frame,
+				 void *cookie,
+				 xlator_t *this,
+				 int32_t op_ret,
+				 int32_t op_errno,
+				 struct iovec *vec,
+				 int32_t count,
+				 struct iatt *stbuf,
+				 struct iobref *iobref,
+				 struct rmw_atom *atom)
+{
+	size_t was_read = 0;
+	uint64_t file_size;
+	crypt_local_t *local = frame->local;
+	struct object_cipher_info *object = &local->info->cinfo;
+
+	struct iovec *partial = atom->get_iovec(frame, 0);
+	struct avec_config *conf = atom->get_config(frame);
+	end_writeback_handler_t end_writeback_partial_block;
+#if DEBUG_CRYPT
+	gf_boolean_t check_last_cblock = _gf_false;
+#endif
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (op_ret < 0)
+		goto exit;
+
+	file_size = local->cur_file_size;
+	was_read = op_ret;
+
+	if (atom->locality == HEAD_ATOM && conf->off_in_head) {
+		/*
+		 * head atom with a non-uptodate gap
+		 * at the beginning
+		 *
+		 * fill the gap with plain text of the
+		 * latest version. Convert a part of hole
+		 * (if any) to zeros.
+		 */
+		int32_t i;
+		int32_t copied = 0;
+		int32_t to_gap; /* amount of data needed to uptodate
+				   the gap at the beginning */
+#if 0
+		int32_t hole = 0; /* The part of the hole which
+				   * got in the head block */
+#endif /* 0 */
+		to_gap = conf->off_in_head;
+
+		if (was_read < to_gap) {
+			if (file_size >
+			    offset_at_head(conf) + was_read) {
+				/*
+				 * It is impossible to uptodate
+				 * head block: too few bytes have
+				 * been read from disk, so that
+				 * partial write is impossible.
+				 *
+				 * It could happen because of many
+				 * reasons: IO errors, (meta)data
+				 * corruption in the local file system,
+				 * etc.
+				 */
+				gf_log(this->name, GF_LOG_WARNING,
+				     "Can not uptodate a gap at the beginning");
+				local->op_ret = -1;
+				local->op_errno = EIO;
+				goto exit;
+			}
+#if 0
+			hole = to_gap - was_read;
+#endif /* 0 */
+			to_gap = was_read;
+		}
+		/*
+		 * uptodate the gap at the beginning
+		 */
+		for (i = 0; i < count && copied < to_gap; i++) {
+			int32_t to_copy;
+
+			to_copy = vec[i].iov_len;
+			if (to_copy > to_gap - copied)
+				to_copy = to_gap - copied;
+
+			memcpy(partial->iov_base, vec[i].iov_base, to_copy);
+			copied += to_copy;
+		}
+#if 0
+		/*
+		 * If possible, convert part of the
+		 * hole, which got in the head block
+		 */
+		ret = TRY_LOCK(&local->hole_lock);
+		if (!ret) {
+			if (local->hole_handled)
+				/*
+				 * already converted by
+				 * crypt_writev_cbk()
+				 */
+				UNLOCK(&local->hole_lock);
+			else {
+				/*
+				 * convert the part of the hole
+				 * which got in the head block
+				 * to zeros.
+				 *
+				 * Update the orig_offset to make
+				 * sure writev_cbk() won't care
+				 * about this part of the hole.
+				 *
+				 */
+				memset(partial->iov_base + to_gap, 0, hole);
+
+				conf->orig_offset -= hole;
+				conf->orig_size += hole;
+				UNLOCK(&local->hole_lock);
+			}
+		}
+		else /*
+		      * conversion is being performed
+		      * by crypt_writev_cbk()
+		      */
+			;
+#endif /* 0 */
+	}
+	if (atom->locality == TAIL_ATOM ||
+	    (!has_tail_block(conf) && conf->off_in_tail)) {
+		/*
+		 * tail atom, or head atom with a non-uptodate
+		 * gap at the end.
+		 *
+		 * fill the gap at the end of the block
+		 * with plain text of the latest version.
+		 * Pad the result, (if needed)
+		 */
+		int32_t i;
+		int32_t to_gap;
+		int copied;
+		off_t off_in_tail;
+		int32_t to_copy;
+
+		off_in_tail = conf->off_in_tail;
+		to_gap = conf->gap_in_tail;
+
+		if (to_gap && was_read < off_in_tail + to_gap) {
+			/*
+			 * It is impossible to uptodate
+			 * the gap at the end: too few bytes
+			 * have been read from disk, so that
+			 * partial write is impossible.
+			 *
+			 * It could happen because of many
+			 * reasons: IO errors, (meta)data
+			 * corruption in the local file system,
+			 * etc.
+			 */
+			gf_log(this->name, GF_LOG_WARNING,
+			       "Can not uptodate a gap at the end");
+			local->op_ret = -1;
+			local->op_errno = EIO;
+			goto exit;
+		}
+		/*
+		 * uptodate the gap at the end
+		 */
+		copied = 0;
+		to_copy = to_gap;
+		for(i = count - 1; i >= 0 && to_copy > 0; i--) {
+			uint32_t from_vec, off_in_vec;
+
+			off_in_vec = 0;
+			from_vec = vec[i].iov_len;
+			if (from_vec > to_copy) {
+				off_in_vec = from_vec - to_copy;
+				from_vec = to_copy;
+			}
+			memcpy(partial->iov_base +
+			       off_in_tail + to_gap - copied - from_vec,
+			       vec[i].iov_base + off_in_vec,
+			       from_vec);
+
+			gf_log(this->name, GF_LOG_DEBUG,
+			       "uptodate %d bytes at tail. Offset at target(source): %d(%d)",
+			       (int)from_vec,
+			       (int)off_in_tail + to_gap - copied - from_vec,
+			       (int)off_in_vec);
+
+			copied += from_vec;
+			to_copy -= from_vec;
+		}
+		partial->iov_len = off_in_tail + to_gap;
+
+		if (object_alg_should_pad(object)) {
+			int32_t resid = 0;
+			resid = partial->iov_len & (object_alg_blksize(object) - 1);
+			if (resid) {
+				/*
+				 * append a new EOF padding
+				 */
+				local->eof_padding_size =
+					object_alg_blksize(object) - resid;
+
+				gf_log(this->name, GF_LOG_DEBUG,
+				       "set padding size %d",
+				       local->eof_padding_size);
+
+				memset(partial->iov_base + partial->iov_len,
+				       1,
+				       local->eof_padding_size);
+				partial->iov_len += local->eof_padding_size;
+#if DEBUG_CRYPT
+				gf_log(this->name, GF_LOG_DEBUG,
+				       "pad cblock with %d zeros:",
+				       local->eof_padding_size);
+				dump_cblock(this,
+					    (unsigned char *)partial->iov_base +
+					    partial->iov_len - object_alg_blksize(object));
+				check_last_cblock = _gf_true;
+#endif
+			}
+		}
+	}
+	/*
+	 * encrypt the whole block
+	 */
+	encrypt_aligned_iov(object,
+			    partial,
+			    1,
+			    atom->offset_at(frame, object));
+#if DEBUG_CRYPT
+	if (check_last_cblock == _gf_true) {
+		gf_log(this->name, GF_LOG_DEBUG,
+		       "encrypt last cblock with offset %llu",
+		       (unsigned long long)atom->offset_at(frame, object));
+		dump_cblock(this, (unsigned char *)partial->iov_base +
+			    partial->iov_len - object_alg_blksize(object));
+	}
+#endif
+	set_local_io_params_writev(frame, object, atom,
+				   atom->offset_at(frame, object),
+				   iovec_get_size(partial, 1));
+	/*
+	 * write the whole block to disk
+	 */
+	end_writeback_partial_block = dispatch_end_writeback(local->fop);
+	conf->cursor ++;
+	STACK_WIND(frame,
+		   end_writeback_partial_block,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->writev,
+		   local->fd,
+		   partial,
+		   1,
+		   atom->offset_at(frame, object),
+		   local->flags,
+		   local->iobref_data,
+		   local->xdata);
+
+	gf_log("crypt", GF_LOG_DEBUG,
+	       "submit partial block: %d bytes from %d offset",
+	       (int)iovec_get_size(partial, 1),
+	       (int)atom->offset_at(frame, object));
+ exit:
+	return 0;
+}
+
+/*
+ * Perform a (read-)modify-write sequence.
+ * This should be performed only after approval
+ * of upper server-side manager, i.e. the caller
+ * needs to make sure this is his turn to rmw.
+ */
+void submit_partial(call_frame_t *frame,
+		    xlator_t *this,
+		    fd_t *fd,
+		    atom_locality_type ltype)
+{
+	int32_t ret;
+	dict_t *dict;
+	struct rmw_atom *atom;
+	crypt_local_t *local = frame->local;
+	struct object_cipher_info *object = &local->info->cinfo;
+
+	atom = atom_by_types(local->active_setup, ltype);
+	/*
+	 * To perform the "read" component of the read-modify-write
+	 * sequence the crypt translator does stack_wind to itself.
+	 * 
+	 * Pass current file size to crypt_readv()
+	 */
+	dict = dict_new();
+	if (!dict) {
+		/*
+		 * FIXME: Handle the error
+		 */
+		gf_log("crypt", GF_LOG_WARNING, "Can not alloc dict");
+		return;
+	}
+	ret = dict_set(dict,
+		       FSIZE_XATTR_PREFIX,
+		       data_from_uint64(local->cur_file_size));
+	if (ret) {
+		/*
+		 * FIXME: Handle the error
+		 */
+		dict_unref(dict);
+		gf_log("crypt", GF_LOG_WARNING, "Can not set dict");
+		goto exit;
+	}
+	STACK_WIND(frame,
+		   atom->rmw,
+		   this,
+		   this->fops->readv, /* crypt_readv */
+		   fd,
+		   atom->count_to_uptodate(frame, object), /* count */
+		   atom->offset_at(frame, object), /* offset to read from */
+		   0,
+		   dict);
+ exit:
+	dict_unref(dict);
+}
+
+/*
+ * submit blocks of FULL_ATOM type
+ */
+void submit_full(call_frame_t *frame, xlator_t *this)
+{
+	crypt_local_t *local = frame->local;
+	struct object_cipher_info *object = &local->info->cinfo;
+	struct rmw_atom *atom = atom_by_types(local->active_setup, FULL_ATOM);
+	uint32_t count; /* total number of full blocks to submit */
+	uint32_t granularity; /* number of blocks to submit in one iteration */
+
+	uint64_t off_in_file; /* start offset in the file, bytes */
+	uint32_t off_in_atom; /* start offset in the atom, blocks */
+	uint32_t blocks_written = 0; /* blocks written for this submit */
+
+	struct avec_config *conf = atom->get_config(frame);
+	end_writeback_handler_t end_writeback_full_block;
+	/*
+	 * Write full blocks by groups of granularity size.
+	 */
+	end_writeback_full_block = dispatch_end_writeback(local->fop);
+
+	if (is_ordered_mode(frame)) {
+		uint32_t skip = has_head_block(conf) ? 1 : 0;
+		count = 1;
+		granularity = 1;
+		/*
+		 * calculate start offset using cursor value;
+		 * here we should take into accout head block,
+		 * which corresponds to cursor value 0.
+		 */
+		off_in_file = atom->offset_at(frame, object) +
+			((conf->cursor - skip) << get_atom_bits(object));
+		off_in_atom = conf->cursor - skip;
+	}
+	else {
+		/*
+		 * in parallel mode
+		 */
+		count = conf->nr_full_blocks;
+		granularity = MAX_IOVEC;
+		off_in_file = atom->offset_at(frame, object);
+		off_in_atom = 0;
+	}
+	while (count) {
+		uint32_t blocks_to_write = count;
+
+		if (blocks_to_write > granularity)
+			blocks_to_write = granularity;
+		if (conf->type == HOLE_ATOM)
+			/*
+			 * reset iovec before encryption
+			 */
+			memset(atom->get_iovec(frame, 0)->iov_base,
+			       0,
+			       get_atom_size(object));
+		/*
+		 * encrypt the group
+		 */
+		encrypt_aligned_iov(object,
+				    atom->get_iovec(frame,
+						    off_in_atom +
+						    blocks_written),
+				    blocks_to_write,
+				    off_in_file + (blocks_written <<
+						   get_atom_bits(object)));
+		
+		set_local_io_params_writev(frame, object, atom,
+		        off_in_file + (blocks_written << get_atom_bits(object)),
+			blocks_to_write <<  get_atom_bits(object));
+
+		conf->cursor += blocks_to_write;
+
+		STACK_WIND(frame,
+			   end_writeback_full_block,
+			   FIRST_CHILD(this),
+			   FIRST_CHILD(this)->fops->writev,
+			   local->fd,
+			   atom->get_iovec(frame, off_in_atom + blocks_written),
+			   blocks_to_write,
+			   off_in_file + (blocks_written << get_atom_bits(object)),
+			   local->flags,
+			   local->iobref_data ? local->iobref_data : local->iobref,
+			   local->xdata);
+
+		gf_log("crypt", GF_LOG_DEBUG, "submit %d full blocks from %d offset",
+		       blocks_to_write,
+		       (int)(off_in_file + (blocks_written << get_atom_bits(object))));
+
+		count -= blocks_to_write;
+		blocks_written += blocks_to_write;
+	}
+	return;
+}
+
+static int32_t rmw_data_head(call_frame_t *frame,
+			     void *cookie,
+			     xlator_t *this,
+			     int32_t op_ret,
+			     int32_t op_errno,
+			     struct iovec *vec,
+			     int32_t count,
+			     struct iatt *stbuf,
+			     struct iobref *iobref,
+			     dict_t *xdata)
+{
+	return rmw_partial_block(frame,
+				 cookie,
+				 this,
+				 op_ret,
+				 op_errno,
+				 vec,
+				 count,
+				 stbuf,
+				 iobref,
+				 atom_by_types(DATA_ATOM, HEAD_ATOM));
+}
+
+static int32_t rmw_data_tail(call_frame_t *frame,
+			     void *cookie,
+			     xlator_t *this,
+			     int32_t op_ret,
+			     int32_t op_errno,
+			     struct iovec *vec,
+			     int32_t count,
+			     struct iatt *stbuf,
+			     struct iobref *iobref,
+			     dict_t *xdata)
+{
+	return rmw_partial_block(frame,
+				 cookie,
+				 this,
+				 op_ret,
+				 op_errno,
+				 vec,
+				 count,
+				 stbuf,
+				 iobref,
+				 atom_by_types(DATA_ATOM, TAIL_ATOM));
+}
+
+static int32_t rmw_hole_head(call_frame_t *frame,
+			     void *cookie,
+			     xlator_t *this,
+			     int32_t op_ret,
+			     int32_t op_errno,
+			     struct iovec *vec,
+			     int32_t count,
+			     struct iatt *stbuf,
+			     struct iobref *iobref,
+			     dict_t *xdata)
+{
+	return rmw_partial_block(frame,
+				 cookie,
+				 this,
+				 op_ret,
+				 op_errno,
+				 vec,
+				 count,
+				 stbuf,
+				 iobref,
+				 atom_by_types(HOLE_ATOM, HEAD_ATOM));
+}
+
+static int32_t rmw_hole_tail(call_frame_t *frame,
+			     void *cookie,
+			     xlator_t *this,
+			     int32_t op_ret,
+			     int32_t op_errno,
+			     struct iovec *vec,
+			     int32_t count,
+			     struct iatt *stbuf,
+			     struct iobref *iobref,
+			     dict_t *xdata)
+{
+	return rmw_partial_block(frame,
+				 cookie,
+				 this,
+				 op_ret,
+				 op_errno,
+				 vec,
+				 count,
+				 stbuf,
+				 iobref,
+				 atom_by_types(HOLE_ATOM, TAIL_ATOM));
+}
+
+/*
+ * atom->count_to_uptodate()
+ */
+static uint32_t count_to_uptodate_head(struct avec_config *conf,
+				       struct object_cipher_info *object)
+{
+	if (conf->acount == 1 && conf->off_in_tail)
+		return get_atom_size(object);
+	else
+		/* there is no need to read the whole head block */
+		return conf->off_in_head;
+}
+
+static uint32_t count_to_uptodate_tail(struct avec_config *conf,
+				       struct object_cipher_info *object)
+{
+	/* we need to read the whole tail block */
+	return get_atom_size(object);
+}
+
+static uint32_t count_to_uptodate_data_head(call_frame_t *frame,
+					    struct object_cipher_info *object)
+{
+	return count_to_uptodate_head(get_data_conf(frame), object);
+}
+
+static uint32_t count_to_uptodate_data_tail(call_frame_t *frame,
+					    struct object_cipher_info *object)
+{
+	return count_to_uptodate_tail(get_data_conf(frame), object);
+}
+
+static uint32_t count_to_uptodate_hole_head(call_frame_t *frame,
+					    struct object_cipher_info *object)
+{
+	return count_to_uptodate_head(get_hole_conf(frame), object);
+}
+
+static uint32_t count_to_uptodate_hole_tail(call_frame_t *frame,
+					    struct object_cipher_info *object)
+{
+	return count_to_uptodate_tail(get_hole_conf(frame), object);
+}
+
+/* atom->get_config() */
+
+static struct avec_config *get_config_data(call_frame_t *frame)
+{
+	return &((crypt_local_t *)frame->local)->data_conf;
+}
+
+static struct avec_config *get_config_hole(call_frame_t *frame)
+{
+	return &((crypt_local_t *)frame->local)->hole_conf;
+}
+
+/*
+ * atom->get_iovec()
+ */
+static struct iovec *get_iovec_hole_head(call_frame_t *frame,
+					 uint32_t count)
+{
+	struct avec_config *conf = get_hole_conf(frame);
+
+	return conf->avec;
+}
+
+static struct iovec *get_iovec_hole_full(call_frame_t *frame,
+					 uint32_t count)
+{
+	struct avec_config *conf = get_hole_conf(frame);
+
+	return conf->avec + (conf->off_in_head ? 1 : 0);
+}
+
+static inline struct iovec *get_iovec_hole_tail(call_frame_t *frame,
+						uint32_t count)
+{
+	struct avec_config *conf = get_hole_conf(frame);
+
+	return conf->avec + (conf->blocks_in_pool - 1);
+}
+
+static inline struct iovec *get_iovec_data_head(call_frame_t *frame,
+						uint32_t count)
+{
+	struct avec_config *conf = get_data_conf(frame);
+
+	return conf->avec;
+}
+
+static inline struct iovec *get_iovec_data_full(call_frame_t *frame,
+						uint32_t count)
+{
+	struct avec_config *conf = get_data_conf(frame);
+
+	return conf->avec + (conf->off_in_head ? 1 : 0) + count;
+}
+
+static inline struct iovec *get_iovec_data_tail(call_frame_t *frame,
+						uint32_t count)
+{
+	struct avec_config *conf = get_data_conf(frame);
+
+	return conf->avec +
+		(conf->off_in_head ? 1 : 0) +
+		conf->nr_full_blocks;
+}
+
+static struct rmw_atom atoms[LAST_DATA_TYPE][LAST_LOCALITY_TYPE] = {
+	[DATA_ATOM][HEAD_ATOM] =
+	{ .locality = HEAD_ATOM,
+	  .rmw = rmw_data_head,
+	  .offset_at = offset_at_data_head,
+	  .offset_in = offset_in_data_head,
+	  .get_iovec = get_iovec_data_head,
+	  .io_size_nopad = io_size_nopad_data_head,
+	  .count_to_uptodate = count_to_uptodate_data_head,
+	  .get_config = get_config_data
+	},
+	[DATA_ATOM][TAIL_ATOM] =
+	{ .locality = TAIL_ATOM,
+	  .rmw = rmw_data_tail,
+	  .offset_at = offset_at_data_tail,
+	  .offset_in = offset_in_tail,
+	  .get_iovec = get_iovec_data_tail,
+	  .io_size_nopad = io_size_nopad_data_tail,
+	  .count_to_uptodate = count_to_uptodate_data_tail,
+	  .get_config = get_config_data
+	},
+	[DATA_ATOM][FULL_ATOM] =
+	{ .locality = FULL_ATOM,
+	  .offset_at = offset_at_data_full,
+	  .offset_in = offset_in_data_full,
+	  .get_iovec = get_iovec_data_full,
+	  .io_size_nopad = io_size_nopad_data_full,
+	  .get_config = get_config_data
+	},
+	[HOLE_ATOM][HEAD_ATOM] =
+	{ .locality = HEAD_ATOM,
+	  .rmw = rmw_hole_head,
+	  .offset_at = offset_at_hole_head,
+	  .offset_in = offset_in_hole_head,
+	  .get_iovec = get_iovec_hole_head,
+	  .io_size_nopad = io_size_nopad_hole_head,
+	  .count_to_uptodate = count_to_uptodate_hole_head,
+	  .get_config = get_config_hole
+	},
+	[HOLE_ATOM][TAIL_ATOM] =
+	{ .locality = TAIL_ATOM,
+	  .rmw = rmw_hole_tail,
+	  .offset_at = offset_at_hole_tail,
+	  .offset_in = offset_in_tail,
+	  .get_iovec = get_iovec_hole_tail,
+	  .io_size_nopad = io_size_nopad_hole_tail,
+	  .count_to_uptodate = count_to_uptodate_hole_tail,
+	  .get_config = get_config_hole
+	},
+	[HOLE_ATOM][FULL_ATOM] =
+	{ .locality = FULL_ATOM,
+	  .offset_at = offset_at_hole_full,
+	  .offset_in = offset_in_hole_full,
+	  .get_iovec = get_iovec_hole_full,
+	  .io_size_nopad = io_size_nopad_hole_full,
+	  .get_config = get_config_hole
+	}
+};
+
+struct rmw_atom *atom_by_types(atom_data_type data,
+			       atom_locality_type locality)
+{
+	return &atoms[data][locality];
+}
+
+/*
+  Local variables:
+  c-indentation-style: "K&R"
+  mode-name: "LC"
+  c-basic-offset: 8
+  tab-width: 8
+  fill-column: 80
+  scroll-step: 1
+  End:
+*/
diff --git a/xlators/encryption/crypt/src/crypt-common.h b/xlators/encryption/crypt/src/crypt-common.h
new file mode 100644
index 000000000..7c212ad5d
--- /dev/null
+++ b/xlators/encryption/crypt/src/crypt-common.h
@@ -0,0 +1,141 @@
+/*
+  Copyright (c) 2008-2013 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 __CRYPT_COMMON_H__
+#define __CRYPT_COMMON_H__
+
+#define INVAL_SUBVERSION_NUMBER (0xff)
+#define CRYPT_INVAL_OP   (GF_FOP_NULL)
+
+#define CRYPTO_FORMAT_PREFIX  "trusted.glusterfs.crypt.att.cfmt"
+#define FSIZE_XATTR_PREFIX    "trusted.glusterfs.crypt.att.size"
+#define SUBREQ_PREFIX         "trusted.glusterfs.crypt.msg.sreq"
+#define FSIZE_MSG_PREFIX      "trusted.glusterfs.crypt.msg.size"
+#define DE_MSG_PREFIX         "trusted.glusterfs.crypt.msg.dent"
+#define REQUEST_ID_PREFIX     "trusted.glusterfs.crypt.msg.rqid"
+#define MSGFLAGS_PREFIX       "trusted.glusterfs.crypt.msg.xfgs"
+
+
+/* messages for crypt_open() */
+#define MSGFLAGS_REQUEST_MTD_RLOCK   1 /* take read lock and don't unlock */
+#define MSGFLAGS_REQUEST_MTD_WLOCK   2 /* take write lock and don't unlock */
+
+#define AES_BLOCK_BITS (4) /* AES_BLOCK_SIZE == 1 << AES_BLOCK_BITS */
+
+#define noop   do {; } while (0)
+#define cassert(cond) ({ switch (-1) { case (cond): case 0: break; } })
+#define __round_mask(x, y) ((__typeof__(x))((y)-1))
+#define round_up(x, y) ((((x)-1) | __round_mask(x, y))+1)
+
+/*
+ * Format of file's metadata
+ */
+struct crypt_format {
+	uint8_t loader_id;  /* version of metadata loader */
+	uint8_t versioned[0]; /* file's metadata of specific version */
+} __attribute__((packed));
+
+typedef enum {
+	AES_CIPHER_ALG,
+	LAST_CIPHER_ALG
+} cipher_alg_t;
+
+typedef enum {
+	XTS_CIPHER_MODE,
+	LAST_CIPHER_MODE
+} cipher_mode_t;
+
+typedef enum {
+	MTD_LOADER_V1,
+	LAST_MTD_LOADER
+} mtd_loader_id;
+
+static inline void msgflags_set_mtd_rlock(uint32_t *flags)
+{
+	*flags |= MSGFLAGS_REQUEST_MTD_RLOCK;
+}
+
+static inline void msgflags_set_mtd_wlock(uint32_t *flags)
+{
+	*flags |= MSGFLAGS_REQUEST_MTD_WLOCK;
+}
+
+static inline gf_boolean_t msgflags_check_mtd_rlock(uint32_t *flags)
+{
+	return *flags & MSGFLAGS_REQUEST_MTD_RLOCK;
+}
+
+static inline gf_boolean_t msgflags_check_mtd_wlock(uint32_t *flags)
+{
+	return *flags & MSGFLAGS_REQUEST_MTD_WLOCK;
+}
+
+static inline gf_boolean_t msgflags_check_mtd_lock(uint32_t *flags)
+{
+	return  msgflags_check_mtd_rlock(flags) ||
+		msgflags_check_mtd_wlock(flags);
+}
+
+/*
+ * returns number of logical blocks occupied
+ * (maybe partially) by @count bytes
+ * at offset @start.
+ */
+static inline off_t logical_blocks_occupied(uint64_t start, off_t count,
+					    int blkbits)
+{
+	return ((start + count - 1) >> blkbits) - (start >> blkbits) + 1;
+}
+
+/*
+ * are two bytes (represented by offsets @off1
+ * and @off2 respectively) in the same logical
+ * block.
+ */
+static inline int in_same_lblock(uint64_t off1, uint64_t off2,
+				 int blkbits)
+{
+	return off1 >> blkbits == off2 >> blkbits;
+}
+
+static inline void dump_cblock(xlator_t *this, unsigned char *buf)
+{
+	gf_log(this->name, GF_LOG_DEBUG,
+	       "dump cblock: %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x",
+	       (buf)[0],
+	       (buf)[1],
+	       (buf)[2],
+	       (buf)[3],
+	       (buf)[4],
+	       (buf)[5],
+	       (buf)[6],
+	       (buf)[7],
+	       (buf)[8],
+	       (buf)[9],
+	       (buf)[10],
+	       (buf)[11],
+	       (buf)[12],
+	       (buf)[13],
+	       (buf)[14],
+	       (buf)[15]);
+}
+
+#endif /* __CRYPT_COMMON_H__ */
+
+/*
+  Local variables:
+  c-indentation-style: "K&R"
+  mode-name: "LC"
+  c-basic-offset: 8
+  tab-width: 8
+  fill-column: 80
+  scroll-step: 1
+  End:
+*/
diff --git a/xlators/encryption/crypt/src/crypt-mem-types.h b/xlators/encryption/crypt/src/crypt-mem-types.h
new file mode 100644
index 000000000..799727573
--- /dev/null
+++ b/xlators/encryption/crypt/src/crypt-mem-types.h
@@ -0,0 +1,43 @@
+/*
+  Copyright (c) 2008-2013 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 __CRYPT_MEM_TYPES_H__
+#define __CRYPT_MEM_TYPES_H__
+
+#include "mem-types.h"
+
+enum gf_crypt_mem_types_ {
+	gf_crypt_mt_priv = gf_common_mt_end + 1,
+	gf_crypt_mt_inode,
+	gf_crypt_mt_data,
+	gf_crypt_mt_mtd,
+	gf_crypt_mt_loc,
+	gf_crypt_mt_iatt,
+	gf_crypt_mt_key,
+	gf_crypt_mt_iovec,
+	gf_crypt_mt_char,
+};
+
+#endif /* __CRYPT_MEM_TYPES_H__ */
+
+/*
+  Local variables:
+  c-indentation-style: "K&R"
+  mode-name: "LC"
+  c-basic-offset: 8
+  tab-width: 8
+  fill-column: 80
+  scroll-step: 1
+  End:
+*/
+
+
+
diff --git a/xlators/encryption/crypt/src/crypt.c b/xlators/encryption/crypt/src/crypt.c
new file mode 100644
index 000000000..db2e6d83c
--- /dev/null
+++ b/xlators/encryption/crypt/src/crypt.c
@@ -0,0 +1,4498 @@
+/*
+  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 <ctype.h>
+#include <sys/uio.h>
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "glusterfs.h"
+#include "xlator.h"
+#include "logging.h"
+#include "defaults.h"
+
+#include "crypt-common.h"
+#include "crypt.h"
+
+static void init_inode_info_head(struct crypt_inode_info *info, fd_t *fd);
+static int32_t init_inode_info_tail(struct crypt_inode_info *info,
+				    struct master_cipher_info *master);
+static int32_t prepare_for_submit_hole(call_frame_t *frame, xlator_t *this,
+				       uint64_t from, off_t size);
+static int32_t load_file_size(call_frame_t *frame, void *cookie,
+			      xlator_t *this, int32_t op_ret, int32_t op_errno,
+			      dict_t *dict, dict_t *xdata);
+static void do_ordered_submit(call_frame_t *frame, xlator_t *this,
+			      atom_data_type dtype);
+static void do_parallel_submit(call_frame_t *frame, xlator_t *this,
+			       atom_data_type dtype);
+static void put_one_call_open(call_frame_t *frame);
+static void put_one_call_readv(call_frame_t *frame, xlator_t *this);
+static void put_one_call_writev(call_frame_t *frame, xlator_t *this);
+static void put_one_call_ftruncate(call_frame_t *frame, xlator_t *this);
+static void free_avec(struct iovec *avec, char **pool, int blocks_in_pool);
+static void free_avec_data(crypt_local_t *local);
+static void free_avec_hole(crypt_local_t *local);
+
+static crypt_local_t *crypt_alloc_local(call_frame_t *frame, xlator_t *this,
+					glusterfs_fop_t fop)
+{
+	crypt_local_t *local = NULL;
+
+	local = mem_get0(this->local_pool);
+	if (!local) {
+		gf_log(this->name, GF_LOG_ERROR, "out of memory");
+		return NULL;
+	}
+	local->fop = fop;
+	LOCK_INIT(&local->hole_lock);
+	LOCK_INIT(&local->call_lock);
+	LOCK_INIT(&local->rw_count_lock);
+
+	frame->local = local;
+	return local;
+}
+
+struct crypt_inode_info *get_crypt_inode_info(inode_t *inode, xlator_t *this)
+{
+	int ret;
+	uint64_t value = 0;
+	struct crypt_inode_info *info;
+
+	ret = inode_ctx_get(inode, this, &value);
+	if (ret == -1) {
+		gf_log (this->name, GF_LOG_WARNING,
+			"Can not get inode info");
+		return NULL;
+	}
+	info = (struct crypt_inode_info *)(long)value;
+	if (info == NULL) {
+		gf_log (this->name, GF_LOG_WARNING,
+			"Can not obtain inode info");
+		return NULL;
+	}
+	return info;
+}
+
+static struct crypt_inode_info *local_get_inode_info(crypt_local_t *local,
+						      xlator_t *this)
+{
+	if (local->info)
+		return local->info;
+	local->info = get_crypt_inode_info(local->fd->inode, this);
+	return local->info;
+}
+
+static struct crypt_inode_info *alloc_inode_info(crypt_local_t *local,
+						 loc_t *loc)
+{
+	struct crypt_inode_info *info;
+
+	info = GF_CALLOC(1, sizeof(*info), gf_crypt_mt_inode);
+	if (!info) {
+		local->op_ret = -1;
+		local->op_errno = ENOMEM;
+		gf_log ("crypt", GF_LOG_WARNING,
+			"Can not allocate inode info");
+		return NULL;
+	}
+	memset(info, 0, sizeof(*info));
+#if DEBUG_CRYPT
+	info->loc = GF_CALLOC(1, sizeof(*loc), gf_crypt_mt_loc);
+	if (!info->loc) {
+		gf_log("crypt", GF_LOG_WARNING, "Can not allocate loc");
+		GF_FREE(info);
+		return NULL;
+	}
+	if (loc_copy(info->loc, loc)){
+		GF_FREE(info->loc);
+		GF_FREE(info);
+		return NULL;
+	}
+#endif /* DEBUG_CRYPT */
+
+	local->info = info;
+	return info;
+}
+
+static void free_inode_info(struct crypt_inode_info *info)
+{
+#if DEBUG_CRYPT
+	loc_wipe(info->loc);
+	GF_FREE(info->loc);
+#endif
+	memset(info, 0, sizeof(*info));
+	GF_FREE(info);
+}
+
+int crypt_forget (xlator_t *this, inode_t *inode)
+{
+        uint64_t ctx_addr = 0;
+        if (!inode_ctx_del (inode, this, &ctx_addr))
+                free_inode_info((struct crypt_inode_info *)(long)ctx_addr);
+        return 0;
+}
+
+#if DEBUG_CRYPT
+static void check_read(call_frame_t *frame, xlator_t *this, int32_t read,
+		       struct iovec *vec, int32_t count, struct iatt *stbuf)
+{
+	crypt_local_t *local = frame->local;
+	struct object_cipher_info *object = get_object_cinfo(local->info);
+	struct avec_config *conf = &local->data_conf;
+	uint32_t resid = stbuf->ia_size & (object_alg_blksize(object) - 1);
+
+	if (read <= 0)
+		return;
+	if (read != iovec_get_size(vec, count))
+		gf_log ("crypt", GF_LOG_DEBUG,
+			"op_ret differs from amount of read bytes");
+
+	if (object_alg_should_pad(object) && (read & (object_alg_blksize(object) - 1)))
+		gf_log ("crypt", GF_LOG_DEBUG,
+			"bad amount of read bytes (!= 0 mod(cblock size))");
+
+	if (conf->aligned_offset + read >
+	    stbuf->ia_size + (resid ? object_alg_blksize(object) - resid : 0))
+		gf_log ("crypt", GF_LOG_DEBUG,
+			"bad amount of read bytes (too large))");
+
+}
+
+#define PT_BYTES_TO_DUMP (32)
+static void dump_plain_text(crypt_local_t *local, struct iovec *avec)
+{
+	int32_t to_dump;
+	char str[PT_BYTES_TO_DUMP + 1];
+
+	if (!avec)
+		return;
+	to_dump = avec->iov_len;
+	if (to_dump > PT_BYTES_TO_DUMP)
+		to_dump = PT_BYTES_TO_DUMP;
+	memcpy(str, avec->iov_base, to_dump);
+	memset(str + to_dump, '0', 1);
+	gf_log("crypt", GF_LOG_DEBUG, "Read file: %s", str);
+}
+
+static int32_t data_conf_invariant(struct avec_config *conf)
+{
+	return conf->acount ==
+		!!has_head_block(conf) +
+		!!has_tail_block(conf)+
+		conf->nr_full_blocks;
+}
+
+static int32_t hole_conf_invariant(struct avec_config *conf)
+{
+	return conf->blocks_in_pool ==
+		!!has_head_block(conf) +
+		!!has_tail_block(conf)+
+		!!has_full_blocks(conf);
+}
+
+static void crypt_check_conf(struct avec_config *conf)
+{
+	int32_t ret = 0;
+	const char *msg;
+
+	switch (conf->type) {
+	case DATA_ATOM:
+		msg = "data";
+		ret = data_conf_invariant(conf);
+		break;
+	case HOLE_ATOM:
+		msg = "hole";
+		ret = hole_conf_invariant(conf);
+		break;
+	default:
+		msg = "unknown";
+	}
+	if (!ret)
+		gf_log("crypt", GF_LOG_DEBUG, "bad %s conf", msg);
+}
+
+static void check_buf(call_frame_t *frame, xlator_t *this, struct iatt *buf)
+{
+	crypt_local_t *local = frame->local;
+	struct object_cipher_info *object = &local->info->cinfo;
+	uint64_t local_file_size;
+
+	switch(local->fop) {
+	case GF_FOP_FTRUNCATE:
+		return;
+	case GF_FOP_WRITE:
+		local_file_size = local->new_file_size;
+		break;
+	case GF_FOP_READ:
+		if (parent_is_crypt_xlator(frame, this))
+			return;
+		local_file_size = local->cur_file_size;
+		break;
+	default:
+		gf_log("crypt", GF_LOG_DEBUG, "bad file operation");
+		return;
+	}
+	if (buf->ia_size != round_up(local_file_size,
+					 object_alg_blksize(object)))
+		gf_log("crypt", GF_LOG_DEBUG,
+		       "bad ia_size in buf (%llu), should be %llu",
+		       (unsigned long long)buf->ia_size,
+		       (unsigned long long)round_up(local_file_size,
+						    object_alg_blksize(object)));
+}
+
+#else
+#define check_read(frame, this, op_ret, vec, count, stbuf) noop
+#define dump_plain_text(local, avec) noop
+#define crypt_check_conf(conf) noop
+#define check_buf(frame, this, buf) noop
+#endif /* DEBUG_CRYPT */
+
+/*
+ * Pre-conditions:
+ * @vec represents a ciphertext of expanded size and
+ * aligned offset.
+ *
+ * Compound a temporal vector @avec with block-aligned
+ * components, decrypt and fix it up to represent a chunk
+ * of data corresponding to the original size and offset.
+ * Pass the result to the next translator.
+ */
+int32_t crypt_readv_cbk(call_frame_t *frame,
+			void *cookie,
+			xlator_t *this,
+			int32_t op_ret,
+			int32_t op_errno,
+			struct iovec *vec,
+			int32_t count,
+			struct iatt *stbuf,
+			struct iobref *iobref,
+			dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+	struct avec_config *conf = &local->data_conf;
+	struct object_cipher_info *object = &local->info->cinfo;
+
+	struct iovec *avec;
+	uint32_t i;
+	uint32_t to_vec;
+	uint32_t to_user;
+
+	check_buf(frame, this, stbuf);
+	check_read(frame, this, op_ret, vec, count, stbuf);
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+	local->iobref = iobref_ref(iobref);
+
+	local->buf = *stbuf;
+	local->buf.ia_size = local->cur_file_size;
+
+	if (op_ret <= 0 || count == 0 || vec[0].iov_len == 0)
+		goto put_one_call;
+
+	if (conf->orig_offset >= local->cur_file_size) {
+		local->op_ret = 0;
+		goto put_one_call;
+	}
+	/* 
+	 * correct config params with real file size
+	 * and actual amount of bytes read
+	 */
+	set_config_offsets(frame, this,
+			   conf->orig_offset, op_ret, DATA_ATOM, 0);
+
+	if (conf->orig_offset + conf->orig_size > local->cur_file_size)
+		conf->orig_size = local->cur_file_size - conf->orig_offset;
+	/*
+	 * calculate amount of data to be returned
+	 * to user.
+	 */
+	to_user = op_ret;
+	if (conf->aligned_offset + to_user <= conf->orig_offset) {
+		gf_log(this->name, GF_LOG_WARNING, "Incomplete read");
+		local->op_ret = -1;
+		local->op_errno = EIO;
+		goto put_one_call;
+	}
+	to_user -= (conf->aligned_offset - conf->orig_offset);
+
+	if (to_user > conf->orig_size)
+		to_user = conf->orig_size;
+	local->rw_count = to_user;
+
+	op_errno = set_config_avec_data(this, local,
+					conf, object, vec, count);
+	if (op_errno) {
+		local->op_ret = -1;
+		local->op_errno = op_errno;
+		goto put_one_call;
+	}
+	avec = conf->avec;
+#if DEBUG_CRYPT
+	if (conf->off_in_tail != 0 &&
+	    conf->off_in_tail < object_alg_blksize(object) &&
+	    object_alg_should_pad(object))
+		gf_log(this->name, GF_LOG_DEBUG, "Bad offset in tail %d",
+		       conf->off_in_tail);
+	if (iovec_get_size(vec, count) != 0 &&
+	    in_same_lblock(conf->orig_offset + iovec_get_size(vec, count) - 1,
+			   local->cur_file_size - 1,
+			   object_alg_blkbits(object))) {
+		gf_log(this->name, GF_LOG_DEBUG, "Compound last cblock");
+		dump_cblock(this,
+			    (unsigned char *)(avec[conf->acount - 1].iov_base) +
+			    avec[conf->acount - 1].iov_len - object_alg_blksize(object));
+		dump_cblock(this,
+			    (unsigned char *)(vec[count - 1].iov_base) +
+			    vec[count - 1].iov_len - object_alg_blksize(object));
+	}
+#endif
+	decrypt_aligned_iov(object, avec,
+			    conf->acount, conf->aligned_offset);
+	/*
+	 * pass proper plain data to user
+	 */
+	avec[0].iov_base += (conf->aligned_offset - conf->orig_offset);
+	avec[0].iov_len  -= (conf->aligned_offset - conf->orig_offset);
+
+	to_vec = to_user;
+	for (i = 0; i < conf->acount; i++) {
+		if (avec[i].iov_len > to_vec)
+			avec[i].iov_len = to_vec;
+		to_vec -= avec[i].iov_len;
+	}
+ put_one_call:
+	put_one_call_readv(frame, this);
+	return 0;
+}
+
+static int32_t do_readv(call_frame_t *frame,
+			void *cookie,
+			xlator_t *this,
+			int32_t op_ret,
+			int32_t op_errno,
+			dict_t *dict,
+			dict_t *xdata)
+{
+	data_t *data;
+	crypt_local_t *local = frame->local;
+
+	if (op_ret < 0)
+		goto error;	
+	/*
+	 * extract regular file size
+	 */
+	data = dict_get(dict, FSIZE_XATTR_PREFIX);
+	if (!data) {
+		gf_log("crypt", GF_LOG_WARNING, "Regular file size not found");
+		op_errno = EIO;
+		goto error;
+	}
+	local->cur_file_size = data_to_uint64(data);
+	
+	get_one_call(frame);
+	STACK_WIND(frame,
+		   crypt_readv_cbk,
+		   FIRST_CHILD (this),
+		   FIRST_CHILD (this)->fops->readv,
+		   local->fd,
+		   /*
+		    * FIXME: read amount can be reduced
+		    */
+		   local->data_conf.expanded_size,
+		   local->data_conf.aligned_offset,
+		   local->flags,
+		   local->xdata);
+	return 0;
+ error:
+	local->op_ret = -1;
+	local->op_errno = op_errno;
+
+	get_one_call(frame);
+	put_one_call_readv(frame, this);
+	return 0;
+}
+
+static int32_t crypt_readv_finodelk_cbk(call_frame_t *frame,
+					void *cookie,
+					xlator_t *this,
+					int32_t op_ret,
+					int32_t op_errno,
+					dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	if (op_ret < 0)
+		goto error;
+	/*
+	 * An access has been granted,
+	 * retrieve file size
+	 */
+	STACK_WIND(frame,
+		   do_readv,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fgetxattr,
+		   local->fd,
+		   FSIZE_XATTR_PREFIX,
+		   NULL);
+	return  0;
+ error:
+	fd_unref(local->fd);
+	if (local->xdata)
+		dict_unref(local->xdata);
+	STACK_UNWIND_STRICT(readv,
+			    frame,
+			    -1,
+			    op_errno,
+			    NULL,
+			    0,
+			    NULL,
+			    NULL,
+			    NULL);
+	return 0;
+}
+
+static int32_t readv_trivial_completion(call_frame_t *frame,
+					void *cookie,
+					xlator_t *this,
+					int32_t op_ret,
+					int32_t op_errno,
+					struct iatt *buf,
+					dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (op_ret < 0) {
+		gf_log(this->name, GF_LOG_WARNING,
+		       "stat failed (%d)", op_errno);
+		goto error;
+	}
+	local->buf = *buf;
+	STACK_WIND(frame,
+		   load_file_size,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->getxattr,
+		   local->loc,
+		   FSIZE_XATTR_PREFIX,
+		   NULL);
+	return 0;	
+ error:
+	STACK_UNWIND_STRICT(readv, frame, op_ret, op_errno,
+			    NULL, 0, NULL, NULL, NULL);
+	return 0;
+}
+
+int32_t crypt_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 ret;
+	crypt_local_t *local;
+	struct crypt_inode_info *info;
+	struct gf_flock lock = {0, };
+
+#if DEBUG_CRYPT
+	gf_log("crypt", GF_LOG_DEBUG, "reading %d bytes from offset %llu",
+	       (int)size, (long long)offset);
+	if (parent_is_crypt_xlator(frame, this))
+		gf_log("crypt", GF_LOG_DEBUG, "parent is crypt");
+#endif	
+	local = crypt_alloc_local(frame, this, GF_FOP_READ);
+	if (!local) {
+		ret = ENOMEM;
+		goto error;
+	}
+	if (size == 0)
+		goto trivial;
+
+	local->fd = fd_ref(fd);
+	local->flags = flags;
+
+	info = local_get_inode_info(local, this);
+	if (info == NULL) {
+		ret = EINVAL;
+		fd_unref(fd);
+		goto error;
+	}
+	if (!object_alg_atomic(&info->cinfo)) {
+		ret = EINVAL;
+		fd_unref(fd);
+		goto error;
+	}
+	set_config_offsets(frame, this, offset, size,
+			   DATA_ATOM, 0);
+	if (parent_is_crypt_xlator(frame, this)) {
+		data_t *data;
+		/*
+		 * We are called by crypt_writev (or cypt_ftruncate)
+		 * to perform the "read" component of the read-modify-write
+		 * (or read-prune-write) sequence for some atom;
+		 *
+		 * don't ask for access:
+		 * it has already been acquired
+		 *
+		 * Retrieve current file size
+		 */
+		if (!xdata) {
+			gf_log("crypt", GF_LOG_WARNING,
+			       "Regular file size hasn't been passed");
+			ret = EIO;
+			goto error;
+		}
+		data = dict_get(xdata, FSIZE_XATTR_PREFIX);
+		if (!data) {
+			gf_log("crypt", GF_LOG_WARNING,
+			       "Regular file size not found");
+			ret = EIO;
+			goto error;
+		}
+		local->old_file_size =
+			local->cur_file_size = data_to_uint64(data);
+
+		get_one_call(frame);
+		STACK_WIND(frame,
+			   crypt_readv_cbk,
+			   FIRST_CHILD(this),
+			   FIRST_CHILD(this)->fops->readv,
+			   local->fd,
+			   /*
+			    * FIXME: read amount can be reduced
+			    */
+			   local->data_conf.expanded_size,
+			   local->data_conf.aligned_offset,
+			   flags,
+			   NULL);
+		return 0;
+	}
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+
+	lock.l_len    = 0;
+        lock.l_start  = 0;
+        lock.l_type   = F_RDLCK;
+        lock.l_whence = SEEK_SET;
+
+	STACK_WIND(frame,
+		   crypt_readv_finodelk_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->finodelk,
+		   this->name,
+		   fd,
+		   F_SETLKW,
+		   &lock,
+		   NULL);
+	return 0;
+ trivial:
+	STACK_WIND(frame,
+		   readv_trivial_completion,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fstat,
+		   fd,
+		   NULL);
+	return 0;
+ error:
+	STACK_UNWIND_STRICT(readv,
+			    frame,
+			    -1,
+			    ret,
+			    NULL,
+			    0,
+			    NULL,
+			    NULL,
+			    NULL);
+	return 0;
+}
+
+void set_local_io_params_writev(call_frame_t *frame,
+				struct object_cipher_info *object,
+				struct rmw_atom *atom,
+				off_t io_offset,
+				uint32_t io_size)
+{
+	crypt_local_t *local = frame->local;
+
+	local->io_offset = io_offset;
+	local->io_size = io_size;
+
+	local->io_offset_nopad =
+		atom->offset_at(frame, object) + atom->offset_in(frame, object);
+
+	gf_log("crypt", GF_LOG_DEBUG,
+	       "set nopad offset to %llu",
+	       (unsigned long long)local->io_offset_nopad);
+
+	local->io_size_nopad = atom->io_size_nopad(frame, object);
+
+	gf_log("crypt", GF_LOG_DEBUG,
+	       "set nopad size to %llu",
+	       (unsigned long long)local->io_size_nopad);
+
+	local->update_disk_file_size = 0;
+	/*
+	 * NOTE: eof_padding_size is 0 for all full atoms;
+	 * For head and tail atoms it will be set up at rmw_partial block()
+	 */
+	local->new_file_size = local->cur_file_size;
+
+	if (local->io_offset_nopad + local->io_size_nopad > local->cur_file_size) {
+
+		local->new_file_size = local->io_offset_nopad + local->io_size_nopad;
+
+		gf_log("crypt", GF_LOG_DEBUG,
+		       "set new file size to %llu",
+		       (unsigned long long)local->new_file_size);
+
+		local->update_disk_file_size = 1;
+	}
+}
+
+void set_local_io_params_ftruncate(call_frame_t *frame,
+				   struct object_cipher_info *object)
+{
+	uint32_t resid;
+	crypt_local_t *local = frame->local;
+	struct avec_config *conf = &local->data_conf;
+
+	resid = conf->orig_offset & (object_alg_blksize(object) - 1);
+	if (resid) {
+		local->eof_padding_size =
+			object_alg_blksize(object) - resid;
+		local->new_file_size = conf->aligned_offset;
+		local->update_disk_file_size = 0;
+		/*
+		 * file size will be updated
+		 * in the ->writev() stack,
+		 * when submitting file tail
+		 */
+	}
+	else {	
+		local->eof_padding_size = 0;
+		local->new_file_size = conf->orig_offset;
+		local->update_disk_file_size = 1;
+		/*
+		 * file size will be updated
+		 * in this ->ftruncate stack
+		 */
+	}
+}
+
+static inline void submit_head(call_frame_t *frame, xlator_t *this)
+{
+	crypt_local_t *local = frame->local;
+	submit_partial(frame, this, local->fd, HEAD_ATOM);
+}
+
+static inline void submit_tail(call_frame_t *frame, xlator_t *this)
+{
+	crypt_local_t *local = frame->local;
+	submit_partial(frame, this, local->fd, TAIL_ATOM);
+}
+
+static void submit_hole(call_frame_t *frame, xlator_t *this)
+{
+	/*
+	 * hole conversion always means
+	 * appended write and goes in ordered fashion
+	 */
+	do_ordered_submit(frame, this, HOLE_ATOM);
+}
+
+static void submit_data(call_frame_t *frame, xlator_t *this)
+{
+	if (is_ordered_mode(frame)) {
+		do_ordered_submit(frame, this, DATA_ATOM);
+		return;
+	}
+	gf_log("crypt", GF_LOG_WARNING, "Bad submit mode");
+	get_nr_calls(frame, nr_calls_data(frame));
+	do_parallel_submit(frame, this, DATA_ATOM);
+	return;
+}
+
+/*
+ * heplers called by writev_cbk, fruncate_cbk in ordered mode
+ */
+
+static inline int32_t should_submit_hole(crypt_local_t *local)
+{
+	struct avec_config *conf = &local->hole_conf;
+
+	return conf->avec != NULL;
+}
+
+static inline int32_t should_resume_submit_hole(crypt_local_t *local)
+{
+	struct avec_config *conf = &local->hole_conf;
+
+	if (local->fop == GF_FOP_WRITE && has_tail_block(conf))
+		/*
+		 * Don't submit a part of hole, which
+		 * fits into a data block:
+ 		 * this part of hole will be converted
+		 * as a gap filled by zeros in data head
+		 * block.
+		 */
+		return conf->cursor < conf->acount - 1;
+	else
+		return conf->cursor < conf->acount;
+}
+
+static inline int32_t should_resume_submit_data(call_frame_t *frame)
+{
+	crypt_local_t *local = frame->local;
+	struct avec_config *conf = &local->data_conf;
+
+	if (is_ordered_mode(frame))
+		return conf->cursor < conf->acount;
+	/*
+	 * parallel writes
+	 */
+	return 0;
+}
+
+static inline int32_t should_submit_data_after_hole(crypt_local_t *local)
+{
+	return local->data_conf.avec != NULL;
+}
+
+static void update_local_file_params(call_frame_t *frame,
+				     xlator_t *this,
+				     struct iatt *prebuf,
+				     struct iatt *postbuf)
+{
+	crypt_local_t *local = frame->local;
+
+	check_buf(frame, this, postbuf);
+
+	local->prebuf = *prebuf;
+	local->postbuf = *postbuf;
+
+	local->prebuf.ia_size  = local->cur_file_size;
+	local->postbuf.ia_size = local->new_file_size;
+
+	local->cur_file_size = local->new_file_size;	
+}
+
+static int32_t end_writeback_writev(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)
+{
+	crypt_local_t *local = frame->local;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (op_ret <= 0) {
+		gf_log(this->name, GF_LOG_WARNING,
+		       "writev iteration failed");
+		goto put_one_call;
+	}
+	/*
+	 * op_ret includes paddings (atom's head, atom's tail and EOF)
+	 */
+	if (op_ret < local->io_size) {
+		gf_log(this->name, GF_LOG_WARNING,
+		       "Incomplete writev iteration");
+		goto put_one_call;
+	}
+	op_ret -= local->eof_padding_size;
+	local->op_ret = op_ret;
+
+	update_local_file_params(frame, this, prebuf, postbuf);
+
+	if (data_write_in_progress(local)) {
+
+		LOCK(&local->rw_count_lock);
+		local->rw_count += op_ret;
+		UNLOCK(&local->rw_count_lock);
+
+		if (should_resume_submit_data(frame))
+			submit_data(frame, this);
+	}
+	else {
+		/*
+		 * hole conversion is going on;
+		 * don't take into account written zeros
+		 */
+		if (should_resume_submit_hole(local))
+			submit_hole(frame, this);
+
+		else if (should_submit_data_after_hole(local))
+			submit_data(frame, this);
+	}
+ put_one_call:
+	put_one_call_writev(frame, this);
+	return 0;
+}
+
+#define crypt_writev_cbk end_writeback_writev
+
+#define HOLE_WRITE_CHUNK_BITS 12
+#define HOLE_WRITE_CHUNK_SIZE (1 << HOLE_WRITE_CHUNK_BITS)
+
+/*
+ * Convert hole of size @size at offset @off to
+ * zeros and prepare respective iovecs for submit.
+ * The hole lock should be held.
+ *
+ * Pre-conditions:
+ * @local->file_size is set and valid.
+ */
+int32_t prepare_for_submit_hole(call_frame_t *frame, xlator_t *this,
+				uint64_t off, off_t size)
+{
+	int32_t ret;
+	crypt_local_t *local = frame->local;
+	struct object_cipher_info *object = &local->info->cinfo;
+
+	set_config_offsets(frame, this, off, size, HOLE_ATOM, 1);
+
+	ret = set_config_avec_hole(this, local,
+				   &local->hole_conf, object, local->fop);
+	crypt_check_conf(&local->hole_conf);
+
+	return ret;
+}
+
+/*
+ * prepare for submit @count bytes at offset @from
+ */
+int32_t prepare_for_submit_data(call_frame_t *frame, xlator_t *this,
+				off_t from, int32_t size, struct iovec *vec,
+				int32_t vec_count, int32_t setup_gap)
+{
+	uint32_t ret;
+	crypt_local_t *local = frame->local;
+	struct object_cipher_info *object = &local->info->cinfo;
+
+	set_config_offsets(frame, this, from, size,
+			   DATA_ATOM, setup_gap);
+
+	ret = set_config_avec_data(this, local,
+				   &local->data_conf, object, vec, vec_count);
+	crypt_check_conf(&local->data_conf);
+
+	return ret;
+}
+
+static void free_avec(struct iovec *avec,
+		      char **pool, int blocks_in_pool)
+{
+	if (!avec)
+		return;
+	GF_FREE(pool);
+	GF_FREE(avec);
+}
+
+static void free_avec_data(crypt_local_t *local)
+{
+	return free_avec(local->data_conf.avec,
+			 local->data_conf.pool,
+			 local->data_conf.blocks_in_pool);
+}
+
+static void free_avec_hole(crypt_local_t *local)
+{
+	return free_avec(local->hole_conf.avec,
+			 local->hole_conf.pool,
+			 local->hole_conf.blocks_in_pool);
+}
+
+
+static void do_parallel_submit(call_frame_t *frame, xlator_t *this,
+			       atom_data_type dtype)
+{
+	crypt_local_t *local = frame->local;
+	struct avec_config *conf;
+
+	local->active_setup = dtype;
+	conf = conf_by_type(frame, dtype);
+
+	if (has_head_block(conf))
+		submit_head(frame, this);
+
+	if (has_full_blocks(conf))
+		submit_full(frame, this);
+
+	if (has_tail_block(conf))
+		submit_tail(frame, this);
+	return;
+}
+
+static void do_ordered_submit(call_frame_t *frame, xlator_t *this,
+			      atom_data_type dtype)
+{
+	crypt_local_t *local = frame->local;
+	struct avec_config *conf;
+
+	local->active_setup = dtype;
+	conf = conf_by_type(frame, dtype);
+
+	if (should_submit_head_block(conf)) {
+		get_one_call_nolock(frame);
+		submit_head(frame, this);
+	}
+	else if (should_submit_full_block(conf)) {
+		get_one_call_nolock(frame);
+		submit_full(frame, this);
+	}
+	else if (should_submit_tail_block(conf)) {
+		get_one_call_nolock(frame);
+		submit_tail(frame, this);
+	}
+	else
+		gf_log("crypt", GF_LOG_DEBUG,
+		       "nothing has been submitted in ordered mode");
+	return;
+}
+
+static int32_t do_writev(call_frame_t *frame,
+			 void *cookie,
+			 xlator_t *this,
+			 int32_t op_ret,
+			 int32_t op_errno,
+			 dict_t *dict,
+			 dict_t *xdata)
+{
+	data_t *data;
+	crypt_local_t *local = frame->local;
+	struct object_cipher_info *object = &local->info->cinfo;
+	/*
+	 * extract regular file size
+	 */
+	data = dict_get(dict, FSIZE_XATTR_PREFIX);
+	if (!data) {
+		gf_log("crypt", GF_LOG_WARNING, "Regular file size not found");
+		op_ret = -1;
+		op_errno = EIO;
+		goto error;
+	}
+	local->old_file_size = local->cur_file_size = data_to_uint64(data);
+
+	set_gap_at_end(frame, object, &local->data_conf, DATA_ATOM);
+
+	if (local->cur_file_size < local->data_conf.orig_offset) {
+		/*
+		 * Set up hole config
+		 */
+		op_errno = prepare_for_submit_hole(frame,
+			       this,
+			       local->cur_file_size,
+			       local->data_conf.orig_offset - local->cur_file_size);
+		if (op_errno) {
+			local->op_ret = -1;
+			local->op_errno = op_errno;
+			goto error;
+		}
+	}
+	if (should_submit_hole(local))
+		submit_hole(frame, this);
+	else 
+		submit_data(frame, this);
+	return 0;
+ error:
+	get_one_call_nolock(frame);
+	put_one_call_writev(frame, this);
+	return 0;
+}
+
+static int32_t crypt_writev_finodelk_cbk(call_frame_t *frame,
+					 void *cookie,
+					 xlator_t *this,
+					 int32_t op_ret,
+					 int32_t op_errno,
+					 dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (op_ret < 0)
+		goto error;
+	/*
+	 * An access has been granted,
+	 * retrieve file size first
+	 */
+	STACK_WIND(frame,
+		   do_writev,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fgetxattr,
+		   local->fd,
+		   FSIZE_XATTR_PREFIX,
+		   NULL);
+	return 0;
+ error:
+	get_one_call_nolock(frame);
+	put_one_call_writev(frame, this);
+	return 0;	
+}
+
+static int32_t writev_trivial_completion(call_frame_t *frame,
+					 void *cookie,
+					 xlator_t *this,
+					 int32_t op_ret,
+					 int32_t op_errno,
+					 struct iatt *buf,
+					 dict_t *dict)
+{
+	crypt_local_t *local = frame->local;
+	
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+	local->prebuf = *buf;
+	local->postbuf = *buf;
+
+	local->prebuf.ia_size = local->cur_file_size;
+	local->postbuf.ia_size = local->cur_file_size;
+
+	get_one_call(frame);
+	put_one_call_writev(frame, this);
+	return 0;
+}
+
+int crypt_writev(call_frame_t *frame,
+		 xlator_t *this,
+		 fd_t *fd,
+		 struct iovec *vec,
+		 int32_t count,
+		 off_t offset,
+		 uint32_t flags,
+		 struct iobref *iobref,
+		 dict_t *xdata)
+{
+	int32_t ret;
+	crypt_local_t *local;
+	struct crypt_inode_info *info;
+	struct gf_flock lock = {0, };
+#if DEBUG_CRYPT
+	gf_log ("crypt", GF_LOG_DEBUG, "writing %d bytes from offset %llu",
+		(int)iovec_get_size(vec, count), (long long)offset);
+#endif
+	local = crypt_alloc_local(frame, this, GF_FOP_WRITE);
+	if (!local) {
+		ret = ENOMEM;
+		goto error;
+	}
+	local->fd = fd_ref(fd);
+
+	if (iobref)
+		local->iobref = iobref_ref(iobref);
+	/*
+	 * to update real file size on the server
+	 */
+	local->xattr = dict_new();
+	if (!local->xattr) {
+		ret = ENOMEM;
+		goto error;
+	}
+	local->flags = flags;
+
+	info = local_get_inode_info(local, this);
+	if (info == NULL) {
+		ret = EINVAL;
+		goto error;
+	}
+	if (!object_alg_atomic(&info->cinfo)) {
+		ret = EINVAL;
+		goto error;
+	}
+	if (iovec_get_size(vec, count) == 0)
+		goto trivial;
+
+	ret = prepare_for_submit_data(frame, this, offset,
+				      iovec_get_size(vec, count),
+				      vec, count, 0 /* don't setup gup
+						       in tail: we don't
+						       know file size yet */);
+	if (ret)
+		goto error;
+
+	if (parent_is_crypt_xlator(frame, this)) {
+		data_t *data;
+		/*
+		 * we are called by shinking crypt_ftruncate(),
+		 * which doesn't perform hole conversion;
+		 *
+		 * don't ask for access:
+		 * it has already been acquired
+		 */
+
+		/*
+		 * extract file size
+		 */
+		if (!xdata) {
+			gf_log("crypt", GF_LOG_WARNING,
+			       "Regular file size hasn't been passed");
+			ret = EIO;
+			goto error;
+		}
+		data = dict_get(xdata, FSIZE_XATTR_PREFIX);
+		if (!data) {
+			gf_log("crypt", GF_LOG_WARNING,
+			       "Regular file size not found");
+			ret = EIO;
+			goto error;
+		}
+		local->old_file_size =
+			local->cur_file_size = data_to_uint64(data);
+
+		submit_data(frame, this);
+		return 0;
+	}
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+	/*
+	 * lock the file and retrieve its size
+	 */
+	lock.l_len    = 0;
+        lock.l_start  = 0;
+        lock.l_type   = F_WRLCK;
+        lock.l_whence = SEEK_SET;
+
+	STACK_WIND(frame,
+		   crypt_writev_finodelk_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->finodelk,
+		   this->name,
+		   fd,
+		   F_SETLKW,
+		   &lock,
+		   NULL);
+	return 0;
+ trivial:
+	STACK_WIND(frame,
+		   writev_trivial_completion,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fstat,
+		   fd,
+		   NULL);
+	return 0;
+ error:
+	if (local && local->fd)
+		fd_unref(fd);
+	if (local && local->iobref)
+		iobref_unref(iobref);
+	if (local && local->xdata)
+		dict_unref(xdata);
+	if (local && local->xattr)
+		dict_unref(local->xattr);
+	if (local && local->info)
+		free_inode_info(local->info);
+
+	STACK_UNWIND_STRICT(writev, frame, -1, ret, NULL, NULL, NULL);
+	return 0;
+}
+
+int32_t prepare_for_prune(call_frame_t *frame, xlator_t *this, uint64_t offset)
+{
+	set_config_offsets(frame, this,
+			   offset,
+			   0, /* count */
+			   DATA_ATOM,
+			   0 /* since we prune, there is no
+				gap in tail to uptodate */);
+	return 0;
+}
+
+/*
+ * Finish the read-prune-modify sequence
+ *
+ * Can be invoked as
+ * 1) ->ftruncate_cbk() for cblock-aligned, or trivial prune
+ * 2) ->writev_cbk() for non-cblock-aligned prune
+ */
+
+static int32_t prune_complete(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)
+{
+	crypt_local_t *local = frame->local;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	update_local_file_params(frame, this, prebuf, postbuf);
+
+	put_one_call_ftruncate(frame, this);
+	return 0;
+}
+
+/*
+ * This is called as ->ftruncate_cbk()
+ *
+ * Perform the "write" component of the
+ * read-prune-write sequence.
+ *
+ * submuit the rest of the file
+ */
+static int32_t prune_submit_file_tail(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)
+{
+	crypt_local_t *local = frame->local;
+	struct avec_config *conf = &local->data_conf;
+	dict_t *dict;
+
+	if (op_ret < 0)
+		goto put_one_call;
+
+	if (local->xdata) {
+		dict_unref(local->xdata);
+		local->xdata = NULL;
+	}
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+
+	dict = dict_new();
+	if (!dict) {
+		op_errno = ENOMEM;
+		goto error;
+	}
+
+	update_local_file_params(frame, this, prebuf, postbuf);
+	local->new_file_size = conf->orig_offset;
+
+	/*
+	 * The rest of the file is a partial block and, hence,
+	 * should be written via RMW sequence, so the crypt xlator
+	 * does STACK_WIND to itself.
+	 *
+	 * Pass current file size to crypt_writev()
+	 */
+	op_errno = dict_set(dict,
+			    FSIZE_XATTR_PREFIX,
+			    data_from_uint64(local->cur_file_size));
+	if (op_errno) {
+		gf_log("crypt", GF_LOG_WARNING,
+		       "can not set key to update file size");
+		dict_unref(dict);
+		goto error;
+	}
+	gf_log("crypt", GF_LOG_DEBUG,
+	       "passing current file size (%llu) to crypt_writev",
+	       (unsigned long long)local->cur_file_size);
+	/*
+	 * Padding will be filled with
+	 * zeros by rmw_partial_block()
+	 */
+	STACK_WIND(frame,
+		   prune_complete,
+		   this,
+		   this->fops->writev, /* crypt_writev */
+		   local->fd,
+		   &local->vec,
+		   1,
+		   conf->aligned_offset, /* offset to write from */
+		   0,
+		   local->iobref,
+		   dict);
+
+	dict_unref(dict);
+	return 0;
+ error:
+	local->op_ret = -1;
+	local->op_errno = op_errno;
+ put_one_call:
+	put_one_call_ftruncate(frame, this);
+	return 0;
+}
+
+/*
+ * This is called as a callback of ->writev() invoked in behalf
+ * of ftruncate(): it can be
+ * 1) ordered writes issued by hole conversion in the case of
+ *    expanded truncate, or
+ * 2) an rmw partial data block issued by non-cblock-aligned
+ * prune.
+ */
+int32_t end_writeback_ftruncate(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)
+{
+	crypt_local_t  *local = frame->local;
+	/*
+	 * if nothing has been written,
+	 * then it must be an error
+	 */
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (op_ret < 0)
+		goto put_one_call;
+
+	update_local_file_params(frame, this, prebuf, postbuf);
+
+	if (data_write_in_progress(local))
+		/* case (2) */
+		goto put_one_call;
+	/* case (1) */
+	if (should_resume_submit_hole(local))
+		submit_hole(frame, this);
+	/*
+	 * case of hole, when we should't resume
+	 */
+ put_one_call:
+	put_one_call_ftruncate(frame, this);
+	return 0;
+}
+
+/*
+ * Perform prune and write components of the
+ * read-prune-write sequence.
+ *
+ * Called as ->readv_cbk()
+ *
+ * Pre-conditions:
+ * @vec contains the latest atom of the file
+ * (plain text)
+ */
+static int32_t prune_write(call_frame_t *frame,
+			   void *cookie,
+			   xlator_t *this,
+			   int32_t op_ret,
+			   int32_t op_errno,
+			   struct iovec *vec,
+			   int32_t count,
+			   struct iatt *stbuf,
+			   struct iobref *iobref,
+			   dict_t *xdata)
+{
+	int32_t i;
+	size_t to_copy;
+	size_t copied = 0;
+	crypt_local_t *local = frame->local;
+	struct avec_config *conf = &local->data_conf;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+	if (op_ret == -1)
+		goto put_one_call;
+
+	/*
+	 * At first, uptodate head block
+	 */
+	if (iovec_get_size(vec, count) < conf->off_in_head) {
+		gf_log(this->name, GF_LOG_WARNING,
+		       "Failed to uptodate head block for prune");
+		local->op_ret = -1;
+		local->op_errno = EIO;
+		goto put_one_call;
+	}
+	local->vec.iov_len = conf->off_in_head;
+	local->vec.iov_base = GF_CALLOC(1, local->vec.iov_len,
+					gf_crypt_mt_data);
+
+	if (local->vec.iov_base == NULL) {
+		local->op_ret = -1;
+		local->op_errno = ENOMEM;
+	}
+	for (i = 0; i < count; i++) {
+		to_copy = vec[i].iov_len;
+		if (to_copy > local->vec.iov_len - copied)
+			to_copy = local->vec.iov_len - copied;
+
+		memcpy((char *)local->vec.iov_base + copied,
+		       vec[i].iov_base,
+		       to_copy);
+		copied += to_copy;
+		if (copied == local->vec.iov_len)
+			break;
+	}
+	/*
+	 * perform prune with aligned offset
+	 * (i.e. at this step we prune a bit
+	 * more then it is needed
+	 */
+	STACK_WIND(frame,
+		   prune_submit_file_tail,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->ftruncate,
+		   local->fd,
+		   conf->aligned_offset,
+		   local->xdata);
+	return 0;
+ put_one_call:
+	put_one_call_ftruncate(frame, this);
+	return 0;
+}
+
+/*
+ * Perform a read-prune-write sequence
+ */
+int32_t read_prune_write(call_frame_t *frame, xlator_t *this)
+{
+	int32_t ret = 0;
+	dict_t *dict = NULL;
+	crypt_local_t  *local = frame->local;
+	struct avec_config *conf = &local->data_conf;
+	struct object_cipher_info *object = &local->info->cinfo;
+
+	set_local_io_params_ftruncate(frame, object);
+	get_one_call_nolock(frame);
+
+	if ((conf->orig_offset & (object_alg_blksize(object) - 1)) == 0) {
+		/*
+		 * cblock-aligned prune:
+		 * we don't need read and write components,
+		 * just cut file body
+		 */
+		gf_log("crypt", GF_LOG_DEBUG,
+		       "prune without RMW (at offset %llu",
+		       (unsigned long long)conf->orig_offset);
+
+		STACK_WIND(frame,
+			   prune_complete,
+			   FIRST_CHILD(this),
+			   FIRST_CHILD(this)->fops->ftruncate,
+			   local->fd,
+			   conf->orig_offset,
+			   local->xdata);
+		return 0;
+	}
+	gf_log("crypt", GF_LOG_DEBUG,
+	       "prune with RMW (at offset %llu",
+	       (unsigned long long)conf->orig_offset);	
+	/*
+	 * We are about to perform the "read" component of the
+	 * read-prune-write sequence. It means that we need to
+	 * read encrypted data from disk and decrypt it.
+	 * So, the crypt translator does STACK_WIND to itself.
+	 * 
+	 * Pass current file size to crypt_readv()
+	 
+	 */
+	dict = dict_new();
+	if (!dict) {
+		gf_log("crypt", GF_LOG_WARNING, "Can not alloc dict");
+		ret = ENOMEM;
+		goto exit;
+	}
+	ret = dict_set(dict,
+		       FSIZE_XATTR_PREFIX,
+		       data_from_uint64(local->cur_file_size));
+	if (ret) {
+		gf_log("crypt", GF_LOG_WARNING, "Can not set dict");
+		goto exit;
+	}
+	STACK_WIND(frame,
+		   prune_write,
+		   this,
+		   this->fops->readv, /* crypt_readv */
+		   local->fd,
+		   get_atom_size(object), /* bytes to read */
+		   conf->aligned_offset, /* offset to read from */
+		   0,
+		   dict);
+ exit:
+	if (dict)
+		dict_unref(dict);
+	return ret;
+}
+
+/*
+ * File prune is more complicated than expand.
+ * First we need to read the latest atom to not lose info
+ * needed for proper update. Also we need to make sure that
+ * every component of read-prune-write sequence leaves data
+ * consistent
+ *
+ * Non-cblock aligned prune is performed as read-prune-write
+ * sequence:
+ *
+ * 1) read the latest atom;
+ * 2) perform cblock-aligned prune
+ * 3) issue a write request for the end-of-file
+ */
+int32_t prune_file(call_frame_t *frame, xlator_t *this, uint64_t offset)
+{
+	int32_t ret;
+
+	ret = prepare_for_prune(frame, this, offset);
+	if (ret)
+		return ret;
+	return read_prune_write(frame, this);
+}
+
+int32_t expand_file(call_frame_t *frame, xlator_t *this,
+		    uint64_t offset)
+{
+	int32_t ret;
+	crypt_local_t *local = frame->local;
+
+	ret = prepare_for_submit_hole(frame, this,
+				      local->old_file_size,
+				      offset - local->old_file_size);
+	if (ret)
+		return ret;
+	submit_hole(frame, this);
+	return 0;
+}
+
+static int32_t ftruncate_trivial_completion(call_frame_t *frame,
+					    void *cookie,
+					    xlator_t *this,
+					    int32_t op_ret,
+					    int32_t op_errno,
+					    struct iatt *buf,
+					    dict_t *dict)
+{
+	crypt_local_t *local = frame->local;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+	local->prebuf = *buf;
+	local->postbuf = *buf;
+
+	local->prebuf.ia_size = local->cur_file_size;
+	local->postbuf.ia_size = local->cur_file_size;
+
+	get_one_call(frame);
+	put_one_call_ftruncate(frame, this);
+	return 0;
+}
+
+static int32_t do_ftruncate(call_frame_t *frame,
+			    void *cookie,
+			    xlator_t *this,
+			    int32_t op_ret,
+			    int32_t op_errno,
+			    dict_t *dict,
+			    dict_t *xdata)
+{
+	data_t *data;
+	crypt_local_t *local = frame->local;
+
+	if (op_ret)
+		goto error;	
+	/*
+	 * extract regular file size
+	 */
+	data = dict_get(dict, FSIZE_XATTR_PREFIX);
+	if (!data) {
+		gf_log("crypt", GF_LOG_WARNING, "Regular file size not found");
+		op_errno = EIO;
+		goto error;
+	}
+	local->old_file_size = local->cur_file_size = data_to_uint64(data);
+
+	if (local->data_conf.orig_offset == local->cur_file_size) {
+#if DEBUG_CRYPT
+		gf_log("crypt", GF_LOG_DEBUG,
+		       "trivial ftruncate (current file size %llu)",
+		       (unsigned long long)local->cur_file_size);
+#endif
+		goto trivial;
+	}
+	else if (local->data_conf.orig_offset < local->cur_file_size) {
+#if DEBUG_CRYPT
+		gf_log("crypt", GF_LOG_DEBUG, "prune from %llu to %llu",
+		       (unsigned long long)local->cur_file_size,
+		       (unsigned long long)local->data_conf.orig_offset);
+#endif
+		op_errno = prune_file(frame,
+				      this,
+				      local->data_conf.orig_offset);
+	}
+	else {
+#if DEBUG_CRYPT
+		gf_log("crypt", GF_LOG_DEBUG, "expand from %llu to %llu",
+		       (unsigned long long)local->cur_file_size,
+		       (unsigned long long)local->data_conf.orig_offset);
+#endif
+		op_errno = expand_file(frame,
+				       this,
+				       local->data_conf.orig_offset);
+	}
+	if (op_errno)
+		goto error;
+	return 0;
+ trivial:
+	STACK_WIND(frame,
+		   ftruncate_trivial_completion,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fstat,
+		   local->fd,
+		   NULL);
+	return 0;
+ error:
+	/*
+	 * finish with ftruncate
+	 */
+	local->op_ret = -1;
+	local->op_errno = op_errno;
+
+	get_one_call_nolock(frame);
+	put_one_call_ftruncate(frame, this);
+	return 0;
+}
+
+static int32_t crypt_ftruncate_finodelk_cbk(call_frame_t *frame,
+					    void *cookie,
+					    xlator_t *this,
+					    int32_t op_ret,
+					    int32_t op_errno,
+					    dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (op_ret < 0)
+		goto error;
+	/*
+	 * An access has been granted,
+	 * retrieve file size first
+	 */
+	STACK_WIND(frame,
+		   do_ftruncate,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fgetxattr,
+		   local->fd,
+		   FSIZE_XATTR_PREFIX,
+		   NULL);
+	return 0;
+ error:
+	get_one_call_nolock(frame);
+	put_one_call_ftruncate(frame, this);
+	return 0;
+}
+
+/*
+ * ftruncate is performed in 2 steps:
+ * . recieve file size;
+ * . expand or prune file.
+ */
+static int32_t crypt_ftruncate(call_frame_t *frame,
+			       xlator_t *this,
+			       fd_t *fd,
+			       off_t offset,
+			       dict_t *xdata)
+{
+	int32_t ret;
+	crypt_local_t *local;
+	struct crypt_inode_info *info;
+	struct gf_flock lock = {0, };
+
+	local = crypt_alloc_local(frame, this, GF_FOP_FTRUNCATE);
+	if (!local) {
+		ret = ENOMEM;
+		goto error;
+	}
+	local->xattr = dict_new();
+	if (!local->xattr) {
+		ret = ENOMEM;
+		goto error;
+	}
+	local->fd = fd_ref(fd);
+	info = local_get_inode_info(local, this);
+	if (info == NULL) {
+		ret = EINVAL;
+		goto error;
+	}
+	if (!object_alg_atomic(&info->cinfo)) {
+		ret = EINVAL;
+		goto error;
+	}
+	local->data_conf.orig_offset = offset;
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+
+	lock.l_len    = 0;
+        lock.l_start  = 0;
+        lock.l_type   = F_WRLCK;
+        lock.l_whence = SEEK_SET;
+
+	STACK_WIND(frame,
+		   crypt_ftruncate_finodelk_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->finodelk,
+		   this->name,
+		   fd,
+		   F_SETLKW,
+		   &lock,
+		   NULL);
+	return 0;
+ error:
+	if (local && local->fd)
+		fd_unref(fd);
+	if (local && local->xdata)
+		dict_unref(xdata);
+	if (local && local->xattr)
+		dict_unref(local->xattr);
+	if (local && local->info)
+		free_inode_info(local->info);
+
+	STACK_UNWIND_STRICT(ftruncate, frame, -1, ret, NULL, NULL, NULL);
+	return 0;
+}
+
+/* ->flush_cbk() */
+int32_t truncate_end(call_frame_t *frame,
+		     void *cookie,
+		     xlator_t *this,
+		     int32_t op_ret,
+		     int32_t op_errno,
+		     dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	STACK_UNWIND_STRICT(truncate,
+			    frame,
+			    op_ret,
+			    op_errno,
+			    &local->prebuf,
+			    &local->postbuf,
+			    local->xdata);
+	return 0;
+}
+
+/* ftruncate_cbk() */
+int32_t truncate_flush(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)
+{
+	crypt_local_t *local = frame->local;
+	fd_t *fd = local->fd;
+	local->prebuf = *prebuf;
+	local->postbuf = *postbuf;
+
+	STACK_WIND(frame,
+		   truncate_end,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->flush,
+		   fd,
+		   NULL);
+	fd_unref(fd);
+	return 0;
+}
+
+/*
+ * is called as ->open_cbk()
+ */
+static int32_t truncate_begin(call_frame_t *frame,
+			      void *cookie,
+			      xlator_t *this,
+			      int32_t op_ret,
+			      int32_t op_errno,
+			      fd_t *fd,
+			      dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	if (op_ret < 0) {
+		fd_unref(fd);
+		STACK_UNWIND_STRICT(truncate,
+				    frame,
+				    op_ret,
+				    op_errno, NULL, NULL, NULL);
+		return 0;
+	}
+	/*
+	 * crypt_truncate() is implemented via crypt_ftruncate(),
+	 * so the crypt xlator does STACK_WIND to itself here
+	 */
+	STACK_WIND(frame,
+		   truncate_flush,
+		   this,
+		   this->fops->ftruncate, /* crypt_ftruncate */
+		   fd,
+		   local->offset,
+		   NULL);
+	return 0;
+}
+
+/*
+ * crypt_truncate() is implemented via crypt_ftruncate() as a
+ * sequence crypt_open() - crypt_ftruncate() - truncate_flush()
+ */
+int32_t crypt_truncate(call_frame_t *frame,
+		       xlator_t *this,
+		       loc_t *loc,
+		       off_t offset,
+		       dict_t *xdata)
+{
+	fd_t *fd;
+	crypt_local_t *local;
+
+#if DEBUG_CRYPT
+	gf_log(this->name, GF_LOG_DEBUG,
+	       "truncate file %s at offset %llu",
+	       loc->path, (unsigned long long)offset);
+#endif
+	local = crypt_alloc_local(frame, this, GF_FOP_TRUNCATE);
+	if (!local)
+		goto error;
+
+	fd = fd_create(loc->inode, frame->root->pid);
+	if (!fd) {
+		gf_log(this->name, GF_LOG_ERROR, "Can not create fd");
+		goto error;
+	}
+	local->fd = fd;
+	local->offset = offset;
+	local->xdata = xdata;
+	STACK_WIND(frame,
+		   truncate_begin,
+		   this,
+		   this->fops->open, /* crypt_open() */
+		   loc,
+		   O_RDWR,
+		   fd,
+		   NULL);
+	return 0;
+ error:
+	STACK_UNWIND_STRICT(truncate, frame, -1, EINVAL, NULL, NULL, NULL);
+	return 0;
+}
+
+end_writeback_handler_t dispatch_end_writeback(glusterfs_fop_t fop)
+{
+	switch (fop) {
+	case GF_FOP_WRITE:
+		return end_writeback_writev;
+	case GF_FOP_FTRUNCATE:
+		return end_writeback_ftruncate;
+	default:
+		gf_log("crypt", GF_LOG_WARNING, "Bad wb operation %d", fop);
+		return NULL;
+	}
+}
+
+/*
+ * true, if the caller needs metadata string
+ */
+static int32_t is_custom_mtd(dict_t *xdata)
+{
+	data_t *data;
+	uint32_t flags;
+
+	if (!xdata)
+		return 0;
+
+	data = dict_get(xdata, MSGFLAGS_PREFIX);
+	if (!data)
+		return 0;
+	if (data->len != sizeof(uint32_t)) {
+		gf_log("crypt", GF_LOG_WARNING,
+		       "Bad msgflags size (%d)", data->len);
+		return -1;
+	}
+	flags = *((uint32_t *)data->data);
+	return msgflags_check_mtd_lock(&flags);
+}
+
+static int32_t crypt_open_done(call_frame_t *frame,
+			       void *cookie,
+			       xlator_t *this,
+			       int32_t op_ret,
+			       int32_t op_errno, dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+	if (op_ret < 0)
+		gf_log(this->name, GF_LOG_WARNING, "mtd unlock failed (%d)",
+		       op_errno);
+	put_one_call_open(frame);
+	return 0;
+}
+
+static void crypt_open_tail(call_frame_t *frame, xlator_t *this)
+{
+	struct gf_flock  lock  = {0, };
+	crypt_local_t *local = frame->local;
+
+        lock.l_type   = F_UNLCK;
+        lock.l_whence = SEEK_SET;
+        lock.l_start  = 0;
+        lock.l_len    = 0;
+        lock.l_pid    = 0;
+
+	STACK_WIND(frame,
+		   crypt_open_done,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->finodelk,
+		   this->name,
+		   local->fd,
+		   F_SETLKW,
+		   &lock,
+		   NULL);
+}
+
+/*
+ * load private inode info at open time
+ * called as ->fgetxattr_cbk()
+ */
+static int load_mtd_open(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 ret;
+	gf_boolean_t upload_info;
+	data_t *mtd;
+	uint64_t value = 0;
+	struct crypt_inode_info *info;
+	crypt_local_t *local = frame->local;
+	crypt_private_t *priv = this->private;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (local->fd->inode->ia_type == IA_IFLNK)
+		goto exit;
+	if (op_ret < 0)
+		goto exit;
+	/*
+	 * first, check for cached info
+	 */
+	ret = inode_ctx_get(local->fd->inode, this, &value);
+	if (ret != -1) {
+		info = (struct crypt_inode_info *)(long)value;
+		if (info == NULL) {
+			gf_log(this->name, GF_LOG_WARNING,
+			       "Inode info expected, but not found");
+			local->op_ret = -1;
+			local->op_errno = EIO;
+			goto exit;
+		}
+		/*
+		 * info has been found in the cache
+		 */
+		upload_info = _gf_false;
+	}
+	else {
+		/*
+		 * info hasn't been found in the cache.
+		 */
+		info = alloc_inode_info(local, local->loc);
+		if (!info) {
+			local->op_ret = -1;
+			local->op_errno = ENOMEM;
+			goto exit;
+		}
+		init_inode_info_head(info, local->fd);
+		upload_info = _gf_true;
+	}
+	/*
+	 * extract metadata
+	 */
+	mtd = dict_get(dict, CRYPTO_FORMAT_PREFIX);
+	if (!mtd) {
+		local->op_ret = -1;
+		local->op_errno = ENOENT;
+		gf_log (this->name, GF_LOG_WARNING,
+			"Format string wasn't found");
+		goto exit;
+	}
+	/*
+	 * authenticate metadata against the path
+	 */	
+	ret = open_format((unsigned char *)mtd->data,
+			  mtd->len,
+			  local->loc,
+			  info,
+			  get_master_cinfo(priv),
+			  local,
+			  upload_info);
+	if (ret) {
+		local->op_ret = -1;
+		local->op_errno = ret;
+		goto exit;
+	}
+	if (upload_info) {
+		ret = init_inode_info_tail(info, get_master_cinfo(priv));
+		if (ret) {
+			local->op_ret = -1;
+			local->op_errno = ret;
+			goto exit;	
+		}
+		ret = inode_ctx_put(local->fd->inode,
+				    this, (uint64_t)(long)info);
+		if (ret == -1) {
+			local->op_ret = -1;
+			local->op_errno = EIO;
+			goto exit;
+		}
+	}
+	if (local->custom_mtd) {
+		/*
+		 * pass the metadata string to the customer
+		 */
+		ret = dict_set_static_bin(local->xdata,
+					  CRYPTO_FORMAT_PREFIX,
+					  mtd->data,
+					  mtd->len);
+		if (ret) {
+			local->op_ret = -1;
+			local->op_errno = ret;
+			goto exit;
+		}
+	}
+ exit:
+	if (!local->custom_mtd)
+		crypt_open_tail(frame, this);
+	else
+		put_one_call_open(frame);
+	return 0;
+}
+
+static int32_t crypt_open_finodelk_cbk(call_frame_t *frame,
+				       void *cookie,
+				       xlator_t *this,
+				       int32_t op_ret,
+				       int32_t op_errno,
+				       dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (op_ret < 0) {
+		gf_log(this->name, GF_LOG_WARNING, "finodelk (LOCK) failed");
+		goto exit;
+	}
+	STACK_WIND(frame,
+		   load_mtd_open,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fgetxattr,
+		   local->fd,
+		   CRYPTO_FORMAT_PREFIX,
+		   NULL);
+	return 0;
+ exit:
+	put_one_call_open(frame);
+	return 0;
+}
+
+/*
+ * verify metadata against the specified pathname
+ */
+static int32_t crypt_open_cbk(call_frame_t *frame,
+			      void *cookie,
+			      xlator_t *this,
+			      int32_t op_ret,
+			      int32_t op_errno,
+			      fd_t *fd,
+			      dict_t *xdata)
+{
+	struct gf_flock lock = {0, };
+	crypt_local_t *local = frame->local;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (local->fd->inode->ia_type == IA_IFLNK)
+		goto exit;
+	if (op_ret < 0)
+		goto exit;
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+	else if (local->custom_mtd){
+		local->xdata = dict_new();
+		if (!local->xdata) {
+			local->op_ret = -1;
+			local->op_errno = ENOMEM;
+			gf_log ("crypt", GF_LOG_ERROR,
+				"Can not get new dict for mtd string");
+			goto exit;
+		}
+	}
+	lock.l_len    = 0;
+        lock.l_start  = 0;
+        lock.l_type   = local->custom_mtd ? F_WRLCK : F_RDLCK;
+        lock.l_whence = SEEK_SET;
+
+	STACK_WIND(frame,
+		   crypt_open_finodelk_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->finodelk,
+		   this->name,
+		   fd,
+		   F_SETLKW,
+		   &lock,
+		   NULL);
+	return 0;
+ exit:
+	put_one_call_open(frame);
+	return 0;
+}
+
+static int32_t crypt_open(call_frame_t *frame,
+			  xlator_t *this,
+			  loc_t *loc,
+			  int32_t flags,
+			  fd_t *fd,
+			  dict_t *xdata)
+{
+	int32_t ret = ENOMEM;
+	crypt_local_t *local;
+
+	local = crypt_alloc_local(frame, this, GF_FOP_OPEN);
+	if (!local)
+		goto error;
+	local->loc = GF_CALLOC(1, sizeof(*loc), gf_crypt_mt_loc);
+	if (!local->loc) {
+		ret = ENOMEM;
+		goto error;
+	}
+	memset(local->loc, 0, sizeof(*local->loc));
+	ret = loc_copy(local->loc, loc);
+	if (ret) {
+		GF_FREE(local->loc);
+		goto error;
+	}
+	local->fd = fd_ref(fd);
+
+	ret = is_custom_mtd(xdata);
+	if (ret < 0) {
+		loc_wipe(local->loc);
+		GF_FREE(local->loc);
+		ret = EINVAL;
+		goto error;
+	}
+	local->custom_mtd = ret;
+
+	if ((flags & O_ACCMODE) == O_WRONLY)
+		/*
+		 * we can't open O_WRONLY, because
+		 * we need to do read-modify-write
+		 */
+		flags = (flags & ~O_ACCMODE) | O_RDWR;
+	/*
+	 * Make sure that out translated offsets
+	 * and counts won't be ignored
+	 */
+	flags &= ~O_APPEND;
+	get_one_call_nolock(frame);
+	STACK_WIND(frame,
+		   crypt_open_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->open,
+		   loc,
+		   flags,
+		   fd,
+		   xdata);
+	return 0;
+ error:
+	STACK_UNWIND_STRICT(open,
+			    frame,
+			    -1,
+			    ret,
+			    NULL,
+			    NULL);
+	return 0;
+}
+
+static int32_t init_inode_info_tail(struct crypt_inode_info *info,
+				    struct master_cipher_info *master)
+{
+	int32_t ret;
+	struct object_cipher_info *object = &info->cinfo;
+
+#if DEBUG_CRYPT
+	gf_log("crypt", GF_LOG_DEBUG, "Init inode info for object %s",
+	       uuid_utoa(info->oid));
+#endif
+	ret = data_cipher_algs[object->o_alg][object->o_mode].set_private(info,
+									master);
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR, "Set private info failed");
+		return ret;
+	}
+	return 0;
+}
+
+/*
+ * Init inode info at ->create() time
+ */
+static void init_inode_info_create(struct crypt_inode_info *info,
+				   struct master_cipher_info *master,
+				   data_t *data)
+{
+	struct object_cipher_info *object;
+
+	info->nr_minor = CRYPT_XLATOR_ID;
+	memcpy(info->oid, data->data, data->len);
+
+	object = &info->cinfo;
+
+	object->o_alg        = master->m_alg;
+	object->o_mode       = master->m_mode;
+	object->o_block_bits = master->m_block_bits;
+	object->o_dkey_size  = master->m_dkey_size;
+}
+
+static void init_inode_info_head(struct crypt_inode_info *info, fd_t *fd)
+{
+	memcpy(info->oid, fd->inode->gfid, sizeof(uuid_t));
+}
+
+static int32_t crypt_create_done(call_frame_t *frame,
+				 void *cookie,
+				 xlator_t *this,
+				 int32_t op_ret,
+				 int32_t op_errno, dict_t *xdata)
+{
+	crypt_private_t *priv = this->private;
+	crypt_local_t *local = frame->local;
+	struct crypt_inode_info *info = local->info;
+	fd_t *local_fd = local->fd;
+	dict_t *local_xdata = local->xdata;
+	inode_t *local_inode = local->inode;
+
+	if (op_ret < 0) {
+		free_inode_info(info);
+		goto unwind;
+	}
+	op_errno = init_inode_info_tail(info, get_master_cinfo(priv));
+	if (op_errno) {
+		op_ret = -1;
+		free_inode_info(info);
+		goto unwind;
+	}
+	/*
+	 * FIXME: drop major subversion number
+	 */
+	op_ret = inode_ctx_put(local->fd->inode, this, (uint64_t)(long)info);
+	if (op_ret == -1) {
+		op_errno = EIO;
+		free_inode_info(info);
+		goto unwind;
+	}
+ unwind:
+	free_format(local);
+	STACK_UNWIND_STRICT(create,
+			    frame,
+			    op_ret,
+			    op_errno,
+			    local_fd,
+			    local_inode,
+			    &local->buf,
+			    &local->prebuf,
+			    &local->postbuf,
+			    local_xdata);
+	fd_unref(local_fd);
+	inode_unref(local_inode);
+	if (local_xdata)
+		dict_unref(local_xdata);
+	return 0;
+}
+
+static int crypt_create_tail(call_frame_t *frame,
+			     void *cookie,
+			     xlator_t *this,
+			     int32_t op_ret,
+			     int32_t op_errno,
+			     dict_t *xdata)
+{
+	struct gf_flock  lock  = {0, };
+	crypt_local_t *local = frame->local;
+	fd_t *local_fd = local->fd;
+	dict_t *local_xdata = local->xdata;
+	inode_t *local_inode = local->inode;
+
+	dict_unref(local->xattr);
+
+	if (op_ret < 0)
+		goto error;
+
+        lock.l_type   = F_UNLCK;
+        lock.l_whence = SEEK_SET;
+        lock.l_start  = 0;
+        lock.l_len    = 0;
+        lock.l_pid    = 0;
+
+	STACK_WIND(frame,
+		   crypt_create_done,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->finodelk,
+		   this->name,
+		   local->fd,
+		   F_SETLKW,
+		   &lock,
+		   NULL);
+	return 0;
+ error:
+	free_inode_info(local->info);
+	free_format(local);
+
+	STACK_UNWIND_STRICT(create,
+			    frame,
+			    op_ret,
+			    op_errno,
+			    local_fd,
+			    local_inode,
+			    &local->buf,
+			    &local->prebuf,
+			    &local->postbuf,
+			    local_xdata);
+
+	fd_unref(local_fd);
+	inode_unref(local_inode);
+	if (local_xdata)
+		dict_unref(local_xdata);
+	return 0;
+}
+
+static int32_t crypt_create_finodelk_cbk(call_frame_t *frame,
+					 void *cookie,
+					 xlator_t *this,
+					 int32_t op_ret,
+					 int32_t op_errno,
+					 dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+	struct crypt_inode_info *info = local->info;
+
+	if (op_ret < 0)
+		goto error;
+
+	STACK_WIND(frame,
+		   crypt_create_tail,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fsetxattr,
+		   local->fd,
+		   local->xattr, /* CRYPTO_FORMAT_PREFIX */
+		   0,
+		   NULL);
+	return 0;
+ error: 
+	free_inode_info(info);
+	free_format(local);
+	fd_unref(local->fd);
+	dict_unref(local->xattr);
+	if (local->xdata)
+		dict_unref(local->xdata);
+
+	STACK_UNWIND_STRICT(create,
+			    frame,
+			    op_ret,
+			    op_errno,
+			    NULL,
+			    NULL,
+			    NULL,
+			    NULL,
+			    NULL,
+			    NULL);
+	return 0;
+}
+
+/*
+ * Create and store crypt-specific format on disk;
+ * Populate cache with private inode info
+ */
+static int32_t crypt_create_cbk(call_frame_t *frame,
+				void *cookie,
+				xlator_t *this,
+				int32_t op_ret,
+				int32_t op_errno,
+				fd_t *fd,
+				inode_t *inode,
+				struct iatt *buf,
+				struct iatt *preparent,
+				struct iatt *postparent,
+				dict_t *xdata)
+{
+	struct gf_flock lock = {0, };
+	crypt_local_t *local = frame->local;
+	struct crypt_inode_info *info = local->info;
+
+	if (op_ret < 0)
+		goto error;
+	if (xdata)
+		local->xdata = dict_ref(xdata);	
+	local->inode = inode_ref(inode);
+	local->buf = *buf;
+	local->prebuf = *preparent;
+	local->postbuf = *postparent;
+
+        lock.l_len    = 0;
+        lock.l_start  = 0;
+        lock.l_type   = F_WRLCK;
+        lock.l_whence = SEEK_SET;
+
+        STACK_WIND(frame,
+		   crypt_create_finodelk_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->finodelk,
+		   this->name,
+		   local->fd,
+		   F_SETLKW,
+		   &lock,
+		   NULL);	
+	return 0;
+ error:
+	free_inode_info(info);
+	free_format(local);
+	fd_unref(local->fd);
+	dict_unref(local->xattr);
+
+	STACK_UNWIND_STRICT(create,
+			    frame,
+			    op_ret,
+			    op_errno,
+			    NULL, NULL, NULL,
+			    NULL, NULL, NULL);
+	return 0;
+}
+
+static int32_t crypt_create(call_frame_t *frame,
+			    xlator_t *this,
+			    loc_t *loc,
+			    int32_t flags,
+			    mode_t mode,
+			    mode_t umask,
+			    fd_t *fd,
+			    dict_t *xdata)
+{
+	int ret;
+	data_t *data;
+	crypt_local_t *local;
+	crypt_private_t *priv;
+	struct master_cipher_info *master;
+	struct crypt_inode_info *info;
+
+	priv = this->private;
+	master = get_master_cinfo(priv);
+
+	if (master_alg_atomic(master)) {
+		/*
+		 * We can't open O_WRONLY, because we
+		 * need to do read-modify-write.
+		 */
+		if ((flags & O_ACCMODE) == O_WRONLY)
+			flags = (flags & ~O_ACCMODE) | O_RDWR;
+		/*
+		 * Make sure that out translated offsets
+		 * and counts won't be ignored
+		 */
+		flags &= ~O_APPEND;
+	}
+	local = crypt_alloc_local(frame, this, GF_FOP_CREATE);
+	if (!local) {
+		ret = ENOMEM;
+		goto error;
+	}
+	data = dict_get(xdata, "gfid-req");
+	if (!data) {
+		ret = EINVAL;
+		gf_log("crypt", GF_LOG_WARNING, "gfid not found");
+		goto error;
+	}
+	if (data->len != sizeof(uuid_t)) {
+		ret = EINVAL;
+		gf_log("crypt", GF_LOG_WARNING,
+		       "bad gfid size (%d), should be %d",
+		       (int)data->len, (int)sizeof(uuid_t));
+		goto error;
+	}
+	info = alloc_inode_info(local, loc);
+	if (!info){
+		ret = ENOMEM;
+		goto error;
+	}
+	/*
+	 * NOTE:
+	 * format has to be created BEFORE
+	 * proceeding to the untrusted server
+	 */
+	ret = alloc_format_create(local);
+	if (ret) {
+		free_inode_info(info);
+		goto error;
+	}
+	init_inode_info_create(info, master, data);
+
+	ret = create_format(local->format,
+			    loc,
+			    info,
+			    master);
+	if (ret) {
+		free_inode_info(info);
+		goto error;		
+	}
+	local->xattr = dict_new();
+	if (!local->xattr) {
+		free_inode_info(info);
+		free_format(local);	
+		goto error;
+	}
+	ret = dict_set_static_bin(local->xattr,
+				  CRYPTO_FORMAT_PREFIX,
+				  local->format,
+				  new_format_size());
+	if (ret) {
+		dict_unref(local->xattr);
+		free_inode_info(info);
+		free_format(local);
+		goto error;
+	}
+	ret = dict_set(local->xattr, FSIZE_XATTR_PREFIX, data_from_uint64(0));
+	if (ret) {
+		dict_unref(local->xattr);
+		free_inode_info(info);
+		free_format(local);
+		goto error;
+	}
+	local->fd = fd_ref(fd);
+
+	STACK_WIND(frame,
+		   crypt_create_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->create,
+		   loc,
+		   flags,
+		   mode,
+		   umask,
+		   fd,
+		   xdata);
+	return 0;
+ error:
+	gf_log("crypt", GF_LOG_WARNING, "can not create file");
+	STACK_UNWIND_STRICT(create,
+			    frame,
+			    -1,
+			    ret,
+			    NULL, NULL, NULL,
+			    NULL, NULL, NULL);
+	return 0;
+}
+
+/*
+ * FIXME: this should depends on the version of format string
+ */
+static int32_t filter_crypt_xattr(dict_t *dict,
+				  char *key, data_t *value, void *data)
+{
+        dict_del(dict, key);
+        return 0;
+}
+
+static int32_t crypt_fsetxattr(call_frame_t *frame,
+			       xlator_t *this,
+			       fd_t *fd,
+			       dict_t *dict,
+			       int32_t flags, dict_t *xdata)
+{
+	dict_foreach_fnmatch(dict, "trusted.glusterfs.crypt*",
+			     filter_crypt_xattr, NULL);
+	STACK_WIND(frame,
+		   default_fsetxattr_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fsetxattr,
+		   fd,
+		   dict,
+		   flags,
+		   xdata);
+	return 0;
+}
+
+/*
+ * TBD: verify file metadata before wind
+ */
+static int32_t crypt_setxattr(call_frame_t *frame,
+			      xlator_t *this,
+			      loc_t *loc,
+			      dict_t *dict,
+			      int32_t flags, dict_t *xdata)
+{
+	dict_foreach_fnmatch(dict, "trusted.glusterfs.crypt*",
+			     filter_crypt_xattr, NULL);
+	STACK_WIND(frame,
+		   default_setxattr_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->setxattr,
+		   loc,
+		   dict,
+		   flags,
+		   xdata);
+	return 0;
+}
+
+/*
+ * called as flush_cbk()
+ */
+static int32_t linkop_end(call_frame_t *frame,
+			  void *cookie,
+			  xlator_t *this,
+			  int32_t op_ret,
+			  int32_t op_errno,
+			  dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+	linkop_unwind_handler_t unwind_fn;
+	unwind_fn = linkop_unwind_dispatch(local->fop);
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (op_ret < 0 &&
+	    op_errno == ENOENT &&
+	    local->loc->inode->ia_type == IA_IFLNK) {
+		local->op_ret = 0;
+		local->op_errno = 0;
+	}
+	unwind_fn(frame);
+	return 0;
+}
+
+/*
+ * unpin inode on the server
+ */
+static int32_t link_flush(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)
+{
+	crypt_local_t *local = frame->local;
+
+	if (op_ret < 0)
+		goto error;
+	if (local->xdata) {
+		dict_unref(local->xdata);
+		local->xdata = NULL;
+	}
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+	local->inode = inode_ref(inode);
+	local->buf = *buf;
+	local->prebuf = *preparent;
+	local->postbuf = *postparent;
+
+	STACK_WIND(frame,
+		   linkop_end,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->flush,
+		   local->fd,
+		   NULL);
+	return 0;
+ error:
+	local->op_ret = -1;
+	local->op_errno = op_errno;
+	link_unwind(frame);
+	return 0;
+}
+
+void link_unwind(call_frame_t *frame)
+{
+	crypt_local_t *local = frame->local;
+	dict_t *xdata;
+	dict_t *xattr;
+	inode_t *inode;
+
+	if (!local) {
+		STACK_UNWIND_STRICT(link,
+				    frame,
+				    -1,
+				    ENOMEM,
+				    NULL,
+				    NULL,
+				    NULL,
+				    NULL,
+				    NULL);
+		return;
+	}
+	xdata = local->xdata;
+	xattr = local->xattr;
+	inode = local->inode;
+
+	if (local->loc){
+		loc_wipe(local->loc);
+		GF_FREE(local->loc);
+	}
+	if (local->newloc) {
+		loc_wipe(local->newloc);
+		GF_FREE(local->newloc);
+	}
+	if (local->fd)
+		fd_unref(local->fd);
+	if (local->format)
+		GF_FREE(local->format);
+
+	STACK_UNWIND_STRICT(link,
+			    frame,
+			    local->op_ret,
+			    local->op_errno,
+			    inode,
+			    &local->buf,
+			    &local->prebuf,
+			    &local->postbuf,
+			    xdata);
+	if (xdata)
+		dict_unref(xdata);
+	if (xattr)
+		dict_unref(xattr);
+	if (inode)
+		inode_unref(inode);
+}
+
+void link_wind(call_frame_t *frame, xlator_t *this)
+{
+	crypt_local_t *local = frame->local;
+
+	STACK_WIND(frame,
+		   link_flush,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->link,
+		   local->loc,
+		   local->newloc,
+		   local->xdata);
+}
+
+/*
+ * unlink()
+ */
+static int32_t unlink_flush(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)
+{
+	crypt_local_t *local = frame->local;
+
+	if (op_ret < 0)
+		goto error;
+	local->prebuf = *preparent;
+	local->postbuf = *postparent;
+	if (local->xdata) {
+		dict_unref(local->xdata);
+		local->xdata = NULL;
+	}
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+
+	STACK_WIND(frame,
+		   linkop_end,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->flush,
+		   local->fd,
+		   NULL);
+	return 0;
+ error:
+	local->op_ret = -1;
+	local->op_errno = op_errno;
+	unlink_unwind(frame);
+	return 0;
+}
+
+void unlink_unwind(call_frame_t *frame)
+{
+	crypt_local_t *local = frame->local;
+	dict_t *xdata;
+	dict_t *xattr;
+
+	if (!local) {
+		STACK_UNWIND_STRICT(unlink,
+				    frame,
+				    -1,
+				    ENOMEM,
+				    NULL,
+				    NULL,
+				    NULL);
+		return;
+	}
+	xdata = local->xdata;
+	xattr = local->xattr;
+	if (local->loc){
+		loc_wipe(local->loc);
+		GF_FREE(local->loc);
+	}
+	if (local->fd)
+		fd_unref(local->fd);
+	if (local->format)
+		GF_FREE(local->format);
+
+	STACK_UNWIND_STRICT(unlink,
+			    frame,
+			    local->op_ret,
+			    local->op_errno,
+			    &local->prebuf,
+			    &local->postbuf,
+			    xdata);
+	if (xdata)
+		dict_unref(xdata);
+	if (xattr)
+		dict_unref(xattr);
+}
+
+void unlink_wind(call_frame_t *frame, xlator_t *this)
+{
+	crypt_local_t *local = frame->local;
+
+	STACK_WIND(frame,
+		   unlink_flush,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->unlink,
+		   local->loc,
+		   local->flags,
+		   local->xdata);
+}
+
+void rename_unwind(call_frame_t *frame)
+{
+	crypt_local_t *local = frame->local;
+	dict_t *xdata;
+	dict_t *xattr;
+	struct iatt *prenewparent;
+	struct iatt *postnewparent;
+
+	if (!local) {
+		STACK_UNWIND_STRICT(rename,
+				    frame,
+				    -1,
+				    ENOMEM,
+				    NULL,
+				    NULL,
+				    NULL,
+				    NULL,
+				    NULL,
+				    NULL);	
+		return;
+	}
+	xdata = local->xdata;
+	xattr = local->xattr;
+	prenewparent = local->prenewparent;
+	postnewparent = local->postnewparent;	
+
+	if (local->loc){
+		loc_wipe(local->loc);
+		GF_FREE(local->loc);
+	}
+	if (local->newloc){
+		loc_wipe(local->newloc);
+		GF_FREE(local->newloc);
+	}
+	if (local->fd)
+		fd_unref(local->fd);
+	if (local->format)
+		GF_FREE(local->format);
+
+	STACK_UNWIND_STRICT(rename,
+			    frame,
+			    local->op_ret,
+			    local->op_errno,
+			    &local->buf,
+			    &local->prebuf,
+			    &local->postbuf,
+			    prenewparent,
+			    postnewparent,
+			    xdata);
+	if (xdata)
+		dict_unref(xdata);
+	if (xattr)
+		dict_unref(xattr);
+	if (prenewparent)
+		GF_FREE(prenewparent);
+	if (postnewparent)
+		GF_FREE(postnewparent);
+}
+
+/*
+ * called as flush_cbk()
+ */
+static int32_t rename_end(call_frame_t *frame,
+			  void *cookie,
+			  xlator_t *this,
+			  int32_t op_ret,
+			  int32_t op_errno,
+			  dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	rename_unwind(frame);
+	return 0;
+}
+
+static int32_t rename_flush(call_frame_t *frame,
+			    void *cookie,
+			    xlator_t *this,
+			    int32_t op_ret,
+			    int32_t op_errno,
+			    struct iatt *buf,
+			    struct iatt *preoldparent,
+			    struct iatt *postoldparent,
+			    struct iatt *prenewparent,
+			    struct iatt *postnewparent,
+			    dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	if (op_ret < 0)
+		goto error;
+	dict_unref(local->xdata);
+	local->xdata = NULL;
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+
+	local->buf = *buf;
+	local->prebuf = *preoldparent;
+	local->postbuf = *postoldparent;
+	if (prenewparent) {
+		local->prenewparent = GF_CALLOC(1, sizeof(*prenewparent),
+						gf_crypt_mt_iatt);
+		if (!local->prenewparent) {
+			op_errno = ENOMEM;
+			goto error;
+		}
+		*local->prenewparent = *prenewparent;
+	}
+	if (postnewparent) {
+		local->postnewparent = GF_CALLOC(1, sizeof(*postnewparent),
+						 gf_crypt_mt_iatt);
+		if (!local->postnewparent) {
+			op_errno = ENOMEM;
+			goto error;
+		}
+		*local->postnewparent = *postnewparent;
+	}
+	STACK_WIND(frame,
+		   rename_end,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->flush,
+		   local->fd,
+		   NULL);
+	return 0;
+ error:
+	local->op_ret = -1;
+	local->op_errno = op_errno;
+	rename_unwind(frame);
+	return 0;
+}
+
+void rename_wind(call_frame_t *frame, xlator_t *this)
+{
+	crypt_local_t *local = frame->local;
+
+	STACK_WIND(frame,
+		   rename_flush,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->rename,
+		   local->loc,
+		   local->newloc,
+		   local->xdata);
+}
+
+static int32_t __do_linkop(call_frame_t *frame,
+			   void *cookie,
+			   xlator_t *this,
+			   int32_t op_ret,
+			   int32_t op_errno, dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+	linkop_wind_handler_t wind_fn;
+	linkop_unwind_handler_t unwind_fn;
+
+	wind_fn = linkop_wind_dispatch(local->fop);
+	unwind_fn = linkop_unwind_dispatch(local->fop);
+
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if (op_ret >= 0)
+		wind_fn(frame, this);
+	else {
+		gf_log(this->name, GF_LOG_WARNING, "mtd unlock failed (%d)",
+		       op_errno);
+		unwind_fn(frame);
+	}
+	return 0;
+}
+
+static int32_t do_linkop(call_frame_t *frame,
+			 void *cookie,
+			 xlator_t *this,
+			 int32_t op_ret,
+			 int32_t op_errno,
+			 dict_t *xdata)
+{
+	struct gf_flock  lock  = {0, };
+	crypt_local_t *local = frame->local;
+	linkop_unwind_handler_t unwind_fn;
+
+	unwind_fn = linkop_unwind_dispatch(local->fop);
+	local->op_ret = op_ret;
+	local->op_errno = op_errno;
+
+	if(op_ret < 0)
+		goto error;
+
+	lock.l_type   = F_UNLCK;
+        lock.l_whence = SEEK_SET;
+        lock.l_start  = 0;
+        lock.l_len    = 0;
+        lock.l_pid    = 0;
+
+	STACK_WIND(frame,
+		   __do_linkop,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->finodelk,
+		   this->name,
+		   local->fd,
+		   F_SETLKW,
+		   &lock,
+		   NULL);
+	return 0;
+ error:		
+	unwind_fn(frame);
+	return 0;
+}
+
+/*
+ * Update the metadata string (against the new pathname);
+ * submit the result
+ */
+static int32_t linkop_begin(call_frame_t *frame,
+			    void *cookie,
+			    xlator_t *this,
+			    int32_t op_ret,
+			    int32_t op_errno,
+			    fd_t *fd,
+			    dict_t *xdata)
+{
+	gf_boolean_t upload_info;
+	crypt_local_t *local = frame->local;
+	crypt_private_t *priv = this->private;
+	struct crypt_inode_info *info;
+	data_t *old_mtd;
+	uint32_t new_mtd_size;
+	uint64_t value = 0;
+	void (*unwind_fn)(call_frame_t *frame);
+	void (*wind_fn)(call_frame_t *frame, xlator_t *this);
+	mtd_op_t mop;
+
+	wind_fn = linkop_wind_dispatch(local->fop);
+	unwind_fn = linkop_unwind_dispatch(local->fop);
+	mop = linkop_mtdop_dispatch(local->fop);
+
+	if (local->fd->inode->ia_type == IA_IFLNK)
+		goto wind;
+	if (op_ret < 0)
+		/*
+		 * verification failed
+		 */
+		goto error;
+
+	old_mtd = dict_get(xdata, CRYPTO_FORMAT_PREFIX);
+	if (!old_mtd) {
+		op_errno = EIO;
+		gf_log (this->name, GF_LOG_DEBUG,
+			"Metadata string wasn't found");
+		goto error;
+	}
+	new_mtd_size = format_size(mop, old_mtd->len);
+	op_errno = alloc_format(local, new_mtd_size);
+	if (op_errno)
+		goto error;
+	/*
+	 * check for cached info 
+	 */
+	op_ret = inode_ctx_get(fd->inode, this, &value);
+	if (op_ret != -1) {
+		info = (struct crypt_inode_info *)(long)value;
+		if (info == NULL) {
+			gf_log (this->name, GF_LOG_WARNING,
+				"Inode info was not found");
+			op_errno = EINVAL;
+			goto error;
+		}
+		/*
+		 * info was found in the cache
+		 */
+		local->info = info;
+		upload_info = _gf_false;
+	}
+	else {
+		/*
+		 * info wasn't found in the cache;
+		 */
+		info = alloc_inode_info(local, local->loc);
+		if (!info)
+			goto error;
+		init_inode_info_head(info, fd);
+		local->info = info;
+		upload_info = _gf_true;
+	}
+	op_errno = open_format((unsigned char *)old_mtd->data,
+			       old_mtd->len,
+			       local->loc,
+			       info,
+			       get_master_cinfo(priv),
+			       local,
+			       upload_info);
+	if (op_errno)
+		goto error;
+	if (upload_info == _gf_true) {
+		op_errno = init_inode_info_tail(info,
+						get_master_cinfo(priv));
+		if (op_errno)
+			goto error;
+		op_errno = inode_ctx_put(fd->inode, this,
+					 (uint64_t)(long)(info));
+		if (op_errno == -1) {
+			op_errno = EIO;
+			goto error;
+		}
+	}
+	/*
+	 * update the format string (append/update/cup a MAC)
+	 */
+	op_errno = update_format(local->format,
+				 (unsigned char *)old_mtd->data,
+				 old_mtd->len,
+				 local->mac_idx,
+				 mop,
+				 local->newloc,
+				 info,
+				 get_master_cinfo(priv),
+				 local);
+	if (op_errno)
+		goto error;
+	/*
+	 * store the new format string on the server
+	 */
+	if (new_mtd_size) {
+		op_errno = dict_set_static_bin(local->xattr,
+					       CRYPTO_FORMAT_PREFIX,
+					       local->format,
+					       new_mtd_size);
+		if (op_errno)
+			goto error;
+	}
+	STACK_WIND(frame,
+		   do_linkop,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->setxattr,
+		   local->loc,
+		   local->xattr,
+		   0,
+		   NULL);
+	return 0;
+ wind:
+	wind_fn(frame, this);
+	return 0;
+ error:
+	local->op_ret = -1;
+	local->op_errno = op_errno;
+	unwind_fn(frame);
+	return 0;
+}
+
+static int32_t linkop_grab_local(call_frame_t *frame,
+				 xlator_t *this,
+				 loc_t *oldloc,
+				 loc_t *newloc,
+				 int flags, dict_t *xdata,
+				 glusterfs_fop_t op)
+{
+	int32_t ret = ENOMEM;
+	fd_t *fd;
+	crypt_local_t *local;
+
+	local = crypt_alloc_local(frame, this, op);
+	if (!local)
+		goto error;
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+
+	fd = fd_create(oldloc->inode, frame->root->pid);
+	if (!fd) {
+		gf_log(this->name, GF_LOG_ERROR, "Can not create fd");
+		goto error;
+	}
+	local->fd = fd;
+	local->flags = flags;
+	local->loc = GF_CALLOC(1, sizeof(*oldloc), gf_crypt_mt_loc);
+	if (!local->loc)
+		goto error;
+	memset(local->loc, 0, sizeof(*local->loc));
+	ret = loc_copy(local->loc, oldloc);
+	if (ret) {
+		GF_FREE(local->loc);
+		local->loc = NULL;
+		goto error;
+	}
+	if (newloc) {
+		local->newloc = GF_CALLOC(1, sizeof(*newloc), gf_crypt_mt_loc);
+		if (!local->newloc) {
+			GF_FREE(local->loc);
+			loc_wipe(local->loc);
+			goto error;
+		}
+		memset(local->newloc, 0, sizeof(*local->newloc));
+		ret = loc_copy(local->newloc, newloc);
+		if (ret) {
+			GF_FREE(local->loc);
+			loc_wipe(local->loc);
+			GF_FREE(local->newloc);
+			goto error;
+		}
+	}
+	local->xattr = dict_new();
+	if (!local->xattr) {
+		gf_log(this->name, GF_LOG_ERROR, "Can not create dict");
+		ret = ENOMEM;
+		goto error;
+	}
+	return 0;
+ error:
+	if (local->xdata)
+		dict_unref(local->xdata);
+	if (local->fd)
+		fd_unref(local->fd);
+	local->fd = 0;
+	local->loc = NULL;
+	local->newloc = NULL;
+
+	local->op_ret = -1;
+	local->op_errno = ret;
+
+	return ret;
+}
+
+/*
+ * read and verify locked metadata against the old pathname (via open);
+ * update the metadata string in accordance with the new pathname;
+ * submit modified metadata;
+ * wind;
+ */
+static int32_t linkop(call_frame_t *frame,
+		      xlator_t *this,
+		      loc_t *oldloc,
+		      loc_t *newloc,
+		      int flags,
+		      dict_t *xdata,
+		      glusterfs_fop_t op)
+{
+	int32_t ret;
+	dict_t *dict;
+	crypt_local_t *local;
+	void (*unwind_fn)(call_frame_t *frame);
+
+	unwind_fn = linkop_unwind_dispatch(op);
+
+	ret = linkop_grab_local(frame, this, oldloc, newloc, flags, xdata, op);
+	local = frame->local;
+	if (ret)
+		goto error;
+	dict = dict_new();
+	if (!dict) {
+		gf_log(this->name, GF_LOG_ERROR, "Can not create dict");
+		ret = ENOMEM;
+		goto error;
+	}
+	/*
+	 * Set a message to crypt_open() that we need
+	 * locked metadata string.
+	 * All link operations (link, unlink, rename)
+	 * need write lock
+	 */
+	msgflags_set_mtd_wlock(&local->msgflags);
+	ret = dict_set_static_bin(dict,
+				  MSGFLAGS_PREFIX,
+				  &local->msgflags,
+				  sizeof(local->msgflags));
+	if (ret) {
+		gf_log(this->name, GF_LOG_ERROR, "Can not set dict");
+		dict_unref(dict);
+		goto error;
+	}
+	/*
+	 * verify metadata against the old pathname
+	 * and retrieve locked metadata string
+	 */
+	STACK_WIND(frame,
+		   linkop_begin,
+		   this,
+		   this->fops->open, /* crypt_open() */
+		   oldloc,
+		   O_RDWR,
+		   local->fd,
+		   dict);
+	dict_unref(dict);
+	return 0;
+ error:
+	local->op_ret = -1;
+	local->op_errno = ret;
+	unwind_fn(frame);
+	return 0;
+}
+
+static int32_t crypt_link(call_frame_t *frame, xlator_t *this,
+			  loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+{
+	return linkop(frame, this, oldloc, newloc, 0, xdata, GF_FOP_LINK);
+}
+
+static int32_t crypt_unlink(call_frame_t *frame, xlator_t *this,
+			    loc_t *loc, int flags, dict_t *xdata)
+{
+	return linkop(frame, this, loc, NULL, flags, xdata, GF_FOP_UNLINK);
+}
+
+static int32_t crypt_rename(call_frame_t *frame, xlator_t *this,
+			    loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+{
+	return linkop(frame, this, oldloc, newloc, 0, xdata, GF_FOP_RENAME);
+}
+
+static void put_one_call_open(call_frame_t *frame)
+{
+	crypt_local_t *local = frame->local;
+	if (put_one_call(local)) {
+		fd_t *fd = local->fd;
+		loc_t *loc = local->loc;
+		dict_t *xdata = local->xdata;
+
+		STACK_UNWIND_STRICT(open,
+				    frame,
+				    local->op_ret,
+				    local->op_errno,
+				    fd,
+				    xdata);
+		fd_unref(fd);
+		if (xdata)
+			dict_unref(xdata);
+		loc_wipe(loc);
+		GF_FREE(loc);
+	}
+}
+
+static int32_t __crypt_readv_done(call_frame_t *frame,
+				  void *cookie,
+				  xlator_t *this,
+				  int32_t op_ret,
+				  int32_t op_errno, dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+	fd_t *local_fd = local->fd;
+	dict_t *local_xdata = local->xdata;
+	/* read deals with data configs only */
+	struct iovec *avec = local->data_conf.avec;
+	char **pool = local->data_conf.pool;
+	int blocks_in_pool = local->data_conf.blocks_in_pool;
+	struct iobref *iobref = local->iobref;
+	struct iobref *iobref_data = local->iobref_data;
+
+	if (op_ret < 0) {
+		gf_log(this->name, GF_LOG_WARNING,
+		       "readv unlock failed (%d)", op_errno);
+		if (local->op_ret >= 0) {
+			local->op_ret = op_ret;
+			local->op_errno = op_errno;
+		}
+	}
+	dump_plain_text(local, avec);
+
+	gf_log("crypt", GF_LOG_DEBUG,
+	       "readv: ret_to_user: %d, iovec len: %d, ia_size: %llu",
+	       (int)(local->rw_count > 0 ? local->rw_count : local->op_ret),
+	       (int)(local->rw_count > 0 ? iovec_get_size(avec, local->data_conf.acount) : 0),
+	       (unsigned long long)local->buf.ia_size);
+
+	STACK_UNWIND_STRICT(readv,
+			    frame,
+			    local->rw_count > 0 ? local->rw_count : local->op_ret,
+			    local->op_errno,
+			    avec,
+			    avec ? local->data_conf.acount : 0,
+			    &local->buf,
+			    local->iobref,
+			    local_xdata);
+
+	free_avec(avec, pool, blocks_in_pool);
+	fd_unref(local_fd);
+	if (local_xdata)
+		dict_unref(local_xdata);
+	if (iobref)
+		iobref_unref(iobref);
+	if (iobref_data)
+		iobref_unref(iobref_data);
+	return 0;
+}
+
+static void crypt_readv_done(call_frame_t *frame, xlator_t *this)
+{
+	if (parent_is_crypt_xlator(frame, this))
+		/*
+		 *  don't unlock (it will be done by the parent)
+		 */
+		__crypt_readv_done(frame, NULL, this, 0, 0, NULL);
+	else {
+		crypt_local_t *local = frame->local;
+		struct gf_flock  lock  = {0, };
+
+		lock.l_type   = F_UNLCK;
+		lock.l_whence = SEEK_SET;
+		lock.l_start  = 0;
+		lock.l_len    = 0;
+		lock.l_pid    = 0;
+
+		STACK_WIND(frame,
+			   __crypt_readv_done,
+			   FIRST_CHILD(this),
+			   FIRST_CHILD(this)->fops->finodelk,
+			   this->name,
+			   local->fd,
+			   F_SETLKW,
+			   &lock,
+			   NULL);
+	}
+}
+
+static void put_one_call_readv(call_frame_t *frame, xlator_t *this)
+{
+	crypt_local_t *local = frame->local;
+	if (put_one_call(local))
+		crypt_readv_done(frame, this);
+}
+
+static int32_t __crypt_writev_done(call_frame_t *frame,
+				   void *cookie,
+				   xlator_t *this,
+				   int32_t op_ret,
+				   int32_t op_errno, dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+	fd_t *local_fd = local->fd;
+	dict_t *local_xdata = local->xdata;
+	int32_t ret_to_user;
+
+	if (local->xattr)
+		dict_unref(local->xattr);
+	/*
+	 * Calculate amout of butes to be returned
+	 * to user. We need to subtract paddings that
+	 * have been written as a part of atom.
+	 */
+	/*
+	 * subtract head padding
+	 */
+	if (local->rw_count == 0)
+		/*
+		 * Nothing has been written, it must be an error
+		 */
+		ret_to_user = local->op_ret;
+	else if (local->rw_count <= local->data_conf.off_in_head) {
+		gf_log("crypt", GF_LOG_WARNING, "Incomplete write");
+		ret_to_user = 0;
+	}
+	else
+		ret_to_user = local->rw_count -
+			local->data_conf.off_in_head;
+	/*
+	 * subtract tail padding
+	 */
+	if (ret_to_user > local->data_conf.orig_size)
+		ret_to_user = local->data_conf.orig_size;
+
+	if (local->iobref)
+		iobref_unref(local->iobref);
+	if (local->iobref_data)
+		iobref_unref(local->iobref_data);
+	free_avec_data(local);
+	free_avec_hole(local);
+
+	gf_log("crypt", GF_LOG_DEBUG,
+	       "writev: ret_to_user: %d", ret_to_user);
+
+	STACK_UNWIND_STRICT(writev,
+			    frame,
+			    ret_to_user,
+			    local->op_errno,
+			    &local->prebuf,
+			    &local->postbuf,
+			    local_xdata);
+	fd_unref(local_fd);
+	if (local_xdata)
+		dict_unref(local_xdata);
+	return 0;
+}
+
+static int32_t crypt_writev_done(call_frame_t *frame,
+				 void *cookie,
+				 xlator_t *this,
+				 int32_t op_ret,
+				 int32_t op_errno,
+				 dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	if (op_ret < 0)
+		gf_log("crypt", GF_LOG_WARNING, "can not update file size");
+
+	if (parent_is_crypt_xlator(frame, this))
+		/*
+		 * don't unlock (it will be done by the parent)
+		 */
+		__crypt_writev_done(frame, NULL, this, 0, 0, NULL);
+	else {
+		struct gf_flock  lock  = {0, };
+
+		lock.l_type   = F_UNLCK;
+		lock.l_whence = SEEK_SET;
+		lock.l_start  = 0;
+		lock.l_len    = 0;
+		lock.l_pid    = 0;
+
+		STACK_WIND(frame,
+			   __crypt_writev_done,
+			   FIRST_CHILD(this),
+			   FIRST_CHILD(this)->fops->finodelk,
+			   this->name,
+			   local->fd,
+			   F_SETLKW,
+			   &lock,
+			   NULL);
+	}
+	return 0;
+}
+
+static void put_one_call_writev(call_frame_t *frame, xlator_t *this)
+{
+	crypt_local_t *local = frame->local;
+	if (put_one_call(local)) {
+		if (local->update_disk_file_size) {
+			int32_t ret;
+			/*
+			 * update file size, unlock the file and unwind
+			 */
+			ret = dict_set(local->xattr,
+				       FSIZE_XATTR_PREFIX,
+				       data_from_uint64(local->cur_file_size));
+			if (ret) {
+				gf_log("crypt", GF_LOG_WARNING,
+				       "can not set key to update file size");
+				crypt_writev_done(frame, NULL,
+						  this, 0, 0, NULL);
+				return;
+			}
+			gf_log("crypt", GF_LOG_DEBUG,
+			       "Updating disk file size to %llu",
+			       (unsigned long long)local->cur_file_size);
+			STACK_WIND(frame,
+				   crypt_writev_done,
+				   FIRST_CHILD(this),
+				   FIRST_CHILD(this)->fops->fsetxattr,
+				   local->fd,
+				   local->xattr, /* CRYPTO_FORMAT_PREFIX */
+				   0,
+				   NULL);
+		}
+		else
+			crypt_writev_done(frame, NULL, this, 0, 0, NULL);
+	}
+}
+
+static int32_t __crypt_ftruncate_done(call_frame_t *frame,
+				      void *cookie,
+				      xlator_t *this,
+				      int32_t op_ret,
+				      int32_t op_errno, dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+	fd_t *local_fd = local->fd;
+	dict_t *local_xdata = local->xdata;
+	char *iobase = local->vec.iov_base;
+
+	if (op_ret < 0) {
+		gf_log(this->name, GF_LOG_WARNING,
+		       "ftruncate unlock failed (%d)", op_errno);
+		if (local->op_ret >= 0) {
+			local->op_ret = op_ret;
+			local->op_errno = op_errno;
+		}
+	}
+	if (local->iobref_data)
+		iobref_unref(local->iobref_data);
+	free_avec_data(local);
+	free_avec_hole(local);
+
+	gf_log("crypt", GF_LOG_DEBUG,
+	       "ftruncate, return to user: presize=%llu, postsize=%llu",
+	       (unsigned long long)local->prebuf.ia_size,
+	       (unsigned long long)local->postbuf.ia_size);
+
+	STACK_UNWIND_STRICT(ftruncate,
+			    frame,
+			    local->op_ret < 0 ? -1 : 0,
+			    local->op_errno,
+			    &local->prebuf,
+			    &local->postbuf,
+			    local_xdata);
+	fd_unref(local_fd);
+	if (local_xdata)
+		dict_unref(local_xdata);
+	if (iobase)
+		GF_FREE(iobase);
+	return 0;
+}
+
+static int32_t crypt_ftruncate_done(call_frame_t *frame,
+				    void *cookie,
+				    xlator_t *this,
+				    int32_t op_ret,
+				    int32_t op_errno,
+				    dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+	struct gf_flock  lock  = {0, };
+
+	dict_unref(local->xattr);
+	if (op_ret < 0)
+		gf_log("crypt", GF_LOG_WARNING, "can not update file size");
+
+	lock.l_type   = F_UNLCK;
+	lock.l_whence = SEEK_SET;
+	lock.l_start  = 0;
+	lock.l_len    = 0;
+	lock.l_pid    = 0;
+
+	STACK_WIND(frame,
+		   __crypt_ftruncate_done,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->finodelk,
+		   this->name,
+		   local->fd,
+		   F_SETLKW,
+		   &lock,
+		   NULL);
+	return 0;
+}
+
+static void put_one_call_ftruncate(call_frame_t *frame, xlator_t *this)
+{
+	crypt_local_t *local = frame->local;
+	if (put_one_call(local)) {
+		if (local->update_disk_file_size) {
+			int32_t ret;
+			/*
+			 * update file size, unlock the file and unwind
+			 */
+			ret = dict_set(local->xattr,
+				       FSIZE_XATTR_PREFIX,
+				       data_from_uint64(local->cur_file_size));
+			if (ret) {
+				gf_log("crypt", GF_LOG_WARNING,
+				       "can not set key to update file size");
+				crypt_ftruncate_done(frame, NULL,
+						     this, 0, 0, NULL);
+				return;
+			}
+			gf_log("crypt", GF_LOG_DEBUG,
+			       "Updating disk file size to %llu",
+			       (unsigned long long)local->cur_file_size);
+			STACK_WIND(frame,
+				   crypt_ftruncate_done,
+				   FIRST_CHILD(this),
+				   FIRST_CHILD(this)->fops->fsetxattr,
+				   local->fd,
+				   local->xattr, /* CRYPTO_FORMAT_PREFIX */
+				   0,
+				   NULL);
+		}
+		else
+			crypt_ftruncate_done(frame, NULL, this, 0, 0, NULL);
+	}
+}
+
+/*
+ * load regular file size for some FOPs
+ */
+static int32_t load_file_size(call_frame_t *frame,
+			      void *cookie,
+			      xlator_t *this,
+			      int32_t op_ret,
+			      int32_t op_errno,
+			      dict_t *dict,
+			      dict_t *xdata)
+{
+	data_t *data;
+	crypt_local_t *local = frame->local;
+
+	dict_t *local_xdata = local->xdata;
+	inode_t *local_inode = local->inode;
+
+	if (op_ret < 0)
+		goto unwind;
+	/*
+	 * load regular file size
+	 */
+	data = dict_get(dict, FSIZE_XATTR_PREFIX);
+	if (!data) {
+		if (local->xdata)
+			dict_unref(local->xdata);
+		gf_log("crypt", GF_LOG_WARNING, "Regular file size not found");
+		op_ret = -1;
+		op_errno = EIO;
+		goto unwind;
+	}
+	local->buf.ia_size = data_to_uint64(data);
+
+	gf_log(this->name, GF_LOG_DEBUG,
+	       "FOP %d: Translate regular file to %llu",
+	       local->fop,
+	       (unsigned long long)local->buf.ia_size);
+ unwind:	
+	if (local->fd)
+		fd_unref(local->fd);
+	if (local->loc) {
+		loc_wipe(local->loc);
+		GF_FREE(local->loc);
+	}
+	switch (local->fop) {
+	case GF_FOP_FSTAT:
+		STACK_UNWIND_STRICT(fstat,
+				    frame,
+				    op_ret,
+				    op_errno,
+				    op_ret >= 0 ? &local->buf : NULL,
+				    local->xdata);
+		break;
+	case GF_FOP_STAT:
+		STACK_UNWIND_STRICT(stat,
+				    frame,
+				    op_ret,
+				    op_errno,
+				    op_ret >= 0 ? &local->buf : NULL,
+				    local->xdata);
+		break;
+	case GF_FOP_LOOKUP:
+		STACK_UNWIND_STRICT(lookup,
+				    frame,
+				    op_ret,
+				    op_errno,
+				    op_ret >= 0 ? local->inode : NULL,
+				    op_ret >= 0 ? &local->buf : NULL,				    
+				    local->xdata,
+				    op_ret >= 0 ? &local->postbuf : NULL);
+		break;
+	case GF_FOP_READ:
+		STACK_UNWIND_STRICT(readv,
+				    frame,
+				    op_ret,
+				    op_errno,
+				    NULL,
+				    0,
+				    op_ret >= 0 ? &local->buf : NULL,
+				    NULL,
+				    NULL);
+		break;
+	default:
+		gf_log(this->name, GF_LOG_WARNING,
+		       "Improper file operation %d", local->fop);
+	}
+	if (local_xdata)
+		dict_unref(local_xdata);
+	if (local_inode)
+		inode_unref(local_inode);
+	return 0;
+}
+
+static int32_t crypt_stat_common_cbk(call_frame_t *frame,
+				     void *cookie,
+				     xlator_t *this,
+				     int32_t op_ret,
+				     int32_t op_errno,
+				     struct iatt *buf, dict_t *xdata)
+{
+	crypt_local_t *local = frame->local;
+
+	if (op_ret < 0)
+		goto unwind;
+	if (!IA_ISREG(buf->ia_type))
+		goto unwind;
+
+	local->buf = *buf;
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+
+	switch (local->fop) {
+	case GF_FOP_FSTAT:
+		STACK_WIND(frame,
+			   load_file_size,
+			   FIRST_CHILD(this),
+			   FIRST_CHILD(this)->fops->fgetxattr,
+			   local->fd,
+			   FSIZE_XATTR_PREFIX,
+			   NULL);
+		break;
+	case GF_FOP_STAT:
+		STACK_WIND(frame,
+			   load_file_size,
+			   FIRST_CHILD(this),
+			   FIRST_CHILD(this)->fops->getxattr,
+			   local->loc,
+			   FSIZE_XATTR_PREFIX,
+			   NULL);
+		break;
+	default:
+		gf_log (this->name, GF_LOG_WARNING,
+			"Improper file operation %d", local->fop);
+	}
+	return 0;
+ unwind:
+	if (local->fd)
+		fd_unref(local->fd);
+	if (local->loc) {
+		loc_wipe(local->loc);
+		GF_FREE(local->loc);
+	}
+	switch (local->fop) {
+	case GF_FOP_FSTAT:
+		STACK_UNWIND_STRICT(fstat,
+				    frame,
+				    op_ret,
+				    op_errno,
+				    op_ret >= 0 ? buf : NULL,
+				    op_ret >= 0 ? xdata : NULL);
+		break;
+	case GF_FOP_STAT:
+		STACK_UNWIND_STRICT(stat,
+				    frame,
+				    op_ret,
+				    op_errno,
+				    op_ret >= 0 ? buf : NULL,
+				    op_ret >= 0 ? xdata : NULL);
+		break;
+	default:
+		gf_log (this->name, GF_LOG_WARNING,
+			"Improper file operation %d", local->fop);
+	}
+	return 0;
+}
+
+static int32_t crypt_fstat(call_frame_t *frame,
+			   xlator_t *this,
+			   fd_t *fd, dict_t *xdata)
+{
+	crypt_local_t *local;
+
+	local = crypt_alloc_local(frame, this, GF_FOP_FSTAT);
+	if (!local)
+		goto error;
+	local->fd = fd_ref(fd);
+	STACK_WIND(frame,
+		   crypt_stat_common_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->fstat,
+		   fd,
+		   xdata);	
+	return 0;
+ error:
+	STACK_UNWIND_STRICT(fstat,
+			    frame,
+			    -1,
+			    ENOMEM,
+			    NULL,
+			    NULL);
+	return 0;
+}
+
+static int32_t crypt_stat(call_frame_t *frame,
+			  xlator_t *this,
+			  loc_t *loc, dict_t *xdata)
+{
+	int32_t ret;
+	crypt_local_t *local;
+
+	local = crypt_alloc_local(frame, this, GF_FOP_STAT);
+	if (!local)
+		goto error;
+	local->loc = GF_CALLOC(1, sizeof(*loc), gf_crypt_mt_loc);
+	if (!local->loc)
+		goto error;
+	memset(local->loc, 0, sizeof(*local->loc));
+	ret = loc_copy(local->loc, loc);
+	if (ret) {
+		GF_FREE(local->loc);
+		goto error;
+	}
+	STACK_WIND(frame,
+		   crypt_stat_common_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->stat,
+		   loc,
+		   xdata);	
+	return 0;
+ error:
+	STACK_UNWIND_STRICT(stat,
+			    frame,
+			    -1,
+			    ENOMEM,
+			    NULL,
+			    NULL);
+	return 0;
+}
+
+static int32_t crypt_lookup_cbk(call_frame_t *frame,
+				void *cookie,
+				xlator_t *this,
+				int32_t op_ret,
+				int32_t op_errno,
+				inode_t *inode,
+				struct iatt *buf, dict_t *xdata,
+				struct iatt *postparent)
+{
+	crypt_local_t *local = frame->local;
+
+	if (op_ret < 0)
+		goto unwind;
+	if (!IA_ISREG(buf->ia_type))
+		goto unwind;
+
+	local->inode = inode_ref(inode);
+	local->buf = *buf;
+	local->postbuf = *postparent;
+	if (xdata)
+		local->xdata = dict_ref(xdata);
+	uuid_copy(local->loc->gfid, buf->ia_gfid);
+
+	STACK_WIND(frame,
+		   load_file_size,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->getxattr,
+		   local->loc,
+		   FSIZE_XATTR_PREFIX,
+		   NULL);
+	return 0;
+ unwind:
+	loc_wipe(local->loc);
+	GF_FREE(local->loc);
+	STACK_UNWIND_STRICT(lookup,
+			    frame,
+			    op_ret,
+			    op_errno,
+			    inode,
+			    buf,
+			    xdata,
+			    postparent);
+	return 0;
+}
+
+static int32_t crypt_lookup(call_frame_t *frame,
+			    xlator_t *this,
+			    loc_t *loc, dict_t *xdata)
+{
+	int32_t ret;
+	crypt_local_t *local;
+
+	local = crypt_alloc_local(frame, this, GF_FOP_LOOKUP);
+	if (!local)
+		goto error;
+	local->loc = GF_CALLOC(1, sizeof(*loc), gf_crypt_mt_loc);
+	if (!local->loc)
+		goto error;
+	memset(local->loc, 0, sizeof(*local->loc));
+	ret = loc_copy(local->loc, loc);
+	if (ret) {
+		GF_FREE(local->loc);
+		goto error;
+	}
+	gf_log(this->name, GF_LOG_DEBUG, "Lookup %s", loc->path);
+	STACK_WIND(frame,
+		   crypt_lookup_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->lookup,
+		   loc,
+		   xdata);
+	return 0;
+ error:
+	STACK_UNWIND_STRICT(lookup,
+			    frame,
+			    -1,
+			    ENOMEM,
+			    NULL,
+			    NULL,
+			    NULL,
+			    NULL);
+	return 0;
+}
+
+/*
+ * for every regular directory entry find its real file size
+ * and update stat's buf properly
+ */
+static int32_t crypt_readdirp_cbk(call_frame_t *frame,
+				  void *cookie,
+				  xlator_t *this,
+				  int32_t op_ret,
+				  int32_t op_errno,
+				  gf_dirent_t *entries, dict_t *xdata)
+{
+	gf_dirent_t *entry = NULL;
+
+	if (op_ret < 0)
+		goto unwind;
+
+	list_for_each_entry (entry, (&entries->list), list) {
+		data_t *data;
+
+		if (!IA_ISREG(entry->d_stat.ia_type))
+			continue;
+		data = dict_get(entry->dict, FSIZE_XATTR_PREFIX);
+		if (!data){
+			gf_log("crypt", GF_LOG_WARNING,
+			       "Regular file size of direntry not found");
+			op_errno = EIO;
+			op_ret = -1;
+			break;
+		}
+		entry->d_stat.ia_size = data_to_uint64(data);
+	}
+ unwind:
+	STACK_UNWIND_STRICT(readdirp, frame, op_ret, op_errno, entries, xdata);
+	return 0;
+}
+
+/*
+ * ->readdirp() fills in-core inodes, so we need to set proper
+ * file sizes for all directory entries of the parent @fd.
+ * Actual updates take place in ->crypt_readdirp_cbk()
+ */
+static int32_t crypt_readdirp(call_frame_t *frame, xlator_t *this,
+			      fd_t *fd, size_t size, off_t offset,
+			      dict_t *xdata)
+{
+	int32_t ret = ENOMEM;
+
+	if (!xdata) {
+		xdata = dict_new();
+		if (!xdata)
+			goto error;
+	}
+	else
+		dict_ref(xdata);
+	/*
+	 * make sure that we'll have real file sizes at ->readdirp_cbk()
+	 */
+	ret = dict_set(xdata, FSIZE_XATTR_PREFIX, data_from_uint64(0));
+	if (ret) {
+		dict_unref(xdata);
+		goto error;
+	}
+	STACK_WIND(frame,
+		   crypt_readdirp_cbk,
+		   FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->readdirp,
+		   fd,
+		   size,
+		   offset,
+		   xdata);
+	dict_unref(xdata);
+	return 0;
+ error:
+	STACK_UNWIND_STRICT(readdirp, frame, -1, ret, NULL, NULL);
+	return 0;
+}
+
+static int32_t crypt_access(call_frame_t *frame,
+			    xlator_t *this,
+			    loc_t *loc,
+			    int32_t mask, dict_t *xdata)
+{
+	gf_log(this->name, GF_LOG_WARNING,
+	       "NFS mounts of encrypted volumes are unsupported");
+	STACK_UNWIND_STRICT(access, frame, -1, EPERM, NULL);
+	return 0;
+}
+
+int32_t master_set_block_size (xlator_t *this, crypt_private_t *priv,
+			       dict_t *options)
+{
+	uint64_t block_size = 0;
+	struct master_cipher_info *master = get_master_cinfo(priv);
+
+	if (options != NULL)
+		GF_OPTION_RECONF("block-size", block_size, options,
+				 size, error);
+	else
+		GF_OPTION_INIT("block-size", block_size, size, error);
+
+	switch (block_size) {
+	case 512:
+		master->m_block_bits = 9;
+		break;
+	case 1024:
+		master->m_block_bits = 10;
+		break;
+	case 2048:
+		master->m_block_bits = 11;
+		break;
+	case 4096:
+		master->m_block_bits = 12;
+		break;
+	default:
+		gf_log("crypt", GF_LOG_ERROR,
+		       "FATAL: unsupported block size %llu",
+		       (unsigned long long)block_size);
+		goto error;
+	}
+	return 0;
+ error:
+	return -1;
+}
+
+int32_t master_set_alg(xlator_t *this, crypt_private_t *priv)
+{
+	struct master_cipher_info *master = get_master_cinfo(priv);
+	master->m_alg = AES_CIPHER_ALG;
+	return 0;
+}
+
+int32_t master_set_mode(xlator_t *this, crypt_private_t *priv)
+{
+	struct master_cipher_info *master = get_master_cinfo(priv);
+	master->m_mode = XTS_CIPHER_MODE;
+	return 0;
+}
+
+/*
+ * set key size in bits to the master info
+ * Pre-conditions: cipher mode in the master info is uptodate.
+ */
+static int master_set_data_key_size (xlator_t *this, crypt_private_t *priv,
+				     dict_t *options)
+{
+	int32_t ret;
+	uint64_t key_size = 0;
+	struct master_cipher_info *master = get_master_cinfo(priv);
+
+	if (options != NULL)
+		GF_OPTION_RECONF("data-key-size", key_size, options,
+				 size, error);
+	else
+		GF_OPTION_INIT("data-key-size", key_size, size, error);
+
+	ret = data_cipher_algs[master->m_alg][master->m_mode].check_key(key_size);
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR,
+		       "FATAL: wrong bin key size %llu for alg %d mode %d",
+		       (unsigned long long)key_size,
+		       (int)master->m_alg,
+		       (int)master->m_mode);
+		goto error;
+	}
+	master->m_dkey_size = key_size;
+	return 0;
+ error:
+	return -1;
+}
+
+static int is_hex(char *s) {
+	return  ('0' <= *s && *s <= '9') || ('a' <= *s && *s <= 'f');
+}
+
+static int parse_hex_buf(xlator_t *this, char *src, unsigned char *dst,
+			 int hex_size)
+{
+	int i;
+	int hex_byte = 0;
+
+	for (i = 0; i < (hex_size / 2); i++) {
+		if (!is_hex(src + i*2) || !is_hex(src + i*2 + 1)) {
+			gf_log("crypt", GF_LOG_ERROR,
+			       "FATAL: not hex symbol in key");
+			return -1;
+		}
+		if (sscanf(src + i*2, "%2x", &hex_byte) != 1) {
+			gf_log("crypt", GF_LOG_ERROR,
+			       "FATAL: can not parse hex key");
+			return -1;
+		}
+		dst[i] = hex_byte & 0xff;
+	}
+	return 0;
+}
+
+/*
+ * Parse options;
+ * install master volume key
+ */
+int32_t master_set_master_vol_key(xlator_t *this, crypt_private_t *priv)
+{
+	int32_t ret;
+	FILE *file = NULL;
+
+	int32_t key_size;
+	char *opt_key_file_pathname = NULL;
+
+	unsigned char bin_buf[MASTER_VOL_KEY_SIZE];
+	char hex_buf[2 * MASTER_VOL_KEY_SIZE];
+
+	struct master_cipher_info *master = get_master_cinfo(priv);
+	/*
+	 * extract master key passed via option
+	 */
+	GF_OPTION_INIT("master-key", opt_key_file_pathname, path, bad_key);
+
+	if (!opt_key_file_pathname) {
+		gf_log(this->name, GF_LOG_ERROR, "FATAL: missing master key");
+		return -1;
+	}
+	gf_log(this->name, GF_LOG_DEBUG, "handling file key %s",
+	       opt_key_file_pathname);
+
+	file = fopen(opt_key_file_pathname, "r");
+	if (file == NULL) {
+		gf_log(this->name, GF_LOG_ERROR,
+		       "FATAL: can not open file with master key");
+		return -1;
+	}
+	/*
+	 * extract hex key
+	 */
+	key_size = fread(hex_buf, 1, sizeof(hex_buf), file);
+	if (key_size < sizeof(hex_buf)) {
+		gf_log(this->name, GF_LOG_ERROR,
+		       "FATAL: master key is too short");
+		goto bad_key;
+	}
+	ret = parse_hex_buf(this, hex_buf, bin_buf, key_size);
+	if (ret)
+		goto bad_key;
+	memcpy(master->m_key, bin_buf, MASTER_VOL_KEY_SIZE);
+	memset(hex_buf, 0, sizeof(hex_buf));
+	fclose(file);
+
+ 	memset(bin_buf, 0, sizeof(bin_buf));
+	return 0;
+ bad_key:
+	gf_log(this->name, GF_LOG_ERROR, "FATAL: bad master key");
+	if (file)
+		fclose(file);
+	memset(bin_buf, 0, sizeof(bin_buf));
+	return -1;
+}
+
+/*
+ * Derive volume key for object-id authentication
+ */
+int32_t master_set_nmtd_vol_key(xlator_t *this, crypt_private_t *priv)
+{
+	return get_nmtd_vol_key(get_master_cinfo(priv));
+}
+
+int32_t crypt_init_xlator(xlator_t *this)
+{
+	int32_t ret;
+	crypt_private_t *priv = this->private; 
+
+	ret = master_set_alg(this, priv);
+	if (ret)
+		return ret;
+	ret = master_set_mode(this, priv);
+	if (ret)
+		return ret;
+	ret = master_set_block_size(this, priv, NULL);
+	if (ret)
+		return ret;
+	ret = master_set_data_key_size(this, priv, NULL);
+	if (ret)
+		return ret;
+	ret = master_set_master_vol_key(this, priv);
+	if (ret)
+		return ret;
+	return master_set_nmtd_vol_key(this, priv);
+}
+
+static int32_t crypt_alloc_private(xlator_t *this)
+{
+	this->private = GF_CALLOC(1, sizeof(crypt_private_t), gf_crypt_mt_priv);
+	if (!this->private) {
+		gf_log("crypt", GF_LOG_ERROR,
+		       "Can not allocate memory for private data");
+		return ENOMEM;
+	}
+	return 0;
+}
+
+static void crypt_free_private(xlator_t *this)
+{
+	crypt_private_t *priv = this->private;
+	if (priv) {
+		memset(priv, 0, sizeof(*priv));
+		GF_FREE(priv);
+	}
+}
+
+int32_t reconfigure (xlator_t *this, dict_t *options)
+{
+	int32_t ret = -1;
+	crypt_private_t *priv = NULL;
+
+	GF_VALIDATE_OR_GOTO ("crypt", this, error);
+	GF_VALIDATE_OR_GOTO (this->name, this->private, error);
+	GF_VALIDATE_OR_GOTO (this->name, options, error);
+
+	priv = this->private;
+
+	ret = master_set_block_size(this, priv, options);
+	if (ret) {
+		gf_log("this->name", GF_LOG_ERROR,
+		       "Failed to reconfure block size");
+		goto error;
+	}
+	ret = master_set_data_key_size(this, priv, options);
+	if (ret) {
+		gf_log("this->name", GF_LOG_ERROR,
+		       "Failed to reconfure data key size");
+		goto error;
+	}
+	return 0;
+ error:
+	return ret;
+}
+
+int32_t init(xlator_t *this)
+{
+	int32_t ret;
+
+	if (!this->children || this->children->next) {
+		gf_log ("crypt", GF_LOG_ERROR,
+			"FATAL: crypt should have exactly one child");
+		return EINVAL;
+	}
+	if (!this->parents) {
+		gf_log (this->name, GF_LOG_WARNING,
+			"dangling volume. check volfile ");
+	}
+	ret = crypt_alloc_private(this);
+	if (ret)
+		return ret;
+	ret = crypt_init_xlator(this);
+	if (ret)
+		goto error;
+	this->local_pool = mem_pool_new(crypt_local_t, 64);
+        if (!this->local_pool) {
+		gf_log(this->name, GF_LOG_ERROR,
+		       "failed to create local_t's memory pool");
+		ret = ENOMEM;
+                goto error;
+        }
+	gf_log ("crypt", GF_LOG_INFO, "crypt xlator loaded");
+	return 0;
+ error:
+	crypt_free_private(this);
+	return ret;
+}
+
+void fini (xlator_t *this)
+{
+	crypt_free_private(this);
+}
+
+struct xlator_fops fops = {
+	.readv        = crypt_readv,
+	.writev       = crypt_writev,
+	.truncate     = crypt_truncate,
+	.ftruncate    = crypt_ftruncate,
+	.setxattr     = crypt_setxattr,
+	.fsetxattr    = crypt_fsetxattr,
+	.link         = crypt_link,
+	.unlink       = crypt_unlink,
+	.rename       = crypt_rename,
+	.open         = crypt_open,
+	.create       = crypt_create,
+	.stat         = crypt_stat,
+	.fstat        = crypt_fstat,
+	.lookup       = crypt_lookup,
+	.readdirp     = crypt_readdirp,
+	.access       = crypt_access   
+};
+
+struct xlator_cbks cbks = {
+	.forget       = crypt_forget
+};
+
+struct volume_options options[] = {
+	{ .key = {"master-key"},
+	  .type = GF_OPTION_TYPE_PATH,
+	  .description = "Pathname of regular file which contains master volume key"
+	},
+	{ .key = {"data-key-size"},
+	  .type = GF_OPTION_TYPE_SIZET,
+	  .description = "Data key size (bits)",
+	  .min = 256,
+	  .max = 512,
+	  .default_value = "256",
+	},
+	{ .key = {"block-size"},
+	  .type = GF_OPTION_TYPE_SIZET,
+	  .description = "Atom size (bits)",
+	  .min = 512,
+	  .max = 4096,
+	  .default_value = "4096"
+	},
+	{ .key  = {NULL} },
+};
+
+/*
+  Local variables:
+  c-indentation-style: "K&R"
+  mode-name: "LC"
+  c-basic-offset: 8
+  tab-width: 8
+  fill-column: 80
+  scroll-step: 1
+  End:
+*/
diff --git a/xlators/encryption/crypt/src/crypt.h b/xlators/encryption/crypt/src/crypt.h
new file mode 100644
index 000000000..01a8542ab
--- /dev/null
+++ b/xlators/encryption/crypt/src/crypt.h
@@ -0,0 +1,899 @@
+/*
+  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 __CRYPT_H__
+#define __CRYPT_H__
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+#include <openssl/aes.h>
+#include <openssl/evp.h>
+#include <openssl/sha.h>
+#include <openssl/hmac.h>
+#include <openssl/cmac.h>
+#include <openssl/modes.h>
+#include "crypt-mem-types.h"
+
+#define CRYPT_XLATOR_ID  (0)
+
+#define MAX_IOVEC_BITS (3)
+#define MAX_IOVEC (1 << MAX_IOVEC_BITS)
+#define KEY_FACTOR_BITS  (6)
+
+#define DEBUG_CRYPT (0)
+#define TRIVIAL_TFM (0)
+
+#define CRYPT_MIN_BLOCK_BITS (9)
+#define CRYPT_MAX_BLOCK_BITS (12)
+
+#define MASTER_VOL_KEY_SIZE (32)
+#define NMTD_VOL_KEY_SIZE (16)
+
+struct crypt_key {
+	uint32_t len;
+	const char *label;	
+};
+
+/*
+ * Add new key types to the end of this
+ * enumeration but before LAST_KEY_TYPE
+ */
+typedef enum {
+	MASTER_VOL_KEY,
+	NMTD_VOL_KEY,
+	NMTD_LINK_KEY,
+	EMTD_FILE_KEY,
+	DATA_FILE_KEY_256,
+	DATA_FILE_KEY_512,
+	LAST_KEY_TYPE
+}crypt_key_type;
+
+struct kderive_context {
+	const unsigned char *pkey;/* parent key */
+	uint32_t pkey_len;        /* parent key size, bits */
+	uint32_t ckey_len;        /* child key size, bits */
+	unsigned char *fid;       /* fixed input data, NIST 800-108, 5.1 */
+	uint32_t fid_len;         /* fid len, bytes */
+	unsigned char *out;       /* contains child keying material */
+	uint32_t out_len;         /* out len, bytes */
+};
+
+typedef enum {
+	DATA_ATOM,
+	HOLE_ATOM,
+	LAST_DATA_TYPE
+}atom_data_type;
+
+typedef enum {
+	HEAD_ATOM,
+	TAIL_ATOM,
+	FULL_ATOM,
+	LAST_LOCALITY_TYPE
+}atom_locality_type;
+
+typedef enum {
+	MTD_CREATE,
+	MTD_APPEND,
+	MTD_OVERWRITE,
+	MTD_CUT,
+	MTD_LAST_OP
+} mtd_op_t;
+
+struct xts128_context {
+	void      *key1, *key2;
+	block128_f block1,block2;
+};
+
+struct object_cipher_info {
+	cipher_alg_t  o_alg;
+        cipher_mode_t o_mode;
+	uint32_t      o_block_bits;
+        uint32_t      o_dkey_size; /* raw data key size in bits */
+	union {
+		struct {
+			unsigned char ivec[16];
+			AES_KEY dkey[2];
+			AES_KEY tkey;    /* key used for tweaking */
+			XTS128_CONTEXT xts;
+		} aes_xts;
+	} u;
+};
+
+struct master_cipher_info {
+	/*
+	 * attributes inherited by newly created regular files
+	 */
+	cipher_alg_t  m_alg;
+        cipher_mode_t m_mode;
+	uint32_t      m_block_bits;
+        uint32_t      m_dkey_size; /* raw key size in bits */
+	/*
+	 * master key
+	 */
+	unsigned char m_key[MASTER_VOL_KEY_SIZE];
+	/* 
+	 * volume key for oid authentication
+	 */
+	unsigned char m_nmtd_key[NMTD_VOL_KEY_SIZE];
+};
+
+/*
+* This info is not changed during file's life
+ */
+struct crypt_inode_info {
+#if DEBUG_CRYPT
+	loc_t *loc; /* pathname that the file has been
+		       opened, or created with */
+#endif
+	uint16_t nr_minor;
+	uuid_t oid;
+	struct object_cipher_info cinfo;
+};
+
+/*
+ * this should locate in secure memory
+ */
+typedef struct {
+	struct master_cipher_info master;
+} crypt_private_t;
+
+static inline struct master_cipher_info *get_master_cinfo(crypt_private_t *priv)
+{
+	return &priv->master;
+}
+
+static inline struct object_cipher_info *get_object_cinfo(struct crypt_inode_info
+							 *info)
+{
+	return &info->cinfo;
+}
+
+/*
+ * this describes layouts and properties
+ * of atoms in an aligned vector
+ */
+struct avec_config {
+	uint32_t atom_size;
+	atom_data_type type;
+	size_t orig_size;
+	off_t orig_offset;
+	size_t expanded_size;
+	off_t aligned_offset;
+
+	uint32_t off_in_head;
+	uint32_t off_in_tail;
+	uint32_t gap_in_tail;
+	uint32_t nr_full_blocks;
+
+	struct iovec *avec; /* aligned vector */
+	uint32_t acount; /* number of avec components. The same
+			  * as number of occupied logical blocks */
+	char **pool;
+	uint32_t blocks_in_pool;
+	uint32_t cursor; /* makes sense only for ordered writes,
+			  * so there is no races on this counter.
+			  *
+			  * Cursor is per-config object, we don't
+			  * reset cursor for atoms of different
+			  * localities (head, tail, full)
+			  */
+};
+
+
+typedef struct {
+	glusterfs_fop_t fop; /* code of FOP this local info built for */
+	fd_t *fd;
+	inode_t *inode;
+	loc_t *loc;
+	int32_t mac_idx;
+	loc_t *newloc;
+	int32_t flags;
+	int32_t wbflags;
+	struct crypt_inode_info *info;
+	struct iobref *iobref;
+	struct iobref *iobref_data;
+	off_t offset;
+
+	uint64_t old_file_size; /* per FOP, retrieved under lock held */
+	uint64_t cur_file_size; /* per iteration, before issuing IOs */
+	uint64_t new_file_size; /* per iteration, after issuing IOs */
+
+	uint64_t io_offset; /* offset of IOs issued per iteration */
+	uint64_t io_offset_nopad; /* offset of user's data in the atom */
+	uint32_t io_size; /* size of IOs issued per iteration */
+	uint32_t io_size_nopad; /* size of user's data in the IOs */
+	uint32_t eof_padding_size; /* size od EOF padding in the IOs */
+
+	gf_lock_t call_lock; /* protect nr_calls from many cbks */
+	int32_t nr_calls;
+
+	atom_data_type active_setup; /* which setup (hole or date)
+					is currently active */
+	/* data setup */
+	struct avec_config data_conf;
+
+	/* hole setup */
+	int hole_conv_in_proggress;
+	gf_lock_t hole_lock; /* protect hole config from many cbks */
+	int hole_handled;
+	struct avec_config hole_conf;
+	struct iatt buf;
+	struct iatt prebuf;
+	struct iatt postbuf;
+	struct iatt *prenewparent;
+	struct iatt *postnewparent;
+	int32_t op_ret;
+	int32_t op_errno;
+	int32_t rw_count; /* total read or written */
+	gf_lock_t rw_count_lock; /* protect the counter above */
+	unsigned char *format; /* for create, update format string */
+	uint32_t format_size;
+	uint32_t msgflags; /* messages for crypt_open() */
+	dict_t *xdata;
+	dict_t *xattr;
+	struct iovec vec; /* contains last file's atom for
+			     read-prune-write sequence */
+	gf_boolean_t custom_mtd;
+	/*
+	 * the next 3 fields are used by readdir and friends
+	 */
+	gf_dirent_t *de; /* directory entry */
+	char *de_path; /* pathname of directory entry */
+	uint32_t de_prefix_len; /* lenght of the parent's pathname */
+	gf_dirent_t *entries;
+
+	uint32_t update_disk_file_size:1;
+} crypt_local_t;
+
+/* This represents a (read)modify-write atom */
+struct rmw_atom {
+	atom_locality_type locality;
+	/*
+	 * read-modify-write sequence of the atom
+	 */
+	int32_t (*rmw)(call_frame_t *frame,
+		       void *cookie,
+		       xlator_t *this,
+		       int32_t op_ret,
+		       int32_t op_errno,
+		       struct iovec *vec,
+		       int32_t count,
+		       struct iatt *stbuf,
+		       struct iobref *iobref,
+		       dict_t *xdata);
+	/*
+	 * offset of the logical block in a file
+	 */
+	loff_t (*offset_at)(call_frame_t *frame,
+			    struct object_cipher_info *object);
+	/*
+	 * IO offset in an atom
+	 */
+	uint32_t (*offset_in)(call_frame_t *frame,
+			      struct object_cipher_info *object);
+	/*
+	 * number of bytes of plain text of this atom that user
+	 * wants to read/write.
+	 * It can be smaller than atom_size in the case of head
+	 * or tail atoms.
+	 */
+	uint32_t (*io_size_nopad)(call_frame_t *frame,
+				  struct object_cipher_info *object);
+	/*
+	 * which iovec represents the atom
+	 */
+	struct iovec *(*get_iovec)(call_frame_t *frame, uint32_t count);
+	/*
+	 * how many bytes of partial block should be uptodated by
+	 * reading from disk.
+	 * This is used to perform a read component of RMW (read-modify-write).
+	 */
+	uint32_t (*count_to_uptodate)(call_frame_t *frame, struct object_cipher_info *object);
+	struct avec_config *(*get_config)(call_frame_t *frame);
+};
+
+struct data_cipher_alg {
+	gf_boolean_t atomic; /* true means that algorithm requires
+				to pad data before cipher transform */
+	gf_boolean_t should_pad; /* true means that algorithm requires
+				    to pad the end of file with extra-data */
+	uint32_t blkbits; /* blksize = 1 << blkbits */
+	/*
+	 * any preliminary sanity checks goes here
+	 */
+	int32_t (*init)(void);
+	/*
+	 * set alg-mode specific inode info
+	 */
+	int32_t (*set_private)(struct crypt_inode_info *info,
+			       struct master_cipher_info *master);
+	/*
+	 * check alg-mode specific data key
+	 */
+	int32_t (*check_key)(uint32_t key_size);
+	void (*set_iv)(off_t offset, struct object_cipher_info *object);
+	int32_t (*encrypt)(const unsigned char *from, unsigned char *to,
+			   size_t length, off_t offset, const int enc,
+			   struct object_cipher_info *object);
+};
+
+/*
+ * version-dependent metadata loader
+ */
+struct crypt_mtd_loader {
+	/*
+	 * return core format size
+	 */
+	size_t (*format_size)(mtd_op_t op, size_t old_size);
+	/*
+	 * pack version-specific metadata of an object
+	 * at ->create()
+	 */
+	int32_t (*create_format)(unsigned char *wire,
+				 loc_t *loc,
+				 struct crypt_inode_info *info,
+				 struct master_cipher_info *master);
+	/*
+	 * extract version-specific metadata of an object
+	 * at ->open() time
+	 */
+	int32_t (*open_format)(unsigned char *wire,
+			       int32_t len,
+			       loc_t *loc,
+			       struct crypt_inode_info *info,
+			       struct master_cipher_info *master,
+			       crypt_local_t *local,
+			       gf_boolean_t load_info);
+	int32_t (*update_format)(unsigned char *new,
+				 unsigned char *old,
+				 size_t old_len,
+				 int32_t mac_idx,
+				 mtd_op_t op,
+				 loc_t *loc,
+				 struct crypt_inode_info *info,
+				 struct master_cipher_info *master,
+				 crypt_local_t *local);
+};
+
+typedef int32_t (*end_writeback_handler_t)(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);
+typedef void (*linkop_wind_handler_t)(call_frame_t *frame, xlator_t *this);
+typedef void (*linkop_unwind_handler_t)(call_frame_t *frame);
+
+
+/* Declarations */
+
+/* keys.c */
+extern struct crypt_key crypt_keys[LAST_KEY_TYPE];
+int32_t get_nmtd_vol_key(struct master_cipher_info *master);
+int32_t get_nmtd_link_key(loc_t *loc,
+			  struct master_cipher_info *master,
+			  unsigned char *result);
+int32_t get_emtd_file_key(struct crypt_inode_info *info,
+			  struct master_cipher_info *master,
+			  unsigned char *result);
+int32_t get_data_file_key(struct crypt_inode_info *info,
+			  struct master_cipher_info *master,
+			  uint32_t keysize,
+			  unsigned char *key);
+/* data.c */
+extern struct data_cipher_alg data_cipher_algs[LAST_CIPHER_ALG][LAST_CIPHER_MODE];
+void encrypt_aligned_iov(struct object_cipher_info *object,
+			 struct iovec *vec,
+			 int count,
+			 off_t off);
+void decrypt_aligned_iov(struct object_cipher_info *object,
+			 struct iovec *vec,
+			 int count,
+			 off_t off);
+int32_t align_iov_by_atoms(xlator_t *this,
+			   crypt_local_t *local,
+			   struct object_cipher_info *object,
+			   struct iovec *vec /* input vector */,
+			   int32_t count /* number of vec components */,
+			   struct iovec *avec /* aligned vector */,
+			   char **blocks /* pool of blocks */,
+			   uint32_t *blocks_allocated,
+			   struct avec_config *conf);
+int32_t set_config_avec_data(xlator_t *this,
+			     crypt_local_t *local,
+			     struct avec_config *conf,
+			     struct object_cipher_info *object,
+			     struct iovec *vec,
+			     int32_t vec_count);
+int32_t set_config_avec_hole(xlator_t *this,
+			     crypt_local_t *local,
+			     struct avec_config *conf,
+			     struct object_cipher_info *object,
+			     glusterfs_fop_t fop);
+void set_gap_at_end(call_frame_t *frame, struct object_cipher_info *object,
+		    struct avec_config *conf, atom_data_type dtype);
+void set_config_offsets(call_frame_t *frame,
+			xlator_t *this,
+			uint64_t offset,
+			uint64_t count,
+			atom_data_type dtype,
+			int32_t setup_gap_in_tail);
+
+/* metadata.c */
+extern struct crypt_mtd_loader mtd_loaders [LAST_MTD_LOADER];
+
+int32_t alloc_format(crypt_local_t *local, size_t size);
+int32_t alloc_format_create(crypt_local_t *local);
+void free_format(crypt_local_t *local);
+size_t format_size(mtd_op_t op, size_t old_size);
+size_t new_format_size(void);
+int32_t open_format(unsigned char *str, int32_t len, loc_t *loc,
+		    struct crypt_inode_info *info,
+		    struct master_cipher_info *master, crypt_local_t *local,
+		    gf_boolean_t load_info);
+int32_t update_format(unsigned char *new, unsigned char *old,
+		      size_t old_len, int32_t mac_idx, mtd_op_t op, loc_t *loc,
+		      struct crypt_inode_info *info,
+		      struct master_cipher_info *master,
+		      crypt_local_t *local);
+int32_t create_format(unsigned char *wire,
+		      loc_t *loc,
+		      struct crypt_inode_info *info,
+		      struct master_cipher_info *master);
+
+/* atom.c */
+struct rmw_atom *atom_by_types(atom_data_type data,
+			       atom_locality_type locality);
+void submit_partial(call_frame_t *frame,
+		    xlator_t *this,
+		    fd_t *fd,
+		    atom_locality_type ltype);
+void submit_full(call_frame_t *frame, xlator_t *this);
+
+/* crypt.c */
+
+end_writeback_handler_t dispatch_end_writeback(glusterfs_fop_t fop);
+static size_t iovec_get_size(struct iovec *vec, uint32_t count);
+void set_local_io_params_writev(call_frame_t *frame,
+				struct object_cipher_info *object,
+				struct rmw_atom *atom, off_t io_offset,
+				uint32_t io_size);
+void link_wind(call_frame_t *frame, xlator_t *this);
+void unlink_wind(call_frame_t *frame, xlator_t *this);
+void link_unwind(call_frame_t *frame);
+void unlink_unwind(call_frame_t *frame);
+void rename_wind(call_frame_t *frame, xlator_t *this);
+void rename_unwind(call_frame_t *frame);
+
+/* Inline functions */
+
+static inline size_t iovec_get_size(struct iovec *vec, uint32_t count)
+{
+	int i;
+	size_t size = 0;
+	for (i = 0; i < count; i++)
+		size += vec[i].iov_len;
+	return size;
+}
+
+static inline int32_t crypt_xlator_id(void)
+{
+	return CRYPT_XLATOR_ID;
+}
+
+static inline mtd_loader_id current_mtd_loader(void)
+{
+	return MTD_LOADER_V1;
+}
+
+static inline uint32_t master_key_size (void)
+{
+	return crypt_keys[MASTER_VOL_KEY].len >> 3;
+}
+
+static inline uint32_t nmtd_vol_key_size (void)
+{
+	return crypt_keys[NMTD_VOL_KEY].len >> 3;
+}
+
+static inline uint32_t alg_mode_blkbits(cipher_alg_t alg,
+					cipher_mode_t mode)
+{
+	return data_cipher_algs[alg][mode].blkbits;
+}
+
+static inline uint32_t alg_mode_blksize(cipher_alg_t alg,
+					cipher_mode_t mode)
+{
+	return 1 << alg_mode_blkbits(alg, mode);
+}
+
+static inline gf_boolean_t alg_mode_atomic(cipher_alg_t alg,
+					   cipher_mode_t mode)
+{
+	return data_cipher_algs[alg][mode].atomic;
+}
+
+static inline gf_boolean_t alg_mode_should_pad(cipher_alg_t alg,
+					       cipher_mode_t mode)
+{
+	return data_cipher_algs[alg][mode].should_pad;
+}
+
+static inline uint32_t master_alg_blksize(struct master_cipher_info *mr)
+{
+	return alg_mode_blksize(mr->m_alg, mr->m_mode);
+}
+
+static inline uint32_t master_alg_blkbits(struct master_cipher_info *mr)
+{
+	return alg_mode_blkbits(mr->m_alg, mr->m_mode);
+}
+
+static inline gf_boolean_t master_alg_atomic(struct master_cipher_info *mr)
+{
+	return alg_mode_atomic(mr->m_alg, mr->m_mode);
+}
+
+static inline gf_boolean_t master_alg_should_pad(struct master_cipher_info *mr)
+{
+	return alg_mode_should_pad(mr->m_alg, mr->m_mode);
+}
+
+static inline uint32_t object_alg_blksize(struct object_cipher_info *ob)
+{
+	return alg_mode_blksize(ob->o_alg, ob->o_mode);
+}
+
+static inline uint32_t object_alg_blkbits(struct object_cipher_info *ob)
+{
+	return alg_mode_blkbits(ob->o_alg, ob->o_mode);
+}
+
+static inline gf_boolean_t object_alg_atomic(struct object_cipher_info *ob)
+{
+	return alg_mode_atomic(ob->o_alg, ob->o_mode);
+}
+
+static inline gf_boolean_t object_alg_should_pad(struct object_cipher_info *ob)
+{
+	return alg_mode_should_pad(ob->o_alg, ob->o_mode);
+}
+
+static inline uint32_t aes_raw_key_size(struct master_cipher_info *master)
+{
+	return master->m_dkey_size >> 3;
+}
+
+static inline struct avec_config *get_hole_conf(call_frame_t *frame)
+{
+	return &(((crypt_local_t *)frame->local)->hole_conf);
+}
+
+static inline struct avec_config *get_data_conf(call_frame_t *frame)
+{
+	return &(((crypt_local_t *)frame->local)->data_conf);
+}
+
+static inline int32_t get_atom_bits (struct object_cipher_info *object)
+{
+	return object->o_block_bits;
+}
+
+static inline int32_t get_atom_size (struct object_cipher_info *object)
+{
+	return 1 << get_atom_bits(object);
+}
+
+static inline int32_t has_head_block(struct avec_config *conf)
+{
+	return conf->off_in_head ||
+		(conf->acount == 1 && conf->off_in_tail);
+}
+
+static inline int32_t has_tail_block(struct avec_config *conf)
+{
+	return conf->off_in_tail && conf->acount > 1;
+}
+
+static inline int32_t has_full_blocks(struct avec_config *conf)
+{
+	return conf->nr_full_blocks;
+}
+
+static inline int32_t should_submit_head_block(struct avec_config *conf)
+{
+	return has_head_block(conf) && (conf->cursor == 0);
+}
+
+static inline int32_t should_submit_tail_block(struct avec_config *conf)
+{
+	return has_tail_block(conf) && (conf->cursor == conf->acount - 1);
+}
+
+static inline int32_t should_submit_full_block(struct avec_config *conf)
+{
+	uint32_t start = has_head_block(conf) ? 1 : 0;
+
+	return has_full_blocks(conf) &&
+		conf->cursor >= start &&
+		conf->cursor < start + conf->nr_full_blocks;
+}
+
+#if DEBUG_CRYPT
+static inline void crypt_check_input_len(size_t len,
+					 struct object_cipher_info *object)
+{
+	if (object_alg_should_pad(object) && (len & (object_alg_blksize(object) - 1)))
+		gf_log ("crypt", GF_LOG_DEBUG, "bad input len: %d", (int)len);
+}
+
+static inline void check_head_block(struct avec_config *conf)
+{
+	if (!has_head_block(conf))
+		gf_log("crypt", GF_LOG_DEBUG, "not a head atom");
+}
+
+static inline void check_tail_block(struct avec_config *conf)
+{
+	if (!has_tail_block(conf))
+		gf_log("crypt", GF_LOG_DEBUG, "not a tail atom");
+}
+
+static inline void check_full_block(struct avec_config *conf)
+{
+	if (!has_full_blocks(conf))
+		gf_log("crypt", GF_LOG_DEBUG, "not a full atom");
+}
+
+static inline void check_cursor_head(struct avec_config *conf)
+{
+	if (!has_head_block(conf))
+		gf_log("crypt",
+		       GF_LOG_DEBUG, "Illegal call of head atom method");
+	else if (conf->cursor != 0)
+		gf_log("crypt",
+		       GF_LOG_DEBUG, "Cursor (%d) is not at head atom",
+		       conf->cursor);
+}
+
+static inline void check_cursor_full(struct avec_config *conf)
+{
+	if (!has_full_blocks(conf))
+		gf_log("crypt",
+		       GF_LOG_DEBUG, "Illegal call of full atom method");
+	if (has_head_block(conf) && (conf->cursor == 0))
+		gf_log("crypt",
+		       GF_LOG_DEBUG, "Cursor is not at full atom");
+}
+
+/*
+ * FIXME: use avec->iov_len to check setup
+ */
+static inline int data_local_invariant(crypt_local_t *local)
+{
+	return 0;
+}
+
+#else
+#define crypt_check_input_len(len, object) noop
+#define check_head_block(conf) noop
+#define check_tail_block(conf) noop
+#define check_full_block(conf) noop
+#define check_cursor_head(conf) noop
+#define check_cursor_full(conf) noop
+
+#endif /* DEBUG_CRYPT */
+
+static inline struct avec_config *conf_by_type(call_frame_t *frame,
+					       atom_data_type dtype)
+{
+	struct avec_config *conf = NULL;
+
+	switch (dtype) {
+	case HOLE_ATOM:
+		conf = get_hole_conf(frame);
+		break;
+	case DATA_ATOM:
+		conf = get_data_conf(frame);
+		break;
+	default:
+		gf_log("crypt", GF_LOG_DEBUG, "bad atom type");
+	}
+	return conf;
+}
+
+static inline uint32_t nr_calls_head(struct avec_config *conf)
+{
+	return has_head_block(conf) ? 1 : 0;
+}
+
+static inline uint32_t nr_calls_tail(struct avec_config *conf)
+{
+	return has_tail_block(conf) ? 1 : 0;
+}
+
+static inline uint32_t nr_calls_full(struct avec_config *conf)
+{
+	switch(conf->type) {
+	case HOLE_ATOM:
+		return has_full_blocks(conf);
+	case DATA_ATOM:
+		return has_full_blocks(conf) ?
+			logical_blocks_occupied(0,
+						conf->nr_full_blocks,
+						MAX_IOVEC_BITS)	: 0;
+	default:
+		gf_log("crypt", GF_LOG_DEBUG, "bad atom data type");
+		return 0;
+	}
+}
+
+static inline uint32_t nr_calls(struct avec_config *conf)
+{
+	return nr_calls_head(conf) + nr_calls_tail(conf) + nr_calls_full(conf);
+}
+
+static inline uint32_t nr_calls_data(call_frame_t *frame)
+{
+	return nr_calls(get_data_conf(frame));
+}
+
+static inline uint32_t nr_calls_hole(call_frame_t *frame)
+{
+	return nr_calls(get_hole_conf(frame));
+}
+
+static inline void get_one_call_nolock(call_frame_t *frame)
+{
+	crypt_local_t *local = frame->local;
+
+	++local->nr_calls;
+
+	//gf_log("crypt", GF_LOG_DEBUG, "get %d calls", 1);
+}
+
+static inline void get_one_call(call_frame_t *frame)
+{
+	crypt_local_t *local = frame->local;
+
+	LOCK(&local->call_lock);
+	get_one_call_nolock(frame);
+	UNLOCK(&local->call_lock);
+}
+
+static inline void get_nr_calls_nolock(call_frame_t *frame, int32_t nr)
+{
+	crypt_local_t *local = frame->local;
+
+	local->nr_calls += nr;
+
+	//gf_log("crypt", GF_LOG_DEBUG, "get %d calls", nr);
+}
+
+static inline void get_nr_calls(call_frame_t *frame, int32_t nr)
+{
+	crypt_local_t *local = frame->local;
+
+	LOCK(&local->call_lock);
+	get_nr_calls_nolock(frame, nr);
+	UNLOCK(&local->call_lock);
+}
+
+static inline int put_one_call(crypt_local_t *local)
+{
+	uint32_t last = 0;
+
+	LOCK(&local->call_lock);
+	if (--local->nr_calls == 0)
+		last = 1;
+
+	//gf_log("crypt", GF_LOG_DEBUG, "put %d calls", 1);
+
+	UNLOCK(&local->call_lock);
+	return last;
+}
+
+static inline int is_appended_write(call_frame_t *frame)
+{
+	crypt_local_t *local = frame->local;
+	struct avec_config *conf = get_data_conf(frame);
+
+	return conf->orig_offset + conf->orig_size > local->old_file_size;
+}
+
+static inline int is_ordered_mode(call_frame_t *frame)
+{
+#if 0
+	crypt_local_t *local = frame->local;
+	return local->fop == GF_FOP_FTRUNCATE ||
+		(local->fop == GF_FOP_WRITE && is_appended_write(frame));
+#endif
+	return 1;
+}
+
+static inline int32_t hole_conv_completed(crypt_local_t *local)
+{
+	struct avec_config *conf = &local->hole_conf;
+	return conf->cursor == conf->acount;
+}
+
+static inline int32_t data_write_in_progress(crypt_local_t *local)
+{
+	return local->active_setup == DATA_ATOM;
+}
+
+static inline int32_t parent_is_crypt_xlator(call_frame_t *frame,
+					     xlator_t *this)
+{
+	return frame->parent->this == this;
+}
+
+static inline linkop_wind_handler_t linkop_wind_dispatch(glusterfs_fop_t fop)
+{
+	switch(fop){
+	case GF_FOP_LINK:
+		return link_wind;
+	case GF_FOP_UNLINK:
+		return unlink_wind;
+	case GF_FOP_RENAME:
+		return rename_wind;
+	default:
+		gf_log("crypt", GF_LOG_ERROR, "Bad link operation %d", fop);
+		return NULL;
+	}
+}
+
+static inline linkop_unwind_handler_t linkop_unwind_dispatch(glusterfs_fop_t fop)
+{
+	switch(fop){
+	case GF_FOP_LINK:
+		return link_unwind;
+	case GF_FOP_UNLINK:
+		return unlink_unwind;
+	case GF_FOP_RENAME:
+		return rename_unwind;
+	default:
+		gf_log("crypt", GF_LOG_ERROR, "Bad link operation %d", fop);
+		return NULL;		
+	}	
+}
+
+static inline mtd_op_t linkop_mtdop_dispatch(glusterfs_fop_t fop)
+{
+	switch (fop) {
+	case GF_FOP_LINK:
+		return MTD_APPEND;
+	case GF_FOP_UNLINK:
+		return MTD_CUT;
+	case GF_FOP_RENAME:
+		return MTD_OVERWRITE;
+	default:
+		gf_log("crypt", GF_LOG_WARNING, "Bad link operation %d", fop);
+		return MTD_LAST_OP;
+	}
+}
+
+#endif /* __CRYPT_H__ */
+
+/*
+  Local variables:
+  c-indentation-style: "K&R"
+  mode-name: "LC"
+  c-basic-offset: 8
+  tab-width: 8
+  fill-column: 80
+  scroll-step: 1
+  End:
+*/
diff --git a/xlators/encryption/crypt/src/data.c b/xlators/encryption/crypt/src/data.c
new file mode 100644
index 000000000..762fa554a
--- /dev/null
+++ b/xlators/encryption/crypt/src/data.c
@@ -0,0 +1,769 @@
+/*
+  Copyright (c) 2008-2013 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 "defaults.h"
+#include "crypt-common.h"
+#include "crypt.h"
+
+static void set_iv_aes_xts(off_t offset, struct object_cipher_info *object)
+{
+	unsigned char *ivec;
+
+	ivec = object->u.aes_xts.ivec;
+
+	/* convert the tweak into a little-endian byte
+	 * array (IEEE P1619/D16, May 2007, section 5.1)
+	 */
+
+	*((uint64_t *)ivec) = htole64(offset);
+
+	/* ivec is padded with zeroes */
+}
+
+static int32_t aes_set_keys_common(unsigned char *raw_key, uint32_t key_size,
+				   AES_KEY *keys)
+{
+	int32_t ret;
+
+	ret = AES_set_encrypt_key(raw_key,
+				  key_size,
+				  &keys[AES_ENCRYPT]);
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR, "Set encrypt key failed");
+		return ret;
+	}
+	ret = AES_set_decrypt_key(raw_key,
+				  key_size,
+				  &keys[AES_DECRYPT]);
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR, "Set decrypt key failed");
+		return ret;
+	}
+	return 0;
+}
+
+/*
+ * set private cipher info for xts mode
+ */
+static int32_t set_private_aes_xts(struct crypt_inode_info *info,
+				   struct master_cipher_info *master)
+{
+	int ret;
+	struct object_cipher_info *object = get_object_cinfo(info);
+	unsigned char *data_key;
+	uint32_t subkey_size;
+
+	/* init tweak value */
+	memset(object->u.aes_xts.ivec, 0, 16);
+
+	data_key = GF_CALLOC(1, object->o_dkey_size, gf_crypt_mt_key);
+	if (!data_key)
+		return ENOMEM;
+
+	/*
+	 * retrieve data keying meterial
+	 */
+	ret = get_data_file_key(info, master, object->o_dkey_size, data_key);
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR, "Failed to retrieve data key");
+		GF_FREE(data_key);
+		return ret;
+	}
+	/*
+	 * parse compound xts key
+	 */
+	subkey_size = object->o_dkey_size >> 4; /* (xts-key-size-in-bytes / 2) */
+	/*
+	 * install key for data encryption
+	 */
+	ret = aes_set_keys_common(data_key,
+				  subkey_size << 3, object->u.aes_xts.dkey);
+	if (ret) {
+		GF_FREE(data_key);
+		return ret;
+	}
+	/*
+	 * set up key used to encrypt tweaks
+	 */
+	ret = AES_set_encrypt_key(data_key + subkey_size,
+				  object->o_dkey_size / 2,
+				  &object->u.aes_xts.tkey);
+	if (ret < 0)
+		gf_log("crypt", GF_LOG_ERROR, "Set tweak key failed");
+
+	GF_FREE(data_key);
+	return ret;
+}
+
+static int32_t aes_xts_init(void)
+{
+	cassert(AES_BLOCK_SIZE == (1 << AES_BLOCK_BITS));
+	return 0;
+}
+
+static int32_t check_key_aes_xts(uint32_t keysize)
+{
+	switch(keysize) {
+	case 256:
+	case 512:
+		return 0;
+	default:
+		break;
+	}
+	return -1;
+}
+
+static int32_t encrypt_aes_xts(const unsigned char *from,
+			       unsigned char *to, size_t length,
+			       off_t offset, const int enc,
+			       struct object_cipher_info *object)
+{
+	XTS128_CONTEXT ctx;
+	if (enc) {
+		ctx.key1 = &object->u.aes_xts.dkey[AES_ENCRYPT];
+		ctx.block1 = (block128_f)AES_encrypt;
+	}
+	else {
+		ctx.key1 = &object->u.aes_xts.dkey[AES_DECRYPT];
+		ctx.block1 = (block128_f)AES_decrypt;
+	}
+	ctx.key2 = &object->u.aes_xts.tkey;
+	ctx.block2 = (block128_f)AES_encrypt;
+
+	return CRYPTO_xts128_encrypt(&ctx,
+				     object->u.aes_xts.ivec,
+				     from,
+				     to,
+				     length, enc);
+}
+
+/*
+ * Cipher input chunk @from of length @len;
+ * @to: result of cipher transform;
+ * @off: offset in a file (must be cblock-aligned);
+ */
+static void cipher_data(struct object_cipher_info *object,
+			char *from,
+			char *to,
+			off_t off,
+			size_t len,
+			const int enc)
+{
+	crypt_check_input_len(len, object);
+
+#if TRIVIAL_TFM && DEBUG_CRYPT
+	return;
+#endif
+	data_cipher_algs[object->o_alg][object->o_mode].set_iv(off, object);
+	data_cipher_algs[object->o_alg][object->o_mode].encrypt
+		((const unsigned char *)from,
+		 (unsigned char *)to,
+		 len,
+		 off,
+		 enc,
+		 object);
+}
+
+#define MAX_CIPHER_CHUNK (1 << 30)
+
+/*
+ * Do cipher (encryption/decryption) transform of a
+ * continuous region of memory.
+ *
+ * @len: a number of bytes to transform;
+ * @buf: data to transform;
+ * @off: offset in a file, should be block-aligned
+ *       for atomic cipher modes and ksize-aligned
+ *       for other modes).
+ * @dir: direction of transform (encrypt/decrypt).
+ */
+static void cipher_region(struct object_cipher_info *object,
+			  char *from,
+			  char *to,
+			  off_t off,
+			  size_t len,
+			  int dir)
+{
+	while (len > 0) {
+		size_t to_cipher;
+
+		to_cipher = len;
+		if (to_cipher > MAX_CIPHER_CHUNK)
+			to_cipher = MAX_CIPHER_CHUNK;
+
+		/* this will reset IV */
+		cipher_data(object,
+			    from,
+			    to,
+			    off,
+			    to_cipher,
+			    dir);
+		from += to_cipher;
+		to   += to_cipher;
+		off  += to_cipher;
+		len  -= to_cipher;
+	}
+}
+
+/*
+ * Do cipher transform (encryption/decryption) of
+ * plaintext/ciphertext represented by @vec.
+ *
+ * Pre-conditions: @vec represents a continuous piece
+ * of data in a file at offset @off to be ciphered
+ * (encrypted/decrypted).
+ * @count is the number of vec's components. All the
+ * components must be block-aligned, the caller is
+ * responsible for this. @dir is "direction" of
+ * transform (encrypt/decrypt).
+ */
+static void cipher_aligned_iov(struct object_cipher_info *object,
+			       struct iovec *vec,
+			       int count,
+			       off_t off,
+			       int32_t dir)
+{
+	int i;
+	int len = 0;
+
+	for (i = 0; i < count; i++) {
+		cipher_region(object,
+			      vec[i].iov_base,
+			      vec[i].iov_base,
+			      off + len,
+			      vec[i].iov_len,
+			      dir);
+		len += vec[i].iov_len;
+	}
+}
+
+void encrypt_aligned_iov(struct object_cipher_info *object,
+			 struct iovec *vec,
+			 int count,
+			 off_t off)
+{
+	cipher_aligned_iov(object, vec, count, off, 1);
+}
+
+void decrypt_aligned_iov(struct object_cipher_info *object,
+			 struct iovec *vec,
+			 int count,
+			 off_t off)
+{
+	cipher_aligned_iov(object, vec, count, off, 0);
+}
+
+#if DEBUG_CRYPT
+static void compound_stream(struct iovec *vec, int count, char *buf, off_t skip)
+{
+	int i;
+	int off = 0;
+	for (i = 0; i < count; i++) {
+		memcpy(buf + off,
+		       vec[i].iov_base + skip,
+		       vec[i].iov_len - skip);
+
+		off += (vec[i].iov_len - skip);
+		skip = 0;
+	}
+}
+
+static void check_iovecs(struct iovec *vec, int cnt,
+			 struct iovec *avec, int acnt, uint32_t off_in_head)
+{
+	char *s1, *s2;
+	uint32_t size, asize;
+
+	size = iovec_get_size(vec, cnt);
+	asize = iovec_get_size(avec, acnt) - off_in_head;
+	if (size != asize) {
+		gf_log("crypt", GF_LOG_DEBUG, "size %d is not eq asize %d",
+		       size, asize);
+		return;
+	}
+	s1 = GF_CALLOC(1, size, gf_crypt_mt_data);
+	if (!s1) {
+		gf_log("crypt", GF_LOG_DEBUG, "Can not allocate stream ");
+		return;
+	}
+	s2 = GF_CALLOC(1, asize, gf_crypt_mt_data);
+	if (!s2) {
+		GF_FREE(s1);
+		gf_log("crypt", GF_LOG_DEBUG, "Can not allocate stream ");
+		return;
+	}
+	compound_stream(vec, cnt, s1, 0);
+	compound_stream(avec, acnt, s2, off_in_head);
+	if (memcmp(s1, s2, size))
+		gf_log("crypt", GF_LOG_DEBUG, "chunks of different data");
+	GF_FREE(s1);
+	GF_FREE(s2);
+}
+
+#else
+#define check_iovecs(vec, count, avec, avecn, off) noop
+#endif /* DEBUG_CRYPT */
+
+static char *data_alloc_block(xlator_t *this, crypt_local_t *local,
+			      int32_t block_size)
+{
+	struct iobuf *iobuf = NULL;
+
+	iobuf = iobuf_get2(this->ctx->iobuf_pool, block_size);
+	if (!iobuf) {
+		gf_log("crypt", GF_LOG_ERROR,
+		       "Failed to get iobuf");
+		return NULL;
+	}
+	if (!local->iobref_data) {
+		local->iobref_data = iobref_new();
+		if (!local->iobref_data) {
+			gf_log("crypt", GF_LOG_ERROR,
+			       "Failed to get iobref");
+			iobuf_unref(iobuf);
+			return NULL;
+		}
+	}
+	iobref_add(local->iobref_data, iobuf);
+	return iobuf->ptr;
+}
+
+/*
+ * Compound @avec, which represent the same data
+ * chunk as @vec, but has aligned components of
+ * specified block size. Alloc blocks, if needed.
+ * In particular, incomplete head and tail blocks
+ * must be allocated.
+ * Put number of allocated blocks to @num_blocks.
+ *
+ * Example:
+ *
+ * input: data chunk represented by 4 components
+ * [AB],[BC],[CD],[DE];
+ * output: 5 logical blocks (0, 1, 2, 3, 4).
+ *
+ *   A     B       C     D               E
+ *   *-----*+------*-+---*----+--------+-*
+ *   |     ||      | |   |    |        | |
+ * *-+-----+*------+-*---+----*--------*-+------*
+ * 0        1        2        3        4
+ *
+ * 0 - incomplete compound (head);
+ * 1, 2 - full compound;
+ * 3 - full non-compound (the case of reuse);
+ * 4 - incomplete non-compound (tail).
+ */
+int32_t align_iov_by_atoms(xlator_t *this,
+			   crypt_local_t *local,
+			   struct object_cipher_info *object,
+			   struct iovec *vec /* input vector */,
+			   int32_t count /* number of vec components */,
+			   struct iovec *avec /* aligned vector */,
+			   char **blocks /* pool of blocks */,
+			   uint32_t *blocks_allocated,
+			   struct avec_config *conf)
+{
+	int vecn = 0;      /* number of the current component in vec */
+	int avecn = 0;     /* number of the current component in avec */
+	off_t vec_off = 0; /* offset in the current vec component,
+			    * i.e. the number of bytes have already
+			    * been copied */
+	int32_t block_size = get_atom_size(object);
+	size_t to_process; /* number of vec's bytes to copy and(or) re-use */
+	int32_t off_in_head = conf->off_in_head;
+
+	to_process = iovec_get_size(vec, count);
+
+	while (to_process > 0) {
+		if (off_in_head ||
+		    vec[vecn].iov_len - vec_off < block_size) {
+			/*
+			 * less than block_size:
+			 * the case of incomplete (head or tail),
+			 * or compound block
+			 */
+			size_t copied = 0;
+			/*
+			 * populate the pool with a new block
+			 */
+			blocks[*blocks_allocated] = data_alloc_block(this,
+								    local,
+								    block_size);
+			if (!blocks[*blocks_allocated])
+				return -ENOMEM;
+			memset(blocks[*blocks_allocated], 0, off_in_head);
+			/*
+			 * fill the block with vec components
+			 */
+			do {
+				size_t to_copy;
+
+				to_copy = vec[vecn].iov_len - vec_off;
+				if (to_copy > block_size - off_in_head)
+					to_copy = block_size - off_in_head;
+
+				memcpy(blocks[*blocks_allocated] + off_in_head + copied,
+				       vec[vecn].iov_base + vec_off,
+				       to_copy);
+
+				copied += to_copy;
+				to_process -= to_copy;
+
+				vec_off += to_copy;
+				if (vec_off == vec[vecn].iov_len) {
+					/* finished with this vecn */
+					vec_off = 0;
+					vecn++;
+				}
+			} while (copied < (block_size - off_in_head) && to_process > 0);
+			/*
+			 * update avec
+			 */
+			avec[avecn].iov_len = off_in_head + copied;
+			avec[avecn].iov_base = blocks[*blocks_allocated];
+
+			(*blocks_allocated)++;
+			off_in_head = 0;
+		} else {
+			/*
+			 * the rest of the current vec component
+			 * is not less than block_size, so reuse
+			 * the memory buffer of the component.
+			 */
+			size_t to_reuse;
+			to_reuse = (to_process > block_size ?
+				    block_size :
+				    to_process);
+			avec[avecn].iov_len = to_reuse;
+			avec[avecn].iov_base = vec[vecn].iov_base + vec_off;
+
+			vec_off += to_reuse;
+			if (vec_off == vec[vecn].iov_len) {
+				/* finished with this vecn */
+				vec_off = 0;
+				vecn++;
+			}
+			to_process -= to_reuse;
+		}
+		avecn++;
+	}
+	check_iovecs(vec, count, avec, avecn, conf->off_in_head);
+	return 0;
+}
+
+/*
+ * allocate and setup aligned vector for data submission
+ * Pre-condition: @conf is set.
+ */
+int32_t set_config_avec_data(xlator_t *this,
+			     crypt_local_t *local,
+			     struct avec_config *conf,
+			     struct object_cipher_info *object,
+			     struct iovec *vec,
+			     int32_t vec_count)
+{
+	int32_t ret = ENOMEM;
+	struct iovec *avec;
+	char **pool;
+	uint32_t blocks_in_pool = 0;
+
+	conf->type = DATA_ATOM;
+
+	avec = GF_CALLOC(conf->acount, sizeof(*avec), gf_crypt_mt_iovec);
+	if (!avec)
+		return ret;
+	pool = GF_CALLOC(conf->acount, sizeof(pool), gf_crypt_mt_char);
+	if (!pool) {
+		GF_FREE(avec);
+		return ret;
+	}
+	if (!vec) {
+		/*
+		 * degenerated case: no data
+		 */
+		pool[0] = data_alloc_block(this, local, get_atom_size(object));
+		if (!pool[0])
+			goto free;
+		blocks_in_pool = 1;
+		avec->iov_base = pool[0];
+		avec->iov_len = conf->off_in_tail;
+	}
+	else {
+		ret = align_iov_by_atoms(this, local, object, vec, vec_count,
+					 avec, pool, &blocks_in_pool, conf);
+		if (ret)
+			goto free;
+	}
+	conf->avec = avec;
+	conf->pool = pool;
+	conf->blocks_in_pool = blocks_in_pool;
+	return 0;
+ free:
+	GF_FREE(avec);
+	GF_FREE(pool);
+	return ret;
+}
+
+/*
+ * allocate and setup aligned vector for hole submission
+ */
+int32_t set_config_avec_hole(xlator_t *this,
+			     crypt_local_t *local,
+			     struct avec_config *conf,
+			     struct object_cipher_info *object,
+			     glusterfs_fop_t fop)
+{
+	uint32_t i, idx;
+	struct iovec *avec;
+	char **pool;
+	uint32_t num_blocks;
+	uint32_t blocks_in_pool = 0;
+
+	conf->type = HOLE_ATOM;
+
+	num_blocks = conf->acount -
+		(conf->nr_full_blocks ? conf->nr_full_blocks - 1 : 0);
+
+	switch (fop) {
+	case GF_FOP_WRITE:
+		/*
+		 * hole goes before data
+		 */
+		if (num_blocks == 1 && conf->off_in_tail != 0)
+		/*
+		 * we won't submit a hole which fits into
+		 * a data atom: this part of hole will be
+		 * submitted with data write
+		 */
+			return 0;
+		break;
+	case GF_FOP_FTRUNCATE:
+		/*
+		 * expanding truncate, hole goes after data,
+		 * and will be submited in any case.
+ 		 */
+		break;
+	default:
+		gf_log("crypt", GF_LOG_WARNING,
+		       "bad file operation %d", fop);
+		return 0;
+	}
+	avec = GF_CALLOC(num_blocks, sizeof(*avec), gf_crypt_mt_iovec);
+	if (!avec)
+		return ENOMEM;
+	pool = GF_CALLOC(num_blocks, sizeof(pool), gf_crypt_mt_char);
+	if (!pool) {
+		GF_FREE(avec);
+		return ENOMEM;
+	}
+	for (i = 0; i < num_blocks; i++) {
+		pool[i] = data_alloc_block(this, local, get_atom_size(object));
+		if (pool[i] == NULL)
+			goto free;
+		blocks_in_pool++;
+	}
+	if (has_head_block(conf)) {
+		/* set head block */
+		idx = 0;
+		avec[idx].iov_base = pool[idx];
+		avec[idx].iov_len = get_atom_size(object);
+		memset(avec[idx].iov_base + conf->off_in_head,
+		       0,
+		       get_atom_size(object) - conf->off_in_head);
+	}
+	if (has_tail_block(conf)) {
+		/* set tail block */
+		idx = num_blocks - 1;
+		avec[idx].iov_base = pool[idx];
+		avec[idx].iov_len = get_atom_size(object);
+		memset(avec[idx].iov_base, 0, conf->off_in_tail);
+	}
+	if (has_full_blocks(conf)) {
+		/* set full block */
+		idx = conf->off_in_head ? 1 : 0;
+		avec[idx].iov_base = pool[idx];
+		avec[idx].iov_len = get_atom_size(object);
+		/*
+		 * since we re-use the buffer,
+		 * zeroes will be set every time
+		 * before encryption, see submit_full()
+		 */
+	}
+	conf->avec = avec;
+	conf->pool = pool;
+	conf->blocks_in_pool = blocks_in_pool;
+	return 0;
+ free:
+	GF_FREE(avec);
+	GF_FREE(pool);
+	return ENOMEM;
+}
+
+/* A helper for setting up config of partial atoms (which
+ * participate in read-modify-write sequence).
+ *
+ * Calculate and setup precise amount of "extra-bytes"
+ * that should be uptodated at the end of partial (not
+ * necessarily tail!) block.
+ *
+ * Pre-condition: local->old_file_size is valid!
+ * @conf contains setup, which is enough for correct calculation
+ * of has_tail_block(), ->get_offset().
+ */
+void set_gap_at_end(call_frame_t *frame, struct object_cipher_info *object,
+		    struct avec_config *conf, atom_data_type dtype)
+{
+	uint32_t to_block;
+	crypt_local_t *local = frame->local;
+	uint64_t old_file_size = local->old_file_size;
+	struct rmw_atom *partial = atom_by_types(dtype,
+						 has_tail_block(conf) ?
+						 TAIL_ATOM : HEAD_ATOM);
+
+	if (old_file_size <= partial->offset_at(frame, object))
+		to_block = 0;
+	else {
+		to_block = old_file_size - partial->offset_at(frame, object);
+		if (to_block > get_atom_size(object))
+			to_block = get_atom_size(object);
+	}
+	if (to_block > conf->off_in_tail)
+		conf->gap_in_tail = to_block - conf->off_in_tail;
+	else
+		/*
+		 * nothing to uptodate
+		 */
+		conf->gap_in_tail = 0;
+}
+
+/*
+ * fill struct avec_config with offsets layouts
+ */
+void set_config_offsets(call_frame_t *frame,
+			xlator_t *this,
+			uint64_t offset,
+			uint64_t count,
+			atom_data_type dtype,
+			int32_t set_gap)
+{
+	crypt_local_t *local;
+	struct object_cipher_info *object;
+	struct avec_config *conf;
+	uint32_t resid;
+
+	uint32_t atom_size;
+	uint32_t atom_bits;
+
+	size_t orig_size;
+	off_t orig_offset;
+	size_t expanded_size;
+	off_t aligned_offset;
+
+	uint32_t off_in_head = 0;
+	uint32_t off_in_tail = 0;
+	uint32_t nr_full_blocks;
+	int32_t size_full_blocks;
+
+	uint32_t acount; /* number of alifned components to write.
+			  * The same as number of occupied logical
+			  * blocks (atoms)
+			  */
+	local = frame->local;
+	object = &local->info->cinfo;
+	conf = (dtype == DATA_ATOM ?
+		get_data_conf(frame) : get_hole_conf(frame));
+
+	orig_offset = offset;
+	orig_size = count;
+
+	atom_size = get_atom_size(object);
+	atom_bits = get_atom_bits(object);
+
+	/*
+	 * Round-down the start,
+	 * round-up the end.
+	 */
+	resid = offset & (uint64_t)(atom_size - 1);
+
+	if (resid)
+		off_in_head = resid;
+	aligned_offset = offset - off_in_head;
+	expanded_size = orig_size + off_in_head;
+
+	/* calculate tail,
+	   expand size forward  */
+	resid = (offset + orig_size) & (uint64_t)(atom_size - 1);
+
+	if (resid) {
+		off_in_tail = resid;
+		expanded_size += (atom_size - off_in_tail);
+	}
+	/*
+	 * calculate number of occupied blocks
+	 */
+	acount = expanded_size >> atom_bits;
+	/*
+	 * calculate number of full blocks
+	 */
+	size_full_blocks = expanded_size;
+	if (off_in_head)
+		size_full_blocks -= atom_size;
+	if (off_in_tail && size_full_blocks > 0)
+		size_full_blocks -= atom_size;
+	nr_full_blocks = size_full_blocks >> atom_bits;
+
+	conf->atom_size = atom_size;
+	conf->orig_size = orig_size;
+	conf->orig_offset = orig_offset;
+	conf->expanded_size = expanded_size;
+	conf->aligned_offset = aligned_offset;
+
+	conf->off_in_head = off_in_head;
+	conf->off_in_tail = off_in_tail;
+	conf->nr_full_blocks = nr_full_blocks;
+	conf->acount = acount;
+	/*
+	 * Finally, calculate precise amount of
+	 * "extra-bytes" that should be uptodated
+	 * at the end.
+	 * Only if RMW is expected.
+	 */
+	if (off_in_tail && set_gap)
+		set_gap_at_end(frame, object, conf, dtype);
+}
+
+struct data_cipher_alg data_cipher_algs[LAST_CIPHER_ALG][LAST_CIPHER_MODE] = {
+	[AES_CIPHER_ALG][XTS_CIPHER_MODE] =
+	{ .atomic = _gf_true,
+	  .should_pad = _gf_true,
+	  .blkbits = AES_BLOCK_BITS,
+	  .init = aes_xts_init,
+	  .set_private = set_private_aes_xts,
+	  .check_key = check_key_aes_xts,
+	  .set_iv = set_iv_aes_xts,
+	  .encrypt = encrypt_aes_xts
+	}
+};
+
+/*
+  Local variables:
+  c-indentation-style: "K&R"
+  mode-name: "LC"
+  c-basic-offset: 8
+  tab-width: 8
+  fill-column: 80
+  scroll-step: 1
+  End:
+*/
diff --git a/xlators/encryption/crypt/src/keys.c b/xlators/encryption/crypt/src/keys.c
new file mode 100644
index 000000000..4a1d3bb5a
--- /dev/null
+++ b/xlators/encryption/crypt/src/keys.c
@@ -0,0 +1,302 @@
+/*
+  Copyright (c) 2008-2013 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 "defaults.h"
+#include "crypt-common.h"
+#include "crypt.h"
+
+/* Key hierarchy
+                  
+                   +----------------+
+                   | MASTER_VOL_KEY |
+                   +-------+--------+
+                           |
+                           |
+          +----------------+----------------+
+          |                |                |
+          |                |                |
+  +-------+------+ +-------+-------+ +------+--------+
+  | NMTD_VOL_KEY | | EMTD_FILE_KEY | | DATA_FILE_KEY |
+  +-------+------+ +---------------+ +---------------+
+          |
+          |
+  +-------+-------+
+  | NMTD_LINK_KEY |
+  +---------------+
+
+ */
+
+#if DEBUG_CRYPT
+static void check_prf_iters(uint32_t num_iters)
+{
+	if (num_iters == 0)
+		gf_log ("crypt", GF_LOG_DEBUG,
+			"bad number of prf iterations : %d", num_iters);
+}
+#else
+#define check_prf_iters(num_iters) noop
+#endif /* DEBUG_CRYPT */
+
+unsigned char crypt_fake_oid[16] =
+	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
+
+/*
+ * derive key in the counter mode using
+ * sha256-based HMAC as PRF, see
+ * NIST Special Publication 800-108, 5.1)
+ */
+
+#define PRF_OUTPUT_SIZE  SHA256_DIGEST_LENGTH
+
+static int32_t kderive_init(struct kderive_context *ctx,
+			    const unsigned char *pkey, /* parent key */
+			    uint32_t pkey_size,  /* parent key size */
+			    const unsigned char *idctx, /* id-context */
+			    uint32_t idctx_size,
+			    crypt_key_type type  /* type of child key */)
+{
+	unsigned char *pos;
+	uint32_t llen = strlen(crypt_keys[type].label);	
+	/*
+	 * Compoud the fixed input data for KDF:
+	 * [i]_2 || Label || 0x00 || Id-Context || [L]_2),
+	 * NIST SP 800-108, 5.1
+	 */
+	ctx->fid_len =
+		sizeof(uint32_t) +
+		llen +
+		1 +
+		idctx_size +
+		sizeof(uint32_t);
+
+	ctx->fid = GF_CALLOC(ctx->fid_len, 1, gf_crypt_mt_key);
+	if (!ctx->fid)
+		return ENOMEM;
+	ctx->out_len = round_up(crypt_keys[type].len >> 3,
+				PRF_OUTPUT_SIZE);
+	ctx->out = GF_CALLOC(ctx->out_len, 1, gf_crypt_mt_key);
+	if (!ctx->out) {
+		GF_FREE(ctx->fid);
+		return ENOMEM;
+	}
+	ctx->pkey = pkey;
+	ctx->pkey_len = pkey_size;
+	ctx->ckey_len = crypt_keys[type].len;
+
+	pos = ctx->fid;
+
+	/* counter will be set up in kderive_rfn() */
+	pos += sizeof(uint32_t);
+
+	memcpy(pos, crypt_keys[type].label, llen);
+	pos += llen;
+
+	/* set up zero octet */
+	*pos = 0;
+	pos += 1;
+
+	memcpy(pos, idctx, idctx_size);
+	pos += idctx_size;
+
+	*((uint32_t *)pos) = htobe32(ctx->ckey_len);
+
+	return 0;
+}
+
+static void kderive_update(struct kderive_context *ctx)
+{	
+	uint32_t i;
+	HMAC_CTX hctx;
+	unsigned char *pos = ctx->out;
+	uint32_t *p_iter = (uint32_t *)ctx->fid;
+	uint32_t num_iters = ctx->out_len / PRF_OUTPUT_SIZE;
+
+	check_prf_iters(num_iters);
+
+	HMAC_CTX_init(&hctx);
+	for (i = 0; i < num_iters; i++) {
+		/*
+		 * update the iteration number in the fid
+		 */
+		*p_iter = htobe32(i);
+		HMAC_Init_ex(&hctx,
+			     ctx->pkey, ctx->pkey_len >> 3,
+			     EVP_sha256(),
+			     NULL);
+		HMAC_Update(&hctx, ctx->fid, ctx->fid_len);
+		HMAC_Final(&hctx, pos, NULL);
+
+		pos += PRF_OUTPUT_SIZE;
+	}
+	HMAC_CTX_cleanup(&hctx);
+}
+
+static void kderive_final(struct kderive_context *ctx, unsigned char *child)
+{
+	memcpy(child, ctx->out, ctx->ckey_len >> 3);
+	GF_FREE(ctx->fid);
+	GF_FREE(ctx->out);
+	memset(ctx, 0, sizeof(*ctx));
+}
+
+/*
+ * derive per-volume key for object ids aithentication
+ */
+int32_t get_nmtd_vol_key(struct master_cipher_info *master)
+{
+	int32_t ret;
+	struct kderive_context ctx;
+
+	ret = kderive_init(&ctx,
+			   master->m_key,
+			   master_key_size(),
+			   crypt_fake_oid, sizeof(uuid_t), NMTD_VOL_KEY);
+	if (ret)
+		return ret;
+	kderive_update(&ctx);
+	kderive_final(&ctx, master->m_nmtd_key);
+	return 0;
+}
+
+/*
+ * derive per-link key for aithentication of non-encrypted
+ * meta-data (nmtd)
+ */
+int32_t get_nmtd_link_key(loc_t *loc,
+			  struct master_cipher_info *master,
+			  unsigned char *result)
+{
+	int32_t ret;
+	struct kderive_context ctx;
+
+	ret = kderive_init(&ctx,
+			   master->m_nmtd_key,
+			   nmtd_vol_key_size(),
+			   (const unsigned char *)loc->path,
+			   strlen(loc->path), NMTD_LINK_KEY);
+	if (ret)
+		return ret;
+	kderive_update(&ctx);
+	kderive_final(&ctx, result);
+	return 0;
+}
+
+/*
+ * derive per-file key for encryption and authentication
+ * of encrypted part of metadata (emtd)
+ */
+int32_t get_emtd_file_key(struct crypt_inode_info *info,
+			  struct master_cipher_info *master,
+			  unsigned char *result)
+{
+	int32_t ret;
+	struct kderive_context ctx;
+
+	ret = kderive_init(&ctx,
+			   master->m_key,
+			   master_key_size(),
+			   info->oid, sizeof(uuid_t), EMTD_FILE_KEY);
+	if (ret)
+		return ret;
+	kderive_update(&ctx);
+	kderive_final(&ctx, result);
+	return 0;
+}
+
+static int32_t data_key_type_by_size(uint32_t keysize, crypt_key_type *type)
+{
+	int32_t ret = 0;
+	switch (keysize) {
+	case 256:
+		*type = DATA_FILE_KEY_256;
+		break;
+	case 512:
+		*type = DATA_FILE_KEY_512;
+		break;
+	default:
+		gf_log("crypt", GF_LOG_ERROR, "Unsupported data key size %d",
+		       keysize);
+		ret = ENOTSUP;
+		break;
+	}
+	return ret;
+}
+
+/*
+ * derive per-file key for data encryption
+ */
+int32_t get_data_file_key(struct crypt_inode_info *info,
+			  struct master_cipher_info *master,
+			  uint32_t keysize,
+			  unsigned char *key)
+{
+	int32_t ret;
+	struct kderive_context ctx;
+	crypt_key_type type;
+
+	ret = data_key_type_by_size(keysize, &type);
+	if (ret)
+		return ret;
+	ret = kderive_init(&ctx,
+			   master->m_key,
+			   master_key_size(),
+			   info->oid, sizeof(uuid_t), type);
+	if (ret)
+		return ret;
+	kderive_update(&ctx);
+	kderive_final(&ctx, key);
+	return 0;
+}
+
+/*
+ * NOTE: Don't change existing keys: it will break compatibility;
+ */
+struct crypt_key crypt_keys[LAST_KEY_TYPE] = {
+	[MASTER_VOL_KEY] =
+	{ .len = MASTER_VOL_KEY_SIZE << 3,
+	  .label = "volume-master",
+	},
+	[NMTD_VOL_KEY] =
+	{ .len = NMTD_VOL_KEY_SIZE << 3,
+	  .label = "volume-nmtd-key-generation"
+	},
+	[NMTD_LINK_KEY] =
+	{ .len = 128,
+	  .label = "link-nmtd-authentication"
+	},
+	[EMTD_FILE_KEY] =
+	{ .len = 128,
+	  .label = "file-emtd-encryption-and-auth"
+	},
+	[DATA_FILE_KEY_256] =
+	{ .len = 256,
+	  .label = "file-data-encryption-256"
+	},
+	[DATA_FILE_KEY_512] =
+	{ .len = 512,
+	  .label = "file-data-encryption-512"
+	}
+};
+
+/*
+  Local variables:
+  c-indentation-style: "K&R"
+  mode-name: "LC"
+  c-basic-offset: 8
+  tab-width: 8
+  fill-column: 80
+  scroll-step: 1
+  End:
+*/
diff --git a/xlators/encryption/crypt/src/metadata.c b/xlators/encryption/crypt/src/metadata.c
new file mode 100644
index 000000000..36b14c055
--- /dev/null
+++ b/xlators/encryption/crypt/src/metadata.c
@@ -0,0 +1,605 @@
+/*
+  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 "defaults.h"
+#include "crypt-common.h"
+#include "crypt.h"
+#include "metadata.h"
+
+int32_t alloc_format(crypt_local_t *local, size_t size)
+{
+	if (size > 0) {
+		local->format = GF_CALLOC(1, size, gf_crypt_mt_mtd);
+		if (!local->format)
+			return ENOMEM;
+	}
+	local->format_size = size;
+	return 0;
+}
+
+int32_t alloc_format_create(crypt_local_t *local)
+{
+	return alloc_format(local, new_format_size());
+}
+
+void free_format(crypt_local_t *local)
+{
+	GF_FREE(local->format);
+}
+
+/*
+ * Check compatibility with extracted metadata
+ */
+static int32_t check_file_metadata(struct crypt_inode_info *info)
+{
+	struct object_cipher_info *object = &info->cinfo;
+
+	if (info->nr_minor != CRYPT_XLATOR_ID) {
+		gf_log("crypt", GF_LOG_WARNING,
+		       "unsupported minor subversion %d", info->nr_minor);
+		return EINVAL;
+	}
+	if (object->o_alg > LAST_CIPHER_ALG) {
+		gf_log("crypt", GF_LOG_WARNING,
+		       "unsupported cipher algorithm %d",
+		       object->o_alg);
+		return EINVAL;
+	}
+	if (object->o_mode > LAST_CIPHER_MODE) {
+		gf_log("crypt", GF_LOG_WARNING,
+		       "unsupported cipher mode %d",
+		       object->o_mode);
+		return EINVAL;
+	}
+	if (object->o_block_bits < CRYPT_MIN_BLOCK_BITS ||
+	    object->o_block_bits > CRYPT_MAX_BLOCK_BITS) {
+		gf_log("crypt", GF_LOG_WARNING, "unsupported block bits %d",
+		       object->o_block_bits);
+		return EINVAL;
+	}
+	/* TBD: check data key size */
+	return 0;
+}
+
+static size_t format_size_v1(mtd_op_t op, size_t old_size)
+{
+	
+	switch (op) {
+	case MTD_CREATE:
+		return sizeof(struct mtd_format_v1);
+	case MTD_OVERWRITE:
+		return old_size;
+	case MTD_APPEND:
+		return old_size + NMTD_8_MAC_SIZE;
+	case MTD_CUT:
+		if (old_size > sizeof(struct mtd_format_v1))
+			return old_size - NMTD_8_MAC_SIZE;
+		else
+			return 0;
+	default:
+		gf_log("crypt", GF_LOG_WARNING, "Bad mtd operation");
+		return 0;
+	}
+}
+
+/*
+ * Calculate size of the updated format string.
+ * Returned zero means that we don't need to update the format string.
+ */
+size_t format_size(mtd_op_t op, size_t old_size)
+{
+	size_t versioned;
+
+	versioned = mtd_loaders[current_mtd_loader()].format_size(op,
+				 old_size - sizeof(struct crypt_format));
+	if (versioned != 0)
+		return versioned + sizeof(struct crypt_format);
+	return 0;
+}
+
+/*
+ * size of the format string of newly created file (nr_links = 1)
+ */
+size_t new_format_size(void)
+{
+	return format_size(MTD_CREATE, 0);
+}
+
+/*
+ * Calculate per-link MAC by pathname
+ */
+static int32_t calc_link_mac_v1(struct mtd_format_v1 *fmt,
+				loc_t *loc,
+				unsigned char *result,
+				struct crypt_inode_info *info,
+				struct master_cipher_info *master)
+{
+	int32_t ret;	
+	unsigned char nmtd_link_key[16];
+	CMAC_CTX *cctx;
+	size_t len;
+
+	ret = get_nmtd_link_key(loc, master, nmtd_link_key);
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR, "Can not get nmtd link key");
+		return -1;
+	}
+	cctx = CMAC_CTX_new();
+	if (!cctx) {
+		gf_log("crypt", GF_LOG_ERROR, "CMAC_CTX_new failed");
+		return -1; 
+	}
+	ret = CMAC_Init(cctx, nmtd_link_key, sizeof(nmtd_link_key),
+			EVP_aes_128_cbc(), 0);
+	if (!ret) {
+		gf_log("crypt", GF_LOG_ERROR, "CMAC_Init failed");
+		CMAC_CTX_free(cctx);
+		return -1;
+	}
+	ret = CMAC_Update(cctx, get_NMTD_V1(info), SIZE_OF_NMTD_V1);
+	if (!ret) {
+		gf_log("crypt", GF_LOG_ERROR, "CMAC_Update failed");
+		CMAC_CTX_free(cctx);
+		return -1;
+	}
+	ret = CMAC_Final(cctx, result, &len);
+	CMAC_CTX_free(cctx);
+	if (!ret) {
+		gf_log("crypt", GF_LOG_ERROR, "CMAC_Final failed");
+		return -1;
+	}
+	return 0;
+}
+
+/*
+ * Create per-link MAC of index @idx by pathname
+ */
+static int32_t create_link_mac_v1(struct mtd_format_v1 *fmt,
+				  uint32_t idx,
+				  loc_t *loc,
+				  struct crypt_inode_info *info,
+				  struct master_cipher_info *master)
+{
+	int32_t ret;
+	unsigned char *mac;
+	unsigned char cmac[16];
+
+	mac = get_NMTD_V1_MAC(fmt) + idx * SIZE_OF_NMTD_V1_MAC;
+
+	ret = calc_link_mac_v1(fmt, loc, cmac, info, master);
+	if (ret)
+		return -1;
+	memcpy(mac, cmac, SIZE_OF_NMTD_V1_MAC);
+	return 0;
+}
+
+static int32_t create_format_v1(unsigned char *wire,
+				loc_t *loc,
+				struct crypt_inode_info *info,
+				struct master_cipher_info *master)
+{
+	int32_t ret;
+	struct mtd_format_v1 *fmt;
+	unsigned char mtd_key[16];
+	AES_KEY EMTD_KEY;
+	unsigned char nmtd_link_key[16];
+	uint32_t ad;
+	GCM128_CONTEXT *gctx;
+
+	fmt = (struct mtd_format_v1 *)wire;
+
+	fmt->minor_id = info->nr_minor;
+	fmt->alg_id = AES_CIPHER_ALG;
+	fmt->dkey_factor = master->m_dkey_size >> KEY_FACTOR_BITS;
+	fmt->block_bits = master->m_block_bits;
+	fmt->mode_id = master->m_mode;
+	/*
+	 * retrieve keys for the parts of metadata
+	 */
+	ret = get_emtd_file_key(info, master, mtd_key);
+	if (ret)
+		return ret;
+	ret = get_nmtd_link_key(loc, master, nmtd_link_key);
+	if (ret)
+		return ret;
+	
+	AES_set_encrypt_key(mtd_key, sizeof(mtd_key)*8, &EMTD_KEY);
+
+	gctx = CRYPTO_gcm128_new(&EMTD_KEY, (block128_f)AES_encrypt);
+
+	/* TBD: Check return values */
+
+	CRYPTO_gcm128_setiv(gctx, info->oid, sizeof(uuid_t));
+
+	ad = htole32(MTD_LOADER_V1);
+	ret = CRYPTO_gcm128_aad(gctx, (const unsigned char *)&ad, sizeof(ad));
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR, " CRYPTO_gcm128_aad failed");
+		CRYPTO_gcm128_release(gctx);
+		return ret;
+	}
+	ret = CRYPTO_gcm128_encrypt(gctx,
+				    get_EMTD_V1(fmt),
+				    get_EMTD_V1(fmt),
+				    SIZE_OF_EMTD_V1);
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR, " CRYPTO_gcm128_encrypt failed");
+		CRYPTO_gcm128_release(gctx);
+		return ret;
+	}
+	/*
+	 * set MAC of encrypted part of metadata
+	 */
+	CRYPTO_gcm128_tag(gctx, get_EMTD_V1_MAC(fmt), SIZE_OF_EMTD_V1_MAC);
+	CRYPTO_gcm128_release(gctx);
+	/*
+	 * set the first MAC of non-encrypted part of metadata 
+	 */
+	return create_link_mac_v1(fmt, 0, loc, info, master);
+}
+
+/*
+ * Called by fops:
+ * ->create();
+ * ->link();
+ *
+ * Pack common and version-specific parts of file's metadata
+ * Pre-conditions: @info contains valid object-id.
+ */
+int32_t create_format(unsigned char *wire,
+		      loc_t *loc,
+		      struct crypt_inode_info *info,
+		      struct master_cipher_info *master)
+{
+	struct crypt_format *fmt = (struct crypt_format *)wire;
+
+	fmt->loader_id = current_mtd_loader();
+
+	wire += sizeof(struct crypt_format);
+	return mtd_loaders[current_mtd_loader()].create_format(wire, loc,
+							       info, master);
+}
+
+/*
+ * Append or overwrite per-link mac of @mac_idx index
+ * in accordance with the new pathname
+ */
+int32_t appov_link_mac_v1(unsigned char *new,
+			  unsigned char *old,
+			  uint32_t old_size,
+			  int32_t mac_idx,
+			  loc_t *loc,
+			  struct crypt_inode_info *info,
+			  struct master_cipher_info *master,
+			  crypt_local_t *local)
+{
+	memcpy(new, old, old_size);
+	return create_link_mac_v1((struct mtd_format_v1 *)new, mac_idx,
+				  loc, info, master);
+}
+
+/*
+ * Cut per-link mac of @mac_idx index
+ */
+static int32_t cut_link_mac_v1(unsigned char *new,
+			       unsigned char *old,			       
+			       uint32_t old_size,
+			       int32_t mac_idx,
+			       loc_t *loc,
+			       struct crypt_inode_info *info,
+			       struct master_cipher_info *master,
+			       crypt_local_t *local)
+{
+	memcpy(new,
+	       old,
+	       sizeof(struct mtd_format_v1) + NMTD_8_MAC_SIZE * (mac_idx - 1));
+
+	memcpy(new + sizeof(struct mtd_format_v1) + NMTD_8_MAC_SIZE * (mac_idx - 1),
+	       old + sizeof(struct mtd_format_v1) + NMTD_8_MAC_SIZE * mac_idx,
+	       old_size - (sizeof(struct mtd_format_v1) + NMTD_8_MAC_SIZE * mac_idx));
+	return 0;
+}
+
+int32_t update_format_v1(unsigned char *new,
+			 unsigned char *old,
+				size_t old_len,
+				int32_t mac_idx, /* of old name */
+				mtd_op_t op,
+				loc_t *loc,
+				struct crypt_inode_info *info,
+				struct master_cipher_info *master,
+				crypt_local_t *local)
+{
+	switch (op) {
+	case MTD_APPEND:
+		mac_idx = 1 + (old_len - sizeof(struct mtd_format_v1))/8;
+	case MTD_OVERWRITE:
+		return appov_link_mac_v1(new, old, old_len, mac_idx,
+					 loc, info, master, local);
+	case MTD_CUT:
+		return cut_link_mac_v1(new, old, old_len, mac_idx,
+				       loc, info, master, local);
+	default:
+		gf_log("crypt", GF_LOG_ERROR, "Bad  mtd operation %d", op);
+		return -1;
+	}
+}
+
+/*
+ * Called by fops:
+ *
+ * ->link()
+ * ->unlink()
+ * ->rename()
+ *
+ */
+int32_t update_format(unsigned char *new,
+		      unsigned char *old,
+		      size_t old_len,
+		      int32_t mac_idx,
+		      mtd_op_t op,
+		      loc_t *loc,
+		      struct crypt_inode_info *info,
+		      struct master_cipher_info *master,
+		      crypt_local_t *local)
+{
+	if (!new)
+		return 0;
+	memcpy(new, old, sizeof(struct crypt_format));
+
+	old += sizeof(struct crypt_format);
+	new += sizeof(struct crypt_format);
+	old_len -= sizeof(struct crypt_format);
+
+	return mtd_loaders[current_mtd_loader()].update_format(new, old,
+							       old_len,
+							       mac_idx, op,
+							       loc, info,
+							       master, local);
+}
+
+/*
+ * Perform preliminary checks of found metadata
+ * Return < 0 on errors;
+ * Return number of object-id MACs (>= 1) on success
+ */
+int32_t check_format_v1(uint32_t len, unsigned char *wire)
+{
+	uint32_t nr_links;
+
+	if (len < sizeof(struct mtd_format_v1)) {
+		gf_log("crypt", GF_LOG_ERROR,
+		       "v1-loader: bad metadata size %d", len);
+		goto error;
+	}
+	len -= sizeof(struct mtd_format_v1);
+	if (len % sizeof(nmtd_8_mac_t)) {
+		gf_log("crypt", GF_LOG_ERROR,
+		       "v1-loader: bad metadata format");
+		goto error;
+	}
+	nr_links = 1 + len / sizeof(nmtd_8_mac_t);
+	if (nr_links > _POSIX_LINK_MAX)
+		goto error;
+	return nr_links;
+ error:
+	return EIO;
+}
+
+/*
+ * Verify per-link MAC specified by index @idx
+ *
+ * return:
+ * -1 on errors;
+ *  0 on failed verification;
+ *  1 on sucessful verification
+ */
+static int32_t verify_link_mac_v1(struct mtd_format_v1 *fmt,
+				  uint32_t idx /* index of the mac to verify */,
+				  loc_t *loc,
+				  struct crypt_inode_info *info,
+				  struct master_cipher_info *master)
+{
+	int32_t ret;
+	unsigned char *mac;
+	unsigned char cmac[16];
+
+	mac = get_NMTD_V1_MAC(fmt) + idx * SIZE_OF_NMTD_V1_MAC;
+
+	ret = calc_link_mac_v1(fmt, loc, cmac, info, master);
+	if (ret)
+		return -1;
+	if (memcmp(cmac, mac, SIZE_OF_NMTD_V1_MAC))
+		return 0;
+	return 1;
+}
+
+/*
+ * Lookup per-link MAC by pathname.
+ *
+ * return index of the MAC, if it was found;
+ * return < 0 on errors, or if the MAC wasn't found
+ */
+static int32_t lookup_link_mac_v1(struct mtd_format_v1 *fmt,
+				  uint32_t nr_macs,
+				  loc_t *loc,
+				  struct crypt_inode_info *info,
+				  struct master_cipher_info *master)
+{
+	int32_t ret;
+	uint32_t idx;
+
+	for (idx = 0; idx < nr_macs; idx++) {
+		ret = verify_link_mac_v1(fmt, idx, loc, info, master);
+		if (ret < 0)
+			return ret;
+		if (ret > 0)
+			return idx;
+	}
+	return -ENOENT;
+}
+
+/*
+ * Extract version-specific part of metadata
+ */
+static int32_t open_format_v1(unsigned char *wire,
+			      int32_t len,
+			      loc_t *loc,
+			      struct crypt_inode_info *info,
+			      struct master_cipher_info *master,
+			      crypt_local_t *local,
+			      gf_boolean_t load_info)
+{
+	int32_t ret;
+	int32_t num_nmtd_macs;
+	struct mtd_format_v1 *fmt;
+	unsigned char mtd_key[16];
+	AES_KEY EMTD_KEY;
+	GCM128_CONTEXT *gctx;
+	uint32_t ad;
+	emtd_8_mac_t gmac;
+	struct object_cipher_info *object;
+
+	num_nmtd_macs = check_format_v1(len, wire);
+	if (num_nmtd_macs <= 0)
+		return EIO;
+	fmt = (struct mtd_format_v1 *)wire;
+
+	ret = lookup_link_mac_v1(fmt, num_nmtd_macs, loc, info, master);
+	if (ret < 0) {
+		gf_log("crypt", GF_LOG_ERROR, "NMTD verification failed");
+		return EINVAL;
+	}
+	local->mac_idx = ret;
+	if (load_info == _gf_false)
+		/* the case of partial open */
+		return 0;
+
+	object = &info->cinfo;
+
+	ret = get_emtd_file_key(info, master, mtd_key);
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR, "Can not retrieve metadata key");
+		return ret;
+	}
+	/*
+	 * decrypt encrypted meta-data
+	 */
+	ret = AES_set_encrypt_key(mtd_key, sizeof(mtd_key)*8, &EMTD_KEY);
+	if (ret < 0) {
+		gf_log("crypt", GF_LOG_ERROR, "Can not set encrypt key");
+		return ret;
+	}
+	gctx = CRYPTO_gcm128_new(&EMTD_KEY, (block128_f)AES_encrypt);
+	if (!gctx) {
+		gf_log("crypt", GF_LOG_ERROR, "Can not alloc gcm context");
+		return ENOMEM;
+	}
+	CRYPTO_gcm128_setiv(gctx, info->oid, sizeof(uuid_t));
+
+	ad = htole32(MTD_LOADER_V1);
+	ret = CRYPTO_gcm128_aad(gctx, (const unsigned char *)&ad, sizeof(ad));
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR, " CRYPTO_gcm128_aad failed");
+		CRYPTO_gcm128_release(gctx);
+		return ret;	
+	}
+	ret = CRYPTO_gcm128_decrypt(gctx,
+				    get_EMTD_V1(fmt),
+				    get_EMTD_V1(fmt),
+				    SIZE_OF_EMTD_V1);
+	if (ret) {
+		gf_log("crypt", GF_LOG_ERROR, " CRYPTO_gcm128_decrypt failed");
+		CRYPTO_gcm128_release(gctx);
+		return ret;
+	}
+	/*
+	 * verify metadata
+	 */
+	CRYPTO_gcm128_tag(gctx, gmac, sizeof(gmac));
+	CRYPTO_gcm128_release(gctx);
+	if (memcmp(gmac, get_EMTD_V1_MAC(fmt), SIZE_OF_EMTD_V1_MAC)) {
+		gf_log("crypt", GF_LOG_ERROR, "EMTD verification failed");
+		return EINVAL;
+	}
+	/*
+	 * load verified metadata to the private part of inode
+	 */
+	info->nr_minor = fmt->minor_id;
+
+	object->o_alg = fmt->alg_id;
+	object->o_dkey_size = fmt->dkey_factor << KEY_FACTOR_BITS;
+	object->o_block_bits = fmt->block_bits;
+	object->o_mode = fmt->mode_id;
+
+	return check_file_metadata(info);
+}
+
+/*
+ * perform metadata authentication against @loc->path;
+ * extract crypt-specific attribtes and populate @info
+ * with them (optional)
+ */
+int32_t open_format(unsigned char *str,
+		    int32_t len,
+		    loc_t *loc,
+		    struct crypt_inode_info *info,
+		    struct master_cipher_info *master,
+		    crypt_local_t *local,
+		    gf_boolean_t load_info)
+{
+	struct crypt_format *fmt;
+	if (len < sizeof(*fmt)) {
+		gf_log("crypt", GF_LOG_ERROR, "Bad core format");
+		return EIO;
+	}
+	fmt = (struct crypt_format *)str;
+
+	if (fmt->loader_id >= LAST_MTD_LOADER) {
+		gf_log("crypt", GF_LOG_ERROR,
+		       "Unsupported loader id %d", fmt->loader_id);
+		return EINVAL;
+	}
+	str += sizeof(*fmt);
+	len -= sizeof(*fmt);
+
+	return mtd_loaders[fmt->loader_id].open_format(str,
+						       len,
+						       loc,
+						       info,
+						       master,
+						       local,
+						       load_info);
+}
+
+struct crypt_mtd_loader mtd_loaders [LAST_MTD_LOADER] = {
+	[MTD_LOADER_V1] = 
+	{.format_size = format_size_v1,
+	 .create_format = create_format_v1,
+	 .open_format = open_format_v1,
+	 .update_format = update_format_v1
+	}
+};
+
+/*
+  Local variables:
+  c-indentation-style: "K&R"
+  mode-name: "LC"
+  c-basic-offset: 8
+  tab-width: 8
+  fill-column: 80
+  scroll-step: 1
+  End:
+*/
diff --git a/xlators/encryption/crypt/src/metadata.h b/xlators/encryption/crypt/src/metadata.h
new file mode 100644
index 000000000..a92f149ef
--- /dev/null
+++ b/xlators/encryption/crypt/src/metadata.h
@@ -0,0 +1,74 @@
+/*
+  Copyright (c) 2008-2013 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 __METADATA_H__
+#define __METADATA_H__
+
+#define NMTD_8_MAC_SIZE   (8)
+#define EMTD_8_MAC_SIZE   (8)
+
+typedef uint8_t nmtd_8_mac_t[NMTD_8_MAC_SIZE];
+typedef uint8_t emtd_8_mac_t[EMTD_8_MAC_SIZE] ;
+
+/*
+ * Version "v1" of file's metadata.
+ * Metadata of this version has 4 components:
+ * 
+ * 1) EMTD (Encrypted part of MeTaData);
+ * 2) NMTD (Non-encrypted part of MeTaData);
+ * 3) EMTD_MAC; (EMTD Message Authentication Code);
+ * 4) Array of per-link NMTD MACs (for every (hard)link it includes
+ *    exactly one MAC)
+ */
+struct mtd_format_v1 {
+	/* EMTD, encrypted part of meta-data */
+	uint8_t            alg_id;  /* cipher algorithm id (only AES for now) */
+        uint8_t            mode_id; /* cipher mode id; (only XTS for now) */
+	uint8_t            block_bits; /* encoded block size */
+	uint8_t            minor_id; /* client translator id */
+	uint8_t            dkey_factor; /* encoded size of the data key */
+	/* MACs */
+	emtd_8_mac_t       gmac; /* MAC of the encrypted meta-data, 8 bytes */
+	nmtd_8_mac_t       omac; /* per-link MACs of the non-encrypted
+				  * meta-data: at least one such MAC is always
+				  * present */
+} __attribute__((packed));
+
+/*
+ * NMTD, the non-encrypted part of metadata of version "v1"
+ * is file's gfid, which is generated on trusted machines.
+ */
+#define SIZE_OF_NMTD_V1 (sizeof(uuid_t))
+#define SIZE_OF_EMTD_V1 (offsetof(struct mtd_format_v1, gmac) -		\
+			 offsetof(struct mtd_format_v1, alg_id))
+#define SIZE_OF_NMTD_V1_MAC (NMTD_8_MAC_SIZE)
+#define SIZE_OF_EMTD_V1_MAC (EMTD_8_MAC_SIZE)
+
+static inline unsigned char *get_EMTD_V1(struct mtd_format_v1 *format)
+{
+        return &format->alg_id;
+}
+
+static inline unsigned char *get_NMTD_V1(struct crypt_inode_info *info)
+{
+        return info->oid;
+}
+
+static inline unsigned char *get_EMTD_V1_MAC(struct mtd_format_v1 *format)
+{
+        return format->gmac;
+}
+
+static inline unsigned char *get_NMTD_V1_MAC(struct mtd_format_v1 *format)
+{
+        return format->omac;
+}
+
+#endif /* __METADATA_H__ */
diff --git a/xlators/features/Makefile.am b/xlators/features/Makefile.am
index 6a73301d7..d2f5ef192 100644
--- a/xlators/features/Makefile.am
+++ b/xlators/features/Makefile.am
@@ -1,4 +1,4 @@
 SUBDIRS = locks quota read-only mac-compat quiesce marker index \
-	  protect $(GLUPY_SUBDIR) # trash path-converter # filter
+	  protect compress changelog gfid-access $(GLUPY_SUBDIR) qemu-block # trash path-converter # filter
 
 CLEANFILES =
diff --git a/xlators/features/changelog/Makefile.am b/xlators/features/changelog/Makefile.am
new file mode 100644
index 000000000..153bb6850
--- /dev/null
+++ b/xlators/features/changelog/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src lib
+
+CLEANFILES =
diff --git a/xlators/storage/bd_map/Makefile.am b/xlators/features/changelog/lib/Makefile.am
index a985f42a8..a985f42a8 100644
--- a/xlators/storage/bd_map/Makefile.am
+++ b/xlators/features/changelog/lib/Makefile.am
diff --git a/xlators/features/changelog/lib/examples/c/get-changes.c b/xlators/features/changelog/lib/examples/c/get-changes.c
new file mode 100644
index 000000000..14562585a
--- /dev/null
+++ b/xlators/features/changelog/lib/examples/c/get-changes.c
@@ -0,0 +1,87 @@
+/*
+   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.
+*/
+
+/**
+ * get set of new changes every 10 seconds (just print the file names)
+ *
+ * Compile it using:
+ *  gcc -o getchanges `pkg-config --cflags libgfchangelog` get-changes.c \
+ *  `pkg-config --libs libgfchangelog`
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/un.h>
+#include <limits.h>
+#include <sys/socket.h>
+#include <sys/types.h>
+#include <errno.h>
+
+#include "changelog.h"
+
+#define handle_error(fn)                                \
+        printf ("%s (reason: %s)\n", fn, strerror (errno))
+
+int
+main (int argc, char ** argv)
+{
+        int     i           = 0;
+        int     ret         = 0;
+        ssize_t nr_changes  = 0;
+        ssize_t changes     = 0;
+        char fbuf[PATH_MAX] = {0,};
+
+        /* get changes for brick "/home/vshankar/export/yow/yow-1" */
+        ret = gf_changelog_register ("/home/vshankar/export/yow/yow-1",
+                                     "/tmp/scratch", "/tmp/change.log", 9, 5);
+        if (ret) {
+                handle_error ("register failed");
+                goto out;
+        }
+
+        while (1) {
+                i = 0;
+                nr_changes = gf_changelog_scan ();
+                if (nr_changes < 0) {
+                        handle_error ("scan(): ");
+                        break;
+                }
+
+                if (nr_changes == 0)
+                        goto next;
+
+                printf ("Got %ld changelog files\n", nr_changes);
+
+                while ( (changes =
+                         gf_changelog_next_change (fbuf, PATH_MAX)) > 0) {
+                        printf ("changelog file [%d]: %s\n", ++i, fbuf);
+
+                        /* process changelog */
+                        /* ... */
+                        /* ... */
+                        /* ... */
+                        /* done processing */
+
+                        ret = gf_changelog_done (fbuf);
+                        if (ret)
+                                handle_error ("gf_changelog_done");
+                }
+
+                if (changes == -1)
+                        handle_error ("gf_changelog_next_change");
+
+        next:
+                sleep (10);
+        }
+
+ out:
+        return ret;
+}
diff --git a/xlators/features/changelog/lib/examples/python/changes.py b/xlators/features/changelog/lib/examples/python/changes.py
new file mode 100644
index 000000000..d21db8eab
--- /dev/null
+++ b/xlators/features/changelog/lib/examples/python/changes.py
@@ -0,0 +1,32 @@
+#!/usr/bin/python
+
+import os
+import sys
+import time
+import libgfchangelog
+
+cl = libgfchangelog.Changes()
+
+def get_changes(brick, scratch_dir, log_file, log_level, interval):
+    change_list = []
+    try:
+        cl.cl_register(brick, scratch_dir, log_file, log_level)
+        while True:
+            cl.cl_scan()
+            change_list = cl.cl_getchanges()
+            if change_list:
+                print change_list
+            for change in change_list:
+                print('done with %s' % (change))
+                cl.cl_done(change)
+            time.sleep(interval)
+    except OSError:
+        ex = sys.exc_info()[1]
+        print ex
+
+if __name__ == '__main__':
+    if len(sys.argv) != 5:
+        print("usage: %s <brick> <scratch-dir> <log-file> <fetch-interval>"
+              % (sys.argv[0]))
+        sys.exit(1)
+    get_changes(sys.argv[1], sys.argv[2], sys.argv[3], 9, int(sys.argv[4]))
diff --git a/xlators/features/changelog/lib/examples/python/libgfchangelog.py b/xlators/features/changelog/lib/examples/python/libgfchangelog.py
new file mode 100644
index 000000000..68ec3baf1
--- /dev/null
+++ b/xlators/features/changelog/lib/examples/python/libgfchangelog.py
@@ -0,0 +1,64 @@
+import os
+from ctypes import *
+from ctypes.util import find_library
+
+class Changes(object):
+    libgfc = CDLL(find_library("gfchangelog"), use_errno=True)
+
+    @classmethod
+    def geterrno(cls):
+        return get_errno()
+
+    @classmethod
+    def raise_oserr(cls):
+        errn = cls.geterrno()
+        raise OSError(errn, os.strerror(errn))
+
+    @classmethod
+    def _get_api(cls, call):
+        return getattr(cls.libgfc, call)
+
+    @classmethod
+    def cl_register(cls, brick, path, log_file, log_level, retries = 0):
+        ret = cls._get_api('gf_changelog_register')(brick, path,
+                                                    log_file, log_level, retries)
+        if ret == -1:
+            cls.raise_oserr()
+
+    @classmethod
+    def cl_scan(cls):
+        ret = cls._get_api('gf_changelog_scan')()
+        if ret == -1:
+            cls.raise_oserr()
+
+    @classmethod
+    def cl_startfresh(cls):
+        ret = cls._get_api('gf_changelog_start_fresh')()
+        if ret == -1:
+            cls.raise_oserr()
+
+    @classmethod
+    def cl_getchanges(cls):
+        """ remove hardcoding for path name length """
+        def clsort(f):
+            return f.split('.')[-1]
+        changes = []
+        buf = create_string_buffer('\0', 4096)
+        call = cls._get_api('gf_changelog_next_change')
+
+        while True:
+            ret = call(buf, 4096)
+            if ret in (0, -1):
+                break;
+            changes.append(buf.raw[:ret-1])
+        if ret == -1:
+            cls.raise_oserr()
+        # cleanup tracker
+        cls.cl_startfresh()
+        return sorted(changes, key=clsort)
+
+    @classmethod
+    def cl_done(cls, clfile):
+        ret = cls._get_api('gf_changelog_done')(clfile)
+        if ret == -1:
+            cls.raise_oserr()
diff --git a/xlators/features/changelog/lib/src/Makefile.am b/xlators/features/changelog/lib/src/Makefile.am
new file mode 100644
index 000000000..fbaaea628
--- /dev/null
+++ b/xlators/features/changelog/lib/src/Makefile.am
@@ -0,0 +1,37 @@
+libgfchangelog_la_CFLAGS = -Wall $(GF_CFLAGS) $(GF_DARWIN_LIBGLUSTERFS_CFLAGS) \
+			-DDATADIR=\"$(localstatedir)\"
+
+libgfchangelog_la_CPPFLAGS = $(GF_CPPFLAGS) -D__USE_FILE_OFFSET64 -fpic \
+				-I../../../src/ -I$(top_srcdir)/libglusterfs/src \
+				-I$(top_srcdir)/xlators/features/changelog/src \
+				-DDATADIR=\"$(localstatedir)\"
+
+libgfchangelog_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+			$(GF_GLUSTERFS_LIBS)
+
+libgfchangelog_la_LDFLAGS = $(GF_LDFLAGS)
+
+libgfchangelogdir = $(includedir)/glusterfs/gfchangelog
+lib_LTLIBRARIES = libgfchangelog.la
+
+CONTRIB_BUILDDIR = $(top_builddir)/contrib
+
+libgfchangelog_la_SOURCES = gf-changelog.c gf-changelog-process.c \
+		gf-changelog-helpers.c $(CONTRIBDIR)/uuid/clear.c \
+		$(CONTRIBDIR)/uuid/copy.c $(CONTRIBDIR)/uuid/gen_uuid.c \
+		$(CONTRIBDIR)/uuid/pack.c $(CONTRIBDIR)/uuid/parse.c \
+		$(CONTRIBDIR)/uuid/unparse.c $(CONTRIBDIR)/uuid/uuid_time.c \
+		$(CONTRIBDIR)/uuid/compare.c $(CONTRIBDIR)/uuid/isnull.c \
+		$(CONTRIBDIR)/uuid/unpack.c
+
+noinst_HEADERS = gf-changelog-helpers.h $(CONTRIBDIR)/uuid/uuidd.h \
+		$(CONTRIBDIR)/uuid/uuid.h $(CONTRIBDIR)/uuid/uuidP.h \
+		$(CONTRIB_BUILDDIR)/uuid/uuid_types.h
+
+libgfchangelog_HEADERS = changelog.h
+
+CLEANFILES =
+CONFIG_CLEAN_FILES = $(CONTRIB_BUILDDIR)/uuid/uuid_types.h
+
+$(top_builddir)/libglusterfs/src/libglusterfs.la:
+	$(MAKE) -C $(top_builddir)/libglusterfs/src/ all
diff --git a/xlators/features/changelog/lib/src/changelog.h b/xlators/features/changelog/lib/src/changelog.h
new file mode 100644
index 000000000..5cddfb583
--- /dev/null
+++ b/xlators/features/changelog/lib/src/changelog.h
@@ -0,0 +1,31 @@
+/*
+   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 _GF_CHANGELOG_H
+#define _GF_CHANGELOG_H
+
+/* API set */
+
+int
+gf_changelog_register (char *brick_path, char *scratch_dir,
+                       char *log_file, int log_levl, int max_reconnects);
+ssize_t
+gf_changelog_scan ();
+
+int
+gf_changelog_start_fresh ();
+
+ssize_t
+gf_changelog_next_change (char *bufptr, size_t maxlen);
+
+int
+gf_changelog_done (char *file);
+
+#endif
diff --git a/xlators/features/changelog/lib/src/gf-changelog-helpers.c b/xlators/features/changelog/lib/src/gf-changelog-helpers.c
new file mode 100644
index 000000000..1eef8bf04
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-helpers.c
@@ -0,0 +1,180 @@
+/*
+   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 "changelog-mem-types.h"
+#include "gf-changelog-helpers.h"
+
+ssize_t gf_changelog_read_path (int fd, char *buffer, size_t bufsize)
+{
+        return read (fd, buffer, bufsize);
+}
+
+size_t
+gf_changelog_write (int fd, char *buffer, size_t len)
+{
+        ssize_t size = 0;
+        size_t writen = 0;
+
+        while (writen < len) {
+                size = write (fd,
+                              buffer + writen, len - writen);
+                if (size <= 0)
+                        break;
+
+                writen += size;
+        }
+
+        return writen;
+}
+
+void
+gf_rfc3986_encode (unsigned char *s, char *enc, char *estr)
+{
+        for (; *s; s++) {
+                if (estr[*s])
+                        sprintf(enc, "%c", estr[*s]);
+                else
+                        sprintf(enc, "%%%02X", *s);
+                while (*++enc);
+        }
+}
+
+/**
+ * thread safe version of readline with buffering
+ * (taken from Unix Network Programming Volume I, W.R. Stevens)
+ *
+ * This is favoured over fgets() as we'd need to ftruncate()
+ * (see gf_changelog_scan() API) to record new changelog files.
+ * stream open functions does have a truncate like api (although
+ * that can be done via @fflush(fp), @ftruncate(fd) and @fseek(fp),
+ * but this involves mixing POSIX file descriptors and stream FILE *).
+ *
+ * NOTE: This implmentation still does work with more than one fd's
+ *       used to perform gf_readline(). For this very reason it's not
+ *       made a part of libglusterfs.
+ */
+
+static pthread_key_t rl_key;
+static pthread_once_t rl_once = PTHREAD_ONCE_INIT;
+
+static void
+readline_destructor (void *ptr)
+{
+        GF_FREE (ptr);
+}
+
+static void
+readline_once (void)
+{
+        pthread_key_create (&rl_key, readline_destructor);
+}
+
+static ssize_t
+my_read (read_line_t *tsd, int fd, char *ptr)
+{
+        if (tsd->rl_cnt <= 0) {
+                if ( (tsd->rl_cnt = read (fd, tsd->rl_buf, MAXLINE)) < 0 )
+                        return -1;
+                else if (tsd->rl_cnt == 0)
+                        return 0;
+                tsd->rl_bufptr = tsd->rl_buf;
+        }
+
+        tsd->rl_cnt--;
+        *ptr = *tsd->rl_bufptr++;
+        return 1;
+}
+
+static int
+gf_readline_init_once (read_line_t **tsd)
+{
+        if (pthread_once (&rl_once, readline_once) != 0)
+                return -1;
+
+        *tsd = pthread_getspecific (rl_key);
+        if (*tsd)
+                goto out;
+
+        *tsd = GF_CALLOC (1, sizeof (**tsd),
+                          gf_changelog_mt_libgfchangelog_rl_t);
+        if (!*tsd)
+                return -1;
+
+        if (pthread_setspecific (rl_key, *tsd) != 0)
+                return -1;
+
+ out:
+        return 0;
+}
+
+ssize_t
+gf_readline (int fd, void *vptr, size_t maxlen)
+{
+        size_t       n   = 0;
+        size_t       rc  = 0;
+        char         c   = ' ';
+        char        *ptr = NULL;
+        read_line_t *tsd = NULL;
+
+        if (gf_readline_init_once (&tsd))
+                return -1;
+
+        ptr = vptr;
+        for (n = 1; n < maxlen; n++) {
+                if ( (rc = my_read (tsd, fd, &c)) == 1 ) {
+                        *ptr++ = c;
+                        if (c == '\n')
+                                break;
+                } else if (rc == 0) {
+                        *ptr = '\0';
+                        return (n - 1);
+                } else
+                        return -1;
+        }
+
+        *ptr = '\0';
+        return n;
+
+}
+
+off_t
+gf_lseek (int fd, off_t offset, int whence)
+{
+        off_t        off = 0;
+        read_line_t *tsd = NULL;
+
+        if (gf_readline_init_once (&tsd))
+                return -1;
+
+        if ( (off = lseek (fd, offset, whence)) == -1)
+                return -1;
+
+        tsd->rl_cnt = 0;
+        tsd->rl_bufptr = tsd->rl_buf;
+
+        return off;
+}
+
+int
+gf_ftruncate (int fd, off_t length)
+{
+        read_line_t *tsd = NULL;
+
+        if (gf_readline_init_once (&tsd))
+                return -1;
+
+        if (ftruncate (fd, 0))
+                return -1;
+
+        tsd->rl_cnt = 0;
+        tsd->rl_bufptr = tsd->rl_buf;
+
+        return 0;
+}
diff --git a/xlators/features/changelog/lib/src/gf-changelog-helpers.h b/xlators/features/changelog/lib/src/gf-changelog-helpers.h
new file mode 100644
index 000000000..3aa6ed7b8
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-helpers.h
@@ -0,0 +1,97 @@
+/*
+   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 _GF_CHANGELOG_HELPERS_H
+#define _GF_CHANGELOG_HELPERS_H
+
+#include <unistd.h>
+#include <dirent.h>
+#include <limits.h>
+#include <pthread.h>
+
+#include <xlator.h>
+
+#define GF_CHANGELOG_TRACKER  "tracker"
+
+#define GF_CHANGELOG_CURRENT_DIR    ".current"
+#define GF_CHANGELOG_PROCESSED_DIR  ".processed"
+#define GF_CHANGELOG_PROCESSING_DIR ".processing"
+
+#ifndef MAXLINE
+#define MAXLINE 4096
+#endif
+
+#define GF_CHANGELOG_FILL_BUFFER(ptr, ascii, off, len) do {     \
+                memcpy (ascii + off, ptr, len);                 \
+                off += len;                                     \
+        } while (0)
+
+typedef struct read_line {
+        int rl_cnt;
+        char *rl_bufptr;
+        char rl_buf[MAXLINE];
+} read_line_t;
+
+typedef struct gf_changelog {
+        xlator_t *this;
+
+        /* 'processing' directory stream */
+        DIR *gfc_dir;
+
+        /* fd to the tracker file */
+        int gfc_fd;
+
+        /* connection retries */
+        int gfc_connretries;
+
+        char gfc_sockpath[PATH_MAX];
+
+        char gfc_brickpath[PATH_MAX];
+
+        /* socket for recieving notifications */
+        int gfc_sockfd;
+
+        char *gfc_working_dir;
+
+        /* RFC 3986 string encoding */
+        char rfc3986[256];
+
+        char gfc_current_dir[PATH_MAX];
+        char gfc_processed_dir[PATH_MAX];
+        char gfc_processing_dir[PATH_MAX];
+
+        pthread_t gfc_changelog_processor;
+} gf_changelog_t;
+
+int
+gf_changelog_notification_init (xlator_t *this, gf_changelog_t *gfc);
+
+void *
+gf_changelog_process (void *data);
+
+ssize_t
+gf_changelog_read_path (int fd, char *buffer, size_t bufsize);
+
+void
+gf_rfc3986_encode (unsigned char *s, char *enc, char *estr);
+
+size_t
+gf_changelog_write (int fd, char *buffer, size_t len);
+
+ssize_t
+gf_readline (int fd, void *vptr, size_t maxlen);
+
+int
+gf_ftruncate (int fd, off_t length);
+
+off_t
+gf_lseek (int fd, off_t offset, int whence);
+
+#endif
diff --git a/xlators/features/changelog/lib/src/gf-changelog-process.c b/xlators/features/changelog/lib/src/gf-changelog-process.c
new file mode 100644
index 000000000..df7204931
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-process.c
@@ -0,0 +1,571 @@
+/*
+   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 <unistd.h>
+#include <pthread.h>
+
+#include "uuid.h"
+#include "globals.h"
+#include "glusterfs.h"
+
+#include "gf-changelog-helpers.h"
+
+/* from the changelog translator */
+#include "changelog-misc.h"
+
+extern int byebye;
+
+/**
+ * number of gfid records after fop number
+ */
+int nr_gfids[] = {
+        [GF_FOP_MKNOD]   = 1,
+        [GF_FOP_MKDIR]   = 1,
+        [GF_FOP_UNLINK]  = 1,
+        [GF_FOP_RMDIR]   = 1,
+        [GF_FOP_SYMLINK] = 1,
+        [GF_FOP_RENAME]  = 2,
+        [GF_FOP_LINK]    = 1,
+        [GF_FOP_CREATE]  = 1,
+};
+
+static char *
+binary_to_ascii (uuid_t uuid)
+{
+        return uuid_utoa (uuid);
+}
+
+static char *
+conv_noop (char *ptr) { return ptr; }
+
+#define VERIFY_SEPARATOR(ptr, plen, perr)       \
+        {                                       \
+                if (*(ptr + plen) != '\0') {    \
+                        perr = 1;               \
+                        break;                  \
+                }                               \
+        }
+
+#define MOVER_MOVE(mover, nleft, bytes)         \
+        {                                       \
+                mover += bytes;                 \
+                nleft -= bytes;                 \
+        }                                       \
+
+#define PARSE_GFID(mov, ptr, le, fn, perr)                      \
+        {                                                       \
+                VERIFY_SEPARATOR (mov, le, perr);               \
+                ptr = fn (mov);                                 \
+                if (!ptr) {                                     \
+                        perr = 1;                               \
+                        break;                                  \
+                }                                               \
+        }
+
+#define FILL_AND_MOVE(pt, buf, of, mo, nl, le)                          \
+        {                                                               \
+                GF_CHANGELOG_FILL_BUFFER (pt, buf, of, strlen (pt));    \
+                MOVER_MOVE (mo, nl, le);                                \
+        }
+
+
+#define PARSE_GFID_MOVE(ptr, uuid, mover, nleft, perr)          \
+        {                                                       \
+                memcpy (uuid, mover, sizeof (uuid_t));          \
+                ptr = binary_to_ascii (uuid);                   \
+                if (!ptr) {                                     \
+                        perr = 1;                               \
+                        break;                                  \
+                }                                               \
+                MOVER_MOVE (mover, nleft, sizeof (uuid_t));     \
+        }                                                       \
+
+#define LINE_BUFSIZE  3*PATH_MAX /* enough buffer for extra chars too */
+
+/**
+ * using mmap() makes parsing easy. fgets() cannot be used here as
+ * the binary gfid could contain a line-feed (0x0A), in that case fgets()
+ * would read an incomplete line and parsing would fail. using POSIX fds
+ * would result is additional code to maintain state in case of partial
+ * reads of data (where multiple entries do not fit extirely in the buffer).
+ *
+ * mmap() gives the flexibility of pointing to an offset in the file
+ * without us worrying about reading it in memory (VM does that for us for
+ * free).
+ */
+
+static int
+gf_changelog_parse_binary (xlator_t *this,
+                           gf_changelog_t *gfc, int from_fd, int to_fd,
+                           size_t start_offset, struct stat *stbuf)
+
+{
+        int     ret              = -1;
+        off_t   off              = 0;
+        off_t   nleft            = 0;
+        uuid_t  uuid             = {0,};
+        char   *ptr              = NULL;
+        char   *bname_start      = NULL;
+        char   *bname_end        = NULL;
+        char   *mover            = NULL;
+        char   *start            = NULL;
+        char    current_mover    = ' ';
+        size_t  blen             = 0;
+        int     parse_err        = 0;
+        char ascii[LINE_BUFSIZE] = {0,};
+
+        nleft = stbuf->st_size;
+
+        start = (char *) mmap (NULL, nleft,
+                               PROT_READ, MAP_PRIVATE, from_fd, 0);
+        if (!start) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "mmap() error (reason: %s)", strerror (errno));
+                goto out;
+        }
+
+        mover = start;
+
+        MOVER_MOVE (mover, nleft, start_offset);
+
+        while (nleft > 0) {
+
+                off = blen = 0;
+                ptr = bname_start = bname_end = NULL;
+
+                current_mover = *mover;
+
+                switch (current_mover) {
+                case 'D':
+                case 'M':
+                        MOVER_MOVE (mover, nleft, 1);
+                        PARSE_GFID_MOVE (ptr, uuid, mover, nleft, parse_err);
+
+                        break;
+
+                case 'E':
+                        MOVER_MOVE (mover, nleft, 1);
+                        PARSE_GFID_MOVE (ptr, uuid, mover, nleft, parse_err);
+
+                        bname_start = mover;
+                        if ( (bname_end = strchr (mover, '\n')) == NULL ) {
+                                parse_err = 1;
+                                break;
+                        }
+
+                        blen = bname_end - bname_start;
+                        MOVER_MOVE (mover, nleft, blen);
+
+                        break;
+
+                default:
+                        parse_err = 1;
+                }
+
+                if (parse_err)
+                        break;
+
+                GF_CHANGELOG_FILL_BUFFER (&current_mover, ascii, off, 1);
+                GF_CHANGELOG_FILL_BUFFER (" ", ascii, off, 1);
+                GF_CHANGELOG_FILL_BUFFER (ptr, ascii, off, strlen (ptr));
+                if (blen)
+                        GF_CHANGELOG_FILL_BUFFER (bname_start,
+                                                  ascii, off, blen);
+                GF_CHANGELOG_FILL_BUFFER ("\n", ascii, off, 1);
+
+                if (gf_changelog_write (to_fd, ascii, off) != off) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "processing binary changelog failed due to "
+                                " error in writing ascii change (reason: %s)",
+                                strerror (errno));
+                        break;
+                }
+
+                MOVER_MOVE (mover, nleft, 1);
+        }
+
+        if ( (nleft == 0) && (!parse_err))
+                ret = 0;
+
+        if (munmap (start, stbuf->st_size))
+                gf_log (this->name, GF_LOG_ERROR,
+                        "munmap() error (reason: %s)", strerror (errno));
+ out:
+        return ret;
+}
+
+/**
+ * ascii decoder:
+ *  - separate out one entry from another
+ *  - use fop name rather than fop number
+ */
+static int
+gf_changelog_parse_ascii (xlator_t *this,
+                          gf_changelog_t *gfc, int from_fd, int to_fd,
+                          size_t start_offset, struct stat *stbuf)
+{
+        int         ng            = 0;
+        int         ret           = -1;
+        int         fop           = 0;
+        int         len           = 0;
+        off_t       off           = 0;
+        off_t       nleft         = 0;
+        char       *ptr           = NULL;
+        char       *eptr          = NULL;
+        char       *start         = NULL;
+        char       *mover         = NULL;
+        int         parse_err     = 0;
+        char        current_mover = ' ';
+        char ascii[LINE_BUFSIZE]  = {0,};
+        const char *fopname       = NULL;
+
+        nleft = stbuf->st_size;
+
+        start = (char *) mmap (NULL, nleft,
+                               PROT_READ, MAP_PRIVATE, from_fd, 0);
+        if (!start) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "mmap() error (reason: %s)", strerror (errno));
+                goto out;
+        }
+
+        mover = start;
+
+        MOVER_MOVE (mover, nleft, start_offset);
+
+        while (nleft > 0) {
+                off = 0;
+                current_mover = *mover;
+
+                GF_CHANGELOG_FILL_BUFFER (&current_mover, ascii, off, 1);
+                GF_CHANGELOG_FILL_BUFFER (" ", ascii, off, 1);
+
+                switch (current_mover) {
+                case 'D':
+                case 'M':
+                        MOVER_MOVE (mover, nleft, 1);
+
+                        /* target gfid */
+                        PARSE_GFID (mover, ptr, UUID_CANONICAL_FORM_LEN,
+                                    conv_noop, parse_err);
+                        FILL_AND_MOVE(ptr, ascii, off,
+                                      mover, nleft, UUID_CANONICAL_FORM_LEN);
+                        break;
+
+                case 'E':
+                        MOVER_MOVE (mover, nleft, 1);
+
+                        /* target gfid */
+                        PARSE_GFID (mover, ptr, UUID_CANONICAL_FORM_LEN,
+                                    conv_noop, parse_err);
+                        FILL_AND_MOVE (ptr, ascii, off,
+                                       mover, nleft, UUID_CANONICAL_FORM_LEN);
+                        FILL_AND_MOVE (" ", ascii, off,
+                                       mover, nleft, 1);
+
+                        /* fop */
+                        len = strlen (mover);
+                        VERIFY_SEPARATOR (mover, len, parse_err);
+
+                        fop = atoi (mover);
+                        if ( (fopname = gf_fop_list[fop]) == NULL) {
+                                parse_err = 1;
+                                break;
+                        }
+
+                        MOVER_MOVE (mover, nleft, len);
+
+                        len = strlen (fopname);
+                        GF_CHANGELOG_FILL_BUFFER (fopname, ascii, off, len);
+
+                        /* pargfid + bname */
+                        ng = nr_gfids[fop];
+                        while (ng-- > 0) {
+                                MOVER_MOVE (mover, nleft, 1);
+                                len = strlen (mover);
+                                GF_CHANGELOG_FILL_BUFFER (" ", ascii, off, 1);
+
+                                PARSE_GFID (mover, ptr, len,
+                                            conv_noop, parse_err);
+                                eptr = calloc (3, strlen (ptr));
+                                if (!eptr) {
+                                        parse_err = 1;
+                                        break;
+                                }
+
+                                gf_rfc3986_encode ((unsigned char *) ptr,
+                                                   eptr, gfc->rfc3986);
+                                FILL_AND_MOVE (eptr, ascii, off,
+                                               mover, nleft, len);
+                                free (eptr);
+                        }
+
+                        break;
+                default:
+                        parse_err = 1;
+                }
+
+                if (parse_err)
+                        break;
+
+                GF_CHANGELOG_FILL_BUFFER ("\n", ascii, off, 1);
+
+                if (gf_changelog_write (to_fd, ascii, off) != off) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "processing ascii changelog failed due to "
+                                " wrror in writing change (reason: %s)",
+                                strerror (errno));
+                        break;
+                }
+
+                MOVER_MOVE (mover, nleft, 1);
+
+        }
+
+        if ( (nleft == 0) && (!parse_err))
+                ret = 0;
+
+        if (munmap (start, stbuf->st_size))
+                gf_log (this->name, GF_LOG_ERROR,
+                        "munmap() error (reason: %s)", strerror (errno));
+
+ out:
+        return ret;
+}
+
+#define COPY_BUFSIZE  8192
+static int
+gf_changelog_copy (xlator_t *this, int from_fd, int to_fd)
+{
+        ssize_t size                  = 0;
+        char   buffer[COPY_BUFSIZE+1] = {0,};
+
+        while (1) {
+                size = read (from_fd, buffer, COPY_BUFSIZE);
+                if (size <= 0)
+                        break;
+
+                if (gf_changelog_write (to_fd,
+                                        buffer, size) != size) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "error processing ascii changlog");
+                        size = -1;
+                        break;
+                }
+        }
+
+        return (size < 0 ? -1 : 0);
+}
+
+static int
+gf_changelog_decode (xlator_t *this, gf_changelog_t *gfc, int from_fd,
+                     int to_fd, struct stat *stbuf, int *zerob)
+{
+        int    ret        = -1;
+        int    encoding   = -1;
+        size_t elen       = 0;
+        char buffer[1024] = {0,};
+
+        CHANGELOG_GET_ENCODING (from_fd, buffer, 1024, encoding, elen);
+        if (encoding == -1) /* unknown encoding */
+                goto out;
+
+        if (!CHANGELOG_VALID_ENCODING (encoding))
+                goto out;
+
+        if (elen == stbuf->st_size) {
+                *zerob = 1;
+                goto out;
+        }
+
+        /**
+         * start processing after the header
+         */
+        lseek (from_fd, elen, SEEK_SET);
+
+        switch (encoding) {
+        case CHANGELOG_ENCODE_BINARY:
+                /**
+                 * this ideally should have been a part of changelog-encoders.c
+                 * (ie. part of the changelog translator).
+                 */
+                ret = gf_changelog_parse_binary (this, gfc, from_fd,
+                                                 to_fd, elen, stbuf);
+                break;
+
+        case CHANGELOG_ENCODE_ASCII:
+                ret = gf_changelog_parse_ascii (this, gfc, from_fd,
+                                                to_fd, elen, stbuf);
+                break;
+        default:
+                ret = gf_changelog_copy (this, from_fd, to_fd);
+        }
+
+ out:
+        return ret;
+}
+
+static int
+gf_changelog_consume (xlator_t *this, gf_changelog_t *gfc, char *from_path)
+{
+        int         ret        = -1;
+        int         fd1        = 0;
+        int         fd2        = 0;
+        int         zerob      = 0;
+        struct stat stbuf      = {0,};
+        char dest[PATH_MAX]    = {0,};
+        char to_path[PATH_MAX] = {0,};
+
+        ret = stat (from_path, &stbuf);
+        if (ret || !S_ISREG(stbuf.st_mode)) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "stat failed on changelog file: %s", from_path);
+                goto out;
+        }
+
+        fd1 = open (from_path, O_RDONLY);
+        if (fd1 < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "cannot open changelog file: %s (reason: %s)",
+                        from_path, strerror (errno));
+                goto out;
+        }
+
+        (void) snprintf (to_path, PATH_MAX, "%s%s",
+                         gfc->gfc_current_dir, basename (from_path));
+        (void) snprintf (dest, PATH_MAX, "%s%s",
+                         gfc->gfc_processing_dir, basename (from_path));
+
+        fd2 = open (to_path, O_CREAT | O_TRUNC | O_RDWR,
+                    S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
+        if (fd2 < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "cannot create ascii changelog file %s (reason %s)",
+                        to_path, strerror (errno));
+                goto close_fd;
+        } else {
+                ret = gf_changelog_decode (this, gfc, fd1,
+                                           fd2, &stbuf, &zerob);
+
+                close (fd2);
+
+                if (!ret) {
+                        /* move it to processing on a successfull
+                           decode */
+                        ret = rename (to_path, dest);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "error moving %s to processing dir"
+                                        " (reason: %s)", to_path,
+                                        strerror (errno));
+                }
+
+                /* remove it from .current if it's an empty file */
+                if (zerob) {
+                        ret = unlink (to_path);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "could not unlink %s (reason: %s",
+                                        to_path, strerror (errno));
+                }
+        }
+
+ close_fd:
+        close (fd1);
+
+ out:
+        return ret;
+}
+
+static char *
+gf_changelog_ext_change (xlator_t *this,
+                         gf_changelog_t *gfc, char *path, size_t readlen)
+{
+        int     alo = 0;
+        int     ret = 0;
+        size_t  len = 0;
+        char   *buf = NULL;
+
+        buf = path;
+        while (len < readlen) {
+                if (*buf == '\0') {
+                        alo = 1;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "processing changelog: %s", path);
+                        ret = gf_changelog_consume (this, gfc, path);
+                }
+
+                if (ret)
+                        break;
+
+                len++; buf++;
+                if (alo) {
+                        alo = 0;
+                        path = buf;
+                }
+        }
+
+        return (ret) ? NULL : path;
+}
+
+void *
+gf_changelog_process (void *data)
+{
+        ssize_t         len      = 0;
+        ssize_t         offlen   = 0;
+        xlator_t       *this     = NULL;
+        char           *sbuf     = NULL;
+        gf_changelog_t *gfc      = NULL;
+        char from_path[PATH_MAX] = {0,};
+
+        gfc = (gf_changelog_t *) data;
+        this = gfc->this;
+
+        pthread_detach (pthread_self());
+
+        for (;;) {
+                len = gf_changelog_read_path (gfc->gfc_sockfd,
+                                              from_path + offlen,
+                                              PATH_MAX - offlen);
+                if (len < 0)
+                        continue; /* ignore it for now */
+
+                if (len == 0) { /* close() from the changelog translator */
+                        gf_log (this->name, GF_LOG_INFO, "close from changelog"
+                                " notification translator.");
+
+                        if (gfc->gfc_connretries != 1) {
+                                if (!gf_changelog_notification_init(this, gfc))
+                                        continue;
+                        }
+
+                        byebye = 1;
+                        break;
+                }
+
+                len += offlen;
+                sbuf = gf_changelog_ext_change (this, gfc, from_path, len);
+                if (!sbuf) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "could not extract changelog filename");
+                        continue;
+                }
+
+                offlen = 0;
+                if (sbuf != (from_path + len)) {
+                        offlen = from_path + len - sbuf;
+                        memmove (from_path, sbuf, offlen);
+                }
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "byebye (%d) from processing thread...", byebye);
+        return NULL;
+}
diff --git a/xlators/features/changelog/lib/src/gf-changelog.c b/xlators/features/changelog/lib/src/gf-changelog.c
new file mode 100644
index 000000000..ca8e373e7
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog.c
@@ -0,0 +1,515 @@
+/*
+   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 <errno.h>
+#include <dirent.h>
+#include <stddef.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+#include <string.h>
+
+#include "globals.h"
+#include "glusterfs.h"
+#include "logging.h"
+
+#include "gf-changelog-helpers.h"
+
+/* from the changelog translator */
+#include "changelog-misc.h"
+#include "changelog-mem-types.h"
+
+int byebye = 0;
+
+static void
+gf_changelog_cleanup (gf_changelog_t *gfc)
+{
+        /* socket */
+        if (gfc->gfc_sockfd != -1)
+                close (gfc->gfc_sockfd);
+        /* tracker fd */
+        if (gfc->gfc_fd != -1)
+                close (gfc->gfc_fd);
+        /* processing dir */
+        if (gfc->gfc_dir)
+                closedir (gfc->gfc_dir);
+
+        if (gfc->gfc_working_dir)
+                free (gfc->gfc_working_dir); /* allocated by realpath */
+}
+
+void
+__attribute__ ((constructor)) gf_changelog_ctor (void)
+{
+        glusterfs_ctx_t *ctx = NULL;
+
+        ctx = glusterfs_ctx_new ();
+        if (!ctx)
+                return;
+
+        if (glusterfs_globals_init (ctx)) {
+                free (ctx);
+                ctx = NULL;
+                return;
+        }
+
+        THIS->ctx = ctx;
+}
+
+void
+__attribute__ ((destructor)) gf_changelog_dtor (void)
+{
+        xlator_t        *this = NULL;
+        glusterfs_ctx_t *ctx  = NULL;
+        gf_changelog_t  *gfc  = NULL;
+
+        this = THIS;
+        if (!this)
+                return;
+
+        ctx = this->ctx;
+        gfc = this->private;
+
+        if (gfc) {
+                gf_changelog_cleanup (gfc);
+                GF_FREE (gfc);
+        }
+
+        if (ctx) {
+                pthread_mutex_destroy (&ctx->lock);
+                free (ctx);
+                ctx = NULL;
+        }
+}
+
+
+static int
+gf_changelog_open_dirs (gf_changelog_t *gfc)
+{
+        int  ret                    = -1;
+        DIR *dir                    = NULL;
+        int  tracker_fd             = 0;
+        char tracker_path[PATH_MAX] = {0,};
+
+        (void) snprintf (gfc->gfc_current_dir, PATH_MAX,
+                         "%s/"GF_CHANGELOG_CURRENT_DIR"/",
+                         gfc->gfc_working_dir);
+        ret = mkdir_p (gfc->gfc_current_dir, 0600, _gf_false);
+        if (ret)
+                goto out;
+
+        (void) snprintf (gfc->gfc_processed_dir, PATH_MAX,
+                         "%s/"GF_CHANGELOG_PROCESSED_DIR"/",
+                         gfc->gfc_working_dir);
+        ret = mkdir_p (gfc->gfc_processed_dir, 0600, _gf_false);
+        if (ret)
+                goto out;
+
+        (void) snprintf (gfc->gfc_processing_dir, PATH_MAX,
+                         "%s/"GF_CHANGELOG_PROCESSING_DIR"/",
+                         gfc->gfc_working_dir);
+        ret = mkdir_p (gfc->gfc_processing_dir, 0600, _gf_false);
+        if (ret)
+                goto out;
+
+        dir = opendir (gfc->gfc_processing_dir);
+        if (!dir) {
+                gf_log ("", GF_LOG_ERROR,
+                        "opendir() error [reason: %s]", strerror (errno));
+                goto out;
+        }
+
+        gfc->gfc_dir = dir;
+
+        (void) snprintf (tracker_path, PATH_MAX,
+                         "%s/"GF_CHANGELOG_TRACKER, gfc->gfc_working_dir);
+
+        tracker_fd = open (tracker_path, O_CREAT | O_APPEND | O_RDWR,
+                           S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
+        if (tracker_fd < 0) {
+                closedir (gfc->gfc_dir);
+                ret = -1;
+                goto out;
+        }
+
+        gfc->gfc_fd = tracker_fd;
+        ret = 0;
+ out:
+        return ret;
+}
+
+int
+gf_changelog_notification_init (xlator_t *this, gf_changelog_t *gfc)
+{
+        int                ret    = 0;
+        int                len    = 0;
+        int                tries  = 0;
+        int                sockfd = 0;
+        struct sockaddr_un remote;
+
+        this = gfc->this;
+
+        if (gfc->gfc_sockfd != -1) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Reconnecting...");
+                close (gfc->gfc_sockfd);
+        }
+
+        sockfd = socket (AF_UNIX, SOCK_STREAM, 0);
+        if (sockfd < 0) {
+                ret = -1;
+                goto out;
+        }
+
+        CHANGELOG_MAKE_SOCKET_PATH (gfc->gfc_brickpath,
+                                    gfc->gfc_sockpath, PATH_MAX);
+        gf_log (this->name, GF_LOG_INFO,
+                "connecting to changelog socket: %s (brick: %s)",
+                gfc->gfc_sockpath, gfc->gfc_brickpath);
+
+        remote.sun_family = AF_UNIX;
+        strcpy (remote.sun_path, gfc->gfc_sockpath);
+
+        len = strlen (remote.sun_path) + sizeof (remote.sun_family);
+
+        while (tries < gfc->gfc_connretries) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "connection attempt %d/%d...",
+                        tries + 1, gfc->gfc_connretries);
+
+                /* initiate a connect */
+                if (connect (sockfd, (struct sockaddr *) &remote, len) == 0) {
+                        gfc->gfc_sockfd = sockfd;
+                        break;
+                }
+
+                tries++;
+                sleep (2);
+        }
+
+        if (tries == gfc->gfc_connretries) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "could not connect to changelog socket!"
+                        " bailing out...");
+                ret = -1;
+        } else
+                gf_log (this->name, GF_LOG_INFO,
+                        "connection successful");
+
+ out:
+        return ret;
+}
+
+int
+gf_changelog_done (char *file)
+{
+        int             ret    = -1;
+        char           *buffer = NULL;
+        xlator_t       *this   = NULL;
+        gf_changelog_t *gfc    = NULL;
+        char to_path[PATH_MAX] = {0,};
+
+        errno = EINVAL;
+
+        this = THIS;
+        if (!this)
+                goto out;
+
+        gfc = (gf_changelog_t *) this->private;
+        if (!gfc)
+                goto out;
+
+        if (!file || !strlen (file))
+                goto out;
+
+        /* make sure 'file' is inside ->gfc_working_dir */
+        buffer = realpath (file, NULL);
+        if (!buffer)
+                goto out;
+
+        if (strncmp (gfc->gfc_working_dir,
+                     buffer, strlen (gfc->gfc_working_dir)))
+                goto out;
+
+        (void) snprintf (to_path, PATH_MAX, "%s%s",
+                         gfc->gfc_processed_dir, basename (buffer));
+        gf_log (this->name, GF_LOG_DEBUG,
+                "moving %s to processed directory", file);
+        ret = rename (buffer, to_path);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "cannot move %s to %s (reason: %s)",
+                        file, to_path, strerror (errno));
+                goto out;
+        }
+
+        ret = 0;
+
+ out:
+        if (buffer)
+                free (buffer); /* allocated by realpath() */
+        return ret;
+}
+
+/**
+ * @API
+ *  for a set of changelogs, start from the begining
+ */
+int
+gf_changelog_start_fresh ()
+{
+        xlator_t *this = NULL;
+        gf_changelog_t *gfc = NULL;
+
+        this = THIS;
+        if (!this)
+                goto out;
+
+        errno = EINVAL;
+
+        gfc = (gf_changelog_t *) this->private;
+        if (!gfc)
+                goto out;
+
+        if (gf_ftruncate (gfc->gfc_fd, 0))
+                goto out;
+
+        return 0;
+
+ out:
+        return -1;
+}
+
+/**
+ * @API
+ * return the next changelog file entry. zero means all chanelogs
+ * consumed.
+ */
+ssize_t
+gf_changelog_next_change (char *bufptr, size_t maxlen)
+{
+        ssize_t         size       = 0;
+        int             tracker_fd = 0;
+        xlator_t       *this       = NULL;
+        gf_changelog_t *gfc        = NULL;
+        char buffer[PATH_MAX]      = {0,};
+
+        errno = EINVAL;
+
+        this = THIS;
+        if (!this)
+                goto out;
+
+        gfc = (gf_changelog_t *) this->private;
+        if (!gfc)
+                goto out;
+
+        tracker_fd = gfc->gfc_fd;
+
+        size = gf_readline (tracker_fd, buffer, maxlen);
+        if (size < 0)
+                goto out;
+        if (size == 0)
+                return 0;
+
+        memcpy (bufptr, buffer, size - 1);
+        *(buffer + size) = '\0';
+
+        return size;
+
+ out:
+        return -1;
+}
+
+/**
+ * @API
+ *  gf_changelog_scan() - scan and generate a list of change entries
+ *
+ * calling this api multiple times (without calling gf_changlog_done())
+ * would result new changelogs(s) being refreshed in the tracker file.
+ * This call also acts as a cancellation point for the consumer.
+ */
+ssize_t
+gf_changelog_scan ()
+{
+        int             ret        = 0;
+        int             tracker_fd = 0;
+        size_t          len        = 0;
+        size_t          off        = 0;
+        xlator_t       *this       = NULL;
+        size_t          nr_entries = 0;
+        gf_changelog_t *gfc        = NULL;
+        struct dirent  *entryp     = NULL;
+        struct dirent  *result     = NULL;
+        char buffer[PATH_MAX]      = {0,};
+
+        this = THIS;
+        if (!this)
+                goto out;
+
+        gfc = (gf_changelog_t *) this->private;
+        if (!gfc)
+                goto out;
+
+        /**
+         * do we need to protect 'byebye' with locks? worst, the
+         * consumer would get notified during next scan().
+         */
+        if (byebye) {
+                errno = ECONNREFUSED;
+                goto out;
+        }
+
+        errno = EINVAL;
+
+        tracker_fd = gfc->gfc_fd;
+
+        if (gf_ftruncate (tracker_fd, 0))
+                goto out;
+
+        len = offsetof(struct dirent, d_name)
+                + pathconf(gfc->gfc_processing_dir, _PC_NAME_MAX) + 1;
+        entryp = GF_CALLOC (1, len,
+                            gf_changelog_mt_libgfchangelog_dirent_t);
+        if (!entryp)
+                goto out;
+
+        rewinddir (gfc->gfc_dir);
+        while (1) {
+                ret = readdir_r (gfc->gfc_dir, entryp, &result);
+                if (ret || !result)
+                        break;
+
+                if ( !strcmp (basename (entryp->d_name), ".")
+                     || !strcmp (basename (entryp->d_name), "..") )
+                        continue;
+
+                nr_entries++;
+
+                GF_CHANGELOG_FILL_BUFFER (gfc->gfc_processing_dir,
+                                          buffer, off,
+                                          strlen (gfc->gfc_processing_dir));
+                GF_CHANGELOG_FILL_BUFFER (entryp->d_name, buffer,
+                                          off, strlen (entryp->d_name));
+                GF_CHANGELOG_FILL_BUFFER ("\n", buffer, off, 1);
+
+                if (gf_changelog_write (tracker_fd, buffer, off) != off) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "error writing changelog filename"
+                                " to tracker file");
+                        break;
+                }
+                off = 0;
+        }
+
+        GF_FREE (entryp);
+
+        if (!result) {
+                if (gf_lseek (tracker_fd, 0, SEEK_SET) != -1)
+                        return nr_entries;
+        }
+ out:
+        return -1;
+}
+
+/**
+ * @API
+ *  gf_changelog_register() - register a client for updates.
+ */
+int
+gf_changelog_register (char *brick_path, char *scratch_dir,
+                       char *log_file, int log_level, int max_reconnects)
+{
+        int             i    = 0;
+        int             ret  = -1;
+        int             errn = 0;
+        xlator_t       *this = NULL;
+        gf_changelog_t *gfc  = NULL;
+
+        this = THIS;
+        if (!this->ctx)
+                goto out;
+
+        errno = ENOMEM;
+
+        gfc = GF_CALLOC (1, sizeof (*gfc),
+                         gf_changelog_mt_libgfchangelog_t);
+        if (!gfc)
+                goto out;
+
+        gfc->this = this;
+
+        gfc->gfc_dir = NULL;
+        gfc->gfc_fd = gfc->gfc_sockfd = -1;
+
+        gfc->gfc_working_dir = realpath (scratch_dir, NULL);
+        if (!gfc->gfc_working_dir) {
+                errn = errno;
+                goto cleanup;
+        }
+
+        ret = gf_changelog_open_dirs (gfc);
+        if (ret) {
+                errn = errno;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "could not create entries in scratch dir");
+                goto cleanup;
+        }
+
+        /* passing ident as NULL means to use default ident for syslog */
+        if (gf_log_init (this->ctx, log_file, NULL))
+                goto cleanup;
+
+        gf_log_set_loglevel ((log_level == -1) ? GF_LOG_INFO :
+                             log_level);
+
+        gfc->gfc_connretries = (max_reconnects <= 0) ? 1 : max_reconnects;
+        (void) strncpy (gfc->gfc_brickpath, brick_path, PATH_MAX);
+
+        ret = gf_changelog_notification_init (this, gfc);
+        if (ret) {
+                errn = errno;
+                goto cleanup;
+        }
+
+        ret = gf_thread_create (&gfc->gfc_changelog_processor,
+				NULL, gf_changelog_process, gfc);
+        if (ret) {
+                errn = errno;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "error creating changelog processor thread"
+                        " new changes won't be recorded!!!");
+                goto cleanup;
+        }
+
+        for (; i < 256; i++) {
+                gfc->rfc3986[i] =
+                        (isalnum(i) || i == '~' ||
+                        i == '-' || i == '.' || i == '_') ? i : 0;
+        }
+
+        ret = 0;
+        this->private = gfc;
+
+        goto out;
+
+ cleanup:
+        gf_changelog_cleanup (gfc);
+        GF_FREE (gfc);
+        this->private = NULL;
+        errno = errn;
+
+ out:
+        return ret;
+}
diff --git a/xlators/features/changelog/src/Makefile.am b/xlators/features/changelog/src/Makefile.am
new file mode 100644
index 000000000..e85031ad4
--- /dev/null
+++ b/xlators/features/changelog/src/Makefile.am
@@ -0,0 +1,19 @@
+xlator_LTLIBRARIES = changelog.la
+
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+noinst_HEADERS = changelog-helpers.h changelog-mem-types.h changelog-rt.h \
+	changelog-misc.h changelog-encoders.h changelog-notifier.h
+
+changelog_la_LDFLAGS = -module -avoidversion
+
+changelog_la_SOURCES = changelog.c changelog-rt.c changelog-helpers.c \
+	changelog-encoders.c changelog-notifier.c
+changelog_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src -fPIC -D_FILE_OFFSET_BITS=64 \
+	-D_GNU_SOURCE -D$(GF_HOST_OS) -shared -nostartfiles -DDATADIR=\"$(localstatedir)\"
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/changelog/src/changelog-encoders.c b/xlators/features/changelog/src/changelog-encoders.c
new file mode 100644
index 000000000..553eec85c
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-encoders.c
@@ -0,0 +1,176 @@
+/*
+   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 "changelog-encoders.h"
+
+size_t
+entry_fn (void *data, char *buffer, gf_boolean_t encode)
+{
+        char    *tmpbuf = NULL;
+        size_t  bufsz  = 0;
+        struct changelog_entry_fields *ce = NULL;
+
+        ce = (struct changelog_entry_fields *) data;
+
+        if (encode) {
+                tmpbuf = uuid_utoa (ce->cef_uuid);
+                CHANGELOG_FILL_BUFFER (buffer, bufsz, tmpbuf, strlen (tmpbuf));
+        } else {
+                CHANGELOG_FILL_BUFFER (buffer, bufsz,
+                                       ce->cef_uuid, sizeof (uuid_t));
+        }
+
+        CHANGELOG_FILL_BUFFER (buffer, bufsz, "/", 1);
+        CHANGELOG_FILL_BUFFER (buffer, bufsz,
+                               ce->cef_bname, strlen (ce->cef_bname));
+        return bufsz;
+}
+
+size_t
+fop_fn (void *data, char *buffer, gf_boolean_t encode)
+{
+        char buf[10]          = {0,};
+        size_t         bufsz = 0;
+        glusterfs_fop_t fop   = 0;
+
+        fop = *(glusterfs_fop_t *) data;
+
+        if (encode) {
+                (void) snprintf (buf, sizeof (buf), "%d", fop);
+                CHANGELOG_FILL_BUFFER (buffer, bufsz, buf, strlen (buf));
+        } else
+                CHANGELOG_FILL_BUFFER (buffer, bufsz, &fop, sizeof (fop));
+
+        return bufsz;
+}
+
+void
+entry_free_fn (void *data)
+{
+        changelog_opt_t *co = data;
+
+        if (!co)
+                return;
+
+        GF_FREE (co->co_entry.cef_bname);
+}
+
+/**
+ * try to write all data in one shot
+ */
+
+static inline void
+changelog_encode_write_xtra (changelog_log_data_t *cld,
+                             char *buffer, size_t *off, gf_boolean_t encode)
+{
+        int              i      = 0;
+        size_t           offset = 0;
+        void            *data   = NULL;
+        changelog_opt_t *co     = NULL;
+
+        offset = *off;
+
+        co = (changelog_opt_t *) cld->cld_ptr;
+
+        for (; i < cld->cld_xtra_records; i++, co++) {
+                CHANGELOG_FILL_BUFFER (buffer, offset, "\0", 1);
+
+                switch (co->co_type) {
+                case CHANGELOG_OPT_REC_FOP:
+                        data = &co->co_fop;
+                        break;
+                case CHANGELOG_OPT_REC_ENTRY:
+                        data = &co->co_entry;
+                        break;
+                }
+
+                if (co->co_convert)
+                        offset += co->co_convert (data,
+                                                  buffer + offset, encode);
+                else /* no coversion: write it out as it is */
+                        CHANGELOG_FILL_BUFFER (buffer, offset,
+                                               data, co->co_len);
+        }
+
+        *off = offset;
+}
+
+int
+changelog_encode_ascii (xlator_t *this, changelog_log_data_t *cld)
+{
+        size_t            off      = 0;
+        size_t            gfid_len = 0;
+        char             *gfid_str = NULL;
+        char             *buffer   = NULL;
+        changelog_priv_t *priv     = NULL;
+
+        priv = this->private;
+
+        gfid_str = uuid_utoa (cld->cld_gfid);
+        gfid_len = strlen (gfid_str);
+
+        /* extra bytes for decorations */
+        buffer = alloca (gfid_len + cld->cld_ptr_len + 10);
+        CHANGELOG_STORE_ASCII (priv, buffer,
+                               off, gfid_str, gfid_len, cld);
+
+        if (cld->cld_xtra_records)
+                changelog_encode_write_xtra (cld, buffer, &off, _gf_true);
+
+        CHANGELOG_FILL_BUFFER (buffer, off, "\0", 1);
+
+        return changelog_write_change (priv, buffer, off);
+}
+
+int
+changelog_encode_binary (xlator_t *this, changelog_log_data_t *cld)
+{
+        size_t            off    = 0;
+        char             *buffer = NULL;
+        changelog_priv_t *priv   = NULL;
+
+        priv = this->private;
+
+        /* extra bytes for decorations */
+        buffer = alloca (sizeof (uuid_t) + cld->cld_ptr_len + 10);
+        CHANGELOG_STORE_BINARY (priv, buffer, off, cld->cld_gfid, cld);
+
+        if (cld->cld_xtra_records)
+                changelog_encode_write_xtra (cld, buffer, &off, _gf_false);
+
+        CHANGELOG_FILL_BUFFER (buffer, off, "\0", 1);
+
+        return changelog_write_change (priv, buffer, off);
+}
+
+static struct changelog_encoder
+cb_encoder[] = {
+        [CHANGELOG_ENCODE_BINARY] =
+        {
+                .encoder = CHANGELOG_ENCODE_BINARY,
+                .encode = changelog_encode_binary,
+        },
+        [CHANGELOG_ENCODE_ASCII] =
+        {
+                .encoder = CHANGELOG_ENCODE_ASCII,
+                .encode = changelog_encode_ascii,
+        },
+};
+
+void
+changelog_encode_change( changelog_priv_t * priv)
+{
+        priv->ce = &cb_encoder[priv->encode_mode];
+}
diff --git a/xlators/features/changelog/src/changelog-encoders.h b/xlators/features/changelog/src/changelog-encoders.h
new file mode 100644
index 000000000..a3efbee05
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-encoders.h
@@ -0,0 +1,46 @@
+/*
+   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 _CHANGELOG_ENCODERS_H
+#define _CHANGELOG_ENCODERS_H
+
+#include "xlator.h"
+#include "defaults.h"
+
+#include "changelog-helpers.h"
+
+#define CHANGELOG_STORE_ASCII(priv, buf, off, gfid, gfid_len, cld) do { \
+                CHANGELOG_FILL_BUFFER (buffer, off,                     \
+                                       priv->maps[cld->cld_type], 1);   \
+                CHANGELOG_FILL_BUFFER (buffer,                          \
+                                       off, gfid, gfid_len);            \
+        } while (0)
+
+#define CHANGELOG_STORE_BINARY(priv, buf, off, gfid, cld) do {          \
+                CHANGELOG_FILL_BUFFER (buffer, off,                     \
+                                       priv->maps[cld->cld_type], 1);   \
+                CHANGELOG_FILL_BUFFER (buffer,                          \
+                                       off, gfid, sizeof (uuid_t));     \
+        } while (0)
+
+size_t
+entry_fn (void *data, char *buffer, gf_boolean_t encode);
+size_t
+fop_fn (void *data, char *buffer, gf_boolean_t encode);
+void
+entry_free_fn (void *data);
+int
+changelog_encode_binary (xlator_t *, changelog_log_data_t *);
+int
+changelog_encode_ascii (xlator_t *, changelog_log_data_t *);
+void
+changelog_encode_change(changelog_priv_t *);
+
+#endif /* _CHANGELOG_ENCODERS_H */
diff --git a/xlators/features/changelog/src/changelog-helpers.c b/xlators/features/changelog/src/changelog-helpers.c
new file mode 100644
index 000000000..7ab0091b5
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-helpers.c
@@ -0,0 +1,693 @@
+/*
+   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 "xlator.h"
+#include "defaults.h"
+#include "logging.h"
+#include "iobuf.h"
+
+#include "changelog-helpers.h"
+#include "changelog-mem-types.h"
+
+#include "changelog-encoders.h"
+#include <pthread.h>
+
+void
+changelog_thread_cleanup (xlator_t *this, pthread_t thr_id)
+{
+        int   ret    = 0;
+        void *retval = NULL;
+
+        /* send a cancel request to the thread */
+        ret = pthread_cancel (thr_id);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "could not cancel thread (reason: %s)",
+                        strerror (errno));
+                goto out;
+        }
+
+        ret = pthread_join (thr_id, &retval);
+        if (ret || (retval != PTHREAD_CANCELED)) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "cancel request not adhered as expected"
+                        " (reason: %s)", strerror (errno));
+        }
+
+ out:
+        return;
+}
+
+inline void *
+changelog_get_usable_buffer (changelog_local_t *local)
+{
+        changelog_log_data_t *cld = NULL;
+
+        cld = &local->cld;
+        if (!cld->cld_iobuf)
+                return NULL;
+
+        return cld->cld_iobuf->ptr;
+}
+
+inline void
+changelog_set_usable_record_and_length (changelog_local_t *local,
+                                        size_t len, int xr)
+{
+        changelog_log_data_t *cld = NULL;
+
+        cld = &local->cld;
+
+        cld->cld_ptr_len = len;
+        cld->cld_xtra_records = xr;
+}
+
+void
+changelog_local_cleanup (xlator_t *xl, changelog_local_t *local)
+{
+        int                   i   = 0;
+        changelog_opt_t      *co  = NULL;
+        changelog_log_data_t *cld = NULL;
+
+        if (!local)
+                return;
+
+        cld = &local->cld;
+
+        /* cleanup dynamic allocation for extra records */
+        if (cld->cld_xtra_records) {
+                co = (changelog_opt_t *) cld->cld_ptr;
+                for (; i < cld->cld_xtra_records; i++, co++)
+                        if (co->co_free)
+                                co->co_free (co);
+        }
+
+        CHANGELOG_IOBUF_UNREF (cld->cld_iobuf);
+
+        if (local->inode)
+                inode_unref (local->inode);
+
+        mem_put (local);
+}
+
+inline int
+changelog_write (int fd, char *buffer, size_t len)
+{
+        ssize_t size = 0;
+        size_t writen = 0;
+
+        while (writen < len) {
+                size = write (fd,
+                              buffer + writen, len - writen);
+                if (size <= 0)
+                        break;
+
+                writen += size;
+        }
+
+        return (writen != len);
+}
+
+static int
+changelog_rollover_changelog (xlator_t *this,
+                              changelog_priv_t *priv, unsigned long ts)
+{
+        int   ret            = -1;
+        int   notify         = 0;
+        char *bname          = NULL;
+        char ofile[PATH_MAX] = {0,};
+        char nfile[PATH_MAX] = {0,};
+
+        if (priv->changelog_fd != -1) {
+                close (priv->changelog_fd);
+                priv->changelog_fd = -1;
+        }
+
+        (void) snprintf (ofile, PATH_MAX,
+                         "%s/"CHANGELOG_FILE_NAME, priv->changelog_dir);
+        (void) snprintf (nfile, PATH_MAX,
+                         "%s/"CHANGELOG_FILE_NAME".%lu",
+                         priv->changelog_dir, ts);
+
+        ret = rename (ofile, nfile);
+        if (!ret)
+                notify = 1;
+
+        if (ret && (errno == ENOENT)) {
+                ret = 0;
+        }
+
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "error renaming %s -> %s (reason %s)",
+                        ofile, nfile, strerror (errno));
+        }
+
+        if (notify) {
+                bname = basename (nfile);
+                gf_log (this->name, GF_LOG_DEBUG, "notifying: %s", bname);
+                ret = changelog_write (priv->wfd, bname, strlen (bname) + 1);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Failed to send file name to notify thread"
+                                " (reason: %s)", strerror (errno));
+                }
+        }
+
+        return ret;
+}
+
+int
+changelog_open (xlator_t *this,
+                changelog_priv_t *priv)
+{
+        int fd                        = 0;
+        int ret                       = -1;
+        int flags                     = 0;
+        char buffer[1024]             = {0,};
+        char changelog_path[PATH_MAX] = {0,};
+
+        (void) snprintf (changelog_path, PATH_MAX,
+                         "%s/"CHANGELOG_FILE_NAME,
+                         priv->changelog_dir);
+
+        flags |= (O_CREAT | O_RDWR);
+        if (priv->fsync_interval == 0)
+                flags |= O_SYNC;
+
+        fd = open (changelog_path, flags,
+                   S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
+        if (fd < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "unable to open/create changelog file %s"
+                        " (reason: %s). change-logging will be"
+                        " inactive", changelog_path, strerror (errno));
+                goto out;
+        }
+
+        priv->changelog_fd = fd;
+
+        (void) snprintf (buffer, 1024, CHANGELOG_HEADER,
+                         CHANGELOG_VERSION_MAJOR,
+                         CHANGELOG_VERSION_MINOR,
+                         priv->ce->encoder);
+        ret = changelog_write_change (priv, buffer, strlen (buffer));
+        if (ret) {
+                close (priv->changelog_fd);
+                priv->changelog_fd = -1;
+                goto out;
+        }
+
+        ret = 0;
+
+ out:
+        return ret;
+}
+
+int
+changelog_start_next_change (xlator_t *this,
+                             changelog_priv_t *priv,
+                             unsigned long ts, gf_boolean_t finale)
+{
+        int ret = -1;
+
+        ret = changelog_rollover_changelog (this, priv, ts);
+
+        if (!ret && !finale)
+                ret = changelog_open (this, priv);
+
+        return ret;
+}
+
+/**
+ * return the length of entry
+ */
+inline size_t
+changelog_entry_length ()
+{
+        return sizeof (changelog_log_data_t);
+}
+
+int
+changelog_fill_rollover_data (changelog_log_data_t *cld, gf_boolean_t is_last)
+{
+        struct timeval tv = {0,};
+
+        cld->cld_type = CHANGELOG_TYPE_ROLLOVER;
+
+        if (gettimeofday (&tv, NULL))
+                return -1;
+
+        cld->cld_roll_time = (unsigned long) tv.tv_sec;
+        cld->cld_finale = is_last;
+        return 0;
+}
+
+int
+changelog_write_change (changelog_priv_t *priv, char *buffer, size_t len)
+{
+        return changelog_write (priv->changelog_fd, buffer, len);
+}
+
+inline int
+changelog_handle_change (xlator_t *this,
+                         changelog_priv_t *priv, changelog_log_data_t *cld)
+{
+        int ret = 0;
+
+        if (CHANGELOG_TYPE_IS_ROLLOVER (cld->cld_type)) {
+                changelog_encode_change(priv);
+                ret = changelog_start_next_change (this, priv,
+                                                   cld->cld_roll_time,
+                                                   cld->cld_finale);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Problem rolling over changelog(s)");
+                goto out;
+        }
+
+        /**
+         * case when there is reconfigure done (disabling changelog) and there
+         * are still fops that have updates in prgress.
+         */
+        if (priv->changelog_fd == -1)
+                return 0;
+
+        if (CHANGELOG_TYPE_IS_FSYNC (cld->cld_type)) {
+                ret = fsync (priv->changelog_fd);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "fsync failed (reason: %s)",
+                                strerror (errno));
+                }
+                goto out;
+        }
+
+        ret = priv->ce->encode (this, cld);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "error writing changelog to disk");
+        }
+
+ out:
+        return ret;
+}
+
+changelog_local_t *
+changelog_local_init (xlator_t *this, inode_t *inode,
+                      uuid_t gfid, int xtra_records,
+                      gf_boolean_t update_flag)
+{
+        changelog_local_t *local = NULL;
+        struct iobuf      *iobuf = NULL;
+
+        /**
+         * We relax the presence of inode if @update_flag is true.
+         * The caller (implmentation of the fop) needs to be careful to
+         * not blindly use local->inode.
+         */
+        if (!update_flag && !inode) {
+                gf_log_callingfn (this->name, GF_LOG_WARNING,
+                                  "inode needed for version checking !!!");
+                goto out;
+        }
+
+        if (xtra_records) {
+                iobuf = iobuf_get2 (this->ctx->iobuf_pool,
+                                    xtra_records * CHANGELOG_OPT_RECORD_LEN);
+                if (!iobuf)
+                        goto out;
+        }
+
+        local = mem_get0 (this->local_pool);
+        if (!local) {
+                CHANGELOG_IOBUF_UNREF (iobuf);
+                goto out;
+        }
+
+        local->update_no_check = update_flag;
+
+        uuid_copy (local->cld.cld_gfid, gfid);
+
+        local->cld.cld_iobuf = iobuf;
+        local->cld.cld_xtra_records = 0; /* set by the caller */
+
+        if (inode)
+                local->inode = inode_ref (inode);
+
+ out:
+        return local;
+}
+
+int
+changelog_forget (xlator_t *this, inode_t *inode)
+{
+        uint64_t ctx_addr = 0;
+        changelog_inode_ctx_t *ctx = NULL;
+
+        inode_ctx_del (inode, this, &ctx_addr);
+        if (!ctx_addr)
+                return 0;
+
+        ctx = (changelog_inode_ctx_t *) (long) ctx_addr;
+        GF_FREE (ctx);
+
+        return 0;
+}
+
+int
+changelog_inject_single_event (xlator_t *this,
+                               changelog_priv_t *priv,
+                               changelog_log_data_t *cld)
+{
+        return priv->cd.dispatchfn (this, priv, priv->cd.cd_data, cld, NULL);
+}
+
+/**
+ * TODO: these threads have many thing in common (wake up after
+ * a certain time etc..). move them into separate routine.
+ */
+void *
+changelog_rollover (void *data)
+{
+        int                     ret   = 0;
+        xlator_t               *this  = NULL;
+        struct timeval          tv    = {0,};
+        changelog_log_data_t    cld   = {0,};
+        changelog_time_slice_t *slice = NULL;
+        changelog_priv_t       *priv  = data;
+
+        this = priv->cr.this;
+        slice = &priv->slice;
+
+        while (1) {
+                tv.tv_sec  = priv->rollover_time;
+                tv.tv_usec = 0;
+
+                ret = select (0, NULL, NULL, NULL, &tv);
+                if (ret)
+                        continue;
+
+                ret = changelog_fill_rollover_data (&cld, _gf_false);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "failed to fill rollover data");
+                        continue;
+                }
+
+                LOCK (&priv->lock);
+                {
+                        ret = changelog_inject_single_event (this, priv, &cld);
+                        if (!ret)
+                                SLICE_VERSION_UPDATE (slice);
+                }
+                UNLOCK (&priv->lock);
+        }
+
+        return NULL;
+}
+
+void *
+changelog_fsync_thread (void *data)
+{
+        int                   ret  = 0;
+        xlator_t             *this = NULL;
+        struct timeval        tv   = {0,};
+        changelog_log_data_t  cld  = {0,};
+        changelog_priv_t     *priv = data;
+
+        this = priv->cf.this;
+        cld.cld_type = CHANGELOG_TYPE_FSYNC;
+
+        while (1) {
+                tv.tv_sec  = priv->fsync_interval;
+                tv.tv_usec = 0;
+
+                ret = select (0, NULL, NULL, NULL, &tv);
+                if (ret)
+                        continue;
+
+                ret = changelog_inject_single_event (this, priv, &cld);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "failed to inject fsync event");
+        }
+
+        return NULL;
+}
+
+/* macros for inode/changelog version checks */
+
+#define INODE_VERSION_UPDATE(priv, inode, iver, slice, type) do {       \
+                LOCK (&inode->lock);                                    \
+                {                                                       \
+                        LOCK (&priv->lock);                             \
+                        {                                               \
+                                *iver = slice->changelog_version[type]; \
+                        }                                               \
+                        UNLOCK (&priv->lock);                           \
+                }                                                       \
+                UNLOCK (&inode->lock);                                  \
+        } while (0)
+
+#define INODE_VERSION_EQUALS_SLICE(priv, ver, slice, type, upd) do {    \
+                LOCK (&priv->lock);                                     \
+                {                                                       \
+                        upd = (ver == slice->changelog_version[type])   \
+                                ? _gf_false : _gf_true;                 \
+                }                                                       \
+                UNLOCK (&priv->lock);                                   \
+        } while (0)
+
+static int
+__changelog_inode_ctx_set (xlator_t *this,
+                           inode_t *inode, changelog_inode_ctx_t *ctx)
+{
+        uint64_t ctx_addr = (uint64_t) ctx;
+        return __inode_ctx_set (inode, this, &ctx_addr);
+}
+
+/**
+ * one shot routine to get the address and the value of a inode version
+ * for a particular type.
+ */
+static changelog_inode_ctx_t *
+__changelog_inode_ctx_get (xlator_t *this,
+                           inode_t *inode, unsigned long **iver,
+                           unsigned long *version, changelog_log_type type)
+{
+        int                    ret      = 0;
+        uint64_t               ctx_addr = 0;
+        changelog_inode_ctx_t *ctx      = NULL;
+
+        ret = __inode_ctx_get (inode, this, &ctx_addr);
+        if (ret < 0)
+                ctx_addr = 0;
+        if (ctx_addr != 0) {
+                ctx = (changelog_inode_ctx_t *) (long)ctx_addr;
+                goto out;
+        }
+
+        ctx = GF_CALLOC (1, sizeof (*ctx), gf_changelog_mt_inode_ctx_t);
+        if (!ctx)
+                goto out;
+
+        ret = __changelog_inode_ctx_set (this, inode, ctx);
+        if (ret) {
+                GF_FREE (ctx);
+                ctx = NULL;
+        }
+
+ out:
+        if (ctx && iver && version) {
+                *iver = CHANGELOG_INODE_VERSION_TYPE (ctx, type);
+                *version = **iver;
+        }
+
+        return ctx;
+}
+
+static changelog_inode_ctx_t *
+changelog_inode_ctx_get (xlator_t *this,
+                         inode_t *inode, unsigned long **iver,
+                         unsigned long *version, changelog_log_type type)
+{
+        changelog_inode_ctx_t *ctx = NULL;
+
+        LOCK (&inode->lock);
+        {
+                ctx = __changelog_inode_ctx_get (this,
+                                                 inode, iver, version, type);
+        }
+        UNLOCK (&inode->lock);
+
+        return ctx;
+}
+
+/**
+ * This is the main update routine. Locking has been made granular so as to
+ * maximize parallelism of fops - I'll try to explain it below using execution
+ * timelines.
+ *
+ * Basically, the contention is between multiple execution threads of this
+ * routine and the roll-over thread. So, instead of having a big lock, we hold
+ * granular locks: inode->lock and priv->lock. Now I'll explain what happens
+ * when there is an update and a roll-over at just about the same time.
+ * NOTE:
+ *  - the dispatcher itself synchronizes updates via it's own lock
+ *  - the slice version in incremented by the roll-over thread
+ *
+ * Case 1: When the rollover thread wins before the inode version can be
+ * compared with the slice version.
+ *
+ *          [updater]                 |             [rollover]
+ *                                    |
+ *                                    |           <SLICE: 1, 1, 1>
+ * <changelog_update>                 |
+ *   <changelog_inode_ctx_get>        |
+ *      <CTX: 1, 1, 1>                |
+ *                                    |         <dispatch-rollover-event>
+ *                                    |         LOCK (&priv->lock)
+ *                                    |            <SLICE_VERSION_UPDATE>
+ *                                    |              <SLICE: 2, 2, 2>
+ *                                    |         UNLOCK (&priv->lock)
+ *                                    |
+ * LOCK (&priv->lock)                 |
+ *   <INODE_VERSION_EQUALS_SLICE>     |
+ *    I: 1 <-> S: 2                   |
+ *    update: true                    |
+ * UNLOCK (&priv->lock)               |
+ *                                    |
+ * <if update == true>                |
+ *  <dispath-update-event>            |
+ *  <INODE_VERSION_UPDATE>            |
+ *   LOCK (&inode->lock)              |
+ *    LOCK (&priv->lock)              |
+ *     <CTX: 2, 1, 1>                 |
+ *    UNLOCK (&priv->lock)            |
+ *   UNLOCK (&inode->lock)            |
+ *
+ * Therefore, the change gets recorded in the next change (no lost change). If
+ * the slice version was ahead of the inode version (say I:1, S: 2), then
+ * anyway the comparison would result in a update (I: 1, S: 3).
+ *
+ * If the rollover time is too less, then there is another contention when the
+ * updater tries to bring up inode version to the slice version (this is also
+ * the case when the roll-over thread wakes up during INODE_VERSION_UPDATE.
+ *
+ *   <CTX: 1, 1, 1>                   |       <SLICE: 2, 2, 2>
+ *                                    |
+ *                                    |
+ * <dispath-update-event>             |
+ * <INODE_VERSION_UPDATE>             |
+ *  LOCK (&inode->lock)               |
+ *   LOCK (&priv->lock)               |
+ *    <CTX: 2, 1, 1>                  |
+ *   UNLOCK (&priv->lock)             |
+ *  UNLOCK (&inode->lock)             |
+ *                                    |         <dispatch-rollover-event>
+ *                                    |         LOCK (&priv->lock)
+ *                                    |            <SLICE_VERSION_UPDATE>
+ *                                    |              <SLICE: 3, 3, 3>
+ *                                    |         UNLOCK (&priv->lock)
+ *
+ *
+ * Case 2: When the fop thread wins
+ *
+ *          [updater]                 |             [rollover]
+ *                                    |
+ *                                    |           <SLICE: 1, 1, 1>
+ * <changelog_update>                 |
+ *   <changelog_inode_ctx_get>        |
+ *      <CTX: 0, 0, 0>                |
+ *                                    |
+ * LOCK (&priv->lock)                 |
+ *   <INODE_VERSION_EQUALS_SLICE>     |
+ *    I: 0 <-> S: 1                   |
+ *    update: true                    |
+ * UNLOCK (&priv->lock)               |
+ *                                    |         <dispatch-rollover-event>
+ *                                    |         LOCK (&priv->lock)
+ *                                    |            <SLICE_VERSION_UPDATE>
+ *                                    |              <SLICE: 2, 2, 2>
+ *                                    |         UNLOCK (&priv->lock)
+ * <if update == true>                |
+ *  <dispath-update-event>            |
+ *  <INODE_VERSION_UPDATE>            |
+ *   LOCK (&inode->lock)              |
+ *    LOCK (&priv->lock)              |
+ *     <CTX: 2, 0, 0>                 |
+ *    UNLOCK (&priv->lock)            |
+ *   UNLOCK (&inode->lock)            |
+ *
+ * Here again, if the inode version was equal to the slice version (I: 1, S: 1)
+ * then there is no need to record an update (as the equality of the two version
+ * signifies an update was recorded in the current time slice).
+ */
+inline void
+changelog_update (xlator_t *this, changelog_priv_t *priv,
+                  changelog_local_t *local, changelog_log_type type)
+{
+        int                     ret        = 0;
+        unsigned long          *iver       = NULL;
+        unsigned long           version    = 0;
+        inode_t                *inode      = NULL;
+        changelog_time_slice_t *slice      = NULL;
+        changelog_inode_ctx_t  *ctx        = NULL;
+        changelog_log_data_t   *cld_0      = NULL;
+        changelog_log_data_t   *cld_1      = NULL;
+        changelog_local_t      *next_local = NULL;
+        gf_boolean_t            need_upd   = _gf_true;
+
+        slice = &priv->slice;
+
+        /**
+         * for fops that do not require inode version checking
+         */
+        if (local->update_no_check)
+                goto update;
+
+        inode = local->inode;
+
+        ctx = changelog_inode_ctx_get (this,
+                                       inode, &iver, &version, type);
+        if (!ctx)
+                goto update;
+
+        INODE_VERSION_EQUALS_SLICE (priv, version, slice, type, need_upd);
+
+ update:
+        if (need_upd) {
+                cld_0 = &local->cld;
+                cld_0->cld_type = type;
+
+                if ( (next_local = local->prev_entry) != NULL ) {
+                        cld_1 = &next_local->cld;
+                        cld_1->cld_type = type;
+                }
+
+                ret = priv->cd.dispatchfn (this, priv,
+                                           priv->cd.cd_data, cld_0, cld_1);
+
+                /**
+                 * update after the dispatcher has successfully done
+                 * it's job.
+                 */
+                if (!local->update_no_check && iver && !ret)
+                        INODE_VERSION_UPDATE (priv, inode, iver, slice, type);
+        }
+
+        return;
+}
diff --git a/xlators/features/changelog/src/changelog-helpers.h b/xlators/features/changelog/src/changelog-helpers.h
new file mode 100644
index 000000000..ad79636b0
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-helpers.h
@@ -0,0 +1,395 @@
+/*
+   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 _CHANGELOG_HELPERS_H
+#define _CHANGELOG_HELPERS_H
+
+#include "locking.h"
+#include "timer.h"
+#include "pthread.h"
+#include "iobuf.h"
+
+#include "changelog-misc.h"
+
+/**
+ * the changelog entry
+ */
+typedef struct changelog_log_data {
+        /* rollover related */
+        unsigned long cld_roll_time;
+
+        /* reopen changelog? */
+        gf_boolean_t cld_finale;
+
+        changelog_log_type cld_type;
+
+        /**
+         * sincd gfid is _always_ a necessity, it's not a part
+         * of the iobuf. by doing this we do not add any overhead
+         * for data and metadata related fops.
+         */
+        uuid_t        cld_gfid;
+
+        /**
+         * iobufs are used for optionals records: pargfid, path,
+         * write offsets etc.. It's the fop implementers job
+         * to allocate (iobuf_get() in the fop) and get unref'ed
+         * in the callback (CHANGELOG_STACK_UNWIND).
+         */
+        struct iobuf *cld_iobuf;
+
+#define cld_ptr cld_iobuf->ptr
+
+        /**
+         * after allocation you can point this to the length of
+         * usable data, but make sure it does not exceed the
+         * the size of the requested iobuf.
+         */
+        size_t        cld_iobuf_len;
+
+#define cld_ptr_len cld_iobuf_len
+
+        /**
+         * number of optional records
+         */
+        int cld_xtra_records;
+} changelog_log_data_t;
+
+/**
+ * holder for dispatch function and private data
+ */
+
+typedef struct changelog_priv changelog_priv_t;
+
+typedef struct changelog_dispatcher {
+        void *cd_data;
+        int (*dispatchfn) (xlator_t *, changelog_priv_t *, void *,
+                           changelog_log_data_t *, changelog_log_data_t *);
+} changelog_dispatcher_t;
+
+struct changelog_bootstrap {
+        changelog_mode_t mode;
+        int (*ctor) (xlator_t *, changelog_dispatcher_t *);
+        int (*dtor) (xlator_t *, changelog_dispatcher_t *);
+};
+
+struct changelog_encoder {
+        changelog_encoder_t encoder;
+        int (*encode) (xlator_t *, changelog_log_data_t *);
+};
+
+
+/* xlator private */
+
+typedef struct changelog_time_slice {
+        /**
+         * just in case we need nanosecond granularity some day.
+         * field is unused as of now (maybe we'd need it later).
+         */
+        struct timeval tv_start;
+
+        /**
+         * version of changelog file, incremented each time changes
+         * rollover.
+         */
+        unsigned long changelog_version[CHANGELOG_MAX_TYPE];
+} changelog_time_slice_t;
+
+typedef struct changelog_rollover {
+        /* rollover thread */
+        pthread_t rollover_th;
+
+        xlator_t *this;
+} changelog_rollover_t;
+
+typedef struct changelog_fsync {
+        /* fsync() thread */
+        pthread_t fsync_th;
+
+        xlator_t *this;
+} changelog_fsync_t;
+
+# define CHANGELOG_MAX_CLIENTS  5
+typedef struct changelog_notify {
+        /* reader end of the pipe */
+        int rfd;
+
+        /* notifier thread */
+        pthread_t notify_th;
+
+        /* unique socket path */
+        char sockpath[PATH_MAX];
+
+        int socket_fd;
+
+        /**
+         * simple array of accept()'ed fds. Not scalable at all
+         * for large number of clients, but it's okay as we have
+         * a ahrd limit in this version (@CHANGELOG_MAX_CLIENTS).
+         */
+        int client_fd[CHANGELOG_MAX_CLIENTS];
+
+        xlator_t *this;
+} changelog_notify_t;
+
+struct changelog_priv {
+        gf_boolean_t active;
+
+        /* to generate unique socket file per brick */
+        char *changelog_brick;
+
+        /* logging directory */
+        char *changelog_dir;
+
+        /* one file for all changelog types */
+        int changelog_fd;
+
+        gf_lock_t lock;
+
+        /* writen end of the pipe */
+        int wfd;
+
+        /* rollover time */
+        int32_t rollover_time;
+
+        /* fsync() interval */
+        int32_t fsync_interval;
+
+        /* changelog type maps */
+        const char *maps[CHANGELOG_MAX_TYPE];
+
+        /* time slicer */
+        changelog_time_slice_t slice;
+
+        /* context of the updater */
+        changelog_dispatcher_t cd;
+
+        /* context of the rollover thread */
+        changelog_rollover_t cr;
+
+        /* context of fsync thread */
+        changelog_fsync_t cf;
+
+        /* context of the notifier thread */
+        changelog_notify_t cn;
+
+        /* operation mode */
+        changelog_mode_t op_mode;
+
+        /* bootstrap routine for 'current' logger */
+        struct changelog_bootstrap *cb;
+
+        /* encoder mode */
+        changelog_encoder_t encode_mode;
+
+        /* encoder */
+        struct changelog_encoder *ce;
+};
+
+struct changelog_local {
+        inode_t              *inode;
+        gf_boolean_t          update_no_check;
+
+        changelog_log_data_t  cld;
+
+        /**
+         * ->prev_entry is used in cases when there needs to be
+         * additional changelog entry for the parent (eg. rename)
+         * It's analogous to ->next in single linked list world,
+         * but we call it as ->prev_entry... ha ha ha
+         */
+        struct changelog_local *prev_entry;
+};
+
+typedef struct changelog_local changelog_local_t;
+
+/* inode version is stored in inode ctx */
+typedef struct changelog_inode_ctx {
+        unsigned long iversion[CHANGELOG_MAX_TYPE];
+} changelog_inode_ctx_t;
+
+#define CHANGELOG_INODE_VERSION_TYPE(ctx, type)  &(ctx->iversion[type])
+
+/**
+ * Optional Records:
+ *  fops that need to save additional information request a array of
+ *  @changelog_opt_t struct. The array is allocated via @iobufs.
+ */
+typedef enum {
+        CHANGELOG_OPT_REC_FOP,
+        CHANGELOG_OPT_REC_ENTRY,
+} changelog_optional_rec_type_t;
+
+struct changelog_entry_fields {
+        uuid_t  cef_uuid;
+        char   *cef_bname;
+};
+
+typedef struct {
+        /**
+         * @co_covert can be used to do post-processing of the record before
+         * it's persisted to the CHANGELOG. If this is NULL, then the record
+         * is persisted as per it's in memory format.
+         */
+        size_t (*co_convert) (void *data, char *buffer, gf_boolean_t encode);
+
+        /* release routines */
+        void (*co_free) (void *data);
+
+        /* type of the field */
+        changelog_optional_rec_type_t co_type;
+
+        /**
+         * sizeof of the 'valid' field in the union. This field is not used if
+         * @co_convert is specified.
+         */
+        size_t co_len;
+
+        union {
+                glusterfs_fop_t co_fop;
+                struct changelog_entry_fields co_entry;
+        };
+} changelog_opt_t;
+
+#define CHANGELOG_OPT_RECORD_LEN  sizeof (changelog_opt_t)
+
+/**
+ * helpers routines
+ */
+
+void
+changelog_thread_cleanup (xlator_t *this, pthread_t thr_id);
+inline void *
+changelog_get_usable_buffer (changelog_local_t *local);
+inline void
+changelog_set_usable_record_and_length (changelog_local_t *local,
+                                        size_t len, int xr);
+void
+changelog_local_cleanup (xlator_t *xl, changelog_local_t *local);
+changelog_local_t *
+changelog_local_init (xlator_t *this, inode_t *inode, uuid_t gfid,
+                      int xtra_records, gf_boolean_t update_flag);
+int
+changelog_start_next_change (xlator_t *this,
+                             changelog_priv_t *priv,
+                             unsigned long ts, gf_boolean_t finale);
+int
+changelog_open (xlator_t *this, changelog_priv_t *priv);
+int
+changelog_fill_rollover_data (changelog_log_data_t *cld, gf_boolean_t is_last);
+int
+changelog_inject_single_event (xlator_t *this,
+                               changelog_priv_t *priv,
+                               changelog_log_data_t *cld);
+inline size_t
+changelog_entry_length ();
+inline int
+changelog_write (int fd, char *buffer, size_t len);
+int
+changelog_write_change (changelog_priv_t *priv, char *buffer, size_t len);
+inline int
+changelog_handle_change (xlator_t *this,
+                         changelog_priv_t *priv, changelog_log_data_t *cld);
+inline void
+changelog_update (xlator_t *this, changelog_priv_t *priv,
+                  changelog_local_t *local, changelog_log_type type);
+void *
+changelog_rollover (void *data);
+void *
+changelog_fsync_thread (void *data);
+int
+changelog_forget (xlator_t *this, inode_t *inode);
+
+/* macros */
+
+#define CHANGELOG_STACK_UNWIND(fop, frame, params ...) do {             \
+                changelog_local_t *__local = NULL;                      \
+                xlator_t          *__xl    = NULL;                      \
+                if (frame) {                                            \
+                        __local      = frame->local;                    \
+                        __xl         = frame->this;                     \
+                        frame->local = NULL;                            \
+                }                                                       \
+                STACK_UNWIND_STRICT (fop, frame, params);               \
+                changelog_local_cleanup (__xl, __local);                \
+                if (__local && __local->prev_entry)                     \
+                        changelog_local_cleanup (__xl,                  \
+                                                 __local->prev_entry);  \
+        } while (0)
+
+#define CHANGELOG_IOBUF_REF(iobuf) do {         \
+                if (iobuf)                      \
+                        iobuf_ref (iobuf);      \
+        } while (0)
+
+#define CHANGELOG_IOBUF_UNREF(iobuf) do {       \
+                if (iobuf)                      \
+                        iobuf_unref (iobuf);    \
+        } while (0)
+
+#define CHANGELOG_FILL_BUFFER(buffer, off, val, len) do {       \
+                memcpy (buffer + off, val, len);                \
+                off += len;                                     \
+        } while (0)
+
+#define SLICE_VERSION_UPDATE(slice) do {                \
+                int i = 0;                              \
+                for (; i < CHANGELOG_MAX_TYPE; i++) {   \
+                        slice->changelog_version[i]++;  \
+                }                                       \
+        } while (0)
+
+#define CHANGLOG_FILL_FOP_NUMBER(co, fop, converter, xlen) do { \
+                co->co_convert = converter;                     \
+                co->co_free = NULL;                             \
+                co->co_type = CHANGELOG_OPT_REC_FOP;            \
+                co->co_fop = fop;                               \
+                xlen += sizeof (fop);                           \
+        } while (0)
+
+#define CHANGELOG_FILL_ENTRY(co, pargfid, bname,                        \
+                             converter, freefn, xlen, label)            \
+        do {                                                            \
+                co->co_convert = converter;                             \
+                co->co_free = freefn;                                   \
+                co->co_type = CHANGELOG_OPT_REC_ENTRY;                  \
+                uuid_copy (co->co_entry.cef_uuid, pargfid);             \
+                co->co_entry.cef_bname = gf_strdup(bname);              \
+                if (!co->co_entry.cef_bname)                            \
+                        goto label;                                     \
+                xlen += (UUID_CANONICAL_FORM_LEN + strlen (bname));     \
+        } while (0)
+
+#define CHANGELOG_INIT(this, local, inode, gfid, xrec)                  \
+        local = changelog_local_init (this, inode, gfid, xrec, _gf_false)
+
+#define CHANGELOG_INIT_NOCHECK(this, local, inode, gfid, xrec)          \
+        local = changelog_local_init (this, inode, gfid, xrec, _gf_true)
+
+#define CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, label) do { \
+                if (!priv->active)                              \
+                        goto label;                             \
+                /* ignore rebalance process's activity. */      \
+                if (frame->root->pid == GF_CLIENT_PID_DEFRAG)   \
+                        goto label;                             \
+        } while (0)
+
+/* ignore internal fops */
+#define CHANGELOG_IF_INTERNAL_FOP_THEN_GOTO(dict, label) do {           \
+                if (dict && dict_get (dict, GLUSTERFS_INTERNAL_FOP_KEY)) \
+                        goto label;                                     \
+        } while (0)
+
+#define CHANGELOG_COND_GOTO(priv, cond, label) do {                    \
+                if (!priv->active || cond)                             \
+                        goto label;                                    \
+        } while (0)
+
+#endif /* _CHANGELOG_HELPERS_H */
diff --git a/xlators/features/changelog/src/changelog-mem-types.h b/xlators/features/changelog/src/changelog-mem-types.h
new file mode 100644
index 000000000..d72464eab
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-mem-types.h
@@ -0,0 +1,29 @@
+/*
+   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 _CHANGELOG_MEM_TYPES_H
+#define _CHANGELOG_MEM_TYPES_H
+
+#include "mem-types.h"
+
+enum gf_changelog_mem_types {
+        gf_changelog_mt_priv_t                  = gf_common_mt_end + 1,
+        gf_changelog_mt_str_t                   = gf_common_mt_end + 2,
+        gf_changelog_mt_batch_t                 = gf_common_mt_end + 3,
+        gf_changelog_mt_rt_t                    = gf_common_mt_end + 4,
+        gf_changelog_mt_inode_ctx_t             = gf_common_mt_end + 5,
+        gf_changelog_mt_libgfchangelog_t        = gf_common_mt_end + 6,
+        gf_changelog_mt_libgfchangelog_rl_t     = gf_common_mt_end + 7,
+        gf_changelog_mt_libgfchangelog_dirent_t = gf_common_mt_end + 8,
+        gf_changelog_mt_changelog_buffer_t      = gf_common_mt_end + 9,
+        gf_changelog_mt_end
+};
+
+#endif
diff --git a/xlators/features/changelog/src/changelog-misc.h b/xlators/features/changelog/src/changelog-misc.h
new file mode 100644
index 000000000..0712a3771
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-misc.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.
+*/
+
+#ifndef _CHANGELOG_MISC_H
+#define _CHANGELOG_MISC_H
+
+#include "glusterfs.h"
+#include "common-utils.h"
+
+#define CHANGELOG_MAX_TYPE  3
+#define CHANGELOG_FILE_NAME "CHANGELOG"
+
+#define CHANGELOG_VERSION_MAJOR  1
+#define CHANGELOG_VERSION_MINOR  0
+
+#define CHANGELOG_UNIX_SOCK  DEFAULT_VAR_RUN_DIRECTORY"/changelog-%s.sock"
+
+/**
+ * header starts with the version and the format of the changelog.
+ * 'version' not much of a use now.
+ */
+#define CHANGELOG_HEADER                                                \
+        "GlusterFS Changelog | version: v%d.%d | encoding : %d\n"
+
+#define CHANGELOG_MAKE_SOCKET_PATH(brick_path, sockpath, len) do {      \
+                char md5_sum[MD5_DIGEST_LENGTH*2+1] = {0,};             \
+                md5_wrapper((unsigned char *) brick_path,               \
+                            strlen(brick_path),                         \
+                            md5_sum);                                   \
+                (void) snprintf (sockpath, len,                         \
+                                 CHANGELOG_UNIX_SOCK, md5_sum);         \
+        } while (0)
+
+/**
+ * ... used by libgfchangelog.
+ */
+#define CHANGELOG_GET_ENCODING(fd, buffer, len, enc, enc_len) do {      \
+                FILE *fp;                                               \
+                int fd_dup, maj, min;                                   \
+                                                                        \
+                enc = -1;                                               \
+                fd_dup = dup (fd);                                      \
+                                                                        \
+                if (fd_dup != -1) {                                     \
+                        fp = fdopen (fd_dup, "r");                      \
+                        if (fp) {                                       \
+                                if (fgets (buffer, len, fp)) {          \
+                                        elen = strlen (buffer);         \
+                                        sscanf (buffer,                 \
+                                                CHANGELOG_HEADER,       \
+                                                &maj, &min, &enc);      \
+                                }                                       \
+                                fclose (fp);                            \
+                        } else {                                        \
+                                close (fd_dup);                         \
+                        }                                               \
+                }                                                       \
+        } while (0)
+
+/**
+ * everything after 'CHANGELOG_TYPE_ENTRY' are internal types
+ * (ie. none of the fops trigger this type of event), hence
+ * CHANGELOG_MAX_TYPE = 3
+ */
+typedef enum {
+        CHANGELOG_TYPE_DATA = 0,
+        CHANGELOG_TYPE_METADATA,
+        CHANGELOG_TYPE_ENTRY,
+        CHANGELOG_TYPE_ROLLOVER,
+        CHANGELOG_TYPE_FSYNC,
+} changelog_log_type;
+
+/* operation modes - RT for now */
+typedef enum {
+        CHANGELOG_MODE_RT = 0,
+} changelog_mode_t;
+
+/* encoder types */
+
+typedef enum {
+        CHANGELOG_ENCODE_MIN = 0,
+        CHANGELOG_ENCODE_BINARY,
+        CHANGELOG_ENCODE_ASCII,
+        CHANGELOG_ENCODE_MAX,
+} changelog_encoder_t;
+
+#define CHANGELOG_VALID_ENCODING(enc)                                   \
+        (enc > CHANGELOG_ENCODE_MIN && enc < CHANGELOG_ENCODE_MAX)
+
+#define CHANGELOG_TYPE_IS_ENTRY(type)  (type == CHANGELOG_TYPE_ENTRY)
+#define CHANGELOG_TYPE_IS_ROLLOVER(type)  (type == CHANGELOG_TYPE_ROLLOVER)
+#define CHANGELOG_TYPE_IS_FSYNC(type)  (type == CHANGELOG_TYPE_FSYNC)
+
+#endif /* _CHANGELOG_MISC_H */
diff --git a/xlators/features/changelog/src/changelog-notifier.c b/xlators/features/changelog/src/changelog-notifier.c
new file mode 100644
index 000000000..1f8b31253
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-notifier.c
@@ -0,0 +1,314 @@
+/*
+   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 "changelog-notifier.h"
+
+#include <pthread.h>
+
+inline static void
+changelog_notify_clear_fd (changelog_notify_t *cn, int i)
+{
+        cn->client_fd[i] = -1;
+}
+
+inline static void
+changelog_notify_save_fd (changelog_notify_t *cn, int i, int fd)
+{
+        cn->client_fd[i] = fd;
+}
+
+static int
+changelog_notify_insert_fd (xlator_t *this, changelog_notify_t *cn, int fd)
+{
+        int i   = 0;
+        int ret = 0;
+
+        for (; i < CHANGELOG_MAX_CLIENTS; i++) {
+                if (cn->client_fd[i] == -1)
+                        break;
+        }
+
+        if (i == CHANGELOG_MAX_CLIENTS) {
+                /**
+                 * this case should not be hit as listen() would limit
+                 * the number of completely established connections.
+                 */
+                gf_log (this->name, GF_LOG_WARNING,
+                        "hit max client limit (%d)", CHANGELOG_MAX_CLIENTS);
+                ret = -1;
+        }
+        else
+                changelog_notify_save_fd (cn, i, fd);
+
+        return ret;
+}
+
+static void
+changelog_notify_fill_rset (changelog_notify_t *cn, fd_set *rset, int *maxfd)
+{
+        int i = 0;
+
+        FD_ZERO (rset);
+
+        FD_SET (cn->socket_fd, rset);
+        *maxfd = cn->socket_fd;
+
+        FD_SET (cn->rfd, rset);
+        *maxfd = max (*maxfd, cn->rfd);
+
+        for (; i < CHANGELOG_MAX_CLIENTS; i++) {
+                if (cn->client_fd[i] != -1) {
+                        FD_SET (cn->client_fd[i], rset);
+                        *maxfd = max (*maxfd, cn->client_fd[i]);
+                }
+        }
+
+        *maxfd = *maxfd + 1;
+}
+
+static int
+changelog_notify_client (changelog_notify_t *cn, char *path, ssize_t len)
+{
+        int i = 0;
+        int ret = 0;
+
+        for (; i < CHANGELOG_MAX_CLIENTS; i++) {
+                if (cn->client_fd[i] == -1)
+                        continue;
+
+                if (changelog_write (cn->client_fd[i],
+                                     path, len)) {
+                        ret = -1;
+
+                        close (cn->client_fd[i]);
+                        changelog_notify_clear_fd (cn, i);
+                }
+        }
+
+        return ret;
+}
+
+static void
+changelog_notifier_init (changelog_notify_t *cn)
+{
+        int i = 0;
+
+        cn->socket_fd = -1;
+
+        for (; i < CHANGELOG_MAX_CLIENTS; i++) {
+                changelog_notify_clear_fd (cn, i);
+        }
+}
+
+static void
+changelog_close_client_conn (changelog_notify_t *cn)
+{
+        int i = 0;
+
+        for (; i < CHANGELOG_MAX_CLIENTS; i++) {
+                if (cn->client_fd[i] == -1)
+                        continue;
+
+                close (cn->client_fd[i]);
+                changelog_notify_clear_fd (cn, i);
+        }
+}
+
+static void
+changelog_notifier_cleanup (void *arg)
+{
+        changelog_notify_t *cn = NULL;
+
+        cn = (changelog_notify_t *) arg;
+
+        changelog_close_client_conn (cn);
+
+        if (cn->socket_fd != -1)
+                close (cn->socket_fd);
+
+        if (cn->rfd)
+                close (cn->rfd);
+
+        if (unlink (cn->sockpath))
+                gf_log ("", GF_LOG_WARNING,
+                        "could not unlink changelog socket file"
+                        " %s (reason: %s", cn->sockpath, strerror (errno));
+}
+
+void *
+changelog_notifier (void *data)
+{
+        int                 i         = 0;
+        int                 fd        = 0;
+        int                 max_fd    = 0;
+        int                 len       = 0;
+        ssize_t             readlen   = 0;
+        xlator_t           *this      = NULL;
+        changelog_priv_t   *priv      = NULL;
+        changelog_notify_t *cn        = NULL;
+        struct sockaddr_un  local     = {0,};
+        char path[PATH_MAX]           = {0,};
+        char abspath[PATH_MAX]        = {0,};
+
+        char buffer;
+        fd_set rset;
+
+        priv = (changelog_priv_t *) data;
+
+        cn = &priv->cn;
+        this = cn->this;
+
+        pthread_cleanup_push (changelog_notifier_cleanup, cn);
+
+        changelog_notifier_init (cn);
+
+        cn->socket_fd = socket (AF_UNIX, SOCK_STREAM, 0);
+        if (cn->socket_fd < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "changelog socket error (reason: %s)",
+                        strerror (errno));
+                goto out;
+        }
+
+        CHANGELOG_MAKE_SOCKET_PATH (priv->changelog_brick,
+                                    cn->sockpath, PATH_MAX);
+        if (unlink (cn->sockpath) < 0) {
+                if (errno != ENOENT) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Could not unlink changelog socket file (%s)"
+                                " (reason: %s)",
+                                CHANGELOG_UNIX_SOCK, strerror (errno));
+                        goto cleanup;
+                }
+        }
+
+        local.sun_family = AF_UNIX;
+        strcpy (local.sun_path, cn->sockpath);
+
+        len = strlen (local.sun_path) + sizeof (local.sun_family);
+
+        /* bind to the unix domain socket */
+        if (bind (cn->socket_fd, (struct sockaddr *) &local, len) < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Could not bind to changelog socket (reason: %s)",
+                        strerror (errno));
+                goto cleanup;
+        }
+
+        /* listen for incoming connections */
+        if (listen (cn->socket_fd, CHANGELOG_MAX_CLIENTS) < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "listen() error on changelog socket (reason: %s)",
+                        strerror (errno));
+                goto cleanup;
+        }
+
+        /**
+         * simple select() on all to-be-read file descriptors. This method
+         * though old school works pretty well when you have a handfull of
+         * fd's to be watched (clients).
+         *
+         * Future TODO: move this to epoll based notification facility if
+         *              number of clients increase.
+         */
+        for (;;) {
+                changelog_notify_fill_rset (cn, &rset, &max_fd);
+
+                if (select (max_fd, &rset, NULL, NULL, NULL) < 0) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "select() returned -1 (reason: %s)",
+                                strerror (errno));
+                        sleep (2);
+                        continue;
+                }
+
+                if (FD_ISSET (cn->socket_fd, &rset)) {
+                        fd = accept (cn->socket_fd, NULL, NULL);
+                        if (fd < 0) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "accept error on changelog socket"
+                                        " (reason: %s)", strerror (errno));
+                        } else if (changelog_notify_insert_fd (this, cn, fd)) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "hit max client limit");
+                        }
+                }
+
+                if (FD_ISSET (cn->rfd, &rset)) {
+                        /**
+                         * read changelog filename and notify all connected
+                         * clients.
+                         */
+                        readlen = 0;
+                        while (readlen < PATH_MAX) {
+                                len = read (cn->rfd, &path[readlen++], 1);
+                                if (len == -1) {
+                                        break;
+                                }
+
+                                if (len == 0) {
+                                        gf_log (this->name, GF_LOG_ERROR,
+                                                "rollover thread sent EOF"
+                                                " on pipe - possibly a crash.");
+                                        /* be blunt and close all connections */
+                                        pthread_exit(NULL);
+                                }
+
+                                if (path[readlen - 1] == '\0')
+                                        break;
+                        }
+
+                        /* should we close all client connections here too? */
+                        if (len < 0 || readlen == PATH_MAX) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "Could not get pathname from rollover"
+                                        " thread or pathname too long");
+                                goto process_rest;
+                        }
+
+                        (void) snprintf (abspath, PATH_MAX,
+                                         "%s/%s", priv->changelog_dir, path);
+                        if (changelog_notify_client (cn, abspath,
+                                                     strlen (abspath) + 1))
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "could not notify some clients with new"
+                                        " changelogs");
+                }
+
+        process_rest:
+                for (i = 0; i < CHANGELOG_MAX_CLIENTS; i++) {
+                        if ( (fd = cn->client_fd[i]) == -1 )
+                                continue;
+
+                        if (FD_ISSET (fd, &rset)) {
+                                /**
+                                 * the only data we accept from the client is a
+                                 * disconnect. Anything else is treated as bogus
+                                 * and is silently discarded (also warned!!!).
+                                 */
+                                if ( (readlen = read (fd, &buffer, 1)) <= 0 ) {
+                                        close (fd);
+                                        changelog_notify_clear_fd (cn, i);
+                                } else {
+                                        /* silently discard data and log */
+                                        gf_log (this->name, GF_LOG_WARNING,
+                                                "misbehaving changelog client");
+                                }
+                        }
+                }
+
+        }
+
+ cleanup:;
+        pthread_cleanup_pop (1);
+
+ out:
+        return NULL;
+}
diff --git a/xlators/features/marker/utils/src/procdiggy.h b/xlators/features/changelog/src/changelog-notifier.h
index 56dfc4eb2..55e728356 100644
--- a/xlators/features/marker/utils/src/procdiggy.h
+++ b/xlators/features/changelog/src/changelog-notifier.h
@@ -1,5 +1,5 @@
 /*
-   Copyright (c) 2011-2012 Red Hat, Inc. <http://www.redhat.com>
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
    This file is licensed to you under your choice of the GNU Lesser
@@ -7,14 +7,13 @@
    later), or the GNU General Public License, version 2 (GPLv2), in all
    cases as published by the Free Software Foundation.
 */
-#ifdef __NetBSD__
-#include <sys/syslimits.h>
-#endif /* __NetBSD__ */
 
-#define PROC "/proc"
+#ifndef _CHANGELOG_NOTIFIER_H
+#define _CHANGELOG_NOTIFIER_H
 
-pid_t pidinfo (pid_t pid, char **name);
+#include "changelog-helpers.h"
 
-int prociter (int (*proch) (pid_t pid, pid_t ppid, char *name, void *data),
-              void *data);
+void *
+changelog_notifier (void *data);
 
+#endif
diff --git a/xlators/features/changelog/src/changelog-rt.c b/xlators/features/changelog/src/changelog-rt.c
new file mode 100644
index 000000000..c147f68ca
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-rt.c
@@ -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 _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "xlator.h"
+#include "defaults.h"
+#include "logging.h"
+
+#include "changelog-rt.h"
+#include "changelog-mem-types.h"
+
+int
+changelog_rt_init (xlator_t *this, changelog_dispatcher_t *cd)
+{
+        changelog_rt_t *crt = NULL;
+
+        crt = GF_CALLOC (1, sizeof (*crt),
+                         gf_changelog_mt_rt_t);
+        if (!crt)
+                return -1;
+
+        LOCK_INIT (&crt->lock);
+
+        cd->cd_data = crt;
+        cd->dispatchfn = &changelog_rt_enqueue;
+
+        return 0;
+}
+
+int
+changelog_rt_fini (xlator_t *this, changelog_dispatcher_t *cd)
+{
+        changelog_rt_t *crt = NULL;
+
+        crt = cd->cd_data;
+
+        LOCK_DESTROY (&crt->lock);
+        GF_FREE (crt);
+
+        return 0;
+}
+
+int
+changelog_rt_enqueue (xlator_t *this, changelog_priv_t *priv, void *cbatch,
+                      changelog_log_data_t *cld_0, changelog_log_data_t *cld_1)
+{
+        int             ret = 0;
+        changelog_rt_t *crt = NULL;
+
+        crt = (changelog_rt_t *) cbatch;
+
+        LOCK (&crt->lock);
+        {
+                ret = changelog_handle_change (this, priv, cld_0);
+                if (!ret && cld_1)
+                        ret = changelog_handle_change (this, priv, cld_1);
+        }
+        UNLOCK (&crt->lock);
+
+        return ret;
+}
diff --git a/xlators/features/changelog/src/changelog-rt.h b/xlators/features/changelog/src/changelog-rt.h
new file mode 100644
index 000000000..1fc2bbc5b
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-rt.h
@@ -0,0 +1,33 @@
+/*
+   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 _CHANGELOG_RT_H
+#define _CHANGELOG_RT_H
+
+#include "locking.h"
+#include "timer.h"
+#include "pthread.h"
+
+#include "changelog-helpers.h"
+
+/* unused as of now - may be you would need it later */
+typedef struct changelog_rt {
+        gf_lock_t lock;
+} changelog_rt_t;
+
+int
+changelog_rt_init (xlator_t *this, changelog_dispatcher_t *cd);
+int
+changelog_rt_fini (xlator_t *this, changelog_dispatcher_t *cd);
+int
+changelog_rt_enqueue (xlator_t *this, changelog_priv_t *priv, void *cbatch,
+                      changelog_log_data_t *cld_0, changelog_log_data_t *cld_1);
+
+#endif /* _CHANGELOG_RT_H */
diff --git a/xlators/features/changelog/src/changelog.c b/xlators/features/changelog/src/changelog.c
new file mode 100644
index 000000000..cea0e8c70
--- /dev/null
+++ b/xlators/features/changelog/src/changelog.c
@@ -0,0 +1,1477 @@
+/*
+   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 "xlator.h"
+#include "defaults.h"
+#include "logging.h"
+#include "iobuf.h"
+
+#include "changelog-rt.h"
+
+#include "changelog-encoders.h"
+#include "changelog-mem-types.h"
+
+#include <pthread.h>
+
+#include "changelog-notifier.h"
+
+static struct changelog_bootstrap
+cb_bootstrap[] = {
+        {
+                .mode = CHANGELOG_MODE_RT,
+                .ctor = changelog_rt_init,
+                .dtor = changelog_rt_fini,
+        },
+};
+
+/* Entry operations - TYPE III */
+
+/**
+ * entry operations do not undergo inode version checking.
+ */
+
+/* {{{ */
+
+/* rmdir */
+
+int32_t
+changelog_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)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (rmdir, frame, op_ret, op_errno,
+                                preparent, postparent, xdata);
+        return 0;
+}
+
+int32_t
+changelog_rmdir (call_frame_t *frame, xlator_t *this,
+                 loc_t *loc, int xflags, dict_t *xdata)
+{
+        size_t            xtra_len = 0;
+        changelog_priv_t *priv     = NULL;
+        changelog_opt_t  *co       = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        CHANGELOG_INIT_NOCHECK (this, frame->local,
+                                NULL, loc->inode->gfid, 2);
+
+        co = changelog_get_usable_buffer (frame->local);
+        if (!co)
+                goto wind;
+
+        CHANGLOG_FILL_FOP_NUMBER (co, frame->root->op, fop_fn, xtra_len);
+
+        co++;
+        CHANGELOG_FILL_ENTRY (co, loc->pargfid, loc->name,
+                              entry_fn, entry_free_fn, xtra_len, wind);
+
+        changelog_set_usable_record_and_length (frame->local, xtra_len, 2);
+
+ wind:
+        STACK_WIND (frame, changelog_rmdir_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->rmdir,
+                    loc, xflags, xdata);
+        return 0;
+}
+
+/* unlink */
+
+int32_t
+changelog_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)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (unlink, frame, op_ret, op_errno,
+                                preparent, postparent, xdata);
+        return 0;
+}
+
+int32_t
+changelog_unlink (call_frame_t *frame, xlator_t *this,
+                  loc_t *loc, int xflags, dict_t *xdata)
+{
+        size_t            xtra_len = 0;
+        changelog_priv_t *priv     = NULL;
+        changelog_opt_t  *co       = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+        CHANGELOG_IF_INTERNAL_FOP_THEN_GOTO (xdata, wind);
+
+        CHANGELOG_INIT_NOCHECK (this, frame->local, NULL, loc->inode->gfid, 2);
+
+        co = changelog_get_usable_buffer (frame->local);
+        if (!co)
+                goto wind;
+
+        CHANGLOG_FILL_FOP_NUMBER (co, frame->root->op, fop_fn, xtra_len);
+
+        co++;
+        CHANGELOG_FILL_ENTRY (co, loc->pargfid, loc->name,
+                              entry_fn, entry_free_fn, xtra_len, wind);
+
+        changelog_set_usable_record_and_length (frame->local, xtra_len, 2);
+
+ wind:
+        STACK_WIND (frame, changelog_unlink_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->unlink,
+                    loc, xflags, xdata);
+        return 0;
+}
+
+/* rename */
+
+int32_t
+changelog_rename_cbk (call_frame_t *frame,
+                      void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno,
+                      struct iatt *buf, struct iatt *preoldparent,
+                      struct iatt *postoldparent, struct iatt *prenewparent,
+                      struct iatt *postnewparent, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (rename, frame, op_ret, op_errno,
+                                buf, preoldparent, postoldparent,
+                                prenewparent, postnewparent, xdata);
+        return 0;
+}
+
+
+int32_t
+changelog_rename (call_frame_t *frame, xlator_t *this,
+                  loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+{
+        size_t            xtra_len  = 0;
+        changelog_priv_t *priv      = NULL;
+        changelog_opt_t  *co        = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        /* 3 == fop + oldloc + newloc */
+        CHANGELOG_INIT_NOCHECK (this, frame->local,
+                                NULL, oldloc->inode->gfid, 3);
+
+        co = changelog_get_usable_buffer (frame->local);
+        if (!co)
+                goto wind;
+
+        CHANGLOG_FILL_FOP_NUMBER (co, frame->root->op, fop_fn, xtra_len);
+
+        co++;
+        CHANGELOG_FILL_ENTRY (co, oldloc->pargfid, oldloc->name,
+                              entry_fn, entry_free_fn, xtra_len, wind);
+
+        co++;
+        CHANGELOG_FILL_ENTRY (co, newloc->pargfid, newloc->name,
+                              entry_fn, entry_free_fn, xtra_len, wind);
+
+        changelog_set_usable_record_and_length (frame->local, xtra_len, 3);
+
+ wind:
+        STACK_WIND (frame, changelog_rename_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->rename,
+                    oldloc, newloc, xdata);
+        return 0;
+}
+
+/* link */
+
+int32_t
+changelog_link_cbk (call_frame_t *frame,
+                    void *cookie, xlator_t *this, int32_t op_ret,
+                    int32_t op_errno, inode_t *inode,
+                    struct iatt *buf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (link, frame, op_ret, op_errno,
+                                inode, buf, preparent, postparent, xdata);
+        return 0;
+}
+
+int32_t
+changelog_link (call_frame_t *frame,
+                xlator_t *this, loc_t *oldloc,
+                loc_t *newloc, dict_t *xdata)
+{
+        size_t            xtra_len = 0;
+        changelog_priv_t *priv     = NULL;
+        changelog_opt_t  *co       = NULL;
+
+        priv = this->private;
+
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+        CHANGELOG_IF_INTERNAL_FOP_THEN_GOTO (xdata, wind);
+
+        CHANGELOG_INIT_NOCHECK (this, frame->local, NULL, oldloc->gfid, 2);
+
+        co = changelog_get_usable_buffer (frame->local);
+        if (!co)
+                goto wind;
+
+        CHANGLOG_FILL_FOP_NUMBER (co, frame->root->op, fop_fn, xtra_len);
+
+        co++;
+        CHANGELOG_FILL_ENTRY (co, newloc->pargfid, newloc->name,
+                              entry_fn, entry_free_fn, xtra_len, wind);
+
+        changelog_set_usable_record_and_length (frame->local, xtra_len, 2);
+
+ wind:
+        STACK_WIND (frame, changelog_link_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->link,
+                    oldloc, newloc, xdata);
+        return 0;
+}
+
+/* mkdir */
+
+int32_t
+changelog_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)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (mkdir, frame, op_ret, op_errno,
+                                inode, buf, preparent, postparent, xdata);
+        return 0;
+}
+
+int32_t
+changelog_mkdir (call_frame_t *frame, xlator_t *this,
+                 loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata)
+{
+        int               ret      = -1;
+        uuid_t            gfid     = {0,};
+        void             *uuid_req = NULL;
+        size_t            xtra_len = 0;
+        changelog_priv_t *priv     = NULL;
+        changelog_opt_t  *co       = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        ret = dict_get_ptr (xdata, "gfid-req", &uuid_req);
+        if (ret) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "failed to get gfid from dict");
+                goto wind;
+        }
+        uuid_copy (gfid, uuid_req);
+
+        CHANGELOG_INIT_NOCHECK (this, frame->local, NULL, gfid, 2);
+
+        co = changelog_get_usable_buffer (frame->local);
+        if (!co)
+                goto wind;
+
+        CHANGLOG_FILL_FOP_NUMBER (co, frame->root->op, fop_fn, xtra_len);
+
+        co++;
+        CHANGELOG_FILL_ENTRY (co, loc->pargfid, loc->name,
+                              entry_fn, entry_free_fn, xtra_len, wind);
+
+        changelog_set_usable_record_and_length (frame->local, xtra_len, 2);
+
+ wind:
+        STACK_WIND (frame, changelog_mkdir_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->mkdir,
+                    loc, mode, umask, xdata);
+        return 0;
+}
+
+/* symlink */
+
+int32_t
+changelog_symlink_cbk (call_frame_t *frame,
+                       void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno,
+                       inode_t *inode, struct iatt *buf, struct iatt *preparent,
+                       struct iatt *postparent, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (symlink, frame, op_ret, op_errno,
+                                inode, buf, preparent, postparent, xdata);
+        return 0;
+}
+
+int32_t
+changelog_symlink (call_frame_t *frame, xlator_t *this,
+                   const char *linkname, loc_t *loc,
+                   mode_t umask, dict_t *xdata)
+{
+        int               ret      = -1;
+        size_t            xtra_len = 0;
+        uuid_t            gfid     = {0,};
+        void             *uuid_req = NULL;
+        changelog_priv_t *priv     = NULL;
+        changelog_opt_t  *co       = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        ret = dict_get_ptr (xdata, "gfid-req", &uuid_req);
+        if (ret) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "failed to get gfid from dict");
+                goto wind;
+        }
+        uuid_copy (gfid, uuid_req);
+
+        CHANGELOG_INIT_NOCHECK (this, frame->local, NULL, gfid, 2);
+
+        co = changelog_get_usable_buffer (frame->local);
+        if (!co)
+                goto wind;
+
+        CHANGLOG_FILL_FOP_NUMBER (co, frame->root->op, fop_fn, xtra_len);
+
+        co++;
+        CHANGELOG_FILL_ENTRY (co, loc->pargfid, loc->name,
+                              entry_fn, entry_free_fn, xtra_len, wind);
+
+        changelog_set_usable_record_and_length (frame->local, xtra_len, 2);
+
+ wind:
+        STACK_WIND (frame, changelog_symlink_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->symlink,
+                    linkname, loc, umask, xdata);
+        return 0;
+}
+
+/* mknod */
+
+int32_t
+changelog_mknod_cbk (call_frame_t *frame,
+                     void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, inode_t *inode,
+                     struct iatt *buf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (mknod, frame, op_ret, op_errno,
+                                inode, buf, preparent, postparent, xdata);
+        return 0;
+}
+
+int32_t
+changelog_mknod (call_frame_t *frame,
+                 xlator_t *this, loc_t *loc,
+                 mode_t mode, dev_t dev, mode_t umask, dict_t *xdata)
+{
+        int               ret      = -1;
+        uuid_t            gfid     = {0,};
+        void             *uuid_req = NULL;
+        size_t            xtra_len = 0;
+        changelog_priv_t *priv     = NULL;
+        changelog_opt_t  *co       = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        ret = dict_get_ptr (xdata, "gfid-req", &uuid_req);
+        if (ret) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "failed to get gfid from dict");
+                goto wind;
+        }
+        uuid_copy (gfid, uuid_req);
+
+        CHANGELOG_INIT_NOCHECK (this, frame->local, NULL, gfid, 2);
+
+        co = changelog_get_usable_buffer (frame->local);
+        if (!co)
+                goto wind;
+
+        CHANGLOG_FILL_FOP_NUMBER (co, frame->root->op, fop_fn, xtra_len);
+
+        co++;
+        CHANGELOG_FILL_ENTRY (co, loc->pargfid, loc->name,
+                              entry_fn, entry_free_fn, xtra_len, wind);
+
+        changelog_set_usable_record_and_length (frame->local, xtra_len, 2);
+
+ wind:
+        STACK_WIND (frame, changelog_mknod_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->mknod,
+                    loc, mode, dev, umask, xdata);
+        return 0;
+}
+
+/* creat */
+
+int32_t
+changelog_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)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (create, frame,
+                                op_ret, op_errno, fd, inode,
+                                buf, preparent, postparent, xdata);
+        return 0;
+}
+
+int32_t
+changelog_create (call_frame_t *frame, xlator_t *this,
+                  loc_t *loc, int32_t flags, mode_t mode,
+                  mode_t umask, fd_t *fd, dict_t *xdata)
+{
+        int               ret      = -1;
+        uuid_t            gfid     = {0,};
+        void             *uuid_req = NULL;
+        changelog_opt_t  *co       = NULL;
+        changelog_priv_t *priv     = NULL;
+        size_t            xtra_len = 0;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        ret = dict_get_ptr (xdata, "gfid-req", &uuid_req);
+        if (ret) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "failed to get gfid from dict");
+                goto wind;
+        }
+        uuid_copy (gfid, uuid_req);
+
+        /* init with two extra records */
+        CHANGELOG_INIT_NOCHECK (this, frame->local, NULL, gfid, 2);
+        if (!frame->local)
+                goto wind;
+
+        co = changelog_get_usable_buffer (frame->local);
+        if (!co)
+                goto wind;
+
+        CHANGLOG_FILL_FOP_NUMBER (co, frame->root->op, fop_fn, xtra_len);
+
+        co++;
+        CHANGELOG_FILL_ENTRY (co, loc->pargfid, loc->name,
+                              entry_fn, entry_free_fn, xtra_len, wind);
+
+        changelog_set_usable_record_and_length (frame->local, xtra_len, 2);
+
+ wind:
+        STACK_WIND (frame, changelog_create_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->create,
+                    loc, flags, mode, umask, fd, xdata);
+        return 0;
+}
+
+/* }}} */
+
+
+/* Metadata modification fops - TYPE II */
+
+/* {{{ */
+
+/* {f}setattr */
+
+int32_t
+changelog_fsetattr_cbk (call_frame_t *frame,
+                        void *cookie, xlator_t *this, int32_t op_ret,
+                        int32_t op_errno, struct iatt *preop_stbuf,
+                        struct iatt *postop_stbuf, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_METADATA);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (fsetattr, frame, op_ret, op_errno,
+                                preop_stbuf, postop_stbuf, xdata);
+
+        return 0;
+
+
+}
+
+int32_t
+changelog_fsetattr (call_frame_t *frame,
+                    xlator_t *this, fd_t *fd,
+                    struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+        changelog_priv_t *priv = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        CHANGELOG_INIT (this, frame->local,
+                        fd->inode, fd->inode->gfid, 0);
+
+ wind:
+        STACK_WIND (frame, changelog_fsetattr_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->fsetattr,
+                    fd, stbuf, valid, xdata);
+        return 0;
+
+
+}
+
+int32_t
+changelog_setattr_cbk (call_frame_t *frame,
+                       void *cookie, xlator_t *this, int32_t op_ret,
+                       int32_t op_errno, struct iatt *preop_stbuf,
+                       struct iatt *postop_stbuf, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_METADATA);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (setattr, frame, op_ret, op_errno,
+                                preop_stbuf, postop_stbuf, xdata);
+
+        return 0;
+}
+
+int32_t
+changelog_setattr (call_frame_t *frame,
+                   xlator_t *this, loc_t *loc,
+                   struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+        changelog_priv_t *priv = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        CHANGELOG_INIT (this, frame->local,
+                        loc->inode, loc->inode->gfid, 0);
+
+ wind:
+        STACK_WIND (frame, changelog_setattr_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->setattr,
+                    loc, stbuf, valid, xdata);
+        return 0;
+}
+
+/* {f}removexattr */
+
+int32_t
+changelog_fremovexattr_cbk (call_frame_t *frame,
+                            void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_METADATA);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (fremovexattr, frame, op_ret, op_errno, xdata);
+
+        return 0;
+}
+
+int32_t
+changelog_fremovexattr (call_frame_t *frame, xlator_t *this,
+                        fd_t *fd, const char *name, dict_t *xdata)
+{
+        changelog_priv_t *priv = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        CHANGELOG_INIT (this, frame->local,
+                        fd->inode, fd->inode->gfid, 0);
+
+ wind:
+        STACK_WIND (frame, changelog_fremovexattr_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->fremovexattr,
+                    fd, name, xdata);
+        return 0;
+}
+
+int32_t
+changelog_removexattr_cbk (call_frame_t *frame,
+                           void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_METADATA);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (removexattr, frame, op_ret, op_errno, xdata);
+
+        return 0;
+}
+
+int32_t
+changelog_removexattr (call_frame_t *frame, xlator_t *this,
+                       loc_t *loc, const char *name, dict_t *xdata)
+{
+        changelog_priv_t *priv = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        CHANGELOG_INIT (this, frame->local,
+                        loc->inode, loc->inode->gfid, 0);
+
+ wind:
+        STACK_WIND (frame, changelog_removexattr_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->removexattr,
+                    loc, name, xdata);
+        return 0;
+}
+
+/* {f}setxattr */
+
+int32_t
+changelog_setxattr_cbk (call_frame_t *frame,
+                        void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_METADATA);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (setxattr, frame, op_ret, op_errno, xdata);
+
+        return 0;
+}
+
+int32_t
+changelog_setxattr (call_frame_t *frame,
+                    xlator_t *this, loc_t *loc,
+                    dict_t *dict, int32_t flags, dict_t *xdata)
+{
+        changelog_priv_t *priv = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        CHANGELOG_INIT (this, frame->local,
+                        loc->inode, loc->inode->gfid, 0);
+
+ wind:
+        STACK_WIND (frame, changelog_setxattr_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->setxattr,
+                    loc, dict, flags, xdata);
+        return 0;
+}
+
+int32_t
+changelog_fsetxattr_cbk (call_frame_t *frame,
+                         void *cookie, xlator_t *this, int32_t op_ret,
+                         int32_t op_errno, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_METADATA);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (fsetxattr, frame, op_ret, op_errno, xdata);
+
+        return 0;
+}
+
+int32_t
+changelog_fsetxattr (call_frame_t *frame,
+                     xlator_t *this, fd_t *fd, dict_t *dict,
+                     int32_t flags, dict_t *xdata)
+{
+        changelog_priv_t *priv = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        CHANGELOG_INIT (this, frame->local,
+                        fd->inode, fd->inode->gfid, 0);
+
+ wind:
+        STACK_WIND (frame, changelog_fsetxattr_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->fsetxattr,
+                    fd, dict, flags, xdata);
+        return 0;
+}
+
+/* }}} */
+
+
+/* Data modification fops - TYPE I */
+
+/* {{{ */
+
+/* {f}truncate() */
+
+int32_t
+changelog_truncate_cbk (call_frame_t *frame,
+                        void *cookie, xlator_t *this, int32_t op_ret,
+                        int32_t op_errno, struct iatt *prebuf,
+                        struct iatt *postbuf, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_DATA);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (truncate, frame,
+                                op_ret, op_errno, prebuf, postbuf, xdata);
+        return 0;
+}
+
+int32_t
+changelog_truncate (call_frame_t *frame,
+                    xlator_t *this, loc_t *loc, off_t offset, dict_t *xdata)
+{
+        changelog_priv_t *priv = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        CHANGELOG_INIT (this, frame->local,
+                        loc->inode, loc->inode->gfid, 0);
+
+ wind:
+        STACK_WIND (frame, changelog_truncate_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->truncate,
+                    loc, offset, xdata);
+        return 0;
+}
+
+int32_t
+changelog_ftruncate_cbk (call_frame_t *frame,
+                         void *cookie, xlator_t *this, int32_t op_ret,
+                         int32_t op_errno, struct iatt *prebuf,
+                         struct iatt *postbuf, dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret < 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_DATA);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (ftruncate, frame,
+                                op_ret, op_errno, prebuf, postbuf, xdata);
+        return 0;
+}
+
+int32_t
+changelog_ftruncate (call_frame_t *frame,
+                     xlator_t *this, fd_t *fd, off_t offset, dict_t *xdata)
+{
+        changelog_priv_t *priv = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        CHANGELOG_INIT (this, frame->local,
+                        fd->inode, fd->inode->gfid, 0);
+
+ wind:
+        STACK_WIND (frame, changelog_ftruncate_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->ftruncate,
+                    fd, offset, xdata);
+        return 0;
+}
+
+/* writev() */
+
+int32_t
+changelog_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf,
+                    dict_t *xdata)
+{
+        changelog_priv_t  *priv  = NULL;
+        changelog_local_t *local = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+
+        CHANGELOG_COND_GOTO (priv, ((op_ret <= 0) || !local), unwind);
+
+        changelog_update (this, priv, local, CHANGELOG_TYPE_DATA);
+
+ unwind:
+        CHANGELOG_STACK_UNWIND (writev, frame,
+                                op_ret, op_errno, prebuf, postbuf, xdata);
+        return 0;
+}
+
+int32_t
+changelog_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)
+{
+        changelog_priv_t *priv = NULL;
+
+        priv = this->private;
+        CHANGELOG_NOT_ACTIVE_THEN_GOTO (frame, priv, wind);
+
+        CHANGELOG_INIT (this, frame->local,
+                        fd->inode, fd->inode->gfid, 0);
+
+ wind:
+        STACK_WIND (frame, changelog_writev_cbk, FIRST_CHILD (this),
+                    FIRST_CHILD (this)->fops->writev, fd, vector,
+                    count, offset, flags, iobref, xdata);
+        return 0;
+}
+
+/* }}} */
+
+/**
+ * The
+ *   - @init ()
+ *   - @fini ()
+ *   - @reconfigure ()
+ *   ... and helper routines
+ */
+
+/**
+ * needed if there are more operation modes in the future.
+ */
+static void
+changelog_assign_opmode (changelog_priv_t *priv, char *mode)
+{
+        if ( strncmp (mode, "realtime", 8) == 0 ) {
+                priv->op_mode = CHANGELOG_MODE_RT;
+        }
+}
+
+static void
+changelog_assign_encoding (changelog_priv_t *priv, char *enc)
+{
+        if ( strncmp (enc, "binary", 6) == 0 ) {
+                priv->encode_mode = CHANGELOG_ENCODE_BINARY;
+        } else if ( strncmp (enc, "ascii", 5) == 0 ) {
+                priv->encode_mode = CHANGELOG_ENCODE_ASCII;
+        }
+}
+
+/* cleanup any helper threads that are running */
+static void
+changelog_cleanup_helper_threads (xlator_t *this, changelog_priv_t *priv)
+{
+        if (priv->cr.rollover_th) {
+                changelog_thread_cleanup (this, priv->cr.rollover_th);
+                priv->cr.rollover_th = 0;
+        }
+
+        if (priv->cf.fsync_th) {
+                changelog_thread_cleanup (this, priv->cf.fsync_th);
+                priv->cf.fsync_th = 0;
+        }
+}
+
+/* spawn helper thread; cleaning up in case of errors */
+static int
+changelog_spawn_helper_threads (xlator_t *this, changelog_priv_t *priv)
+{
+        int ret = 0;
+
+        priv->cr.this = this;
+        ret = gf_thread_create (&priv->cr.rollover_th,
+				NULL, changelog_rollover, priv);
+        if (ret)
+                goto out;
+
+        if (priv->fsync_interval) {
+                priv->cf.this = this;
+                ret = gf_thread_create (&priv->cf.fsync_th,
+					NULL, changelog_fsync_thread, priv);
+        }
+
+        if (ret)
+                changelog_cleanup_helper_threads (this, priv);
+
+ out:
+        return ret;
+}
+
+/* cleanup the notifier thread */
+static int
+changelog_cleanup_notifier (xlator_t *this, changelog_priv_t *priv)
+{
+        int ret = 0;
+
+        if (priv->cn.notify_th) {
+                changelog_thread_cleanup (this, priv->cn.notify_th);
+                priv->cn.notify_th = 0;
+
+                ret = close (priv->wfd);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "error closing writer end of notifier pipe"
+                                " (reason: %s)", strerror (errno));
+        }
+
+        return ret;
+}
+
+/* spawn the notifier thread - nop if already running */
+static int
+changelog_spawn_notifier (xlator_t *this, changelog_priv_t *priv)
+{
+        int ret        = 0;
+        int flags      = 0;
+        int pipe_fd[2] = {0, 0};
+
+        if (priv->cn.notify_th)
+                goto out; /* notifier thread already running */
+
+        ret = pipe (pipe_fd);
+        if (ret == -1) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Cannot create pipe (reason: %s)", strerror (errno));
+                goto out;
+        }
+
+        /* writer is non-blocking */
+        flags = fcntl (pipe_fd[1], F_GETFL);
+        flags |= O_NONBLOCK;
+
+        ret = fcntl (pipe_fd[1], F_SETFL, flags);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to set O_NONBLOCK flag");
+                goto out;
+        }
+
+        priv->wfd = pipe_fd[1];
+
+        priv->cn.this = this;
+        priv->cn.rfd  = pipe_fd[0];
+
+        ret = gf_thread_create (&priv->cn.notify_th,
+				NULL, changelog_notifier, priv);
+
+ out:
+        return ret;
+}
+
+int32_t
+mem_acct_init (xlator_t *this)
+{
+        int     ret = -1;
+
+        if (!this)
+                return ret;
+
+        ret = xlator_mem_acct_init (this, gf_changelog_mt_end + 1);
+
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_WARNING, "Memory accounting"
+                        " init failed");
+                return ret;
+        }
+
+        return ret;
+}
+
+static int
+changelog_init (xlator_t *this, changelog_priv_t *priv)
+{
+        int                  i   = 0;
+        int                  ret = -1;
+        struct timeval       tv  = {0,};
+        changelog_log_data_t cld = {0,};
+
+        ret = gettimeofday (&tv, NULL);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "gettimeofday() failure");
+                goto out;
+        }
+
+        priv->slice.tv_start = tv;
+
+        priv->maps[CHANGELOG_TYPE_DATA]     = "D ";
+        priv->maps[CHANGELOG_TYPE_METADATA] = "M ";
+        priv->maps[CHANGELOG_TYPE_ENTRY]    = "E ";
+
+        for (; i < CHANGELOG_MAX_TYPE; i++) {
+                /* start with version 1 */
+                priv->slice.changelog_version[i] = 1;
+        }
+
+        if (!priv->active)
+                return ret;
+
+        /* spawn the notifier thread */
+        ret = changelog_spawn_notifier (this, priv);
+        if (ret)
+                goto out;
+
+        /**
+         * start with a fresh changelog file every time. this is done
+         * in case there was an encoding change. so... things are kept
+         * simple here.
+         */
+        ret = changelog_fill_rollover_data (&cld, _gf_false);
+        if (ret)
+                goto out;
+
+        LOCK (&priv->lock);
+        {
+                ret = changelog_inject_single_event (this, priv, &cld);
+        }
+        UNLOCK (&priv->lock);
+
+        /* ... and finally spawn the helpers threads */
+        ret = changelog_spawn_helper_threads (this, priv);
+
+ out:
+        return ret;
+}
+
+int
+reconfigure (xlator_t *this, dict_t *options)
+{
+        int                     ret            = 0;
+        char                   *tmp            = NULL;
+        changelog_priv_t       *priv           = NULL;
+        gf_boolean_t            active_earlier = _gf_true;
+        gf_boolean_t            active_now     = _gf_true;
+        changelog_time_slice_t *slice          = NULL;
+        changelog_log_data_t    cld            = {0,};
+
+        priv = this->private;
+        if (!priv)
+                goto out;
+
+        ret = -1;
+        active_earlier = priv->active;
+
+        /* first stop the rollover and the fsync thread */
+        changelog_cleanup_helper_threads (this, priv);
+
+        GF_OPTION_RECONF ("changelog-dir", tmp, options, str, out);
+        if (!tmp) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "\"changelog-dir\" option is not set");
+                goto out;
+        }
+
+        GF_FREE (priv->changelog_dir);
+        priv->changelog_dir = gf_strdup (tmp);
+        if (!priv->changelog_dir)
+                goto out;
+
+        ret = mkdir_p (priv->changelog_dir, 0600, _gf_true);
+        if (ret)
+                goto out;
+
+        GF_OPTION_RECONF ("changelog", active_now, options, bool, out);
+
+        /**
+         * changelog_handle_change() handles changes that could possibly
+         * have been submit changes before changelog deactivation.
+         */
+        if (!active_now)
+                priv->active = _gf_false;
+
+        GF_OPTION_RECONF ("op-mode", tmp, options, str, out);
+        changelog_assign_opmode (priv, tmp);
+
+        tmp = NULL;
+
+        GF_OPTION_RECONF ("encoding", tmp, options, str, out);
+        changelog_assign_encoding (priv, tmp);
+
+        GF_OPTION_RECONF ("rollover-time",
+                          priv->rollover_time, options, int32, out);
+        GF_OPTION_RECONF ("fsync-interval",
+                          priv->fsync_interval, options, int32, out);
+
+        if (active_now || active_earlier) {
+                ret = changelog_fill_rollover_data (&cld, !active_now);
+                if (ret)
+                        goto out;
+
+                slice = &priv->slice;
+
+                LOCK (&priv->lock);
+                {
+                        ret = changelog_inject_single_event (this, priv, &cld);
+                        if (!ret && active_now)
+                                SLICE_VERSION_UPDATE (slice);
+                }
+                UNLOCK (&priv->lock);
+
+                if (ret)
+                        goto out;
+
+                if (active_now) {
+                        ret = changelog_spawn_notifier (this, priv);
+                        if (!ret)
+                                ret = changelog_spawn_helper_threads (this,
+                                                                      priv);
+                } else
+                        ret = changelog_cleanup_notifier (this, priv);
+        }
+
+ out:
+        if (ret) {
+                ret = changelog_cleanup_notifier (this, priv);
+        } else {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "changelog reconfigured");
+                if (active_now)
+                        priv->active = _gf_true;
+        }
+
+        return ret;
+}
+
+int32_t
+init (xlator_t *this)
+{
+        int               ret  = -1;
+        char             *tmp  = NULL;
+        changelog_priv_t *priv = NULL;
+
+        GF_VALIDATE_OR_GOTO ("changelog", this, out);
+
+        if (!this->children || this->children->next) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "translator needs a single subvolume");
+                goto out;
+        }
+
+        if (!this->parents) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "dangling volume. please check volfile");
+                goto out;
+        }
+
+        priv = GF_CALLOC (1, sizeof (*priv), gf_changelog_mt_priv_t);
+        if (!priv)
+                goto out;
+
+        this->local_pool = mem_pool_new (changelog_local_t, 64);
+        if (!this->local_pool) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to create local memory pool");
+                goto out;
+        }
+
+        LOCK_INIT (&priv->lock);
+
+        GF_OPTION_INIT ("changelog-brick", tmp, str, out);
+        if (!tmp) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "\"changelog-brick\" option is not set");
+                goto out;
+        }
+
+        priv->changelog_brick = gf_strdup (tmp);
+        if (!priv->changelog_brick)
+                goto out;
+        tmp = NULL;
+
+        GF_OPTION_INIT ("changelog-dir", tmp, str, out);
+        if (!tmp) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "\"changelog-dir\" option is not set");
+                goto out;
+        }
+
+        priv->changelog_dir = gf_strdup (tmp);
+        if (!priv->changelog_dir)
+                goto out;
+        tmp = NULL;
+
+        /**
+         * create the directory even if change-logging would be inactive
+         * so that consumers can _look_ into it (finding nothing...)
+         */
+        ret = mkdir_p (priv->changelog_dir, 0600, _gf_true);
+        if (ret)
+                goto out;
+
+        GF_OPTION_INIT ("changelog", priv->active, bool, out);
+
+        GF_OPTION_INIT ("op-mode", tmp, str, out);
+        changelog_assign_opmode (priv, tmp);
+
+        tmp = NULL;
+
+        GF_OPTION_INIT ("encoding", tmp, str, out);
+        changelog_assign_encoding (priv, tmp);
+
+        GF_OPTION_INIT ("rollover-time", priv->rollover_time, int32, out);
+
+        GF_OPTION_INIT ("fsync-interval", priv->fsync_interval, int32, out);
+
+        changelog_encode_change(priv);
+
+        GF_ASSERT (cb_bootstrap[priv->op_mode].mode == priv->op_mode);
+        priv->cb = &cb_bootstrap[priv->op_mode];
+
+        /* ... now bootstrap the logger */
+        ret = priv->cb->ctor (this, &priv->cd);
+        if (ret)
+                goto out;
+
+        priv->changelog_fd = -1;
+        ret = changelog_init (this, priv);
+        if (ret)
+                goto out;
+
+        gf_log (this->name, GF_LOG_DEBUG, "changelog translator loaded");
+
+ out:
+        if (ret) {
+                if (this->local_pool)
+                        mem_pool_destroy (this->local_pool);
+                if (priv->cb) {
+                        ret = priv->cb->dtor (this, &priv->cd);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "error in cleanup during init()");
+                }
+                GF_FREE (priv->changelog_brick);
+                GF_FREE (priv->changelog_dir);
+                GF_FREE (priv);
+                this->private = NULL;
+        } else
+                this->private = priv;
+
+        return ret;
+}
+
+void
+fini (xlator_t *this)
+{
+        int               ret  = -1;
+        changelog_priv_t *priv = NULL;
+
+        priv = this->private;
+
+        if (priv) {
+                ret = priv->cb->dtor (this, &priv->cd);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "error in fini");
+                mem_pool_destroy (this->local_pool);
+                GF_FREE (priv->changelog_brick);
+                GF_FREE (priv->changelog_dir);
+                GF_FREE (priv);
+        }
+
+        this->private = NULL;
+
+        return;
+}
+
+struct xlator_fops fops = {
+        .mknod        = changelog_mknod,
+        .mkdir        = changelog_mkdir,
+        .create       = changelog_create,
+        .symlink      = changelog_symlink,
+        .writev       = changelog_writev,
+        .truncate     = changelog_truncate,
+        .ftruncate    = changelog_ftruncate,
+        .link         = changelog_link,
+        .rename       = changelog_rename,
+        .unlink       = changelog_unlink,
+        .rmdir        = changelog_rmdir,
+        .setattr      = changelog_setattr,
+        .fsetattr     = changelog_fsetattr,
+        .setxattr     = changelog_setxattr,
+        .fsetxattr    = changelog_fsetxattr,
+        .removexattr  = changelog_removexattr,
+        .fremovexattr = changelog_fremovexattr,
+};
+
+struct xlator_cbks cbks = {
+        .forget = changelog_forget,
+};
+
+struct volume_options options[] = {
+        {.key = {"changelog"},
+         .type = GF_OPTION_TYPE_BOOL,
+         .default_value = "off",
+         .description = "enable/disable change-logging"
+        },
+        {.key = {"changelog-brick"},
+         .type = GF_OPTION_TYPE_PATH,
+         .description = "brick path to generate unique socket file name."
+                       " should be the export directory of the volume strictly."
+        },
+        {.key = {"changelog-dir"},
+         .type = GF_OPTION_TYPE_PATH,
+         .description = "directory for the changelog files"
+        },
+        {.key = {"op-mode"},
+         .type = GF_OPTION_TYPE_STR,
+         .default_value = "realtime",
+         .value = {"realtime"},
+         .description = "operation mode - futuristic operation modes"
+        },
+        {.key = {"encoding"},
+         .type = GF_OPTION_TYPE_STR,
+         .default_value = "ascii",
+         .value = {"binary", "ascii"},
+         .description = "encoding type for changelogs"
+        },
+        {.key = {"rollover-time"},
+         .default_value = "60",
+         .type = GF_OPTION_TYPE_TIME,
+         .description = "time to switch to a new changelog file (in seconds)"
+        },
+        {.key = {"fsync-interval"},
+         .type = GF_OPTION_TYPE_TIME,
+         .default_value = "0",
+         .description = "do not open CHANGELOG file with O_SYNC mode."
+                        " instead perform fsync() at specified intervals"
+        },
+        {.key = {NULL}
+        },
+};
diff --git a/xlators/features/compress/Makefile.am b/xlators/features/compress/Makefile.am
new file mode 100644
index 000000000..a985f42a8
--- /dev/null
+++ b/xlators/features/compress/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/compress/src/Makefile.am b/xlators/features/compress/src/Makefile.am
new file mode 100644
index 000000000..4a64b52a9
--- /dev/null
+++ b/xlators/features/compress/src/Makefile.am
@@ -0,0 +1,17 @@
+xlator_LTLIBRARIES = cdc.la
+
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+noinst_HEADERS = cdc.h cdc-mem-types.h
+
+cdc_la_LDFLAGS = -module -avoidversion $(LIBZ_LIBS)
+
+cdc_la_SOURCES = cdc.c cdc-helper.c
+cdc_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -D$(GF_HOST_OS) \
+-shared -nostartfiles  $(LIBZ_CFLAGS)
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/compress/src/cdc-helper.c b/xlators/features/compress/src/cdc-helper.c
new file mode 100644
index 000000000..54432ff45
--- /dev/null
+++ b/xlators/features/compress/src/cdc-helper.c
@@ -0,0 +1,547 @@
+/*
+   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 "glusterfs.h"
+#include "logging.h"
+
+#include "cdc.h"
+#include "cdc-mem-types.h"
+
+#ifdef HAVE_LIB_Z
+#include "zlib.h"
+#endif
+
+#ifdef HAVE_LIB_Z
+/* gzip header looks something like this
+ * (RFC 1950)
+ *
+ * +---+---+---+---+---+---+---+---+---+---+
+ * |ID1|ID2|CM |FLG|     MTIME     |XFL|OS |
+ * +---+---+---+---+---+---+---+---+---+---+
+ *
+ * Data is usually sent without this header i.e
+ * Data sent = <compressed-data> + trailer(8)
+ * The trailer contains the checksum.
+ *
+ * gzip_header is added only during debugging.
+ * Refer to the function cdc_dump_iovec_to_disk
+ */
+static const char gzip_header[10] =
+        {
+                '\037', '\213', Z_DEFLATED, 0,
+                0,      0,      0,          0,
+                0,      GF_CDC_OS_ID
+        };
+
+static int32_t
+cdc_next_iovec (xlator_t *this, cdc_info_t *ci)
+{
+        int ret = -1;
+
+        ci->ncount++;
+        /* check for iovec overflow -- should not happen */
+        if (ci->ncount == MAX_IOVEC) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Zlib output buffer overflow"
+                        " ->ncount (%d) | ->MAX_IOVEC (%d)",
+                        ci->ncount, MAX_IOVEC);
+                goto out;
+        }
+
+        ret = 0;
+
+ out:
+        return ret;
+}
+
+static void
+cdc_put_long (unsigned char *string, unsigned long x)
+{
+        string[0] = (unsigned char) (x & 0xff);
+        string[1] = (unsigned char) ((x & 0xff00) >> 8);
+        string[2] = (unsigned char) ((x & 0xff0000) >> 16);
+        string[3] = (unsigned char) ((x & 0xff000000) >> 24);
+}
+
+static unsigned long
+cdc_get_long (unsigned char *buf)
+{
+        return ((unsigned long) buf[0])
+                | (((unsigned long) buf[1]) << 8)
+                | (((unsigned long) buf[2]) << 16)
+                | (((unsigned long) buf[3]) << 24);
+}
+
+static int32_t
+cdc_init_gzip_trailer (xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci)
+{
+        int   ret = -1;
+        char *buf = NULL;
+
+        ret = cdc_next_iovec (this, ci);
+        if (ret)
+                goto out;
+
+        buf = CURR_VEC(ci).iov_base =
+                (char *) GF_CALLOC (1, GF_CDC_VALIDATION_SIZE,
+                                    gf_cdc_mt_gzip_trailer_t);
+
+        if (!CURR_VEC(ci).iov_base)
+                goto out;
+
+        CURR_VEC(ci).iov_len = GF_CDC_VALIDATION_SIZE;
+
+        cdc_put_long ((unsigned char *)&buf[0], ci->crc);
+        cdc_put_long ((unsigned char *)&buf[4], ci->stream.total_in);
+
+        ret = 0;
+
+ out:
+        return ret;
+}
+
+static int32_t
+cdc_alloc_iobuf_and_init_vec (xlator_t *this,
+                              cdc_priv_t *priv, cdc_info_t *ci,
+                              int size)
+{
+        int           ret       = -1;
+        int           alloc_len = 0;
+        struct iobuf *iobuf     = NULL;
+
+        ret = cdc_next_iovec (this, ci);
+        if (ret)
+                goto out;
+
+        alloc_len = size ? size : ci->buffer_size;
+
+        iobuf = iobuf_get2 (this->ctx->iobuf_pool, alloc_len);
+        if (!iobuf)
+                goto out;
+
+        ret = iobref_add (ci->iobref, iobuf);
+        if (ret)
+                goto out;
+
+        /* Initialize this iovec */
+        CURR_VEC(ci).iov_base = iobuf->ptr;
+        CURR_VEC(ci).iov_len  = alloc_len;
+
+        ret = 0;
+
+ out:
+        return ret;
+}
+
+static void
+cdc_init_zlib_output_stream (cdc_priv_t *priv, cdc_info_t *ci, int size)
+{
+        ci->stream.next_out  = (unsigned char *) CURR_VEC(ci).iov_base;
+        ci->stream.avail_out = size ? size : ci->buffer_size;
+}
+
+/* This routine is for testing and debugging only.
+ * Data written = header(10) + <compressed-data> + trailer(8)
+ * So each gzip dump file is at least 18 bytes in size.
+ */
+void
+cdc_dump_iovec_to_disk (xlator_t *this, cdc_info_t *ci, const char *file)
+{
+        int    i      = 0;
+        int    fd     = 0;
+        size_t writen = 0;
+        size_t total_writen = 0;
+
+        fd = open (file, O_WRONLY|O_CREAT|O_TRUNC, 0777 );
+        if (fd < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Cannot open file: %s", file);
+                return;
+        }
+
+        writen = write (fd, (char *) gzip_header, 10);
+        total_writen += writen;
+        for (i = 0; i < ci->ncount; i++) {
+                writen = write (fd, (char *) ci->vec[i].iov_base, ci->vec[i].iov_len);
+                total_writen += writen;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                        "dump'd %zu bytes to %s", total_writen, GF_CDC_DEBUG_DUMP_FILE );
+
+        close (fd);
+}
+
+static int32_t
+cdc_flush_libz_buffer (cdc_priv_t *priv, xlator_t *this, cdc_info_t *ci,
+                       int (*libz_func)(z_streamp, int),
+                       int flush)
+{
+        int32_t ret = Z_OK;
+        int done = 0;
+        unsigned int deflate_len = 0;
+
+        for (;;) {
+                deflate_len = ci->buffer_size - ci->stream.avail_out;
+
+                if (deflate_len != 0) {
+                        CURR_VEC(ci).iov_len = deflate_len;
+
+                        ret = cdc_alloc_iobuf_and_init_vec (this, priv, ci, 0);
+                        if (ret) {
+                                ret = Z_MEM_ERROR;
+                                break;
+                        }
+
+                        /* Re-position Zlib output buffer */
+                        cdc_init_zlib_output_stream (priv, ci, 0);
+                }
+
+                if (done) {
+                        ci->ncount--;
+                        break;
+                }
+
+                ret = libz_func (&ci->stream, flush);
+
+                if (ret == Z_BUF_ERROR) {
+                        ret = Z_OK;
+                        ci->ncount--;
+                        break;
+                }
+
+                done = (ci->stream.avail_out != 0 || ret == Z_STREAM_END);
+
+                if (ret != Z_OK && ret != Z_STREAM_END)
+                        break;
+        }
+
+        return ret;
+}
+
+static int32_t
+do_cdc_compress (struct iovec *vec, xlator_t *this, cdc_priv_t *priv,
+                 cdc_info_t *ci)
+{
+        int ret = -1;
+
+        /* Initialize defalte */
+        ret = deflateInit2 (&ci->stream, priv->cdc_level, Z_DEFLATED,
+                            priv->window_size, priv->mem_level,
+                            Z_DEFAULT_STRATEGY);
+
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "unable to init Zlib (retval: %d)", ret);
+                goto out;
+        }
+
+        ret = cdc_alloc_iobuf_and_init_vec (this, priv, ci, 0);
+        if (ret)
+                goto out;
+
+        /* setup output buffer */
+        cdc_init_zlib_output_stream (priv, ci, 0);
+
+        /* setup input buffer */
+        ci->stream.next_in  = (unsigned char *) vec->iov_base;
+        ci->stream.avail_in = vec->iov_len;
+
+        ci->crc = crc32 (ci->crc, (const Bytef *) vec->iov_base, vec->iov_len);
+
+        gf_log (this->name, GF_LOG_DEBUG, "crc=%lu len=%d buffer_size=%d",
+                ci->crc, ci->stream.avail_in, ci->buffer_size);
+
+        /* compress !! */
+        while (ci->stream.avail_in != 0) {
+                if (ci->stream.avail_out == 0) {
+
+                        CURR_VEC(ci).iov_len = ci->buffer_size;
+
+                        ret = cdc_alloc_iobuf_and_init_vec (this, priv, ci, 0);
+                        if (ret)
+                                break;
+
+                        /* Re-position Zlib output buffer */
+                        cdc_init_zlib_output_stream (priv, ci, 0);
+                }
+
+                ret = deflate (&ci->stream, Z_NO_FLUSH);
+                if (ret != Z_OK)
+                        break;
+        }
+
+ out:
+        return ret;
+}
+
+int32_t
+cdc_compress (xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci,
+              dict_t **xdata)
+{
+        int ret = -1;
+        int i   = 0;
+
+        ci->iobref = iobref_new ();
+        if (!ci->iobref)
+                goto out;
+
+        if (!*xdata) {
+                *xdata = dict_new ();
+                if (!*xdata) {
+                        gf_log (this->name, GF_LOG_ERROR, "Cannot allocate xdata"
+                                " dict");
+                        goto out;
+                }
+        }
+
+        /* data */
+        for (i = 0; i < ci->count; i++) {
+                ret = do_cdc_compress (&ci->vector[i], this, priv, ci);
+                if (ret != Z_OK)
+                        goto deflate_cleanup_out;
+        }
+
+        /* flush zlib buffer */
+        ret = cdc_flush_libz_buffer (priv, this, ci, deflate, Z_FINISH);
+        if (!(ret == Z_OK || ret == Z_STREAM_END)) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Compression Error: ret (%d)", ret);
+                ret = -1;
+                goto deflate_cleanup_out;
+        }
+
+        /* trailer */
+        ret = cdc_init_gzip_trailer (this, priv, ci);
+        if (ret)
+                goto deflate_cleanup_out;
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Compressed %ld to %ld bytes",
+                ci->stream.total_in, ci->stream.total_out);
+
+        ci->nbytes = ci->stream.total_out + GF_CDC_VALIDATION_SIZE;
+
+        /* set deflated canary value for identification */
+        ret = dict_set_int32 (*xdata, GF_CDC_DEFLATE_CANARY_VAL, 1);
+        if (ret) {
+                /* Send uncompressed data if we can't _tell_ the client
+                 * that deflated data is on it's way. So, we just log
+                 * the faliure and continue as usual.
+                 */
+                 gf_log (this->name, GF_LOG_ERROR,
+                 "Data deflated, but could not set canary"
+                 " value in dict for identification");
+        }
+
+        /* This is to be used in testing */
+        if ( priv->debug ) {
+                cdc_dump_iovec_to_disk (this, ci, GF_CDC_DEBUG_DUMP_FILE );
+        }
+
+ deflate_cleanup_out:
+        (void) deflateEnd(&ci->stream);
+
+ out:
+        return ret;
+}
+
+
+/* deflate content is checked by the presence of a canary
+ * value in the dict as the key
+ */
+static int32_t
+cdc_check_content_for_deflate (dict_t *xdata)
+{
+        return dict_get (xdata, GF_CDC_DEFLATE_CANARY_VAL) ? -1 : 0;
+}
+
+static unsigned long
+cdc_extract_crc (char *trailer)
+{
+        return cdc_get_long ((unsigned char *) &trailer[0]);
+}
+
+static unsigned long
+cdc_extract_size (char *trailer)
+{
+        return cdc_get_long ((unsigned char *) &trailer[4]);
+}
+
+static int32_t
+cdc_validate_inflate (cdc_info_t *ci, unsigned long crc,
+                      unsigned long len)
+{
+        return !((crc == ci->crc)
+                 /* inflated length is hidden inside
+                  * Zlib stream struct */
+                 && (len == ci->stream.total_out));
+}
+
+static int32_t
+do_cdc_decompress (xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci)
+{
+        int            ret          = -1;
+        int            i            = 0;
+        int            len          = 0;
+        char          *inflte       = NULL;
+        char          *trailer      = NULL;
+        struct iovec   vec          = {0,};
+        unsigned long  computed_crc = 0;
+        unsigned long  computed_len = 0;
+
+        ret = inflateInit2 (&ci->stream, priv->window_size);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Zlib: Unable to initialize inflate");
+                goto out;
+        }
+
+        vec = THIS_VEC(ci, 0);
+
+        trailer = (char *) (((char *) vec.iov_base) + vec.iov_len
+                - GF_CDC_VALIDATION_SIZE);
+
+        /* CRC of uncompressed data */
+        computed_crc = cdc_extract_crc (trailer);
+
+        /* size of uncomrpessed data */
+        computed_len = cdc_extract_size (trailer);
+
+        gf_log (this->name, GF_LOG_DEBUG, "crc=%lu len=%lu buffer_size=%d",
+                computed_crc, computed_len, ci->buffer_size);
+
+        inflte = vec.iov_base ;
+        len = vec.iov_len - GF_CDC_VALIDATION_SIZE;
+
+        /* allocate buffer of the original length of the data */
+        ret = cdc_alloc_iobuf_and_init_vec (this, priv, ci, 0);
+        if (ret)
+                goto out;
+
+        /* setup output buffer */
+        cdc_init_zlib_output_stream (priv, ci, 0);
+
+        /* setup input buffer */
+        ci->stream.next_in = (unsigned char *) inflte;
+        ci->stream.avail_in = len;
+
+        while (ci->stream.avail_in != 0) {
+                if (ci->stream.avail_out == 0) {
+                        CURR_VEC(ci).iov_len = ci->buffer_size;
+
+                        ret = cdc_alloc_iobuf_and_init_vec (this, priv, ci, 0);
+                        if (ret)
+                                break;
+
+                        /* Re-position Zlib output buffer */
+                        cdc_init_zlib_output_stream (priv, ci, 0);
+                }
+
+                ret = inflate (&ci->stream, Z_NO_FLUSH);
+                if (ret == Z_STREAM_ERROR)
+                        break;
+        }
+
+        /* flush zlib buffer */
+        ret = cdc_flush_libz_buffer (priv, this, ci, inflate, Z_SYNC_FLUSH);
+        if (!(ret == Z_OK || ret == Z_STREAM_END)) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Decompression Error: ret (%d)", ret);
+                ret = -1;
+                goto out;
+        }
+
+        /* compute CRC of the uncompresses data to check for
+         * correctness */
+
+        for (i = 0; i < ci->ncount; i++) {
+                ci->crc = crc32 (ci->crc,
+                                 (const Bytef *) ci->vec[i].iov_base,
+                                 ci->vec[i].iov_len);
+        }
+
+        /* validate inflated data */
+        ret = cdc_validate_inflate (ci, computed_crc, computed_len);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Checksum or length mismatched in inflated data");
+        }
+
+ out:
+        return ret;
+}
+
+int32_t
+cdc_decompress (xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci,
+                dict_t *xdata)
+{
+        int32_t ret = -1;
+
+        /* check for deflate content */
+        if (!cdc_check_content_for_deflate (xdata)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Content not deflated, passing through ...");
+                goto passthrough_out;
+        }
+
+        ci->iobref = iobref_new ();
+        if (!ci->iobref)
+                goto passthrough_out;
+
+        /* do we need to do this? can we assume that one iovec
+         * will hold per request data everytime?
+         *
+         * server/client protocol seems to deal with a single
+         * iovec even if op_ret > 1M. So, it looks ok to
+         * assume that a single iovec will contain all the
+         * data (This saves us a lot from finding the trailer
+         * and the data since it could have been split-up onto
+         * two adjacent iovec's.
+         *
+         * But, in case this translator is loaded above quick-read
+         * for some reason, then it's entirely possible that we get
+         * multiple iovec's...
+         *
+         * This case (handled below) is not tested. (by loading the
+         * xlator below quick-read)
+         */
+
+        /* @@ I_HOPE_THIS_IS_NEVER_HIT */
+        if (ci->count > 1) {
+                gf_log (this->name, GF_LOG_WARNING, "unable to handle"
+                        " multiple iovecs (%d in number)", ci->count);
+                goto inflate_cleanup_out;
+                /* TODO: coallate all iovecs in one */
+        }
+
+        ret = do_cdc_decompress (this, priv, ci);
+        if (ret)
+                goto inflate_cleanup_out;
+
+        ci->nbytes = ci->stream.total_out;
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Inflated %ld to %ld bytes",
+                ci->stream.total_in, ci->stream.total_out);
+
+ inflate_cleanup_out:
+        (void) inflateEnd (&ci->stream);
+
+ passthrough_out:
+        return ret;
+}
+
+#endif
diff --git a/xlators/features/compress/src/cdc-mem-types.h b/xlators/features/compress/src/cdc-mem-types.h
new file mode 100644
index 000000000..efa008059
--- /dev/null
+++ b/xlators/features/compress/src/cdc-mem-types.h
@@ -0,0 +1,22 @@
+/*
+   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 __CDC_MEM_TYPES_H
+#define __CDC_MEM_TYPES_H
+
+#include "mem-types.h"
+
+enum gf_cdc_mem_types {
+        gf_cdc_mt_priv_t         = gf_common_mt_end + 1,
+        gf_cdc_mt_vec_t          = gf_common_mt_end + 2,
+        gf_cdc_mt_gzip_trailer_t = gf_common_mt_end + 3,
+};
+
+#endif
diff --git a/xlators/features/compress/src/cdc.c b/xlators/features/compress/src/cdc.c
new file mode 100644
index 000000000..eb7d87c56
--- /dev/null
+++ b/xlators/features/compress/src/cdc.c
@@ -0,0 +1,342 @@
+/*
+   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/uio.h>
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "xlator.h"
+#include "defaults.h"
+#include "logging.h"
+
+#include "cdc.h"
+#include "cdc-mem-types.h"
+
+static void
+cdc_cleanup_iobref (cdc_info_t *ci)
+{
+        assert(ci->iobref != NULL);
+        iobref_clear (ci->iobref);
+}
+
+int32_t
+cdc_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno,
+               struct iovec *vector, int32_t count,
+               struct iatt *stbuf, struct iobref *iobref,
+               dict_t *xdata)
+{
+        int         ret   = -1;
+        cdc_priv_t *priv  = NULL;
+        cdc_info_t  ci    = {0,};
+
+        GF_VALIDATE_OR_GOTO ("cdc", this, default_out);
+        GF_VALIDATE_OR_GOTO (this->name, frame, default_out);
+
+        priv = this->private;
+
+        if (op_ret <= 0)
+                goto default_out;
+
+        if ( (priv->min_file_size != 0)
+             && (op_ret < priv->min_file_size) )
+                goto default_out;
+
+        ci.count       = count;
+        ci.ibytes      = op_ret;
+        ci.vector      = vector;
+        ci.buf         = NULL;
+        ci.iobref      = NULL;
+        ci.ncount      = 0;
+        ci.crc         = 0;
+        ci.buffer_size = GF_CDC_DEF_BUFFERSIZE;
+
+/* A readv compresses on the server side and decompresses on the client side
+ */
+        if (priv->op_mode == GF_CDC_MODE_SERVER) {
+                ret = cdc_compress (this, priv, &ci, &xdata);
+        } else if (priv->op_mode == GF_CDC_MODE_CLIENT) {
+                ret = cdc_decompress (this, priv, &ci, xdata);
+        } else {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Invalid operation mode (%d)", priv->op_mode);
+        }
+
+        if (ret)
+                goto default_out;
+
+        STACK_UNWIND_STRICT (readv, frame, ci.nbytes, op_errno,
+                             ci.vec, ci.ncount, stbuf, iobref,
+                             xdata);
+        cdc_cleanup_iobref (&ci);
+        return 0;
+
+ default_out:
+        STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno,
+                             vector, count, stbuf, iobref, xdata);
+        return 0;
+}
+
+int32_t
+cdc_readv (call_frame_t *frame, xlator_t *this,
+           fd_t *fd, size_t size, off_t offset, uint32_t flags,
+           dict_t *xdata)
+{
+        fop_readv_cbk_t cbk = NULL;
+
+#ifdef HAVE_LIB_Z
+        cbk = cdc_readv_cbk;
+#else
+        cbk = default_readv_cbk;
+#endif
+        STACK_WIND (frame, cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->readv,
+                    fd, size, offset, flags, xdata);
+        return 0;
+}
+
+int32_t
+cdc_writev_cbk (call_frame_t *frame,
+                void *cookie,
+                xlator_t *this,
+                int32_t op_ret,
+                int32_t op_errno,
+                struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+
+	STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+	return 0;
+}
+
+int32_t
+cdc_writev (call_frame_t *frame,
+            xlator_t *this,
+            fd_t *fd,
+            struct iovec *vector,
+            int32_t count,
+            off_t offset,
+            uint32_t flags,
+            struct iobref *iobref, dict_t *xdata)
+{
+	int	     ret   = -1;
+	cdc_priv_t  *priv  = NULL;
+	cdc_info_t   ci    = {0,};
+	size_t       isize = 0;
+
+	GF_VALIDATE_OR_GOTO ("cdc", this, default_out);
+	GF_VALIDATE_OR_GOTO (this->name, frame, default_out);
+
+	priv = this->private;
+
+	isize =  iov_length(vector, count);
+
+	if (isize <= 0)
+                goto default_out;
+
+        if ( (priv->min_file_size != 0)
+             && (isize < priv->min_file_size) )
+                goto default_out;
+
+        ci.count       = count;
+        ci.ibytes      = isize;
+        ci.vector      = vector;
+        ci.buf         = NULL;
+        ci.iobref      = NULL;
+        ci.ncount      = 0;
+        ci.crc         = 0;
+        ci.buffer_size = GF_CDC_DEF_BUFFERSIZE;
+
+/* A writev compresses on the client side and decompresses on the server side
+ */
+	if (priv->op_mode == GF_CDC_MODE_CLIENT) {
+		ret = cdc_compress (this, priv, &ci, &xdata);
+	} else if (priv->op_mode == GF_CDC_MODE_SERVER) {
+		ret = cdc_decompress (this, priv, &ci, xdata);
+	} else {
+		gf_log (this->name, GF_LOG_ERROR, "Invalid operation mode (%d) ", priv->op_mode);
+	}
+
+	if (ret)
+		goto default_out;
+
+	STACK_WIND (frame,
+                    cdc_writev_cbk,
+                    FIRST_CHILD (this),
+                    FIRST_CHILD (this)->fops->writev,
+                    fd, ci.vec, ci.ncount, offset, flags,
+                    iobref, xdata);
+
+        cdc_cleanup_iobref (&ci);
+        return 0;
+
+ default_out:
+	STACK_WIND (frame,
+                    cdc_writev_cbk,
+                    FIRST_CHILD (this),
+                    FIRST_CHILD (this)->fops->writev,
+                    fd, vector, count, offset, flags,
+                    iobref, xdata);
+	return 0;
+}
+
+int32_t
+init (xlator_t *this)
+{
+        int         ret      = -1;
+        char       *temp_str = NULL;
+        cdc_priv_t *priv     = NULL;
+
+        GF_VALIDATE_OR_GOTO ("cdc", this, err);
+
+        if (!this->children || this->children->next) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Need subvolume == 1");
+                goto err;
+        }
+
+        if (!this->parents) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Dangling volume. Check volfile");
+        }
+
+        priv = GF_CALLOC (1, sizeof (*priv), gf_cdc_mt_priv_t);
+        if (!priv) {
+                goto err;
+        }
+
+        /* Check if debug mode is turned on */
+        GF_OPTION_INIT ("debug", priv->debug, bool, err);
+        if( priv->debug ) {
+                gf_log (this->name, GF_LOG_DEBUG, "CDC debug option turned on");
+        }
+
+        /* Set Gzip Window Size */
+        GF_OPTION_INIT ("window-size", priv->window_size, int32, err);
+        if ( (priv->window_size > GF_CDC_MAX_WINDOWSIZE)
+             || (priv->window_size < GF_CDC_DEF_WINDOWSIZE) ) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Invalid gzip window size (%d), using default",
+                        priv->window_size);
+                priv->window_size = GF_CDC_DEF_WINDOWSIZE;
+        }
+
+        /* Set Gzip (De)Compression Level */
+        GF_OPTION_INIT ("compression-level", priv->cdc_level, int32, err);
+        if ( ((priv->cdc_level < 1) || (priv->cdc_level > 9))
+             && (priv->cdc_level != GF_CDC_DEF_COMPRESSION) ) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Invalid gzip (de)compression level (%d),"
+                        " using default", priv->cdc_level);
+                priv->cdc_level = GF_CDC_DEF_COMPRESSION;
+        }
+
+        /* Set Gzip Memory Level */
+        GF_OPTION_INIT ("mem-level", priv->mem_level, int32, err);
+        if ( (priv->mem_level < 1) || (priv->mem_level > 9) ) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Invalid gzip memory level, using the default");
+                priv->mem_level = GF_CDC_DEF_MEMLEVEL;
+        }
+
+        /* Set min file size to enable compression */
+        GF_OPTION_INIT ("min-size", priv->min_file_size, int32, err);
+
+        /* Mode of operation - Server/Client */
+        ret = dict_get_str (this->options, "mode", &temp_str);
+        if (ret) {
+                gf_log (this->name, GF_LOG_CRITICAL,
+                        "Operation mode not specified !!");
+                goto err;
+        }
+
+        if (GF_CDC_MODE_IS_CLIENT (temp_str)) {
+                priv->op_mode = GF_CDC_MODE_CLIENT;
+        } else if (GF_CDC_MODE_IS_SERVER (temp_str)) {
+                priv->op_mode = GF_CDC_MODE_SERVER;
+        } else {
+                gf_log (this->name, GF_LOG_CRITICAL,
+                        "Bogus operation mode (%s) specified", temp_str);
+                goto err;
+        }
+
+        this->private = priv;
+        gf_log (this->name, GF_LOG_DEBUG, "CDC xlator loaded in (%s) mode",temp_str);
+        return 0;
+
+ err:
+        if (priv)
+                GF_FREE (priv);
+
+        return -1;
+}
+
+void
+fini (xlator_t *this)
+{
+        cdc_priv_t *priv = this->private;
+
+        if (priv)
+                GF_FREE (priv);
+        this->private = NULL;
+        return;
+}
+
+struct xlator_fops fops = {
+        .readv  = cdc_readv,
+        .writev = cdc_writev,
+};
+
+struct xlator_cbks cbks = {
+};
+
+struct volume_options options[] = {
+        { .key  = {"window-size"},
+          .default_value = "-15",
+          .type = GF_OPTION_TYPE_INT,
+          .description = "Size of the zlib history buffer."
+        },
+        { .key  = {"mem-level"},
+          .default_value = "8",
+          .type = GF_OPTION_TYPE_INT,
+          .description = "Memory allocated for internal compression state.\
+                          1 uses minimum memory but is slow and reduces \
+                          compression ratio; memLevel=9 uses maximum memory \
+                          for optimal speed. The default value is 8."
+        },
+        { .key  = {"compression-level"},
+          .default_value = "-1",
+          .type = GF_OPTION_TYPE_INT,
+          .description = "Compression levels \
+                          0 : no compression, 1 : best speed, \
+                          9 : best compression, -1 : default compression "
+        },
+        { .key  = {"min-size"},
+          .default_value = "0",
+          .type = GF_OPTION_TYPE_INT,
+          .description = "Data is compressed only when its size exceeds this."
+        },
+        { .key  = {"mode"},
+          .value = {"server", "client"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "Set on the basis of where the xlator is loaded."
+        },
+        { .key = {"debug"},
+          .default_value = "false",
+          .type = GF_OPTION_TYPE_BOOL,
+          .description = "This is used in testing. Will dump compressed data \
+                          to disk as a gzip file."
+        },
+        { .key  = {NULL}
+        },
+};
diff --git a/xlators/features/compress/src/cdc.h b/xlators/features/compress/src/cdc.h
new file mode 100644
index 000000000..71f4d2317
--- /dev/null
+++ b/xlators/features/compress/src/cdc.h
@@ -0,0 +1,107 @@
+/*
+   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 __CDC_H
+#define __CDC_H
+
+#ifdef HAVE_LIB_Z
+#include "zlib.h"
+#endif
+
+#include "xlator.h"
+
+#ifndef MAX_IOVEC
+#define MAX_IOVEC 16
+#endif
+
+typedef struct cdc_priv {
+        int window_size;
+        int mem_level;
+        int cdc_level;
+        int min_file_size;
+        int op_mode;
+        gf_boolean_t debug;
+        gf_lock_t lock;
+} cdc_priv_t;
+
+typedef struct cdc_info {
+        /* input bits */
+        int            count;
+        int32_t        ibytes;
+        struct iovec  *vector;
+        struct iatt   *buf;
+
+        /* output bits */
+        int            ncount;
+        int            nbytes;
+        int            buffer_size;
+        struct iovec   vec[MAX_IOVEC];
+        struct iobref *iobref;
+
+        /* zlib bits */
+#ifdef HAVE_LIB_Z
+        z_stream      stream;
+#endif
+        unsigned long crc;
+} cdc_info_t;
+
+#define NVEC(ci) (ci->ncount - 1)
+#define CURR_VEC(ci) ci->vec[ci->ncount - 1]
+#define THIS_VEC(ci, i) ci->vector[i]
+
+/* Gzip defaults */
+#define GF_CDC_DEF_WINDOWSIZE  -15 /* default value */
+#define GF_CDC_MAX_WINDOWSIZE  -8  /* max value     */
+
+#ifdef HAVE_LIB_Z
+#define GF_CDC_DEF_COMPRESSION Z_DEFAULT_COMPRESSION
+#else
+#define GF_CDC_DEF_COMPRESSION -1
+#endif
+
+#define GF_CDC_DEF_MEMLEVEL    8
+#define GF_CDC_DEF_BUFFERSIZE  262144 // 256K - default compression buffer size
+
+/* Operation mode
+ * If xlator is loaded on client, readv decompresses and writev compresses
+ * If xlator is loaded on server, readv compresses and writev decompresses
+ */
+#define GF_CDC_MODE_CLIENT   0
+#define GF_CDC_MODE_SERVER   1
+
+/* min size of data to do cmpression
+ * 0 == compress even 1byte
+ */
+#define GF_CDC_MIN_CHUNK_SIZE 0
+
+#define GF_CDC_VALIDATION_SIZE 8
+
+#define GF_CDC_OS_ID 0xFF
+#define GF_CDC_DEFLATE_CANARY_VAL "deflate"
+#define GF_CDC_DEBUG_DUMP_FILE "/tmp/cdcdump.gz"
+
+#define GF_CDC_MODE_IS_CLIENT(m) \
+        (strcmp (m, "client") == 0)
+
+#define GF_CDC_MODE_IS_SERVER(m) \
+        (strcmp (m, "server") == 0)
+
+int32_t
+cdc_compress (xlator_t *this,
+              cdc_priv_t *priv,
+              cdc_info_t *ci,
+              dict_t **xdata);
+int32_t
+cdc_decompress (xlator_t *this,
+                cdc_priv_t *priv,
+                cdc_info_t *ci,
+                dict_t *xdata);
+
+#endif
diff --git a/xlators/features/gfid-access/Makefile.am b/xlators/features/gfid-access/Makefile.am
new file mode 100644
index 000000000..af437a64d
--- /dev/null
+++ b/xlators/features/gfid-access/Makefile.am
@@ -0,0 +1 @@
+SUBDIRS = src
diff --git a/xlators/features/gfid-access/src/Makefile.am b/xlators/features/gfid-access/src/Makefile.am
new file mode 100644
index 000000000..db53affaa
--- /dev/null
+++ b/xlators/features/gfid-access/src/Makefile.am
@@ -0,0 +1,15 @@
+xlator_LTLIBRARIES = gfid-access.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+gfid_access_la_LDFLAGS = -module -avoid-version
+
+gfid_access_la_SOURCES = gfid-access.c
+gfid_access_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = gfid-access.h gfid-access-mem-types.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/gfid-access/src/gfid-access-mem-types.h b/xlators/features/gfid-access/src/gfid-access-mem-types.h
new file mode 100644
index 000000000..168d67b43
--- /dev/null
+++ b/xlators/features/gfid-access/src/gfid-access-mem-types.h
@@ -0,0 +1,23 @@
+/*
+   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 _GFID_ACCESS_MEM_TYPES_H
+#define _GFID_ACCESS_MEM_TYPES_H
+
+#include "mem-types.h"
+
+enum gf_changelog_mem_types {
+        gf_gfid_access_mt_priv_t = gf_common_mt_end + 1,
+        gf_gfid_access_mt_gfid_t,
+        gf_gfid_access_mt_end
+};
+
+#endif
+
diff --git a/xlators/features/gfid-access/src/gfid-access.c b/xlators/features/gfid-access/src/gfid-access.c
new file mode 100644
index 000000000..da0ba7e50
--- /dev/null
+++ b/xlators/features/gfid-access/src/gfid-access.c
@@ -0,0 +1,1172 @@
+/*
+   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 "gfid-access.h"
+#include "inode.h"
+#include "byte-order.h"
+
+
+
+void
+ga_newfile_args_free (ga_newfile_args_t *args)
+{
+        if (!args)
+                goto out;
+
+        GF_FREE (args->bname);
+
+        if (S_ISLNK (args->st_mode) && args->args.symlink.linkpath) {
+                GF_FREE (args->args.symlink.linkpath);
+                args->args.symlink.linkpath = NULL;
+        }
+
+        mem_put (args);
+out:
+        return;
+}
+
+
+void
+ga_heal_args_free (ga_heal_args_t *args)
+{
+        if (!args)
+                goto out;
+
+        GF_FREE (args->bname);
+
+        mem_put (args);
+out:
+        return;
+}
+
+
+ga_newfile_args_t *
+ga_newfile_parse_args (xlator_t *this, data_t *data)
+{
+        ga_newfile_args_t *args     = NULL;
+        ga_private_t      *priv     = NULL;
+        int                len      = 0;
+        int                blob_len = 0;
+        int                min_len  = 0;
+        void              *blob     = NULL;
+
+        priv = this->private;
+
+        blob = data->data;
+        blob_len = data->len;
+
+        min_len = sizeof (args->uid) + sizeof (args->gid) + sizeof (args->gfid)
+                + sizeof (args->st_mode) + 2 + 2;
+        if (blob_len < min_len) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Invalid length: Total length is less "
+                        "than minimum length.");
+                goto err;
+        }
+
+        args = mem_get0 (priv->newfile_args_pool);
+        if (args == NULL)
+                goto err;
+
+        args->uid = ntoh32 (*(uint32_t *)blob);
+        blob += sizeof (uint32_t);
+        blob_len -= sizeof (uint32_t);
+
+        args->gid = ntoh32 (*(uint32_t *)blob);
+        blob += sizeof (uint32_t);
+        blob_len -= sizeof (uint32_t);
+
+        memcpy (args->gfid, blob, sizeof (args->gfid));
+        blob += sizeof (args->gfid);
+        blob_len -= sizeof (args->gfid);
+
+        args->st_mode = ntoh32 (*(uint32_t *)blob);
+        blob += sizeof (uint32_t);
+        blob_len -= sizeof (uint32_t);
+
+        len = strnlen (blob, blob_len);
+        if (len == blob_len)
+        if (len == blob_len) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "gfid: %s. No null byte present.",
+                        args->gfid);
+                goto err;
+        }
+
+        args->bname = GF_CALLOC (1, (len + 1), gf_common_mt_char);
+        if (args->bname == NULL)
+                goto err;
+
+        memcpy (args->bname, blob, (len + 1));
+        blob += (len + 1);
+        blob_len -= (len + 1);
+
+        if (S_ISDIR (args->st_mode)) {
+                if (blob_len < sizeof (uint32_t)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "gfid: %s. Invalid length",
+                                args->gfid);
+                        goto err;
+                }
+                args->args.mkdir.mode = ntoh32 (*(uint32_t *)blob);
+                blob += sizeof (uint32_t);
+                blob_len -= sizeof (uint32_t);
+
+                if (blob_len < sizeof (uint32_t)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "gfid: %s. Invalid length",
+                                args->gfid);
+                        goto err;
+                }
+                args->args.mkdir.umask = ntoh32 (*(uint32_t *)blob);
+                blob += sizeof (uint32_t);
+                blob_len -= sizeof (uint32_t);
+                if (blob_len < 0) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "gfid: %s. Invalid length",
+                                args->gfid);
+                        goto err;
+                }
+        } else if (S_ISLNK (args->st_mode)) {
+                len = strnlen (blob, blob_len);
+                if (len == blob_len) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "gfid: %s. Invalid length",
+                                args->gfid);
+                        goto err;
+                }
+                args->args.symlink.linkpath = GF_CALLOC (1, len + 1,
+                                                         gf_common_mt_char);
+                if (args->args.symlink.linkpath == NULL)
+                        goto err;
+
+                memcpy (args->args.symlink.linkpath, blob, (len + 1));
+                blob += (len + 1);
+                blob_len -= (len + 1);
+        } else {
+                if (blob_len < sizeof (uint32_t)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "gfid: %s. Invalid length",
+                                args->gfid);
+                        goto err;
+                }
+                args->args.mknod.mode = ntoh32 (*(uint32_t *)blob);
+                blob += sizeof (uint32_t);
+                blob_len -= sizeof (uint32_t);
+
+                if (blob_len < sizeof (uint32_t)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "gfid: %s. Invalid length",
+                                args->gfid);
+                        goto err;
+                }
+                args->args.mknod.rdev = ntoh32 (*(uint32_t *)blob);
+                blob += sizeof (uint32_t);
+                blob_len -= sizeof (uint32_t);
+
+                if (blob_len < sizeof (uint32_t)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "gfid: %s. Invalid length",
+                                args->gfid);
+                        goto err;
+                }
+                args->args.mknod.umask = ntoh32 (*(uint32_t *)blob);
+                blob += sizeof (uint32_t);
+                blob_len -= sizeof (uint32_t);
+        }
+
+        if (blob_len) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "gfid: %s. Invalid length",
+                        args->gfid);
+                goto err;
+        }
+
+        return args;
+
+err:
+        if (args)
+                ga_newfile_args_free (args);
+
+        return NULL;
+}
+
+ga_heal_args_t *
+ga_heal_parse_args (xlator_t *this, data_t *data)
+{
+        ga_heal_args_t *args     = NULL;
+        ga_private_t   *priv     = NULL;
+        void           *blob     = NULL;
+        int             len      = 0;
+        int             blob_len = 0;
+
+        blob = data->data;
+        blob_len = data->len;
+
+        priv = this->private;
+
+        /* bname should at least contain a character */
+        if (blob_len < (sizeof (args->gfid) + 2))
+                goto err;
+
+        args = mem_get0 (priv->heal_args_pool);
+        if (!args)
+                goto err;
+
+        memcpy (args->gfid, blob, sizeof (args->gfid));
+        blob += sizeof (args->gfid);
+        blob_len -= sizeof (args->gfid);
+
+        len = strnlen (blob, blob_len);
+        if (len == blob_len)
+                goto err;
+
+        args->bname = GF_CALLOC (1, len + 1, gf_common_mt_char);
+        if (!args->bname)
+                goto err;
+
+        memcpy (args->bname, blob, len);
+        blob_len -= (len + 1);
+
+        if (blob_len)
+                goto err;
+
+        return args;
+
+err:
+        if (args)
+                ga_heal_args_free (args);
+
+        return NULL;
+}
+
+static int32_t
+ga_fill_tmp_loc (loc_t *loc, xlator_t *this, char *gfid,
+                 char *bname, dict_t *xdata, loc_t *new_loc)
+{
+        int       ret    = -1;
+        uint64_t  value  = 0;
+        inode_t  *parent = NULL;
+
+        parent = loc->inode;
+        ret = inode_ctx_get (loc->inode, this, &value);
+        if (!ret) {
+                parent = (void *)value;
+        }
+
+        /* parent itself should be looked up */
+        uuid_copy (new_loc->pargfid, parent->gfid);
+        new_loc->parent = inode_ref (parent);
+
+        new_loc->inode = inode_grep (parent->table, parent, bname);
+        if (!new_loc->inode)
+                new_loc->inode = inode_new (parent->table);
+
+        loc_path (new_loc, bname);
+        new_loc->name = basename (new_loc->path);
+
+        /* As GFID would not be set on the entry yet, lets not send entry
+           gfid in the request */
+        /*uuid_copy (new_loc->gfid, (const unsigned char *)gfid); */
+
+        ret = dict_set_static_bin (xdata, "gfid-req", gfid, 16);
+        if (ret < 0)
+                goto out;
+
+        ret = 0;
+
+out:
+        return ret;
+}
+
+
+
+static gf_boolean_t
+__is_gfid_access_dir (uuid_t gfid)
+{
+        uuid_t  aux_gfid;
+
+        memset (aux_gfid, 0, 16);
+        aux_gfid[15] = GF_AUX_GFID;
+
+        if (uuid_compare (gfid, aux_gfid) == 0)
+                return _gf_true;
+
+        return _gf_false;
+}
+
+int32_t
+ga_forget (xlator_t *this, inode_t *inode)
+{
+        int       ret = -1;
+        uint64_t  value = 0;
+        inode_t  *tmp_inode = NULL;
+
+        ret = inode_ctx_del (inode, this, &value);
+        if (ret)
+                goto out;
+
+        tmp_inode = (void *)value;
+        inode_unref (tmp_inode);
+
+out:
+        return 0;
+}
+
+
+static int
+ga_heal_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+             int32_t op_ret, int32_t op_errno,
+             inode_t *inode, struct iatt *stat, dict_t *dict,
+             struct iatt *postparent)
+{
+        call_frame_t *orig_frame = NULL;
+
+        orig_frame = frame->local;
+        frame->local = NULL;
+
+        /* don't worry about inode linking and other stuff. They'll happen on
+         * the next lookup.
+         */
+        STACK_DESTROY (frame->root);
+
+        STACK_UNWIND_STRICT (setxattr, orig_frame, op_ret, op_errno, dict);
+
+        return 0;
+}
+
+static int
+ga_newentry_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno,
+                 inode_t *inode, struct iatt *buf,
+                 struct iatt *preparent, struct iatt *postparent,
+                 dict_t *xdata)
+{
+        call_frame_t *orig_frame = NULL;
+
+        orig_frame = frame->local;
+        frame->local = NULL;
+
+        /* don't worry about inode linking and other stuff. They'll happen on
+         * the next lookup.
+         */
+        STACK_DESTROY (frame->root);
+
+        STACK_UNWIND_STRICT (setxattr, orig_frame, op_ret, op_errno, xdata);
+
+        return 0;
+}
+
+int32_t
+ga_new_entry (call_frame_t *frame, xlator_t *this, loc_t *loc, data_t *data,
+              dict_t *xdata)
+{
+        int                ret       = -1;
+        ga_newfile_args_t *args      = NULL;
+        loc_t              tmp_loc   = {0,};
+        call_frame_t      *new_frame = NULL;
+        mode_t             mode      = 0;
+
+        args = ga_newfile_parse_args (this, data);
+        if (!args)
+                goto out;
+
+        if (!xdata)
+                xdata = dict_new ();
+
+        ret = ga_fill_tmp_loc (loc, this, args->gfid,
+                               args->bname, xdata, &tmp_loc);
+        if (ret)
+                goto out;
+
+        new_frame = copy_frame (frame);
+        if (!new_frame)
+                goto out;
+        new_frame->local = (void *)frame;
+
+        new_frame->root->uid = args->uid;
+        new_frame->root->gid = args->gid;
+
+        if (S_ISDIR (args->st_mode)) {
+                STACK_WIND (new_frame, ga_newentry_cbk,
+                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir,
+                            &tmp_loc, args->args.mkdir.mode,
+                            args->args.mkdir.umask, xdata);
+        } else if (S_ISLNK (args->st_mode)) {
+                STACK_WIND (new_frame, ga_newentry_cbk,
+                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->symlink,
+                            args->args.symlink.linkpath,
+                            &tmp_loc, 0, xdata);
+        } else {
+                /* use 07777 (4 7s) for considering the Sticky bits etc) */
+                mode = (S_IFMT & args->st_mode) |
+                        (07777 | args->args.mknod.mode);;
+
+                STACK_WIND (new_frame, ga_newentry_cbk,
+                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->mknod,
+                            &tmp_loc, mode,
+                            args->args.mknod.rdev, args->args.mknod.umask,
+                            xdata);
+        }
+
+        ret = 0;
+out:
+        ga_newfile_args_free (args);
+
+        return ret;
+}
+
+int32_t
+ga_heal_entry (call_frame_t *frame, xlator_t *this, loc_t *loc, data_t *data,
+               dict_t *xdata)
+{
+        int             ret       = -1;
+        ga_heal_args_t *args      = NULL;
+        loc_t           tmp_loc   = {0,};
+        call_frame_t   *new_frame = NULL;
+
+        args = ga_heal_parse_args (this, data);
+        if (!args)
+                goto out;
+
+        if (!xdata)
+                xdata = dict_new ();
+
+        ret = ga_fill_tmp_loc (loc, this, args->gfid, args->bname,
+                               xdata, &tmp_loc);
+        if (ret)
+                goto out;
+
+        new_frame = copy_frame (frame);
+        if (!new_frame)
+                goto out;
+        new_frame->local = (void *)frame;
+
+        STACK_WIND (new_frame, ga_heal_cbk, FIRST_CHILD (this),
+                    FIRST_CHILD(this)->fops->lookup,
+                    &tmp_loc, xdata);
+
+        ret = 0;
+out:
+        if (args)
+                ga_heal_args_free (args);
+
+        return ret;
+}
+
+int32_t
+ga_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno,
+                 dict_t *xdata)
+{
+        STACK_UNWIND_STRICT (setxattr, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+int32_t
+ga_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+             int32_t flags, dict_t *xdata)
+{
+
+        data_t  *data     = NULL;
+        int      op_errno = ENOMEM;
+        int      ret      = 0;
+        inode_t *unref    = NULL;
+
+        if ((loc->name && !strcmp (GF_GFID_DIR, loc->name)) &&
+            ((loc->parent &&
+              __is_root_gfid (loc->parent->gfid)) ||
+             __is_root_gfid (loc->pargfid))) {
+                op_errno = EPERM;
+                goto err;
+        }
+
+        data = dict_get (dict, GF_FUSE_AUX_GFID_NEWFILE);
+        if (data) {
+                ret = ga_new_entry (frame, this, loc, data, xdata);
+                if (ret)
+                        goto err;
+                return 0;
+        }
+
+        data = dict_get (dict, GF_FUSE_AUX_GFID_HEAL);
+        if (data) {
+                ret = ga_heal_entry (frame, this, loc, data, xdata);
+                if (ret)
+                        goto err;
+                return 0;
+        }
+
+        //If the inode is a virtual inode change the inode otherwise perform
+        //the operation on same inode
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, loc, unref, wind);
+
+wind:
+        STACK_WIND (frame, ga_setxattr_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->setxattr, loc, dict, flags,
+                    xdata);
+        if (unref)
+                inode_unref (unref);
+
+        return 0;
+err:
+        STACK_UNWIND_STRICT (setxattr, frame, -1, op_errno, xdata);
+        return 0;
+}
+
+
+int32_t
+ga_virtual_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, inode_t *inode,
+                       struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+        int       j           = 0;
+        int       i           = 0;
+        int       ret         = 0;
+        uint64_t  temp_ino    = 0;
+        inode_t  *cbk_inode   = NULL;
+        inode_t  *true_inode  = NULL;
+        uuid_t    random_gfid = {0,};
+
+        if (frame->local)
+                cbk_inode = frame->local;
+        else
+                cbk_inode = inode;
+
+        frame->local = NULL;
+        if (op_ret)
+                goto unwind;
+
+        if (!IA_ISDIR (buf->ia_type))
+                goto unwind;
+
+        /* need to send back a different inode for linking in itable */
+        if (cbk_inode == inode) {
+                /* check if the inode is in the 'itable' or
+                   if its just previously discover()'d inode */
+                true_inode = inode_find (inode->table, buf->ia_gfid);
+                if (!true_inode) {
+                        cbk_inode = inode_new (inode->table);
+
+                        if (!cbk_inode) {
+                                op_ret = -1;
+                                op_errno = ENOMEM;
+                                goto unwind;
+                        }
+                        /* the inode is not present in itable, ie, the actual
+                           path is not yet looked up. Use the current inode
+                           itself for now */
+                        inode_ref (inode);
+                } else {
+                        /* 'inode_ref()' has been done in inode_find() */
+                        inode = true_inode;
+                }
+
+                ret = inode_ctx_put (cbk_inode, this, (uint64_t)inode);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "failed to set the inode ctx with"
+                                "the actual inode");
+                        if (inode)
+                                inode_unref (inode);
+                }
+                inode = NULL;
+        }
+
+        if (!uuid_is_null (cbk_inode->gfid)) {
+                /* if the previous linked inode is used, use the
+                   same gfid */
+                uuid_copy (random_gfid, cbk_inode->gfid);
+        } else {
+                /* replace the buf->ia_gfid to a random gfid
+                   for directory, for files, what we received is fine */
+                uuid_generate (random_gfid);
+        }
+
+        uuid_copy (buf->ia_gfid, random_gfid);
+
+        for (i = 15; i > (15 - 8); i--) {
+                temp_ino += (uint64_t)(buf->ia_gfid[i]) << j;
+                j += 8;
+        }
+        buf->ia_ino = temp_ino;
+
+unwind:
+        STACK_UNWIND_STRICT (lookup, frame, op_ret, op_errno, cbk_inode, buf,
+                             xdata, postparent);
+
+        return 0;
+}
+
+int32_t
+ga_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+        ga_private_t *priv = NULL;
+
+        /* if the entry in question is not 'root',
+           then follow the normal path */
+        if (op_ret || !__is_root_gfid(buf->ia_gfid))
+                goto unwind;
+
+        priv = this->private;
+
+        /* do we need to copy root stbuf everytime? */
+        /* mostly yes, as we want to have the 'stat' info show latest
+           in every _cbk() */
+
+        /* keep the reference for root stat buf */
+        priv->root_stbuf = *buf;
+        priv->gfiddir_stbuf = priv->root_stbuf;
+        priv->gfiddir_stbuf.ia_gfid[15] = GF_AUX_GFID;
+        priv->gfiddir_stbuf.ia_ino = GF_AUX_GFID;
+
+unwind:
+        STACK_UNWIND_STRICT (lookup, frame, op_ret, op_errno, inode, buf,
+                             xdata, postparent);
+        return 0;
+}
+
+int32_t
+ga_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+        ga_private_t *priv     = NULL;
+        int           ret      = -1;
+        uuid_t        tmp_gfid = {0,};
+        loc_t         tmp_loc  = {0,};
+        uint64_t      value    = 0;
+        inode_t      *inode    = NULL;
+        inode_t      *true_inode    = NULL;
+        int32_t       op_errno = ENOENT;
+
+        /* if its discover(), no need for any action here */
+        if (!loc->name)
+                goto wind;
+
+        /* if its revalidate, and inode is not of type directory,
+           proceed with 'wind' */
+        if (loc->inode && loc->inode->ia_type &&
+            !IA_ISDIR (loc->inode->ia_type))
+                goto wind;
+
+        priv = this->private;
+
+        /* need to check if the lookup is on virtual dir */
+        if ((loc->name && !strcmp (GF_GFID_DIR, loc->name)) &&
+            ((loc->parent && __is_root_gfid (loc->parent->gfid)) ||
+             __is_root_gfid (loc->pargfid))) {
+                /* this means, the query is on '/.gfid', return the fake stat,
+                   and say success */
+
+                STACK_UNWIND_STRICT (lookup, frame, 0, 0, loc->inode,
+                                     &priv->gfiddir_stbuf, xdata,
+                                     &priv->root_stbuf);
+                return 0;
+        }
+
+        /* now, check if the lookup() is on an existing entry,
+           but on gfid-path */
+        if (!((loc->parent && __is_gfid_access_dir (loc->parent->gfid)) ||
+              __is_gfid_access_dir (loc->pargfid)))
+                goto wind;
+
+        /* make sure the 'basename' is actually a 'canonical-gfid',
+           otherwise, return error */
+        ret = uuid_parse (loc->name, tmp_gfid);
+        if (ret)
+                goto err;
+
+        /* if its fresh lookup, go ahead and send it down, if not,
+           for directory, we need indirection to actual dir inode */
+        if (!(loc->inode && loc->inode->ia_type))
+                goto discover;
+
+        /* revalidate on directory */
+        ret = inode_ctx_get (loc->inode, this, &value);
+        if (ret)
+                goto err;
+
+        inode = (void *)value;
+
+        /* valid inode, already looked up, work on that */
+        if (inode->ia_type)
+                goto discover;
+
+        /* check if the inode is in the 'itable' or
+           if its just previously discover()'d inode */
+        true_inode = inode_find (loc->inode->table, tmp_gfid);
+        if (true_inode) {
+                /* time do another lookup and update the context
+                   with proper inode */
+                op_errno = ESTALE;
+                goto err;
+        }
+
+discover:
+        /* for the virtual entries, we don't need to send 'gfid-req' key, as
+           for these entries, we don't want to 'set' a new gfid */
+        if (xdata)
+                dict_del (xdata, "gfid-req");
+
+        uuid_copy (tmp_loc.gfid, tmp_gfid);
+
+        /* if revalidate, then we need to have the proper reference */
+        if (inode) {
+                tmp_loc.inode = inode_ref (inode);
+                frame->local = loc->inode;
+        } else {
+                tmp_loc.inode = inode_ref (loc->inode);
+        }
+
+        STACK_WIND (frame, ga_virtual_lookup_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->lookup, &tmp_loc, xdata);
+
+        inode_unref (tmp_loc.inode);
+
+        return 0;
+
+wind:
+        /* used for all the normal lookup path */
+        STACK_WIND (frame, ga_lookup_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->lookup, loc, xdata);
+
+        return 0;
+
+err:
+        STACK_UNWIND_STRICT (lookup, frame, -1, op_errno, loc->inode,
+                             &priv->gfiddir_stbuf, xdata,
+                             &priv->root_stbuf);
+        return 0;
+}
+
+int
+ga_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          mode_t umask, dict_t *xdata)
+{
+        int op_errno = 0;
+
+        GFID_ACCESS_ENTRY_OP_CHECK (loc, op_errno, err);
+
+        STACK_WIND (frame, default_mkdir_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->mkdir, loc, mode, umask,
+                    xdata);
+
+        return 0;
+
+err:
+        STACK_UNWIND_STRICT (mkdir, frame, -1, op_errno, loc->inode,
+                             NULL, NULL, NULL, xdata);
+        return 0;
+}
+
+
+int
+ga_create (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+           mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+        int op_errno = 0;
+
+        GFID_ACCESS_ENTRY_OP_CHECK (loc, op_errno, err);
+
+        STACK_WIND (frame, default_create_cbk,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->create,
+                    loc, flags, mode, umask, fd, xdata);
+        return 0;
+err:
+        STACK_UNWIND_STRICT (create, frame, -1, op_errno, NULL,
+                             NULL, NULL, NULL, NULL, xdata);
+
+        return 0;
+
+}
+
+int
+ga_symlink (call_frame_t *frame, xlator_t *this, const char *linkname,
+            loc_t *loc, mode_t umask, dict_t *xdata)
+{
+        int op_errno = 0;
+
+        GFID_ACCESS_ENTRY_OP_CHECK (loc, op_errno, err);
+
+        STACK_WIND (frame, default_symlink_cbk,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->symlink,
+                    linkname, loc, umask, xdata);
+        return 0;
+err:
+        STACK_UNWIND_STRICT (symlink, frame, -1, op_errno, NULL,
+                             NULL, NULL, NULL, xdata);
+
+        return 0;
+}
+
+int
+ga_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dev_t rdev, mode_t umask, dict_t *xdata)
+{
+        int op_errno = 0;
+
+        GFID_ACCESS_ENTRY_OP_CHECK (loc, op_errno, err);
+
+        STACK_WIND (frame, default_mknod_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->mknod, loc, mode, rdev,
+                    umask, xdata);
+
+        return 0;
+err:
+        STACK_UNWIND_STRICT (mknod, frame, -1, op_errno, NULL,
+                             NULL, NULL, NULL, xdata);
+
+        return 0;
+}
+
+int
+ga_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flag,
+          dict_t *xdata)
+{
+        int op_errno = 0;
+        inode_t *unref = NULL;
+
+        GFID_ACCESS_ENTRY_OP_CHECK (loc, op_errno, err);
+
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, loc, unref, wind);
+
+wind:
+        STACK_WIND (frame, default_rmdir_cbk,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->rmdir,
+                    loc, flag, xdata);
+        if (unref)
+                inode_unref (unref);
+
+        return 0;
+err:
+        STACK_UNWIND_STRICT (rmdir, frame, -1, op_errno, NULL,
+                             NULL, xdata);
+
+        return 0;
+}
+
+int
+ga_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t xflag,
+           dict_t *xdata)
+{
+        int op_errno = 0;
+        inode_t *unref = NULL;
+
+        GFID_ACCESS_ENTRY_OP_CHECK (loc, op_errno, err);
+
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, loc, unref, wind);
+
+wind:
+        STACK_WIND (frame, default_unlink_cbk,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->unlink,
+                    loc, xflag, xdata);
+
+        if (unref)
+                inode_unref (unref);
+
+        return 0;
+err:
+        STACK_UNWIND_STRICT (unlink, frame, -1, op_errno, NULL,
+                             NULL, xdata);
+
+        return 0;
+}
+
+int
+ga_rename (call_frame_t *frame, xlator_t *this,
+           loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+{
+        int op_errno = 0;
+        inode_t *oldloc_unref = NULL;
+        inode_t *newloc_unref = NULL;
+
+        GFID_ACCESS_ENTRY_OP_CHECK (oldloc, op_errno, err);
+        GFID_ACCESS_ENTRY_OP_CHECK (newloc, op_errno, err);
+
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, oldloc, oldloc_unref,
+                                         handle_newloc);
+
+handle_newloc:
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, newloc, newloc_unref, wind);
+
+wind:
+        STACK_WIND (frame, default_rename_cbk,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->rename,
+                    oldloc, newloc, xdata);
+
+        if (oldloc_unref)
+                inode_unref (oldloc_unref);
+
+        if (newloc_unref)
+                inode_unref (newloc_unref);
+
+        return 0;
+err:
+        STACK_UNWIND_STRICT (rename, frame, -1, op_errno, NULL,
+                             NULL, NULL, NULL, NULL, xdata);
+
+        return 0;
+}
+
+
+int
+ga_link (call_frame_t *frame, xlator_t *this,
+         loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+{
+        int op_errno = 0;
+        inode_t *oldloc_unref = NULL;
+        inode_t *newloc_unref = NULL;
+
+        GFID_ACCESS_ENTRY_OP_CHECK (oldloc, op_errno, err);
+        GFID_ACCESS_ENTRY_OP_CHECK (newloc, op_errno, err);
+
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, oldloc, oldloc_unref,
+                                         handle_newloc);
+
+handle_newloc:
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, newloc, newloc_unref, wind);
+
+wind:
+        STACK_WIND (frame, default_link_cbk,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->link,
+                    oldloc, newloc, xdata);
+
+        if (oldloc_unref)
+                inode_unref (oldloc_unref);
+
+        if (newloc_unref)
+                inode_unref (newloc_unref);
+
+        return 0;
+err:
+        STACK_UNWIND_STRICT (link, frame, -1, op_errno, NULL,
+                             NULL, NULL, NULL, xdata);
+
+        return 0;
+}
+
+int32_t
+ga_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc,
+            fd_t *fd, dict_t *xdata)
+{
+        int op_errno = 0;
+
+        GFID_ACCESS_ENTRY_OP_CHECK (loc, op_errno, err);
+
+        /* also check if the loc->inode itself is virtual
+           inode, if yes, return with failure, mainly because we
+           can't handle all the readdirp and other things on it. */
+        if (inode_ctx_get (loc->inode, this, NULL) == 0) {
+                op_errno = ENOTSUP;
+                goto err;
+        }
+
+        STACK_WIND (frame, default_opendir_cbk,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->opendir,
+                    loc, fd, xdata);
+        return 0;
+err:
+        STACK_UNWIND_STRICT (opendir, frame, -1, op_errno, NULL, xdata);
+
+        return 0;
+}
+
+int32_t
+ga_getxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
+             const char *name, dict_t *xdata)
+{
+        inode_t *unref = NULL;
+
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, loc, unref, wind);
+
+wind:
+        STACK_WIND (frame, default_getxattr_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+
+        if (unref)
+                inode_unref (unref);
+
+        return 0;
+}
+
+int32_t
+ga_stat (call_frame_t *frame, xlator_t *this, loc_t *loc,
+         dict_t *xdata)
+{
+        inode_t *unref = NULL;
+
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, loc, unref, wind);
+
+wind:
+        STACK_WIND (frame, default_stat_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->stat, loc, xdata);
+        if (unref)
+                inode_unref (unref);
+
+        return 0;
+}
+
+int32_t
+ga_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
+            struct iatt *stbuf, int32_t valid,
+            dict_t *xdata)
+{
+        inode_t *unref = NULL;
+
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, loc, unref, wind);
+
+wind:
+        STACK_WIND (frame, default_setattr_cbk, FIRST_CHILD (this),
+                    FIRST_CHILD (this)->fops->setattr, loc, stbuf, valid,
+                    xdata);
+        if (unref)
+                inode_unref (unref);
+
+        return 0;
+}
+
+int32_t
+ga_removexattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *name, dict_t *xdata)
+{
+        inode_t *unref = NULL;
+
+        GFID_ACCESS_GET_VALID_DIR_INODE (this, loc, unref, wind);
+
+wind:
+        STACK_WIND (frame, default_removexattr_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->removexattr, loc, name,
+                    xdata);
+        if (unref)
+                inode_unref (unref);
+
+        return 0;
+}
+
+
+int32_t
+mem_acct_init (xlator_t *this)
+{
+        int     ret = -1;
+
+        if (!this)
+                return ret;
+
+        ret = xlator_mem_acct_init (this, gf_gfid_access_mt_end + 1);
+
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_WARNING, "Memory accounting"
+                        " init failed");
+                return ret;
+        }
+
+        return ret;
+}
+
+int32_t
+init (xlator_t *this)
+{
+        ga_private_t *priv = NULL;
+        int ret = -1;
+
+        if (!this->children || this->children->next) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "not configured with exactly one child. exiting");
+                goto out;
+        }
+
+        /* This can be the top of graph in certain cases */
+        if (!this->parents) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "dangling volume. check volfile ");
+        }
+
+        /* TODO: define a mem-type structure */
+        priv = GF_CALLOC (1, sizeof (*priv), gf_gfid_access_mt_priv_t);
+        if (!priv)
+                goto out;
+
+        priv->newfile_args_pool = mem_pool_new (ga_newfile_args_t, 512);
+        if (!priv->newfile_args_pool)
+                goto out;
+
+        priv->heal_args_pool = mem_pool_new (ga_heal_args_t, 512);
+        if (!priv->heal_args_pool)
+                goto out;
+
+        this->private = priv;
+
+        ret = 0;
+out:
+        if (ret && priv) {
+                if (priv->newfile_args_pool)
+                        mem_pool_destroy (priv->newfile_args_pool);
+                GF_FREE (priv);
+        }
+
+        return ret;
+}
+
+void
+fini (xlator_t *this)
+{
+        ga_private_t *priv = NULL;
+        priv = this->private;
+        this->private = NULL;
+
+        if (priv) {
+                if (priv->newfile_args_pool)
+                        mem_pool_destroy (priv->newfile_args_pool);
+                if (priv->heal_args_pool)
+                        mem_pool_destroy (priv->heal_args_pool);
+                GF_FREE (priv);
+        }
+
+        return;
+}
+
+
+struct xlator_fops fops = {
+        .lookup = ga_lookup,
+
+        /* entry fops */
+        .mkdir   = ga_mkdir,
+        .mknod   = ga_mknod,
+        .create  = ga_create,
+        .symlink = ga_symlink,
+        .link    = ga_link,
+        .unlink  = ga_unlink,
+        .rmdir   = ga_rmdir,
+        .rename  = ga_rename,
+
+        /* handle any other directory operations here */
+        .opendir  = ga_opendir,
+        .stat     = ga_stat,
+        .setattr  = ga_setattr,
+        .getxattr = ga_getxattr,
+        .removexattr = ga_removexattr,
+
+        /* special fop to handle more entry creations */
+        .setxattr = ga_setxattr,
+};
+
+struct xlator_cbks cbks = {
+        .forget = ga_forget,
+};
+
+struct volume_options options[] = {
+        /* This translator doesn't take any options, or provide any options */
+        { .key  = {NULL} },
+};
diff --git a/xlators/features/gfid-access/src/gfid-access.h b/xlators/features/gfid-access/src/gfid-access.h
new file mode 100644
index 000000000..e13c9b724
--- /dev/null
+++ b/xlators/features/gfid-access/src/gfid-access.h
@@ -0,0 +1,128 @@
+/*
+   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 __GFID_ACCESS_H__
+#define __GFID_ACCESS_H__
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "glusterfs.h"
+#include "logging.h"
+#include "dict.h"
+#include "xlator.h"
+#include "defaults.h"
+#include "gfid-access-mem-types.h"
+
+#define UUID_CANONICAL_FORM_LEN 36
+
+#define GF_FUSE_AUX_GFID_NEWFILE "glusterfs.gfid.newfile"
+#define GF_FUSE_AUX_GFID_HEAL    "glusterfs.gfid.heal"
+
+#define GF_GFID_KEY "GLUSTERFS_GFID"
+#define GF_GFID_DIR ".gfid"
+#define GF_AUX_GFID 0xd
+
+#define GFID_ACCESS_GET_VALID_DIR_INODE(x,l,unref,lbl) do {             \
+                int       ret       = 0;                                \
+                uint64_t  value     = 0;                                \
+                inode_t  *tmp_inode = NULL;                             \
+                                                                        \
+                /* if its an entry operation, on the virtual */         \
+                /* directory inode as parent, we need to handle */      \
+                /* it properly */                                       \
+                if (l->parent) {                                        \
+                        ret = inode_ctx_get (l->parent, x, &value);     \
+                        if (ret)                                        \
+                                goto lbl;                               \
+                        tmp_inode = (inode_t *)value;                   \
+                        unref = inode_ref (tmp_inode);                  \
+                        l->parent = tmp_inode;                          \
+                        /* if parent is virtual, no need to handle */   \
+                        /* loc->inode */                                \
+                        break;                                          \
+                }                                                       \
+                                                                        \
+                /* if its an inode operation, on the virtual */         \
+                /* directory inode itself, we need to handle */         \
+                /* it properly */                                       \
+                if (l->inode) {                                         \
+                        ret = inode_ctx_get (l->inode, x, &value);      \
+                        if (ret)                                        \
+                                goto lbl;                               \
+                        tmp_inode = (inode_t *)value;                   \
+                        unref = inode_ref (tmp_inode);                  \
+                        l->inode = tmp_inode;                           \
+                }                                                       \
+                                                                        \
+        } while (0)
+
+#define GFID_ACCESS_ENTRY_OP_CHECK(loc,err,lbl)    do {                 \
+                /* need to check if the lookup is on virtual dir */     \
+                if ((loc->name && !strcmp (GF_GFID_DIR, loc->name)) &&  \
+                    ((loc->parent &&                                    \
+                      __is_root_gfid (loc->parent->gfid)) ||            \
+                      __is_root_gfid (loc->pargfid))) {                 \
+                        err = EEXIST;                                   \
+                        goto lbl;                                       \
+                }                                                       \
+                                                                        \
+                /* now, check if the lookup() is on an existing */      \
+                /* entry, but on gfid-path */                           \
+                if ((loc->parent &&                                     \
+                     __is_gfid_access_dir (loc->parent->gfid)) ||       \
+                    __is_gfid_access_dir (loc->pargfid)) {              \
+                        err = EPERM;                                    \
+                        goto lbl;                                       \
+                }                                                       \
+        } while (0)
+
+
+typedef struct {
+        unsigned int  uid;
+        unsigned int  gid;
+        char          gfid[UUID_CANONICAL_FORM_LEN + 1];
+        unsigned int  st_mode;
+        char         *bname;
+
+        union {
+                struct _symlink_in {
+                        char     *linkpath;
+                } __attribute__ ((__packed__)) symlink;
+
+                struct _mknod_in {
+                        unsigned int mode;
+                        unsigned int rdev;
+                        unsigned int umask;
+                } __attribute__ ((__packed__)) mknod;
+
+                struct _mkdir_in {
+                        unsigned int mode;
+                        unsigned int umask;
+                } __attribute__ ((__packed__)) mkdir;
+        } __attribute__ ((__packed__)) args;
+} __attribute__((__packed__)) ga_newfile_args_t;
+
+typedef struct {
+        char      gfid[UUID_CANONICAL_FORM_LEN + 1];
+        char     *bname; /* a null terminated basename */
+} __attribute__((__packed__)) ga_heal_args_t;
+
+struct ga_private {
+        /* root inode's stbuf */
+        struct iatt root_stbuf;
+        struct iatt gfiddir_stbuf;
+        struct mem_pool *newfile_args_pool;
+        struct mem_pool *heal_args_pool;
+};
+typedef struct ga_private ga_private_t;
+
+#endif /* __GFID_ACCESS_H__ */
diff --git a/xlators/features/index/src/index.c b/xlators/features/index/src/index.c
index 3245076ec..9253120f3 100644
--- a/xlators/features/index/src/index.c
+++ b/xlators/features/index/src/index.c
@@ -15,8 +15,10 @@
 #include "index.h"
 #include "options.h"
 #include "glusterfs3-xdr.h"
+#include "syncop.h"
 
 #define XATTROP_SUBDIR "xattrop"
+#define BASE_INDICES_HOLDER_SUBDIR "base_indices_holder"
 
 call_stub_t *
 __index_dequeue (struct list_head *callstubs)
@@ -242,20 +244,40 @@ check_delete_stale_index_file (xlator_t *this, char *filename)
 {
         int             ret = 0;
         struct stat     st = {0};
+        struct stat     base_index_st = {0};
         char            filepath[PATH_MAX] = {0};
+        char            filepath_under_base_indices_holder[PATH_MAX] = {0};
         index_priv_t    *priv = NULL;
 
         priv = this->private;
+        if (priv->to_be_healed_states != synced_state)
+                return;
+
         make_file_path (priv->index_basepath, XATTROP_SUBDIR,
                         filename, filepath, sizeof (filepath));
+
+        make_file_path (priv->index_basepath, BASE_INDICES_HOLDER_SUBDIR,
+                        filename, filepath_under_base_indices_holder,
+                        sizeof (filepath_under_base_indices_holder));
+
+
+        ret = stat (filepath_under_base_indices_holder, &base_index_st);
+        if (ret) {
+                gf_log (THIS->name, GF_LOG_ERROR, "Base index is not created"
+                        "under index/base_indices_holder");
+                return;
+        }
+
         ret = stat (filepath, &st);
-        if (!ret && st.st_nlink == 1)
+        if (!ret && st.st_nlink == 2) {
                 unlink (filepath);
+                unlink (filepath_under_base_indices_holder);
+        }
 }
 
 static int
 index_fill_readdir (fd_t *fd, DIR *dir, off_t off,
-                    size_t size, gf_dirent_t *entries)
+                    size_t size, gf_dirent_t *entries, readdir_directory type)
 {
         off_t     in_case = -1;
         size_t    filled = 0;
@@ -298,7 +320,8 @@ index_fill_readdir (fd_t *fd, DIR *dir, off_t off,
                 }
 
                 if (!strncmp (entry->d_name, XATTROP_SUBDIR"-",
-                              strlen (XATTROP_SUBDIR"-"))) {
+                              strlen (XATTROP_SUBDIR"-")) &&
+                              (type == INDEX_XATTROP)) {
                         check_delete_stale_index_file (this, entry->d_name);
                         continue;
                 }
@@ -337,16 +360,175 @@ out:
 }
 
 int
+sync_base_indices (void *index_priv)
+{
+        index_priv_t    *priv = NULL;
+        DIR             *dir_base_holder  = NULL;
+        DIR             *xattrop_dir = NULL;
+        struct dirent   *entry = NULL;
+        char            base_indices_holder[PATH_MAX] = {0};
+        char            xattrop_directory[PATH_MAX] = {0};
+        char            base_index_path[PATH_MAX] = {0};
+        char            xattrop_index_path[PATH_MAX] = {0};
+        int             ret = 0;
+
+        priv = index_priv;
+
+        snprintf (base_indices_holder, PATH_MAX, "%s/%s", priv->index_basepath,
+                  BASE_INDICES_HOLDER_SUBDIR);
+        snprintf (xattrop_directory, PATH_MAX, "%s/%s", priv->index_basepath,
+                  XATTROP_SUBDIR);
+
+        if ((dir_base_holder = opendir(base_indices_holder)) == NULL) {
+                ret = -1;
+                goto out;
+        }
+        if ((xattrop_dir = opendir (xattrop_directory)) == NULL) {
+                ret = -1;
+                goto out;
+        }
+
+        priv->to_be_healed_states = sync_started;
+        while ((entry = readdir(xattrop_dir)) != NULL) {
+                if (!strcmp (entry->d_name, ".") ||
+                    !strcmp (entry->d_name, "..")) {
+                        continue;
+                }
+                if (strncmp (entry->d_name, XATTROP_SUBDIR"-",
+                    strlen (XATTROP_SUBDIR"-"))) {
+                        continue;
+                }
+                if (!strncmp (entry->d_name, XATTROP_SUBDIR"-",
+                    strlen (XATTROP_SUBDIR"-"))) {
+
+                    snprintf (xattrop_index_path, PATH_MAX, "%s/%s",
+                              xattrop_directory, entry->d_name);
+
+                    snprintf (base_index_path, PATH_MAX, "%s/%s",
+                              base_indices_holder, entry->d_name);
+
+                    ret = link (xattrop_index_path, base_index_path);
+                    if (ret && errno != EEXIST)
+                        goto out;
+
+                }
+        }
+        ret = closedir (xattrop_dir);
+        if (ret)
+                goto out;
+        ret = closedir (dir_base_holder);
+        if (ret)
+                goto out;
+
+        ret = 0;
+out:
+        return ret;
+
+}
+
+int
+base_indices_syncing_done (int ret, call_frame_t *frame, void *data)
+{
+        index_priv_t         *priv = NULL;
+        priv = data;
+
+        if (!priv)
+                goto out;
+
+        if (ret) {
+                priv->to_be_healed_states = sync_not_started;
+        } else {
+                priv->to_be_healed_states = synced_state;
+        }
+
+        STACK_DESTROY (frame->root);
+
+out:
+        return 0;
+}
+
+int
+sync_base_indices_from_xattrop (xlator_t *this)
+{
+
+        index_priv_t          *priv = NULL;
+        char                  base_indices_holder[PATH_MAX] =  {0};
+        int                   ret = 0;
+        struct stat           st = {0};
+        DIR                   *dir = NULL;
+        struct dirent         *entry = NULL;
+        call_frame_t          *frame = NULL;
+
+        priv = this->private;
+
+        if (priv->to_be_healed_states != sync_not_started) {
+                ret = -1;
+                goto out;
+        }
+
+        snprintf (base_indices_holder, PATH_MAX, "%s/%s", priv->index_basepath,
+                  BASE_INDICES_HOLDER_SUBDIR);
+
+        ret = stat (base_indices_holder, &st);
+
+        if (ret && (errno != ENOENT)) {
+                goto out;
+        } else if (errno == ENOENT) {
+                ret = index_dir_create (this, BASE_INDICES_HOLDER_SUBDIR);
+                if (ret)
+                        goto out;
+        } else {
+                if ((dir = opendir (base_indices_holder)) == NULL) {
+                        ret = -1;
+                        goto out;
+                }
+                while ((entry = readdir (dir)) != NULL) {
+                        if (!strcmp (entry->d_name, ".") ||
+                            !strcmp (entry->d_name,"..")) {
+                                continue;
+                        }
+                        ret = unlink (entry->d_name);
+                        if (ret)
+                                goto out;
+                }
+                closedir (dir);
+        }
+
+        /*At this point of time we have index/base_indicies_holder directory
+         *is with no entries*/
+
+        frame = create_frame (this, this->ctx->pool);
+        if (!frame) {
+                ret = -1;
+                goto out;
+        }
+        set_lk_owner_from_ptr (&frame->root->lk_owner, frame->root);
+
+        frame->root->pid = LOW_PRIO_PROC_PID;
+
+        ret = synctask_new (this->ctx->env, sync_base_indices,
+                            base_indices_syncing_done,frame, priv);
+
+
+
+out:
+        return ret;
+
+}
+
+int
 index_add (xlator_t *this, uuid_t gfid, const char *subdir)
 {
         int32_t           op_errno = 0;
         char              gfid_path[PATH_MAX] = {0};
         char              index_path[PATH_MAX] = {0};
+        char              base_path[PATH_MAX] = {0};
         int               ret = 0;
         uuid_t            index = {0};
         index_priv_t      *priv = NULL;
         struct stat       st = {0};
         int               fd = 0;
+        int               index_created = 0;
 
         priv = this->private;
         GF_ASSERT_AND_GOTO_WITH_ERROR (this->name, !uuid_is_null (gfid),
@@ -364,9 +546,11 @@ index_add (xlator_t *this, uuid_t gfid, const char *subdir)
         ret = link (index_path, gfid_path);
         if (!ret || (errno == EEXIST))  {
                 ret = 0;
+                index_created = 1;
                 goto out;
         }
 
+
         op_errno = errno;
         if (op_errno == ENOENT) {
                 ret = index_dir_create (this, subdir);
@@ -398,10 +582,36 @@ index_add (xlator_t *this, uuid_t gfid, const char *subdir)
                         "add to index (%s)", uuid_utoa (gfid),
                         strerror (errno));
                 goto out;
+        } else {
+                index_created = 1;
+        }
+
+        if (priv->to_be_healed_states != sync_not_started) {
+                 make_index_path (priv->index_basepath,
+                                  GF_BASE_INDICES_HOLDER_GFID,
+                                  index, base_path, sizeof (base_path));
+                 ret = link (index_path, base_path);
+                 if (ret)
+                         goto out;
         }
 
         ret = 0;
 out:
+        /*If base_indices_holder is not created: create and sync
+         *If directory is present: delete contents and start syncing
+         *If syncing is in progress :No need to do any thing
+         *If syncing is done: No need to do anything*/
+        if (!ret) {
+                switch (priv->to_be_healed_states) {
+                        case sync_not_started:
+                                ret = sync_base_indices_from_xattrop (this);
+                                break;
+                        case sync_started:
+                        case synced_state:
+                                /*No need to do anything*/
+                                break;
+                }
+        }
         return ret;
 }
 
@@ -737,8 +947,18 @@ index_getxattr_wrapper (call_frame_t *frame, xlator_t *this,
                 goto done;
         }
 
-        ret = dict_set_static_bin (xattr, (char*)name, priv->xattrop_vgfid,
-                                   sizeof (priv->xattrop_vgfid));
+        if (!strcmp (name, GF_XATTROP_INDEX_GFID)) {
+
+                ret = dict_set_static_bin (xattr, (char*)name,
+                                           priv->xattrop_vgfid,
+                                           sizeof (priv->xattrop_vgfid));
+
+        } else if (!strcmp (name, GF_BASE_INDICES_HOLDER_GFID)) {
+
+                ret = dict_set_static_bin (xattr, (char*)name,
+                                           priv->base_indices_holder_vgfid,
+                                      sizeof (priv->base_indices_holder_vgfid));
+        }
         if (ret) {
                 ret = -ENOMEM;
                 gf_log (THIS->name, GF_LOG_ERROR, "xattrop index "
@@ -782,6 +1002,15 @@ index_lookup_wrapper (call_frame_t *frame, xlator_t *this,
         } else if (!uuid_compare (loc->pargfid, priv->xattrop_vgfid)) {
                 make_file_path (priv->index_basepath, XATTROP_SUBDIR,
                                 loc->name, path, sizeof (path));
+        } else if (!uuid_compare (loc->gfid,priv->base_indices_holder_vgfid)){
+                make_index_dir_path (priv->index_basepath,
+                                     BASE_INDICES_HOLDER_SUBDIR, path,
+                                    sizeof (path));
+                is_dir = _gf_true;
+        } else if (!uuid_compare (loc->pargfid, priv->base_indices_holder_vgfid)) {
+                make_file_path (priv->index_basepath,
+                                BASE_INDICES_HOLDER_SUBDIR,loc->name, path,
+                                sizeof (path));
         }
 
         ret = lstat (path, &lstatbuf);
@@ -803,10 +1032,14 @@ index_lookup_wrapper (call_frame_t *frame, xlator_t *this,
         }
 
         iatt_from_stat (&stbuf, &lstatbuf);
-        if (is_dir)
+        if (is_dir && !uuid_compare (loc->gfid, priv->xattrop_vgfid)) {
                 uuid_copy (stbuf.ia_gfid, priv->xattrop_vgfid);
-        else
+        } else if (is_dir &&
+                !uuid_compare (loc->gfid, priv->base_indices_holder_vgfid)) {
+                uuid_copy (stbuf.ia_gfid, priv->base_indices_holder_vgfid);
+        } else {
                 uuid_generate (stbuf.ia_gfid);
+        }
         stbuf.ia_ino = -1;
         op_ret = 0;
 done:
@@ -818,6 +1051,44 @@ done:
 }
 
 int32_t
+base_indices_readdir_wrapper (call_frame_t *frame, xlator_t *this,
+                              fd_t *fd, size_t size, off_t off, dict_t *xdata)
+{
+        index_priv_t    *priv = NULL;
+        char            base_indices_holder[PATH_MAX] = {0};
+        DIR             *dir = NULL;
+        int32_t         op_ret = -1;
+        int32_t         op_errno = 0;
+        int             count = 0;
+        gf_dirent_t     entries;
+
+        priv = this->private;
+
+        make_index_dir_path (priv->index_basepath, BASE_INDICES_HOLDER_SUBDIR,
+                             base_indices_holder, sizeof (base_indices_holder));
+
+        dir = opendir (base_indices_holder);
+        if (!dir) {
+                op_errno = EINVAL;
+                goto done;
+        }
+
+
+        INIT_LIST_HEAD (&entries.list);
+
+        count = index_fill_readdir (fd, dir, off, size, &entries,
+                                    BASE_INDICES_HOLDER);
+        /* pick ENOENT to indicate EOF */
+        op_errno = errno;
+        op_ret = count;
+        closedir (dir);
+done:
+        STACK_UNWIND_STRICT (readdir, frame, op_ret, op_errno, &entries, xdata);
+        gf_dirent_free (&entries);
+        return 0;
+}
+
+int32_t
 index_readdir_wrapper (call_frame_t *frame, xlator_t *this,
                        fd_t *fd, size_t size, off_t off, dict_t *xdata)
 {
@@ -848,7 +1119,8 @@ index_readdir_wrapper (call_frame_t *frame, xlator_t *this,
                 goto done;
         }
 
-        count = index_fill_readdir (fd, dir, off, size, &entries);
+        count = index_fill_readdir (fd, dir, off, size, &entries,
+                                    INDEX_XATTROP);
 
         /* pick ENOENT to indicate EOF */
         op_errno = errno;
@@ -915,7 +1187,10 @@ index_getxattr (call_frame_t *frame, xlator_t *this,
 {
         call_stub_t     *stub = NULL;
 
-        if (!name || strcmp (GF_XATTROP_INDEX_GFID, name))
+        if (!name)
+                goto out;
+        if (strcmp (GF_XATTROP_INDEX_GFID, name) &&
+            strcmp (GF_BASE_INDICES_HOLDER_GFID, name))
                 goto out;
 
         stub = fop_getxattr_stub (frame, index_getxattr_wrapper, loc, name,
@@ -942,7 +1217,9 @@ index_lookup (call_frame_t *frame, xlator_t *this,
         priv = this->private;
 
         if (uuid_compare (loc->gfid, priv->xattrop_vgfid) &&
-            uuid_compare (loc->pargfid, priv->xattrop_vgfid))
+            uuid_compare (loc->pargfid, priv->xattrop_vgfid) &&
+            uuid_compare (loc->gfid, priv->base_indices_holder_vgfid) &&
+            uuid_compare (loc->pargfid, priv->base_indices_holder_vgfid))
                 goto normal;
 
         stub = fop_lookup_stub (frame, index_lookup_wrapper, loc, xattr_req);
@@ -968,10 +1245,19 @@ index_readdir (call_frame_t *frame, xlator_t *this,
         index_priv_t    *priv = NULL;
 
         priv = this->private;
-        if (uuid_compare (fd->inode->gfid, priv->xattrop_vgfid))
+        if (uuid_compare (fd->inode->gfid, priv->xattrop_vgfid) &&
+            uuid_compare (fd->inode->gfid, priv->base_indices_holder_vgfid))
                 goto out;
-        stub = fop_readdir_stub (frame, index_readdir_wrapper, fd, size, off,
-                                 xdata);
+
+        if (!uuid_compare (fd->inode->gfid, priv->xattrop_vgfid)) {
+                stub = fop_readdir_stub (frame, index_readdir_wrapper, fd, size, 
+                                         off, xdata);
+        } else if (!uuid_compare (fd->inode->gfid,
+                                  priv->base_indices_holder_vgfid)) {
+                stub = fop_readdir_stub (frame, base_indices_readdir_wrapper,
+                                         fd, size, off, xdata);
+        }
+
         if (!stub) {
                 STACK_UNWIND_STRICT (readdir, frame, -1, ENOMEM, NULL, NULL);
                 return 0;
@@ -1075,10 +1361,13 @@ init (xlator_t *this)
         GF_OPTION_INIT ("index-base", priv->index_basepath, path, out);
         uuid_generate (priv->index);
         uuid_generate (priv->xattrop_vgfid);
+        /*base_indices_holder is a directory which contains hard links to
+         * all base indices inside indices/xattrop directory*/
+        uuid_generate (priv->base_indices_holder_vgfid);
         INIT_LIST_HEAD (&priv->callstubs);
 
         this->private = priv;
-        ret = pthread_create (&thread, &w_attr, index_worker, this);
+        ret = gf_thread_create (&thread, &w_attr, index_worker, this);
         if (ret) {
                 gf_log (this->name, GF_LOG_WARNING, "Failed to create "
                         "worker thread, aborting");
diff --git a/xlators/features/index/src/index.h b/xlators/features/index/src/index.h
index 661dcdbc4..d6dcb1c23 100644
--- a/xlators/features/index/src/index.h
+++ b/xlators/features/index/src/index.h
@@ -36,14 +36,28 @@ typedef struct index_fd_ctx {
         DIR *dir;
 } index_fd_ctx_t;
 
+typedef enum {
+        sync_not_started,
+        sync_started,
+        synced_state,
+} to_be_healed_states_t;
+
+typedef enum {
+        INDEX_XATTROP,
+        BASE_INDICES_HOLDER,
+} readdir_directory;
+
 typedef struct index_priv {
         char *index_basepath;
         uuid_t index;
         gf_lock_t lock;
         uuid_t xattrop_vgfid;//virtual gfid of the xattrop index dir
+        uuid_t base_indices_holder_vgfid; //virtual gfid of the
+                                          //to_be_healed_xattrop directory
         struct list_head callstubs;
         pthread_mutex_t mutex;
         pthread_cond_t  cond;
+        to_be_healed_states_t to_be_healed_states;
 } index_priv_t;
 
 #define INDEX_STACK_UNWIND(fop, frame, params ...)      \
diff --git a/xlators/features/locks/src/Makefile.am b/xlators/features/locks/src/Makefile.am
index 8908c1f52..0f79731b4 100644
--- a/xlators/features/locks/src/Makefile.am
+++ b/xlators/features/locks/src/Makefile.am
@@ -11,6 +11,7 @@ noinst_HEADERS = locks.h common.h locks-mem-types.h clear.h
 
 AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
 
+
 AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS)
 
 CLEANFILES = 
diff --git a/xlators/features/locks/src/clear.c b/xlators/features/locks/src/clear.c
index 5790a99ce..124b9ad0f 100644
--- a/xlators/features/locks/src/clear.c
+++ b/xlators/features/locks/src/clear.c
@@ -339,6 +339,7 @@ blkd:
                                    -1, EAGAIN);
                 STACK_UNWIND_STRICT (entrylk, elock->frame, -1, EAGAIN, NULL);
                 GF_FREE ((char *) elock->basename);
+                GF_FREE (elock->connection_id);
                 GF_FREE (elock);
         }
 
@@ -379,8 +380,8 @@ out:
 
 int
 clrlk_clear_lks_in_all_domains (xlator_t *this, pl_inode_t *pl_inode,
-                          clrlk_args *args, int *blkd, int *granted,
-                          int *op_errno)
+                                clrlk_args *args, int *blkd, int *granted,
+                                int *op_errno)
 {
         pl_dom_list_t   *dom            = NULL;
         int             ret             = -1;
diff --git a/xlators/features/locks/src/common.c b/xlators/features/locks/src/common.c
index 9c21bddb9..b3309580d 100644
--- a/xlators/features/locks/src/common.c
+++ b/xlators/features/locks/src/common.c
@@ -35,6 +35,7 @@ __insert_and_merge (pl_inode_t *pl_inode, posix_lock_t *lock);
 static int
 pl_send_prelock_unlock (xlator_t *this, pl_inode_t *pl_inode,
                         posix_lock_t *old_lock);
+
 static pl_dom_list_t *
 __allocate_domain (const char *volume)
 {
@@ -75,8 +76,8 @@ get_domain (pl_inode_t *pl_inode, const char *volume)
 {
         pl_dom_list_t *dom = NULL;
 
-        GF_VALIDATE_OR_GOTO (POSIX_LOCKS, pl_inode, out);
-        GF_VALIDATE_OR_GOTO (POSIX_LOCKS, volume, out);
+        GF_VALIDATE_OR_GOTO ("posix-locks", pl_inode, out);
+        GF_VALIDATE_OR_GOTO ("posix-locks", volume, out);
 
         pthread_mutex_lock (&pl_inode->mutex);
         {
@@ -92,9 +93,9 @@ get_domain (pl_inode_t *pl_inode, const char *volume)
 unlock:
         pthread_mutex_unlock (&pl_inode->mutex);
         if (dom) {
-                gf_log (POSIX_LOCKS, GF_LOG_TRACE, "Domain %s found", volume);
+                gf_log ("posix-locks", GF_LOG_TRACE, "Domain %s found", volume);
         } else {
-                gf_log (POSIX_LOCKS, GF_LOG_TRACE, "Domain %s not found", volume);
+                gf_log ("posix-locks", GF_LOG_TRACE, "Domain %s not found", volume);
         }
 out:
         return dom;
@@ -135,10 +136,10 @@ __pl_inode_is_empty (pl_inode_t *pl_inode)
 void
 pl_print_locker (char *str, int size, xlator_t *this, call_frame_t *frame)
 {
-        snprintf (str, size, "Pid=%llu, lk-owner=%s, Transport=%p, Frame=%llu",
+        snprintf (str, size, "Pid=%llu, lk-owner=%s, Client=%p, Frame=%llu",
                   (unsigned long long) frame->root->pid,
                   lkowner_utoa (&frame->root->lk_owner),
-                  (void *)frame->root->trans,
+                  frame->root->client,
                   (unsigned long long) frame->root->unique);
 }
 
@@ -462,14 +463,14 @@ unlock:
 
 /* Create a new posix_lock_t */
 posix_lock_t *
-new_posix_lock (struct gf_flock *flock, void *transport, pid_t client_pid,
+new_posix_lock (struct gf_flock *flock, client_t *client, pid_t client_pid,
                 gf_lkowner_t *owner, fd_t *fd)
 {
         posix_lock_t *lock = NULL;
 
-        GF_VALIDATE_OR_GOTO (POSIX_LOCKS, flock, out);
-        GF_VALIDATE_OR_GOTO (POSIX_LOCKS, transport, out);
-        GF_VALIDATE_OR_GOTO (POSIX_LOCKS, fd, out);
+        GF_VALIDATE_OR_GOTO ("posix-locks", flock, out);
+        GF_VALIDATE_OR_GOTO ("posix-locks", client, out);
+        GF_VALIDATE_OR_GOTO ("posix-locks", fd, out);
 
         lock = GF_CALLOC (1, sizeof (posix_lock_t),
                           gf_locks_mt_posix_lock_t);
@@ -485,7 +486,7 @@ new_posix_lock (struct gf_flock *flock, void *transport, pid_t client_pid,
         else
                 lock->fl_end = flock->l_start + flock->l_len - 1;
 
-        lock->transport  = transport;
+        lock->client     = client;
         lock->fd_num     = fd_to_fdnum (fd);
         lock->fd         = fd;
         lock->client_pid = client_pid;
@@ -565,7 +566,7 @@ same_owner (posix_lock_t *l1, posix_lock_t *l2)
 {
 
         return (is_same_lkowner (&l1->owner, &l2->owner) &&
-                (l1->transport  == l2->transport));
+                (l1->client == l2->client));
 
 }
 
@@ -694,7 +695,7 @@ subtract_locks (posix_lock_t *big, posix_lock_t *small)
         }
 
         GF_ASSERT (0);
-        gf_log (POSIX_LOCKS, GF_LOG_ERROR, "Unexpected case in subtract_locks");
+        gf_log ("posix-locks", GF_LOG_ERROR, "Unexpected case in subtract_locks");
 
 out:
         if (v.locks[0]) {
@@ -812,7 +813,7 @@ __insert_and_merge (pl_inode_t *pl_inode, posix_lock_t *lock)
                                 sum = add_locks (lock, conf);
 
                                 sum->fl_type    = lock->fl_type;
-                                sum->transport  = lock->transport;
+                                sum->client     = lock->client;
                                 sum->fd_num     = lock->fd_num;
                                 sum->client_pid = lock->client_pid;
                                 sum->owner      = lock->owner;
@@ -830,7 +831,7 @@ __insert_and_merge (pl_inode_t *pl_inode, posix_lock_t *lock)
                                 sum = add_locks (lock, conf);
 
                                 sum->fl_type    = conf->fl_type;
-                                sum->transport  = conf->transport;
+                                sum->client     = conf->client;
                                 sum->fd_num     = conf->fd_num;
                                 sum->client_pid = conf->client_pid;
                                 sum->owner      = conf->owner;
@@ -988,7 +989,7 @@ pl_send_prelock_unlock (xlator_t *this, pl_inode_t *pl_inode,
         flock.l_len    = old_lock->user_flock.l_len;
 
 
-        unlock_lock = new_posix_lock (&flock, old_lock->transport,
+        unlock_lock = new_posix_lock (&flock, old_lock->client,
                                       old_lock->client_pid, &old_lock->owner,
                                       old_lock->fd);
         GF_VALIDATE_OR_GOTO (this->name, unlock_lock, out);
@@ -1097,3 +1098,124 @@ pl_getlk (pl_inode_t *pl_inode, posix_lock_t *lock)
 
         return conf;
 }
+
+
+struct _lock_table *
+pl_lock_table_new (void)
+{
+        struct _lock_table *new = NULL;
+
+        new = GF_CALLOC (1, sizeof (struct _lock_table), gf_common_mt_lock_table);
+        if (new == NULL) {
+                goto out;
+        }
+        INIT_LIST_HEAD (&new->entrylk_lockers);
+        INIT_LIST_HEAD (&new->inodelk_lockers);
+        LOCK_INIT (&new->lock);
+out:
+        return new;
+}
+
+
+int
+pl_add_locker (struct _lock_table *table, const char *volume,
+               loc_t *loc, fd_t *fd, pid_t pid, gf_lkowner_t *owner,
+               glusterfs_fop_t type)
+{
+        int32_t         ret = -1;
+        struct _locker *new = NULL;
+
+        GF_VALIDATE_OR_GOTO ("lock-table", table, out);
+        GF_VALIDATE_OR_GOTO ("lock-table", volume, out);
+
+        new = GF_CALLOC (1, sizeof (struct _locker), gf_common_mt_locker);
+        if (new == NULL) {
+                goto out;
+        }
+        INIT_LIST_HEAD (&new->lockers);
+
+        new->volume = gf_strdup (volume);
+
+        if (fd == NULL) {
+                loc_copy (&new->loc, loc);
+        } else {
+                new->fd = fd_ref (fd);
+        }
+
+        new->pid   = pid;
+        new->owner = *owner;
+
+        LOCK (&table->lock);
+        {
+                if (type == GF_FOP_ENTRYLK)
+                        list_add_tail (&new->lockers, &table->entrylk_lockers);
+                else
+                        list_add_tail (&new->lockers, &table->inodelk_lockers);
+        }
+        UNLOCK (&table->lock);
+out:
+        return ret;
+}
+
+int
+pl_del_locker (struct _lock_table *table, const char *volume,
+               loc_t *loc, fd_t *fd, gf_lkowner_t *owner, glusterfs_fop_t type)
+{
+        struct _locker    *locker = NULL;
+        struct _locker    *tmp = NULL;
+        int32_t            ret = -1;
+        struct list_head  *head = NULL;
+        struct list_head   del;
+
+        GF_VALIDATE_OR_GOTO ("lock-table", table, out);
+        GF_VALIDATE_OR_GOTO ("lock-table", volume, out);
+
+        INIT_LIST_HEAD (&del);
+
+        LOCK (&table->lock);
+        {
+                if (type == GF_FOP_ENTRYLK) {
+                        head = &table->entrylk_lockers;
+                } else {
+                        head = &table->inodelk_lockers;
+                }
+
+                list_for_each_entry_safe (locker, tmp, head, lockers) {
+                        if (!is_same_lkowner (&locker->owner, owner) ||
+                            strcmp (locker->volume, volume))
+                                continue;
+
+                        /*
+                         * It is possible for inodelk lock to come on anon-fd
+                         * and inodelk unlock to come on normal fd in case of
+                         * client re-opens. So don't check for fds to be equal.
+                         */
+                        if (locker->fd && fd)
+                                list_move_tail (&locker->lockers, &del);
+                        else if (locker->loc.inode && loc &&
+                                 (locker->loc.inode == loc->inode))
+                                list_move_tail (&locker->lockers, &del);
+                }
+        }
+        UNLOCK (&table->lock);
+
+        tmp = NULL;
+        locker = NULL;
+
+        list_for_each_entry_safe (locker, tmp, &del, lockers) {
+                list_del_init (&locker->lockers);
+                if (locker->fd)
+                        fd_unref (locker->fd);
+                else
+                        loc_wipe (&locker->loc);
+
+                GF_FREE (locker->volume);
+                GF_FREE (locker);
+        }
+
+        ret = 0;
+out:
+        return ret;
+
+}
+
diff --git a/xlators/features/locks/src/common.h b/xlators/features/locks/src/common.h
index 94ef4b2b4..db19ec978 100644
--- a/xlators/features/locks/src/common.h
+++ b/xlators/features/locks/src/common.h
@@ -14,12 +14,14 @@
 /*dump locks format strings */
 #define RANGE_FMT               "type=%s, whence=%hd, start=%llu, len=%llu"
 #define ENTRY_FMT               "type=%s on basename=%s"
-#define DUMP_GEN_FMT            "pid = %llu, owner=%s, transport=%p, "
+#define DUMP_GEN_FMT            "pid = %llu, owner=%s, client=%p"
 #define GRNTD_AT                "granted at %s"
 #define BLKD_AT                 "blocked at %s"
-#define DUMP_BLKD_FMT           DUMP_GEN_FMT", "BLKD_AT
-#define DUMP_GRNTD_FMT          DUMP_GEN_FMT", "GRNTD_AT
-#define DUMP_BLKD_GRNTD_FMT     DUMP_GEN_FMT", "BLKD_AT", "GRNTD_AT
+#define CONN_ID                 "connection-id=%s"
+#define DUMP_BLKD_FMT           DUMP_GEN_FMT", "CONN_ID", "BLKD_AT
+#define DUMP_GRNTD_FMT          DUMP_GEN_FMT", "CONN_ID", "GRNTD_AT
+#define DUMP_BLKD_GRNTD_FMT     DUMP_GEN_FMT", "CONN_ID", "BLKD_AT", "GRNTD_AT
+
 #define ENTRY_BLKD_FMT          ENTRY_FMT", "DUMP_BLKD_FMT
 #define ENTRY_GRNTD_FMT         ENTRY_FMT", "DUMP_GRNTD_FMT
 #define ENTRY_BLKD_GRNTD_FMT    ENTRY_FMT", "DUMP_BLKD_GRNTD_FMT
@@ -29,8 +31,24 @@
 #define RANGE_BLKD_GRNTD_FMT    RANGE_FMT", "DUMP_BLKD_GRNTD_FMT
 
 #define SET_FLOCK_PID(flock, lock) ((flock)->l_pid = lock->client_pid)
+
+struct _locker {
+        struct list_head  lockers;
+        char             *volume;
+        loc_t             loc;
+        fd_t             *fd;
+        gf_lkowner_t      owner;
+        pid_t             pid;
+};
+
+struct _lock_table {
+        struct list_head  inodelk_lockers;
+        struct list_head  entrylk_lockers;
+        gf_lock_t         lock;
+};
+
 posix_lock_t *
-new_posix_lock (struct gf_flock *flock, void *transport, pid_t client_pid,
+new_posix_lock (struct gf_flock *flock, client_t *client, pid_t client_pid,
                 gf_lkowner_t *owner, fd_t *fd);
 
 pl_inode_t *
@@ -63,7 +81,8 @@ pl_dom_list_t *
 get_domain (pl_inode_t *pl_inode, const char *volume);
 
 void
-grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom);
+grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode,
+                           pl_dom_list_t *dom);
 
 void
 __delete_inode_lock (pl_inode_lock_t *lock);
@@ -78,9 +97,9 @@ grant_blocked_entry_locks (xlator_t *this, pl_inode_t *pl_inode,
 void pl_update_refkeeper (xlator_t *this, inode_t *inode);
 
 int32_t
-__get_inodelk_count (xlator_t *this, pl_inode_t *pl_inode);
+__get_inodelk_count (xlator_t *this, pl_inode_t *pl_inode, char *domname);
 int32_t
-get_inodelk_count (xlator_t *this, inode_t *inode);
+get_inodelk_count (xlator_t *this, inode_t *inode, char *domname);
 
 int32_t
 __get_entrylk_count (xlator_t *this, pl_inode_t *pl_inode);
@@ -143,6 +162,26 @@ pl_verify_reservelk (xlator_t *this, pl_inode_t *pl_inode,
                      posix_lock_t *lock, int can_block);
 int
 pl_reserve_unlock (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *reqlock);
+
 uint32_t
 check_entrylk_on_basename (xlator_t *this, inode_t *parent, char *basename);
+
+int32_t
+pl_add_locker (struct _lock_table *table, const char *volume,
+               loc_t *loc,
+               fd_t *fd,
+               pid_t pid,
+               gf_lkowner_t *owner,
+               glusterfs_fop_t type);
+
+int32_t
+pl_del_locker (struct _lock_table *table, const char *volume,
+               loc_t *loc,
+               fd_t *fd,
+               gf_lkowner_t *owner,
+               glusterfs_fop_t type);
+
+struct _lock_table *
+pl_lock_table_new (void);
+
 #endif /* __COMMON_H__ */
diff --git a/xlators/features/locks/src/entrylk.c b/xlators/features/locks/src/entrylk.c
index d934a8b94..0785dc547 100644
--- a/xlators/features/locks/src/entrylk.c
+++ b/xlators/features/locks/src/entrylk.c
@@ -25,7 +25,7 @@
 
 static pl_entry_lock_t *
 new_entrylk_lock (pl_inode_t *pinode, const char *basename, entrylk_type type,
-                  void *trans, pid_t client_pid, gf_lkowner_t *owner,
+                  client_t *client, pid_t client_pid, gf_lkowner_t *owner,
                   const char *volume)
 
 {
@@ -39,7 +39,7 @@ new_entrylk_lock (pl_inode_t *pinode, const char *basename, entrylk_type type,
 
         newlock->basename   = basename ? gf_strdup (basename) : NULL;
         newlock->type       = type;
-        newlock->trans      = trans;
+        newlock->trans      = client;
         newlock->volume     = volume;
         newlock->client_pid = client_pid;
         newlock->owner      = *owner;
@@ -305,18 +305,15 @@ __find_most_matching_lock (pl_dom_list_t *dom, const char *basename)
 
 int
 __lock_name (pl_inode_t *pinode, const char *basename, entrylk_type type,
-             call_frame_t *frame, pl_dom_list_t *dom, xlator_t *this, int nonblock)
+             call_frame_t *frame, pl_dom_list_t *dom, xlator_t *this,
+             int nonblock, char *conn_id)
 {
         pl_entry_lock_t *lock       = NULL;
         pl_entry_lock_t *conf       = NULL;
-        void            *trans      = NULL;
-        pid_t            client_pid = 0;
         int              ret        = -EINVAL;
 
-        trans = frame->root->trans;
-        client_pid = frame->root->pid;
-
-        lock = new_entrylk_lock (pinode, basename, type, trans, client_pid,
+        lock = new_entrylk_lock (pinode, basename, type,
+                                 frame->root->client, frame->root->pid,
                                  &frame->root->lk_owner, dom->domain);
         if (!lock) {
                 ret = -ENOMEM;
@@ -325,12 +322,17 @@ __lock_name (pl_inode_t *pinode, const char *basename, entrylk_type type,
 
         lock->frame   = frame;
         lock->this    = this;
-        lock->trans   = trans;
+        lock->trans   = frame->root->client;
+
+        if (conn_id) {
+                lock->connection_id = gf_strdup (conn_id);
+        }
 
         conf = __lock_grantable (dom, basename, type);
         if (conf) {
                 ret = -EAGAIN;
                 if (nonblock){
+                        GF_FREE (lock->connection_id);
                         GF_FREE ((char *)lock->basename);
                         GF_FREE (lock);
                         goto out;
@@ -350,6 +352,7 @@ __lock_name (pl_inode_t *pinode, const char *basename, entrylk_type type,
         if ( __blocked_lock_conflict (dom, basename, type) && !(__owner_has_lock (dom, lock))) {
                 ret = -EAGAIN;
                 if (nonblock) {
+                        GF_FREE (lock->connection_id);
                         GF_FREE ((char *) lock->basename);
                         GF_FREE (lock);
                         goto out;
@@ -480,13 +483,15 @@ __grant_blocked_entry_locks (xlator_t *this, pl_inode_t *pl_inode,
                         pl_inode, bl->basename);
 
                 bl_ret = __lock_name (pl_inode, bl->basename, bl->type,
-                                      bl->frame, dom, bl->this, 0);
+                                      bl->frame, dom, bl->this, 0,
+                                      bl->connection_id);
 
                 if (bl_ret == 0) {
                         list_add (&bl->blocked_locks, granted);
                 } else {
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "should never happen");
+                        GF_FREE (bl->connection_id);
                         GF_FREE ((char *)bl->basename);
                         GF_FREE (bl);
                 }
@@ -507,7 +512,8 @@ grant_blocked_entry_locks (xlator_t *this, pl_inode_t *pl_inode,
 
         pthread_mutex_lock (&pl_inode->mutex);
         {
-                __grant_blocked_entry_locks (this, pl_inode, dom, &granted_list);
+                __grant_blocked_entry_locks (this, pl_inode, dom,
+                                             &granted_list);
         }
         pthread_mutex_unlock (&pl_inode->mutex);
 
@@ -520,24 +526,26 @@ grant_blocked_entry_locks (xlator_t *this, pl_inode_t *pl_inode,
 
                 STACK_UNWIND_STRICT (entrylk, lock->frame, 0, 0, NULL);
 
-		GF_FREE ((char *)lock->basename);
-		GF_FREE (lock);
+                GF_FREE (lock->connection_id);
+                GF_FREE ((char *)lock->basename);
+                GF_FREE (lock);
         }
 
         GF_FREE ((char *)unlocked->basename);
+        GF_FREE (unlocked->connection_id);
         GF_FREE (unlocked);
 
         return;
 }
 
 /**
- * release_entry_locks_for_transport: release all entry locks from this
- * transport for this loc_t
+ * release_entry_locks_for_client: release all entry locks from this
+ * client for this loc_t
  */
 
 static int
-release_entry_locks_for_transport (xlator_t *this, pl_inode_t *pinode,
-                                   pl_dom_list_t *dom, void *trans)
+release_entry_locks_for_client (xlator_t *this, pl_inode_t *pinode,
+                                pl_dom_list_t *dom, client_t *client)
 {
         pl_entry_lock_t  *lock = NULL;
         pl_entry_lock_t  *tmp = NULL;
@@ -551,14 +559,14 @@ release_entry_locks_for_transport (xlator_t *this, pl_inode_t *pinode,
         {
                 list_for_each_entry_safe (lock, tmp, &dom->blocked_entrylks,
                                           blocked_locks) {
-                        if (lock->trans != trans)
+                        if (lock->trans != client)
                                 continue;
 
                         list_del_init (&lock->blocked_locks);
 
                         gf_log (this->name, GF_LOG_TRACE,
                                 "releasing lock on  held by "
-                                "{transport=%p}",trans);
+                                "{client=%p}", client);
 
                         list_add (&lock->blocked_locks, &released);
 
@@ -566,16 +574,17 @@ release_entry_locks_for_transport (xlator_t *this, pl_inode_t *pinode,
 
                 list_for_each_entry_safe (lock, tmp, &dom->entrylk_list,
                                           domain_list) {
-                        if (lock->trans != trans)
+                        if (lock->trans != client)
                                 continue;
 
                         list_del_init (&lock->domain_list);
 
                         gf_log (this->name, GF_LOG_TRACE,
                                 "releasing lock on  held by "
-                                "{transport=%p}",trans);
+                                "{client=%p}", client);
 
                         GF_FREE ((char *)lock->basename);
+                        GF_FREE (lock->connection_id);
                         GF_FREE (lock);
                 }
 
@@ -591,6 +600,7 @@ release_entry_locks_for_transport (xlator_t *this, pl_inode_t *pinode,
                 STACK_UNWIND_STRICT (entrylk, lock->frame, -1, EAGAIN, NULL);
 
                 GF_FREE ((char *)lock->basename);
+                GF_FREE (lock->connection_id);
                 GF_FREE (lock);
 
         }
@@ -601,6 +611,7 @@ release_entry_locks_for_transport (xlator_t *this, pl_inode_t *pinode,
                 STACK_UNWIND_STRICT (entrylk, lock->frame, 0, 0, NULL);
 
                 GF_FREE ((char *)lock->basename);
+                GF_FREE (lock->connection_id);
                 GF_FREE (lock);
         }
 
@@ -611,19 +622,23 @@ release_entry_locks_for_transport (xlator_t *this, pl_inode_t *pinode,
 int
 pl_common_entrylk (call_frame_t *frame, xlator_t *this,
                    const char *volume, inode_t *inode, const char *basename,
-                   entrylk_cmd cmd, entrylk_type type, loc_t *loc, fd_t *fd)
-{
-        int32_t  op_ret   = -1;
-        int32_t  op_errno = 0;
+                   entrylk_cmd cmd, entrylk_type type, loc_t *loc, fd_t *fd,
+                   dict_t *xdata)
 
-        void *        transport = NULL;
-
-        pl_inode_t *     pinode   = NULL;
+{
+        int32_t          op_ret   = -1;
+        int32_t          op_errno = 0;
         int              ret      = -1;
-        pl_entry_lock_t *unlocked = NULL;
         char             unwind   = 1;
+        GF_UNUSED int    dict_ret = -1;
+        pl_inode_t      *pinode   = NULL;
+        pl_entry_lock_t *unlocked = NULL;
+        pl_dom_list_t   *dom      = NULL;
+        char            *conn_id  = NULL;
+        pl_ctx_t        *ctx      = NULL;
 
-        pl_dom_list_t          *dom = NULL;
+        if (xdata)
+                dict_ret = dict_get_str (xdata, "connection-id", &conn_id);
 
         pinode = pl_inode_get (this, inode);
         if (!pinode) {
@@ -639,18 +654,17 @@ pl_common_entrylk (call_frame_t *frame, xlator_t *this,
 
         entrylk_trace_in (this, frame, volume, fd, loc, basename, cmd, type);
 
-        transport = frame->root->trans;
-
         if (frame->root->lk_owner.len == 0) {
                 /*
                   this is a special case that means release
-                  all locks from this transport
+                  all locks from this client
                 */
 
                 gf_log (this->name, GF_LOG_TRACE,
-                        "Releasing locks for transport %p", transport);
+                        "Releasing locks for client %p", frame->root->client);
 
-                release_entry_locks_for_transport (this, pinode, dom, transport);
+                release_entry_locks_for_client (this, pinode, dom,
+                                                frame->root->client);
                 op_ret = 0;
 
                 goto out;
@@ -661,7 +675,7 @@ pl_common_entrylk (call_frame_t *frame, xlator_t *this,
                 pthread_mutex_lock (&pinode->mutex);
                 {
                         ret = __lock_name (pinode, basename, type,
-                                           frame, dom, this, 0);
+                                           frame, dom, this, 0, conn_id);
                 }
                 pthread_mutex_unlock (&pinode->mutex);
 
@@ -686,7 +700,7 @@ pl_common_entrylk (call_frame_t *frame, xlator_t *this,
                 pthread_mutex_lock (&pinode->mutex);
                 {
                         ret = __lock_name (pinode, basename, type,
-                                           frame, dom, this, 1);
+                                           frame, dom, this, 1, conn_id);
                 }
                 pthread_mutex_unlock (&pinode->mutex);
 
@@ -723,6 +737,24 @@ out:
                 entrylk_trace_out (this, frame, volume, fd, loc, basename,
                                    cmd, type, op_ret, op_errno);
 
+                ctx = pl_ctx_get (frame->root->client, this);
+
+                if (ctx == NULL) {
+                        gf_log (this->name, GF_LOG_INFO, "pl_ctx_get() failed");
+                        goto unwind;
+                }
+
+                if (cmd == ENTRYLK_UNLOCK)
+                        pl_del_locker (ctx->ltable, volume, loc, fd,
+                                       &frame->root->lk_owner,
+                                       GF_FOP_ENTRYLK);
+                else
+                        pl_add_locker (ctx->ltable, volume, loc, fd,
+                                       frame->root->pid,
+                                       &frame->root->lk_owner,
+                                       GF_FOP_ENTRYLK);
+
+unwind:
                 STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno, NULL);
         } else {
                 entrylk_trace_block (this, frame, volume, fd, loc, basename,
@@ -742,10 +774,10 @@ out:
 int
 pl_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)
 {
-
-        pl_common_entrylk (frame, this, volume, loc->inode, basename, cmd, type, loc, NULL);
+        pl_common_entrylk (frame, this, volume, loc->inode, basename, cmd,
+                           type, loc, NULL, xdata);
 
         return 0;
 }
@@ -760,10 +792,10 @@ pl_entrylk (call_frame_t *frame, xlator_t *this,
 int
 pl_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)
 {
-
-        pl_common_entrylk (frame, this, volume, fd->inode, basename, cmd, type, NULL, fd);
+        pl_common_entrylk (frame, this, volume, fd->inode, basename, cmd,
+                           type, NULL, fd, xdata);
 
         return 0;
 }
diff --git a/xlators/features/locks/src/inodelk.c b/xlators/features/locks/src/inodelk.c
index 42350e59a..508523e11 100644
--- a/xlators/features/locks/src/inodelk.c
+++ b/xlators/features/locks/src/inodelk.c
@@ -39,8 +39,10 @@ inline void
 __pl_inodelk_unref (pl_inode_lock_t *lock)
 {
         lock->ref--;
-        if (!lock->ref)
+        if (!lock->ref) {
+                GF_FREE (lock->connection_id);
                 GF_FREE (lock);
+        }
 }
 
 /* Check if 2 inodelks are conflicting on type. Only 2 shared locks don't conflict */
@@ -122,7 +124,7 @@ static inline int
 same_inodelk_owner (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
 {
         return (is_same_lkowner (&l1->owner, &l2->owner) &&
-                (l1->transport  == l2->transport));
+                (l1->client == l2->client));
 }
 
 /* Returns true if the 2 inodelks conflict with each other */
@@ -292,7 +294,7 @@ __inode_unlock_lock (xlator_t *this, pl_inode_lock_t *lock, pl_dom_list_t *dom)
                         " Matching lock not found for unlock %llu-%llu, by %s "
                         "on %p", (unsigned long long)lock->fl_start,
                         (unsigned long long)lock->fl_end,
-                        lkowner_utoa (&lock->owner), lock->transport);
+                        lkowner_utoa (&lock->owner), lock->client);
                 goto out;
         }
         __delete_inode_lock (conf);
@@ -300,7 +302,7 @@ __inode_unlock_lock (xlator_t *this, pl_inode_lock_t *lock, pl_dom_list_t *dom)
                 " Matching lock found for unlock %llu-%llu, by %s on %p",
                 (unsigned long long)lock->fl_start,
                 (unsigned long long)lock->fl_end, lkowner_utoa (&lock->owner),
-                lock->transport);
+                lock->client);
 
 out:
         return conf;
@@ -333,7 +335,8 @@ __grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode,
 
 /* Grant all inodelks blocked on a lock */
 void
-grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom)
+grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode,
+                           pl_dom_list_t *dom)
 {
         struct list_head granted;
         pl_inode_lock_t *lock;
@@ -372,10 +375,10 @@ grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *
         pthread_mutex_unlock (&pl_inode->mutex);
 }
 
-/* Release all inodelks from this transport */
+/* Release all inodelks from this client */
 static int
-release_inode_locks_of_transport (xlator_t *this, pl_dom_list_t *dom,
-                                  inode_t *inode, void *trans)
+release_inode_locks_of_client (xlator_t *this, pl_dom_list_t *dom,
+                                  inode_t *inode, client_t *client)
 {
         pl_inode_lock_t *tmp = NULL;
         pl_inode_lock_t *l = NULL;
@@ -395,7 +398,7 @@ release_inode_locks_of_transport (xlator_t *this, pl_dom_list_t *dom,
         {
 
                 list_for_each_entry_safe (l, tmp, &dom->blocked_inodelks, blocked_locks) {
-                        if (l->transport != trans)
+                        if (l->client != client)
                                 continue;
 
                         list_del_init (&l->blocked_locks);
@@ -408,8 +411,8 @@ release_inode_locks_of_transport (xlator_t *this, pl_dom_list_t *dom,
 
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "releasing blocking lock on %s held by "
-                                "{transport=%p, pid=%"PRId64" lk-owner=%s}",
-                                file, trans, (uint64_t) l->client_pid,
+                                "{client=%p, pid=%"PRId64" lk-owner=%s}",
+                                file, client, (uint64_t) l->client_pid,
                                 lkowner_utoa (&l->owner));
 
                         list_add (&l->blocked_locks, &released);
@@ -420,7 +423,7 @@ release_inode_locks_of_transport (xlator_t *this, pl_dom_list_t *dom,
                 }
 
                 list_for_each_entry_safe (l, tmp, &dom->inodelk_list, list) {
-                        if (l->transport != trans)
+                        if (l->client != client)
                                 continue;
 
                         inode_path (inode, NULL, &path);
@@ -431,8 +434,8 @@ release_inode_locks_of_transport (xlator_t *this, pl_dom_list_t *dom,
 
                         gf_log (this->name, GF_LOG_DEBUG,
                                 "releasing granted lock on %s held by "
-                                "{transport=%p, pid=%"PRId64" lk-owner=%s}",
-                                file, trans, (uint64_t) l->client_pid,
+                                "{client=%p, pid=%"PRId64" lk-owner=%s}",
+                                file, client, (uint64_t) l->client_pid,
                                 lkowner_utoa (&l->owner));
 
                         if (path) {
@@ -515,8 +518,9 @@ out:
 
 /* Create a new inode_lock_t */
 pl_inode_lock_t *
-new_inode_lock (struct gf_flock *flock, void *transport, pid_t client_pid,
-                gf_lkowner_t *owner, const char *volume)
+new_inode_lock (struct gf_flock *flock, client_t *client, pid_t client_pid,
+                call_frame_t *frame, xlator_t *this, const char *volume,
+                char *conn_id)
 
 {
         pl_inode_lock_t *lock = NULL;
@@ -535,10 +539,16 @@ new_inode_lock (struct gf_flock *flock, void *transport, pid_t client_pid,
         else
                 lock->fl_end = flock->l_start + flock->l_len - 1;
 
-        lock->transport  = transport;
+        lock->client     = client;
         lock->client_pid = client_pid;
         lock->volume     = volume;
-        lock->owner      = *owner;
+        lock->owner      = frame->root->lk_owner;
+        lock->frame      = frame;
+        lock->this       = this;
+
+        if (conn_id) {
+                lock->connection_id = gf_strdup (conn_id);
+        }
 
         INIT_LIST_HEAD (&lock->list);
         INIT_LIST_HEAD (&lock->blocked_locks);
@@ -547,21 +557,58 @@ new_inode_lock (struct gf_flock *flock, void *transport, pid_t client_pid,
         return lock;
 }
 
+int32_t
+_pl_convert_volume (const char *volume, char **res)
+{
+        char    *mdata_vol = NULL;
+        int     ret = 0;
+
+        mdata_vol = strrchr (volume, ':');
+        //if the volume already ends with :metadata don't bother
+        if (mdata_vol && (strcmp (mdata_vol, ":metadata") == 0))
+                return 0;
+
+        ret = gf_asprintf (res, "%s:metadata", volume);
+        if (ret <= 0)
+                return ENOMEM;
+        return 0;
+}
+
+int32_t
+_pl_convert_volume_for_special_range (struct gf_flock *flock,
+                                      const char *volume, char **res)
+{
+        int32_t     ret = 0;
+
+        if ((flock->l_start == LLONG_MAX -1) &&
+            (flock->l_len == 0)) {
+                ret = _pl_convert_volume (volume, res);
+        }
+
+        return ret;
+}
+
 /* Common inodelk code called from pl_inodelk and pl_finodelk */
 int
 pl_common_inodelk (call_frame_t *frame, xlator_t *this,
                    const char *volume, inode_t *inode, int32_t cmd,
-                   struct gf_flock *flock, loc_t *loc, fd_t *fd)
+                   struct gf_flock *flock, loc_t *loc, fd_t *fd, dict_t *xdata)
 {
         int32_t           op_ret     = -1;
         int32_t           op_errno   = 0;
         int               ret        = -1;
+        GF_UNUSED int     dict_ret   = -1;
         int               can_block  = 0;
-        pid_t             client_pid = -1;
-        void *            transport  = NULL;
         pl_inode_t *      pinode     = NULL;
         pl_inode_lock_t * reqlock    = NULL;
         pl_dom_list_t *   dom        = NULL;
+        char             *res        = NULL;
+        char             *res1       = NULL;
+        char             *conn_id    = NULL;
+        pl_ctx_t         *ctx        = NULL;
+
+        if (xdata)
+                dict_ret = dict_get_str (xdata, "connection-id", &conn_id);
 
         VALIDATE_OR_GOTO (frame, out);
         VALIDATE_OR_GOTO (inode, unwind);
@@ -572,10 +619,13 @@ pl_common_inodelk (call_frame_t *frame, xlator_t *this,
                 goto unwind;
         }
 
-        pl_trace_in (this, frame, fd, loc, cmd, flock, volume);
+        op_errno = _pl_convert_volume_for_special_range (flock, volume, &res);
+        if (op_errno)
+                goto unwind;
+        if (res)
+                volume = res;
 
-        transport  = frame->root->trans;
-        client_pid = frame->root->pid;
+        pl_trace_in (this, frame, fd, loc, cmd, flock, volume);
 
         pinode = pl_inode_get (this, inode);
         if (!pinode) {
@@ -592,19 +642,26 @@ pl_common_inodelk (call_frame_t *frame, xlator_t *this,
         if (frame->root->lk_owner.len == 0) {
                 /*
                   special case: this means release all locks
-                  from this transport
+                  from this client
                 */
                 gf_log (this->name, GF_LOG_TRACE,
-                        "Releasing all locks from transport %p", transport);
-
-                release_inode_locks_of_transport (this, dom, inode, transport);
+                        "Releasing all locks from client %p", frame->root->client);
+
+                release_inode_locks_of_client (this, dom, inode, frame->root->client);
+                _pl_convert_volume (volume, &res1);
+                if (res1) {
+                        dom = get_domain (pinode, res1);
+                        if (dom)
+                                release_inode_locks_of_client (this, dom,
+                                                        inode, frame->root->client);
+                }
 
                 op_ret = 0;
                 goto unwind;
         }
 
-        reqlock = new_inode_lock (flock, transport, client_pid,
-                                  &frame->root->lk_owner, volume);
+        reqlock = new_inode_lock (flock, frame->root->client, frame->root->pid,
+                                  frame, this, volume, conn_id);
 
         if (!reqlock) {
                 op_ret = -1;
@@ -612,14 +669,10 @@ pl_common_inodelk (call_frame_t *frame, xlator_t *this,
                 goto unwind;
         }
 
-        reqlock->frame = frame;
-        reqlock->this  = this;
 
         switch (cmd) {
         case F_SETLKW:
                 can_block = 1;
-                reqlock->frame = frame;
-                reqlock->this  = this;
 
                 /* fall through */
 
@@ -651,6 +704,23 @@ pl_common_inodelk (call_frame_t *frame, xlator_t *this,
 
         op_ret = 0;
 
+        ctx = pl_ctx_get (frame->root->client, this);
+
+        if (ctx == NULL) {
+                gf_log (this->name, GF_LOG_INFO, "pl_ctx_get() failed");
+                goto unwind;
+        }
+
+        if (flock->l_type == F_UNLCK)
+                pl_del_locker (ctx->ltable, volume, loc, fd,
+                               &frame->root->lk_owner,
+                               GF_FOP_INODELK);
+        else
+                pl_add_locker (ctx->ltable, volume, loc, fd,
+                               frame->root->pid,
+                               &frame->root->lk_owner,
+                               GF_FOP_INODELK);
+
 unwind:
         if ((inode != NULL) && (flock !=NULL)) {
                 pl_update_refkeeper (this, inode);
@@ -659,53 +729,78 @@ unwind:
 
         STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno, NULL);
 out:
+        GF_FREE (res);
+        GF_FREE (res1);
         return 0;
 }
 
 int
 pl_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)
 {
-
-        pl_common_inodelk (frame, this, volume, loc->inode, cmd, flock, loc, NULL);
+        pl_common_inodelk (frame, this, volume, loc->inode, cmd, flock,
+                           loc, NULL, xdata);
 
         return 0;
 }
 
 int
 pl_finodelk (call_frame_t *frame, xlator_t *this,
-             const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock)
+             const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock,
+             dict_t *xdata)
 {
-
-        pl_common_inodelk (frame, this, volume, fd->inode, cmd, flock, NULL, fd);
+        pl_common_inodelk (frame, this, volume, fd->inode, cmd, flock,
+                           NULL, fd, xdata);
 
         return 0;
 
 }
 
+static inline int32_t
+__get_inodelk_dom_count (pl_dom_list_t *dom)
+{
+        pl_inode_lock_t     *lock   = NULL;
+        int32_t             count   = 0;
+
+        list_for_each_entry (lock, &dom->inodelk_list, list) {
+                count++;
+        }
+        list_for_each_entry (lock, &dom->blocked_inodelks, blocked_locks) {
+                count++;
+        }
+        return count;
+}
 
+/* Returns the no. of locks (blocked/granted) held on a given domain name
+ * If @domname is NULL, returns the no. of locks in all the domains present.
+ * If @domname is non-NULL and non-existent, returns 0 */
 int32_t
-__get_inodelk_count (xlator_t *this, pl_inode_t *pl_inode)
+__get_inodelk_count (xlator_t *this, pl_inode_t *pl_inode, char *domname)
 {
         int32_t            count  = 0;
-        pl_inode_lock_t   *lock   = NULL;
         pl_dom_list_t     *dom    = NULL;
 
         list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
-                list_for_each_entry (lock, &dom->inodelk_list, list) {
-                        count++;
-                }
-                list_for_each_entry (lock, &dom->blocked_inodelks, blocked_locks) {
-                        count++;
-                }
+                if (domname) {
+                        if (strcmp (domname, dom->domain) == 0) {
+                                count = __get_inodelk_dom_count (dom);
+                                goto out;
+                        }
+
+                } else {
+                    /* Counting locks from all domains */
+                        count += __get_inodelk_dom_count (dom);
 
+                }
         }
 
+out:
         return count;
 }
 
 int32_t
-get_inodelk_count (xlator_t *this, inode_t *inode)
+get_inodelk_count (xlator_t *this, inode_t *inode, char *domname)
 {
         pl_inode_t   *pl_inode = NULL;
         uint64_t      tmp_pl_inode = 0;
@@ -721,7 +816,7 @@ get_inodelk_count (xlator_t *this, inode_t *inode)
 
         pthread_mutex_lock (&pl_inode->mutex);
         {
-                count = __get_inodelk_count (this, pl_inode);
+                count = __get_inodelk_count (this, pl_inode, domname);
         }
         pthread_mutex_unlock (&pl_inode->mutex);
 
diff --git a/xlators/features/locks/src/locks.h b/xlators/features/locks/src/locks.h
index 7ffc67e1b..76fc941d7 100644
--- a/xlators/features/locks/src/locks.h
+++ b/xlators/features/locks/src/locks.h
@@ -19,10 +19,10 @@
 #include "stack.h"
 #include "call-stub.h"
 #include "locks-mem-types.h"
+#include "client_t.h"
 
 #include "lkowner.h"
 
-#define POSIX_LOCKS "posix-locks"
 struct __pl_fd;
 
 struct __posix_lock {
@@ -33,7 +33,7 @@ struct __posix_lock {
         off_t              fl_end;
 
         short              blocked;    /* waiting to acquire */
-        struct gf_flock       user_flock; /* the flock supplied by the user */
+        struct gf_flock    user_flock; /* the flock supplied by the user */
         xlator_t          *this;       /* required for blocked locks */
         unsigned long      fd_num;
 
@@ -46,7 +46,7 @@ struct __posix_lock {
         /* These two together serve to uniquely identify each process
            across nodes */
 
-        void              *transport;     /* to identify client node */
+        void              *client;     /* to identify client node */
         gf_lkowner_t       owner;
         pid_t              client_pid;    /* pid of client process */
 };
@@ -63,7 +63,7 @@ struct __pl_inode_lock {
 
         const char        *volume;
 
-        struct gf_flock       user_flock; /* the flock supplied by the user */
+        struct gf_flock    user_flock; /* the flock supplied by the user */
         xlator_t          *this;       /* required for blocked locks */
         fd_t              *fd;
 
@@ -75,9 +75,11 @@ struct __pl_inode_lock {
         /* These two together serve to uniquely identify each process
            across nodes */
 
-        void              *transport;     /* to identify client node */
+        void              *client;     /* to identify client node */
         gf_lkowner_t       owner;
         pid_t              client_pid;    /* pid of client process */
+
+        char              *connection_id; /* stores the client connection id */
 };
 typedef struct __pl_inode_lock pl_inode_lock_t;
 
@@ -115,7 +117,9 @@ struct __entry_lock {
 
         void             *trans;
         gf_lkowner_t      owner;
-        pid_t             client_pid;    /* pid of client process */
+	pid_t             client_pid;    /* pid of client process */
+
+        char             *connection_id; /* stores the client connection id */
 };
 typedef struct __entry_lock pl_entry_lock_t;
 
@@ -152,9 +156,11 @@ typedef struct {
         char           *brickname;
 } posix_locks_private_t;
 
+
 typedef struct {
         gf_boolean_t   entrylk_count_req;
         gf_boolean_t   inodelk_count_req;
+        gf_boolean_t   inodelk_dom_count_req;
         gf_boolean_t   posixlk_count_req;
         gf_boolean_t   parent_entrylk_req;
 
@@ -166,8 +172,21 @@ typedef struct {
         enum {TRUNCATE, FTRUNCATE} op;
 } pl_local_t;
 
+
 typedef struct {
         struct list_head locks_list;
 } pl_fdctx_t;
 
+
+typedef struct _locks_ctx {
+        gf_lock_t            ltable_lock; /* only for replace, 
+                                             ltable has its own internal 
+                                             lock for operations */
+        struct _lock_table  *ltable;
+} pl_ctx_t;
+
+
+pl_ctx_t *
+pl_ctx_get (client_t *client, xlator_t *xlator);
+
 #endif /* __POSIX_LOCKS_H__ */
diff --git a/xlators/features/locks/src/posix.c b/xlators/features/locks/src/posix.c
index 2bc5f8581..7bfb38a51 100644
--- a/xlators/features/locks/src/posix.c
+++ b/xlators/features/locks/src/posix.c
@@ -51,7 +51,7 @@ pl_new_fdctx ()
 
         fdctx = GF_CALLOC (1, sizeof (*fdctx),
                            gf_locks_mt_pl_fdctx_t);
-        GF_VALIDATE_OR_GOTO (POSIX_LOCKS, fdctx, out);
+        GF_VALIDATE_OR_GOTO ("posix-locks", fdctx, out);
 
         INIT_LIST_HEAD (&fdctx->locks_list);
 
@@ -66,7 +66,7 @@ pl_check_n_create_fdctx (xlator_t *this, fd_t *fd)
         uint64_t    tmp   = 0;
         pl_fdctx_t *fdctx = NULL;
 
-        GF_VALIDATE_OR_GOTO (POSIX_LOCKS, this, out);
+        GF_VALIDATE_OR_GOTO ("posix-locks", this, out);
         GF_VALIDATE_OR_GOTO (this->name, fd, out);
 
         LOCK (&fd->lock);
@@ -119,7 +119,7 @@ pl_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 static int
 truncate_allowed (pl_inode_t *pl_inode,
-                  void *transport, pid_t client_pid,
+                  client_t *client, pid_t client_pid,
                   gf_lkowner_t *owner, off_t offset)
 {
         posix_lock_t *l = NULL;
@@ -128,7 +128,7 @@ truncate_allowed (pl_inode_t *pl_inode,
 
         region.fl_start   = offset;
         region.fl_end     = LLONG_MAX;
-        region.transport  = transport;
+        region.client     = client;
         region.client_pid = client_pid;
         region.owner      = *owner;
 
@@ -139,7 +139,7 @@ truncate_allowed (pl_inode_t *pl_inode,
                             && locks_overlap (&region, l)
                             && !same_owner (&region, l)) {
                                 ret = 0;
-                                gf_log (POSIX_LOCKS, GF_LOG_TRACE, "Truncate "
+                                gf_log ("posix-locks", GF_LOG_TRACE, "Truncate "
                                         "allowed");
                                 break;
                         }
@@ -186,7 +186,7 @@ truncate_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         if (priv->mandatory
             && pl_inode->mandatory
-            && !truncate_allowed (pl_inode, frame->root->trans,
+            && !truncate_allowed (pl_inode, frame->root->client,
                                   frame->root->pid, &frame->root->lk_owner,
                                   local->offset)) {
                 op_ret   = -1;
@@ -347,7 +347,7 @@ delete_locks_of_fd (xlator_t *this, pl_inode_t *pl_inode, fd_t *fd)
 
 static void
 __delete_locks_of_owner (pl_inode_t *pl_inode,
-                         void *transport, gf_lkowner_t *owner)
+                         client_t *client, gf_lkowner_t *owner)
 {
         posix_lock_t *tmp = NULL;
         posix_lock_t *l = NULL;
@@ -357,7 +357,7 @@ __delete_locks_of_owner (pl_inode_t *pl_inode,
         list_for_each_entry_safe (l, tmp, &pl_inode->ext_list, list) {
                 if (l->blocked)
                         continue;
-                if ((l->transport == transport) &&
+                if ((l->client == client) &&
                     is_same_lkowner (&l->owner, owner)) {
                         gf_log ("posix-locks", GF_LOG_TRACE,
                                 " Flushing lock"
@@ -810,7 +810,7 @@ pl_migrate_locks (call_frame_t *frame, fd_t *newfd, uint64_t oldfd_num,
                 list_for_each_entry (l, &pl_inode->ext_list, list) {
                         if (l->fd_num == oldfd_num) {
                                 l->fd_num = newfd_num;
-                                l->transport = frame->root->trans;
+                                l->client = frame->root->client;
                         }
                 }
         }
@@ -983,7 +983,7 @@ pl_flush (call_frame_t *frame, xlator_t *this,
         }
         pthread_mutex_lock (&pl_inode->mutex);
         {
-                __delete_locks_of_owner (pl_inode, frame->root->trans,
+                __delete_locks_of_owner (pl_inode, frame->root->client,
                                          &frame->root->lk_owner);
         }
         pthread_mutex_unlock (&pl_inode->mutex);
@@ -1178,7 +1178,7 @@ pl_readv (call_frame_t *frame, xlator_t *this,
         if (priv->mandatory && pl_inode->mandatory) {
                 region.fl_start   = offset;
                 region.fl_end     = offset + size - 1;
-                region.transport  = frame->root->trans;
+                region.client     = frame->root->client;
                 region.fd_num     = fd_to_fdnum(fd);
                 region.client_pid = frame->root->pid;
                 region.owner      = frame->root->lk_owner;
@@ -1272,7 +1272,7 @@ pl_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
         if (priv->mandatory && pl_inode->mandatory) {
                 region.fl_start   = offset;
                 region.fl_end     = offset + iov_length (vector, count) - 1;
-                region.transport  = frame->root->trans;
+                region.client     = frame->root->client;
                 region.fd_num     = fd_to_fdnum(fd);
                 region.client_pid = frame->root->pid;
                 region.owner      = frame->root->lk_owner;
@@ -1353,7 +1353,7 @@ lock_dup (posix_lock_t *lock)
 {
         posix_lock_t *new_lock = NULL;
 
-        new_lock = new_posix_lock (&lock->user_flock, lock->transport,
+        new_lock = new_posix_lock (&lock->user_flock, lock->client,
                                    lock->client_pid, &lock->owner,
                                    (fd_t *)lock->fd_num);
         return new_lock;
@@ -1513,8 +1513,6 @@ int
 pl_lk (call_frame_t *frame, xlator_t *this,
        fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
-        void         *transport  = NULL;
-        pid_t         client_pid = 0;
         pl_inode_t   *pl_inode   = NULL;
         int           op_ret     = 0;
         int           op_errno   = 0;
@@ -1523,9 +1521,6 @@ pl_lk (call_frame_t *frame, xlator_t *this,
         posix_lock_t *conf       = NULL;
         int           ret        = 0;
 
-        transport  = frame->root->trans;
-        client_pid = frame->root->pid;
-
         if ((flock->l_start < 0) || (flock->l_len < 0)) {
                 op_ret = -1;
                 op_errno = EINVAL;
@@ -1539,7 +1534,7 @@ pl_lk (call_frame_t *frame, xlator_t *this,
                 goto unwind;
         }
 
-        reqlock = new_posix_lock (flock, transport, client_pid,
+        reqlock = new_posix_lock (flock, frame->root->client, frame->root->pid,
                                   &frame->root->lk_owner, fd);
 
         if (!reqlock) {
@@ -1764,6 +1759,7 @@ pl_forget (xlator_t *this,
                                         list_del_init (&entry_l->domain_list);
 
                                         GF_FREE ((char *)entry_l->basename);
+                                        GF_FREE (entry_l->connection_id);
                                         GF_FREE (entry_l);
                                 }
 
@@ -1797,6 +1793,7 @@ pl_forget (xlator_t *this,
 
                 STACK_UNWIND_STRICT (entrylk, entry_l->frame, -1, 0, NULL);
                 GF_FREE ((char *)entry_l->basename);
+                GF_FREE (entry_l->connection_id);
                 GF_FREE (entry_l);
 
         }
@@ -1946,19 +1943,34 @@ pl_entrylk_xattr_fill (xlator_t *this, inode_t *inode,
 }
 
 void
-pl_inodelk_xattr_fill (xlator_t *this, inode_t *inode,
-                       dict_t *dict)
+pl_inodelk_xattr_fill (xlator_t *this, inode_t *inode, dict_t *dict,
+                       gf_boolean_t per_dom)
 {
         int32_t     count = 0;
         int         ret   = -1;
+        char        *domname = NULL;
+
+
+        if (per_dom){
+                ret = dict_get_str (dict, GLUSTERFS_INODELK_DOM_COUNT,
+                                    &domname);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to get "
+                                "value for key %s",GLUSTERFS_INODELK_DOM_COUNT);
+                        goto out;
+                }
+        }
+
+        count = get_inodelk_count (this, inode, domname);
 
-        count = get_inodelk_count (this, inode);
         ret = dict_set_int32 (dict, GLUSTERFS_INODELK_COUNT, count);
         if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        " dict_set failed on key %s", GLUSTERFS_INODELK_COUNT);
+                gf_log (this->name, GF_LOG_DEBUG, "Failed to set count for "
+                        "key %s", GLUSTERFS_INODELK_COUNT);
         }
 
+out:
+        return;
 }
 
 void
@@ -2003,7 +2015,9 @@ pl_lookup_cbk (call_frame_t *frame,
         if (local->entrylk_count_req)
                 pl_entrylk_xattr_fill (this, inode, xdata);
         if (local->inodelk_count_req)
-                pl_inodelk_xattr_fill (this, inode, xdata);
+                pl_inodelk_xattr_fill (this, inode, xdata, _gf_false);
+        if (local->inodelk_dom_count_req)
+                pl_inodelk_xattr_fill (this, inode, xdata, _gf_true);
         if (local->posixlk_count_req)
                 pl_posixlk_xattr_fill (this, inode, xdata);
 
@@ -2050,6 +2064,8 @@ pl_lookup (call_frame_t *frame,
                         local->entrylk_count_req = 1;
                 if (dict_get (xdata, GLUSTERFS_INODELK_COUNT))
                         local->inodelk_count_req = 1;
+                if (dict_get (xdata, GLUSTERFS_INODELK_DOM_COUNT))
+                        local->inodelk_dom_count_req = 1;
                 if (dict_get (xdata, GLUSTERFS_POSIXLK_COUNT))
                         local->posixlk_count_req = 1;
                 if (dict_get (xdata, GLUSTERFS_PARENT_ENTRYLK))
@@ -2088,7 +2104,11 @@ pl_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 if (local->entrylk_count_req)
                         pl_entrylk_xattr_fill (this, entry->inode, entry->dict);
                 if (local->inodelk_count_req)
-                        pl_inodelk_xattr_fill (this, entry->inode, entry->dict);
+                        pl_inodelk_xattr_fill (this, entry->inode, entry->dict,
+                                               _gf_false);
+                if (local->inodelk_dom_count_req)
+                        pl_inodelk_xattr_fill (this, entry->inode, entry->dict,
+                                               _gf_true);
                 if (local->posixlk_count_req)
                         pl_posixlk_xattr_fill (this, entry->inode, entry->dict);
         }
@@ -2117,6 +2137,8 @@ pl_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
                         local->entrylk_count_req = 1;
                 if (dict_get (dict, GLUSTERFS_INODELK_COUNT))
                         local->inodelk_count_req = 1;
+                if (dict_get (dict, GLUSTERFS_INODELK_DOM_COUNT))
+                        local->inodelk_dom_count_req = 1;
                 if (dict_get (dict, GLUSTERFS_POSIXLK_COUNT))
                         local->posixlk_count_req = 1;
         }
@@ -2136,8 +2158,8 @@ out:
 
 void
 pl_dump_lock (char *str, int size, struct gf_flock *flock,
-              gf_lkowner_t *owner, void *trans, time_t *granted_time,
-              time_t *blkd_time, gf_boolean_t active)
+              gf_lkowner_t *owner, void *trans, char *conn_id,
+              time_t *granted_time, time_t *blkd_time, gf_boolean_t active)
 {
         char  *type_str    = NULL;
         char   granted[32] = {0,};
@@ -2165,7 +2187,7 @@ pl_dump_lock (char *str, int size, struct gf_flock *flock,
                                   (unsigned long long) flock->l_start,
                                   (unsigned long long) flock->l_len,
                                   (unsigned long long) flock->l_pid,
-                                  lkowner_utoa (owner), trans,
+                                  lkowner_utoa (owner), trans, conn_id,
                                   ctime_r (granted_time, granted));
                 } else {
                         snprintf (str, size, RANGE_BLKD_GRNTD_FMT,
@@ -2173,7 +2195,7 @@ pl_dump_lock (char *str, int size, struct gf_flock *flock,
                                   (unsigned long long) flock->l_start,
                                   (unsigned long long) flock->l_len,
                                   (unsigned long long) flock->l_pid,
-                                  lkowner_utoa (owner), trans,
+                                  lkowner_utoa (owner), trans, conn_id,
                                   ctime_r (blkd_time, blocked),
                                   ctime_r (granted_time, granted));
                 }
@@ -2184,7 +2206,7 @@ pl_dump_lock (char *str, int size, struct gf_flock *flock,
                           (unsigned long long) flock->l_start,
                           (unsigned long long) flock->l_len,
                           (unsigned long long) flock->l_pid,
-                          lkowner_utoa (owner), trans,
+                          lkowner_utoa (owner), trans, conn_id,
                           ctime_r (blkd_time, blocked));
         }
 
@@ -2222,6 +2244,7 @@ __dump_entrylks (pl_inode_t *pl_inode)
                                           "ENTRYLK_WRLCK", lock->basename,
                                           (unsigned long long) lock->client_pid,
                                           lkowner_utoa (&lock->owner), lock->trans,
+                                          lock->connection_id,
                                           ctime_r (&lock->granted_time.tv_sec, granted));
                         } else {
                                 snprintf (tmp, 256, ENTRY_BLKD_GRNTD_FMT,
@@ -2229,6 +2252,7 @@ __dump_entrylks (pl_inode_t *pl_inode)
                                           "ENTRYLK_WRLCK", lock->basename,
                                           (unsigned long long) lock->client_pid,
                                           lkowner_utoa (&lock->owner), lock->trans,
+                                          lock->connection_id,
                                           ctime_r (&lock->blkd_time.tv_sec, blocked),
                                           ctime_r (&lock->granted_time.tv_sec, granted));
                         }
@@ -2248,6 +2272,7 @@ __dump_entrylks (pl_inode_t *pl_inode)
                                   "ENTRYLK_WRLCK", lock->basename,
                                   (unsigned long long) lock->client_pid,
                                   lkowner_utoa (&lock->owner), lock->trans,
+                                  lock->connection_id,
                                   ctime_r (&lock->blkd_time.tv_sec, blocked));
 
                         gf_proc_dump_write(key, tmp);
@@ -2298,7 +2323,7 @@ __dump_inodelks (pl_inode_t *pl_inode)
                         SET_FLOCK_PID (&lock->user_flock, lock);
                         pl_dump_lock (tmp, 256, &lock->user_flock,
                                       &lock->owner,
-                                      lock->transport,
+                                      lock->client, lock->connection_id,
                                       &lock->granted_time.tv_sec,
                                       &lock->blkd_time.tv_sec,
                                       _gf_true);
@@ -2315,7 +2340,7 @@ __dump_inodelks (pl_inode_t *pl_inode)
                         SET_FLOCK_PID (&lock->user_flock, lock);
                         pl_dump_lock (tmp, 256, &lock->user_flock,
                                       &lock->owner,
-                                      lock->transport,
+                                      lock->client, lock->connection_id,
                                       0, &lock->blkd_time.tv_sec,
                                       _gf_false);
                         gf_proc_dump_write(key, tmp);
@@ -2356,7 +2381,7 @@ __dump_posixlks (pl_inode_t *pl_inode)
                                      count,
                                      lock->blocked ? "BLOCKED" : "ACTIVE");
               pl_dump_lock (tmp, 256, &lock->user_flock,
-                            &lock->owner, lock->transport,
+                            &lock->owner, lock->client, NULL,
                             &lock->granted_time.tv_sec, &lock->blkd_time.tv_sec,
                             (lock->blocked)? _gf_false: _gf_true);
               gf_proc_dump_write(key, tmp);
@@ -2433,7 +2458,7 @@ unlock:
                         __dump_entrylks (pl_inode);
                 }
 
-                count = __get_inodelk_count (this, pl_inode);
+                count = __get_inodelk_count (this, pl_inode, NULL);
                 if (count) {
                         gf_proc_dump_write("inodelk-count", "%d", count);
                         __dump_inodelks (pl_inode);
@@ -2480,6 +2505,124 @@ mem_acct_init (xlator_t *this)
         return ret;
 }
 
+
+pl_ctx_t*
+pl_ctx_get (client_t *client, xlator_t *xlator)
+{
+        void *tmp = NULL;
+        pl_ctx_t *ctx = NULL;
+
+        client_ctx_get (client, xlator, &tmp);
+
+        ctx = tmp;
+
+        if (ctx != NULL)
+                goto out;
+
+        ctx = GF_CALLOC (1, sizeof (pl_ctx_t), gf_locks_mt_posix_lock_t);
+
+        if (ctx == NULL)
+                goto out;
+
+        ctx->ltable = pl_lock_table_new();
+
+        if (ctx->ltable == NULL) {
+                GF_FREE (ctx);
+                ctx = NULL;
+                goto out;
+        }
+
+        LOCK_INIT (&ctx->ltable_lock);
+
+        if (client_ctx_set (client, xlator, ctx) != 0) {
+                LOCK_DESTROY (&ctx->ltable_lock);
+                GF_FREE (ctx->ltable);
+                GF_FREE (ctx);
+                ctx = NULL;
+        }
+out:
+        return ctx;
+}
+
+static void
+ltable_delete_locks (struct _lock_table *ltable)
+{
+        struct _locker *locker = NULL;
+        struct _locker *tmp    = NULL;
+
+        list_for_each_entry_safe (locker, tmp, &ltable->inodelk_lockers, lockers) {
+                if (locker->fd)
+                        pl_del_locker (ltable, locker->volume, &locker->loc,
+                                       locker->fd, &locker->owner,
+                                       GF_FOP_INODELK);
+                GF_FREE (locker->volume);
+                GF_FREE (locker);
+        }
+
+        list_for_each_entry_safe (locker, tmp, &ltable->entrylk_lockers, lockers) {
+                if (locker->fd)
+                        pl_del_locker (ltable, locker->volume, &locker->loc,
+                                       locker->fd, &locker->owner,
+                                       GF_FOP_ENTRYLK);
+                GF_FREE (locker->volume);
+                GF_FREE (locker);
+        }
+        GF_FREE (ltable);
+}
+
+
+static int32_t
+destroy_cbk (xlator_t *this, client_t *client)
+{
+        void     *tmp       = NULL;
+        pl_ctx_t *locks_ctx = NULL;
+
+        client_ctx_del (client, this, &tmp);
+
+        if (tmp == NULL)
+                return 0
+;
+        locks_ctx = tmp;
+        if (locks_ctx->ltable)
+                ltable_delete_locks (locks_ctx->ltable);
+
+        LOCK_DESTROY (&locks_ctx->ltable_lock);
+        GF_FREE (locks_ctx);
+
+        return 0;
+}
+
+
+static int32_t
+disconnect_cbk (xlator_t *this, client_t *client)
+{
+        int32_t              ret       = 0;
+        pl_ctx_t            *locks_ctx = NULL;
+        struct _lock_table  *ltable    = NULL;
+
+        locks_ctx = pl_ctx_get (client, this);
+        if (locks_ctx == NULL) {
+                gf_log (this->name, GF_LOG_INFO, "pl_ctx_get() failed");
+                goto out;
+        }
+
+        LOCK (&locks_ctx->ltable_lock);
+        {
+                if (locks_ctx->ltable) {
+                        ltable = locks_ctx->ltable;
+                        locks_ctx->ltable = pl_lock_table_new ();
+                }
+        }
+        UNLOCK (&locks_ctx->ltable_lock);
+
+        if (ltable)
+                ltable_delete_locks (ltable);
+
+out:
+        return ret;
+}
+
+
 int
 init (xlator_t *this)
 {
@@ -2508,7 +2651,7 @@ init (xlator_t *this)
                 gf_log (this->name, GF_LOG_CRITICAL,
                         "'locks' translator is not loaded over a storage "
                         "translator");
-                goto out;;
+                goto out;
         }
 
         priv = GF_CALLOC (1, sizeof (*priv),
@@ -2610,9 +2753,11 @@ struct xlator_dumpops dumpops = {
 };
 
 struct xlator_cbks cbks = {
-        .forget      = pl_forget,
-        .release     = pl_release,
-        .releasedir  = pl_releasedir,
+        .forget            = pl_forget,
+        .release           = pl_release,
+        .releasedir        = pl_releasedir,
+        .client_destroy    = destroy_cbk,
+        .client_disconnect = disconnect_cbk,
 };
 
 
diff --git a/xlators/features/marker/Makefile.am b/xlators/features/marker/Makefile.am
index a6ba2de16..a985f42a8 100644
--- a/xlators/features/marker/Makefile.am
+++ b/xlators/features/marker/Makefile.am
@@ -1,3 +1,3 @@
-SUBDIRS = src @SYNCDAEMON_SUBDIR@
+SUBDIRS = src
 
 CLEANFILES =
diff --git a/xlators/features/marker/src/marker.c b/xlators/features/marker/src/marker.c
index 82d9066d5..6a2c85691 100644
--- a/xlators/features/marker/src/marker.c
+++ b/xlators/features/marker/src/marker.c
@@ -185,6 +185,8 @@ marker_local_unref (marker_local_t *local)
 
         loc_wipe (&local->loc);
         loc_wipe (&local->parent_loc);
+        if (local->xdata)
+                dict_unref (local->xdata);
 
         if (local->oplocal) {
                 marker_local_unref (local->oplocal);
@@ -466,11 +468,16 @@ marker_create_frame (xlator_t *this, marker_local_t *local)
 int32_t
 marker_xtime_update_marks (xlator_t *this, marker_local_t *local)
 {
+        marker_conf_t *priv = NULL;
+
         GF_VALIDATE_OR_GOTO ("marker", this, out);
         GF_VALIDATE_OR_GOTO (this->name, local, out);
 
-        if ((local->pid == GF_CLIENT_PID_GSYNCD) ||
-            (local->pid == GF_CLIENT_PID_DEFRAG))
+        priv = this->private;
+
+        if ((local->pid == GF_CLIENT_PID_GSYNCD
+             && !(priv->feature_enabled & GF_XTIME_GSYNC_FORCE))
+            || (local->pid == GF_CLIENT_PID_DEFRAG))
                 goto out;
 
         marker_gettimeofday (local);
@@ -833,7 +840,7 @@ marker_unlink_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         STACK_WIND (frame, marker_unlink_cbk, FIRST_CHILD(this),
                     FIRST_CHILD(this)->fops->unlink, &local->loc, local->xflag,
-                    NULL);
+                    local->xdata);
         return 0;
 err:
         frame->local = NULL;
@@ -858,6 +865,8 @@ marker_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
 
         local = mem_get0 (this->local_pool);
         local->xflag = xflag;
+        if (xdata)
+                local->xdata = dict_ref (xdata);
         MARKER_INIT_LOCAL (frame, local);
 
         ret = loc_copy (&local->loc, loc);
@@ -1808,6 +1817,210 @@ err:
 }
 
 
+int32_t
+marker_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)
+{
+        marker_local_t     *local   = NULL;
+        marker_conf_t      *priv    = NULL;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_TRACE, "%s occurred while "
+                        "fallocating a file ", strerror (op_errno));
+        }
+
+        local = (marker_local_t *) frame->local;
+
+        frame->local = NULL;
+
+        STACK_UNWIND_STRICT (fallocate, frame, op_ret, op_errno, prebuf,
+                             postbuf, xdata);
+
+        if (op_ret == -1 || local == NULL)
+                goto out;
+
+        priv = this->private;
+
+        if (priv->feature_enabled & GF_QUOTA)
+                mq_initiate_quota_txn (this, &local->loc);
+
+        if (priv->feature_enabled & GF_XTIME)
+                marker_xtime_update_marks (this, local);
+out:
+        marker_local_unref (local);
+
+        return 0;
+}
+
+int32_t
+marker_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          ret   = 0;
+        marker_local_t  *local = NULL;
+        marker_conf_t   *priv  = NULL;
+
+        priv = this->private;
+
+        if (priv->feature_enabled == 0)
+                goto wind;
+
+        local = mem_get0 (this->local_pool);
+
+        MARKER_INIT_LOCAL (frame, local);
+
+        ret = marker_inode_loc_fill (fd->inode, &local->loc);
+
+        if (ret == -1)
+                goto err;
+wind:
+        STACK_WIND (frame, marker_fallocate_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+		    xdata);
+        return 0;
+err:
+        STACK_UNWIND_STRICT (fallocate, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+int32_t
+marker_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)
+{
+        marker_local_t     *local   = NULL;
+        marker_conf_t      *priv    = NULL;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_TRACE, "%s occurred during discard",
+                        strerror (op_errno));
+        }
+
+        local = (marker_local_t *) frame->local;
+
+        frame->local = NULL;
+
+        STACK_UNWIND_STRICT (discard, frame, op_ret, op_errno, prebuf,
+                             postbuf, xdata);
+
+        if (op_ret == -1 || local == NULL)
+                goto out;
+
+        priv = this->private;
+
+        if (priv->feature_enabled & GF_QUOTA)
+                mq_initiate_quota_txn (this, &local->loc);
+
+        if (priv->feature_enabled & GF_XTIME)
+                marker_xtime_update_marks (this, local);
+out:
+        marker_local_unref (local);
+
+        return 0;
+}
+
+int32_t
+marker_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+	       size_t len, dict_t *xdata)
+{
+        int32_t          ret   = 0;
+        marker_local_t  *local = NULL;
+        marker_conf_t   *priv  = NULL;
+
+        priv = this->private;
+
+        if (priv->feature_enabled == 0)
+                goto wind;
+
+        local = mem_get0 (this->local_pool);
+
+        MARKER_INIT_LOCAL (frame, local);
+
+        ret = marker_inode_loc_fill (fd->inode, &local->loc);
+
+        if (ret == -1)
+                goto err;
+wind:
+        STACK_WIND (frame, marker_discard_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+        return 0;
+err:
+        STACK_UNWIND_STRICT (discard, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+        return 0;
+}
+
+int32_t
+marker_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)
+{
+        marker_local_t     *local   = NULL;
+        marker_conf_t      *priv    = NULL;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_TRACE, "%s occurred during zerofill",
+                        strerror (op_errno));
+        }
+
+        local = (marker_local_t *) frame->local;
+
+        frame->local = NULL;
+
+        STACK_UNWIND_STRICT (zerofill, frame, op_ret, op_errno, prebuf,
+                             postbuf, xdata);
+
+        if (op_ret == -1 || local == NULL)
+                goto out;
+
+        priv = this->private;
+
+        if (priv->feature_enabled & GF_QUOTA)
+                mq_initiate_quota_txn (this, &local->loc);
+
+        if (priv->feature_enabled & GF_XTIME)
+                marker_xtime_update_marks (this, local);
+out:
+        marker_local_unref (local);
+
+        return 0;
+}
+
+int32_t
+marker_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               size_t len, dict_t *xdata)
+{
+        int32_t          ret   = 0;
+        marker_local_t  *local = NULL;
+        marker_conf_t   *priv  = NULL;
+
+        priv = this->private;
+
+        if (priv->feature_enabled == 0)
+                goto wind;
+
+        local = mem_get0 (this->local_pool);
+
+        MARKER_INIT_LOCAL (frame, local);
+
+        ret = marker_inode_loc_fill (fd->inode, &local->loc);
+
+        if (ret == -1)
+                goto err;
+wind:
+        STACK_WIND (frame, marker_zerofill_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+        return 0;
+err:
+        STACK_UNWIND_STRICT (zerofill, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
 /* when a call from the special client is received on
  * key trusted.glusterfs.volume-mark with value "RESET"
  * or if the value is 0length, update the change the
@@ -2452,7 +2665,7 @@ out:
 int32_t
 reconfigure (xlator_t *this, dict_t *options)
 {
-        int32_t         ret     = -1;
+        int32_t         ret     = 0;
         data_t         *data    = NULL;
         gf_boolean_t    flag    = _gf_false;
         marker_conf_t  *priv    = NULL;
@@ -2493,11 +2706,17 @@ reconfigure (xlator_t *this, dict_t *options)
                                         "xtime updation will fail");
                         } else {
                                 priv->feature_enabled |= GF_XTIME;
+                                data = dict_get (options, "gsync-force-xtime");
+                                if (!data)
+                                        goto out;
+                                ret = gf_string2boolean (data->data, &flag);
+                                if (ret == 0 && flag)
+                                        priv->feature_enabled |= GF_XTIME_GSYNC_FORCE;
                         }
                 }
         }
 out:
-        return 0;
+        return ret;
 }
 
 
@@ -2553,9 +2772,16 @@ init (xlator_t *this)
                                 goto err;
 
                         priv->feature_enabled |= GF_XTIME;
+                        data = dict_get (options, "gsync-force-xtime");
+                        if (!data)
+                                goto cont;
+                        ret = gf_string2boolean (data->data, &flag);
+                        if (ret == 0 && flag)
+                                priv->feature_enabled |= GF_XTIME_GSYNC_FORCE;
                 }
         }
 
+ cont:
         this->local_pool = mem_pool_new (marker_local_t, 128);
         if (!this->local_pool) {
                 gf_log (this->name, GF_LOG_ERROR,
@@ -2617,6 +2843,9 @@ struct xlator_fops fops = {
         .removexattr = marker_removexattr,
         .getxattr    = marker_getxattr,
         .readdirp    = marker_readdirp,
+	.fallocate   = marker_fallocate,
+	.discard     = marker_discard,
+        .zerofill    = marker_zerofill,
 };
 
 struct xlator_cbks cbks = {
@@ -2628,5 +2857,6 @@ struct volume_options options[] = {
         {.key = {"timestamp-file"}},
         {.key = {"quota"}},
         {.key = {"xtime"}},
+        {.key = {"gsync-force-xtime"}},
         {.key = {NULL}}
 };
diff --git a/xlators/features/marker/src/marker.h b/xlators/features/marker/src/marker.h
index 63491ab37..1a58f8cfc 100644
--- a/xlators/features/marker/src/marker.h
+++ b/xlators/features/marker/src/marker.h
@@ -28,8 +28,9 @@
 #define TIMESTAMP_FILE      "timestamp-file"
 
 enum {
-        GF_QUOTA=1,
-        GF_XTIME=2
+        GF_QUOTA             = 1,
+        GF_XTIME             = 2,
+        GF_XTIME_GSYNC_FORCE = 4,
 };
 
 /*initialize the local variable*/
@@ -110,6 +111,7 @@ struct marker_local{
         inode_contribution_t *contri;
 
         int xflag;
+        dict_t *xdata;
 };
 typedef struct marker_local marker_local_t;
 
diff --git a/xlators/features/marker/utils/Makefile.am b/xlators/features/marker/utils/Makefile.am
deleted file mode 100644
index 556951d9f..000000000
--- a/xlators/features/marker/utils/Makefile.am
+++ /dev/null
@@ -1,3 +0,0 @@
-SUBDIRS = syncdaemon src
-
-CLEANFILES =
diff --git a/xlators/features/marker/utils/src/Makefile.am b/xlators/features/marker/utils/src/Makefile.am
deleted file mode 100644
index 9e410cda6..000000000
--- a/xlators/features/marker/utils/src/Makefile.am
+++ /dev/null
@@ -1,26 +0,0 @@
-gsyncddir = $(libexecdir)/glusterfs
-
-gsyncd_PROGRAMS = gsyncd
-
-gsyncd_SOURCES = gsyncd.c procdiggy.c
-
-gsyncd_LDADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
-	$(GF_GLUSTERFS_LIBS)
-
-gsyncd_LDFLAGS = $(GF_LDFLAGS)
-
-noinst_HEADERS = procdiggy.h
-
-AM_CPPFLAGS = $(GF_CPPFLAGS) \
-	-I$(top_srcdir)/libglusterfs/src\
-	-DGSYNCD_PREFIX=\"$(libexecdir)/glusterfs\"\
-	-DUSE_LIBGLUSTERFS\
-	-DSBIN_DIR=\"$(sbindir)\" -DPYTHON=\"$(PYTHON)\"
-
-AM_CFLAGS = -Wall $(GF_CFLAGS)
-
-
-CLEANFILES =
-
-$(top_builddir)/libglusterfs/src/libglusterfs.la:
-	$(MAKE) -C $(top_builddir)/libglusterfs/src/ all
diff --git a/xlators/features/marker/utils/src/gsyncd.c b/xlators/features/marker/utils/src/gsyncd.c
deleted file mode 100644
index 9c4a5bdff..000000000
--- a/xlators/features/marker/utils/src/gsyncd.c
+++ /dev/null
@@ -1,367 +0,0 @@
-/*
-   Copyright (c) 2011-2012 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 <stdlib.h>
-#include <stdio.h>
-#include <unistd.h>
-#include <string.h>
-#include <sys/param.h> /* for PATH_MAX */
-
-/* NOTE (USE_LIBGLUSTERFS):
- * ------------------------
- * When USE_LIBGLUSTERFS debugging sumbol is passed; perform
- * glusterfs translator like initialization so that glusterfs
- * globals, contexts are valid when glustefs api's are invoked.
- * We unconditionally pass then while building gsyncd binary.
- */
-#ifdef USE_LIBGLUSTERFS
-#include "glusterfs.h"
-#include "globals.h"
-#endif
-
-#include "common-utils.h"
-#include "run.h"
-#include "procdiggy.h"
-
-#define _GLUSTERD_CALLED_ "_GLUSTERD_CALLED_"
-#define _GSYNCD_DISPATCHED_ "_GSYNCD_DISPATCHED_"
-#define GSYNCD_CONF "geo-replication/gsyncd.conf"
-#define GSYNCD_PY "gsyncd.py"
-#define RSYNC "rsync"
-
-int restricted = 0;
-
-static int
-duplexpand (void **buf, size_t tsiz, size_t *len)
-{
-        size_t osiz = tsiz * *len;
-        char *p = realloc (*buf, osiz << 1);
-        if (!p) {
-                free(*buf);
-                return -1;
-        }
-
-        memset (p + osiz, 0, osiz);
-        *buf = p;
-        *len <<= 1;
-
-        return 0;
-}
-
-static int
-str2argv (char *str, char ***argv)
-{
-        char *p         = NULL;
-        char *savetok   = NULL;
-        int argc        = 0;
-        size_t argv_len = 32;
-        int ret         = 0;
-
-        assert (str);
-        str = strdup (str);
-        if (!str)
-                return -1;
-
-        *argv = calloc (argv_len, sizeof (**argv));
-        if (!*argv)
-                goto error;
-
-        while ((p = strtok_r (str, " ", &savetok))) {
-                str = NULL;
-
-                argc++;
-                if (argc == argv_len) {
-                        ret = duplexpand ((void *)argv,
-                                          sizeof (**argv),
-                                          &argv_len);
-                        if (ret == -1)
-                                goto error;
-                }
-                (*argv)[argc - 1] = p;
-        }
-
-        return argc;
-
- error:
-        fprintf (stderr, "out of memory\n");
-        return -1;
-}
-
-static int
-invoke_gsyncd (int argc, char **argv)
-{
-        char config_file[PATH_MAX] = {0,};
-        size_t gluster_workdir_len = 0;
-        runner_t runner            = {0,};
-        int i                      = 0;
-        int j                      = 0;
-        char *nargv[argc + 4];
-        char *python = NULL;
-
-        if (restricted) {
-                size_t len;
-                /* in restricted mode we forcibly use the system-wide config */
-                runinit (&runner);
-                runner_add_args (&runner, SBIN_DIR"/gluster",
-                                 "--log-file=-", "system::", "getwd",
-                                 NULL);
-                runner_redir (&runner, STDOUT_FILENO, RUN_PIPE);
-                if (runner_start (&runner) == 0 &&
-                    fgets (config_file, PATH_MAX,
-                           runner_chio (&runner, STDOUT_FILENO)) != NULL &&
-                    (len = strlen (config_file)) &&
-                    config_file[len - 1] == '\n' &&
-                    runner_end (&runner) == 0)
-                        gluster_workdir_len = len - 1;
-
-                if (gluster_workdir_len) {
-                        if (gluster_workdir_len + 1 + strlen (GSYNCD_CONF) + 1 >
-                            PATH_MAX)
-                                goto error;
-                        config_file[gluster_workdir_len] = '/';
-                        strcat (config_file, GSYNCD_CONF);
-                } else
-                        goto error;
-
-                if (setenv ("_GSYNCD_RESTRICTED_", "1", 1) == -1)
-                        goto error;
-        }
-
-        if (chdir ("/") == -1)
-                goto error;
-
-        j = 0;
-        python = getenv("PYTHON");
-        if(!python)
-                python = PYTHON;
-        nargv[j++] = python;
-        nargv[j++] = GSYNCD_PREFIX"/python/syncdaemon/"GSYNCD_PY;
-        for (i = 1; i < argc; i++)
-                nargv[j++] = argv[i];
-        if (config_file[0]) {
-                nargv[j++] = "-c";
-                nargv[j++] = config_file;
-        }
-        nargv[j++] = NULL;
-
-        execvp (python, nargv);
-
-        fprintf (stderr, "exec of '%s' failed\n", python);
-        return 127;
-
- error:
-        fprintf (stderr, "gsyncd initializaion failed\n");
-        return 1;
-}
-
-
-static int
-find_gsyncd (pid_t pid, pid_t ppid, char *name, void *data)
-{
-        char buf[NAME_MAX * 2] = {0,};
-        char path[PATH_MAX]    = {0,};
-        char *p                = NULL;
-        int zeros              = 0;
-        int ret                = 0;
-        int fd                 = -1;
-        pid_t *pida            = (pid_t *)data;
-
-        if (ppid != pida[0])
-                return 0;
-
-        sprintf (path, PROC"/%d/cmdline", pid);
-        fd = open (path, O_RDONLY);
-        if (fd == -1)
-                return 0;
-        ret = read (fd, buf, sizeof (buf));
-        close (fd);
-        if (ret == -1)
-                return 0;
-        for (zeros = 0, p = buf; zeros < 2 && p < buf + ret; p++)
-                zeros += !*p;
-
-        ret = 0;
-        switch (zeros) {
-        case 2:
-                if ((strcmp (basename (buf), basename (PYTHON)) ||
-                     strcmp (basename (buf + strlen (buf) + 1), GSYNCD_PY)) == 0) {
-                        ret = 1;
-                        break;
-                }
-                /* fallthrough */
-        case 1:
-                if (strcmp (basename (buf), GSYNCD_PY) == 0)
-                        ret = 1;
-        }
-
-        if (ret == 1) {
-                if (pida[1] != -1) {
-                        fprintf (stderr, GSYNCD_PY" sibling is not unique");
-                        return -1;
-                }
-                pida[1] = pid;
-        }
-
-        return 0;
-}
-
-static int
-invoke_rsync (int argc, char **argv)
-{
-        int i                  = 0;
-        char path[PATH_MAX]    = {0,};
-        pid_t pid              = -1;
-        pid_t ppid             = -1;
-        pid_t pida[]           = {-1, -1};
-        char *name             = NULL;
-        char buf[PATH_MAX + 1] = {0,};
-        int ret                = 0;
-
-        assert (argv[argc] == NULL);
-
-        if (argc < 2 || strcmp (argv[1], "--server") != 0)
-                goto error;
-
-        for (i = 2; i < argc && argv[i][0] == '-'; i++);
-
-        if (!(i == argc - 2 && strcmp (argv[i], ".") == 0 && argv[i + 1][0] == '/')) {
-                fprintf (stderr, "need an rsync invocation without protected args\n");
-                goto error;
-        }
-
-        /* look up sshd we are spawned from */
-        for (pid = getpid () ;; pid = ppid) {
-                ppid = pidinfo (pid, &name);
-                if (ppid < 0) {
-                        fprintf (stderr, "sshd ancestor not found\n");
-                        goto error;
-                }
-                if (strcmp (name, "sshd") == 0) {
-                        GF_FREE (name);
-                        break;
-                }
-                GF_FREE (name);
-        }
-        /* look up "ssh-sibling" gsyncd */
-        pida[0] = pid;
-        ret = prociter (find_gsyncd, pida);
-        if (ret == -1 || pida[1] == -1) {
-                fprintf (stderr, "gsyncd sibling not found\n");
-                goto error;
-        }
-        /* check if rsync target matches gsyncd target */
-        sprintf (path, PROC"/%d/cwd", pida[1]);
-        ret = readlink (path, buf, sizeof (buf));
-        if (ret == -1 || ret == sizeof (buf))
-                goto error;
-        if (strcmp (argv[argc - 1], "/") == 0 /* root dir cannot be a target */ ||
-            (strcmp (argv[argc - 1], path) /* match against gluster target */ &&
-             strcmp (argv[argc - 1], buf) /* match against file target */) != 0) {
-                fprintf (stderr, "rsync target does not match "GEOREP" session\n");
-                goto error;
-        }
-
-        argv[0] = RSYNC;
-
-        execvp (RSYNC, argv);
-
-        fprintf (stderr, "exec of "RSYNC" failed\n");
-        return 127;
-
- error:
-        fprintf (stderr, "disallowed "RSYNC" invocation\n");
-        return 1;
-}
-
-
-struct invocable {
-        char *name;
-        int (*invoker) (int argc, char **argv);
-};
-
-struct invocable invocables[] = {
-        { "rsync",  invoke_rsync  },
-        { "gsyncd", invoke_gsyncd },
-        { NULL, NULL}
-};
-
-int
-main (int argc, char **argv)
-{
-        char *evas          = NULL;
-        struct invocable *i = NULL;
-        char *b             = NULL;
-        char *sargv         = NULL;
-
-#ifdef USE_LIBGLUSTERFS
-        glusterfs_ctx_t *ctx = NULL;
-
-        ctx = glusterfs_ctx_new ();
-        if (!ctx)
-                return ENOMEM;
-
-        if (glusterfs_globals_init (ctx))
-                return 1;
-
-        THIS->ctx = ctx;
-#endif
-
-        evas = getenv (_GLUSTERD_CALLED_);
-        if (evas && strcmp (evas, "1") == 0)
-                /* OK, we know glusterd called us, no need to look for further config
-                 * ... altough this conclusion should not inherit to our children
-                 */
-                unsetenv (_GLUSTERD_CALLED_);
-        else {
-                /* we regard all gsyncd invocations unsafe
-                 * that do not come from glusterd and
-                 * therefore restrict it
-                 */
-                restricted = 1;
-
-                if (!getenv (_GSYNCD_DISPATCHED_)) {
-                        evas = getenv ("SSH_ORIGINAL_COMMAND");
-                        if (evas)
-                                sargv = evas;
-                        else {
-                                evas = getenv ("SHELL");
-                                if (evas && strcmp (basename (evas), "gsyncd") == 0 &&
-                                    argc == 3 && strcmp (argv[1], "-c") == 0)
-                                        sargv = argv[2];
-                        }
-                }
-
-        }
-
-        if (!(sargv && restricted))
-                return invoke_gsyncd (argc, argv);
-
-        argc = str2argv (sargv, &argv);
-        if (argc == -1 || setenv (_GSYNCD_DISPATCHED_, "1", 1) == -1) {
-                fprintf (stderr, "internal error\n");
-                return 1;
-        }
-
-        b = basename (argv[0]);
-        for (i = invocables; i->name; i++) {
-                if (strcmp (b, i->name) == 0)
-                        return i->invoker (argc, argv);
-        }
-
-        fprintf (stderr, "invoking %s in restricted SSH session is not allowed\n",
-                 b);
-
-        return 1;
-}
diff --git a/xlators/features/marker/utils/src/procdiggy.c b/xlators/features/marker/utils/src/procdiggy.c
deleted file mode 100644
index 1eba414c1..000000000
--- a/xlators/features/marker/utils/src/procdiggy.c
+++ /dev/null
@@ -1,121 +0,0 @@
-/*
-   Copyright (c) 2011-2012 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 <stdlib.h>
-#include <stdio.h>
-#include <unistd.h>
-#include <string.h>
-#include <ctype.h>
-#include <sys/param.h> /* for PATH_MAX */
-
-#include "common-utils.h"
-#include "procdiggy.h"
-
-pid_t
-pidinfo (pid_t pid, char **name)
-{
-        char buf[NAME_MAX * 2] = {0,};
-        FILE *f                = NULL;
-        char path[PATH_MAX]    = {0,};
-        char *p                = NULL;
-        int ret                = 0;
-
-        sprintf (path, PROC"/%d/status", pid);
-
-        f = fopen (path, "r");
-        if (!f)
-                return -1;
-
-        if (name)
-                *name = NULL;
-        for (;;) {
-                size_t len;
-                memset (buf, 0, sizeof (buf));
-                if (fgets (buf, sizeof (buf), f) == NULL ||
-                    (len = strlen (buf)) == 0 ||
-                    buf[len - 1] != '\n') {
-                        pid = -1;
-                        goto out;
-                }
-                buf[len - 1] = '\0';
-
-                if (name && !*name) {
-                        p = strtail (buf, "Name:");
-                        if (p) {
-                                while (isspace (*++p));
-                                *name = gf_strdup (p);
-                                if (!*name) {
-                                        pid = -2;
-                                        goto out;
-                                }
-                                continue;
-                        }
-                }
-
-                p = strtail (buf, "PPid:");
-                if (p)
-                        break;
-        }
-
-        while (isspace (*++p));
-        ret = gf_string2int (p, &pid);
-        if (ret == -1)
-                pid = -1;
-
- out:
-        fclose (f);
-        if (pid == -1 && name && *name)
-                GF_FREE (name);
-        if (pid == -2)
-                fprintf (stderr, "out of memory\n");
-        return pid;
-}
-
-int
-prociter (int (*proch) (pid_t pid, pid_t ppid, char *tmpname, void *data),
-          void *data)
-{
-        char *name        = NULL;
-        DIR *d            = NULL;
-        struct dirent *de = NULL;
-        pid_t pid         = -1;
-        pid_t ppid        = -1;
-        int ret           = 0;
-
-        d = opendir (PROC);
-        if (!d)
-                return -1;
-        while (errno = 0, de = readdir (d)) {
-                if (gf_string2int (de->d_name, &pid) != -1 && pid >= 0) {
-                        ppid = pidinfo (pid, &name);
-                        switch (ppid) {
-                        case -1: continue;
-                        case -2: ret = -1; break;
-                        }
-                        ret = proch (pid, ppid, name, data);
-                        GF_FREE (name);
-                        if (ret)
-                                break;
-                }
-        }
-        closedir (d);
-        if (!de && errno) {
-                fprintf (stderr, "failed to traverse "PROC" (%s)\n",
-                         strerror (errno));
-                ret = -1;
-        }
-
-        return ret;
-}
diff --git a/xlators/features/marker/utils/syncdaemon/Makefile.am b/xlators/features/marker/utils/syncdaemon/Makefile.am
deleted file mode 100644
index cc7cee102..000000000
--- a/xlators/features/marker/utils/syncdaemon/Makefile.am
+++ /dev/null
@@ -1,6 +0,0 @@
-syncdaemondir = $(libexecdir)/glusterfs/python/syncdaemon
-
-syncdaemon_PYTHON = gconf.py gsyncd.py __init__.py master.py README.md repce.py resource.py configinterface.py syncdutils.py monitor.py libcxattr.py \
-                    $(top_builddir)/contrib/ipaddr-py/ipaddr.py
-
-CLEANFILES =
diff --git a/xlators/features/marker/utils/syncdaemon/README.md b/xlators/features/marker/utils/syncdaemon/README.md
deleted file mode 100644
index d45006932..000000000
--- a/xlators/features/marker/utils/syncdaemon/README.md
+++ /dev/null
@@ -1,81 +0,0 @@
-gsycnd, the Gluster Syncdaemon
-==============================
-
-REQUIREMENTS
-------------
-
-_gsyncd_ is a program which can operate either in _master_ or in _slave_ mode.
-Requirements are categorized according to this.
-
-* supported OS is GNU/Linux
-* Python >= 2.5, or 2.4 with Ctypes (see below) (both)
-* OpenSSH >= 4.0 (master) / SSH2 compliant sshd (eg. openssh) (slave)
-* rsync (both)
-* glusterfs with marker support (master); glusterfs (optional on slave)
-* FUSE; for supported versions consult glusterfs
-
-INSTALLATION
-------------
-
-As of now, the supported way of operation is running from the source directory.
-
-If you use Python 2.4.x, you need to install the [Ctypes module](http://python.net/crew/theller/ctypes/).
-
-CONFIGURATION
--------------
-
-gsyncd tunables are a subset of the long command-line options; for listing them,
-type
-
-    gsyncd.py --help
-
-and see the long options up to "--config-file". (The leading double dash should be omitted;
-interim underscores and dashes are interchangeable.) The set of options bear some resemblance
-to those of glusterfs and rsync.
-
-The config file format matches the following syntax:
-
-      <option1>: <value1>
-      <option2>: <value2>
-      # comment
-
-By default (unless specified by the option `-c`), gsyncd looks for config file at _conf/gsyncd.conf_
-in the source tree.
-
-USAGE
------
-
-gsyncd is a utilitly for continous mirroring, ie. it mirrors master to slave incrementally.
-Assume we have a gluster volume _pop_ at localhost. We try to set up the following mirrors
-for it with gysncd:
-
-1. _/data/mirror_
-2. local gluster volume _yow_
-3. _/data/far_mirror_ at example.com
-4. gluster volume _moz_ at example.com
-
-The respective gsyncd invocations are (demoing some syntax sugaring):
-
-1.
-
-      gsyncd.py gluster://localhost:pop file:///data/mirror
-
-  or short form
-
-      gsyncd.py :pop /data/mirror
-
-2. `gsyncd :pop :yow`
-3.
-
-       gsyncd.py :pop ssh://example.com:/data/far_mirror
-
-  or short form
-
-       gsyncd.py :pop example.com:/data/far_mirror
-
-4. `gsyncd.py :pop example.com::moz`
-
-gsyncd has to be available on both sides; it's location on the remote side has to be specified
-via the "--remote-gsyncd" option (or "remote-gsyncd" config file parameter). (This option can also be
-used for setting options on the remote side, although the suggested mode of operation is to
-set parameters like log file / pid file in the configuration file.)
diff --git a/xlators/features/marker/utils/syncdaemon/__codecheck.py b/xlators/features/marker/utils/syncdaemon/__codecheck.py
deleted file mode 100644
index e3386afba..000000000
--- a/xlators/features/marker/utils/syncdaemon/__codecheck.py
+++ /dev/null
@@ -1,46 +0,0 @@
-import os
-import os.path
-import sys
-import tempfile
-import shutil
-
-ipd = tempfile.mkdtemp(prefix = 'codecheck-aux')
-
-try:
-    # add a fake ipaddr module, we don't want to
-    # deal with the real one (just test our code)
-    f = open(os.path.join(ipd, 'ipaddr.py'), 'w')
-    f.write("""
-class IPAddress(object):
-    pass
-class IPNetwork(list):
-    pass
-""")
-    f.close()
-    sys.path.append(ipd)
-
-    fl = os.listdir(os.path.dirname(sys.argv[0]) or '.')
-    fl.sort()
-    for f in fl:
-        if f[-3:] != '.py' or f[0] == '_':
-            continue
-        m = f[:-3]
-        sys.stdout.write('importing %s ...' %  m)
-        __import__(m)
-        print(' OK.')
-
-    def sys_argv_set(a):
-        sys.argv = sys.argv[:1] + a
-
-    gsyncd = sys.modules['gsyncd']
-    for a in [['--help'], ['--version'], ['--canonicalize-escape-url', '/foo']]:
-        print('>>> invoking program with args: %s' % ' '.join(a))
-        pid = os.fork()
-        if not pid:
-            sys_argv_set(a)
-            gsyncd.main()
-        _, r = os.waitpid(pid, 0)
-        if r:
-            raise RuntimeError('invocation failed')
-finally:
-    shutil.rmtree(ipd)
diff --git a/xlators/features/marker/utils/syncdaemon/__init__.py b/xlators/features/marker/utils/syncdaemon/__init__.py
deleted file mode 100644
index e69de29bb..000000000
--- a/xlators/features/marker/utils/syncdaemon/__init__.py
+++ /dev/null
diff --git a/xlators/features/marker/utils/syncdaemon/configinterface.py b/xlators/features/marker/utils/syncdaemon/configinterface.py
deleted file mode 100644
index e55bec519..000000000
--- a/xlators/features/marker/utils/syncdaemon/configinterface.py
+++ /dev/null
@@ -1,224 +0,0 @@
-try:
-    import ConfigParser
-except ImportError:
-    # py 3
-    import configparser as ConfigParser
-import re
-from string import Template
-
-from syncdutils import escape, unescape, norm, update_file, GsyncdError
-
-SECT_ORD = '__section_order__'
-SECT_META = '__meta__'
-config_version = 2.0
-
-re_type = type(re.compile(''))
-
-
-class MultiDict(object):
-    """a virtual dict-like class which functions as the union of underlying dicts"""
-
-    def __init__(self, *dd):
-        self.dicts = dd
-
-    def __getitem__(self, key):
-        val = None
-        for d in self.dicts:
-            if d.get(key):
-                val = d[key]
-        if not val:
-            raise KeyError(key)
-        return val
-
-
-class GConffile(object):
-    """A high-level interface to ConfigParser which flattens the two-tiered
-       config layout by implenting automatic section dispatch based on initial
-       parameters.
-
-    Also ensure section ordering in terms of their time of addition -- a compat
-    hack for Python < 2.7.
-    """
-
-    def _normconfig(self):
-        """normalize config keys by s/-/_/g"""
-        for n, s in self.config._sections.items():
-            if n.find('__') == 0:
-                continue
-            s2 = type(s)()
-            for k, v in s.items():
-                if k.find('__') != 0:
-                    k = norm(k)
-                s2[k] = v
-            self.config._sections[n] = s2
-
-    def __init__(self, path, peers, *dd):
-        """
-        - .path: location of config file
-        - .config: underlying ConfigParser instance
-        - .peers: on behalf of whom we flatten .config
-          (master, or master-slave url pair)
-        - .auxdicts: template subtituents
-        """
-        self.peers = peers
-        self.path = path
-        self.auxdicts = dd
-        self.config = ConfigParser.RawConfigParser()
-        self.config.read(path)
-        self._normconfig()
-
-    def section(self, rx=False):
-        """get the section name of the section representing .peers in .config"""
-        peers = self.peers
-        if not peers:
-            peers = ['.', '.']
-            rx = True
-        if rx:
-            st = 'peersrx'
-        else:
-            st = 'peers'
-        return ' '.join([st] + [escape(u) for u in peers])
-
-    @staticmethod
-    def parse_section(section):
-        """retrieve peers sequence encoded by section name
-           (as urls or regexen, depending on section type)
-        """
-        sl = section.split()
-        st = sl.pop(0)
-        sl = [unescape(u) for u in sl]
-        if st == 'peersrx':
-            sl = [re.compile(u) for u in sl]
-        return sl
-
-    def ord_sections(self):
-        """Return an ordered list of sections.
-
-        Ordering happens based on the auxiliary
-        SECT_ORD section storing indices for each
-        section added through the config API.
-
-        To not to go corrupt in case of manually
-        written config files, we take care to append
-        also those sections which are not registered
-        in SECT_ORD.
-
-        Needed for python 2.{4,5,6} where ConfigParser
-        cannot yet order sections/options internally.
-        """
-        so = {}
-        if self.config.has_section(SECT_ORD):
-            so = self.config._sections[SECT_ORD]
-        so2 = {}
-        for k, v in so.items():
-            if k != '__name__':
-                so2[k] = int(v)
-        tv = 0
-        if so2:
-            tv = max(so2.values()) + 1
-        ss = [s for s in self.config.sections() if s.find('__') != 0]
-        for s in ss:
-            if s in so.keys():
-                continue
-            so2[s] = tv
-            tv += 1
-        def scmp(x, y):
-            return cmp(*(so2[s] for s in (x, y)))
-        ss.sort(scmp)
-        return ss
-
-    def update_to(self, dct, allow_unresolved=False):
-        """update @dct from key/values of ours.
-
-        key/values are collected from .config by filtering the regexp sections
-        according to match, and from .section. The values are treated as templates,
-        which are substituted from .auxdicts and (in case of regexp sections)
-        match groups.
-        """
-        if not self.peers:
-            raise GsyncdError('no peers given, cannot select matching options')
-        def update_from_sect(sect, mud):
-            for k, v in self.config._sections[sect].items():
-                if k == '__name__':
-                    continue
-                if allow_unresolved:
-                    dct[k] = Template(v).safe_substitute(mud)
-                else:
-                    dct[k] = Template(v).substitute(mud)
-        for sect in self.ord_sections():
-            sp = self.parse_section(sect)
-            if isinstance(sp[0], re_type) and len(sp) == len(self.peers):
-                match = True
-                mad = {}
-                for i in range(len(sp)):
-                    m = sp[i].search(self.peers[i])
-                    if not m:
-                        match = False
-                        break
-                    for j in range(len(m.groups())):
-                        mad['match%d_%d' % (i+1, j+1)] = m.groups()[j]
-                if match:
-                    update_from_sect(sect, MultiDict(dct, mad, *self.auxdicts))
-        if self.config.has_section(self.section()):
-            update_from_sect(self.section(), MultiDict(dct, *self.auxdicts))
-
-    def get(self, opt=None):
-        """print the matching key/value pairs from .config,
-           or if @opt given, the value for @opt (according to the
-           logic described in .update_to)
-        """
-        d = {}
-        self.update_to(d, allow_unresolved = True)
-        if opt:
-            opt = norm(opt)
-            v = d.get(opt)
-            if v:
-                print(v)
-        else:
-            for k, v in d.iteritems():
-                if k == '__name__':
-                    continue
-                print("%s: %s" % (k, v))
-
-    def write(self, trfn, opt, *a, **kw):
-        """update on-disk config transactionally
-
-        @trfn is the transaction function
-        """
-        def mergeconf(f):
-            self.config = ConfigParser.RawConfigParser()
-            self.config.readfp(f)
-            self._normconfig()
-            if not self.config.has_section(SECT_META):
-                self.config.add_section(SECT_META)
-            self.config.set(SECT_META, 'version', config_version)
-            return trfn(norm(opt), *a, **kw)
-        def updateconf(f):
-            self.config.write(f)
-        update_file(self.path, updateconf, mergeconf)
-
-    def _set(self, opt, val, rx=False):
-        """set @opt to @val in .section"""
-        sect = self.section(rx)
-        if not self.config.has_section(sect):
-            self.config.add_section(sect)
-            # regarding SECT_ORD, cf. ord_sections
-            if not self.config.has_section(SECT_ORD):
-                self.config.add_section(SECT_ORD)
-            self.config.set(SECT_ORD, sect, len(self.config._sections[SECT_ORD]))
-        self.config.set(sect, opt, val)
-        return True
-
-    def set(self, opt, *a, **kw):
-        """perform ._set transactionally"""
-        self.write(self._set, opt, *a, **kw)
-
-    def _delete(self, opt, rx=False):
-        """delete @opt from .section"""
-        sect = self.section(rx)
-        if self.config.has_section(sect):
-            return self.config.remove_option(sect, opt)
-
-    def delete(self, opt, *a, **kw):
-        """perform ._delete transactionally"""
-        self.write(self._delete, opt, *a, **kw)
diff --git a/xlators/features/marker/utils/syncdaemon/gconf.py b/xlators/features/marker/utils/syncdaemon/gconf.py
deleted file mode 100644
index 146c72a18..000000000
--- a/xlators/features/marker/utils/syncdaemon/gconf.py
+++ /dev/null
@@ -1,20 +0,0 @@
-import os
-
-class GConf(object):
-    """singleton class to store globals
-       shared between gsyncd modules"""
-
-    ssh_ctl_dir = None
-    ssh_ctl_args = None
-    cpid = None
-    pid_file_owned = False
-    log_exit = False
-    permanent_handles = []
-    log_metadata = {}
-
-    @classmethod
-    def setup_ssh_ctl(cls, ctld):
-        cls.ssh_ctl_dir = ctld
-        cls.ssh_ctl_args = ["-oControlMaster=auto", "-S", os.path.join(ctld, "gsycnd-ssh-%r@%h:%p")]
-
-gconf = GConf()
diff --git a/xlators/features/marker/utils/syncdaemon/gsyncd.py b/xlators/features/marker/utils/syncdaemon/gsyncd.py
deleted file mode 100644
index 387900e6c..000000000
--- a/xlators/features/marker/utils/syncdaemon/gsyncd.py
+++ /dev/null
@@ -1,419 +0,0 @@
-#!/usr/bin/env python
-
-import os
-import os.path
-import sys
-import time
-import logging
-import signal
-import optparse
-import fcntl
-import fnmatch
-from optparse import OptionParser, SUPPRESS_HELP
-from logging import Logger
-from errno import EEXIST, ENOENT
-
-from ipaddr import IPAddress, IPNetwork
-
-from gconf import gconf
-from syncdutils import FreeObject, norm, grabpidfile, finalize, log_raise_exception
-from syncdutils import GsyncdError, select, set_term_handler, privileged
-from configinterface import GConffile
-import resource
-from monitor import monitor
-
-class GLogger(Logger):
-    """Logger customizations for gsyncd.
-
-    It implements a log format similar to that of glusterfs.
-    """
-
-    def makeRecord(self, name, level, *a):
-        rv = Logger.makeRecord(self, name, level, *a)
-        rv.nsecs = (rv.created - int(rv.created)) * 1000000
-        fr = sys._getframe(4)
-        callee = fr.f_locals.get('self')
-        if callee:
-            ctx = str(type(callee)).split("'")[1].split('.')[-1]
-        else:
-            ctx = '<top>'
-        if not hasattr(rv, 'funcName'):
-            rv.funcName = fr.f_code.co_name
-        rv.lvlnam = logging.getLevelName(level)[0]
-        rv.ctx = ctx
-        return rv
-
-    @classmethod
-    def setup(cls, **kw):
-        lbl = kw.get('label', "")
-        if lbl:
-            lbl = '(' + lbl + ')'
-        lprm = {'datefmt': "%Y-%m-%d %H:%M:%S",
-                'format': "[%(asctime)s.%(nsecs)d] %(lvlnam)s [%(module)s" + lbl + ":%(lineno)s:%(funcName)s] %(ctx)s: %(message)s"}
-        lprm.update(kw)
-        lvl = kw.get('level', logging.INFO)
-        lprm['level'] = lvl
-        logging.root = cls("root", lvl)
-        logging.setLoggerClass(cls)
-        logging.getLogger().handlers = []
-        logging.basicConfig(**lprm)
-
-    @classmethod
-    def _gsyncd_loginit(cls, **kw):
-        lkw = {}
-        if gconf.log_level:
-            lkw['level'] = gconf.log_level
-        if kw.get('log_file'):
-            if kw['log_file'] in ('-', '/dev/stderr'):
-                lkw['stream'] = sys.stderr
-            elif kw['log_file'] == '/dev/stdout':
-                lkw['stream'] = sys.stdout
-            else:
-                lkw['filename'] = kw['log_file']
-
-        cls.setup(label=kw.get('label'), **lkw)
-
-        lkw.update({'saved_label': kw.get('label')})
-        gconf.log_metadata = lkw
-        gconf.log_exit = True
-
-def startup(**kw):
-    """set up logging, pidfile grabbing, daemonization"""
-    if getattr(gconf, 'pid_file', None) and kw.get('go_daemon') != 'postconn':
-        if not grabpidfile():
-            sys.stderr.write("pidfile is taken, exiting.\n")
-            sys.exit(2)
-        gconf.pid_file_owned = True
-
-    if kw.get('go_daemon') == 'should':
-        x, y = os.pipe()
-        gconf.cpid = os.fork()
-        if gconf.cpid:
-            os.close(x)
-            sys.exit()
-        os.close(y)
-        os.setsid()
-        dn = os.open(os.devnull, os.O_RDWR)
-        for f in (sys.stdin, sys.stdout, sys.stderr):
-            os.dup2(dn, f.fileno())
-        if getattr(gconf, 'pid_file', None):
-            if not grabpidfile(gconf.pid_file + '.tmp'):
-                raise GsyncdError("cannot grab temporary pidfile")
-            os.rename(gconf.pid_file + '.tmp', gconf.pid_file)
-        # wait for parent to terminate
-        # so we can start up with
-        # no messing from the dirty
-        # ol' bustard
-        select((x,), (), ())
-        os.close(x)
-
-    GLogger._gsyncd_loginit(**kw)
-
-def main():
-    """main routine, signal/exception handling boilerplates"""
-    gconf.starttime = time.time()
-    set_term_handler()
-    GLogger.setup()
-    excont = FreeObject(exval = 0)
-    try:
-        try:
-            main_i()
-        except:
-            log_raise_exception(excont)
-    finally:
-        finalize(exval = excont.exval)
-
-def main_i():
-    """internal main routine
-
-    parse command line, decide what action will be taken;
-    we can either:
-    - query/manipulate configuration
-    - format gsyncd urls using gsyncd's url parsing engine
-    - start service in following modes, in given stages:
-      - monitor: startup(), monitor()
-      - master: startup(), connect_remote(), connect(), service_loop()
-      - slave: startup(), connect(), service_loop()
-    """
-    rconf = {'go_daemon': 'should'}
-
-    def store_abs(opt, optstr, val, parser):
-        if val and val != '-':
-            val = os.path.abspath(val)
-        setattr(parser.values, opt.dest, val)
-    def store_local(opt, optstr, val, parser):
-        rconf[opt.dest] = val
-    def store_local_curry(val):
-        return lambda o, oo, vx, p: store_local(o, oo, val, p)
-    def store_local_obj(op, dmake):
-        return lambda o, oo, vx, p: store_local(o, oo, FreeObject(op=op, **dmake(vx)), p)
-
-    op = OptionParser(usage="%prog [options...] <master> <slave>", version="%prog 0.0.1")
-    op.add_option('--gluster-command-dir', metavar='DIR',   default='')
-    op.add_option('--gluster-log-file',    metavar='LOGF',  default=os.devnull, type=str, action='callback', callback=store_abs)
-    op.add_option('--gluster-log-level',   metavar='LVL')
-    op.add_option('--gluster-params',      metavar='PRMS',  default='')
-    op.add_option('--gluster-cli-options', metavar='OPTS',  default='--log-file=-')
-    op.add_option('--mountbroker',         metavar='LABEL')
-    op.add_option('-p', '--pid-file',      metavar='PIDF',  type=str, action='callback', callback=store_abs)
-    op.add_option('-l', '--log-file',      metavar='LOGF',  type=str, action='callback', callback=store_abs)
-    op.add_option('--log-file-mbr',        metavar='LOGF',  type=str, action='callback', callback=store_abs)
-    op.add_option('--state-file',          metavar='STATF', type=str, action='callback', callback=store_abs)
-    op.add_option('--ignore-deletes',      default=False, action='store_true')
-    op.add_option('--use-rsync-xattrs',    default=False, action='store_true')
-    op.add_option('-L', '--log-level',     metavar='LVL')
-    op.add_option('-r', '--remote-gsyncd', metavar='CMD',   default=os.path.abspath(sys.argv[0]))
-    op.add_option('--volume-id',           metavar='UUID')
-    op.add_option('--session-owner',       metavar='ID')
-    op.add_option('-s', '--ssh-command',   metavar='CMD',   default='ssh')
-    op.add_option('--rsync-command',       metavar='CMD',   default='rsync')
-    op.add_option('--rsync-options',       metavar='OPTS',  default='--sparse')
-    op.add_option('--rsync-ssh-options',   metavar='OPTS',  default='--compress')
-    op.add_option('--timeout',             metavar='SEC',   type=int, default=120)
-    op.add_option('--connection-timeout',  metavar='SEC',   type=int, default=60, help=SUPPRESS_HELP)
-    op.add_option('--sync-jobs',           metavar='N',     type=int, default=3)
-    op.add_option('--turns',               metavar='N',     type=int, default=0, help=SUPPRESS_HELP)
-    op.add_option('--allow-network',       metavar='IPS',   default='')
-    op.add_option('--socketdir',           metavar='DIR')
-    op.add_option('--state-socket-unencoded', metavar='SOCKF', type=str, action='callback', callback=store_abs)
-    op.add_option('--checkpoint',          metavar='LABEL', default='')
-    # tunables for failover/failback mechanism:
-    # None   - gsyncd behaves as normal
-    # blind  - gsyncd works with xtime pairs to identify
-    #          candidates for synchronization
-    # wrapup - same as normal mode but does not assign
-    #          xtimes to orphaned files
-    # see crawl() for usage of the above tunables
-    op.add_option('--special-sync-mode', type=str, help=SUPPRESS_HELP)
-
-    op.add_option('-c', '--config-file',   metavar='CONF',  type=str, action='callback', callback=store_local)
-    # duh. need to specify dest or value will be mapped to None :S
-    op.add_option('--monitor', dest='monitor', action='callback', callback=store_local_curry(True))
-    op.add_option('--feedback-fd', dest='feedback_fd', type=int, help=SUPPRESS_HELP, action='callback', callback=store_local)
-    op.add_option('--listen', dest='listen', help=SUPPRESS_HELP,      action='callback', callback=store_local_curry(True))
-    op.add_option('-N', '--no-daemon', dest="go_daemon",    action='callback', callback=store_local_curry('dont'))
-    op.add_option('--debug', dest="go_daemon",              action='callback', callback=lambda *a: (store_local_curry('dont')(*a),
-                                                                                                    setattr(a[-1].values, 'log_file', '-'),
-                                                                                                    setattr(a[-1].values, 'log_level', 'DEBUG'))),
-
-    for a in ('check', 'get'):
-        op.add_option('--config-' + a,      metavar='OPT',  type=str, dest='config', action='callback',
-                      callback=store_local_obj(a, lambda vx: {'opt': vx}))
-    op.add_option('--config-get-all', dest='config', action='callback', callback=store_local_obj('get', lambda vx: {'opt': None}))
-    for m in ('', '-rx', '-glob'):
-        # call this code 'Pythonic' eh?
-        # have to define a one-shot local function to be able to inject (a value depending on the)
-        # iteration variable into the inner lambda
-        def conf_mod_opt_regex_variant(rx):
-            op.add_option('--config-set' + m,   metavar='OPT VAL', type=str, nargs=2, dest='config', action='callback',
-                          callback=store_local_obj('set', lambda vx: {'opt': vx[0], 'val': vx[1], 'rx': rx}))
-            op.add_option('--config-del' + m,   metavar='OPT',  type=str, dest='config', action='callback',
-                          callback=store_local_obj('del', lambda vx: {'opt': vx, 'rx': rx}))
-        conf_mod_opt_regex_variant(m and m[1:] or False)
-
-    op.add_option('--normalize-url',           dest='url_print', action='callback', callback=store_local_curry('normal'))
-    op.add_option('--canonicalize-url',        dest='url_print', action='callback', callback=store_local_curry('canon'))
-    op.add_option('--canonicalize-escape-url', dest='url_print', action='callback', callback=store_local_curry('canon_esc'))
-
-    tunables = [ norm(o.get_opt_string()[2:]) for o in op.option_list if o.callback in (store_abs, 'store_true', None) and o.get_opt_string() not in ('--version', '--help') ]
-    remote_tunables = [ 'listen', 'go_daemon', 'timeout', 'session_owner', 'config_file', 'use_rsync_xattrs' ]
-    rq_remote_tunables = { 'listen': True }
-
-    # precedence for sources of values: 1) commandline, 2) cfg file, 3) defaults
-    # -- for this to work out we need to tell apart defaults from explicitly set
-    # options... so churn out the defaults here and call the parser with virgin
-    # values container.
-    defaults = op.get_default_values()
-    opts, args = op.parse_args(values=optparse.Values())
-    confdata = rconf.get('config')
-    if not (len(args) == 2 or \
-            (len(args) == 1 and rconf.get('listen')) or \
-            (len(args) <= 2 and confdata) or \
-            rconf.get('url_print')):
-        sys.stderr.write("error: incorrect number of arguments\n\n")
-        sys.stderr.write(op.get_usage() + "\n")
-        sys.exit(1)
-
-    restricted = os.getenv('_GSYNCD_RESTRICTED_')
-
-    if restricted:
-        allopts = {}
-        allopts.update(opts.__dict__)
-        allopts.update(rconf)
-        bannedtuns = set(allopts.keys()) - set(remote_tunables)
-        if bannedtuns:
-            raise GsyncdError('following tunables cannot be set with restricted SSH invocaton: ' + \
-                              ', '.join(bannedtuns))
-        for k, v in rq_remote_tunables.items():
-            if not k in allopts or allopts[k] != v:
-                raise GsyncdError('tunable %s is not set to value %s required for restricted SSH invocaton' % \
-                                  (k, v))
-
-    confrx = getattr(confdata, 'rx', None)
-    if confrx:
-        # peers are regexen, don't try to parse them
-        if confrx == 'glob':
-            args = [ '\A' + fnmatch.translate(a) for a in args ]
-        canon_peers = args
-        namedict = {}
-    else:
-        rscs = [resource.parse_url(u) for u in args]
-        dc = rconf.get('url_print')
-        if dc:
-            for r in rscs:
-                print(r.get_url(**{'normal': {},
-                                   'canon': {'canonical': True},
-                                   'canon_esc': {'canonical': True, 'escaped': True}}[dc]))
-            return
-        local = remote = None
-        if rscs:
-            local = rscs[0]
-            if len(rscs) > 1:
-                remote = rscs[1]
-            if not local.can_connect_to(remote):
-                raise GsyncdError("%s cannot work with %s" % (local.path, remote and remote.path))
-        pa = ([], [], [])
-        urlprms = ({}, {'canonical': True}, {'canonical': True, 'escaped': True})
-        for x in rscs:
-            for i in range(len(pa)):
-                pa[i].append(x.get_url(**urlprms[i]))
-        peers, canon_peers, canon_esc_peers = pa
-        # creating the namedict, a dict representing various ways of referring to / repreenting
-        # peers to be fillable in config templates
-        mods = (lambda x: x, lambda x: x[0].upper() + x[1:], lambda x: 'e' + x[0].upper() + x[1:])
-        if remote:
-            rmap = { local: ('local', 'master'), remote: ('remote', 'slave') }
-        else:
-            rmap = { local: ('local', 'slave') }
-        namedict = {}
-        for i in range(len(rscs)):
-            x = rscs[i]
-            for name in rmap[x]:
-                for j in range(3):
-                    namedict[mods[j](name)] = pa[j][i]
-                if x.scheme == 'gluster':
-                    namedict[name + 'vol'] = x.volume
-    if not 'config_file' in rconf:
-        rconf['config_file'] = os.path.join(os.path.dirname(sys.argv[0]), "conf/gsyncd.conf")
-    gcnf = GConffile(rconf['config_file'], canon_peers, defaults.__dict__, opts.__dict__, namedict)
-
-    checkpoint_change = False
-    if confdata:
-        opt_ok = norm(confdata.opt) in tunables + [None]
-        if confdata.op == 'check':
-            if opt_ok:
-                sys.exit(0)
-            else:
-                sys.exit(1)
-        elif not opt_ok:
-            raise GsyncdError("not a valid option: " + confdata.opt)
-        if confdata.op == 'get':
-            gcnf.get(confdata.opt)
-        elif confdata.op == 'set':
-            gcnf.set(confdata.opt, confdata.val, confdata.rx)
-        elif confdata.op == 'del':
-            gcnf.delete(confdata.opt, confdata.rx)
-        # when modifying checkpoint, it's important to make a log
-        # of that, so in that case we go on to set up logging even
-        # if its just config invocation
-        if confdata.opt == 'checkpoint' and confdata.op in ('set', 'del') and \
-           not confdata.rx:
-            checkpoint_change = True
-        if not checkpoint_change:
-            return
-
-    gconf.__dict__.update(defaults.__dict__)
-    gcnf.update_to(gconf.__dict__)
-    gconf.__dict__.update(opts.__dict__)
-    gconf.configinterface = gcnf
-
-    if restricted and gconf.allow_network:
-        ssh_conn = os.getenv('SSH_CONNECTION')
-        if not ssh_conn:
-            #legacy env var
-            ssh_conn = os.getenv('SSH_CLIENT')
-        if ssh_conn:
-            allowed_networks = [ IPNetwork(a) for a in gconf.allow_network.split(',') ]
-            client_ip = IPAddress(ssh_conn.split()[0])
-            allowed = False
-            for nw in allowed_networks:
-                if client_ip in nw:
-                    allowed = True
-                    break
-            if not allowed:
-                raise GsyncdError("client IP address is not allowed")
-
-    ffd = rconf.get('feedback_fd')
-    if ffd:
-        fcntl.fcntl(ffd, fcntl.F_SETFD, fcntl.FD_CLOEXEC)
-
-    #normalize loglevel
-    lvl0 = gconf.log_level
-    if isinstance(lvl0, str):
-        lvl1 = lvl0.upper()
-        lvl2 = logging.getLevelName(lvl1)
-        # I have _never_ _ever_ seen such an utterly braindead
-        # error condition
-        if lvl2 == "Level " + lvl1:
-            raise GsyncdError('cannot recognize log level "%s"' % lvl0)
-        gconf.log_level = lvl2
-
-    if not privileged() and gconf.log_file_mbr:
-        gconf.log_file = gconf.log_file_mbr
-
-    if checkpoint_change:
-        try:
-            GLogger._gsyncd_loginit(log_file=gconf.log_file, label='conf')
-            if confdata.op == 'set':
-                logging.info('checkpoint %s set' % confdata.val)
-            elif confdata.op == 'del':
-                logging.info('checkpoint info was reset')
-        except IOError:
-            if sys.exc_info()[1].errno == ENOENT:
-                # directory of log path is not present,
-                # which happens if we get here from
-                # a peer-multiplexed "config-set checkpoint"
-                # (as that directory is created only on the
-                # original node)
-                pass
-            else:
-                raise
-        return
-
-    go_daemon = rconf['go_daemon']
-    be_monitor = rconf.get('monitor')
-
-    if not be_monitor and isinstance(remote, resource.SSH) and \
-       go_daemon == 'should':
-        go_daemon = 'postconn'
-        log_file = None
-    else:
-        log_file = gconf.log_file
-    if be_monitor:
-        label = 'monitor'
-    elif remote:
-        #master
-        label = ''
-    else:
-        label = 'slave'
-    startup(go_daemon=go_daemon, log_file=log_file, label=label)
-
-    if be_monitor:
-        return monitor()
-
-    logging.info("syncing: %s" % " -> ".join(peers))
-    resource.Popen.init_errhandler()
-    if remote:
-        go_daemon = remote.connect_remote(go_daemon=go_daemon)
-        if go_daemon:
-            startup(go_daemon=go_daemon, log_file=gconf.log_file)
-            # complete remote connection in child
-            remote.connect_remote(go_daemon='done')
-    local.connect()
-    if ffd:
-        os.close(ffd)
-    local.service_loop(*[r for r in [remote] if r])
-
-
-if __name__ == "__main__":
-    main()
diff --git a/xlators/features/marker/utils/syncdaemon/libcxattr.py b/xlators/features/marker/utils/syncdaemon/libcxattr.py
deleted file mode 100644
index f0a9d2292..000000000
--- a/xlators/features/marker/utils/syncdaemon/libcxattr.py
+++ /dev/null
@@ -1,72 +0,0 @@
-import os
-from ctypes import *
-from ctypes.util import find_library
-
-class Xattr(object):
-    """singleton that wraps the extended attribues system
-       interface for python using ctypes
-
-       Just implement it to the degree we need it, in particular
-       - we need just the l*xattr variants, ie. we never want symlinks to be
-         followed
-       - don't need size discovery for getxattr, as we always know the exact
-         sizes we expect
-    """
-
-    libc = CDLL(find_library("libc"))
-
-    @classmethod
-    def geterrno(cls):
-        return c_int.in_dll(cls.libc, 'errno').value
-
-    @classmethod
-    def raise_oserr(cls):
-        errn = cls.geterrno()
-        raise OSError(errn, os.strerror(errn))
-
-    @classmethod
-    def _query_xattr(cls, path, siz, syscall, *a):
-        if siz:
-            buf = create_string_buffer('\0' * siz)
-        else:
-            buf = None
-        ret = getattr(cls.libc, syscall)(*((path,) + a + (buf, siz)))
-        if ret == -1:
-            cls.raise_oserr()
-        if siz:
-            return buf.raw[:ret]
-        else:
-            return ret
-
-    @classmethod
-    def lgetxattr(cls, path, attr, siz=0):
-        return cls._query_xattr( path, siz, 'lgetxattr', attr)
-
-    @classmethod
-    def llistxattr(cls, path, siz=0):
-        ret = cls._query_xattr(path, siz, 'llistxattr')
-        if isinstance(ret, str):
-            ret = ret.split('\0')
-        return ret
-
-    @classmethod
-    def lsetxattr(cls, path, attr, val):
-        ret = cls.libc.lsetxattr(path, attr, val, len(val), 0)
-        if ret == -1:
-            cls.raise_oserr()
-
-    @classmethod
-    def lremovexattr(cls, path, attr):
-        ret = cls.libc.lremovexattr(path, attr)
-        if ret == -1:
-            cls.raise_oserr()
-
-    @classmethod
-    def llistxattr_buf(cls, path):
-        """listxattr variant with size discovery"""
-        size = cls.llistxattr(path)
-        if size == -1:
-            cls.raise_oserr()
-        if size == 0:
-            return []
-        return cls.llistxattr(path, size)
diff --git a/xlators/features/marker/utils/syncdaemon/master.py b/xlators/features/marker/utils/syncdaemon/master.py
deleted file mode 100644
index f903f3059..000000000
--- a/xlators/features/marker/utils/syncdaemon/master.py
+++ /dev/null
@@ -1,961 +0,0 @@
-import os
-import sys
-import time
-import stat
-import random
-import signal
-import logging
-import socket
-import errno
-import re
-from errno import ENOENT, ENODATA, EPIPE
-from threading import currentThread, Condition, Lock
-from datetime import datetime
-try:
-    from hashlib import md5 as md5
-except ImportError:
-    # py 2.4
-    from md5 import new as md5
-
-from gconf import gconf
-from syncdutils import FreeObject, Thread, GsyncdError, boolify, \
-                       escape, unescape, select
-
-URXTIME = (-1, 0)
-
-# Utility functions to help us to get to closer proximity
-# of the DRY principle (no, don't look for elevated or
-# perspectivistic things here)
-
-def _xtime_now():
-    t = time.time()
-    sec = int(t)
-    nsec = int((t - sec) * 1000000)
-    return (sec, nsec)
-
-def _volinfo_hook_relax_foreign(self):
-    volinfo_sys = self.get_sys_volinfo()
-    fgn_vi = volinfo_sys[self.KFGN]
-    if fgn_vi:
-        expiry = fgn_vi['timeout'] - int(time.time()) + 1
-        logging.info('foreign volume info found, waiting %d sec for expiry' % \
-                     expiry)
-        time.sleep(expiry)
-        volinfo_sys = self.get_sys_volinfo()
-    self.volinfo_state, state_change = self.volinfo_state_machine(self.volinfo_state,
-                                                                  volinfo_sys)
-    if self.inter_master:
-        raise GsyncdError("cannot be intermediate master in special mode")
-    return (volinfo_sys, state_change)
-
-
-# The API!
-
-def gmaster_builder():
-    """produce the GMaster class variant corresponding
-       to sync mode"""
-    this = sys.modules[__name__]
-    modemixin = gconf.special_sync_mode
-    if not modemixin:
-        modemixin = 'normal'
-    logging.info('setting up master for %s sync mode' % modemixin)
-    modemixin = getattr(this, modemixin.capitalize() + 'Mixin')
-    sendmarkmixin = boolify(gconf.use_rsync_xattrs) and SendmarkRsyncMixin or SendmarkNormalMixin
-    purgemixin = boolify(gconf.ignore_deletes) and PurgeNoopMixin or PurgeNormalMixin
-    class _GMaster(GMasterBase, modemixin, sendmarkmixin, purgemixin):
-        pass
-    return _GMaster
-
-
-# Mixin classes that implement the data format
-# and logic particularities of the certain
-# sync modes
-
-class NormalMixin(object):
-    """normal geo-rep behavior"""
-
-    minus_infinity = URXTIME
-
-    # following staticmethods ideally would be
-    # methods of an xtime object (in particular,
-    # implementing the hooks needed for comparison
-    # operators), but at this point we don't yet
-    # have a dedicated xtime class
-
-    @staticmethod
-    def serialize_xtime(xt):
-        return "%d.%d" % tuple(xt)
-
-    @staticmethod
-    def deserialize_xtime(xt):
-        return tuple(int(x) for x in xt.split("."))
-
-    @staticmethod
-    def native_xtime(xt):
-        return xt
-
-    @staticmethod
-    def xtime_geq(xt0, xt1):
-        return xt0 >= xt1
-
-    def make_xtime_opts(self, is_master, opts):
-        if not 'create' in opts:
-            opts['create'] = is_master and not self.inter_master
-        if not 'default_xtime' in opts:
-            if is_master and self.inter_master:
-                opts['default_xtime'] = ENODATA
-            else:
-                opts['default_xtime'] = URXTIME
-
-    def xtime_low(self, server, path, **opts):
-        xt = server.xtime(path, self.uuid)
-        if isinstance(xt, int) and xt != ENODATA:
-            return xt
-        if xt == ENODATA or xt < self.volmark:
-            if opts['create']:
-                xt = _xtime_now()
-                server.set_xtime(path, self.uuid, xt)
-            else:
-                xt = opts['default_xtime']
-        return xt
-
-    def keepalive_payload_hook(self, timo, gap):
-        # first grab a reference as self.volinfo
-        # can be changed in main thread
-        vi = self.volinfo
-        if vi:
-            # then have a private copy which we can mod
-            vi = vi.copy()
-            vi['timeout'] = int(time.time()) + timo
-        else:
-            # send keep-alives more frequently to
-            # avoid a delay in announcing our volume info
-            # to slave if it becomes established in the
-            # meantime
-            gap = min(10, gap)
-        return (vi, gap)
-
-    def volinfo_hook(self):
-        volinfo_sys = self.get_sys_volinfo()
-        self.volinfo_state, state_change = self.volinfo_state_machine(self.volinfo_state,
-                                                                      volinfo_sys)
-        return (volinfo_sys, state_change)
-
-    def xtime_reversion_hook(self, path, xtl, xtr):
-        if xtr > xtl:
-            raise GsyncdError("timestamp corruption for " + path)
-
-    def need_sync(self, e, xte, xtrd):
-        return xte > xtrd
-
-    def set_slave_xtime(self, path, mark):
-        self.slave.server.set_xtime(path, self.uuid, mark)
-
-class WrapupMixin(NormalMixin):
-    """a variant that differs from normal in terms
-       of ignoring non-indexed files"""
-
-    @staticmethod
-    def make_xtime_opts(is_master, opts):
-        if not 'create' in opts:
-            opts['create'] = False
-        if not 'default_xtime' in opts:
-            opts['default_xtime'] = URXTIME
-
-    @staticmethod
-    def keepalive_payload_hook(timo, gap):
-        return (None, gap)
-
-    def volinfo_hook(self):
-        return _volinfo_hook_relax_foreign(self)
-
-class BlindMixin(object):
-    """Geo-rep flavor using vectored xtime.
-
-    Coordinates are the master, slave uuid pair;
-    in master coordinate behavior is normal,
-    in slave coordinate we force synchronization
-    on any value difference (these are in disjunctive
-    relation, ie. if either orders the entry to be
-    synced, it shall be synced.
-    """
-
-    minus_infinity = (URXTIME, None)
-
-    @staticmethod
-    def serialize_xtime(xt):
-        a = []
-        for x in xt:
-            if not x:
-                x = ('None', '')
-            a.extend(x)
-        return '.'.join(str(n) for n in a)
-
-    @staticmethod
-    def deserialize_xtime(xt):
-        a = xt.split(".")
-        a = (tuple(a[0:2]), tuple(a[3:4]))
-        b = []
-        for p in a:
-            if p[0] == 'None':
-                p = None
-            else:
-                p = tuple(int(x) for x in p)
-            b.append(p)
-        return tuple(b)
-
-    @staticmethod
-    def native_xtime(xt):
-        return xt[0]
-
-    @staticmethod
-    def xtime_geq(xt0, xt1):
-        return (not xt1[0] or xt0[0] >= xt1[0]) and \
-               (not xt1[1] or xt0[1] >= xt1[1])
-
-    @property
-    def ruuid(self):
-        if self.volinfo_r:
-            return self.volinfo_r['uuid']
-
-    @staticmethod
-    def make_xtime_opts(is_master, opts):
-        if not 'create' in opts:
-            opts['create'] = is_master
-        if not 'default_xtime' in opts:
-            opts['default_xtime'] = URXTIME
-
-    def xtime_low(self, server, path, **opts):
-        xtd = server.xtime_vec(path, self.uuid, self.ruuid)
-        if isinstance(xtd, int):
-            return xtd
-        xt = (xtd[self.uuid], xtd[self.ruuid])
-        if not xt[1] and (not xt[0] or xt[0] < self.volmark):
-            if opts['create']:
-                # not expected, but can happen if file originates
-                # from interrupted gsyncd transfer
-                logging.warn('have to fix up missing xtime on ' + path)
-                xt0 = _xtime_now()
-                server.set_xtime(path, self.uuid, xt0)
-            else:
-                xt0 = opts['default_xtime']
-            xt = (xt0, xt[1])
-        return xt
-
-    @staticmethod
-    def keepalive_payload_hook(timo, gap):
-        return (None, gap)
-
-    def volinfo_hook(self):
-        res = _volinfo_hook_relax_foreign(self)
-        volinfo_r_new = self.slave.server.native_volume_info()
-        if volinfo_r_new['retval']:
-            raise GsyncdError("slave is corrupt")
-        if getattr(self, 'volinfo_r', None):
-            if self.volinfo_r['uuid'] != volinfo_r_new['uuid']:
-                raise GsyncdError("uuid mismatch on slave")
-        self.volinfo_r = volinfo_r_new
-        return res
-
-    def xtime_reversion_hook(self, path, xtl, xtr):
-        if not isinstance(xtr[0], int) and \
-          (isinstance(xtl[0], int) or xtr[0] > xtl[0]):
-            raise GsyncdError("timestamp corruption for " + path)
-
-    def need_sync(self, e, xte, xtrd):
-        if xte[0]:
-            if not xtrd[0] or xte[0] > xtrd[0]:
-                # there is outstanding diff at 0th pos,
-                # we can short-cut to true
-                return True
-        # we arrived to this point by either of these
-        # two possiblilites:
-        # - no outstanding difference at 0th pos,
-        #   wanna see 1st pos if he raises veto
-        #   against "no need to sync" proposal
-        # - no data at 0th pos, 1st pos will have
-        #   to decide (due to xtime assignment,
-        #   in this case 1st pos does carry data
-        #   -- iow, if 1st pos did not have data,
-        #   and 0th neither, 0th would have been
-        #   force-feeded)
-        if not xte[1]:
-            # no data, no veto
-            return False
-        # the hard work: for 1st pos,
-        # the conduct is fetch corresponding
-        # slave data and do a "blind" comparison
-        # (ie. do not care who is newer, we trigger
-        # sync on non-identical xitmes)
-        xtr = self.xtime(e, self.slave)
-        return isinstance(xtr, int) or xte[1] != xtr[1]
-
-    def set_slave_xtime(self, path, mark):
-        xtd = {}
-        for (u, t) in zip((self.uuid, self.ruuid), mark):
-            if t:
-                xtd[u] = t
-        self.slave.server.set_xtime_vec(path, xtd)
-
-
-# Further mixins for certain tunable behaviors
-
-class SendmarkNormalMixin(object):
-
-    def sendmark_regular(self, *a, **kw):
-        return self.sendmark(*a, **kw)
-
-class SendmarkRsyncMixin(object):
-
-    def sendmark_regular(self, *a, **kw):
-        pass
-
-
-class PurgeNormalMixin(object):
-
-    def purge_missing(self, path, names):
-        self.slave.server.purge(path, names)
-
-class PurgeNoopMixin(object):
-
-    def purge_missing(self, path, names):
-        pass
-
-
-
-class GMasterBase(object):
-    """abstract class impementling master role"""
-
-    KFGN = 0
-    KNAT = 1
-
-    def get_sys_volinfo(self):
-        """query volume marks on fs root
-
-        err out on multiple foreign masters
-        """
-        fgn_vis, nat_vi = self.master.server.foreign_volume_infos(), \
-                          self.master.server.native_volume_info()
-        fgn_vi = None
-        if fgn_vis:
-            if len(fgn_vis) > 1:
-                raise GsyncdError("cannot work with multiple foreign masters")
-            fgn_vi = fgn_vis[0]
-        return fgn_vi, nat_vi
-
-    @property
-    def uuid(self):
-        if self.volinfo:
-            return self.volinfo['uuid']
-
-    @property
-    def volmark(self):
-        if self.volinfo:
-            return self.volinfo['volume_mark']
-
-    @property
-    def inter_master(self):
-        """decide if we are an intermediate master
-        in a cascading setup
-        """
-        return self.volinfo_state[self.KFGN] and True or False
-
-    def xtime(self, path, *a, **opts):
-        """get amended xtime
-
-        as of amending, we can create missing xtime, or
-        determine a valid value if what we get is expired
-        (as of the volume mark expiry); way of amendig
-        depends on @opts and on subject of query (master
-        or slave).
-        """
-        if a:
-            rsc = a[0]
-        else:
-            rsc = self.master
-        self.make_xtime_opts(rsc == self.master, opts)
-        return self.xtime_low(rsc.server, path, **opts)
-
-    def __init__(self, master, slave):
-        self.master = master
-        self.slave = slave
-        self.jobtab = {}
-        self.syncer = Syncer(slave)
-        # crawls vs. turns:
-        # - self.crawls is simply the number of crawl() invocations on root
-        # - one turn is a maximal consecutive sequence of crawls so that each
-        #   crawl in it detects a change to be synced
-        # - self.turns is the number of turns since start
-        # - self.total_turns is a limit so that if self.turns reaches it, then
-        #   we exit (for diagnostic purposes)
-        # so, eg., if the master fs changes unceasingly, self.turns will remain 0.
-        self.crawls = 0
-        self.turns = 0
-        self.total_turns = int(gconf.turns)
-        self.lastreport = {'crawls': 0, 'turns': 0}
-        self.start = None
-        self.change_seen = None
-        self.syncTime=0
-        self.lastSyncTime=0
-        self.crawlStartTime=0
-        self.crawlTime=0
-        self.filesSynced=0
-        self.bytesSynced=0
-        # the authoritative (foreign, native) volinfo pair
-        # which lets us deduce what to do when we refetch
-        # the volinfos from system
-        uuid_preset = getattr(gconf, 'volume_id', None)
-        self.volinfo_state = (uuid_preset and {'uuid': uuid_preset}, None)
-        # the actual volinfo we make use of
-        self.volinfo = None
-        self.terminate = False
-        self.checkpoint_thread = None
-
-    @classmethod
-    def _checkpt_param(cls, chkpt, prm, xtimish=True):
-        """use config backend to lookup a parameter belonging to
-           checkpoint @chkpt"""
-        cprm = getattr(gconf, 'checkpoint_' + prm, None)
-        if not cprm:
-            return
-        chkpt_mapped, val = cprm.split(':', 1)
-        if unescape(chkpt_mapped) != chkpt:
-            return
-        if xtimish:
-            val = cls.deserialize_xtime(val)
-        return val
-
-    @classmethod
-    def _set_checkpt_param(cls, chkpt, prm, val, xtimish=True):
-        """use config backend to store a parameter associated
-           with checkpoint @chkpt"""
-        if xtimish:
-            val = cls.serialize_xtime(val)
-        gconf.configinterface.set('checkpoint_' + prm, "%s:%s" % (escape(chkpt), val))
-
-    @staticmethod
-    def humantime(*tpair):
-        """format xtime-like (sec, nsec) pair to human readable format"""
-        ts = datetime.fromtimestamp(float('.'.join(str(n) for n in tpair))).\
-               strftime("%Y-%m-%d %H:%M:%S")
-        if len(tpair) > 1:
-            ts += '.' + str(tpair[1])
-        return ts
-
-    def get_extra_info(self):
-        str_info="\nFile synced : %d" %(self.filesSynced)
-        str_info+="\nBytes Synced : %d KB" %(self.syncer.bytesSynced)
-        str_info+="\nSync Time : %f seconds" %(self.syncTime)
-        self.crawlTime=datetime.now()-self.crawlStartTime
-        years , days =divmod(self.crawlTime.days,365.25)
-        years=int(years)
-        days=int(days)
-
-        date=""
-        m, s = divmod(self.crawlTime.seconds, 60)
-        h, m = divmod(m, 60)
-
-        if years!=0 :
-                date+=str(years)+" year "
-        if days!=0 :
-                date+=str(days)+" day "
-        if h!=0 :
-                date+=str(h)+" H : "
-        if m!=0 or h!=0 :
-                date+=str(m)+" M : "
-
-        date+=str(s)+" S"
-        self.crawlTime=date
-        str_info+="\nCrawl Time : %s" %(str(self.crawlTime))
-        str_info+="\n\0"
-        return str_info
-
-    def checkpt_service(self, chan, chkpt, tgt):
-        """checkpoint service loop
-
-        monitor and verify checkpoint status for @chkpt, and listen
-        for incoming requests for whom we serve a pretty-formatted
-        status report"""
-        if not chkpt:
-            # dummy loop for the case when there is no checkpt set
-            while True:
-                select([chan], [], [])
-                conn, _ = chan.accept()
-                conn.send(self.get_extra_info())
-                conn.close()
-        completed = self._checkpt_param(chkpt, 'completed', xtimish=False)
-        if completed:
-            completed = tuple(int(x) for x in completed.split('.'))
-        while True:
-            s,_,_ = select([chan], [], [], (not completed) and 5 or None)
-            # either request made and we re-check to not
-            # give back stale data, or we still hunting for completion
-            if self.native_xtime(tgt) and self.native_xtime(tgt) < self.volmark:
-                # indexing has been reset since setting the checkpoint
-                status = "is invalid"
-            else:
-                xtr = self.xtime('.', self.slave)
-                if isinstance(xtr, int):
-                    raise GsyncdError("slave root directory is unaccessible (%s)",
-                                      os.strerror(xtr))
-                ncompleted = self.xtime_geq(xtr, tgt)
-                if completed and not ncompleted: # stale data
-                    logging.warn("completion time %s for checkpoint %s became stale" % \
-                                 (self.humantime(*completed), chkpt))
-                    completed = None
-                    gconf.confdata.delete('checkpoint-completed')
-                if ncompleted and not completed: # just reaching completion
-                    completed = "%.6f" % time.time()
-                    self._set_checkpt_param(chkpt, 'completed', completed, xtimish=False)
-                    completed = tuple(int(x) for x in completed.split('.'))
-                    logging.info("checkpoint %s completed" % chkpt)
-                status = completed and \
-                  "completed at " + self.humantime(completed[0]) or \
-                  "not reached yet"
-            if s:
-                conn = None
-                try:
-                    conn, _ = chan.accept()
-                    try:
-                        conn.send("  | checkpoint %s %s %s" % (chkpt, status,self.get_extra_info()))
-                    except:
-                        exc = sys.exc_info()[1]
-                        if (isinstance(exc, OSError) or isinstance(exc, IOError)) and \
-                           exc.errno == EPIPE:
-                            logging.debug('checkpoint client disconnected')
-                        else:
-                            raise
-                finally:
-                    if conn:
-                        conn.close()
-
-    def start_checkpoint_thread(self):
-        """prepare and start checkpoint service"""
-        if self.checkpoint_thread or not (
-          getattr(gconf, 'state_socket_unencoded', None) and getattr(gconf, 'socketdir', None)
-        ):
-            return
-        chan = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
-        state_socket = os.path.join(gconf.socketdir, md5(gconf.state_socket_unencoded).hexdigest() + ".socket")
-        try:
-            os.unlink(state_socket)
-        except:
-            if sys.exc_info()[0] == OSError:
-                pass
-        chan.bind(state_socket)
-        chan.listen(1)
-        checkpt_tgt = None
-        if gconf.checkpoint:
-            checkpt_tgt = self._checkpt_param(gconf.checkpoint, 'target')
-            if not checkpt_tgt:
-                checkpt_tgt = self.xtime('.')
-                if isinstance(checkpt_tgt, int):
-                    raise GsyncdError("master root directory is unaccessible (%s)",
-                                      os.strerror(checkpt_tgt))
-                self._set_checkpt_param(gconf.checkpoint, 'target', checkpt_tgt)
-            logging.debug("checkpoint target %s has been determined for checkpoint %s" % \
-                          (repr(checkpt_tgt), gconf.checkpoint))
-        t = Thread(target=self.checkpt_service, args=(chan, gconf.checkpoint, checkpt_tgt))
-        t.start()
-        self.checkpoint_thread = t
-
-    def crawl_loop(self):
-        """start the keep-alive thread and iterate .crawl"""
-        timo = int(gconf.timeout or 0)
-        if timo > 0:
-            def keep_alive():
-                while True:
-                    vi, gap = self.keepalive_payload_hook(timo, timo * 0.5)
-                    self.slave.server.keep_alive(vi)
-                    time.sleep(gap)
-            t = Thread(target=keep_alive)
-            t.start()
-        self.lastreport['time'] = time.time()
-        self.crawlStartTime=datetime.now()
-        while not self.terminate:
-            self.crawl()
-
-    def add_job(self, path, label, job, *a, **kw):
-        """insert @job function to job table at @path with @label"""
-        if self.jobtab.get(path) == None:
-            self.jobtab[path] = []
-        self.jobtab[path].append((label, a, lambda : job(*a, **kw)))
-
-    def add_failjob(self, path, label):
-        """invoke .add_job with a job that does nothing just fails"""
-        logging.debug('salvaged: ' + label)
-        self.add_job(path, label, lambda: False)
-
-    def wait(self, path, *args):
-        """perform jobs registered for @path
-
-        Reset jobtab entry for @path,
-        determine success as the conjuction of
-        success of all the jobs. In case of
-        success, call .sendmark on @path
-        """
-        jobs = self.jobtab.pop(path, [])
-        succeed = True
-        for j in jobs:
-            ret = j[-1]()
-            if not ret:
-                succeed = False
-        if succeed:
-            self.sendmark(path, *args)
-        return succeed
-
-    def sendmark(self, path, mark, adct=None):
-        """update slave side xtime for @path to master side xtime
-
-        also can send a setattr payload (see Server.setattr).
-        """
-        if adct:
-            self.slave.server.setattr(path, adct)
-        self.set_slave_xtime(path, mark)
-
-    @staticmethod
-    def volinfo_state_machine(volinfo_state, volinfo_sys):
-        """compute new volinfo_state from old one and incoming
-           as of current system state, also indicating if there was a
-           change regarding which volume mark is the authoritative one
-
-        @volinfo_state, @volinfo_sys are pairs of volume mark dicts
-        (foreign, native).
-
-        Note this method is marked as static, ie. the computation is
-        pure, without reliance on any excess implicit state. State
-        transitions which are deemed as ambiguous or banned will raise
-        an exception.
-
-        """
-        # store the value below "boxed" to emulate proper closures
-        # (variables of the enclosing scope are available inner functions
-        # provided they are no reassigned; mutation is OK).
-        param = FreeObject(relax_mismatch = False, state_change = None, index=-1)
-        def select_vi(vi0, vi):
-            param.index += 1
-            if vi and (not vi0 or vi0['uuid'] == vi['uuid']):
-                if not vi0 and not param.relax_mismatch:
-                    param.state_change = param.index
-                # valid new value found; for the rest, we are graceful about
-                # uuid mismatch
-                param.relax_mismatch = True
-                return vi
-            if vi0 and vi and vi0['uuid'] != vi['uuid'] and not param.relax_mismatch:
-                # uuid mismatch for master candidate, bail out
-                raise GsyncdError("aborting on uuid change from %s to %s" % \
-                                   (vi0['uuid'], vi['uuid']))
-            # fall back to old
-            return vi0
-        newstate = tuple(select_vi(*vip) for vip in zip(volinfo_state, volinfo_sys))
-        srep = lambda vi: vi and vi['uuid'][0:8]
-        logging.debug('(%s, %s) << (%s, %s) -> (%s, %s)' % \
-                      tuple(srep(vi) for vi in volinfo_state + volinfo_sys + newstate))
-        return newstate, param.state_change
-
-    def crawl(self, path='.', xtl=None):
-        """crawling...
-
-          Standing around
-          All the right people
-          Crawling
-          Tennis on Tuesday
-          The ladder is long
-          It is your nature
-          You've gotta suntan
-          Football on Sunday
-          Society boy
-
-        Recursively walk the master side tree and check if updates are
-        needed due to xtime differences. One invocation of crawl checks
-        children of @path and do a recursive enter only on
-        those directory children where there is an update needed.
-
-        Way of updates depend on file type:
-        - for symlinks, sync them directy and synchronously
-        - for regular children, register jobs for @path (cf. .add_job) to start
-          and wait on their rsync
-        - for directory children, register a job for @path which waits (.wait)
-          on jobs for the given child
-        (other kind of filesystem nodes are not considered)
-
-        Those slave side children which do not exist on master are simply
-        purged (see Server.purge).
-
-        Behavior is fault tolerant, synchronization is adaptive: if some action fails,
-        just go on relentlessly, adding a fail job (see .add_failjob) which will prevent
-        the .sendmark on @path, so when the next crawl will arrive to @path it will not
-        see it as up-to-date and  will try to sync it again. While this semantics can be
-        supported by funky design principles (http://c2.com/cgi/wiki?LazinessImpatienceHubris),
-        the ultimate reason which excludes other possibilities is simply transience: we cannot
-        assert that the file systems (master / slave) underneath do not change and actions
-        taken upon some condition will not lose their context by the time they are performed.
-        """
-        if path == '.':
-            if self.start:
-                self.crawls += 1
-                logging.debug("... crawl #%d done, took %.6f seconds" % \
-                              (self.crawls, time.time() - self.start))
-            time.sleep(1)
-            self.start = time.time()
-            should_display_info = self.start - self.lastreport['time'] >= 60
-            if should_display_info:
-                logging.info("completed %d crawls, %d turns",
-                             self.crawls - self.lastreport['crawls'],
-                             self.turns - self.lastreport['turns'])
-                self.lastreport.update(crawls = self.crawls,
-                                       turns = self.turns,
-                                       time = self.start)
-            volinfo_sys, state_change = self.volinfo_hook()
-            if self.inter_master:
-                self.volinfo = volinfo_sys[self.KFGN]
-            else:
-                self.volinfo = volinfo_sys[self.KNAT]
-            if state_change == self.KFGN or (state_change == self.KNAT and not self.inter_master):
-                logging.info('new master is %s', self.uuid)
-                if self.volinfo:
-                    logging.info("%s master with volume id %s ..." % \
-                                 (self.inter_master and "intermediate" or "primary",
-                                  self.uuid))
-            if state_change == self.KFGN:
-               gconf.configinterface.set('volume_id', self.uuid)
-            if self.volinfo:
-                if self.volinfo['retval']:
-                    raise GsyncdError ("master is corrupt")
-                self.start_checkpoint_thread()
-            else:
-                if should_display_info or self.crawls == 0:
-                    if self.inter_master:
-                        logging.info("waiting for being synced from %s ..." % \
-                                     self.volinfo_state[self.KFGN]['uuid'])
-                    else:
-                        logging.info("waiting for volume info ...")
-                return
-        logging.debug("entering " + path)
-        if not xtl:
-            xtl = self.xtime(path)
-            if isinstance(xtl, int):
-                self.add_failjob(path, 'no-local-node')
-                return
-        xtr = self.xtime(path, self.slave)
-        if isinstance(xtr, int):
-            if xtr != ENOENT:
-                self.slave.server.purge(path)
-            try:
-                self.slave.server.mkdir(path)
-            except OSError:
-                self.add_failjob(path, 'no-remote-node')
-                return
-            xtr = self.minus_infinity
-        else:
-            self.xtime_reversion_hook(path, xtl, xtr)
-            if xtl == xtr:
-                if path == '.' and self.change_seen:
-                    self.turns += 1
-                    self.change_seen = False
-                    if self.total_turns:
-                        logging.info("finished turn #%s/%s" % \
-                                     (self.turns, self.total_turns))
-                        if self.turns == self.total_turns:
-                            logging.info("reached turn limit")
-                            self.terminate = True
-                return
-        if path == '.':
-            self.change_seen = True
-        try:
-            dem = self.master.server.entries(path)
-        except OSError:
-            self.add_failjob(path, 'local-entries-fail')
-            return
-        random.shuffle(dem)
-        try:
-            des = self.slave.server.entries(path)
-        except OSError:
-            self.slave.server.purge(path)
-            try:
-                self.slave.server.mkdir(path)
-                des = self.slave.server.entries(path)
-            except OSError:
-                self.add_failjob(path, 'remote-entries-fail')
-                return
-        dd = set(des) - set(dem)
-        if dd:
-            self.purge_missing(path, dd)
-        chld = []
-        for e in dem:
-            e = os.path.join(path, e)
-            xte = self.xtime(e)
-            if isinstance(xte, int):
-                logging.warn("irregular xtime for %s: %s" % (e, errno.errorcode[xte]))
-            elif self.need_sync(e, xte, xtr):
-                chld.append((e, xte))
-        def indulgently(e, fnc, blame=None):
-            if not blame:
-                blame = path
-            try:
-                return fnc(e)
-            except (IOError, OSError):
-                ex = sys.exc_info()[1]
-                if ex.errno == ENOENT:
-                    logging.warn("salvaged ENOENT for " + e)
-                    self.add_failjob(blame, 'by-indulgently')
-                    return False
-                else:
-                    raise
-        for e, xte in chld:
-            st = indulgently(e, lambda e: os.lstat(e))
-            if st == False:
-                continue
-            mo = st.st_mode
-            adct = {'own': (st.st_uid, st.st_gid)}
-            if stat.S_ISLNK(mo):
-                if indulgently(e, lambda e: self.slave.server.symlink(os.readlink(e), e)) == False:
-                    continue
-                self.sendmark(e, xte, adct)
-            elif stat.S_ISREG(mo):
-                logging.debug("syncing %s ..." % e)
-                pb = self.syncer.add(e)
-                timeA=datetime.now()
-                def regjob(e, xte, pb):
-                    if pb.wait():
-                        logging.debug("synced " + e)
-                        self.sendmark_regular(e, xte)
-
-                        timeB=datetime.now()
-                        self.lastSyncTime=timeB-timeA
-                        self.syncTime=(self.syncTime+self.lastSyncTime.microseconds)/(10.0**6)
-                        self.filesSynced=self.filesSynced+1
-                        return True
-                    else:
-                        logging.warn("failed to sync " + e)
-                self.add_job(path, 'reg', regjob, e, xte, pb)
-            elif stat.S_ISDIR(mo):
-                adct['mode'] = mo
-                if indulgently(e, lambda e: (self.add_job(path, 'cwait', self.wait, e, xte, adct),
-                                             self.crawl(e, xte),
-                                             True)[-1], blame=e) == False:
-                    continue
-            else:
-                # ignore fifos, sockets and special files
-                pass
-        if path == '.':
-            self.wait(path, xtl)
-
-class BoxClosedErr(Exception):
-    pass
-
-class PostBox(list):
-    """synchronized collection for storing things thought of as "requests" """
-
-    def __init__(self, *a):
-        list.__init__(self, *a)
-        # too bad Python stdlib does not have read/write locks...
-        # it would suffivce to grab the lock in .append as reader, in .close as writer
-        self.lever = Condition()
-        self.open = True
-        self.done = False
-
-    def wait(self):
-        """wait on requests to be processed"""
-        self.lever.acquire()
-        if not self.done:
-            self.lever.wait()
-        self.lever.release()
-        return self.result
-
-    def wakeup(self, data):
-        """wake up requestors with the result"""
-        self.result = data
-        self.lever.acquire()
-        self.done = True
-        self.lever.notifyAll()
-        self.lever.release()
-
-    def append(self, e):
-        """post a request"""
-        self.lever.acquire()
-        if not self.open:
-            raise BoxClosedErr
-        list.append(self, e)
-        self.lever.release()
-
-    def close(self):
-        """prohibit the posting of further requests"""
-        self.lever.acquire()
-        self.open = False
-        self.lever.release()
-
-class Syncer(object):
-    """a staged queue to relay rsync requests to rsync workers
-
-    By "staged queue" its meant that when a consumer comes to the
-    queue, it takes _all_ entries, leaving the queue empty.
-    (I don't know if there is an official term for this pattern.)
-
-    The queue uses a PostBox to accumulate incoming items.
-    When a consumer (rsync worker) comes, a new PostBox is
-    set up and the old one is passed on to the consumer.
-
-    Instead of the simplistic scheme of having one big lock
-    which synchronizes both the addition of new items and
-    PostBox exchanges, use a separate lock to arbitrate consumers,
-    and rely on PostBox's synchronization mechanisms take
-    care about additions.
-
-    There is a corner case racy situation, producers vs. consumers,
-    which is not handled by this scheme: namely, when the PostBox
-    exchange occurs in between being passed to the producer for posting
-    and the post placement. But that's what Postbox.close is for:
-    such a posting will find the PostBox closed, in which case
-    the producer can re-try posting against the actual PostBox of
-    the queue.
-
-    To aid accumlation of items in the PostBoxen before grabbed
-    by an rsync worker, the worker goes to sleep a bit after
-    each completed syncjob.
-    """
-
-    def __init__(self, slave):
-        """spawn worker threads"""
-        self.slave = slave
-        self.lock = Lock()
-        self.pb = PostBox()
-        self.bytesSynced=0
-        for i in range(int(gconf.sync_jobs)):
-            t = Thread(target=self.syncjob)
-            t.start()
-
-    def syncjob(self):
-        """the life of a worker"""
-        while True:
-            pb = None
-            while True:
-                self.lock.acquire()
-                if self.pb:
-                    pb, self.pb = self.pb, PostBox()
-                self.lock.release()
-                if pb:
-                    break
-                time.sleep(0.5)
-            pb.close()
-            po = self.slave.rsync(pb)
-            if po.returncode == 0:
-                regEx=re.search('\ *total\ *transferred\ *file\ *size:\ *(\d+)\ *bytes\ *',po.stdout.read(),re.IGNORECASE)
-                if regEx:
-                        self.bytesSynced+=(int(regEx.group(1)))/1024
-                ret = True
-            elif po.returncode in (23, 24):
-                # partial transfer (cf. rsync(1)), that's normal
-                ret = False
-            else:
-                po.errfail()
-            pb.wakeup(ret)
-
-    def add(self, e):
-        while True:
-            pb = self.pb
-            try:
-                pb.append(e)
-                return pb
-            except BoxClosedErr:
-                pass
diff --git a/xlators/features/marker/utils/syncdaemon/monitor.py b/xlators/features/marker/utils/syncdaemon/monitor.py
deleted file mode 100644
index b8956dcc2..000000000
--- a/xlators/features/marker/utils/syncdaemon/monitor.py
+++ /dev/null
@@ -1,129 +0,0 @@
-import os
-import sys
-import time
-import signal
-import logging
-from gconf import gconf
-from syncdutils import update_file, select, waitpid, set_term_handler
-
-class Monitor(object):
-    """class which spawns and manages gsyncd workers"""
-
-    def __init__(self):
-        self.state = None
-
-    def set_state(self, state):
-        """set the state that can be used by external agents
-           like glusterd for status reporting"""
-        if state == self.state:
-            return
-        self.state = state
-        logging.info('new state: %s' % state)
-        if getattr(gconf, 'state_file', None):
-            update_file(gconf.state_file, lambda f: f.write(state + '\n'))
-
-    def monitor(self):
-        """the monitor loop
-
-        Basic logic is a blantantly simple blunt heuristics:
-        if spawned client survives 60 secs, it's considered OK.
-        This servers us pretty well as it's not vulneralbe to
-        any kind of irregular behavior of the child...
-
-        ... well, except for one: if children is hung up on
-        waiting for some event, it can survive aeons, still
-        will be defunct. So we tweak the above logic to
-        expect the worker to send us a signal within 60 secs
-        (in the form of closing its end of a pipe). The worker
-        does this when it's done with the setup stage
-        ready to enter the service loop (note it's the setup
-        stage which is vulnerable to hangs -- the full
-        blown worker blows up on EPIPE if the net goes down,
-        due to the keep-alive thread)
-        """
-        def sigcont_handler(*a):
-            """
-            Re-init logging and send group kill signal
-            """
-            md = gconf.log_metadata
-            logging.shutdown()
-            lcls = logging.getLoggerClass()
-            lcls.setup(label=md.get('saved_label'), **md)
-            pid = os.getpid()
-            os.kill(-pid, signal.SIGUSR1)
-        signal.signal(signal.SIGUSR1, lambda *a: ())
-        signal.signal(signal.SIGCONT, sigcont_handler)
-
-        argv = sys.argv[:]
-        for o in ('-N', '--no-daemon', '--monitor'):
-            while o in argv:
-                argv.remove(o)
-        argv.extend(('-N', '-p', ''))
-        argv.insert(0, os.path.basename(sys.executable))
-
-        self.set_state('starting...')
-        ret = 0
-        def nwait(p, o=0):
-            p2, r = waitpid(p, o)
-            if not p2:
-                return
-            return r
-        def exit_signalled(s):
-            """ child teminated due to receipt of SIGUSR1 """
-            return (os.WIFSIGNALED(s) and (os.WTERMSIG(s) == signal.SIGUSR1))
-        def exit_status(s):
-            if os.WIFEXITED(s):
-                return os.WEXITSTATUS(s)
-            return 1
-        conn_timeout = int(gconf.connection_timeout)
-        while ret in (0, 1):
-            logging.info('-' * conn_timeout)
-            logging.info('starting gsyncd worker')
-            pr, pw = os.pipe()
-            cpid = os.fork()
-            if cpid == 0:
-                os.close(pr)
-                os.execv(sys.executable, argv + ['--feedback-fd', str(pw)])
-            os.close(pw)
-            t0 = time.time()
-            so = select((pr,), (), (), conn_timeout)[0]
-            os.close(pr)
-            if so:
-                ret = nwait(cpid, os.WNOHANG)
-                if ret != None:
-                    logging.debug("worker died before establishing connection")
-                else:
-                    logging.debug("worker seems to be connected (?? racy check)")
-                    while time.time() < t0 + conn_timeout:
-                        ret = nwait(cpid, os.WNOHANG)
-                        if ret != None:
-                            logging.debug("worker died in startup phase")
-                            break
-                        time.sleep(1)
-            else:
-                logging.debug("worker not confirmed in %d sec, aborting it" % \
-                              conn_timeout)
-                # relax one SIGTERM by setting a handler that sets back
-                # standard handler
-                set_term_handler(lambda *a: set_term_handler())
-                # give a chance to graceful exit
-                os.kill(-os.getpid(), signal.SIGTERM)
-                time.sleep(1)
-                os.kill(cpid, signal.SIGKILL)
-                ret = nwait(cpid)
-            if ret == None:
-                self.set_state('OK')
-                ret = nwait(cpid)
-            if exit_signalled(ret):
-                ret = 0
-            else:
-                ret = exit_status(ret)
-                if ret in (0,1):
-                    self.set_state('faulty')
-            time.sleep(10)
-        self.set_state('inconsistent')
-        return ret
-
-def monitor():
-    """oh yeah, actually Monitor is used as singleton, too"""
-    return Monitor().monitor()
diff --git a/xlators/features/marker/utils/syncdaemon/repce.py b/xlators/features/marker/utils/syncdaemon/repce.py
deleted file mode 100644
index 755fb61df..000000000
--- a/xlators/features/marker/utils/syncdaemon/repce.py
+++ /dev/null
@@ -1,225 +0,0 @@
-import os
-import sys
-import time
-import logging
-from threading import Condition
-try:
-    import thread
-except ImportError:
-    # py 3
-    import _thread as thread
-try:
-    from Queue import Queue
-except ImportError:
-    # py 3
-    from queue import Queue
-try:
-    import cPickle as pickle
-except ImportError:
-    # py 3
-    import pickle
-
-from syncdutils import Thread, select
-
-pickle_proto = -1
-repce_version = 1.0
-
-def ioparse(i, o):
-    if isinstance(i, int):
-        i = os.fdopen(i)
-    # rely on duck typing for recognizing
-    # streams as that works uniformly
-    # in py2 and py3
-    if hasattr(o, 'fileno'):
-        o = o.fileno()
-    return (i, o)
-
-def send(out, *args):
-    """pickle args and write out wholly in one syscall
-
-    ie. not use the ability of pickle to dump directly to
-    a stream, as that would potentially mess up messages
-    by interleaving them
-    """
-    os.write(out, pickle.dumps(args, pickle_proto))
-
-def recv(inf):
-    """load an object from input stream"""
-    return pickle.load(inf)
-
-
-class RepceServer(object):
-    """RePCe is Hungarian for canola, http://hu.wikipedia.org/wiki/Repce
-
-    ... also our homebrewed RPC backend where the transport layer is
-    reduced to a pair of filehandles.
-
-    This is the server component.
-    """
-
-    def __init__(self, obj, i, o, wnum=6):
-        """register a backend object .obj to which incoming messages
-           are dispatched, also incoming/outcoming streams
-        """
-        self.obj = obj
-        self.inf, self.out = ioparse(i, o)
-        self.wnum = wnum
-        self.q = Queue()
-
-    def service_loop(self):
-        """fire up worker threads, get messages and dispatch among them"""
-        for i in range(self.wnum):
-            t = Thread(target=self.worker)
-            t.start()
-        try:
-            while True:
-                self.q.put(recv(self.inf))
-        except EOFError:
-            logging.info("terminating on reaching EOF.")
-
-    def worker(self):
-        """life of a worker
-
-        Get message, extract its id, method name and arguments
-        (kwargs not supported), call method on .obj.
-        Send back message id + return value.
-        If method call throws an exception, rescue it, and send
-        back the exception as result (with flag marking it as
-        exception).
-        """
-        while True:
-            in_data = self.q.get(True)
-            rid = in_data[0]
-            rmeth = in_data[1]
-            exc = False
-            if rmeth == '__repce_version__':
-                res = repce_version
-            else:
-              try:
-                  res = getattr(self.obj, rmeth)(*in_data[2:])
-              except:
-                  res = sys.exc_info()[1]
-                  exc = True
-                  logging.exception("call failed: ")
-            send(self.out, rid, exc, res)
-
-
-class RepceJob(object):
-    """class representing message status we can use
-    for waiting on reply"""
-
-    def __init__(self, cbk):
-        """
-        - .rid: (process-wise) unique id
-        - .cbk: what we do upon receiving reply
-        """
-        self.rid = (os.getpid(), thread.get_ident(), time.time())
-        self.cbk = cbk
-        self.lever = Condition()
-        self.done = False
-
-    def __repr__(self):
-        return ':'.join([str(x) for x in self.rid])
-
-    def wait(self):
-        self.lever.acquire()
-        if not self.done:
-            self.lever.wait()
-        self.lever.release()
-        return self.result
-
-    def wakeup(self, data):
-        self.result = data
-        self.lever.acquire()
-        self.done = True
-        self.lever.notify()
-        self.lever.release()
-
-
-class RepceClient(object):
-    """RePCe is Hungarian for canola, http://hu.wikipedia.org/wiki/Repce
-
-    ... also our homebrewed RPC backend where the transport layer is
-    reduced to a pair of filehandles.
-
-    This is the client component.
-    """
-
-    def __init__(self, i, o):
-        self.inf, self.out = ioparse(i, o)
-        self.jtab = {}
-        t = Thread(target = self.listen)
-        t.start()
-
-    def listen(self):
-        while True:
-            select((self.inf,), (), ())
-            rid, exc, res = recv(self.inf)
-            rjob = self.jtab.pop(rid)
-            if rjob.cbk:
-                rjob.cbk(rjob, [exc, res])
-
-    def push(self, meth, *args, **kw):
-        """wrap arguments in a RepceJob, send them to server
-           and return the RepceJob
-
-           @cbk to pass on RepceJob can be given as kwarg.
-        """
-        cbk = kw.get('cbk')
-        if not cbk:
-            def cbk(rj, res):
-                if res[0]:
-                    raise res[1]
-        rjob = RepceJob(cbk)
-        self.jtab[rjob.rid] = rjob
-        logging.debug("call %s %s%s ..." % (repr(rjob), meth, repr(args)))
-        send(self.out, rjob.rid, meth, *args)
-        return rjob
-
-    def __call__(self, meth, *args):
-        """RePCe client is callabe, calling it implements a synchronous remote call
-
-        We do a .push with a cbk which does a wakeup upon receiving anwser, then wait
-        on the RepceJob.
-        """
-        rjob = self.push(meth, *args, **{'cbk': lambda rj, res: rj.wakeup(res)})
-        exc, res = rjob.wait()
-        if exc:
-            logging.error('call %s (%s) failed on peer with %s' % (repr(rjob), meth, str(type(res).__name__)))
-            raise res
-        logging.debug("call %s %s -> %s" % (repr(rjob), meth, repr(res)))
-        return res
-
-    class mprx(object):
-        """method proxy, standard trick to implement rubyesque method_missing
-           in Python
-
-        A class is a closure factory, you know what I mean, or go read some SICP.
-        """
-        def __init__(self, ins, meth):
-            self.ins = ins
-            self.meth = meth
-
-        def __call__(self, *a):
-            return self.ins(self.meth, *a)
-
-    def __getattr__(self, meth):
-        """this implements transparent method dispatch to remote object,
-           so that you don't need to call the RepceClient instance like
-
-             rclient('how_old_are_you_if_born_in', 1979)
-
-           but you can make it into an ordinary method call like
-
-             rclient.how_old_are_you_if_born_in(1979)
-        """
-        return self.mprx(self, meth)
-
-    def __version__(self):
-        """used in handshake to verify compatibility"""
-        d = {'proto': self('__repce_version__')}
-        try:
-            d['object'] = self('version')
-        except AttributeError:
-            pass
-        return d
diff --git a/xlators/features/marker/utils/syncdaemon/resource.py b/xlators/features/marker/utils/syncdaemon/resource.py
deleted file mode 100644
index 73102fbcb..000000000
--- a/xlators/features/marker/utils/syncdaemon/resource.py
+++ /dev/null
@@ -1,972 +0,0 @@
-import re
-import os
-import sys
-import stat
-import time
-import fcntl
-import errno
-import struct
-import socket
-import logging
-import tempfile
-import threading
-import subprocess
-from errno import EEXIST, ENOENT, ENODATA, ENOTDIR, ELOOP, EISDIR
-from select import error as SelectError
-
-from gconf import gconf
-import repce
-from repce import RepceServer, RepceClient
-from  master import gmaster_builder
-import syncdutils
-from syncdutils import GsyncdError, select, privileged, boolify
-
-UrlRX  = re.compile('\A(\w+)://([^ *?[]*)\Z')
-HostRX = re.compile('[a-z\d](?:[a-z\d.-]*[a-z\d])?', re.I)
-UserRX = re.compile("[\w!\#$%&'*+-\/=?^_`{|}~]+")
-
-def sup(x, *a, **kw):
-    """a rubyesque "super" for python ;)
-
-    invoke caller method in parent class with given args.
-    """
-    return getattr(super(type(x), x), sys._getframe(1).f_code.co_name)(*a, **kw)
-
-def desugar(ustr):
-    """transform sugared url strings to standard <scheme>://<urlbody> form
-
-    parsing logic enforces the constraint that sugared forms should contatin
-    a ':' or a '/', which ensures that sugared urls do not conflict with
-    gluster volume names.
-    """
-    m = re.match('([^:]*):(.*)', ustr)
-    if m:
-        if not m.groups()[0]:
-            return "gluster://localhost" + ustr
-        elif '@' in m.groups()[0] or re.search('[:/]', m.groups()[1]):
-            return "ssh://" + ustr
-        else:
-            return "gluster://" + ustr
-    else:
-        if ustr[0] != '/':
-            raise GsyncdError("cannot resolve sugared url '%s'" % ustr)
-        ap = os.path.normpath(ustr)
-        if ap.startswith('//'):
-            ap = ap[1:]
-        return "file://" + ap
-
-def gethostbyname(hnam):
-    """gethostbyname wrapper"""
-    try:
-        return socket.gethostbyname(hnam)
-    except socket.gaierror:
-        ex = sys.exc_info()[1]
-        raise GsyncdError("failed to resolve %s: %s" % \
-                          (hnam, ex.strerror))
-
-def parse_url(ustr):
-    """instantiate an url object by scheme-to-class dispatch
-
-    The url classes taken into consideration are the ones in
-    this module whose names are full-caps.
-    """
-    m = UrlRX.match(ustr)
-    if not m:
-        ustr = desugar(ustr)
-    m = UrlRX.match(ustr)
-    if not m:
-        raise GsyncdError("malformed url")
-    sch, path = m.groups()
-    this = sys.modules[__name__]
-    if not hasattr(this, sch.upper()):
-        raise GsyncdError("unknown url scheme " + sch)
-    return getattr(this, sch.upper())(path)
-
-
-class _MetaXattr(object):
-    """singleton class, a lazy wrapper around the
-    libcxattr module
-
-    libcxattr (a heavy import due to ctypes) is
-    loaded only when when the single
-    instance is tried to be used.
-
-    This reduces runtime for those invocations
-    which do not need filesystem manipulation
-    (eg. for config, url parsing)
-    """
-
-    def __getattr__(self, meth):
-        from libcxattr import Xattr as LXattr
-        xmeth = [ m for m in dir(LXattr) if m[0] != '_' ]
-        if not meth in xmeth:
-            return
-        for m in xmeth:
-            setattr(self, m, getattr(LXattr, m))
-        return getattr(self, meth)
-
-Xattr = _MetaXattr()
-
-
-class Popen(subprocess.Popen):
-    """customized subclass of subprocess.Popen with a ring
-    buffer for children error output"""
-
-    @classmethod
-    def init_errhandler(cls):
-        """start the thread which handles children's error output"""
-        cls.errstore = {}
-        def tailer():
-            while True:
-                errstore = cls.errstore.copy()
-                try:
-                    poe, _ ,_ = select([po.stderr for po in errstore], [], [], 1)
-                except (ValueError, SelectError):
-                    continue
-                for po in errstore:
-                    if po.stderr not in poe:
-                        continue
-                    po.lock.acquire()
-                    try:
-                        if po.on_death_row:
-                            continue
-                        la = errstore[po]
-                        try:
-                            fd = po.stderr.fileno()
-                        except ValueError:  # file is already closed
-                            continue
-                        l = os.read(fd, 1024)
-                        if not l:
-                            continue
-                        tots = len(l)
-                        for lx in la:
-                            tots += len(lx)
-                        while tots > 1<<20 and la:
-                            tots -= len(la.pop(0))
-                        la.append(l)
-                    finally:
-                        po.lock.release()
-        t = syncdutils.Thread(target = tailer)
-        t.start()
-        cls.errhandler = t
-
-    @classmethod
-    def fork(cls):
-        """fork wrapper that restarts errhandler thread in child"""
-        pid = os.fork()
-        if not pid:
-            cls.init_errhandler()
-        return pid
-
-    def __init__(self, args, *a, **kw):
-        """customizations for subprocess.Popen instantiation
-
-        - 'close_fds' is taken to be the default
-        - if child's stderr is chosen to be managed,
-          register it with the error handler thread
-        """
-        self.args = args
-        if 'close_fds' not in kw:
-            kw['close_fds'] = True
-        self.lock = threading.Lock()
-        self.on_death_row = False
-        try:
-            sup(self, args, *a, **kw)
-        except:
-            ex = sys.exc_info()[1]
-            if not isinstance(ex, OSError):
-                raise
-            raise GsyncdError("""execution of "%s" failed with %s (%s)""" % \
-                              (args[0], errno.errorcode[ex.errno], os.strerror(ex.errno)))
-        if kw.get('stderr') == subprocess.PIPE:
-            assert(getattr(self, 'errhandler', None))
-            self.errstore[self] = []
-
-    def errlog(self):
-        """make a log about child's failure event"""
-        filling = ""
-        if self.elines:
-            filling = ", saying:"
-        logging.error("""command "%s" returned with %s%s""" % \
-                      (" ".join(self.args), repr(self.returncode), filling))
-        lp = ''
-        def logerr(l):
-            logging.error(self.args[0] + "> " + l)
-        for l in self.elines:
-            ls = l.split('\n')
-            ls[0] = lp + ls[0]
-            lp = ls.pop()
-            for ll in ls:
-                logerr(ll)
-        if lp:
-            logerr(lp)
-
-    def errfail(self):
-        """fail nicely if child did not terminate with success"""
-        self.errlog()
-        syncdutils.finalize(exval = 1)
-
-    def terminate_geterr(self, fail_on_err = True):
-        """kill child, finalize stderr harvesting (unregister
-        from errhandler, set up .elines), fail on error if
-        asked for
-        """
-        self.lock.acquire()
-        try:
-            self.on_death_row = True
-        finally:
-            self.lock.release()
-        elines = self.errstore.pop(self)
-        if self.poll() == None:
-            self.terminate()
-            if self.poll() == None:
-                time.sleep(0.1)
-                self.kill()
-                self.wait()
-        while True:
-            if not select([self.stderr],[],[],0.1)[0]:
-                break
-            b = os.read(self.stderr.fileno(), 1024)
-            if b:
-                elines.append(b)
-            else:
-                break
-        self.stderr.close()
-        self.elines = elines
-        if fail_on_err and self.returncode != 0:
-            self.errfail()
-
-
-class Server(object):
-    """singleton implemening those filesystem access primitives
-       which are needed for geo-replication functionality
-
-    (Singleton in the sense it's a class which has only static
-    and classmethods and is used directly, without instantiation.)
-    """
-
-    GX_NSPACE = (privileged() and "trusted" or "system") + ".glusterfs"
-    NTV_FMTSTR = "!" + "B"*19 + "II"
-    FRGN_XTRA_FMT = "I"
-    FRGN_FMTSTR = NTV_FMTSTR + FRGN_XTRA_FMT
-
-    def _pathguard(f):
-        """decorator method that checks
-        the path argument of the decorated
-        functions to make sure it does not
-        point out of the managed tree
-        """
-
-        fc = getattr(f, 'func_code', None)
-        if not fc:
-            # python 3
-            fc = f.__code__
-        pi = list(fc.co_varnames).index('path')
-        def ff(*a):
-            path = a[pi]
-            ps = path.split('/')
-            if path[0] == '/' or '..' in ps:
-                raise ValueError('unsafe path')
-            return f(*a)
-        return ff
-
-    @staticmethod
-    @_pathguard
-    def entries(path):
-        """directory entries in an array"""
-        # prevent symlinks being followed
-        if not stat.S_ISDIR(os.lstat(path).st_mode):
-            raise OSError(ENOTDIR, os.strerror(ENOTDIR))
-        return os.listdir(path)
-
-    @classmethod
-    @_pathguard
-    def purge(cls, path, entries=None):
-        """force-delete subtrees
-
-        If @entries is not specified, delete
-        the whole subtree under @path (including
-        @path).
-
-        Otherwise, @entries should be a
-        a sequence of children of @path, and
-        the effect is identical with a joint
-        @entries-less purge on them, ie.
-
-        for e in entries:
-            cls.purge(os.path.join(path, e))
-        """
-        me_also = entries == None
-        if not entries:
-            try:
-                # if it's a symlink, prevent
-                # following it
-                try:
-                    os.unlink(path)
-                    return
-                except OSError:
-                    ex = sys.exc_info()[1]
-                    if ex.errno == EISDIR:
-                        entries = os.listdir(path)
-                    else:
-                        raise
-            except OSError:
-                ex = sys.exc_info()[1]
-                if ex.errno in (ENOTDIR, ENOENT, ELOOP):
-                    try:
-                        os.unlink(path)
-                        return
-                    except OSError:
-                        ex = sys.exc_info()[1]
-                        if ex.errno == ENOENT:
-                            return
-                        raise
-                else:
-                    raise
-        for e in entries:
-            cls.purge(os.path.join(path, e))
-        if me_also:
-            os.rmdir(path)
-
-    @classmethod
-    @_pathguard
-    def _create(cls, path, ctor):
-        """path creation backend routine"""
-        try:
-            ctor(path)
-        except OSError:
-            ex = sys.exc_info()[1]
-            if ex.errno == EEXIST:
-                cls.purge(path)
-                return ctor(path)
-            raise
-
-    @classmethod
-    @_pathguard
-    def mkdir(cls, path):
-        cls._create(path, os.mkdir)
-
-    @classmethod
-    @_pathguard
-    def symlink(cls, lnk, path):
-        cls._create(path, lambda p: os.symlink(lnk, p))
-
-    @classmethod
-    @_pathguard
-    def xtime(cls, path, uuid):
-        """query xtime extended attribute
-
-        Return xtime of @path for @uuid as a pair of integers.
-        "Normal" errors due to non-existent @path or extended attribute
-        are tolerated and errno is returned in such a case.
-        """
-
-        try:
-            return struct.unpack('!II', Xattr.lgetxattr(path, '.'.join([cls.GX_NSPACE, uuid, 'xtime']), 8))
-        except OSError:
-            ex = sys.exc_info()[1]
-            if ex.errno in (ENOENT, ENODATA, ENOTDIR):
-                return ex.errno
-            else:
-                raise
-
-    @classmethod
-    def xtime_vec(cls, path, *uuids):
-        """vectored version of @xtime
-
-        accepts a list of uuids and returns a dictionary
-        with uuid as key(s) and xtime as value(s)
-        """
-        xt = {}
-        for uuid in uuids:
-            xtu = cls.xtime(path, uuid)
-            if xtu == ENODATA:
-                xtu = None
-            if isinstance(xtu, int):
-                return xtu
-            xt[uuid] = xtu
-        return xt
-
-    @classmethod
-    @_pathguard
-    def set_xtime(cls, path, uuid, mark):
-        """set @mark as xtime for @uuid on @path"""
-        Xattr.lsetxattr(path, '.'.join([cls.GX_NSPACE, uuid, 'xtime']), struct.pack('!II', *mark))
-
-    @classmethod
-    def set_xtime_vec(cls, path, mark_dct):
-        """vectored (or dictered) version of set_xtime
-
-        ignore values that match @ignore
-        """
-        for u,t in mark_dct.items():
-            cls.set_xtime(path, u, t)
-
-    @staticmethod
-    @_pathguard
-    def setattr(path, adct):
-        """set file attributes
-
-        @adct is a dict, where 'own', 'mode' and 'times'
-        keys are looked for and values used to perform
-        chown, chmod or utimes on @path.
-        """
-        own = adct.get('own')
-        if own:
-            os.lchown(path, *own)
-        mode = adct.get('mode')
-        if mode:
-            os.chmod(path, stat.S_IMODE(mode))
-        times = adct.get('times')
-        if times:
-            os.utime(path, times)
-
-    @staticmethod
-    def pid():
-        return os.getpid()
-
-    last_keep_alive = 0
-    @classmethod
-    def keep_alive(cls, dct):
-        """process keepalive messages.
-
-        Return keep-alive counter (number of received keep-alive
-        messages).
-
-        Now the "keep-alive" message can also have a payload which is
-        used to set a foreign volume-mark on the underlying file system.
-        """
-        if dct:
-            key = '.'.join([cls.GX_NSPACE, 'volume-mark', dct['uuid']])
-            val = struct.pack(cls.FRGN_FMTSTR,
-                              *(dct['version']  +
-                                tuple(int(x,16) for x in re.findall('(?:[\da-f]){2}', dct['uuid'])) +
-                                (dct['retval'],) + dct['volume_mark'][0:2] + (dct['timeout'],)))
-            Xattr.lsetxattr('.', key, val)
-        cls.last_keep_alive += 1
-        return cls.last_keep_alive
-
-    @staticmethod
-    def version():
-        """version used in handshake"""
-        return 1.0
-
-
-class SlaveLocal(object):
-    """mix-in class to implement some factes of a slave server
-
-    ("mix-in" is sort of like "abstract class", ie. it's not
-    instantiated just included in the ancesty DAG. I use "mix-in"
-    to indicate that it's not used as an abstract base class,
-    rather just taken in to implement additional functionality
-    on the basis of the assumed availability of certain interfaces.)
-    """
-
-    def can_connect_to(self, remote):
-        """determine our position in the connectibility matrix"""
-        return not remote
-
-    def service_loop(self):
-        """start a RePCe server serving self's server
-
-        stop servicing if a timeout is configured and got no
-        keep-alime in that inteval
-        """
-
-        if boolify(gconf.use_rsync_xattrs) and not privileged():
-            raise GsyncdError("using rsync for extended attributes is not supported")
-
-        repce = RepceServer(self.server, sys.stdin, sys.stdout, int(gconf.sync_jobs))
-        t = syncdutils.Thread(target=lambda: (repce.service_loop(),
-                                              syncdutils.finalize()))
-        t.start()
-        logging.info("slave listening")
-        if gconf.timeout and int(gconf.timeout) > 0:
-            while True:
-                lp = self.server.last_keep_alive
-                time.sleep(int(gconf.timeout))
-                if lp == self.server.last_keep_alive:
-                    logging.info("connection inactive for %d seconds, stopping" % int(gconf.timeout))
-                    break
-        else:
-            select((), (), ())
-
-class SlaveRemote(object):
-    """mix-in class to implement an interface to a remote slave"""
-
-    def connect_remote(self, rargs=[], **opts):
-        """connects to a remote slave
-
-        Invoke an auxiliary utility (slave gsyncd, possibly wrapped)
-        which sets up the connection and set up a RePCe client to
-        communicate throuh its stdio.
-        """
-        slave = opts.get('slave', self.url)
-        extra_opts = []
-        so = getattr(gconf, 'session_owner', None)
-        if so:
-            extra_opts += ['--session-owner', so]
-        if boolify(gconf.use_rsync_xattrs):
-            extra_opts.append('--use-rsync-xattrs')
-        po = Popen(rargs + gconf.remote_gsyncd.split() + extra_opts + \
-                   ['-N', '--listen', '--timeout', str(gconf.timeout), slave],
-                   stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
-        gconf.transport = po
-        return self.start_fd_client(po.stdout, po.stdin, **opts)
-
-    def start_fd_client(self, i, o, **opts):
-        """set up RePCe client, handshake with server
-
-        It's cut out as a separate method to let
-        subclasses hook into client startup
-        """
-        self.server = RepceClient(i, o)
-        rv = self.server.__version__()
-        exrv = {'proto': repce.repce_version, 'object': Server.version()}
-        da0 = (rv, exrv)
-        da1 = ({}, {})
-        for i in range(2):
-            for k, v in da0[i].iteritems():
-                da1[i][k] = int(v)
-        if da1[0] != da1[1]:
-            raise GsyncdError("RePCe major version mismatch: local %s, remote %s" % (exrv, rv))
-
-    def rsync(self, files, *args):
-        """invoke rsync"""
-        if not files:
-            raise GsyncdError("no files to sync")
-        logging.debug("files: " + ", ".join(files))
-        argv = gconf.rsync_command.split() + \
-               ['-aR0', '--files-from=-', '--super','--stats', '--numeric-ids', '--no-implied-dirs'] + \
-               gconf.rsync_options.split() + (boolify(gconf.use_rsync_xattrs) and ['--xattrs'] or []) + \
-               ['.'] + list(args)
-        po = Popen(argv, stdin=subprocess.PIPE,stdout=subprocess.PIPE,stderr=subprocess.PIPE)
-        for f in files:
-            po.stdin.write(f)
-            po.stdin.write('\0')
-
-        po.stdin.close()
-        po.wait()
-        po.terminate_geterr(fail_on_err = False)
-
-        return po
-
-
-class AbstractUrl(object):
-    """abstract base class for url scheme classes"""
-
-    def __init__(self, path, pattern):
-        m = re.search(pattern, path)
-        if not m:
-            raise GsyncdError("malformed path")
-        self.path = path
-        return m.groups()
-
-    @property
-    def scheme(self):
-        return type(self).__name__.lower()
-
-    def canonical_path(self):
-        return self.path
-
-    def get_url(self, canonical=False, escaped=False):
-        """format self's url in various styles"""
-        if canonical:
-            pa = self.canonical_path()
-        else:
-            pa = self.path
-        u = "://".join((self.scheme, pa))
-        if escaped:
-            u = syncdutils.escape(u)
-        return u
-
-    @property
-    def url(self):
-        return self.get_url()
-
-
-  ### Concrete resource classes ###
-
-
-class FILE(AbstractUrl, SlaveLocal, SlaveRemote):
-    """scheme class for file:// urls
-
-    can be used to represent a file slave server
-    on slave side, or interface to a remote file
-    file server on master side
-    """
-
-    class FILEServer(Server):
-        """included server flavor"""
-        pass
-
-    server = FILEServer
-
-    def __init__(self, path):
-        sup(self, path, '^/')
-
-    def connect(self):
-        """inhibit the resource beyond"""
-        os.chdir(self.path)
-
-    def rsync(self, files):
-        return sup(self, files, self.path)
-
-
-class GLUSTER(AbstractUrl, SlaveLocal, SlaveRemote):
-    """scheme class for gluster:// urls
-
-    can be used to represent a gluster slave server
-    on slave side, or interface to a remote gluster
-    slave on master side, or to represent master
-    (slave-ish features come from the mixins, master
-    functionality is outsourced to GMaster from master)
-    """
-
-    class GLUSTERServer(Server):
-        "server enhancements for a glusterfs backend"""
-
-        @classmethod
-        def _attr_unpack_dict(cls, xattr, extra_fields = ''):
-            """generic volume mark fetching/parsing backed"""
-            fmt_string = cls.NTV_FMTSTR + extra_fields
-            buf = Xattr.lgetxattr('.', xattr, struct.calcsize(fmt_string))
-            vm = struct.unpack(fmt_string, buf)
-            m = re.match('(.{8})(.{4})(.{4})(.{4})(.{12})', "".join(['%02x' % x for x in vm[2:18]]))
-            uuid = '-'.join(m.groups())
-            volinfo = {  'version': vm[0:2],
-                         'uuid'   : uuid,
-                         'retval' : vm[18],
-                         'volume_mark': vm[19:21],
-                      }
-            if extra_fields:
-                return volinfo, vm[-len(extra_fields):]
-            else:
-                return volinfo
-
-        @classmethod
-        def foreign_volume_infos(cls):
-            """return list of valid (not expired) foreign volume marks"""
-            dict_list = []
-            xattr_list = Xattr.llistxattr_buf('.')
-            for ele in xattr_list:
-                if ele.find('.'.join([cls.GX_NSPACE, 'volume-mark', ''])) == 0:
-                    d, x = cls._attr_unpack_dict(ele, cls.FRGN_XTRA_FMT)
-                    now = int(time.time())
-                    if x[0] > now:
-                        logging.debug("volinfo[%s] expires: %d (%d sec later)" % \
-                                      (d['uuid'], x[0], x[0] - now))
-                        d['timeout'] = x[0]
-                        dict_list.append(d)
-                    else:
-                        try:
-                            Xattr.lremovexattr('.', ele)
-                        except OSError:
-                            pass
-            return dict_list
-
-        @classmethod
-        def native_volume_info(cls):
-            """get the native volume mark of the underlying gluster volume"""
-            try:
-                return cls._attr_unpack_dict('.'.join([cls.GX_NSPACE, 'volume-mark']))
-            except OSError:
-                ex = sys.exc_info()[1]
-                if ex.errno != ENODATA:
-                    raise
-
-    server = GLUSTERServer
-
-    def __init__(self, path):
-        self.host, self.volume = sup(self, path, '^(%s):(.+)' % HostRX.pattern)
-
-    def canonical_path(self):
-        return ':'.join([gethostbyname(self.host), self.volume])
-
-    def can_connect_to(self, remote):
-        """determine our position in the connectibility matrix"""
-        return True
-
-    class Mounter(object):
-        """Abstract base class for mounter backends"""
-
-        def __init__(self, params):
-            self.params = params
-            self.mntpt = None
-
-        @classmethod
-        def get_glusterprog(cls):
-            return os.path.join(gconf.gluster_command_dir, cls.glusterprog)
-
-        def umount_l(self, d):
-            """perform lazy umount"""
-            po = Popen(self.make_umount_argv(d), stderr=subprocess.PIPE)
-            po.wait()
-            return po
-
-        @classmethod
-        def make_umount_argv(cls, d):
-            raise NotImplementedError
-
-        def make_mount_argv(self, *a):
-            raise NotImplementedError
-
-        def cleanup_mntpt(self, *a):
-            pass
-
-        def handle_mounter(self, po):
-            po.wait()
-
-        def inhibit(self, *a):
-            """inhibit a gluster filesystem
-
-            Mount glusterfs over a temporary mountpoint,
-            change into the mount, and lazy unmount the
-            filesystem.
-            """
-
-            mpi, mpo = os.pipe()
-            mh = Popen.fork()
-            if mh:
-                os.close(mpi)
-                fcntl.fcntl(mpo, fcntl.F_SETFD, fcntl.FD_CLOEXEC)
-                d = None
-                margv = self.make_mount_argv(*a)
-                if self.mntpt:
-                    # mntpt is determined pre-mount
-                    d = self.mntpt
-                    os.write(mpo, d + '\0')
-                po = Popen(margv, **self.mountkw)
-                self.handle_mounter(po)
-                po.terminate_geterr()
-                logging.debug('auxiliary glusterfs mount in place')
-                if not d:
-                    # mntpt is determined during mount
-                    d = self.mntpt
-                    os.write(mpo, d + '\0')
-                os.write(mpo, 'M')
-                t = syncdutils.Thread(target=lambda: os.chdir(d))
-                t.start()
-                tlim = gconf.starttime + int(gconf.connection_timeout)
-                while True:
-                    if not t.isAlive():
-                        break
-                    if time.time() >= tlim:
-                        syncdutils.finalize(exval = 1)
-                    time.sleep(1)
-                os.close(mpo)
-                _, rv = syncdutils.waitpid(mh, 0)
-                if rv:
-                    rv = (os.WIFEXITED(rv) and os.WEXITSTATUS(rv) or 0) - \
-                         (os.WIFSIGNALED(rv) and os.WTERMSIG(rv) or 0)
-                    logging.warn('stale mount possibly left behind on ' + d)
-                    raise GsyncdError("cleaning up temp mountpoint %s failed with status %d" % \
-                                      (d, rv))
-            else:
-                rv = 0
-                try:
-                    os.setsid()
-                    os.close(mpo)
-                    mntdata = ''
-                    while True:
-                        c = os.read(mpi, 1)
-                        if not c:
-                            break
-                        mntdata += c
-                    if mntdata:
-                        mounted = False
-                        if mntdata[-1] == 'M':
-                            mntdata = mntdata[:-1]
-                            assert(mntdata)
-                            mounted = True
-                        assert(mntdata[-1] == '\0')
-                        mntpt = mntdata[:-1]
-                        assert(mntpt)
-                        if mounted:
-                            po = self.umount_l(mntpt)
-                            po.terminate_geterr(fail_on_err = False)
-                            if po.returncode != 0:
-                                po.errlog()
-                                rv = po.returncode
-                        self.cleanup_mntpt(mntpt)
-                except:
-                    logging.exception('mount cleanup failure:')
-                    rv = 200
-                os._exit(rv)
-            logging.debug('auxiliary glusterfs mount prepared')
-
-    class DirectMounter(Mounter):
-        """mounter backend which calls mount(8), umount(8) directly"""
-
-        mountkw = {'stderr': subprocess.PIPE}
-        glusterprog = 'glusterfs'
-
-        @staticmethod
-        def make_umount_argv(d):
-            return ['umount', '-l', d]
-
-        def make_mount_argv(self):
-            self.mntpt = tempfile.mkdtemp(prefix = 'gsyncd-aux-mount-')
-            return [self.get_glusterprog()] + ['--' + p for p in self.params] + [self.mntpt]
-
-        def cleanup_mntpt(self, mntpt = None):
-            if not mntpt:
-                mntpt = self.mntpt
-            os.rmdir(mntpt)
-
-    class MountbrokerMounter(Mounter):
-        """mounter backend using the mountbroker gluster service"""
-
-        mountkw = {'stderr': subprocess.PIPE, 'stdout': subprocess.PIPE}
-        glusterprog = 'gluster'
-
-        @classmethod
-        def make_cli_argv(cls):
-            return [cls.get_glusterprog()] + gconf.gluster_cli_options.split() + ['system::']
-
-        @classmethod
-        def make_umount_argv(cls, d):
-            return cls.make_cli_argv() + ['umount', d, 'lazy']
-
-        def make_mount_argv(self, label):
-            return self.make_cli_argv() + \
-                   ['mount', label, 'user-map-root=' + syncdutils.getusername()] + self.params
-
-        def handle_mounter(self, po):
-            self.mntpt = po.stdout.readline()[:-1]
-            po.stdout.close()
-            sup(self, po)
-            if po.returncode != 0:
-                # if cli terminated with error due to being
-                # refused by glusterd, what it put
-                # out on stdout is a diagnostic message
-                logging.error('glusterd answered: %s' % self.mntpt)
-
-    def connect(self):
-        """inhibit the resource beyond
-
-        Choose mounting backend (direct or mountbroker),
-        set up glusterfs parameters and perform the mount
-        with given backend
-        """
-
-        label = getattr(gconf, 'mountbroker', None)
-        if not label and not privileged():
-            label = syncdutils.getusername()
-        mounter = label and self.MountbrokerMounter or self.DirectMounter
-        params = gconf.gluster_params.split() + \
-                   (gconf.gluster_log_level and ['log-level=' + gconf.gluster_log_level] or []) + \
-                   ['log-file=' + gconf.gluster_log_file, 'volfile-server=' + self.host,
-                    'volfile-id=' + self.volume, 'client-pid=-1']
-        mounter(params).inhibit(*[l for l in [label] if l])
-
-    def connect_remote(self, *a, **kw):
-        sup(self, *a, **kw)
-        self.slavedir = "/proc/%d/cwd" % self.server.pid()
-
-    def service_loop(self, *args):
-        """enter service loop
-
-        - if slave given, instantiate GMaster and
-          pass control to that instance, which implements
-          master behavior
-        - else do that's what's inherited
-        """
-        if args:
-            gmaster_builder()(self, args[0]).crawl_loop()
-        else:
-            sup(self, *args)
-
-    def rsync(self, files):
-        return sup(self, files, self.slavedir)
-
-
-class SSH(AbstractUrl, SlaveRemote):
-    """scheme class for ssh:// urls
-
-    interface to remote slave on master side
-    implementing an ssh based proxy
-    """
-
-    def __init__(self, path):
-        self.remote_addr, inner_url = sup(self, path,
-                                          '^((?:%s@)?%s):(.+)' % tuple([ r.pattern for r in (UserRX, HostRX) ]))
-        self.inner_rsc = parse_url(inner_url)
-
-    def canonical_path(self):
-        m = re.match('([^@]+)@(.+)', self.remote_addr)
-        if m:
-            u, h = m.groups()
-        else:
-            u, h = syncdutils.getusername(), self.remote_addr
-        remote_addr = '@'.join([u, gethostbyname(h)])
-        return ':'.join([remote_addr, self.inner_rsc.get_url(canonical=True)])
-
-    def can_connect_to(self, remote):
-        """determine our position in the connectibility matrix"""
-        return False
-
-    def start_fd_client(self, *a, **opts):
-        """customizations for client startup
-
-        - be a no-op if we are to daemonize (client startup is deferred
-          to post-daemon stage)
-        - determine target url for rsync after consulting server
-        """
-        if opts.get('deferred'):
-            return a
-        sup(self, *a)
-        ityp = type(self.inner_rsc)
-        if ityp == FILE:
-            slavepath = self.inner_rsc.path
-        elif ityp == GLUSTER:
-            slavepath = "/proc/%d/cwd" % self.server.pid()
-        else:
-            raise NotImplementedError
-        self.slaveurl = ':'.join([self.remote_addr, slavepath])
-
-    def connect_remote(self, go_daemon=None):
-        """connect to inner slave url through outer ssh url
-
-        Wrap the connecting utility in ssh.
-
-        Much care is put into daemonizing: in that case
-        ssh is started before daemonization, but
-        RePCe client is to be created after that (as ssh
-        interactive password auth would be defeated by
-        a daemonized ssh, while client should be present
-        only in the final process). In that case the action
-        is taken apart to two parts, this method is ivoked
-        once pre-daemon, once post-daemon. Use @go_daemon
-        to deiced what part to perform.
-
-        [NB. ATM gluster product does not makes use of interactive
-        authentication.]
-        """
-        if go_daemon == 'done':
-            return self.start_fd_client(*self.fd_pair)
-        gconf.setup_ssh_ctl(tempfile.mkdtemp(prefix='gsyncd-aux-ssh-'))
-        deferred = go_daemon == 'postconn'
-        ret = sup(self, gconf.ssh_command.split() + gconf.ssh_ctl_args + [self.remote_addr], slave=self.inner_rsc.url, deferred=deferred)
-        if deferred:
-            # send a message to peer so that we can wait for
-            # the answer from which we know connection is
-            # established and we can proceed with daemonization
-            # (doing that too early robs the ssh passwd prompt...)
-            # However, we'd better not start the RepceClient
-            # before daemonization (that's not preserved properly
-            # in daemon), we just do a an ad-hoc linear put/get.
-            i, o = ret
-            inf = os.fdopen(i)
-            repce.send(o, None, '__repce_version__')
-            select((inf,), (), ())
-            repce.recv(inf)
-            # hack hack hack: store a global reference to the file
-            # to save it from getting GC'd which implies closing it
-            gconf.permanent_handles.append(inf)
-            self.fd_pair = (i, o)
-            return 'should'
-
-    def rsync(self, files):
-        return sup(self, files, '-e', " ".join(gconf.ssh_command.split() + gconf.ssh_ctl_args),
-                   *(gconf.rsync_ssh_options.split() + [self.slaveurl]))
diff --git a/xlators/features/marker/utils/syncdaemon/syncdutils.py b/xlators/features/marker/utils/syncdaemon/syncdutils.py
deleted file mode 100644
index 0764c0790..000000000
--- a/xlators/features/marker/utils/syncdaemon/syncdutils.py
+++ /dev/null
@@ -1,288 +0,0 @@
-import os
-import sys
-import pwd
-import time
-import fcntl
-import shutil
-import logging
-from threading import Lock, Thread as baseThread
-from errno import EACCES, EAGAIN, EPIPE, ENOTCONN, ECONNABORTED, EINTR, errorcode
-from signal import signal, SIGTERM, SIGKILL
-from time import sleep
-import select as oselect
-from os import waitpid as owaitpid
-try:
-    from cPickle import PickleError
-except ImportError:
-    # py 3
-    from pickle import PickleError
-
-from gconf import gconf
-
-try:
-    # py 3
-    from urllib import parse as urllib
-except ImportError:
-    import urllib
-
-def escape(s):
-    """the chosen flavor of string escaping, used all over
-       to turn whatever data to creatable representation"""
-    return urllib.quote_plus(s)
-
-def unescape(s):
-    """inverse of .escape"""
-    return urllib.unquote_plus(s)
-
-def norm(s):
-    if s:
-        return s.replace('-', '_')
-
-def update_file(path, updater, merger = lambda f: True):
-    """update a file in a transaction-like manner"""
-
-    fr = fw = None
-    try:
-        fd = os.open(path, os.O_CREAT|os.O_RDWR)
-        try:
-            fr = os.fdopen(fd, 'r+b')
-        except:
-            os.close(fd)
-            raise
-        fcntl.lockf(fr, fcntl.LOCK_EX)
-        if not merger(fr):
-            return
-
-        tmpp = path + '.tmp.' + str(os.getpid())
-        fd = os.open(tmpp, os.O_CREAT|os.O_EXCL|os.O_WRONLY)
-        try:
-            fw = os.fdopen(fd, 'wb', 0)
-        except:
-            os.close(fd)
-            raise
-        updater(fw)
-        os.fsync(fd)
-        os.rename(tmpp, path)
-    finally:
-        for fx in (fr, fw):
-            if fx:
-                fx.close()
-
-def grabfile(fname, content=None):
-    """open @fname + contest for its fcntl lock
-
-    @content: if given, set the file content to it
-    """
-    # damn those messy open() mode codes
-    fd = os.open(fname, os.O_CREAT|os.O_RDWR)
-    f = os.fdopen(fd, 'r+b', 0)
-    try:
-        fcntl.lockf(f, fcntl.LOCK_EX|fcntl.LOCK_NB)
-    except:
-        ex = sys.exc_info()[1]
-        f.close()
-        if isinstance(ex, IOError) and ex.errno in (EACCES, EAGAIN):
-            # cannot grab, it's taken
-            return
-        raise
-    if content:
-        try:
-            f.truncate()
-            f.write(content)
-        except:
-            f.close()
-            raise
-    gconf.permanent_handles.append(f)
-    return f
-
-def grabpidfile(fname=None, setpid=True):
-    """.grabfile customization for pid files"""
-    if not fname:
-        fname = gconf.pid_file
-    content = None
-    if setpid:
-        content = str(os.getpid()) + '\n'
-    return grabfile(fname, content=content)
-
-final_lock = Lock()
-
-def finalize(*a, **kw):
-    """all those messy final steps we go trough upon termination
-
-    Do away with pidfile, ssh control dir and logging.
-    """
-    final_lock.acquire()
-    if getattr(gconf, 'pid_file', None):
-        rm_pidf = gconf.pid_file_owned
-        if gconf.cpid:
-            # exit path from parent branch of daemonization
-            rm_pidf = False
-            while True:
-                f = grabpidfile(setpid=False)
-                if not f:
-                    # child has already taken over pidfile
-                    break
-                if os.waitpid(gconf.cpid, os.WNOHANG)[0] == gconf.cpid:
-                    # child has terminated
-                    rm_pidf = True
-                    break;
-                time.sleep(0.1)
-        if rm_pidf:
-            try:
-                os.unlink(gconf.pid_file)
-            except:
-                ex = sys.exc_info()[1]
-                if ex.errno == ENOENT:
-                    pass
-                else:
-                    raise
-    if gconf.ssh_ctl_dir and not gconf.cpid:
-        shutil.rmtree(gconf.ssh_ctl_dir)
-    if getattr(gconf, 'state_socket', None):
-        try:
-            os.unlink(gconf.state_socket)
-        except:
-            if sys.exc_info()[0] == OSError:
-                pass
-    if gconf.log_exit:
-        logging.info("exiting.")
-    sys.stdout.flush()
-    sys.stderr.flush()
-    os._exit(kw.get('exval', 0))
-
-def log_raise_exception(excont):
-    """top-level exception handler
-
-    Try to some fancy things to cover up we face with an error.
-    Translate some weird sounding but well understood exceptions
-    into human-friendly lingo
-    """
-    is_filelog = False
-    for h in logging.getLogger().handlers:
-        fno = getattr(getattr(h, 'stream', None), 'fileno', None)
-        if fno and not os.isatty(fno()):
-            is_filelog = True
-
-    exc = sys.exc_info()[1]
-    if isinstance(exc, SystemExit):
-        excont.exval = exc.code or 0
-        raise
-    else:
-        logtag = None
-        if isinstance(exc, GsyncdError):
-            if is_filelog:
-                logging.error(exc.args[0])
-            sys.stderr.write('failure: ' + exc.args[0] + '\n')
-        elif isinstance(exc, PickleError) or isinstance(exc, EOFError) or \
-             ((isinstance(exc, OSError) or isinstance(exc, IOError)) and \
-              exc.errno == EPIPE):
-            logging.error('connection to peer is broken')
-            if hasattr(gconf, 'transport'):
-                gconf.transport.wait()
-                if gconf.transport.returncode == 127:
-                    logging.warn("!!!!!!!!!!!!!")
-                    logging.warn('!!! getting "No such file or directory" errors '
-                                 "is most likely due to MISCONFIGURATION, please consult "
-                                 "http://access.redhat.com/knowledge/docs/en-US/Red_Hat_Storage/2.0/html/Administration_Guide/chap-User_Guide-Geo_Rep-Preparation-Settingup_Environment.html")
-                    logging.warn("!!!!!!!!!!!!!")
-                gconf.transport.terminate_geterr()
-        elif isinstance(exc, OSError) and exc.errno in (ENOTCONN, ECONNABORTED):
-            logging.error('glusterfs session went down [%s]', errorcode[exc.errno])
-        else:
-            logtag = "FAIL"
-        if not logtag and logging.getLogger().isEnabledFor(logging.DEBUG):
-            logtag = "FULL EXCEPTION TRACE"
-        if logtag:
-            logging.exception(logtag + ": ")
-            sys.stderr.write("failed with %s.\n" % type(exc).__name__)
-        excont.exval = 1
-        sys.exit(excont.exval)
-
-
-class FreeObject(object):
-    """wildcard class for which any attribute can be set"""
-
-    def __init__(self, **kw):
-        for k,v in kw.items():
-            setattr(self, k, v)
-
-class Thread(baseThread):
-    """thread class flavor for gsyncd
-
-    - always a daemon thread
-    - force exit for whole program if thread
-      function coughs up an exception
-    """
-    def __init__(self, *a, **kw):
-        tf = kw.get('target')
-        if tf:
-            def twrap(*aa):
-                excont = FreeObject(exval = 0)
-                try:
-                    tf(*aa)
-                except:
-                    try:
-                        log_raise_exception(excont)
-                    finally:
-                        finalize(exval = excont.exval)
-            kw['target'] = twrap
-        baseThread.__init__(self, *a, **kw)
-        self.setDaemon(True)
-
-class GsyncdError(Exception):
-    pass
-
-def getusername(uid = None):
-    if uid == None:
-        uid = os.geteuid()
-    return pwd.getpwuid(uid).pw_name
-
-def privileged():
-    return os.geteuid() == 0
-
-def boolify(s):
-    """
-    Generic string to boolean converter
-
-    return
-    - Quick return if string 's' is of type bool
-    - True if it's in true_list
-    - False if it's in false_list
-    - Warn if it's not present in either and return False
-    """
-    true_list  = ['true', 'yes', '1', 'on']
-    false_list = ['false', 'no', '0', 'off']
-
-    if isinstance(s, bool):
-        return s
-
-    rv = False
-    lstr = s.lower()
-    if lstr in true_list:
-        rv = True
-    elif not lstr in false_list:
-        logging.warn("Unknown string (%s) in string to boolean conversion defaulting to False\n" % (s))
-
-    return rv
-
-def eintr_wrap(func, exc, *a):
-    """
-    wrapper around syscalls resilient to interrupt caused
-    by signals
-    """
-    while True:
-        try:
-            return func(*a)
-        except exc:
-            ex = sys.exc_info()[1]
-            if not ex.args[0] == EINTR:
-                raise
-
-def select(*a):
-    return eintr_wrap(oselect.select, oselect.error, *a)
-
-def waitpid (*a):
-    return eintr_wrap(owaitpid, OSError, *a)
-
-def set_term_handler(hook=lambda *a: finalize(*a, **{'exval': 1})):
-    signal(SIGTERM, hook)
diff --git a/xlators/features/qemu-block/Makefile.am b/xlators/features/qemu-block/Makefile.am
new file mode 100644
index 000000000..af437a64d
--- /dev/null
+++ b/xlators/features/qemu-block/Makefile.am
@@ -0,0 +1 @@
+SUBDIRS = src
diff --git a/xlators/features/qemu-block/src/Makefile.am b/xlators/features/qemu-block/src/Makefile.am
new file mode 100644
index 000000000..08a7b62a0
--- /dev/null
+++ b/xlators/features/qemu-block/src/Makefile.am
@@ -0,0 +1,155 @@
+if ENABLE_QEMU_BLOCK
+xlator_LTLIBRARIES = qemu-block.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+qemu_block_la_LDFLAGS = -module -avoid-version
+qemu_block_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la $(GLIB_LIBS) -lz -lrt
+
+qemu_block_la_SOURCES_qemu = \
+	$(CONTRIBDIR)/qemu/qemu-coroutine.c \
+	$(CONTRIBDIR)/qemu/qemu-coroutine-lock.c \
+	$(CONTRIBDIR)/qemu/qemu-coroutine-sleep.c \
+	$(CONTRIBDIR)/qemu/coroutine-ucontext.c \
+	$(CONTRIBDIR)/qemu/block.c \
+	$(CONTRIBDIR)/qemu/nop-symbols.c
+
+qemu_block_la_SOURCES_qemu_util = \
+	$(CONTRIBDIR)/qemu/util/aes.c \
+	$(CONTRIBDIR)/qemu/util/bitmap.c \
+	$(CONTRIBDIR)/qemu/util/bitops.c \
+	$(CONTRIBDIR)/qemu/util/cutils.c \
+	$(CONTRIBDIR)/qemu/util/error.c \
+	$(CONTRIBDIR)/qemu/util/hbitmap.c \
+	$(CONTRIBDIR)/qemu/util/iov.c \
+	$(CONTRIBDIR)/qemu/util/module.c \
+	$(CONTRIBDIR)/qemu/util/oslib-posix.c \
+	$(CONTRIBDIR)/qemu/util/qemu-option.c \
+	$(CONTRIBDIR)/qemu/util/qemu-error.c \
+	$(CONTRIBDIR)/qemu/util/qemu-thread-posix.c \
+	$(CONTRIBDIR)/qemu/util/unicode.c \
+	$(CONTRIBDIR)/qemu/util/hexdump.c
+
+qemu_block_la_SOURCES_qemu_block = \
+	$(CONTRIBDIR)/qemu/block/snapshot.c \
+	$(CONTRIBDIR)/qemu/block/qcow2-cache.c \
+	$(CONTRIBDIR)/qemu/block/qcow2-cluster.c \
+	$(CONTRIBDIR)/qemu/block/qcow2-refcount.c \
+	$(CONTRIBDIR)/qemu/block/qcow2-snapshot.c \
+	$(CONTRIBDIR)/qemu/block/qcow2.c \
+	$(CONTRIBDIR)/qemu/block/qed-check.c \
+	$(CONTRIBDIR)/qemu/block/qed-cluster.c \
+	$(CONTRIBDIR)/qemu/block/qed-gencb.c \
+	$(CONTRIBDIR)/qemu/block/qed-l2-cache.c \
+	$(CONTRIBDIR)/qemu/block/qed-table.c \
+	$(CONTRIBDIR)/qemu/block/qed.c
+
+qemu_block_la_SOURCES_qemu_qobject = \
+	$(CONTRIBDIR)/qemu/qobject/json-lexer.c \
+	$(CONTRIBDIR)/qemu/qobject/json-parser.c \
+	$(CONTRIBDIR)/qemu/qobject/json-streamer.c \
+	$(CONTRIBDIR)/qemu/qobject/qbool.c \
+	$(CONTRIBDIR)/qemu/qobject/qdict.c \
+	$(CONTRIBDIR)/qemu/qobject/qerror.c \
+	$(CONTRIBDIR)/qemu/qobject/qfloat.c \
+	$(CONTRIBDIR)/qemu/qobject/qint.c \
+	$(CONTRIBDIR)/qemu/qobject/qjson.c \
+	$(CONTRIBDIR)/qemu/qobject/qlist.c \
+	$(CONTRIBDIR)/qemu/qobject/qstring.c
+
+qemu_block_la_SOURCES = \
+	$(qemu_block_la_SOURCES_qemu) \
+	$(qemu_block_la_SOURCES_qemu_util) \
+	$(qemu_block_la_SOURCES_qemu_block) \
+	$(qemu_block_la_SOURCES_qemu_qobject) \
+	bdrv-xlator.c \
+	coroutine-synctask.c \
+	bh-syncop.c \
+	monitor-logging.c \
+	clock-timer.c \
+	qemu-block.c \
+	qb-coroutines.c
+
+noinst_HEADERS_qemu = \
+	$(CONTRIBDIR)/qemu/config-host.h \
+	$(CONTRIBDIR)/qemu/qapi-types.h \
+	$(CONTRIBDIR)/qemu/qmp-commands.h \
+	$(CONTRIBDIR)/qemu/trace/generated-tracers.h \
+	$(CONTRIBDIR)/qemu/include/config.h \
+	$(CONTRIBDIR)/qemu/include/glib-compat.h \
+	$(CONTRIBDIR)/qemu/include/qemu-common.h \
+	$(CONTRIBDIR)/qemu/include/trace.h \
+	$(CONTRIBDIR)/qemu/include/block/coroutine.h \
+	$(CONTRIBDIR)/qemu/include/block/aio.h \
+	$(CONTRIBDIR)/qemu/include/block/block.h \
+	$(CONTRIBDIR)/qemu/include/block/block_int.h \
+	$(CONTRIBDIR)/qemu/include/block/blockjob.h \
+	$(CONTRIBDIR)/qemu/include/block/coroutine.h \
+	$(CONTRIBDIR)/qemu/include/block/coroutine_int.h \
+	$(CONTRIBDIR)/qemu/include/block/snapshot.h \
+	$(CONTRIBDIR)/qemu/include/exec/cpu-common.h \
+	$(CONTRIBDIR)/qemu/include/exec/hwaddr.h \
+	$(CONTRIBDIR)/qemu/include/exec/poison.h \
+	$(CONTRIBDIR)/qemu/include/fpu/softfloat.h \
+	$(CONTRIBDIR)/qemu/include/migration/migration.h \
+	$(CONTRIBDIR)/qemu/include/migration/qemu-file.h \
+	$(CONTRIBDIR)/qemu/include/migration/vmstate.h \
+	$(CONTRIBDIR)/qemu/include/monitor/monitor.h \
+	$(CONTRIBDIR)/qemu/include/monitor/readline.h \
+	$(CONTRIBDIR)/qemu/include/qapi/error.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/json-lexer.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/json-parser.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/json-streamer.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/qbool.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/qdict.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/qerror.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/qfloat.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/qint.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/qjson.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/qlist.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/qobject.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/qstring.h \
+	$(CONTRIBDIR)/qemu/include/qapi/qmp/types.h \
+	$(CONTRIBDIR)/qemu/include/qemu/aes.h \
+	$(CONTRIBDIR)/qemu/include/qemu/atomic.h \
+	$(CONTRIBDIR)/qemu/include/qemu/bitmap.h \
+	$(CONTRIBDIR)/qemu/include/qemu/bitops.h \
+	$(CONTRIBDIR)/qemu/include/qemu/bswap.h \
+	$(CONTRIBDIR)/qemu/include/qemu/compiler.h \
+	$(CONTRIBDIR)/qemu/include/qemu/error-report.h \
+	$(CONTRIBDIR)/qemu/include/qemu/event_notifier.h \
+	$(CONTRIBDIR)/qemu/include/qemu/hbitmap.h \
+	$(CONTRIBDIR)/qemu/include/qemu/host-utils.h \
+	$(CONTRIBDIR)/qemu/include/qemu/iov.h \
+	$(CONTRIBDIR)/qemu/include/qemu/main-loop.h \
+	$(CONTRIBDIR)/qemu/include/qemu/module.h \
+	$(CONTRIBDIR)/qemu/include/qemu/notify.h \
+	$(CONTRIBDIR)/qemu/include/qemu/option.h \
+	$(CONTRIBDIR)/qemu/include/qemu/option_int.h \
+	$(CONTRIBDIR)/qemu/include/qemu/osdep.h \
+	$(CONTRIBDIR)/qemu/include/qemu/queue.h \
+	$(CONTRIBDIR)/qemu/include/qemu/sockets.h \
+	$(CONTRIBDIR)/qemu/include/qemu/thread-posix.h \
+	$(CONTRIBDIR)/qemu/include/qemu/thread.h \
+	$(CONTRIBDIR)/qemu/include/qemu/timer.h \
+	$(CONTRIBDIR)/qemu/include/qemu/typedefs.h \
+	$(CONTRIBDIR)/qemu/include/sysemu/sysemu.h \
+	$(CONTRIBDIR)/qemu/include/sysemu/os-posix.h \
+	$(CONTRIBDIR)/qemu/block/qcow2.h \
+	$(CONTRIBDIR)/qemu/block/qed.h
+
+noinst_HEADERS = \
+	$(noinst_HEADERS_qemu) \
+	qemu-block.h \
+	qemu-block-memory-types.h \
+	qb-coroutines.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+            -I$(CONTRIBDIR)/qemu \
+            -I$(CONTRIBDIR)/qemu/include \
+            -DGLUSTER_XLATOR
+
+AM_CFLAGS = -fno-strict-aliasing -Wall $(GF_CFLAGS) $(GLIB_CFLAGS)
+
+CLEANFILES =
+
+endif
diff --git a/xlators/features/qemu-block/src/bdrv-xlator.c b/xlators/features/qemu-block/src/bdrv-xlator.c
new file mode 100644
index 000000000..106c59775
--- /dev/null
+++ b/xlators/features/qemu-block/src/bdrv-xlator.c
@@ -0,0 +1,397 @@
+/*
+  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 "inode.h"
+#include "syncop.h"
+#include "qemu-block.h"
+#include "block/block_int.h"
+
+typedef struct BDRVGlusterState {
+	inode_t *inode;
+} BDRVGlusterState;
+
+static QemuOptsList runtime_opts = {
+	.name = "gluster",
+	.head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
+	.desc = {
+		{
+			.name = "filename",
+			.type = QEMU_OPT_STRING,
+			.help = "GFID of file",
+		},
+		{ /* end of list */ }
+	},
+};
+
+inode_t *
+qb_inode_from_filename (const char *filename)
+{
+	const char *iptr = NULL;
+	inode_t *inode = NULL;
+
+	iptr = filename + 17;
+	sscanf (iptr, "%p", &inode);
+
+	return inode;
+}
+
+
+int
+qb_inode_to_filename (inode_t *inode, char *filename, int size)
+{
+	return snprintf (filename, size, "gluster://inodep:%p", inode);
+}
+
+
+static fd_t *
+fd_from_bs (BlockDriverState *bs)
+{
+	BDRVGlusterState *s = bs->opaque;
+
+	return fd_anonymous (s->inode);
+}
+
+
+static int
+qemu_gluster_open (BlockDriverState *bs, QDict *options, int bdrv_flags)
+{
+	inode_t *inode = NULL;
+	BDRVGlusterState *s = bs->opaque;
+	QemuOpts *opts = NULL;
+	Error *local_err = NULL;
+	const char *filename = NULL;
+	char gfid_str[128];
+	int ret;
+	qb_conf_t *conf = THIS->private;
+
+	opts = qemu_opts_create_nofail(&runtime_opts);
+	qemu_opts_absorb_qdict(opts, options, &local_err);
+	if (error_is_set(&local_err)) {
+		qerror_report_err(local_err);
+		error_free(local_err);
+		return -EINVAL;
+	}
+
+	filename = qemu_opt_get(opts, "filename");
+
+	/*
+	 * gfid:<gfid> format means we're opening a backing image.
+	 */
+	ret = sscanf(filename, "gluster://gfid:%s", gfid_str);
+	if (ret) {
+		loc_t loc = {0,};
+		struct iatt buf = {0,};
+		uuid_t gfid;
+
+		uuid_parse(gfid_str, gfid);
+
+		loc.inode = inode_find(conf->root_inode->table, gfid);
+		if (!loc.inode) {
+			loc.inode = inode_new(conf->root_inode->table);
+			uuid_copy(loc.inode->gfid, gfid);
+		}
+
+		uuid_copy(loc.gfid, loc.inode->gfid);
+		ret = syncop_lookup(FIRST_CHILD(THIS), &loc, NULL, &buf, NULL,
+				    NULL);
+		if (ret) {
+			loc_wipe(&loc);
+			return -errno;
+		}
+
+		s->inode = inode_ref(loc.inode);
+		loc_wipe(&loc);
+	} else {
+		inode = qb_inode_from_filename (filename);
+		if (!inode)
+			return -EINVAL;
+
+		s->inode = inode_ref(inode);
+	}
+
+	return 0;
+}
+
+
+static int
+qemu_gluster_create (const char *filename, QEMUOptionParameter *options)
+{
+	uint64_t total_size = 0;
+	inode_t *inode = NULL;
+	fd_t *fd = NULL;
+	struct iatt stat = {0, };
+	int ret = 0;
+
+	inode = qb_inode_from_filename (filename);
+	if (!inode)
+		return -EINVAL;
+
+	while (options && options->name) {
+		if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
+			total_size = options->value.n / BDRV_SECTOR_SIZE;
+		}
+		options++;
+	}
+
+	fd = fd_anonymous (inode);
+	if (!fd)
+		return -ENOMEM;
+
+	ret = syncop_fstat (FIRST_CHILD(THIS), fd, &stat);
+	if (ret) {
+		fd_unref (fd);
+		return -errno;
+	}
+
+	if (stat.ia_size) {
+		/* format ONLY if the filesize is 0 bytes */
+		fd_unref (fd);
+		return -EFBIG;
+	}
+
+	if (total_size) {
+		ret = syncop_ftruncate (FIRST_CHILD(THIS), fd, total_size);
+		if (ret) {
+			fd_unref (fd);
+			return -errno;
+		}
+	}
+
+	fd_unref (fd);
+	return 0;
+}
+
+
+static int
+qemu_gluster_co_readv (BlockDriverState *bs, int64_t sector_num, int nb_sectors,
+		       QEMUIOVector *qiov)
+{
+	fd_t *fd = NULL;
+	off_t offset = 0;
+	size_t size = 0;
+	struct iovec *iov = NULL;
+	int count = 0;
+	struct iobref *iobref = NULL;
+	int ret = 0;
+
+	fd = fd_from_bs (bs);
+	if (!fd)
+		return -EIO;
+
+	offset = sector_num * BDRV_SECTOR_SIZE;
+	size = nb_sectors * BDRV_SECTOR_SIZE;
+
+	ret = syncop_readv (FIRST_CHILD(THIS), fd, size, offset, 0,
+			    &iov, &count, &iobref);
+	if (ret < 0) {
+		ret = -errno;
+		goto out;
+	}
+
+	iov_copy (qiov->iov, qiov->niov, iov, count); /* *choke!* */
+
+out:
+	GF_FREE (iov);
+	if (iobref)
+		iobref_unref (iobref);
+	fd_unref (fd);
+	return ret;
+}
+
+
+static int
+qemu_gluster_co_writev (BlockDriverState *bs, int64_t sector_num, int nb_sectors,
+			QEMUIOVector *qiov)
+{
+	fd_t *fd = NULL;
+	off_t offset = 0;
+	size_t size = 0;
+	struct iobref *iobref = NULL;
+	struct iobuf *iobuf = NULL;
+	struct iovec iov = {0, };
+	int ret = -ENOMEM;
+
+	fd = fd_from_bs (bs);
+	if (!fd)
+		return -EIO;
+
+	offset = sector_num * BDRV_SECTOR_SIZE;
+	size = nb_sectors * BDRV_SECTOR_SIZE;
+
+	iobuf = iobuf_get2 (THIS->ctx->iobuf_pool, size);
+	if (!iobuf)
+		goto out;
+
+	iobref = iobref_new ();
+	if (!iobref) {
+		iobuf_unref (iobuf);
+		goto out;
+	}
+
+	iobref_add (iobref, iobuf);
+
+	iov_unload (iobuf_ptr (iobuf), qiov->iov, qiov->niov); /* *choke!* */
+
+	iov.iov_base = iobuf_ptr (iobuf);
+	iov.iov_len = size;
+
+	ret = syncop_writev (FIRST_CHILD(THIS), fd, &iov, 1, offset, iobref, 0);
+	if (ret < 0)
+		ret = -errno;
+
+out:
+	if (iobuf)
+		iobuf_unref (iobuf);
+	if (iobref)
+		iobref_unref (iobref);
+	fd_unref (fd);
+	return ret;
+}
+
+
+static int
+qemu_gluster_co_flush (BlockDriverState *bs)
+{
+	fd_t *fd = NULL;
+	int ret = 0;
+
+	fd = fd_from_bs (bs);
+
+	ret = syncop_flush (FIRST_CHILD(THIS), fd);
+
+	fd_unref (fd);
+
+	return ret;
+}
+
+
+static int
+qemu_gluster_co_fsync (BlockDriverState *bs)
+{
+	fd_t *fd = NULL;
+	int ret = 0;
+
+	fd = fd_from_bs (bs);
+
+	ret = syncop_fsync (FIRST_CHILD(THIS), fd, 0);
+
+	fd_unref (fd);
+
+	return ret;
+}
+
+
+static int
+qemu_gluster_truncate (BlockDriverState *bs, int64_t offset)
+{
+	fd_t *fd = NULL;
+	int ret = 0;
+
+	fd = fd_from_bs (bs);
+
+	ret = syncop_ftruncate (FIRST_CHILD(THIS), fd, offset);
+
+	fd_unref (fd);
+
+	if (ret < 0)
+		return ret;
+
+	return ret;
+}
+
+
+static int64_t
+qemu_gluster_getlength (BlockDriverState *bs)
+{
+	fd_t *fd = NULL;
+	int ret = 0;
+	struct iatt iatt = {0, };
+
+	fd = fd_from_bs (bs);
+
+	ret = syncop_fstat (FIRST_CHILD(THIS), fd, &iatt);
+	if (ret < 0)
+		return -1;
+
+	return iatt.ia_size;
+}
+
+
+static int64_t
+qemu_gluster_allocated_file_size (BlockDriverState *bs)
+{
+	fd_t *fd = NULL;
+	int ret = 0;
+	struct iatt iatt = {0, };
+
+	fd = fd_from_bs (bs);
+
+	ret = syncop_fstat (FIRST_CHILD(THIS), fd, &iatt);
+	if (ret < 0)
+		return -1;
+
+	return iatt.ia_blocks * 512;
+}
+
+
+static void
+qemu_gluster_close (BlockDriverState *bs)
+{
+	BDRVGlusterState *s = NULL;
+
+	s = bs->opaque;
+
+	inode_unref (s->inode);
+
+	return;
+}
+
+
+static QEMUOptionParameter qemu_gluster_create_options[] = {
+	{
+		.name = BLOCK_OPT_SIZE,
+		.type = OPT_SIZE,
+		.help = "Virtual disk size"
+	},
+	{ NULL }
+};
+
+
+static BlockDriver bdrv_gluster = {
+	.format_name                  = "gluster",
+	.protocol_name                = "gluster",
+	.instance_size                = sizeof(BDRVGlusterState),
+	.bdrv_file_open               = qemu_gluster_open,
+	.bdrv_close                   = qemu_gluster_close,
+	.bdrv_create                  = qemu_gluster_create,
+	.bdrv_getlength               = qemu_gluster_getlength,
+	.bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
+	.bdrv_co_readv                = qemu_gluster_co_readv,
+	.bdrv_co_writev               = qemu_gluster_co_writev,
+	.bdrv_co_flush_to_os          = qemu_gluster_co_flush,
+	.bdrv_co_flush_to_disk        = qemu_gluster_co_fsync,
+	.bdrv_truncate                = qemu_gluster_truncate,
+	.create_options               = qemu_gluster_create_options,
+};
+
+
+static void bdrv_gluster_init(void)
+{
+	bdrv_register(&bdrv_gluster);
+}
+
+
+block_init(bdrv_gluster_init);
diff --git a/xlators/features/qemu-block/src/bh-syncop.c b/xlators/features/qemu-block/src/bh-syncop.c
new file mode 100644
index 000000000..e8686f6d4
--- /dev/null
+++ b/xlators/features/qemu-block/src/bh-syncop.c
@@ -0,0 +1,48 @@
+/*
+  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 "glusterfs.h"
+#include "logging.h"
+#include "dict.h"
+#include "xlator.h"
+#include "syncop.h"
+#include "qemu-block-memory-types.h"
+
+#include "block/aio.h"
+
+void
+qemu_bh_schedule (QEMUBH *bh)
+{
+	return;
+}
+
+void
+qemu_bh_cancel (QEMUBH *bh)
+{
+	return;
+}
+
+void
+qemu_bh_delete (QEMUBH *bh)
+{
+
+}
+
+QEMUBH *
+qemu_bh_new (QEMUBHFunc *cb, void *opaque)
+{
+	return NULL;
+}
diff --git a/xlators/features/qemu-block/src/clock-timer.c b/xlators/features/qemu-block/src/clock-timer.c
new file mode 100644
index 000000000..fcbec6ad1
--- /dev/null
+++ b/xlators/features/qemu-block/src/clock-timer.c
@@ -0,0 +1,60 @@
+/*
+  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 "glusterfs.h"
+#include "logging.h"
+#include "dict.h"
+#include "xlator.h"
+#include "syncop.h"
+#include "qemu-block-memory-types.h"
+
+#include "qemu/timer.h"
+
+QEMUClock *vm_clock;
+int use_rt_clock = 0;
+
+QEMUTimer *qemu_new_timer (QEMUClock *clock, int scale,
+			   QEMUTimerCB *cb, void *opaque)
+{
+	return NULL;
+}
+
+int64_t qemu_get_clock_ns (QEMUClock *clock)
+{
+	return 0;
+}
+
+void qemu_mod_timer (QEMUTimer *ts, int64_t expire_time)
+{
+	return;
+}
+
+void qemu_free_timer (QEMUTimer *ts)
+{
+
+}
+
+void qemu_del_timer (QEMUTimer *ts)
+{
+
+}
+
+bool qemu_aio_wait()
+{
+	synctask_wake (synctask_get());
+	synctask_yield (synctask_get());
+	return 0;
+}
diff --git a/xlators/features/qemu-block/src/coroutine-synctask.c b/xlators/features/qemu-block/src/coroutine-synctask.c
new file mode 100644
index 000000000..e43988a95
--- /dev/null
+++ b/xlators/features/qemu-block/src/coroutine-synctask.c
@@ -0,0 +1,116 @@
+/*
+  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 "glusterfs.h"
+#include "logging.h"
+#include "dict.h"
+#include "xlator.h"
+#include "syncop.h"
+#include "qemu-block-memory-types.h"
+
+#include "qemu-block.h"
+
+/*
+ * This code serves as the bridge from the main glusterfs context to the qemu
+ * coroutine context via synctask. We create a single threaded syncenv with a
+ * single synctask responsible for processing a queue of coroutines. The qemu
+ * code invoked from within the synctask function handlers uses the ucontext
+ * coroutine implementation and scheduling logic internal to qemu. This
+ * effectively donates a thread of execution to qemu and its internal coroutine
+ * management.
+ *
+ * NOTE: The existence of concurrent synctasks has proven quite racy with regard
+ * to qemu coroutine management, particularly related to the lifecycle
+ * differences with top-level synctasks and internally created coroutines and
+ * interactions with qemu-internal queues (and locks, in turn). We explicitly
+ * disallow this scenario, via the queue, until it is more well supported.
+ */
+
+static struct {
+	struct list_head	queue;
+	gf_lock_t		lock;
+	struct synctask		*task;
+} qb_co;
+
+static void
+init_qbco()
+{
+	INIT_LIST_HEAD(&qb_co.queue);
+	LOCK_INIT(&qb_co.lock);
+}
+
+static int
+synctask_nop_cbk (int ret, call_frame_t *frame, void *opaque)
+{
+	return 0;
+}
+
+static int
+qb_synctask_wrap (void *opaque)
+{
+	qb_local_t *qb_local, *tmp;
+
+	LOCK(&qb_co.lock);
+
+	while (!list_empty(&qb_co.queue)) {
+		list_for_each_entry_safe(qb_local, tmp, &qb_co.queue, list) {
+			list_del_init(&qb_local->list);
+			break;
+		}
+
+		UNLOCK(&qb_co.lock);
+
+		qb_local->synctask_fn(qb_local);
+		/* qb_local is now unwound and gone! */
+
+		LOCK(&qb_co.lock);
+	}
+
+	qb_co.task = NULL;
+
+	UNLOCK(&qb_co.lock);
+
+	return 0;
+}
+
+int
+qb_coroutine (call_frame_t *frame, synctask_fn_t fn)
+{
+	qb_local_t *qb_local = NULL;
+	qb_conf_t *qb_conf = NULL;
+	static int init = 0;
+
+	qb_local = frame->local;
+	qb_local->synctask_fn = fn;
+	qb_conf = frame->this->private;
+
+	if (!init) {
+		init = 1;
+		init_qbco();
+	}
+
+	LOCK(&qb_co.lock);
+
+	if (!qb_co.task)
+		qb_co.task = synctask_create(qb_conf->env, qb_synctask_wrap,
+					     synctask_nop_cbk, frame, NULL);
+
+	list_add_tail(&qb_local->list, &qb_co.queue);
+
+	UNLOCK(&qb_co.lock);
+
+	return 0;
+}
diff --git a/xlators/features/qemu-block/src/monitor-logging.c b/xlators/features/qemu-block/src/monitor-logging.c
new file mode 100644
index 000000000..d37c37f0f
--- /dev/null
+++ b/xlators/features/qemu-block/src/monitor-logging.c
@@ -0,0 +1,50 @@
+/*
+  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 "glusterfs.h"
+#include "logging.h"
+#include "dict.h"
+#include "xlator.h"
+#include "qemu-block-memory-types.h"
+
+#include "block/block_int.h"
+
+Monitor *cur_mon;
+
+int
+monitor_cur_is_qmp()
+{
+	/* No QMP support here */
+	return 0;
+}
+
+void
+monitor_set_error (Monitor *mon, QError *qerror)
+{
+	/* NOP here */
+	return;
+}
+
+
+void
+monitor_vprintf(Monitor *mon, const char *fmt, va_list ap)
+{
+	char buf[4096];
+
+	vsnprintf(buf, sizeof(buf), fmt, ap);
+
+	gf_log (THIS->name, GF_LOG_ERROR, "%s", buf);
+}
diff --git a/xlators/features/qemu-block/src/qb-coroutines.c b/xlators/features/qemu-block/src/qb-coroutines.c
new file mode 100644
index 000000000..7c52adb21
--- /dev/null
+++ b/xlators/features/qemu-block/src/qb-coroutines.c
@@ -0,0 +1,662 @@
+/*
+  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 "glusterfs.h"
+#include "logging.h"
+#include "dict.h"
+#include "xlator.h"
+#include "inode.h"
+#include "call-stub.h"
+#include "defaults.h"
+#include "qemu-block-memory-types.h"
+#include "qemu-block.h"
+#include "qb-coroutines.h"
+
+
+int
+qb_format_and_resume (void *opaque)
+{
+	qb_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	call_stub_t *stub = NULL;
+	inode_t *inode = NULL;
+	char filename[64];
+	char base_filename[128];
+	int use_base = 0;
+	qb_inode_t *qb_inode = NULL;
+	Error *local_err = NULL;
+	fd_t *fd = NULL;
+	dict_t *xattr = NULL;
+	qb_conf_t *qb_conf = NULL;
+	int ret = -1;
+
+	local = opaque;
+	frame = local->frame;
+	stub = local->stub;
+	inode = local->inode;
+	qb_conf = frame->this->private;
+
+	qb_inode_to_filename (inode, filename, 64);
+
+	qb_inode = qb_inode_ctx_get (frame->this, inode);
+
+	/*
+	 * See if the caller specified a backing image.
+	 */
+	if (!uuid_is_null(qb_inode->backing_gfid) || qb_inode->backing_fname) {
+		loc_t loc = {0,};
+		char gfid_str[64];
+		struct iatt buf;
+
+		if (!uuid_is_null(qb_inode->backing_gfid)) {
+			loc.inode = inode_find(qb_conf->root_inode->table,
+					qb_inode->backing_gfid);
+			if (!loc.inode) {
+				loc.inode = inode_new(qb_conf->root_inode->table);
+				uuid_copy(loc.inode->gfid,
+					qb_inode->backing_gfid);
+			}
+			uuid_copy(loc.gfid, loc.inode->gfid);
+		} else if (qb_inode->backing_fname) {
+			loc.inode = inode_new(qb_conf->root_inode->table);
+			loc.name = qb_inode->backing_fname;
+			loc.parent = inode_parent(inode, NULL, NULL);
+			loc_path(&loc, loc.name);
+		}
+
+		/*
+		 * Lookup the backing image. Verify existence and/or get the
+		 * gfid if we don't already have it.
+		 */
+		ret = syncop_lookup(FIRST_CHILD(frame->this), &loc, NULL, &buf,
+				    NULL, NULL);
+		GF_FREE(qb_inode->backing_fname);
+		if (ret) {
+			loc_wipe(&loc);
+			ret = errno;
+			goto err;
+		}
+
+		uuid_copy(qb_inode->backing_gfid, buf.ia_gfid);
+		loc_wipe(&loc);
+
+		/*
+		 * We pass the filename of the backing image into the qemu block
+		 * subsystem as the associated gfid. This is embedded into the
+		 * clone image and passed along to the gluster bdrv backend when
+		 * the block subsystem needs to operate on the backing image on
+		 * behalf of the clone.
+		 */
+		uuid_unparse(qb_inode->backing_gfid, gfid_str);
+		snprintf(base_filename, sizeof(base_filename),
+			 "gluster://gfid:%s", gfid_str);
+		use_base = 1;
+	}
+
+	bdrv_img_create (filename, qb_inode->fmt,
+			 use_base ? base_filename : NULL, 0, 0, qb_inode->size,
+			 0, &local_err, true);
+
+	if (error_is_set (&local_err)) {
+		gf_log (frame->this->name, GF_LOG_ERROR, "%s",
+			error_get_pretty (local_err));
+		error_free (local_err);
+		QB_STUB_UNWIND (stub, -1, EIO);
+		return 0;
+	}
+
+	fd = fd_anonymous (inode);
+	if (!fd) {
+		gf_log (frame->this->name, GF_LOG_ERROR,
+			"could not create anonymous fd for %s",
+			uuid_utoa (inode->gfid));
+		QB_STUB_UNWIND (stub, -1, ENOMEM);
+		return 0;
+	}
+
+	xattr = dict_new ();
+	if (!xattr) {
+		gf_log (frame->this->name, GF_LOG_ERROR,
+			"could not allocate xattr dict for %s",
+			uuid_utoa (inode->gfid));
+		QB_STUB_UNWIND (stub, -1, ENOMEM);
+		fd_unref (fd);
+		return 0;
+	}
+
+	ret = dict_set_str (xattr, qb_conf->qb_xattr_key, local->fmt);
+	if (ret) {
+		gf_log (frame->this->name, GF_LOG_ERROR,
+			"could not dict_set for %s",
+			uuid_utoa (inode->gfid));
+		QB_STUB_UNWIND (stub, -1, ENOMEM);
+		fd_unref (fd);
+		dict_unref (xattr);
+		return 0;
+	}
+
+	ret = syncop_fsetxattr (FIRST_CHILD(THIS), fd, xattr, 0);
+	if (ret) {
+		ret = errno;
+		gf_log (frame->this->name, GF_LOG_ERROR,
+			"failed to setxattr for %s",
+			uuid_utoa (inode->gfid));
+		QB_STUB_UNWIND (stub, -1, ret);
+		fd_unref (fd);
+		dict_unref (xattr);
+		return 0;
+	}
+
+	fd_unref (fd);
+	dict_unref (xattr);
+
+	QB_STUB_UNWIND (stub, 0, 0);
+
+	return 0;
+
+err:
+	QB_STUB_UNWIND(stub, -1, ret);
+	return 0;
+}
+
+
+static BlockDriverState *
+qb_bs_create (inode_t *inode, const char *fmt)
+{
+	char filename[64];
+	BlockDriverState *bs = NULL;
+	BlockDriver *drv = NULL;
+	int op_errno = 0;
+	int ret = 0;
+
+	bs = bdrv_new (uuid_utoa (inode->gfid));
+	if (!bs) {
+		op_errno = ENOMEM;
+		gf_log (THIS->name, GF_LOG_ERROR,
+			"could not allocate @bdrv for gfid:%s",
+			uuid_utoa (inode->gfid));
+		goto err;
+	}
+
+	drv = bdrv_find_format (fmt);
+	if (!drv) {
+		op_errno = EINVAL;
+		gf_log (THIS->name, GF_LOG_ERROR,
+			"Unknown file format: %s for gfid:%s",
+			fmt, uuid_utoa (inode->gfid));
+		goto err;
+	}
+
+	qb_inode_to_filename (inode, filename, 64);
+
+	ret = bdrv_open (bs, filename, NULL, BDRV_O_RDWR, drv);
+	if (ret < 0) {
+		op_errno = -ret;
+		gf_log (THIS->name, GF_LOG_ERROR,
+			"Unable to bdrv_open() gfid:%s (%s)",
+			uuid_utoa (inode->gfid), strerror (op_errno));
+		goto err;
+	}
+
+	return bs;
+err:
+	errno = op_errno;
+	return NULL;
+}
+
+
+int
+qb_co_open (void *opaque)
+{
+	qb_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	call_stub_t *stub = NULL;
+	inode_t *inode = NULL;
+	qb_inode_t *qb_inode = NULL;
+
+	local = opaque;
+	frame = local->frame;
+	stub = local->stub;
+	inode = local->inode;
+
+	qb_inode = qb_inode_ctx_get (frame->this, inode);
+	if (!qb_inode->bs) {
+		/* FIXME: we need locks around this when
+		   enabling multithreaded syncop/coroutine
+		   for qemu-block
+		*/
+
+		qb_inode->bs = qb_bs_create (inode, qb_inode->fmt);
+		if (!qb_inode->bs) {
+			QB_STUB_UNWIND (stub, -1, errno);
+			return 0;
+		}
+	}
+	qb_inode->refcnt++;
+
+	QB_STUB_RESUME (stub);
+
+	return 0;
+}
+
+
+int
+qb_co_writev (void *opaque)
+{
+	qb_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	call_stub_t *stub = NULL;
+	inode_t *inode = NULL;
+	qb_inode_t *qb_inode = NULL;
+	QEMUIOVector qiov = {0, };
+	int ret = 0;
+
+	local = opaque;
+	frame = local->frame;
+	stub = local->stub;
+	inode = local->inode;
+
+	qb_inode = qb_inode_ctx_get (frame->this, inode);
+	if (!qb_inode->bs) {
+		/* FIXME: we need locks around this when
+		   enabling multithreaded syncop/coroutine
+		   for qemu-block
+		*/
+
+		qb_inode->bs = qb_bs_create (inode, qb_inode->fmt);
+		if (!qb_inode->bs) {
+			QB_STUB_UNWIND (stub, -1, errno);
+			return 0;
+		}
+	}
+
+	qemu_iovec_init_external (&qiov, stub->args.vector, stub->args.count);
+
+	ret = bdrv_pwritev (qb_inode->bs, stub->args.offset, &qiov);
+
+	if (ret < 0) {
+		QB_STUB_UNWIND (stub, -1, -ret);
+	} else {
+		QB_STUB_UNWIND (stub, ret, 0);
+	}
+
+	return 0;
+}
+
+
+int
+qb_co_readv (void *opaque)
+{
+	qb_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	call_stub_t *stub = NULL;
+	inode_t *inode = NULL;
+	qb_inode_t *qb_inode = NULL;
+	struct iobuf *iobuf = NULL;
+	struct iobref *iobref = NULL;
+	struct iovec iov = {0, };
+	int ret = 0;
+
+	local = opaque;
+	frame = local->frame;
+	stub = local->stub;
+	inode = local->inode;
+
+	qb_inode = qb_inode_ctx_get (frame->this, inode);
+	if (!qb_inode->bs) {
+		/* FIXME: we need locks around this when
+		   enabling multithreaded syncop/coroutine
+		   for qemu-block
+		*/
+
+		qb_inode->bs = qb_bs_create (inode, qb_inode->fmt);
+		if (!qb_inode->bs) {
+			QB_STUB_UNWIND (stub, -1, errno);
+			return 0;
+		}
+	}
+
+	if (stub->args.offset >= qb_inode->size) {
+		QB_STUB_UNWIND (stub, 0, 0);
+		return 0;
+	}
+
+	iobuf = iobuf_get2 (frame->this->ctx->iobuf_pool, stub->args.size);
+	if (!iobuf) {
+		QB_STUB_UNWIND (stub, -1, ENOMEM);
+		return 0;
+	}
+
+	iobref = iobref_new ();
+	if (!iobref) {
+		QB_STUB_UNWIND (stub, -1, ENOMEM);
+		iobuf_unref (iobuf);
+		return 0;
+	}
+
+	if (iobref_add (iobref, iobuf) < 0) {
+		iobuf_unref (iobuf);
+		iobref_unref (iobref);
+		QB_STUB_UNWIND (stub, -1, ENOMEM);
+		return 0;
+	}
+
+	ret = bdrv_pread (qb_inode->bs, stub->args.offset, iobuf_ptr (iobuf),
+			  stub->args.size);
+
+	if (ret < 0) {
+		QB_STUB_UNWIND (stub, -1, -ret);
+		iobref_unref (iobref);
+		return 0;
+	}
+
+	iov.iov_base = iobuf_ptr (iobuf);
+	iov.iov_len = ret;
+
+	stub->args_cbk.vector = iov_dup (&iov, 1);
+	stub->args_cbk.count = 1;
+	stub->args_cbk.iobref = iobref;
+
+	QB_STUB_UNWIND (stub, ret, 0);
+
+	return 0;
+}
+
+
+int
+qb_co_fsync (void *opaque)
+{
+	qb_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	call_stub_t *stub = NULL;
+	inode_t *inode = NULL;
+	qb_inode_t *qb_inode = NULL;
+	int ret = 0;
+
+	local = opaque;
+	frame = local->frame;
+	stub = local->stub;
+	inode = local->inode;
+
+	qb_inode = qb_inode_ctx_get (frame->this, inode);
+	if (!qb_inode->bs) {
+		/* FIXME: we need locks around this when
+		   enabling multithreaded syncop/coroutine
+		   for qemu-block
+		*/
+
+		qb_inode->bs = qb_bs_create (inode, qb_inode->fmt);
+		if (!qb_inode->bs) {
+			QB_STUB_UNWIND (stub, -1, errno);
+			return 0;
+		}
+	}
+
+	ret = bdrv_flush (qb_inode->bs);
+
+	if (ret < 0) {
+		QB_STUB_UNWIND (stub, -1, -ret);
+	} else {
+		QB_STUB_UNWIND (stub, ret, 0);
+	}
+
+	return 0;
+}
+
+
+static void
+qb_update_size_xattr (xlator_t *this, fd_t *fd, const char *fmt, off_t offset)
+{
+	char val[QB_XATTR_VAL_MAX];
+	qb_conf_t *qb_conf = NULL;
+	dict_t *xattr = NULL;
+
+	qb_conf = this->private;
+
+	snprintf (val, QB_XATTR_VAL_MAX, "%s:%llu",
+		  fmt, (long long unsigned) offset);
+
+	xattr = dict_new ();
+	if (!xattr)
+		return;
+
+	if (dict_set_str (xattr, qb_conf->qb_xattr_key, val) != 0) {
+		dict_unref (xattr);
+		return;
+	}
+
+	syncop_fsetxattr (FIRST_CHILD(this), fd, xattr, 0);
+	dict_unref (xattr);
+}
+
+
+int
+qb_co_truncate (void *opaque)
+{
+	qb_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	call_stub_t *stub = NULL;
+	inode_t *inode = NULL;
+	qb_inode_t *qb_inode = NULL;
+	int ret = 0;
+	off_t offset = 0;
+	xlator_t *this = NULL;
+
+	this = THIS;
+
+	local = opaque;
+	frame = local->frame;
+	stub = local->stub;
+	inode = local->inode;
+
+	qb_inode = qb_inode_ctx_get (frame->this, inode);
+	if (!qb_inode->bs) {
+		/* FIXME: we need locks around this when
+		   enabling multithreaded syncop/coroutine
+		   for qemu-block
+		*/
+
+		qb_inode->bs = qb_bs_create (inode, qb_inode->fmt);
+		if (!qb_inode->bs) {
+			QB_STUB_UNWIND (stub, -1, errno);
+			return 0;
+		}
+	}
+
+	syncop_fstat (FIRST_CHILD(this), local->fd, &stub->args_cbk.prestat);
+	stub->args_cbk.prestat.ia_size = qb_inode->size;
+
+	ret = bdrv_truncate (qb_inode->bs, stub->args.offset);
+	if (ret < 0)
+		goto out;
+
+	offset = bdrv_getlength (qb_inode->bs);
+
+	qb_inode->size = offset;
+
+	syncop_fstat (FIRST_CHILD(this), local->fd, &stub->args_cbk.poststat);
+	stub->args_cbk.poststat.ia_size = qb_inode->size;
+
+	qb_update_size_xattr (this, local->fd, qb_inode->fmt, qb_inode->size);
+
+out:
+	if (ret < 0) {
+		QB_STUB_UNWIND (stub, -1, -ret);
+	} else {
+		QB_STUB_UNWIND (stub, ret, 0);
+	}
+
+	return 0;
+}
+
+
+int
+qb_co_close (void *opaque)
+{
+	qb_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	inode_t *inode = NULL;
+	qb_inode_t *qb_inode = NULL;
+	BlockDriverState *bs = NULL;
+
+	local = opaque;
+	inode = local->inode;
+
+	qb_inode = qb_inode_ctx_get (THIS, inode);
+
+	if (!--qb_inode->refcnt) {
+		bs = qb_inode->bs;
+		qb_inode->bs = NULL;
+		bdrv_delete (bs);
+	}
+
+	frame = local->frame;
+	frame->local = NULL;
+	qb_local_free (THIS, local);
+	STACK_DESTROY (frame->root);
+
+	return 0;
+}
+
+
+int
+qb_snapshot_create (void *opaque)
+{
+	qb_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	call_stub_t *stub = NULL;
+	inode_t *inode = NULL;
+	qb_inode_t *qb_inode = NULL;
+	QEMUSnapshotInfo sn;
+	struct timeval tv = {0, };
+	int ret = 0;
+
+	local = opaque;
+	frame = local->frame;
+	stub = local->stub;
+	inode = local->inode;
+
+	qb_inode = qb_inode_ctx_get (frame->this, inode);
+	if (!qb_inode->bs) {
+		/* FIXME: we need locks around this when
+		   enabling multithreaded syncop/coroutine
+		   for qemu-block
+		*/
+
+		qb_inode->bs = qb_bs_create (inode, qb_inode->fmt);
+		if (!qb_inode->bs) {
+			QB_STUB_UNWIND (stub, -1, errno);
+			return 0;
+		}
+	}
+
+	memset (&sn, 0, sizeof (sn));
+        pstrcpy (sn.name, sizeof(sn.name), local->name);
+        gettimeofday (&tv, NULL);
+        sn.date_sec = tv.tv_sec;
+        sn.date_nsec = tv.tv_usec * 1000;
+
+        ret = bdrv_snapshot_create (qb_inode->bs, &sn);
+	if (ret < 0) {
+		QB_STUB_UNWIND (stub, -1, -ret);
+	} else {
+		QB_STUB_UNWIND (stub, ret, 0);
+	}
+
+	return 0;
+}
+
+
+int
+qb_snapshot_delete (void *opaque)
+{
+	qb_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	call_stub_t *stub = NULL;
+	inode_t *inode = NULL;
+	qb_inode_t *qb_inode = NULL;
+	int ret = 0;
+
+	local = opaque;
+	frame = local->frame;
+	stub = local->stub;
+	inode = local->inode;
+
+	qb_inode = qb_inode_ctx_get (frame->this, inode);
+	if (!qb_inode->bs) {
+		/* FIXME: we need locks around this when
+		   enabling multithreaded syncop/coroutine
+		   for qemu-block
+		*/
+
+		qb_inode->bs = qb_bs_create (inode, qb_inode->fmt);
+		if (!qb_inode->bs) {
+			QB_STUB_UNWIND (stub, -1, errno);
+			return 0;
+		}
+	}
+
+	ret = bdrv_snapshot_delete (qb_inode->bs, local->name);
+
+	if (ret < 0) {
+		QB_STUB_UNWIND (stub, -1, -ret);
+	} else {
+		QB_STUB_UNWIND (stub, ret, 0);
+	}
+
+	return 0;
+}
+
+
+int
+qb_snapshot_goto (void *opaque)
+{
+	qb_local_t *local = NULL;
+	call_frame_t *frame = NULL;
+	call_stub_t *stub = NULL;
+	inode_t *inode = NULL;
+	qb_inode_t *qb_inode = NULL;
+	int ret = 0;
+
+	local = opaque;
+	frame = local->frame;
+	stub = local->stub;
+	inode = local->inode;
+
+	qb_inode = qb_inode_ctx_get (frame->this, inode);
+	if (!qb_inode->bs) {
+		/* FIXME: we need locks around this when
+		   enabling multithreaded syncop/coroutine
+		   for qemu-block
+		*/
+
+		qb_inode->bs = qb_bs_create (inode, qb_inode->fmt);
+		if (!qb_inode->bs) {
+			QB_STUB_UNWIND (stub, -1, errno);
+			return 0;
+		}
+	}
+
+	ret = bdrv_snapshot_goto (qb_inode->bs, local->name);
+
+	if (ret < 0) {
+		QB_STUB_UNWIND (stub, -1, -ret);
+	} else {
+		QB_STUB_UNWIND (stub, ret, 0);
+	}
+
+	return 0;
+}
diff --git a/xlators/features/qemu-block/src/qb-coroutines.h b/xlators/features/qemu-block/src/qb-coroutines.h
new file mode 100644
index 000000000..583319f3b
--- /dev/null
+++ b/xlators/features/qemu-block/src/qb-coroutines.h
@@ -0,0 +1,30 @@
+/*
+  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 __QB_COROUTINES_H
+#define __QB_COROUTINES_H
+
+#include "syncop.h"
+#include "call-stub.h"
+#include "block/block_int.h"
+#include "monitor/monitor.h"
+
+int qb_format_and_resume (void *opaque);
+int qb_snapshot_create (void *opaque);
+int qb_snapshot_delete (void *opaque);
+int qb_snapshot_goto (void *opaque);
+int qb_co_open (void *opaque);
+int qb_co_close (void *opaque);
+int qb_co_writev (void *opaque);
+int qb_co_readv (void *opaque);
+int qb_co_fsync (void *opaque);
+int qb_co_truncate (void *opaque);
+
+#endif /* __QB_COROUTINES_H */
diff --git a/xlators/features/qemu-block/src/qemu-block-memory-types.h b/xlators/features/qemu-block/src/qemu-block-memory-types.h
new file mode 100644
index 000000000..267b3893f
--- /dev/null
+++ b/xlators/features/qemu-block/src/qemu-block-memory-types.h
@@ -0,0 +1,25 @@
+/*
+  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 __QB_MEM_TYPES_H__
+#define __QB_MEM_TYPES_H__
+
+#include "mem-types.h"
+
+enum gf_qb_mem_types_ {
+        gf_qb_mt_qb_conf_t   = gf_common_mt_end + 1,
+        gf_qb_mt_qb_inode_t,
+        gf_qb_mt_qb_local_t,
+	gf_qb_mt_coroutinesynctask_t,
+        gf_qb_mt_end
+};
+#endif
+
diff --git a/xlators/features/qemu-block/src/qemu-block.c b/xlators/features/qemu-block/src/qemu-block.c
new file mode 100644
index 000000000..48bbf3140
--- /dev/null
+++ b/xlators/features/qemu-block/src/qemu-block.c
@@ -0,0 +1,1140 @@
+/*
+  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 "glusterfs.h"
+#include "logging.h"
+#include "dict.h"
+#include "xlator.h"
+#include "inode.h"
+#include "call-stub.h"
+#include "defaults.h"
+#include "qemu-block-memory-types.h"
+#include "qemu-block.h"
+#include "qb-coroutines.h"
+
+
+qb_inode_t *
+__qb_inode_ctx_get (xlator_t *this, inode_t *inode)
+{
+        uint64_t    value    = 0;
+        qb_inode_t *qb_inode = NULL;
+
+        __inode_ctx_get (inode, this, &value);
+        qb_inode = (qb_inode_t *)(unsigned long) value;
+
+        return qb_inode;
+}
+
+
+qb_inode_t *
+qb_inode_ctx_get (xlator_t *this, inode_t *inode)
+{
+        qb_inode_t *qb_inode = NULL;
+
+        LOCK (&inode->lock);
+        {
+                qb_inode = __qb_inode_ctx_get (this, inode);
+        }
+        UNLOCK (&inode->lock);
+
+        return qb_inode;
+}
+
+
+qb_inode_t *
+qb_inode_ctx_del (xlator_t *this, inode_t *inode)
+{
+        uint64_t    value    = 0;
+        qb_inode_t *qb_inode = NULL;
+
+        inode_ctx_del (inode, this, &value);
+        qb_inode = (qb_inode_t *)(unsigned long) value;
+
+        return qb_inode;
+}
+
+
+int
+qb_inode_cleanup (xlator_t *this, inode_t *inode, int warn)
+{
+	qb_inode_t *qb_inode = NULL;
+
+	qb_inode = qb_inode_ctx_del (this, inode);
+
+	if (!qb_inode)
+		return 0;
+
+	if (warn)
+		gf_log (this->name, GF_LOG_WARNING,
+			"inode %s no longer block formatted",
+			uuid_utoa (inode->gfid));
+
+	/* free (qb_inode->bs); */
+
+	GF_FREE (qb_inode);
+
+	return 0;
+}
+
+
+int
+qb_iatt_fixup (xlator_t *this, inode_t *inode, struct iatt *iatt)
+{
+	qb_inode_t *qb_inode = NULL;
+
+	qb_inode = qb_inode_ctx_get (this, inode);
+	if (!qb_inode)
+		return 0;
+
+	iatt->ia_size = qb_inode->size;
+
+	return 0;
+}
+
+
+int
+qb_format_extract (xlator_t *this, char *format, inode_t *inode)
+{
+	char       *s, *save;
+	uint64_t    size = 0;
+	char        fmt[QB_XATTR_VAL_MAX+1] = {0, };
+	qb_inode_t *qb_inode = NULL;
+	char *formatstr = NULL;
+	uuid_t gfid = {0,};
+	char gfid_str[64] = {0,};
+	int ret;
+
+	strncpy(fmt, format, QB_XATTR_VAL_MAX);
+
+	s = strtok_r(fmt, ":", &save);
+	if (!s)
+		goto invalid;
+	formatstr = gf_strdup(s);
+
+	s = strtok_r(NULL, ":", &save);
+	if (!s)
+		goto invalid;
+	if (gf_string2bytesize (s, &size))
+		goto invalid;
+	if (!size)
+		goto invalid;
+
+	s = strtok_r(NULL, "\0", &save);
+	if (s && !strncmp(s, "<gfid:", strlen("<gfid:"))) {
+		/*
+		 * Check for valid gfid backing image specifier.
+		 */
+		if (strlen(s) + 1 > sizeof(gfid_str))
+			goto invalid;
+		ret = sscanf(s, "<gfid:%[^>]s", gfid_str);
+		if (ret == 1) {
+			ret = uuid_parse(gfid_str, gfid);
+			if (ret < 0)
+				goto invalid;
+		}
+	}
+
+	qb_inode = qb_inode_ctx_get (this, inode);
+	if (!qb_inode)
+		qb_inode = GF_CALLOC (1, sizeof (*qb_inode),
+				      gf_qb_mt_qb_inode_t);
+	if (!qb_inode) {
+		GF_FREE(formatstr);
+		return ENOMEM;
+	}
+
+	strncpy(qb_inode->fmt, formatstr, QB_XATTR_VAL_MAX);
+	qb_inode->size = size;
+
+	/*
+	 * If a backing gfid was not specified, interpret any remaining bytes
+	 * associated with a backing image as a filename local to the parent
+	 * directory. The format processing will validate further.
+	 */
+	if (!uuid_is_null(gfid))
+		uuid_copy(qb_inode->backing_gfid, gfid);
+	else if (s)
+		qb_inode->backing_fname = gf_strdup(s);
+
+	inode_ctx_set (inode, this, (void *)&qb_inode);
+
+	GF_FREE(formatstr);
+
+	return 0;
+
+invalid:
+	GF_FREE(formatstr);
+
+	gf_log (this->name, GF_LOG_WARNING,
+		"invalid format '%s' in inode %s", format,
+		uuid_utoa (inode->gfid));
+	return EINVAL;
+}
+
+
+void
+qb_local_free (xlator_t *this, qb_local_t *local)
+{
+	if (local->inode)
+		inode_unref (local->inode);
+	if (local->fd)
+		fd_unref (local->fd);
+	GF_FREE (local);
+}
+
+
+int
+qb_local_init (call_frame_t *frame)
+{
+	qb_local_t *qb_local = NULL;
+
+	qb_local = GF_CALLOC (1, sizeof (*qb_local), gf_qb_mt_qb_local_t);
+	if (!qb_local)
+		return -1;
+	INIT_LIST_HEAD(&qb_local->list);
+
+	qb_local->frame = frame;
+	frame->local = qb_local;
+
+	return 0;
+}
+
+
+int
+qb_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)
+{
+	char *format = NULL;
+	qb_conf_t *conf = NULL;
+
+	conf = this->private;
+
+	if (op_ret == -1)
+		goto out;
+
+	/*
+	 * Cache the root inode for dealing with backing images. The format
+	 * coroutine and the gluster qemu backend driver both use the root inode
+	 * table to verify and/or redirect I/O to the backing image via
+	 * anonymous fd's.
+	 */
+	if (!conf->root_inode && __is_root_gfid(inode->gfid))
+		conf->root_inode = inode_ref(inode);
+
+	if (!xdata)
+		goto out;
+
+	if (dict_get_str (xdata, conf->qb_xattr_key, &format))
+		goto out;
+
+	if (!format) {
+		qb_inode_cleanup (this, inode, 1);
+		goto out;
+	}
+
+	op_errno = qb_format_extract (this, format, inode);
+	if (op_errno)
+		op_ret = -1;
+
+	qb_iatt_fixup (this, inode, buf);
+out:
+	QB_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, buf,
+			 xdata, postparent);
+	return 0;
+}
+
+
+int
+qb_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+	qb_conf_t *conf = NULL;
+
+	conf = this->private;
+
+	xdata = xdata ? dict_ref (xdata) : dict_new ();
+
+	if (!xdata)
+		goto enomem;
+
+	if (dict_set_int32 (xdata, conf->qb_xattr_key, 0))
+		goto enomem;
+
+	STACK_WIND (frame, qb_lookup_cbk, FIRST_CHILD(this),
+		    FIRST_CHILD(this)->fops->lookup, loc, xdata);
+	dict_unref (xdata);
+	return 0;
+enomem:
+	QB_STACK_UNWIND (lookup, frame, -1, ENOMEM, 0, 0, 0, 0);
+	if (xdata)
+		dict_unref (xdata);
+	return 0;
+}
+
+
+int
+qb_setxattr_format (call_frame_t *frame, xlator_t *this, call_stub_t *stub,
+		    dict_t *xattr, inode_t *inode)
+{
+	char *format = NULL;
+	int op_errno = 0;
+	qb_local_t *qb_local = NULL;
+	data_t *data = NULL;
+	qb_inode_t *qb_inode;
+
+	if (!(data = dict_get (xattr, "trusted.glusterfs.block-format"))) {
+		QB_STUB_RESUME (stub);
+		return 0;
+	}
+
+	format = alloca (data->len + 1);
+	memcpy (format, data->data, data->len);
+	format[data->len] = 0;
+
+	op_errno = qb_format_extract (this, format, inode);
+	if (op_errno) {
+		QB_STUB_UNWIND (stub, -1, op_errno);
+		return 0;
+	}
+	qb_inode = qb_inode_ctx_get(this, inode);
+
+	qb_local = frame->local;
+
+	qb_local->stub = stub;
+	qb_local->inode = inode_ref (inode);
+
+	snprintf(qb_local->fmt, QB_XATTR_VAL_MAX, "%s:%lu", qb_inode->fmt,
+		 qb_inode->size);
+
+	qb_coroutine (frame, qb_format_and_resume);
+
+	return 0;
+}
+
+
+int
+qb_setxattr_snapshot_create (call_frame_t *frame, xlator_t *this,
+			     call_stub_t *stub, dict_t *xattr, inode_t *inode)
+{
+	qb_local_t *qb_local = NULL;
+	char *name = NULL;
+	data_t *data = NULL;
+
+	if (!(data = dict_get (xattr, "trusted.glusterfs.block-snapshot-create"))) {
+		QB_STUB_RESUME (stub);
+		return 0;
+	}
+
+	name = alloca (data->len + 1);
+	memcpy (name, data->data, data->len);
+	name[data->len] = 0;
+
+	qb_local = frame->local;
+
+	qb_local->stub = stub;
+	qb_local->inode = inode_ref (inode);
+	strncpy (qb_local->name, name, 128);
+
+	qb_coroutine (frame, qb_snapshot_create);
+
+	return 0;
+}
+
+
+int
+qb_setxattr_snapshot_delete (call_frame_t *frame, xlator_t *this,
+			     call_stub_t *stub, dict_t *xattr, inode_t *inode)
+{
+	qb_local_t *qb_local = NULL;
+	char *name = NULL;
+	data_t *data = NULL;
+
+	if (!(data = dict_get (xattr, "trusted.glusterfs.block-snapshot-delete"))) {
+		QB_STUB_RESUME (stub);
+		return 0;
+	}
+
+	name = alloca (data->len + 1);
+	memcpy (name, data->data, data->len);
+	name[data->len] = 0;
+
+	qb_local = frame->local;
+
+	qb_local->stub = stub;
+	qb_local->inode = inode_ref (inode);
+	strncpy (qb_local->name, name, 128);
+
+	qb_coroutine (frame, qb_snapshot_delete);
+
+	return 0;
+}
+
+int
+qb_setxattr_snapshot_goto (call_frame_t *frame, xlator_t *this,
+			   call_stub_t *stub, dict_t *xattr, inode_t *inode)
+{
+	qb_local_t *qb_local = NULL;
+	char *name = NULL;
+	data_t *data = NULL;
+
+	if (!(data = dict_get (xattr, "trusted.glusterfs.block-snapshot-goto"))) {
+		QB_STUB_RESUME (stub);
+		return 0;
+	}
+
+	name = alloca (data->len + 1);
+	memcpy (name, data->data, data->len);
+	name[data->len] = 0;
+
+	qb_local = frame->local;
+
+	qb_local->stub = stub;
+	qb_local->inode = inode_ref (inode);
+	strncpy (qb_local->name, name, 128);
+
+	qb_coroutine (frame, qb_snapshot_goto);
+
+	return 0;
+}
+
+
+int
+qb_setxattr_common (call_frame_t *frame, xlator_t *this, call_stub_t *stub,
+		    dict_t *xattr, inode_t *inode)
+{
+	data_t *data = NULL;
+
+	if ((data = dict_get (xattr, "trusted.glusterfs.block-format"))) {
+		qb_setxattr_format (frame, this, stub, xattr, inode);
+		return 0;
+	}
+
+	if ((data = dict_get (xattr, "trusted.glusterfs.block-snapshot-create"))) {
+		qb_setxattr_snapshot_create (frame, this, stub, xattr, inode);
+		return 0;
+	}
+
+	if ((data = dict_get (xattr, "trusted.glusterfs.block-snapshot-delete"))) {
+		qb_setxattr_snapshot_delete (frame, this, stub, xattr, inode);
+		return 0;
+	}
+
+	if ((data = dict_get (xattr, "trusted.glusterfs.block-snapshot-goto"))) {
+		qb_setxattr_snapshot_goto (frame, this, stub, xattr, inode);
+		return 0;
+	}
+
+	QB_STUB_RESUME (stub);
+
+	return 0;
+}
+
+
+int
+qb_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr,
+	     int flags, dict_t *xdata)
+{
+	call_stub_t *stub = NULL;
+
+	if (qb_local_init (frame) != 0)
+		goto enomem;
+
+	stub = fop_setxattr_stub (frame, default_setxattr_resume, loc, xattr,
+				  flags, xdata);
+	if (!stub)
+		goto enomem;
+
+	qb_setxattr_common (frame, this, stub, xattr, loc->inode);
+
+	return 0;
+enomem:
+	QB_STACK_UNWIND (setxattr, frame, -1, ENOMEM, 0);
+	return 0;
+}
+
+
+int
+qb_fsetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xattr,
+	      int flags, dict_t *xdata)
+{
+	call_stub_t *stub = NULL;
+
+	if (qb_local_init (frame) != 0)
+		goto enomem;
+
+	stub = fop_fsetxattr_stub (frame, default_fsetxattr_resume, fd, xattr,
+				   flags, xdata);
+	if (!stub)
+		goto enomem;
+
+	qb_setxattr_common (frame, this, stub, xattr, fd->inode);
+
+	return 0;
+enomem:
+	QB_STACK_UNWIND (fsetxattr, frame, -1, ENOMEM, 0);
+	return 0;
+}
+
+
+int
+qb_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+	     int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
+{
+	call_stub_t *stub = NULL;
+	qb_local_t *qb_local = NULL;
+
+	qb_local = frame->local;
+
+	if (op_ret < 0)
+		goto unwind;
+
+	if (!qb_inode_ctx_get (this, qb_local->inode))
+		goto unwind;
+
+	stub = fop_open_cbk_stub (frame, NULL, op_ret, op_errno, fd, xdata);
+	if (!stub) {
+		op_ret = -1;
+		op_errno = ENOMEM;
+		goto unwind;
+	}
+
+	qb_local->stub = stub;
+
+	qb_coroutine (frame, qb_co_open);
+
+	return 0;
+unwind:
+	QB_STACK_UNWIND (open, frame, op_ret, op_errno, fd, xdata);
+	return 0;
+}
+
+
+int
+qb_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+	 fd_t *fd, dict_t *xdata)
+{
+	qb_local_t *qb_local = NULL;
+	qb_inode_t *qb_inode = NULL;
+
+	qb_inode = qb_inode_ctx_get (this, loc->inode);
+	if (!qb_inode) {
+		STACK_WIND (frame, default_open_cbk, FIRST_CHILD(this),
+			    FIRST_CHILD(this)->fops->open, loc, flags, fd,
+			    xdata);
+		return 0;
+	}
+
+	if (qb_local_init (frame) != 0)
+		goto enomem;
+
+	qb_local = frame->local;
+
+	qb_local->inode = inode_ref (loc->inode);
+	qb_local->fd = fd_ref (fd);
+
+	STACK_WIND (frame, qb_open_cbk, FIRST_CHILD(this),
+		    FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+	return 0;
+enomem:
+	QB_STACK_UNWIND (open, frame, -1, ENOMEM, 0, 0);
+	return 0;
+}
+
+
+int
+qb_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+	   int count, off_t offset, uint32_t flags, struct iobref *iobref,
+	   dict_t *xdata)
+{
+	qb_local_t *qb_local = NULL;
+	qb_inode_t *qb_inode = NULL;
+
+	qb_inode = qb_inode_ctx_get (this, fd->inode);
+	if (!qb_inode) {
+		STACK_WIND (frame, default_writev_cbk, FIRST_CHILD(this),
+			    FIRST_CHILD(this)->fops->writev, fd, vector, count,
+			    offset, flags, iobref, xdata);
+		return 0;
+	}
+
+	if (qb_local_init (frame) != 0)
+		goto enomem;
+
+	qb_local = frame->local;
+
+	qb_local->inode = inode_ref (fd->inode);
+	qb_local->fd = fd_ref (fd);
+
+	qb_local->stub = fop_writev_stub (frame, NULL, fd, vector, count,
+					  offset, flags, iobref, xdata);
+	if (!qb_local->stub)
+		goto enomem;
+
+	qb_coroutine (frame, qb_co_writev);
+
+	return 0;
+enomem:
+	QB_STACK_UNWIND (writev, frame, -1, ENOMEM, 0, 0, 0);
+	return 0;
+}
+
+
+int
+qb_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+	  off_t offset, uint32_t flags, dict_t *xdata)
+{
+	qb_local_t *qb_local = NULL;
+	qb_inode_t *qb_inode = NULL;
+
+	qb_inode = qb_inode_ctx_get (this, fd->inode);
+	if (!qb_inode) {
+		STACK_WIND (frame, default_readv_cbk, FIRST_CHILD(this),
+			    FIRST_CHILD(this)->fops->readv, fd, size, offset,
+			    flags, xdata);
+		return 0;
+	}
+
+	if (qb_local_init (frame) != 0)
+		goto enomem;
+
+	qb_local = frame->local;
+
+	qb_local->inode = inode_ref (fd->inode);
+	qb_local->fd = fd_ref (fd);
+
+	qb_local->stub = fop_readv_stub (frame, NULL, fd, size, offset,
+					  flags, xdata);
+	if (!qb_local->stub)
+		goto enomem;
+
+	qb_coroutine (frame, qb_co_readv);
+
+	return 0;
+enomem:
+	QB_STACK_UNWIND (readv, frame, -1, ENOMEM, 0, 0, 0, 0, 0);
+	return 0;
+}
+
+
+int
+qb_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int dsync,
+	  dict_t *xdata)
+{
+	qb_local_t *qb_local = NULL;
+	qb_inode_t *qb_inode = NULL;
+
+	qb_inode = qb_inode_ctx_get (this, fd->inode);
+	if (!qb_inode) {
+		STACK_WIND (frame, default_fsync_cbk, FIRST_CHILD(this),
+			    FIRST_CHILD(this)->fops->fsync, fd, dsync, xdata);
+		return 0;
+	}
+
+	if (qb_local_init (frame) != 0)
+		goto enomem;
+
+	qb_local = frame->local;
+
+	qb_local->inode = inode_ref (fd->inode);
+	qb_local->fd = fd_ref (fd);
+
+	qb_local->stub = fop_fsync_stub (frame, NULL, fd, dsync, xdata);
+
+	if (!qb_local->stub)
+		goto enomem;
+
+	qb_coroutine (frame, qb_co_fsync);
+
+	return 0;
+enomem:
+	QB_STACK_UNWIND (fsync, frame, -1, ENOMEM, 0, 0, 0);
+	return 0;
+}
+
+
+int
+qb_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+	qb_local_t *qb_local = NULL;
+	qb_inode_t *qb_inode = NULL;
+
+	qb_inode = qb_inode_ctx_get (this, fd->inode);
+	if (!qb_inode) {
+		STACK_WIND (frame, default_flush_cbk, FIRST_CHILD(this),
+			    FIRST_CHILD(this)->fops->flush, fd, xdata);
+		return 0;
+	}
+
+	if (qb_local_init (frame) != 0)
+		goto enomem;
+
+	qb_local = frame->local;
+
+	qb_local->inode = inode_ref (fd->inode);
+	qb_local->fd = fd_ref (fd);
+
+	qb_local->stub = fop_flush_stub (frame, NULL, fd, xdata);
+
+	if (!qb_local->stub)
+		goto enomem;
+
+	qb_coroutine (frame, qb_co_fsync);
+
+	return 0;
+enomem:
+	QB_STACK_UNWIND (flush, frame, -1, ENOMEM, 0);
+	return 0;
+}
+
+static int32_t
+qb_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+		int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+		dict_t *xdata)
+{
+	qb_conf_t *conf = this->private;
+	gf_dirent_t *entry;
+	char *format;
+
+	list_for_each_entry(entry, &entries->list, list) {
+		if (!entry->inode || !entry->dict)
+			continue;
+
+		format = NULL;
+		if (dict_get_str(entry->dict, conf->qb_xattr_key, &format))
+			continue;
+
+		if (!format) {
+			qb_inode_cleanup(this, entry->inode, 1);
+			continue;
+		}
+
+		if (qb_format_extract(this, format, entry->inode))
+			continue;
+
+		qb_iatt_fixup(this, entry->inode, &entry->d_stat);
+	}
+
+	STACK_UNWIND_STRICT(readdirp, frame, op_ret, op_errno, entries, xdata);
+	return 0;
+}
+
+static int32_t
+qb_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+	    off_t off, dict_t *xdata)
+{
+	qb_conf_t *conf = this->private;
+
+	xdata = xdata ? dict_ref(xdata) : dict_new();
+	if (!xdata)
+		goto enomem;
+
+	if (dict_set_int32 (xdata, conf->qb_xattr_key, 0))
+		goto enomem;
+
+	STACK_WIND(frame, qb_readdirp_cbk, FIRST_CHILD(this),
+		   FIRST_CHILD(this)->fops->readdirp, fd, size, off, xdata);
+
+	dict_unref(xdata);
+	return 0;
+
+enomem:
+	QB_STACK_UNWIND(readdirp, frame, -1, ENOMEM, NULL, NULL);
+	if (xdata)
+		dict_unref(xdata);
+	return 0;
+}
+
+int
+qb_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+	     dict_t *xdata)
+{
+	qb_local_t *qb_local = NULL;
+	qb_inode_t *qb_inode = NULL;
+
+	qb_inode = qb_inode_ctx_get (this, loc->inode);
+	if (!qb_inode) {
+		STACK_WIND (frame, default_truncate_cbk, FIRST_CHILD(this),
+			    FIRST_CHILD(this)->fops->truncate, loc, offset,
+			    xdata);
+		return 0;
+	}
+
+	if (qb_local_init (frame) != 0)
+		goto enomem;
+
+	qb_local = frame->local;
+
+	qb_local->inode = inode_ref (loc->inode);
+	qb_local->fd = fd_anonymous (loc->inode);
+
+	qb_local->stub = fop_truncate_stub (frame, NULL, loc, offset, xdata);
+
+	if (!qb_local->stub)
+		goto enomem;
+
+	qb_coroutine (frame, qb_co_truncate);
+
+	return 0;
+enomem:
+	QB_STACK_UNWIND (truncate, frame, -1, ENOMEM, 0, 0, 0);
+	return 0;
+}
+
+
+int
+qb_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+	     dict_t *xdata)
+{
+	qb_local_t *qb_local = NULL;
+	qb_inode_t *qb_inode = NULL;
+
+	qb_inode = qb_inode_ctx_get (this, fd->inode);
+	if (!qb_inode) {
+		STACK_WIND (frame, default_ftruncate_cbk, FIRST_CHILD(this),
+			    FIRST_CHILD(this)->fops->ftruncate, fd, offset,
+			    xdata);
+		return 0;
+	}
+
+	if (qb_local_init (frame) != 0)
+		goto enomem;
+
+	qb_local = frame->local;
+
+	qb_local->inode = inode_ref (fd->inode);
+	qb_local->fd = fd_ref (fd);
+
+	qb_local->stub = fop_ftruncate_stub (frame, NULL, fd, offset, xdata);
+
+	if (!qb_local->stub)
+		goto enomem;
+
+	qb_coroutine (frame, qb_co_truncate);
+
+	return 0;
+enomem:
+	QB_STACK_UNWIND (ftruncate, frame, -1, ENOMEM, 0, 0, 0);
+	return 0;
+}
+
+
+int
+qb_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+	     int op_ret, int op_errno, struct iatt *iatt, dict_t *xdata)
+{
+	inode_t *inode = NULL;
+
+	inode = frame->local;
+	frame->local = NULL;
+
+	if (inode) {
+		qb_iatt_fixup (this, inode, iatt);
+		inode_unref (inode);
+	}
+
+	QB_STACK_UNWIND (stat, frame, op_ret, op_errno, iatt, xdata);
+
+	return 0;
+}
+
+int
+qb_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+	if (qb_inode_ctx_get (this, loc->inode))
+		frame->local = inode_ref (loc->inode);
+
+	STACK_WIND (frame, qb_stat_cbk, FIRST_CHILD(this),
+		    FIRST_CHILD(this)->fops->stat, loc, xdata);
+	return 0;
+}
+
+
+int
+qb_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+	      int op_ret, int op_errno, struct iatt *iatt, dict_t *xdata)
+{
+	inode_t *inode = NULL;
+
+	inode = frame->local;
+	frame->local = NULL;
+
+	if (inode) {
+		qb_iatt_fixup (this, inode, iatt);
+		inode_unref (inode);
+	}
+
+	QB_STACK_UNWIND (fstat, frame, op_ret, op_errno, iatt, xdata);
+
+	return 0;
+}
+
+
+int
+qb_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+	if (qb_inode_ctx_get (this, fd->inode))
+		frame->local = inode_ref (fd->inode);
+
+	STACK_WIND (frame, qb_fstat_cbk, FIRST_CHILD(this),
+		    FIRST_CHILD(this)->fops->fstat, fd, xdata);
+	return 0;
+}
+
+
+int
+qb_setattr_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)
+{
+	inode_t *inode = NULL;
+
+	inode = frame->local;
+	frame->local = NULL;
+
+	if (inode) {
+		qb_iatt_fixup (this, inode, pre);
+		qb_iatt_fixup (this, inode, post);
+		inode_unref (inode);
+	}
+
+	QB_STACK_UNWIND (setattr, frame, op_ret, op_errno, pre, post, xdata);
+
+	return 0;
+}
+
+
+int
+qb_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *buf,
+	    int valid, dict_t *xdata)
+{
+	if (qb_inode_ctx_get (this, loc->inode))
+		frame->local = inode_ref (loc->inode);
+
+	STACK_WIND (frame, qb_setattr_cbk, FIRST_CHILD(this),
+		    FIRST_CHILD(this)->fops->setattr, loc, buf, valid, xdata);
+	return 0;
+}
+
+
+int
+qb_fsetattr_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)
+{
+	inode_t *inode = NULL;
+
+	inode = frame->local;
+	frame->local = NULL;
+
+	if (inode) {
+		qb_iatt_fixup (this, inode, pre);
+		qb_iatt_fixup (this, inode, post);
+		inode_unref (inode);
+	}
+
+	QB_STACK_UNWIND (fsetattr, frame, op_ret, op_errno, pre, post, xdata);
+
+	return 0;
+}
+
+
+int
+qb_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *buf,
+	     int valid, dict_t *xdata)
+{
+	if (qb_inode_ctx_get (this, fd->inode))
+		frame->local = inode_ref (fd->inode);
+
+	STACK_WIND (frame, qb_setattr_cbk, FIRST_CHILD(this),
+		    FIRST_CHILD(this)->fops->fsetattr, fd, buf, valid, xdata);
+	return 0;
+}
+
+
+int
+qb_forget (xlator_t *this, inode_t *inode)
+{
+	return qb_inode_cleanup (this, inode, 0);
+}
+
+
+int
+qb_release (xlator_t *this, fd_t *fd)
+{
+	call_frame_t *frame = NULL;
+
+	frame = create_frame (this, this->ctx->pool);
+	if (!frame) {
+		gf_log (this->name, GF_LOG_ERROR,
+			"Could not allocate frame. "
+			"Leaking QEMU BlockDriverState");
+		return -1;
+	}
+
+	if (qb_local_init (frame) != 0) {
+		gf_log (this->name, GF_LOG_ERROR,
+			"Could not allocate local. "
+			"Leaking QEMU BlockDriverState");
+		STACK_DESTROY (frame->root);
+		return -1;
+	}
+
+	if (qb_coroutine (frame, qb_co_close) != 0) {
+		gf_log (this->name, GF_LOG_ERROR,
+			"Could not allocate coroutine. "
+			"Leaking QEMU BlockDriverState");
+		qb_local_free (this, frame->local);
+		frame->local = NULL;
+		STACK_DESTROY (frame->root);
+	}
+
+	return 0;
+}
+
+int
+mem_acct_init (xlator_t *this)
+{
+        int     ret = -1;
+
+        ret = xlator_mem_acct_init (this, gf_qb_mt_end + 1);
+
+        if (ret)
+                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init "
+                        "failed");
+        return ret;
+}
+
+
+int
+reconfigure (xlator_t *this, dict_t *options)
+{
+	return 0;
+}
+
+
+int
+init (xlator_t *this)
+{
+        qb_conf_t *conf    = NULL;
+        int32_t    ret     = -1;
+	static int bdrv_inited = 0;
+
+        if (!this->children || this->children->next) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "FATAL: qemu-block (%s) not configured with exactly "
+                        "one child", this->name);
+                goto out;
+        }
+
+        conf = GF_CALLOC (1, sizeof (*conf), gf_qb_mt_qb_conf_t);
+        if (!conf)
+                goto out;
+
+        /* configure 'option window-size <size>' */
+        GF_OPTION_INIT ("default-password", conf->default_password, str, out);
+
+	/* qemu coroutines use "co_mutex" for synchronizing among themselves.
+	   However "co_mutex" itself is not threadsafe if the coroutine framework
+	   is multithreaded (which usually is not). However synctasks are
+	   fundamentally multithreaded, so for now create a syncenv which has
+	   scaling limits set to max 1 thread so that the qemu coroutines can
+	   execute "safely".
+
+	   Future work: provide an implementation of "co_mutex" which is
+	   threadsafe and use the global multithreaded ctx->env syncenv.
+	*/
+	conf->env = syncenv_new (0, 1, 1);
+
+        this->private = conf;
+
+        ret = 0;
+
+	snprintf (conf->qb_xattr_key, QB_XATTR_KEY_MAX, QB_XATTR_KEY_FMT,
+		  this->name);
+
+	cur_mon = (void *) 1;
+
+	if (!bdrv_inited) {
+		bdrv_init ();
+		bdrv_inited = 1;
+	}
+
+out:
+        if (ret)
+                GF_FREE (conf);
+
+        return ret;
+}
+
+
+void
+fini (xlator_t *this)
+{
+        qb_conf_t *conf = NULL;
+
+        conf = this->private;
+
+        this->private = NULL;
+
+	if (conf->root_inode)
+		inode_unref(conf->root_inode);
+        GF_FREE (conf);
+
+	return;
+}
+
+
+struct xlator_fops fops = {
+	.lookup      = qb_lookup,
+	.fsetxattr   = qb_fsetxattr,
+	.setxattr    = qb_setxattr,
+	.open        = qb_open,
+        .writev      = qb_writev,
+        .readv       = qb_readv,
+        .fsync       = qb_fsync,
+        .truncate    = qb_truncate,
+        .ftruncate   = qb_ftruncate,
+        .stat        = qb_stat,
+        .fstat       = qb_fstat,
+	.setattr     = qb_setattr,
+	.fsetattr    = qb_fsetattr,
+        .flush       = qb_flush,
+/*
+	.getxattr    = qb_getxattr,
+	.fgetxattr   = qb_fgetxattr
+*/
+	.readdirp    = qb_readdirp,
+};
+
+
+struct xlator_cbks cbks = {
+        .forget   = qb_forget,
+	.release  = qb_release,
+};
+
+
+struct xlator_dumpops dumpops = {
+};
+
+
+struct volume_options options[] = {
+        { .key  = {"default-password"},
+          .type = GF_OPTION_TYPE_STR,
+          .default_value = "",
+          .description = "Default password for the AES encrypted block images."
+        },
+        { .key = {NULL} },
+};
diff --git a/xlators/features/qemu-block/src/qemu-block.h b/xlators/features/qemu-block/src/qemu-block.h
new file mode 100644
index 000000000..c95f2799a
--- /dev/null
+++ b/xlators/features/qemu-block/src/qemu-block.h
@@ -0,0 +1,109 @@
+/*
+  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 __QEMU_BLOCK_H
+#define __QEMU_BLOCK_H
+
+#include "syncop.h"
+#include "call-stub.h"
+#include "block/block_int.h"
+#include "monitor/monitor.h"
+
+/* QB_XATTR_KEY_FMT is the on-disk xattr stored in the inode which
+   indicates that the file must be "interpreted" by the block format
+   logic. The value of the key is of the pattern:
+
+   "format:virtual_size"
+
+   e.g
+
+   "qcow2:20GB" or "qed:100GB"
+
+   The format and virtual size are colon separated. The format is
+   a case sensitive string which qemu recognizes. virtual_size is
+   specified as a size which glusterfs recognizes as size (i.e.,
+   value accepted by gf_string2bytesize())
+*/
+#define QB_XATTR_KEY_FMT "trusted.glusterfs.%s.format"
+
+#define QB_XATTR_KEY_MAX 64
+
+#define QB_XATTR_VAL_MAX 64
+
+
+typedef struct qb_inode {
+	char fmt[QB_XATTR_VAL_MAX]; /* this is only the format, not "format:size" */
+	size_t size; /* virtual size in bytes */
+	BlockDriverState *bs;
+	int refcnt;
+	uuid_t backing_gfid;
+	char *backing_fname;
+} qb_inode_t;
+
+
+typedef struct qb_conf {
+	Monitor *mon;
+	struct syncenv *env;
+	char qb_xattr_key[QB_XATTR_KEY_MAX];
+	char *default_password;
+	inode_t *root_inode;
+} qb_conf_t;
+
+
+typedef struct qb_local {
+	call_frame_t      *frame; /* backpointer */
+	call_stub_t       *stub;
+	inode_t           *inode;
+	fd_t              *fd;
+	char               fmt[QB_XATTR_VAL_MAX+1];
+	char               name[256];
+	synctask_fn_t      synctask_fn;
+	struct list_head   list;
+} qb_local_t;
+
+void qb_local_free (xlator_t *this, qb_local_t *local);
+int qb_coroutine (call_frame_t *frame, synctask_fn_t fn);
+inode_t *qb_inode_from_filename (const char *filename);
+int qb_inode_to_filename (inode_t *inode, char *filename, int size);
+int qb_format_extract (xlator_t *this, char *format, inode_t *inode);
+
+qb_inode_t *qb_inode_ctx_get (xlator_t *this, inode_t *inode);
+
+#define QB_STACK_UNWIND(typ, frame, args ...) do {	\
+	qb_local_t *__local = frame->local;		\
+	xlator_t *__this = frame->this;			\
+							\
+	frame->local = NULL;				\
+	STACK_UNWIND_STRICT (typ, frame, args);		\
+	if (__local)					\
+		qb_local_free (__this, __local);	\
+	} while (0)
+
+#define QB_STUB_UNWIND(stub, op_ret, op_errno) do {	\
+	qb_local_t *__local = stub->frame->local;	\
+	xlator_t *__this = stub->frame->this;		\
+							\
+	stub->frame->local = NULL;			\
+	call_unwind_error (stub, op_ret, op_errno);	\
+	if (__local)					\
+		qb_local_free (__this, __local);	\
+	} while (0)
+
+#define QB_STUB_RESUME(stub_errno) do {			\
+	qb_local_t *__local = stub->frame->local;	\
+	xlator_t *__this = stub->frame->this;		\
+							\
+	stub->frame->local = NULL;			\
+	call_resume (stub);				\
+	if (__local)					\
+		qb_local_free (__this, __local);	\
+	} while (0)
+
+#endif /* !__QEMU_BLOCK_H */
diff --git a/xlators/features/quiesce/src/quiesce.c b/xlators/features/quiesce/src/quiesce.c
index 73eb91947..24c7dc6ed 100644
--- a/xlators/features/quiesce/src/quiesce.c
+++ b/xlators/features/quiesce/src/quiesce.c
@@ -111,7 +111,7 @@ void
 gf_quiesce_enqueue (xlator_t *this, call_stub_t *stub)
 {
         quiesce_priv_t *priv    = NULL;
-        struct timeval  timeout = {0,};
+        struct timespec timeout = {0,};
 
         priv = this->private;
         if (!priv) {
@@ -129,7 +129,7 @@ gf_quiesce_enqueue (xlator_t *this, call_stub_t *stub)
 
         if (!priv->timer) {
                 timeout.tv_sec = 20;
-                timeout.tv_usec = 0;
+                timeout.tv_nsec = 0;
 
                 priv->timer = gf_timer_call_after (this->ctx,
                                                    timeout,
@@ -2492,7 +2492,7 @@ notify (xlator_t *this, int event, void *data, ...)
 {
         int             ret     = 0;
         quiesce_priv_t *priv    = NULL;
-        struct timeval  timeout = {0,};
+        struct timespec timeout = {0,};
 
         priv = this->private;
         if (!priv)
@@ -2525,7 +2525,7 @@ notify (xlator_t *this, int event, void *data, ...)
                 if (priv->timer)
                         break;
                 timeout.tv_sec = 20;
-                timeout.tv_usec = 0;
+                timeout.tv_nsec = 0;
 
                 priv->timer = gf_timer_call_after (this->ctx,
                                                    timeout,
diff --git a/xlators/features/quota/src/quota.c b/xlators/features/quota/src/quota.c
index 4ea54cca8..c527e7ca7 100644
--- a/xlators/features/quota/src/quota.c
+++ b/xlators/features/quota/src/quota.c
@@ -2969,6 +2969,177 @@ err:
         return 0;
 }
 
+int32_t
+quota_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                  ret            = 0;
+        uint64_t                 ctx_int        = 0;
+        quota_inode_ctx_t       *ctx            = NULL;
+        quota_local_t           *local          = NULL;
+        quota_dentry_t          *dentry         = NULL;
+        int64_t                  delta          = 0;
+
+        local = frame->local;
+
+        if ((op_ret < 0) || (local == NULL)) {
+                goto out;
+        }
+
+        ret = inode_ctx_get (local->loc.inode, this, &ctx_int);
+        if (ret) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to get the context", local->loc.path);
+                goto out;
+        }
+
+        ctx = (quota_inode_ctx_t *)(unsigned long) ctx_int;
+
+        if (ctx == NULL) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "quota context not set in %s (gfid:%s)",
+                        local->loc.path, uuid_utoa (local->loc.inode->gfid));
+                goto out;
+        }
+
+        LOCK (&ctx->lock);
+        {
+                ctx->buf = *postbuf;
+        }
+        UNLOCK (&ctx->lock);
+
+        list_for_each_entry (dentry, &ctx->parents, next) {
+                delta = (postbuf->ia_blocks - prebuf->ia_blocks) * 512;
+                quota_update_size (this, local->loc.inode,
+                                   dentry->name, dentry->par, delta);
+        }
+
+out:
+        QUOTA_STACK_UNWIND (fallocate, frame, op_ret, op_errno, prebuf, postbuf,
+                            xdata);
+
+        return 0;
+}
+
+int32_t
+quota_fallocate_helper(call_frame_t *frame, xlator_t *this, fd_t *fd,
+		       int32_t mode, off_t offset, size_t len, dict_t *xdata)
+{
+        quota_local_t *local    = NULL;
+        int32_t        op_errno = EINVAL;
+
+        local = frame->local;
+        if (local == NULL) {
+                gf_log (this->name, GF_LOG_WARNING, "local is NULL");
+                goto unwind;
+        }
+
+        if (local->op_ret == -1) {
+                op_errno = local->op_errno;
+                goto unwind;
+        }
+
+        STACK_WIND (frame, quota_fallocate_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+		    xdata);
+        return 0;
+
+unwind:
+        QUOTA_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
+int32_t
+quota_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            ret     = -1, op_errno = EINVAL;
+        int32_t            parents = 0;
+        quota_local_t     *local   = NULL;
+        quota_inode_ctx_t *ctx     = NULL;
+        quota_priv_t      *priv    = NULL;
+        call_stub_t       *stub    = NULL;
+        quota_dentry_t    *dentry  = NULL;
+
+        GF_ASSERT (frame);
+        GF_VALIDATE_OR_GOTO ("quota", this, unwind);
+        GF_VALIDATE_OR_GOTO (this->name, fd, unwind);
+
+        local = quota_local_new ();
+        if (local == NULL) {
+                goto unwind;
+        }
+
+        frame->local = local;
+        local->loc.inode = inode_ref (fd->inode);
+
+        ret = quota_inode_ctx_get (fd->inode, -1, this, NULL, NULL, &ctx, 0);
+        if (ctx == NULL) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "quota context not set in inode (gfid:%s)",
+                        uuid_utoa (fd->inode->gfid));
+                goto unwind;
+        }
+
+        stub = fop_fallocate_stub(frame, quota_fallocate_helper, fd, mode, offset, len,
+				  xdata); 
+        if (stub == NULL) {
+                op_errno = ENOMEM;
+                goto unwind;
+        }
+
+        priv = this->private;
+        GF_VALIDATE_OR_GOTO (this->name, priv, unwind);
+
+        LOCK (&ctx->lock);
+        {
+                list_for_each_entry (dentry, &ctx->parents, next) {
+                        parents++;
+                }
+        }
+        UNLOCK (&ctx->lock);
+
+	/*
+	 * Note that by using len as the delta we're assuming the range from
+	 * offset to offset+len has not already been allocated. This can result
+	 * in ENOSPC errors attempting to allocate an already allocated range.
+	 */
+        local->delta = len;
+        local->stub = stub;
+        local->link_count = parents;
+
+        list_for_each_entry (dentry, &ctx->parents, next) {
+                ret = quota_check_limit (frame, fd->inode, this, dentry->name,
+                                         dentry->par);
+                if (ret == -1) {
+                        break;
+                }
+        }
+
+        stub = NULL;
+
+        LOCK (&local->lock);
+        {
+                local->link_count = 0;
+                if (local->validate_count == 0) {
+                        stub = local->stub;
+                        local->stub = NULL;
+                }
+        }
+        UNLOCK (&local->lock);
+
+        if (stub != NULL) {
+                call_resume (stub);
+        }
+
+        return 0;
+
+unwind:
+        QUOTA_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
 
 int32_t
 mem_acct_init (xlator_t *this)
@@ -3304,6 +3475,7 @@ struct xlator_fops fops = {
         .removexattr  = quota_removexattr,
         .fremovexattr = quota_fremovexattr,
         .readdirp     = quota_readdirp,
+	.fallocate    = quota_fallocate,
 };
 
 struct xlator_cbks cbks = {
diff --git a/xlators/lib/src/libxlator.c b/xlators/lib/src/libxlator.c
index 624a929d0..9e5357255 100644
--- a/xlators/lib/src/libxlator.c
+++ b/xlators/lib/src/libxlator.c
@@ -11,6 +11,34 @@
 #include "libxlator.h"
 
 
+int marker_xtime_default_gauge[] = {
+        [MCNT_FOUND]    =  1,
+        [MCNT_NOTFOUND] = -1,
+        [MCNT_ENODATA]  = -1,
+        [MCNT_ENOTCONN] = -1,
+        [MCNT_ENOENT]   = -1,
+        [MCNT_EOTHER]   = -1,
+};
+
+int marker_uuid_default_gauge[] = {
+        [MCNT_FOUND]    =  1,
+        [MCNT_NOTFOUND] =  0,
+        [MCNT_ENODATA]  =  0,
+        [MCNT_ENOTCONN] =  0,
+        [MCNT_ENOENT]   =  0,
+        [MCNT_EOTHER]   =  0,
+};
+
+static int marker_idx_errno_map[] = {
+        [MCNT_FOUND]    = EINVAL,
+        [MCNT_NOTFOUND] = EINVAL,
+        [MCNT_ENOENT]   = ENOENT,
+        [MCNT_ENOTCONN] = ENOTCONN,
+        [MCNT_ENODATA]  = ENODATA,
+        [MCNT_EOTHER]   = EINVAL,
+        [MCNT_MAX]      = 0,
+};
+
 /*Copy the contents of oldtimebuf to newtimbuf*/
 static void
 update_timebuf (uint32_t *oldtimbuf, uint32_t *newtimebuf)
@@ -47,22 +75,61 @@ match_uuid_local (const char *name, char *uuid)
 static void
 marker_local_incr_errcount (xl_marker_local_t *local, int op_errno)
 {
+        marker_result_idx_t i = -1;
+
         if (!local)
                 return;
 
         switch (op_errno) {
                 case ENODATA:
-                        local->enodata_count++;
+                        i = MCNT_ENODATA;
                         break;
                 case ENOENT:
-                        local->enoent_count++;
+                        i = MCNT_ENOENT;
                         break;
                 case ENOTCONN:
-                        local->enotconn_count++;
+                        i = MCNT_ENOTCONN;
                         break;
                 default:
+                        i = MCNT_EOTHER;
                         break;
         }
+
+        local->count[i]++;
+}
+
+static int
+evaluate_marker_results (int *gauge, int *count)
+{
+        int i = 0;
+        int op_errno = 0;
+        gf_boolean_t sane = _gf_true;
+
+        /* check if the policy of the gauge is violated;
+         * if yes, try to get the best errno, ie. look
+         * for the first position where there is a more
+         * specific kind of vioilation than the generic EINVAL
+         */
+        for (i = 0; i < MCNT_MAX; i++) {
+                if (sane) {
+                        if ((gauge[i] > 0 && count[i] < gauge[i]) ||
+                            (gauge[i] < 0 && count[i] >= -gauge[i])) {
+                                sane = _gf_false;
+                                /* generic action: adopt corresponding errno */
+                                op_errno = marker_idx_errno_map[i];
+                        }
+                } else {
+                        /* already insane; trying to get a more informative
+                         * errno by checking subsequent counters
+                         */
+                        if (count[i] > 0)
+                                op_errno = marker_idx_errno_map[i];
+                }
+                if (op_errno && op_errno != EINVAL)
+                        break;
+        }
+
+        return op_errno;
 }
 
 /* Aggregate all the <volid>.xtime attrs of the cluster and send the max*/
@@ -98,14 +165,10 @@ cluster_markerxtime_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         {
                 callcnt = --local->call_count;
 
-                if (local->esomerr)
-                        goto unlock;
-
                 vol_uuid = local->vol_uuid;
 
                 if (op_ret) {
                         marker_local_incr_errcount (local, op_errno);
-                        local->esomerr = op_errno;
                         goto unlock;
                 }
 
@@ -119,11 +182,11 @@ cluster_markerxtime_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 if (dict_get_ptr (dict, marker_xattr, (void **)&net_timebuf)) {
                         gf_log (this->name, GF_LOG_WARNING,
                                 "Unable to get <uuid>.xtime attr");
-                        local->noxtime_count++;
+                        local->count[MCNT_NOTFOUND]++;
                         goto unlock;
                 }
 
-                if (local->has_xtime) {
+                if (local->count[MCNT_FOUND]) {
                         get_hosttime (net_timebuf, host_timebuf);
                         if ( (host_timebuf[0]>local->host_timebuf[0]) ||
                                 (host_timebuf[0] == local->host_timebuf[0] &&
@@ -135,7 +198,7 @@ cluster_markerxtime_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 } else {
                         get_hosttime (net_timebuf, local->host_timebuf);
                         update_timebuf (net_timebuf, local->net_timebuf);
-                        local->has_xtime = _gf_true;
+                        local->count[MCNT_FOUND]++;
                 }
 
         }
@@ -147,7 +210,7 @@ unlock:
                 op_errno = 0;
                 need_unwind = 1;
 
-                if (local->has_xtime) {
+                if (local->count[MCNT_FOUND]) {
                         if (!dict)
                                 dict = dict_new();
 
@@ -160,17 +223,9 @@ unlock:
                         }
                 }
 
-                if (local->noxtime_count)
-                        goto out;
-
-                if (local->enodata_count || local->enotconn_count ||
-                                              local->enoent_count) {
+                op_errno = evaluate_marker_results (local->gauge, local->count);
+                if (op_errno)
                         op_ret = -1;
-                        op_errno = local->enodata_count? ENODATA:
-                                        local->enotconn_count? ENOTCONN:
-                                        local->enoent_count? ENOENT:
-                                        local->esomerr;
-                }
         }
 
 out:
@@ -228,7 +283,7 @@ cluster_markeruuid_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 if (ret)
                         goto unlock;
 
-                if (marker_has_volinfo (local)) {
+                if (local->count[MCNT_FOUND]) {
                         if ((local->volmark->major != volmark->major) ||
                             (local->volmark->minor != volmark->minor)) {
                                 op_ret = -1;
@@ -257,6 +312,7 @@ cluster_markeruuid_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         uuid_unparse (volmark->uuid, vol_uuid);
                         if (volmark->retval)
                                 local->retval = volmark->retval;
+                        local->count[MCNT_FOUND]++;
                 }
         }
 unlock:
@@ -267,7 +323,7 @@ unlock:
                 op_errno = 0;
                 need_unwind = 1;
 
-                if (marker_has_volinfo (local)) {
+                if (local->count[MCNT_FOUND]) {
                         if (!dict)
                                 dict = dict_new();
 
@@ -277,11 +333,10 @@ unlock:
                                 op_ret = -1;
                                 op_errno = ENOMEM;
                         }
-                } else {
-                        op_ret = -1;
-                        op_errno = local->enotconn_count? ENOTCONN:
-                               local->enoent_count? ENOENT:EINVAL;
                 }
+                op_errno = evaluate_marker_results (local->gauge, local->count);
+                if (op_errno)
+                        op_ret = -1;
         }
 
  out:
@@ -303,7 +358,7 @@ cluster_getmarkerattr (call_frame_t *frame,xlator_t *this, loc_t *loc,
                        const char *name, void *xl_local,
                        xlator_specf_unwind_t xl_specf_getxattr_unwind,
                        xlator_t **sub_volumes, int count, int type,
-                       char *vol_uuid)
+                       int *gauge, char *vol_uuid)
 {
         int                i     = 0;
         xl_marker_local_t  *local = NULL;
@@ -326,6 +381,7 @@ cluster_getmarkerattr (call_frame_t *frame,xlator_t *this, loc_t *loc,
         local->call_count = count;
         local->xl_specf_unwind = xl_specf_getxattr_unwind;
         local->vol_uuid = vol_uuid;
+        memcpy (local->gauge, gauge, sizeof (local->gauge));
 
         frame->local = local;
 
@@ -357,3 +413,58 @@ err:
         return -1;
 
 }
+
+int
+gf_get_min_stime (xlator_t *this, dict_t *dst, char *key, data_t *value)
+{
+        int ret = -1;
+        uint32_t *net_timebuf = NULL;
+        uint32_t *value_timebuf = NULL;
+        uint32_t host_timebuf[2] = {0,};
+        uint32_t host_value_timebuf[2] = {0,};
+
+        /* stime should be minimum of all the other nodes */
+        ret = dict_get_bin (dst, key, (void **)&net_timebuf);
+        if (ret < 0) {
+                net_timebuf = GF_CALLOC (1, sizeof (int64_t),
+                                           gf_common_mt_char);
+                if (!net_timebuf)
+                        goto out;
+
+                ret = dict_set_bin (dst, key, net_timebuf, sizeof (int64_t));
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "key=%s: dict set failed", key);
+                        goto error;
+                }
+        }
+
+        value_timebuf = data_to_bin (value);
+        if (!value_timebuf) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "key=%s: getting value of stime failed", key);
+                ret = -1;
+                goto out;
+        }
+
+        get_hosttime (value_timebuf, host_value_timebuf);
+        get_hosttime (net_timebuf, host_timebuf);
+
+        /* can't use 'min()' macro here as we need to compare two fields
+           in the array, selectively */
+        if ((host_value_timebuf[0] > host_timebuf[0]) ||
+            ((host_value_timebuf[0] == host_timebuf[0]) &&
+             (host_value_timebuf[1] > host_timebuf[1]))) {
+                update_timebuf (value_timebuf, net_timebuf);
+        }
+
+        ret = 0;
+out:
+        return ret;
+error:
+        /* To be used only when net_timebuf is not set in the dict */
+        if (net_timebuf)
+                GF_FREE (net_timebuf);
+
+        return ret;
+}
diff --git a/xlators/lib/src/libxlator.h b/xlators/lib/src/libxlator.h
index 7e75c9482..1d5e1657f 100644
--- a/xlators/lib/src/libxlator.h
+++ b/xlators/lib/src/libxlator.h
@@ -48,6 +48,69 @@ struct volume_mark {
         uint32_t usec;
 }__attribute__ ((__packed__));
 
+
+/*
+ * The enumerated type here
+ * is used to index two kind
+ * of integer arrays:
+ * - gauges
+ * - counters
+
+ * A counter is used internally,
+ * in getxattr callbacks, to count
+ * the results, categorized as
+ * the enum names suggest. So values
+ * in the counter are always non-negative.
+
+ * Gauges are part of the API.
+ * The caller passes one to the
+ * top-level aggregator function,
+ * cluster_getmarkerattr(). The gauge
+ * defines an evaluation policy for the
+ * counter. That is, at the
+ * end of the aggregation process
+ * the gauge is matched against the
+ * counter, and the policy
+ * represented by the gauge decides
+ * whether to return with success or failure,
+ * and in latter case, what particular failure
+ * case (errno).
+
+ * The rules are the following: for some index i,
+ * - if gauge[i] == 0, no requirement is set
+ *   against counter[i];
+ * - if gauge[i] > 0, counter[i] >= gauge[i]
+ *   is required;
+ * - if gauge[i] < 0, counter[i] < |gauge[i]|
+ *   is required.
+
+ * If the requirement is not met, then i is mapped
+ * to the respective errno (MCNT_ENOENT -> ENOENT),
+ * or in lack of that, EINVAL.
+
+ * Cf. evaluate_marker_results() and marker_idx_errno_map[]
+ * in libxlator.c
+
+ * We provide two default gauges, one inteded for xtime
+ * aggregation, other for volume mark aggregation. The
+ * policies they represent agree with the hard-coded
+ * one prior to gauges. Cf. marker_xtime_default_gauge
+ * and marker_uuid_default_gauge in libxlator.c
+ */
+
+typedef enum {
+        MCNT_FOUND,
+        MCNT_NOTFOUND,
+        MCNT_ENODATA,
+        MCNT_ENOTCONN,
+        MCNT_ENOENT,
+        MCNT_EOTHER,
+        MCNT_MAX
+} marker_result_idx_t;
+
+extern int marker_xtime_default_gauge[];
+extern int marker_uuid_default_gauge[];
+
 struct marker_str {
         struct volume_mark    *volmark;
         data_t                *data;
@@ -55,13 +118,8 @@ struct marker_str {
         uint32_t               host_timebuf[2];
         uint32_t               net_timebuf[2];
         int32_t                call_count;
-        unsigned               has_xtime:1;
-        int32_t                enoent_count;
-        int32_t                enotconn_count;
-        int32_t                enodata_count;
-        int32_t                noxtime_count;
-
-        int                    esomerr;
+        int                    gauge[MCNT_MAX];
+        int                    count[MCNT_MAX];
 
         xlator_specf_unwind_t  xl_specf_unwind;
         void                  *xl_local;
@@ -71,15 +129,6 @@ struct marker_str {
 
 typedef struct marker_str xl_marker_local_t;
 
-static inline gf_boolean_t
-marker_has_volinfo (xl_marker_local_t *marker)
-{
-       if (marker->volmark)
-                return _gf_true;
-       else
-                return _gf_false;
-}
-
 int32_t
 cluster_markerxtime_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                          int op_ret, int op_errno, dict_t *dict, dict_t *xdata);
@@ -93,12 +142,12 @@ cluster_getmarkerattr (call_frame_t *frame,xlator_t *this, loc_t *loc,
                        const char *name, void *xl_local,
                        xlator_specf_unwind_t xl_specf_getxattr_unwind,
                        xlator_t **sub_volumes, int count, int type,
-                       char *vol_uuid);
+                       int *gauge, char *vol_uuid);
 
 int
 match_uuid_local (const char *name, char *uuid);
 
-
-
+int
+gf_get_min_stime (xlator_t *this, dict_t *dst, char *key, data_t *value);
 
 #endif /* !_LIBXLATOR_H */
diff --git a/xlators/mgmt/glusterd/src/Makefile.am b/xlators/mgmt/glusterd/src/Makefile.am
index a6f49ae01..933c44019 100644
--- a/xlators/mgmt/glusterd/src/Makefile.am
+++ b/xlators/mgmt/glusterd/src/Makefile.am
@@ -11,7 +11,9 @@ glusterd_la_SOURCES = glusterd.c glusterd-handler.c glusterd-sm.c \
 	glusterd-volgen.c glusterd-rebalance.c glusterd-quota.c \
 	glusterd-geo-rep.c glusterd-replace-brick.c glusterd-log-ops.c \
 	glusterd-volume-ops.c glusterd-brick-ops.c glusterd-mountbroker.c \
-	glusterd-syncop.c glusterd-hooks.c glusterd-volume-set.c
+	glusterd-syncop.c glusterd-hooks.c glusterd-volume-set.c \
+        glusterd-locks.c glusterd-snapshot.c glusterd-mgmt-handler.c \
+        glusterd-mgmt.c
 
 glusterd_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
 		     $(top_builddir)/rpc/xdr/src/libgfxdr.la \
@@ -21,7 +23,8 @@ glusterd_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
 noinst_HEADERS = glusterd.h glusterd-utils.h glusterd-op-sm.h \
 	glusterd-sm.h glusterd-store.h glusterd-mem-types.h \
 	glusterd-pmap.h glusterd-volgen.h glusterd-mountbroker.h \
-	glusterd-syncop.h glusterd-hooks.h
+	glusterd-syncop.h glusterd-hooks.h glusterd-locks.h \
+	glusterd-mgmt.h
 
 AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
 	-I$(rpclibdir) -I$(CONTRIBDIR)/rbtree \
diff --git a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
index 912a6a798..596503c21 100644
--- a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
+++ b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
@@ -546,6 +546,72 @@ glusterd_handle_add_brick (rpcsvc_request_t *req)
         return glusterd_big_locked_handler (req, __glusterd_handle_add_brick);
 }
 
+static int
+subvol_matcher_init (int **subvols, int count)
+{
+        int ret = -1;
+
+        *subvols = GF_CALLOC (count, sizeof(int), gf_gld_mt_int);
+        if (*subvols)
+                ret = 0;
+
+        return ret;
+}
+
+static void
+subvol_matcher_update (int *subvols, glusterd_volinfo_t *volinfo,
+                       glusterd_brickinfo_t *brickinfo)
+{
+        glusterd_brickinfo_t *tmp        = NULL;
+        int32_t               sub_volume = 0;
+        int                   pos        = 0;
+
+        list_for_each_entry (tmp, &volinfo->bricks, brick_list) {
+
+                if (strcmp (tmp->hostname, brickinfo->hostname) ||
+                    strcmp (tmp->path, brickinfo->path)) {
+                        pos++;
+                        continue;
+                }
+                gf_log (THIS->name, GF_LOG_DEBUG, LOGSTR_FOUND_BRICK,
+                                        brickinfo->hostname, brickinfo->path,
+                                        volinfo->volname);
+                sub_volume = (pos / volinfo->dist_leaf_count);
+                subvols[sub_volume]++;
+                break;
+        }
+
+}
+
+static int
+subvol_matcher_verify (int *subvols, glusterd_volinfo_t *volinfo, char *err_str,
+                       size_t err_len, char *vol_type)
+{
+        int i = 0;
+        int ret = 0;
+
+       do {
+
+                if (subvols[i] % volinfo->dist_leaf_count == 0) {
+                        continue;
+                } else {
+                        ret = -1;
+                        snprintf (err_str, err_len,
+                                "Bricks not from same subvol for %s", vol_type);
+                        gf_log (THIS->name, GF_LOG_ERROR, "%s", err_str);
+                        break;
+                }
+        } while (++i < volinfo->subvol_count);
+
+        return ret;
+}
+
+static void
+subvol_matcher_destroy (int *subvols)
+{
+        GF_FREE (subvols);
+}
+
 int
 __glusterd_handle_remove_brick (rpcsvc_request_t *req)
 {
@@ -559,10 +625,7 @@ __glusterd_handle_remove_brick (rpcsvc_request_t *req)
         int                       i                = 1;
         glusterd_volinfo_t       *volinfo          = NULL;
         glusterd_brickinfo_t     *brickinfo        = NULL;
-        int32_t                   pos              = 0;
-        int32_t                   sub_volume       = 0;
-        int32_t                   sub_volume_start = 0;
-        int32_t                   sub_volume_end   = 0;
+        int                      *subvols          = NULL;
         glusterd_brickinfo_t     *tmp              = NULL;
         char                      err_str[2048]    = {0};
         gf_cli_rsp                rsp              = {0,};
@@ -681,17 +744,6 @@ __glusterd_handle_remove_brick (rpcsvc_request_t *req)
                 goto out;
 	}
 
-        /*Do not allow remove-brick if the volume is a replicate volume*/
-        if ((volinfo->type == GF_CLUSTER_TYPE_REPLICATE) &&
-            (volinfo->brick_count == volinfo->replica_count)) {
-                snprintf (err_str, sizeof(err_str),
-                          "Removing brick from a replicate volume "
-                           "is not allowed");
-                gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
-                ret = -1;
-                goto out;
-        }
-
 	if (!replica_count &&
             (volinfo->type == GF_CLUSTER_TYPE_STRIPE_REPLICATE) &&
             (volinfo->brick_count == volinfo->dist_leaf_count)) {
@@ -738,6 +790,14 @@ __glusterd_handle_remove_brick (rpcsvc_request_t *req)
         }
 
         strcpy (brick_list, " ");
+
+        if ((volinfo->type != GF_CLUSTER_TYPE_NONE) &&
+            (volinfo->subvol_count > 1)) {
+                ret = subvol_matcher_init (&subvols, volinfo->subvol_count);
+                if (ret)
+                        goto out;
+        }
+
         while ( i <= count) {
                 snprintf (key, sizeof (key), "brick%d", i);
                 ret = dict_get_str (dict, key, &brick);
@@ -805,38 +865,18 @@ __glusterd_handle_remove_brick (rpcsvc_request_t *req)
                         goto out;
                 }
 
-                pos = 0;
-                list_for_each_entry (tmp, &volinfo->bricks, brick_list) {
-
-                        if (strcmp (tmp->hostname,brickinfo->hostname) ||
-                            strcmp (tmp->path, brickinfo->path)) {
-                                pos++;
-                                continue;
-                        }
+                /* Find which subvolume the brick belongs to */
+                subvol_matcher_update (subvols, volinfo, brickinfo);
+        }
 
-                        gf_log (this->name, GF_LOG_DEBUG, LOGSTR_FOUND_BRICK,
-                                brickinfo->hostname, brickinfo->path,
-                                volinfo->volname);
-                        if (!sub_volume && (volinfo->dist_leaf_count > 1)) {
-                                sub_volume = (pos / volinfo->dist_leaf_count) + 1;
-                                sub_volume_start = (volinfo->dist_leaf_count *
-                                                    (sub_volume - 1));
-                                sub_volume_end = (volinfo->dist_leaf_count *
-                                                  sub_volume) - 1;
-                        } else {
-                                if (pos < sub_volume_start ||
-                                    pos >sub_volume_end) {
-                                        ret = -1;
-                                        snprintf (err_str, sizeof (err_str),
-                                                  "Bricks not from same subvol "
-                                                  "for %s", vol_type);
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s", err_str);
-                                        goto out;
-                                }
-                        }
-                        break;
-                }
+        /* Check if the bricks belong to the same subvolumes.*/
+        if ((volinfo->type != GF_CLUSTER_TYPE_NONE) &&
+            (volinfo->subvol_count > 1)) {
+                ret = subvol_matcher_verify (subvols, volinfo,
+                                             err_str, sizeof(err_str),
+                                             vol_type);
+                if (ret)
+                        goto out;
         }
 
         ret = glusterd_op_begin_synctask (req, GD_OP_REMOVE_BRICK, dict);
@@ -859,6 +899,7 @@ out:
         }
 
         GF_FREE (brick_list);
+        subvol_matcher_destroy (subvols);
         free (cli_req.dict.dict_val); //its malloced by xdr
 
         return ret;
@@ -871,23 +912,121 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
                                             __glusterd_handle_remove_brick);
 }
 
+static int
+_glusterd_restart_gsync_session (dict_t *this, char *key,
+                                 data_t *value, void *data)
+{
+        char                          *slave      = NULL;
+        char                          *slave_buf  = NULL;
+        char                          *path_list  = NULL;
+        char                          *slave_vol  = NULL;
+        char                          *slave_ip   = NULL;
+        char                          *conf_path  = NULL;
+        char                         **errmsg     = NULL;
+        int                            ret        = -1;
+        glusterd_gsync_status_temp_t  *param      = NULL;
+        gf_boolean_t                   is_running = _gf_false;
+
+        param = (glusterd_gsync_status_temp_t *)data;
+
+        GF_ASSERT (param);
+        GF_ASSERT (param->volinfo);
+
+        slave = strchr(value->data, ':');
+        if (slave) {
+                slave++;
+                slave_buf = gf_strdup (slave);
+                if (!slave_buf) {
+                        gf_log ("", GF_LOG_ERROR,
+                                "Failed to gf_strdup");
+                        ret = -1;
+                        goto out;
+                }
+        }
+        else
+                return 0;
+
+        ret = dict_set_dynstr (param->rsp_dict, "slave", slave_buf);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to store slave");
+                if (slave_buf)
+                        GF_FREE(slave_buf);
+                goto out;
+        }
+
+        ret = glusterd_get_slave_details_confpath (param->volinfo,
+                                                   param->rsp_dict,
+                                                   &slave_ip, &slave_vol,
+                                                   &conf_path, errmsg);
+        if (ret) {
+                if (*errmsg)
+                        gf_log ("", GF_LOG_ERROR, "%s", *errmsg);
+                else
+                        gf_log ("", GF_LOG_ERROR,
+                                "Unable to fetch slave or confpath details.");
+                goto out;
+        }
+
+        /* In cases that gsyncd is not running, we will not invoke it
+         * because of add-brick. */
+        ret = glusterd_check_gsync_running_local (param->volinfo->volname,
+                                                  slave, conf_path,
+                                                  &is_running);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "gsync running validation failed.");
+                goto out;
+        }
+        if (_gf_false == is_running) {
+                gf_log ("", GF_LOG_DEBUG, "gsync session for %s and %s is"
+                        " not running on this node. Hence not restarting.",
+                        param->volinfo->volname, slave);
+                ret = 0;
+                goto out;
+        }
+
+        ret = glusterd_get_local_brickpaths (param->volinfo, &path_list);
+        if (!path_list) {
+                gf_log ("", GF_LOG_DEBUG, "This node not being part of"
+                        " volume should not be running gsyncd. Hence"
+                        " no gsyncd process to restart.");
+                ret = 0;
+                goto out;
+        }
+
+        ret = glusterd_check_restart_gsync_session (param->volinfo, slave,
+                                                    param->rsp_dict, path_list,
+                                                    conf_path, 0);
+        if (ret)
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to restart gsync session.");
+
+out:
+        gf_log ("", GF_LOG_DEBUG, "Returning %d.", ret);
+        return ret;
+}
+
 /* op-sm */
 
 int
 glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
                                 char  *bricks, dict_t *dict)
 {
-        glusterd_brickinfo_t *brickinfo     = NULL;
-        char                 *brick         = NULL;
-        int32_t               i             = 1;
-        char                 *brick_list    = NULL;
-        char                 *free_ptr1     = NULL;
-        char                 *free_ptr2     = NULL;
-        char                 *saveptr       = NULL;
-        int32_t               ret           = -1;
-        int32_t               stripe_count  = 0;
-        int32_t               replica_count = 0;
-        int32_t               type          = 0;
+        char                         *brick          = NULL;
+        int32_t                       i              = 1;
+        char                         *brick_list     = NULL;
+        char                         *free_ptr1      = NULL;
+        char                         *free_ptr2      = NULL;
+        char                         *saveptr        = NULL;
+        int32_t                       ret            = -1;
+        int32_t                       stripe_count   = 0;
+        int32_t                       replica_count  = 0;
+        int32_t                       type           = 0;
+        glusterd_brickinfo_t         *brickinfo      = NULL;
+        glusterd_gsync_status_temp_t  param          = {0, };
+        gf_boolean_t                  restart_needed = 0;
+        char                          msg[1024] __attribute__((unused)) = {0, };
+        int                           caps           = 0;
 
         GF_ASSERT (volinfo);
 
@@ -968,13 +1107,40 @@ glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
 
         if (count)
                 brick = strtok_r (brick_list+1, " \n", &saveptr);
+#ifdef HAVE_BD_XLATOR
+        if (brickinfo->vg[0])
+                caps = CAPS_BD | CAPS_THIN |
+                        CAPS_OFFLOAD_COPY | CAPS_OFFLOAD_SNAPSHOT;
+#endif
 
         while (i <= count) {
-
                 ret = glusterd_volume_brickinfo_get_by_brick (brick, volinfo,
                                                               &brickinfo);
                 if (ret)
                         goto out;
+#ifdef HAVE_BD_XLATOR
+                /* Check for VG/thin pool if its BD volume */
+                if (brickinfo->vg[0]) {
+                        ret = glusterd_is_valid_vg (brickinfo, 0, msg);
+                        if (ret) {
+                                gf_log (THIS->name, GF_LOG_CRITICAL, "%s", msg);
+                                goto out;
+                        }
+                        /* if anyone of the brick does not have thin support,
+                           disable it for entire volume */
+                        caps &= brickinfo->caps;
+                } else
+                        caps = 0;
+#endif
+
+                if (uuid_is_null (brickinfo->uuid)) {
+                        ret = glusterd_resolve_brick (brickinfo);
+                        if (ret) {
+                                gf_log ("", GF_LOG_ERROR, FMTSTR_RESOLVE_BRICK,
+                                        brickinfo->hostname, brickinfo->path);
+                                goto out;
+                        }
+                }
 
                 ret = glusterd_brick_start (volinfo, brickinfo,
                                             _gf_true);
@@ -982,8 +1148,27 @@ glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
                         goto out;
                 i++;
                 brick = strtok_r (NULL, " \n", &saveptr);
+
+                /* Check if the brick is added in this node, and set
+                 * the restart_needed flag. */
+                if ((!uuid_compare (brickinfo->uuid, MY_UUID)) &&
+                    !restart_needed) {
+                        restart_needed = 1;
+                        gf_log ("", GF_LOG_DEBUG,
+                                "Restart gsyncd session, if it's already "
+                                "running.");
+                }
         }
 
+        /* If the restart_needed flag is set, restart gsyncd sessions for that
+         * particular master with all the slaves. */
+        if (restart_needed) {
+                param.rsp_dict = dict;
+                param.volinfo = volinfo;
+                dict_foreach (volinfo->gsync_slaves,
+                              _glusterd_restart_gsync_session, &param);
+        }
+        volinfo->caps = caps;
 out:
         GF_FREE (free_ptr1);
         GF_FREE (free_ptr2);
@@ -1080,15 +1265,6 @@ glusterd_op_stage_add_brick (dict_t *dict, char **op_errstr)
                 goto out;
         }
 
-        if (volinfo->backend == GD_VOL_BK_BD) {
-                snprintf (msg, sizeof (msg), "Add brick is not supported for "
-                           "Block backend volume %s.", volname);
-                gf_log (THIS->name, GF_LOG_ERROR, "%s", msg);
-                *op_errstr = gf_strdup (msg);
-                ret = -1;
-                goto out;
-        }
-
         ret = glusterd_validate_volume_id (dict, volinfo);
         if (ret)
                 goto out;
@@ -1166,6 +1342,18 @@ glusterd_op_stage_add_brick (dict_t *dict, char **op_errstr)
                 }
 
                 if (!uuid_compare (brickinfo->uuid, MY_UUID)) {
+#ifdef HAVE_BD_XLATOR
+                        if (brickinfo->vg[0]) {
+                                ret = glusterd_is_valid_vg (brickinfo, 1, msg);
+                                if (ret) {
+                                        gf_log (THIS->name, GF_LOG_ERROR, "%s",
+                                                msg);
+                                        *op_errstr = gf_strdup (msg);
+                                        goto out;
+                                }
+                        }
+#endif
+
                         ret = glusterd_validate_and_create_brickpath (brickinfo,
                                                           volinfo->volume_id,
                                                           op_errstr, is_force);
@@ -1255,6 +1443,16 @@ glusterd_op_stage_remove_brick (dict_t *dict, char **op_errstr)
 
         case GF_OP_CMD_START:
         {
+                if ((volinfo->type == GF_CLUSTER_TYPE_REPLICATE) &&
+                    dict_get (dict, "replica-count")) {
+                        snprintf (msg, sizeof(msg), "Migration of data is not "
+                                  "needed when reducing replica count. Use the"
+                                  " 'force' option");
+                        errstr = gf_strdup (msg);
+                        gf_log (this->name, GF_LOG_ERROR, "%s", errstr);
+                        goto out;
+                }
+
                 if (GLUSTERD_STATUS_STARTED != volinfo->status) {
                         snprintf (msg, sizeof (msg), "Volume %s needs to be "
                                   "started before remove-brick (you can use "
@@ -1264,6 +1462,17 @@ glusterd_op_stage_remove_brick (dict_t *dict, char **op_errstr)
                         gf_log (this->name, GF_LOG_ERROR, "%s", errstr);
                         goto out;
                 }
+                if (!gd_is_remove_brick_committed (volinfo)) {
+                        snprintf (msg, sizeof (msg), "An earlier remove-brick "
+                                  "task exists for volume %s. Either commit it"
+                                  " or stop it before starting a new task.",
+                                  volinfo->volname);
+                        errstr = gf_strdup (msg);
+                        gf_log (this->name, GF_LOG_ERROR, "Earlier remove-brick"
+                                " task exists for volume %s.",
+                                volinfo->volname);
+                        goto out;
+                }
                 if (glusterd_is_defrag_on(volinfo)) {
                         errstr = gf_strdup("Rebalance is in progress. Please "
                                            "retry after completion");
@@ -1271,7 +1480,7 @@ glusterd_op_stage_remove_brick (dict_t *dict, char **op_errstr)
                         goto out;
                 }
 
-                if (is_origin_glusterd ()) {
+                if (is_origin_glusterd (dict)) {
                         ret = glusterd_generate_and_set_task_id
                                 (dict, GF_REMOVE_BRICK_TID_KEY);
                         if (ret) {
@@ -1458,15 +1667,6 @@ glusterd_op_add_brick (dict_t *dict, char **op_errstr)
                 goto out;
         }
 
-        /* Need to reset the defrag/rebalance status accordingly */
-        switch (volinfo->rebal.defrag_status) {
-        case GF_DEFRAG_STATUS_FAILED:
-        case GF_DEFRAG_STATUS_COMPLETE:
-                volinfo->rebal.defrag_status = 0;
-        default:
-                break;
-        }
-
         ret = glusterd_store_volinfo (volinfo, GLUSTERD_VOLINFO_VER_AC_INCREMENT);
         if (ret)
                 goto out;
@@ -1498,6 +1698,8 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
         glusterd_brickinfo_t    *tmp           = NULL;
         char                    *task_id_str   = NULL;
         xlator_t                *this          = NULL;
+        dict_t                  *bricks_dict   = NULL;
+        char                    *brick_tmpstr  = NULL;
 
         this = THIS;
         GF_ASSERT (this);
@@ -1525,7 +1727,7 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
         /* Set task-id, if available, in ctx dict for operations other than
          * start
          */
-        if (is_origin_glusterd () && (cmd != GF_OP_CMD_START)) {
+        if (is_origin_glusterd (dict) && (cmd != GF_OP_CMD_START)) {
                 if (!uuid_is_null (volinfo->rebal.rebalance_id)) {
                         ret = glusterd_copy_uuid_to_dict
                                 (volinfo->rebal.rebalance_id, dict,
@@ -1538,9 +1740,14 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
                 }
         }
 
-        /* Clear task-id on commmitting/stopping of remove-brick operation */
-        if ((cmd != GF_OP_CMD_START) || (cmd != GF_OP_CMD_STATUS))
+        /* Clear task-id, rebal.op and stored bricks on commmitting/stopping
+         * remove-brick */
+        if ((cmd != GF_OP_CMD_START) || (cmd != GF_OP_CMD_STATUS)) {
                 uuid_clear (volinfo->rebal.rebalance_id);
+                volinfo->rebal.op = GD_OP_NONE;
+                dict_unref (volinfo->rebal.dict);
+                volinfo->rebal.dict = NULL;
+        }
 
         ret = -1;
         switch (cmd) {
@@ -1587,6 +1794,7 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
                         ret = 0;
                 } else {
                         uuid_parse (task_id_str, volinfo->rebal.rebalance_id) ;
+                        volinfo->rebal.op = GD_OP_REMOVE_BRICK;
                 }
                 force = 0;
                 break;
@@ -1623,7 +1831,20 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
                 goto out;
         }
 
-
+        /* Save the list of bricks for later usage. Right now this is required
+         * for displaying the task parameters with task status in volume status.
+         */
+        bricks_dict = dict_new ();
+        if (!bricks_dict) {
+                ret = -1;
+                goto out;
+        }
+        ret = dict_set_int32 (bricks_dict, "count", count);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Failed to save remove-brick count");
+                goto out;
+        }
         while ( i <= count) {
                 snprintf (key, 256, "brick%d", i);
                 ret = dict_get_str (dict, key, &brick);
@@ -1633,6 +1854,21 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
                         goto out;
                 }
 
+                brick_tmpstr = gf_strdup (brick);
+                if (!brick_tmpstr) {
+                        ret = -1;
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Failed to duplicate brick name");
+                        goto out;
+                }
+                ret = dict_set_dynstr (bricks_dict, key, brick_tmpstr);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Failed to add brick to dict");
+                        goto out;
+                }
+                brick_tmpstr = NULL;
+
                 ret = glusterd_op_perform_remove_brick (volinfo, brick, force,
                                                         &need_rebalance);
                 if (ret)
@@ -1646,6 +1882,7 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
                         volinfo->replica_count, replica_count,
                         volinfo->volname);
                 volinfo->replica_count = replica_count;
+                volinfo->sub_count = replica_count;
                 volinfo->dist_leaf_count = glusterd_get_dist_leaf_count (volinfo);
                 volinfo->subvol_count = (volinfo->brick_count /
                                          volinfo->dist_leaf_count);
@@ -1662,6 +1899,8 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
                         }
                 }
         }
+        volinfo->rebal.dict = bricks_dict;
+        bricks_dict = NULL;
 
         ret = glusterd_create_volfiles_and_notify_services (volinfo);
         if (ret) {
@@ -1706,5 +1945,9 @@ out:
         if (ret && err_str[0] && op_errstr)
                 *op_errstr = gf_strdup (err_str);
 
+        GF_FREE (brick_tmpstr);
+        if (bricks_dict)
+                dict_unref (bricks_dict);
+
         return ret;
 }
diff --git a/xlators/mgmt/glusterd/src/glusterd-geo-rep.c b/xlators/mgmt/glusterd/src/glusterd-geo-rep.c
index 7e2ab833f..5786694bd 100644
--- a/xlators/mgmt/glusterd/src/glusterd-geo-rep.c
+++ b/xlators/mgmt/glusterd/src/glusterd-geo-rep.c
@@ -25,6 +25,40 @@
 
 #include <signal.h>
 
+static int
+dict_get_param (dict_t *dict, char *key, char **param);
+
+static int
+glusterd_get_statefile_name (glusterd_volinfo_t *volinfo, char *slave,
+                             char *conf_path, char **statefile);
+
+static int
+glusterd_get_slave_info (char *slave, char **slave_ip,
+                         char **slave_vol, char **op_errstr);
+
+static int
+glusterd_gsync_read_frm_status (char *path, char *buf, size_t blen);
+
+struct gsync_config_opt_vals_ gsync_confopt_vals[] = {
+        {.op_name        = "change_detector",
+         .no_of_pos_vals = 2,
+         .case_sensitive = _gf_true,
+         .values         = {"xsync", "changelog"},
+        },
+        {.op_name        = "special_sync_mode",
+         .no_of_pos_vals = 2,
+         .case_sensitive = _gf_true,
+         .values         = {"partial", "recover"}
+        },
+        {.op_name        = "log-level",
+         .no_of_pos_vals = 5,
+         .case_sensitive = _gf_false,
+         .values         = {"critical", "error", "warning", "info", "debug"}
+        },
+        {.op_name = NULL,
+        },
+};
+
 static char *gsync_reserved_opts[] = {
         "gluster-command-dir",
         "pid-file",
@@ -33,10 +67,156 @@ static char *gsync_reserved_opts[] = {
         "session-owner",
         "state-socket-unencoded",
         "socketdir",
+        "ignore-deletes",
+        "local-id",
+        "local-path",
+        "slave-id",
         NULL
 };
 
 int
+__glusterd_handle_sys_exec (rpcsvc_request_t *req)
+{
+        int32_t                 ret     = 0;
+        dict_t                  *dict   = NULL;
+        gf_cli_req              cli_req = {{0},};
+        glusterd_op_t           cli_op = GD_OP_SYS_EXEC;
+        glusterd_conf_t         *priv   = NULL;
+        char                    *host_uuid = NULL;
+        char                    err_str[2048] = {0,};
+        xlator_t                *this = NULL;
+
+        GF_ASSERT (req);
+
+        this = THIS;
+        GF_ASSERT (this);
+        priv = this->private;
+        GF_ASSERT (priv);
+
+        ret = xdr_to_generic (req->msg[0], &cli_req,
+                              (xdrproc_t)xdr_gf_cli_req);
+        if (ret < 0) {
+                req->rpc_err = GARBAGE_ARGS;
+                goto out;
+        }
+
+        if (cli_req.dict.dict_len) {
+                dict = dict_new ();
+                if (!dict)
+                        goto out;
+
+
+                ret = dict_unserialize (cli_req.dict.dict_val,
+                                        cli_req.dict.dict_len,
+                                        &dict);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR, "failed to "
+                                "unserialize req-buffer to dictionary");
+                        snprintf (err_str, sizeof (err_str), "Unable to decode "
+                                  "the command");
+                        goto out;
+               } else {
+                        dict->extra_stdfree = cli_req.dict.dict_val;
+                }
+
+                host_uuid = gf_strdup (uuid_utoa(MY_UUID));
+                if (host_uuid == NULL) {
+                        snprintf (err_str, sizeof (err_str), "Failed to get "
+                                  "the uuid of local glusterd");
+                        ret = -1;
+                        goto out;
+                }
+
+                ret = dict_set_dynstr (dict, "host-uuid", host_uuid);
+                if (ret)
+                        goto out;
+        }
+
+        ret = glusterd_op_begin_synctask (req, cli_op, dict);
+
+out:
+        if (ret) {
+                if (err_str[0] == '\0')
+                        snprintf (err_str, sizeof (err_str),
+                                  "Operation failed");
+                ret = glusterd_op_send_cli_response (cli_op, ret, 0, req,
+                                                     dict, err_str);
+        }
+        return ret;
+}
+
+int
+__glusterd_handle_copy_file (rpcsvc_request_t *req)
+{
+        int32_t                 ret     = 0;
+        dict_t                  *dict   = NULL;
+        gf_cli_req              cli_req = {{0},};
+        glusterd_op_t           cli_op = GD_OP_COPY_FILE;
+        glusterd_conf_t         *priv   = NULL;
+        char                    *host_uuid = NULL;
+        char                    err_str[2048] = {0,};
+        xlator_t                *this = NULL;
+
+        GF_ASSERT (req);
+
+        this = THIS;
+        GF_ASSERT (this);
+        priv = this->private;
+        GF_ASSERT (priv);
+
+        ret = xdr_to_generic (req->msg[0], &cli_req,
+                              (xdrproc_t)xdr_gf_cli_req);
+        if (ret < 0) {
+                req->rpc_err = GARBAGE_ARGS;
+                goto out;
+        }
+
+        if (cli_req.dict.dict_len) {
+                dict = dict_new ();
+                if (!dict)
+                        goto out;
+
+
+                ret = dict_unserialize (cli_req.dict.dict_val,
+                                        cli_req.dict.dict_len,
+                                        &dict);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR, "failed to "
+                                "unserialize req-buffer to dictionary");
+                        snprintf (err_str, sizeof (err_str), "Unable to decode "
+                                  "the command");
+                        goto out;
+               } else {
+                        dict->extra_stdfree = cli_req.dict.dict_val;
+                }
+
+                host_uuid = gf_strdup (uuid_utoa(MY_UUID));
+                if (host_uuid == NULL) {
+                        snprintf (err_str, sizeof (err_str), "Failed to get "
+                                  "the uuid of local glusterd");
+                        ret = -1;
+                        goto out;
+                }
+
+                ret = dict_set_dynstr (dict, "host-uuid", host_uuid);
+                if (ret)
+                        goto out;
+        }
+
+        ret = glusterd_op_begin_synctask (req, cli_op, dict);
+
+out:
+        if (ret) {
+                if (err_str[0] == '\0')
+                        snprintf (err_str, sizeof (err_str),
+                                  "Operation failed");
+                ret = glusterd_op_send_cli_response (cli_op, ret, 0, req,
+                                                     dict, err_str);
+        }
+        return ret;
+}
+
+int
 __glusterd_handle_gsync_set (rpcsvc_request_t *req)
 {
         int32_t                 ret     = 0;
@@ -100,13 +280,13 @@ __glusterd_handle_gsync_set (rpcsvc_request_t *req)
         ret = dict_get_str (dict, "master", &master);
         if (ret < 0) {
                 gf_log (this->name, GF_LOG_INFO, "master not found, while "
-                        "handling"GEOREP" options");
+                        "handling "GEOREP" options");
                 master = "(No Master)";
         }
 
         ret = dict_get_str (dict, "slave", &slave);
         if (ret < 0) {
-                gf_log (this->name, GF_LOG_INFO, "slave not not found, while"
+                gf_log (this->name, GF_LOG_INFO, "slave not found, while "
                         "handling "GEOREP" options");
                 slave = "(No Slave)";
         }
@@ -120,6 +300,10 @@ __glusterd_handle_gsync_set (rpcsvc_request_t *req)
         }
 
         switch (type) {
+        case GF_GSYNC_OPTION_TYPE_CREATE:
+                strncpy (operation, "create", sizeof (operation));
+                cli_op = GD_OP_GSYNC_CREATE;
+                break;
 
         case GF_GSYNC_OPTION_TYPE_START:
                 strncpy (operation, "start", sizeof (operation));
@@ -136,12 +320,9 @@ __glusterd_handle_gsync_set (rpcsvc_request_t *req)
         case GF_GSYNC_OPTION_TYPE_STATUS:
                 strncpy (operation, "status", sizeof (operation));
                 break;
-        case GF_GSYNC_OPTION_TYPE_ROTATE:
-                strncpy (operation, "rotate", sizeof(operation));
-                break;
         }
 
-        ret = glusterd_op_begin_synctask (req, GD_OP_GSYNC_SET, dict);
+        ret = glusterd_op_begin_synctask (req, cli_op, dict);
 
 out:
         if (ret) {
@@ -154,6 +335,17 @@ out:
         return ret;
 }
 
+int
+glusterd_handle_sys_exec (rpcsvc_request_t *req)
+{
+        return glusterd_big_locked_handler (req, __glusterd_handle_sys_exec);
+}
+
+int
+glusterd_handle_copy_file (rpcsvc_request_t *req)
+{
+        return glusterd_big_locked_handler (req, __glusterd_handle_copy_file);
+}
 
 int
 glusterd_handle_gsync_set (rpcsvc_request_t *req)
@@ -444,7 +636,7 @@ _fcbk_conftodict (char *resbuf, size_t blen, FILE *fp, void *data)
 }
 
 static int
-glusterd_gsync_get_config (char *master, char *slave, char *gl_workdir, dict_t *dict)
+glusterd_gsync_get_config (char *master, char *slave, char *conf_path, dict_t *dict)
 {
         /* key + value, where value must be able to accommodate a path */
         char resbuf[256 + PATH_MAX] = {0,};
@@ -452,7 +644,7 @@ glusterd_gsync_get_config (char *master, char *slave, char *gl_workdir, dict_t *
 
         runinit (&runner);
         runner_add_args  (&runner, GSYNCD_PREFIX"/gsyncd", "-c", NULL);
-        runner_argprintf (&runner, "%s/"GSYNC_CONF, gl_workdir);
+        runner_argprintf (&runner, "%s", conf_path);
         runner_argprintf (&runner, ":%s", master);
         runner_add_args  (&runner, slave, "--config-get-all", NULL);
 
@@ -462,13 +654,13 @@ glusterd_gsync_get_config (char *master, char *slave, char *gl_workdir, dict_t *
 
 static int
 glusterd_gsync_get_param_file (char *prmfile, const char *param, char *master,
-                               char *slave, char *gl_workdir)
+                               char *slave, char *conf_path)
 {
         runner_t            runner = {0,};
 
         runinit (&runner);
         runner_add_args  (&runner, GSYNCD_PREFIX"/gsyncd", "-c", NULL);
-        runner_argprintf (&runner, "%s/"GSYNC_CONF, gl_workdir);
+        runner_argprintf (&runner, "%s", conf_path);
         runner_argprintf (&runner, ":%s", master);
         runner_add_args  (&runner, slave, "--config-get", NULL);
         runner_argprintf (&runner, "%s-file", param);
@@ -477,83 +669,14 @@ glusterd_gsync_get_param_file (char *prmfile, const char *param, char *master,
 }
 
 static int
-glusterd_gsync_get_session_owner (char *master, char *slave, char *session_owner,
-                                  char *gl_workdir)
-{
-        runner_t runner = {0,};
-
-        runinit(&runner);
-        runner_add_args  (&runner, GSYNCD_PREFIX"/gsyncd", "-c", NULL);
-        runner_argprintf (&runner, "%s/"GSYNC_CONF, gl_workdir);
-        runner_argprintf (&runner, ":%s", master);
-        runner_add_args  (&runner, slave, "--config-get", "session-owner",
-                          NULL);
-
-        return glusterd_query_extutil (session_owner, &runner);
-}
-
-/* check whether @slave is local or remote. normalized
- * urls starting with ssh are considered to be remote
- * @returns
- *    1 if slave is remote
- *    0 is slave is local
- */
-static int
-glusterd_gsync_slave_is_remote (char *slave)
-{
-        int   ret     = 0;
-        char *ssh_pos = NULL;
-
-        ssh_pos = strstr(slave, "ssh://");
-        if ( ssh_pos && ((ssh_pos - slave) == 0) )
-                ret = 1;
-
-        return ret;
-}
-
-static int
-glusterd_gsync_get_slave_log_file (char *master, char *slave, char *log_file)
-{
-        int              ret        = -1;
-        runner_t         runner     = {0,};
-        char uuid_str[64]           = {0,};
-        glusterd_conf_t *priv       = NULL;
-        char            *gl_workdir = NULL;
-
-        GF_ASSERT(THIS);
-        GF_ASSERT(THIS->private);
-
-        priv = THIS->private;
-
-        GF_VALIDATE_OR_GOTO("gsyncd", master, out);
-        GF_VALIDATE_OR_GOTO("gsyncd", slave, out);
-
-        gl_workdir = priv->workdir;
-
-        /* get the session owner for the master-slave session */
-        ret = glusterd_gsync_get_session_owner (master, slave, uuid_str,
-                                                gl_workdir);
-        if (ret)
-                goto out;
-
-        /* get the log file for the slave */
-        runinit(&runner);
-        runner_add_args  (&runner, GSYNCD_PREFIX"/gsyncd", "-c", NULL);
-        runner_argprintf (&runner, "%s/"GSYNC_CONF, gl_workdir);
-        runner_argprintf (&runner, "--session-owner=%s", uuid_str);
-        runner_add_args  (&runner, slave, "--config-get", "log-file", NULL);
-
-        ret = glusterd_query_extutil (log_file, &runner);
-
- out:
-        return ret;
-}
-
-static int
-gsyncd_getpidfile (char *master, char *slave, char *pidfile)
+gsyncd_getpidfile (char *master, char *slave, char *pidfile, char *conf_path)
 {
         int                ret             = -1;
         glusterd_conf_t    *priv  = NULL;
+        char               *confpath = NULL;
+        char                conf_buf[PATH_MAX] = "";
+        struct stat         stbuf = {0,};
+
 
         GF_ASSERT (THIS);
         GF_ASSERT (THIS->private);
@@ -563,8 +686,22 @@ gsyncd_getpidfile (char *master, char *slave, char *pidfile)
         GF_VALIDATE_OR_GOTO ("gsync", master, out);
         GF_VALIDATE_OR_GOTO ("gsync", slave, out);
 
+        ret = lstat (conf_path, &stbuf);
+        if (!ret) {
+                gf_log ("", GF_LOG_DEBUG, "Using passed config template(%s).",
+                        conf_path);
+                confpath = conf_path;
+        } else {
+                ret = snprintf (conf_buf, sizeof(conf_buf) - 1,
+                                "%s/"GSYNC_CONF_TEMPLATE, priv->workdir);
+                conf_buf[ret] = '\0';
+                confpath = conf_buf;
+                gf_log ("", GF_LOG_DEBUG, "Using default config template(%s).",
+                        confpath);
+        }
+
         ret = glusterd_gsync_get_param_file (pidfile, "pid", master,
-                                              slave, priv->workdir);
+                                              slave, confpath);
         if (ret == -1) {
                 ret = -2;
                 gf_log ("", GF_LOG_WARNING, "failed to create the pidfile string");
@@ -578,32 +715,6 @@ gsyncd_getpidfile (char *master, char *slave, char *pidfile)
 }
 
 static int
-glusterd_gsyncd_getlogfile (char *master, char *slave, char *log_file)
-{
-        int                ret             = -1;
-        glusterd_conf_t    *priv  = NULL;
-
-        GF_ASSERT (THIS);
-        GF_ASSERT (THIS->private);
-
-        priv = THIS->private;
-
-        GF_VALIDATE_OR_GOTO ("gsync", master, out);
-        GF_VALIDATE_OR_GOTO ("gsync", slave, out);
-
-        ret = glusterd_gsync_get_param_file (log_file, "log", master,
-                                              slave, priv->workdir);
-        if (ret == -1) {
-                ret = -2;
-                gf_log ("", GF_LOG_WARNING, "failed to gsyncd logfile");
-                goto out;
-        }
-
- out:
-        return ret;
-}
-
-static int
 gsync_status_byfd (int fd)
 {
         GF_ASSERT (fd >= -1);
@@ -620,12 +731,12 @@ gsync_status_byfd (int fd)
  * return -1 when not running
  */
 int
-gsync_status (char *master, char *slave, int *status)
+gsync_status (char *master, char *slave, char *conf_path, int *status)
 {
         char pidfile[PATH_MAX] = {0,};
         int  fd                = -1;
 
-        fd = gsyncd_getpidfile (master, slave, pidfile);
+        fd = gsyncd_getpidfile (master, slave, pidfile, conf_path);
         if (fd == -2)
                 return -1;
 
@@ -662,16 +773,48 @@ out:
 }
 
 static int
-gsync_verify_config_options (dict_t *dict, char **op_errstr)
+glusterd_verify_gsyncd_spawn (char *master, char *slave)
 {
-        char  **resopt    = NULL;
-        int     i         = 0;
-        char   *subop     = NULL;
-        char   *slave     = NULL;
-        char   *op_name   = NULL;
-        char   *op_value  = NULL;
-        char   *t         = NULL;
-        gf_boolean_t banned = _gf_true;
+        int                 ret = 0;
+        runner_t            runner = {0,};
+
+        runinit (&runner);
+        runner_add_args (&runner, GSYNCD_PREFIX"/gsyncd",
+                         "--verify", "spawning", NULL);
+        runner_argprintf (&runner, ":%s", master);
+        runner_add_args  (&runner, slave, NULL);
+        runner_redir (&runner, STDOUT_FILENO, RUN_PIPE);
+        ret = runner_start (&runner);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "spawning child failed");
+                ret = -1;
+                goto out;
+        }
+
+        if (runner_end (&runner) != 0)
+                ret = -1;
+
+out:
+        gf_log ("", GF_LOG_DEBUG, "returning %d", ret);
+        return ret;
+}
+
+static int
+gsync_verify_config_options (dict_t *dict, char **op_errstr, char *volname)
+{
+        char                         **resopt           = NULL;
+        int                            i                = 0;
+        int                            ret              = -1;
+        char                          *subop            = NULL;
+        char                          *slave            = NULL;
+        char                          *op_name          = NULL;
+        char                          *op_value         = NULL;
+        char                          *t                = NULL;
+        char                           errmsg[PATH_MAX] = "";
+        gf_boolean_t                   banned           = _gf_true;
+        gf_boolean_t                   op_match         = _gf_true;
+        gf_boolean_t                   val_match        = _gf_true;
+        struct gsync_config_opt_vals_ *conf_vals        = NULL;
 
         if (dict_get_str (dict, "subop", &subop) != 0) {
                 gf_log ("", GF_LOG_WARNING, "missing subop");
@@ -695,6 +838,12 @@ gsync_verify_config_options (dict_t *dict, char **op_errstr)
         }
 
         if (runcmd (GSYNCD_PREFIX"/gsyncd", "--config-check", op_name, NULL)) {
+                ret = glusterd_verify_gsyncd_spawn (volname, slave);
+                if (ret) {
+                        gf_log ("", GF_LOG_ERROR, "Unable to spawn gsyncd");
+                        return 0;
+                }
+
                 gf_log ("", GF_LOG_WARNING, "Invalid option %s", op_name);
                 *op_errstr = gf_strdup ("Invalid option");
 
@@ -739,35 +888,109 @@ gsync_verify_config_options (dict_t *dict, char **op_errstr)
                 }
         }
 
+        /* Check options in gsync_confopt_vals for invalid values */
+        for (conf_vals = gsync_confopt_vals; conf_vals->op_name; conf_vals++) {
+                op_match = _gf_true;
+                for (i = 0; conf_vals->op_name[i] && op_name[i]; i++) {
+                        if (conf_vals->op_name[i] == op_name[i] ||
+                            (conf_vals->op_name[i] == '_' && op_name[i] == '-'))
+                                continue;
+                        op_match = _gf_false;
+                }
+
+                if (op_match) {
+                        val_match = _gf_false;
+                        for (i = 0; i < conf_vals->no_of_pos_vals; i++) {
+                                if(conf_vals->case_sensitive){
+                                        if (!strcmp (conf_vals->values[i], op_value))
+                                                val_match = _gf_true;
+                                } else {
+                                        if (!strcasecmp (conf_vals->values[i], op_value))
+                                                val_match = _gf_true;
+                                }
+                        }
+
+                        if (!val_match) {
+                                ret = snprintf (errmsg, sizeof(errmsg) - 1,
+                                                "Invalid values (%s) for"
+                                                " option %s", op_value,
+                                                op_name);
+                                errmsg[ret] = '\0';
+
+                                gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                                *op_errstr = gf_strdup (errmsg);
+                                return -1;
+                        }
+                }
+        }
+
         return 0;
 }
 
 static int
 glusterd_get_gsync_status_mst_slv (glusterd_volinfo_t *volinfo,
-                                   char *slave, dict_t *rsp_dict, char *node);
+                                   char *slave, char *conf_path,
+                                   dict_t *rsp_dict, char *node);
 
 static int
 _get_status_mst_slv (dict_t *this, char *key, data_t *value, void *data)
 {
         glusterd_gsync_status_temp_t  *param = NULL;
         char                          *slave = NULL;
-        int                           ret = 0;
+        char                          *slave_buf = NULL;
+        char                          *slave_ip  = NULL;
+        char                          *slave_vol = NULL;
+        char                          *errmsg    = NULL;
+        char                           conf_path[PATH_MAX] = "";
+        int                           ret = -1;
+        glusterd_conf_t              *priv = NULL;
 
         param = (glusterd_gsync_status_temp_t *)data;
 
         GF_ASSERT (param);
         GF_ASSERT (param->volinfo);
 
+        if (THIS)
+                priv = THIS->private;
+        if (priv == NULL) {
+                gf_log ("", GF_LOG_ERROR, "priv of glusterd not present");
+                goto out;
+        }
+
         slave = strchr(value->data, ':');
-        if (slave)
-                slave ++;
-        else
+        if (!slave)
                 return 0;
+        slave++;
+
+        ret = glusterd_get_slave_info (slave, &slave_ip, &slave_vol, &errmsg);
+        if (ret) {
+                if (errmsg)
+                        gf_log ("", GF_LOG_ERROR, "Unable to fetch "
+                                "slave details. Error: %s", errmsg);
+                else
+                        gf_log ("", GF_LOG_ERROR,
+                                "Unable to fetch slave details.");
+                ret = -1;
+                goto out;
+        }
+
+        ret = snprintf (conf_path, sizeof(conf_path) - 1,
+                        "%s/"GEOREP"/%s_%s_%s/gsyncd.conf",
+                        priv->workdir, param->volinfo->volname,
+                        slave_ip, slave_vol);
+        conf_path[ret] = '\0';
 
         ret = glusterd_get_gsync_status_mst_slv(param->volinfo,
-                                                slave, param->rsp_dict,
+                                                slave, conf_path,
+                                                param->rsp_dict,
                                                 param->node);
-        return 0;
+out:
+
+        if (slave_buf)
+                GF_FREE(slave_buf);
+
+        gf_log ("", GF_LOG_DEBUG, "Returning %d.", ret);
+        return ret;
 }
 
 
@@ -788,19 +1011,22 @@ static int
 glusterd_remove_slave_in_info (glusterd_volinfo_t *volinfo, char *slave,
                                char **op_errstr)
 {
+        int   zero_slave_entries = _gf_true;
         int   ret = 0;
         char *slavekey = NULL;
 
         GF_ASSERT (volinfo);
         GF_ASSERT (slave);
 
-        ret = glusterd_get_slave (volinfo, slave, &slavekey);
-        if (ret < 0) {
-                ret++;
-                goto out;
-        }
-
-        dict_del (volinfo->gsync_slaves, slavekey);
+        do {
+                ret = glusterd_get_slave (volinfo, slave, &slavekey);
+                if (ret < 0 && zero_slave_entries) {
+                        ret++;
+                        goto out;
+                }
+                zero_slave_entries = _gf_false;
+                dict_del (volinfo->gsync_slaves, slavekey);
+        } while (ret >= 0);
 
         ret = glusterd_store_volinfo (volinfo,
                                       GLUSTERD_VOLINFO_VER_AC_INCREMENT);
@@ -853,8 +1079,9 @@ glusterd_gsync_get_uuid (char *slave, glusterd_volinfo_t *vol,
         return ret;
 }
 
-static int
+int
 glusterd_check_gsync_running_local (char *master, char *slave,
+                                    char *conf_path,
                                     gf_boolean_t *is_run)
 {
         int                 ret    = -1;
@@ -865,7 +1092,7 @@ glusterd_check_gsync_running_local (char *master, char *slave,
         GF_ASSERT (is_run);
 
         *is_run = _gf_false;
-        ret = gsync_status (master, slave, &ret_status);
+        ret = gsync_status (master, slave, conf_path, &ret_status);
         if (ret == 0 && ret_status == 0) {
                 *is_run = _gf_true;
         } else if (ret == -1) {
@@ -882,7 +1109,8 @@ glusterd_check_gsync_running_local (char *master, char *slave,
 
 static int
 glusterd_store_slave_in_info (glusterd_volinfo_t *volinfo, char *slave,
-                              char *host_uuid, char **op_errstr)
+                              char *host_uuid, char **op_errstr,
+                              gf_boolean_t is_force)
 {
         int    ret = 0;
         int    maxslv = 0;
@@ -905,7 +1133,8 @@ glusterd_store_slave_in_info (glusterd_volinfo_t *volinfo, char *slave,
         case -1:
                                 break;
         default:
-                GF_ASSERT (ret > 0);
+                if (!is_force)
+                        GF_ASSERT (ret > 0);
                 ret = dict_get_str (volinfo->gsync_slaves, slavekey, &slaveentry);
                 GF_ASSERT (ret == 0);
 
@@ -914,13 +1143,23 @@ glusterd_store_slave_in_info (glusterd_volinfo_t *volinfo, char *slave,
                  * assert an uuid mismatch
                  */
                 t = strtail (slaveentry, host_uuid);
-                GF_ASSERT (!t || *t != ':');
+                if (!is_force)
+                        GF_ASSERT (!t || *t != ':');
+
+                if (is_force) {
+                        gf_log ("", GF_LOG_DEBUG, GEOREP" has already been "
+                                "invoked for the %s (master) and %s (slave)."
+                                " Allowing without saving info again due to"
+                                " force command.", volinfo->volname, slave);
+                        ret = 0;
+                        goto out;
+                }
 
                 gf_log ("", GF_LOG_ERROR, GEOREP" has already been invoked for "
                                           "the %s (master) and %s (slave) "
                                           "from a different machine",
                                            volinfo->volname, slave);
-                *op_errstr = gf_strdup (GEOREP" already running in an an"
+                *op_errstr = gf_strdup (GEOREP" already running in "
                                         "another machine");
                 ret = -1;
                 goto out;
@@ -953,23 +1192,26 @@ glusterd_store_slave_in_info (glusterd_volinfo_t *volinfo, char *slave,
         return ret;
 }
 
-
 static int
 glusterd_op_verify_gsync_start_options (glusterd_volinfo_t *volinfo,
-                                        char *slave, char **op_errstr)
+                                        char *slave, char *conf_path,
+                                        char *statefile, char **op_errstr,
+                                        gf_boolean_t is_force)
 {
         int                     ret = -1;
         gf_boolean_t            is_running = _gf_false;
         char                    msg[2048] = {0};
         uuid_t                  uuid = {0};
-        glusterd_conf_t         *priv = NULL;
-        xlator_t                *this = NULL;
+        glusterd_conf_t        *priv = NULL;
+        xlator_t               *this = NULL;
+        struct stat             stbuf = {0,};
 
         this = THIS;
 
         GF_ASSERT (volinfo);
         GF_ASSERT (slave);
         GF_ASSERT (op_errstr);
+        GF_ASSERT (conf_path);
         GF_ASSERT (this && this->private);
 
         priv  = this->private;
@@ -979,26 +1221,56 @@ glusterd_op_verify_gsync_start_options (glusterd_volinfo_t *volinfo,
                           "before "GEOREP" start", volinfo->volname);
                 goto out;
         }
+
+        ret = lstat (statefile, &stbuf);
+        if (ret) {
+                snprintf (msg, sizeof (msg), "Session between %s and %s has"
+                          " not been created. Please create session and retry.",
+                          volinfo->volname, slave);
+                gf_log ("", GF_LOG_ERROR, "%s", msg);
+                *op_errstr = gf_strdup (msg);
+                goto out;
+        }
+
+        /* Check if the gsync slave info is stored. If not
+         * session has not been created */
+        ret = glusterd_gsync_get_uuid (slave, volinfo, uuid);
+        if (ret) {
+                snprintf (msg, sizeof (msg), "Session between %s and %s has"
+                          " not been created. Please create session and retry.",
+                          volinfo->volname, slave);
+                gf_log ("", GF_LOG_ERROR, "%s", msg);
+                goto out;
+        }
+
+        if (is_force) {
+                ret = 0;
+                goto out;
+        }
+
         /*Check if the gsync is already started in cmd. inited host
          * If so initiate add it into the glusterd's priv*/
-        ret = glusterd_gsync_get_uuid (slave, volinfo, uuid);
-        if ((ret == 0) && (uuid_compare (MY_UUID, uuid) == 0)) {
-                ret = glusterd_check_gsync_running_local (volinfo->volname,
-                                                          slave, &is_running);
-                if (ret) {
-                        snprintf (msg, sizeof (msg), GEOREP" start option "
-                                  "validation failed ");
-                        goto out;
-                }
-                if (_gf_true == is_running) {
-                        snprintf (msg, sizeof (msg), GEOREP " session between"
-                                  " %s & %s already started", volinfo->volname,
-                                  slave);
-                        ret = -1;
-                        goto out;
-                }
+        ret = glusterd_check_gsync_running_local (volinfo->volname,
+                                                  slave, conf_path,
+                                                  &is_running);
+        if (ret) {
+                snprintf (msg, sizeof (msg), GEOREP" start option "
+                          "validation failed ");
+                goto out;
+        }
+        if (_gf_true == is_running) {
+                snprintf (msg, sizeof (msg), GEOREP " session between"
+                          " %s & %s already started", volinfo->volname,
+                          slave);
+                ret = -1;
+                goto out;
+        }
+
+        ret = glusterd_verify_gsyncd_spawn (volinfo->volname, slave);
+        if (ret) {
+                snprintf (msg, sizeof (msg), "Unable to spawn gsyncd");
+                gf_log ("", GF_LOG_ERROR, "%s", msg);
         }
-        ret = 0;
 out:
         if (ret && (msg[0] != '\0')) {
                 *op_errstr = gf_strdup (msg);
@@ -1024,11 +1296,13 @@ glusterd_check_gsync_running (glusterd_volinfo_t *volinfo, gf_boolean_t *flag)
 
 static int
 glusterd_op_verify_gsync_running (glusterd_volinfo_t *volinfo,
-                                  char *slave, char **op_errstr)
+                                  char *slave, char *conf_path,
+                                  char **op_errstr)
 {
-        int                     ret = -1;
-        char                    msg[2048] = {0};
-        uuid_t                  uuid = {0};
+        int        pfd               = -1;
+        int        ret               = -1;
+        char       msg[2048]         = {0};
+        char       pidfile[PATH_MAX] = {0,};
 
         GF_ASSERT (THIS && THIS->private);
         GF_ASSERT (volinfo);
@@ -1041,13 +1315,26 @@ glusterd_op_verify_gsync_running (glusterd_volinfo_t *volinfo,
 
                 goto out;
         }
-        ret = glusterd_gsync_get_uuid (slave, volinfo, uuid);
-        if (ret == -1) {
-                snprintf (msg, sizeof (msg), GEOREP" session between %s & %s"
-                          " not active", volinfo->volname, slave);
+
+        pfd = gsyncd_getpidfile (volinfo->volname, slave, pidfile, conf_path);
+        if (pfd == -2) {
+                gf_log ("", GF_LOG_ERROR, GEOREP" stop validation "
+                        "failed for %s & %s", volinfo->volname, slave);
+                ret = -1;
+                goto out;
+        }
+        if (gsync_status_byfd (pfd) == -1) {
+                snprintf (msg, sizeof (msg), GEOREP" session b/w %s & %s is not"
+                          " running on this node.", volinfo->volname, slave);
+                gf_log ("", GF_LOG_ERROR, "%s", msg);
+                ret = -1;
+                /* monitor gsyncd already dead */
                 goto out;
         }
 
+        if (pfd < 0)
+                goto out;
+
         ret = 0;
 out:
         if (ret && (msg[0] != '\0')) {
@@ -1066,6 +1353,18 @@ glusterd_verify_gsync_status_opts (dict_t *dict, char **op_errstr)
         gf_boolean_t       exists = _gf_false;
         glusterd_volinfo_t *volinfo = NULL;
         int                ret = 0;
+        char               *conf_path = NULL;
+        char               *slave_ip = NULL;
+        char               *slave_vol = NULL;
+        glusterd_conf_t    *priv = NULL;
+
+        if (THIS)
+                priv = THIS->private;
+        if (priv == NULL) {
+                gf_log ("", GF_LOG_ERROR, "priv of glusterd not present");
+                *op_errstr = gf_strdup ("glusterd defunct");
+                goto out;
+        }
 
         ret = dict_get_str (dict, "master", &volname);
         if (ret < 0) {
@@ -1090,16 +1389,25 @@ glusterd_verify_gsync_status_opts (dict_t *dict, char **op_errstr)
                 goto out;
         }
 
- out:
+        ret = glusterd_get_slave_details_confpath (volinfo, dict, &slave_ip,
+                                                   &slave_vol, &conf_path,
+                                                   op_errstr);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to fetch slave  or confpath details.");
+                ret = -1;
+                goto out;
+        }
+
+out:
         gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
         return ret;
-
 }
 
 
 int
 glusterd_op_gsync_args_get (dict_t *dict, char **op_errstr,
-                            char **master, char **slave)
+                            char **master, char **slave, char **host_uuid)
 {
 
         int             ret = -1;
@@ -1124,6 +1432,14 @@ glusterd_op_gsync_args_get (dict_t *dict, char **op_errstr,
                 }
         }
 
+        if (host_uuid) {
+                ret = dict_get_str (dict, "host-uuid", host_uuid);
+                if (ret < 0) {
+                        gf_log ("", GF_LOG_WARNING, "host_uuid not found");
+                        *op_errstr = gf_strdup ("host_uuid not found");
+                        goto out;
+                }
+        }
 
         ret = 0;
 out:
@@ -1132,17 +1448,647 @@ out:
 }
 
 int
+glusterd_op_stage_sys_exec (dict_t *dict, char **op_errstr)
+{
+        char             errmsg[PATH_MAX]          = "";
+        char            *command                   = NULL;
+        char             command_path[PATH_MAX]    = "";
+        struct stat      st                        = {0,};
+        int              ret                       = -1;
+        glusterd_conf_t *conf                      = NULL;
+        xlator_t        *this                      = NULL;
+
+        this = THIS;
+        GF_ASSERT (this);
+        conf = this->private;
+        GF_ASSERT (conf);
+
+        if (conf->op_version < 2) {
+                gf_log ("", GF_LOG_ERROR, "Op Version not supported.");
+                snprintf (errmsg, sizeof(errmsg), "One or more nodes do not"
+                          " support the required op version.");
+                *op_errstr = gf_strdup (errmsg);
+                ret = -1;
+                goto out;
+        }
+
+        ret = dict_get_str (dict, "command", &command);
+        if (ret) {
+                strcpy (errmsg, "internal error");
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to get command from dict");
+                goto out;
+        }
+
+        /* enforce local occurrence of the command */
+        if (strchr (command, '/')) {
+                strcpy (errmsg, "invalid command name");
+                ret = -1;
+                goto out;
+        }
+
+        sprintf (command_path, GSYNCD_PREFIX"/peer_%s", command);
+        /* check if it's executable */
+        ret = access (command_path, X_OK);
+        if (!ret)
+                /* check if it's a regular file */
+                ret = stat (command_path, &st);
+        if (!ret && !S_ISREG (st.st_mode))
+                ret = -1;
+
+out:
+        if (ret) {
+                if (errmsg[0] == '\0')
+                        snprintf (errmsg, sizeof (errmsg), "%s not found.",
+                                  command);
+                *op_errstr = gf_strdup (errmsg);
+                gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+        }
+
+        gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
+        return ret;
+}
+
+int
+glusterd_op_stage_copy_file (dict_t *dict, char **op_errstr)
+{
+        char             abs_filename[PATH_MAX] = "";
+        char             errmsg[PATH_MAX]       = "";
+        char            *filename               = NULL;
+        char            *host_uuid              = NULL;
+        char             uuid_str [64]          = {0};
+        int              ret                    = -1;
+        glusterd_conf_t *priv                   = NULL;
+        struct stat      stbuf                  = {0,};
+
+        if (THIS)
+                priv = THIS->private;
+        if (priv == NULL) {
+                gf_log ("", GF_LOG_ERROR, "priv of glusterd not present");
+                *op_errstr = gf_strdup ("glusterd defunct");
+                goto out;
+        }
+
+        if (priv->op_version < 2) {
+                gf_log ("", GF_LOG_ERROR, "Op Version not supported.");
+                snprintf (errmsg, sizeof(errmsg), "One or more nodes do not"
+                          " support the required op version.");
+                *op_errstr = gf_strdup (errmsg);
+                ret = -1;
+                goto out;
+        }
+
+        ret = dict_get_str (dict, "host-uuid", &host_uuid);
+        if (ret < 0) {
+                gf_log ("", GF_LOG_ERROR, "Unable to fetch"
+                        " host-uuid from dict.");
+                goto out;
+        }
+
+        uuid_utoa_r (MY_UUID, uuid_str);
+        if (!strcmp (uuid_str, host_uuid)) {
+                ret = dict_get_str (dict, "source", &filename);
+                if (ret < 0) {
+                        gf_log ("", GF_LOG_ERROR, "Unable to fetch"
+                                " filename from dict.");
+                        *op_errstr = gf_strdup ("command unsuccessful");
+                        goto out;
+                }
+                snprintf (abs_filename, sizeof(abs_filename),
+                          "%s/%s", priv->workdir, filename);
+
+                ret = lstat (abs_filename, &stbuf);
+                if (ret) {
+                        snprintf (errmsg, sizeof (errmsg), "Source file"
+                                  " does not exist in %s", priv->workdir);
+                        *op_errstr = gf_strdup (errmsg);
+                        goto out;
+                }
+
+                if (!S_ISREG(stbuf.st_mode)) {
+                        snprintf (errmsg, sizeof (errmsg), "Source file"
+                                 " is not a regular file.");
+                        *op_errstr = gf_strdup (errmsg);
+                        gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                        ret = -1;
+                        goto out;
+                }
+        }
+
+        ret = 0;
+out:
+        gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
+        return ret;
+}
+
+static int
+glusterd_get_statefile_name (glusterd_volinfo_t *volinfo, char *slave,
+                             char *conf_path, char **statefile)
+{
+        glusterd_conf_t *priv = NULL;
+        int              ret = -1;
+        char            *master    = NULL;
+        char            *buf       = NULL;
+        dict_t          *confd = NULL;
+        char            *confpath = NULL;
+        char             conf_buf[PATH_MAX] = "";
+        struct stat      stbuf = {0,};
+
+        GF_ASSERT (THIS);
+        GF_ASSERT (THIS->private);
+        GF_ASSERT (volinfo);
+
+        master = volinfo->volname;
+
+        confd = dict_new ();
+        if (!confd) {
+                gf_log ("", GF_LOG_ERROR, "Unable to create new dict");
+                goto out;
+        }
+
+        priv = THIS->private;
+
+        ret = lstat (conf_path, &stbuf);
+        if (!ret) {
+                gf_log ("", GF_LOG_INFO, "Using passed config template(%s).",
+                        conf_path);
+                confpath = conf_path;
+        } else {
+                ret = snprintf (conf_buf, sizeof(conf_buf) - 1,
+                                "%s/"GSYNC_CONF_TEMPLATE, priv->workdir);
+                conf_buf[ret] = '\0';
+                confpath = conf_buf;
+                gf_log ("", GF_LOG_INFO, "Using default config template(%s).",
+                        confpath);
+        }
+
+        ret = glusterd_gsync_get_config (master, slave, confpath,
+                                         confd);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "Unable to get configuration data"
+                        "for %s(master), %s(slave)", master, slave);
+                goto out;
+
+        }
+
+        ret = dict_get_param (confd, "state_file", &buf);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "Unable to get state_file's name.");
+                goto out;
+        }
+
+        *statefile = gf_strdup(buf);
+        if (!*statefile) {
+                gf_log ("", GF_LOG_ERROR, "Unable to gf_strdup.");
+                ret = -1;
+                goto out;
+        }
+
+        ret = 0;
+ out:
+        if (confd)
+                dict_destroy (confd);
+
+        gf_log ("", GF_LOG_DEBUG, "Returning %d ", ret);
+        return ret;
+}
+
+static int
+glusterd_create_status_file (char *master, char *slave, char *slave_ip,
+                             char *slave_vol, char *status)
+{
+        int                ret    = -1;
+        runner_t           runner = {0,};
+        glusterd_conf_t   *priv   = NULL;
+
+        if (THIS)
+                priv = THIS->private;
+        if (priv == NULL) {
+                gf_log ("", GF_LOG_ERROR, "priv of glusterd not present");
+                goto out;
+        }
+
+        if (!status) {
+                gf_log ("", GF_LOG_ERROR, "Status Empty");
+                goto out;
+        }
+        gf_log ("", GF_LOG_DEBUG, "slave = %s", slave);
+
+        runinit (&runner);
+        runner_add_args (&runner, GSYNCD_PREFIX"/gsyncd", "--create",
+                         status, "-c", NULL);
+        runner_argprintf (&runner, "%s/"GEOREP"/%s_%s_%s/gsyncd.conf",
+                          priv->workdir, master, slave_ip, slave_vol);
+        runner_argprintf (&runner, ":%s", master);
+        runner_add_args  (&runner, slave, NULL);
+        synclock_unlock (&priv->big_lock);
+        ret = runner_run (&runner);
+        synclock_lock (&priv->big_lock);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "Creating status file failed.");
+                ret = -1;
+                goto out;
+        }
+
+        ret = 0;
+out:
+        gf_log ("", GF_LOG_DEBUG, "returning %d", ret);
+        return ret;
+}
+
+static int
+glusterd_verify_slave (char *volname, char *slave_ip, char *slave,
+                       char **op_errstr, gf_boolean_t *is_force_blocker)
+{
+        int32_t          ret                     = -1;
+        runner_t         runner                  = {0,};
+        char             log_file_path[PATH_MAX] = "";
+        char             buf[PATH_MAX]           = "";
+        char            *tmp                     = NULL;
+        char            *save_ptr                = NULL;
+        glusterd_conf_t *priv                    = NULL;
+
+        GF_ASSERT (volname);
+        GF_ASSERT (slave_ip);
+        GF_ASSERT (slave);
+
+        if (THIS)
+                priv = THIS->private;
+        if (priv == NULL) {
+                gf_log ("", GF_LOG_ERROR, "priv of glusterd not present");
+                goto out;
+        }
+
+        snprintf (log_file_path, sizeof(log_file_path),
+                  DEFAULT_LOG_FILE_DIRECTORY"/create_verify_log");
+
+        runinit (&runner);
+        runner_add_args  (&runner, GSYNCD_PREFIX"/gverify.sh", NULL);
+        runner_argprintf (&runner, "%s", volname);
+        runner_argprintf (&runner, "%s", slave_ip);
+        runner_argprintf (&runner, "%s", slave);
+        runner_argprintf (&runner, "%s", log_file_path);
+        runner_redir (&runner, STDOUT_FILENO, RUN_PIPE);
+        synclock_unlock (&priv->big_lock);
+        ret = runner_run (&runner);
+        synclock_lock (&priv->big_lock);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "Not a valid slave");
+                ret = glusterd_gsync_read_frm_status (log_file_path,
+                                                      buf, sizeof(buf));
+                if (ret) {
+                        gf_log ("", GF_LOG_ERROR, "Unable to read from %s",
+                                log_file_path);
+                        goto out;
+                }
+
+                /* Tokenize the error message from gverify.sh to figure out
+                 * if the error is a force blocker or not. */
+                tmp = strtok_r (buf, "|", &save_ptr);
+                if (!strcmp (tmp, "FORCE_BLOCKER"))
+                        *is_force_blocker = 1;
+                else {
+                        /* No FORCE_BLOCKER flag present so all that is
+                         * present is the error message. */
+                        *is_force_blocker = 0;
+                        if (tmp)
+                                *op_errstr = gf_strdup (tmp);
+                        ret = -1;
+                        goto out;
+                }
+
+                /* Copy rest of the error message to op_errstr */
+                tmp = strtok_r (NULL, "|", &save_ptr);
+                if (tmp)
+                        *op_errstr = gf_strdup (tmp);
+                ret = -1;
+                goto out;
+        }
+        ret = 0;
+out:
+        unlink (log_file_path);
+        gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
+        return ret;
+}
+
+int
+glusterd_mountbroker_check (char **slave_ip, char **op_errstr)
+{
+        int   ret             = -1;
+        char *tmp             = NULL;
+        char *save_ptr        = NULL;
+        char *username        = NULL;
+        char *host            = NULL;
+        char errmsg[PATH_MAX] = "";
+
+        GF_ASSERT (slave_ip);
+        GF_ASSERT (*slave_ip);
+
+        /* Checking if hostname has user specified */
+        host = strstr (*slave_ip, "@");
+        if (!host) {
+                gf_log ("", GF_LOG_DEBUG, "No username provided.");
+                ret = 0;
+                goto out;
+        } else {
+                /* Moving the host past the '@' and checking if the
+                 * actual hostname also has '@' */
+                host++;
+                if (strstr (host, "@")) {
+                        gf_log ("", GF_LOG_DEBUG, "host = %s", host);
+                        ret = snprintf (errmsg, sizeof(errmsg) - 1,
+                                        "Invalid Hostname (%s).", host);
+                        errmsg[ret] = '\0';
+                        gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                        ret = -1;
+                        if (op_errstr)
+                                *op_errstr = gf_strdup (errmsg);
+                        goto out;
+                }
+
+                /* Fetching the username and hostname
+                 * and checking if the username is non-root */
+                username = strtok_r (*slave_ip, "@", &save_ptr);
+                tmp = strtok_r (NULL, "@", &save_ptr);
+                if (strcmp (username, "root")) {
+                        ret = snprintf (errmsg, sizeof(errmsg) - 1,
+                                        "Non-root username (%s@%s) not allowed.",
+                                        username, tmp);
+                        errmsg[ret] = '\0';
+                        if (op_errstr)
+                                *op_errstr = gf_strdup (errmsg);
+                        gf_log ("", GF_LOG_ERROR,
+                                "Non-Root username not allowed.");
+                        ret = -1;
+                        goto out;
+                }
+
+                *slave_ip = gf_strdup (tmp);
+                if (!*slave_ip) {
+                        gf_log ("", GF_LOG_ERROR, "Out of memory");
+                        ret = -1;
+                        goto out;
+                }
+        }
+
+        ret = 0;
+out:
+        gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
+        return ret;
+}
+
+int
+glusterd_op_stage_gsync_create (dict_t *dict, char **op_errstr)
+{
+        char               *down_peerstr              = NULL;
+        char               *slave                     = NULL;
+        char               *volname                   = NULL;
+        char               *host_uuid                 = NULL;
+        char               *statefile                 = NULL;
+        char               *slave_ip                  = NULL;
+        char               *slave_vol                 = NULL;
+        char               *conf_path                 = NULL;
+        char                errmsg[PATH_MAX]          = "";
+        char                common_pem_file[PATH_MAX] = "";
+        char                hook_script[PATH_MAX]     = "";
+        char                uuid_str [64]             = "";
+        int                 ret                       = -1;
+        int                 is_pem_push               = -1;
+        gf_boolean_t        is_force                  = -1;
+        gf_boolean_t        is_force_blocker          = -1;
+        gf_boolean_t        exists                    = _gf_false;
+        glusterd_conf_t    *conf                      = NULL;
+        glusterd_volinfo_t *volinfo                   = NULL;
+        struct stat         stbuf                     = {0,};
+        xlator_t           *this                      = NULL;
+
+        this = THIS;
+        GF_ASSERT (this);
+        conf = this->private;
+        GF_ASSERT (conf);
+
+        ret = glusterd_op_gsync_args_get (dict, op_errstr, &volname,
+                                          &slave, &host_uuid);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "Unable to fetch arguments");
+                gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
+                return -1;
+        }
+
+        if (conf->op_version < 2) {
+                gf_log ("", GF_LOG_ERROR, "Op Version not supported.");
+                snprintf (errmsg, sizeof(errmsg), "One or more nodes do not"
+                          " support the required op version.");
+                *op_errstr = gf_strdup (errmsg);
+                ret = -1;
+                goto out;
+        }
+
+        exists = glusterd_check_volume_exists (volname);
+        ret = glusterd_volinfo_find (volname, &volinfo);
+        if ((ret) || (!exists)) {
+                gf_log ("", GF_LOG_WARNING, "volume name does not exist");
+                snprintf (errmsg, sizeof(errmsg), "Volume name %s does not"
+                          " exist", volname);
+                *op_errstr = gf_strdup (errmsg);
+                gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
+                return -1;
+        }
+
+        ret = glusterd_get_slave_details_confpath (volinfo, dict, &slave_ip,
+                                                   &slave_vol, &conf_path,
+                                                   op_errstr);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to fetch slave or confpath details.");
+                ret = -1;
+                goto out;
+        }
+
+        is_force = dict_get_str_boolean (dict, "force", _gf_false);
+
+        uuid_utoa_r (MY_UUID, uuid_str);
+        if (!strcmp (uuid_str, host_uuid)) {
+                ret = glusterd_are_vol_all_peers_up (volinfo,
+                                                     &conf->peers,
+                                                     &down_peerstr);
+                if ((ret == _gf_false) && !is_force) {
+                        snprintf (errmsg, sizeof (errmsg), "Peer %s,"
+                                  " which is a part of %s volume, is"
+                                  " down. Please bring up the peer and"
+                                  " retry.", down_peerstr,
+                                  volinfo->volname);
+                        gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                        *op_errstr = gf_strdup (errmsg);
+                        GF_FREE (down_peerstr);
+                        down_peerstr = NULL;
+                        gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
+                        return -1;
+                } else if (ret == _gf_false) {
+                        gf_log ("", GF_LOG_INFO, "Peer %s,"
+                                " which is a part of %s volume, is"
+                                " down. Force creating geo-rep session."
+                                " On bringing up the peer, re-run"
+                                " \"gluster system:: execute"
+                                " gsec_create\" and \"gluster volume"
+                                " geo-replication %s %s create push-pem"
+                                " force\"", down_peerstr, volinfo->volname,
+                                volinfo->volname, slave);
+                }
+
+                /* Checking if slave host is pingable, has proper passwordless
+                 * ssh login setup, slave volume is created, slave vol is empty,
+                 * and if it has enough memory and bypass in case of force if
+                 * the error is not a force blocker */
+                ret = glusterd_verify_slave (volname, slave_ip, slave_vol,
+                                             op_errstr, &is_force_blocker);
+                if (ret) {
+                        if (is_force && !is_force_blocker) {
+                                gf_log ("", GF_LOG_INFO, "%s is not a valid slave"
+                                        " volume. Error: %s. Force creating geo-rep"
+                                        " session.", slave, *op_errstr);
+                        } else {
+                                gf_log ("", GF_LOG_ERROR,
+                                        "%s is not a valid slave volume. Error: %s",
+                                        slave, *op_errstr);
+                                ret = -1;
+                                goto out;
+                        }
+                }
+
+                ret = dict_get_int32 (dict, "push_pem", &is_pem_push);
+                if (!ret && is_pem_push) {
+                        ret = snprintf (common_pem_file,
+                                        sizeof(common_pem_file) - 1,
+                                        "%s"GLUSTERD_COMMON_PEM_PUB_FILE,
+                                        conf->workdir);
+                        common_pem_file[ret] = '\0';
+
+                        ret = snprintf (hook_script, sizeof(hook_script) - 1,
+                                        "%s"GLUSTERD_CREATE_HOOK_SCRIPT,
+                                        conf->workdir);
+                        hook_script[ret] = '\0';
+
+                        ret = lstat (common_pem_file, &stbuf);
+                        if (ret) {
+                                snprintf (errmsg, sizeof (errmsg), "%s"
+                                          " required for push-pem is"
+                                          " not present. Please run"
+                                          " \"gluster system:: execute"
+                                          " gsec_create\"", common_pem_file);
+                                gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                                *op_errstr = gf_strdup (errmsg);
+                                ret = -1;
+                                goto out;
+                        }
+
+                        ret = lstat (hook_script, &stbuf);
+                        if (ret) {
+                                snprintf (errmsg, sizeof (errmsg),
+                                          "The hook-script (%s) required "
+                                          "for push-pem is not present. "
+                                          "Please install the hook-script "
+                                          "and retry", hook_script);
+                                gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                                *op_errstr = gf_strdup (errmsg);
+                                ret = -1;
+                                goto out;
+                        }
+
+                        if (!S_ISREG(stbuf.st_mode)) {
+                                snprintf (errmsg, sizeof (errmsg), "%s"
+                                          " required for push-pem is"
+                                          " not a regular file. Please run"
+                                          " \"gluster system:: execute"
+                                          " gsec_create\"", common_pem_file);
+                                gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                                ret = -1;
+                                goto out;
+                        }
+                }
+        }
+
+        ret = glusterd_get_statefile_name (volinfo, slave, conf_path, &statefile);
+        if (ret) {
+                if (!strstr(slave, "::"))
+                        snprintf (errmsg, sizeof (errmsg),
+                                  "%s is not a valid slave url.", slave);
+                else
+                        snprintf (errmsg, sizeof (errmsg), "Please check gsync "
+                                  "config file. Unable to get statefile's name");
+                gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                ret = -1;
+                goto out;
+        }
+
+        ret = dict_set_str (dict, "statefile", statefile);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to store statefile path");
+                goto out;
+        }
+
+        ret = lstat (statefile, &stbuf);
+        if (!ret && !is_force) {
+                snprintf (errmsg, sizeof (errmsg), "Session between %s"
+                          " and %s is already created.",
+                          volinfo->volname, slave);
+                gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                ret = -1;
+                goto out;
+        } else if (!ret)
+                gf_log ("", GF_LOG_INFO, "Session between %s"
+                        " and %s is already created. Force"
+                        " creating again.", volinfo->volname, slave);
+
+        ret = glusterd_verify_gsyncd_spawn (volinfo->volname, slave);
+        if (ret) {
+                snprintf (errmsg, sizeof (errmsg), "Unable to spawn gsyncd.");
+                gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                goto out;
+        }
+
+        ret = 0;
+out:
+
+        if (ret && errmsg[0] != '\0')
+                *op_errstr = gf_strdup (errmsg);
+
+        gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
+        return ret;
+}
+
+int
 glusterd_op_stage_gsync_set (dict_t *dict, char **op_errstr)
 {
         int                     ret     = 0;
         int                     type    = 0;
         char                    *volname = NULL;
         char                    *slave   = NULL;
+        char                    *slave_ip = NULL;
+        char                    *slave_vol = NULL;
+        char                    *down_peerstr = NULL;
+        char                    *statefile    = NULL;
+        char                    *path_list    = NULL;
+        char                    *conf_path    = NULL;
         gf_boolean_t            exists   = _gf_false;
         glusterd_volinfo_t      *volinfo = NULL;
         char                    errmsg[PATH_MAX] = {0,};
         dict_t                  *ctx = NULL;
+        gf_boolean_t            is_force = 0;
+        gf_boolean_t            is_force_blocker = -1;
+        gf_boolean_t            is_running = _gf_false;
+        uuid_t                  uuid = {0};
+        char                    uuid_str [64] = {0};
+        char                    *host_uuid = NULL;
+        xlator_t                *this = NULL;
+        glusterd_conf_t         *conf = NULL;
+        struct stat             stbuf = {0,};
 
+        this = THIS;
+        GF_ASSERT (this);
+        conf = this->private;
+        GF_ASSERT (conf);
 
         ret = dict_get_int32 (dict, "type", &type);
         if (ret < 0) {
@@ -1151,25 +2097,26 @@ glusterd_op_stage_gsync_set (dict_t *dict, char **op_errstr)
                 goto out;
         }
 
-        switch (type) {
-        case GF_GSYNC_OPTION_TYPE_STATUS:
+        if (type == GF_GSYNC_OPTION_TYPE_STATUS) {
                 ret = glusterd_verify_gsync_status_opts (dict, op_errstr);
-
                 goto out;
-        case GF_GSYNC_OPTION_TYPE_CONFIG:
-                ret = gsync_verify_config_options (dict, op_errstr);
+        }
 
+        ret = glusterd_op_gsync_args_get (dict, op_errstr,
+                                          &volname, &slave, &host_uuid);
+        if (ret)
                 goto out;
 
-        case GF_GSYNC_OPTION_TYPE_ROTATE:
-                /* checks same as status mode */
-                ret = glusterd_verify_gsync_status_opts(dict, op_errstr);
-                goto out;
-        }
+        uuid_utoa_r (MY_UUID, uuid_str);
 
-        ret = glusterd_op_gsync_args_get (dict, op_errstr, &volname, &slave);
-        if (ret)
+        if (conf->op_version < 2) {
+                gf_log ("", GF_LOG_ERROR, "Op Version not supported.");
+                snprintf (errmsg, sizeof(errmsg), "One or more nodes do not"
+                          " support the required op version.");
+               *op_errstr = gf_strdup (errmsg);
+                ret = -1;
                 goto out;
+        }
 
         exists = glusterd_check_volume_exists (volname);
         ret = glusterd_volinfo_find (volname, &volinfo);
@@ -1182,12 +2129,96 @@ glusterd_op_stage_gsync_set (dict_t *dict, char **op_errstr)
                 goto out;
         }
 
+        ret = glusterd_get_slave_details_confpath (volinfo, dict, &slave_ip,
+                                                   &slave_vol, &conf_path,
+                                                   op_errstr);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to fetch slave or confpath details.");
+                ret = -1;
+                goto out;
+        }
+
+        ret = glusterd_get_statefile_name (volinfo, slave, conf_path, &statefile);
+        if (ret) {
+                /* Checking if slave host is pingable, has proper passwordless
+                 * ssh login setup */
+                ret = glusterd_verify_slave (volname, slave_ip, slave_vol,
+                                             op_errstr, &is_force_blocker);
+                if (ret) {
+                        gf_log ("", GF_LOG_ERROR,
+                                "%s is not a valid slave volume. Error: %s",
+                                slave, *op_errstr);
+                        goto out;
+                }
+
+                if (!strstr(slave, "::"))
+                        snprintf (errmsg, sizeof (errmsg),
+                                  "%s is not a valid slave url.", slave);
+                else
+                        snprintf (errmsg, sizeof (errmsg),
+                                  "Unable to get statefile's name");
+                gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                ret = -1;
+                goto out;
+        }
+
+        ret = dict_set_str (dict, "statefile", statefile);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to store statefile path");
+                goto out;
+        }
+
+        is_force = dict_get_str_boolean (dict, "force", _gf_false);
+
+        /* Allowing stop force to bypass the statefile check
+         * as this command acts as a fail safe method to stop geo-rep
+         * session. */
+        if ((type == GF_GSYNC_OPTION_TYPE_CONFIG) ||
+            ((type == GF_GSYNC_OPTION_TYPE_STOP) && !is_force) ||
+            (type == GF_GSYNC_OPTION_TYPE_DELETE)) {
+                ret = lstat (statefile, &stbuf);
+                if (ret) {
+                        snprintf (errmsg, sizeof(errmsg), "Geo-replication"
+                                  " session between %s and %s does not exist.",
+                                  volinfo->volname, slave);
+                        gf_log ("", GF_LOG_ERROR, "%s. statefile = %s",
+                                errmsg, statefile);
+                        *op_errstr = gf_strdup (errmsg);
+                        ret = -1;
+                        goto out;
+                }
+        }
+
+        /* Check if all peers that are a part of the volume are up or not */
+        if ((type == GF_GSYNC_OPTION_TYPE_DELETE) ||
+            ((type == GF_GSYNC_OPTION_TYPE_STOP) && !is_force)) {
+                if (!strcmp (uuid_str, host_uuid)) {
+                        ret = glusterd_are_vol_all_peers_up (volinfo,
+                                                             &conf->peers,
+                                                             &down_peerstr);
+                        if (ret == _gf_false) {
+                                snprintf (errmsg, sizeof (errmsg), "Peer %s,"
+                                          " which is a part of %s volume, is"
+                                          " down. Please bring up the peer and"
+                                          " retry.", down_peerstr,
+                                          volinfo->volname);
+                                *op_errstr = gf_strdup (errmsg);
+                                ret = -1;
+                                GF_FREE (down_peerstr);
+                                down_peerstr = NULL;
+                                goto out;
+                        }
+                }
+        }
+
         switch (type) {
         case GF_GSYNC_OPTION_TYPE_START:
                 /* don't attempt to start gsync if replace-brick is
                  * in progress */
                 if (glusterd_is_rb_ongoing (volinfo)) {
-                        snprintf (errmsg, sizeof(errmsg),"replace-brick is in"
+                        snprintf (errmsg, sizeof(errmsg), "replace-brick is in"
                                    " progress, not starting geo-replication");
                         *op_errstr = gf_strdup (errmsg);
                         ret = -1;
@@ -1195,16 +2226,18 @@ glusterd_op_stage_gsync_set (dict_t *dict, char **op_errstr)
                 }
 
                 ret = glusterd_op_verify_gsync_start_options (volinfo, slave,
-                                                              op_errstr);
+                                                              conf_path, statefile,
+                                                              op_errstr, is_force);
                 if (ret)
                         goto out;
                 ctx = glusterd_op_get_ctx();
                 if (ctx) {
-                        /*gsyncd does a fuse mount to start the geo-rep session*/
+                        /* gsyncd does a fuse mount to start
+                         * the geo-rep session */
                         if (!glusterd_is_fuse_available ()) {
-                                gf_log ("glusterd", GF_LOG_ERROR, "Unable to open"
-                                        " /dev/fuse (%s), geo-replication start"
-                                        " failed", strerror (errno));
+                                gf_log ("glusterd", GF_LOG_ERROR, "Unable to "
+                                        "open /dev/fuse (%s), geo-replication "
+                                        "start failed", strerror (errno));
                                 snprintf (errmsg, sizeof(errmsg),
                                           "fuse unvailable");
                                 *op_errstr = gf_strdup (errmsg);
@@ -1215,17 +2248,72 @@ glusterd_op_stage_gsync_set (dict_t *dict, char **op_errstr)
                 break;
 
         case GF_GSYNC_OPTION_TYPE_STOP:
-                ret = glusterd_op_verify_gsync_running (volinfo, slave,
-                                                        op_errstr);
+                if (!is_force) {
+                        ret = glusterd_op_verify_gsync_running (volinfo, slave,
+                                                                conf_path,
+                                                                op_errstr);
+                        if (ret) {
+                                ret = glusterd_get_local_brickpaths (volinfo,
+                                                                     &path_list);
+                                if (path_list)
+                                        ret = -1;
+                        }
+                }
+                break;
+
+        case GF_GSYNC_OPTION_TYPE_CONFIG:
+                ret = gsync_verify_config_options (dict, op_errstr, volname);
+                goto out;
+                break;
+
+        case GF_GSYNC_OPTION_TYPE_DELETE:
+                /* Check if the gsync session is still running
+                 * If so ask the user to stop geo-replication first.*/
+                ret = glusterd_gsync_get_uuid (slave, volinfo, uuid);
+                if (ret) {
+                        snprintf (errmsg, sizeof(errmsg), "Geo-replication"
+                                  " session between %s and %s does not exist.",
+                                  volinfo->volname, slave);
+                        gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                        *op_errstr = gf_strdup (errmsg);
+                        ret = -1;
+                        goto out;
+                } else {
+                        ret = glusterd_check_gsync_running_local (volinfo->volname,
+                                                                  slave, conf_path,
+                                                                  &is_running);
+                        if (_gf_true == is_running) {
+                                snprintf (errmsg, sizeof (errmsg), GEOREP
+                                          " session between %s & %s is "
+                                          "still active. Please stop the "
+                                          "session and retry.",
+                                          volinfo->volname, slave);
+                                gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                                *op_errstr = gf_strdup (errmsg);
+                                ret = -1;
+                                goto out;
+                        }
+                }
+
+                ret = glusterd_verify_gsyncd_spawn (volinfo->volname, slave);
+                if (ret) {
+                        snprintf (errmsg, sizeof (errmsg),
+                                  "Unable to spawn gsyncd");
+                        *op_errstr = gf_strdup (errmsg);
+                        gf_log ("", GF_LOG_ERROR, "%s", errmsg);
+                }
+
                 break;
         }
 
 out:
+        gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
         return ret;
 }
 
 static int
-stop_gsync (char *master, char *slave, char **msg)
+stop_gsync (char *master, char *slave, char **msg,
+            char *conf_path, gf_boolean_t is_force)
 {
         int32_t         ret     = 0;
         int             pfd     = -1;
@@ -1237,19 +2325,16 @@ stop_gsync (char *master, char *slave, char **msg)
         GF_ASSERT (THIS);
         GF_ASSERT (THIS->private);
 
-        pfd = gsyncd_getpidfile (master, slave, pidfile);
-        if (pfd == -2) {
+        pfd = gsyncd_getpidfile (master, slave, pidfile, conf_path);
+        if (pfd == -2 && !is_force) {
                 gf_log ("", GF_LOG_ERROR, GEOREP" stop validation "
                         " failed for %s & %s", master, slave);
                 ret = -1;
                 goto out;
         }
-        if (gsync_status_byfd (pfd) == -1) {
+        if (gsync_status_byfd (pfd) == -1 && !is_force) {
                 gf_log ("", GF_LOG_ERROR, "gsyncd b/w %s & %s is not"
                         " running", master, slave);
-                if (msg)
-                        *msg = gf_strdup ("Warning: "GEOREP" session was "
-                                          "defunct at stop time");
                 /* monitor gsyncd already dead */
                 goto out;
         }
@@ -1284,16 +2369,16 @@ stop_gsync (char *master, char *slave, char **msg)
 
 out:
         sys_close (pfd);
+
+        if (is_force)
+                ret = 0;
         return ret;
 }
 
 static int
-glusterd_check_restart_gsync_session (glusterd_volinfo_t *volinfo, char *slave,
-                                      dict_t *resp_dict);
-
-static int
 glusterd_gsync_configure (glusterd_volinfo_t *volinfo, char *slave,
-                          dict_t *dict, dict_t *resp_dict, char **op_errstr)
+                          char *path_list, dict_t *dict,
+                          dict_t *resp_dict, char **op_errstr)
 {
         int32_t         ret     = -1;
         char            *op_name = NULL;
@@ -1302,6 +2387,10 @@ glusterd_gsync_configure (glusterd_volinfo_t *volinfo, char *slave,
         glusterd_conf_t *priv   = NULL;
         char            *subop  = NULL;
         char            *master = NULL;
+        char            *conf_path = NULL;
+        char            *slave_ip  = NULL;
+        char            *slave_vol = NULL;
+        struct stat      stbuf     = {0, };
 
         GF_ASSERT (slave);
         GF_ASSERT (op_errstr);
@@ -1336,10 +2425,17 @@ glusterd_gsync_configure (glusterd_volinfo_t *volinfo, char *slave,
                 goto out;
         }
 
+        ret = dict_get_str (dict, "conf_path", &conf_path);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to fetch conf file path.");
+                goto out;
+        }
+
         master = "";
         runinit (&runner);
         runner_add_args (&runner, GSYNCD_PREFIX"/gsyncd", "-c", NULL);
-        runner_argprintf (&runner, "%s/"GSYNC_CONF, priv->workdir);
+        runner_argprintf (&runner, "%s", conf_path);
         if (volinfo) {
                 master = volinfo->volname;
                 runner_argprintf (&runner, ":%s", master);
@@ -1362,13 +2458,46 @@ glusterd_gsync_configure (glusterd_volinfo_t *volinfo, char *slave,
 
                 goto out;
         }
+
+        if (!strcmp (op_name, "state_file")) {
+
+                ret = lstat (op_value, &stbuf);
+                if (ret) {
+                        ret = dict_get_str (dict, "slave_ip", &slave_ip);
+                        if (ret) {
+                                gf_log ("", GF_LOG_ERROR,
+                                        "Unable to fetch slave IP.");
+                                goto out;
+                        }
+
+                        ret = dict_get_str (dict, "slave_vol", &slave_vol);
+                        if (ret) {
+                                gf_log ("", GF_LOG_ERROR,
+                                        "Unable to fetch slave volume name.");
+                                goto out;
+                        }
+
+                        ret = glusterd_create_status_file (volinfo->volname, slave,
+                                                           slave_ip, slave_vol,
+                                                           "Switching Status File");
+                        if (ret || lstat (op_value, &stbuf)) {
+                                gf_log ("", GF_LOG_ERROR, "Unable to create %s"
+                                        ". Error : %s", op_value,
+                                        strerror (errno));
+                                ret = -1;
+                                goto out;
+                        }
+                }
+        }
+
         ret = 0;
         gf_asprintf (op_errstr, "config-%s successful", subop);
 
 out:
         if (!ret && volinfo) {
                 ret = glusterd_check_restart_gsync_session (volinfo, slave,
-                                                            resp_dict);
+                                                            resp_dict, path_list,
+                                                            conf_path, 0);
                 if (ret)
                         *op_errstr = gf_strdup ("internal error");
         }
@@ -1494,13 +2623,14 @@ dict_get_param (dict_t *dict, char *key, char **param)
 }
 
 static int
-glusterd_read_status_file (char *master, char *slave,
-                           dict_t *dict, char *node)
+glusterd_read_status_file (glusterd_volinfo_t *volinfo, char *slave,
+                           char *conf_path, dict_t *dict, char *node)
 {
         glusterd_conf_t *priv = NULL;
         int              ret = 0;
         char            *statefile = NULL;
-        char             buf[1024] = {0, };
+        char            *master    = NULL;
+        char             buf[1024] = "defunct";
         char             nds[1024] = {0, };
         char             mst[1024] = {0, };
         char             slv[1024] = {0, };
@@ -1510,34 +2640,38 @@ glusterd_read_status_file (char *master, char *slave,
         int              gsync_count = 0;
         int              status = 0;
         char *dyn_node = NULL;
+        char *path_list = NULL;
 
         GF_ASSERT (THIS);
         GF_ASSERT (THIS->private);
+        GF_ASSERT (volinfo);
+
+        master = volinfo->volname;
 
         confd = dict_new ();
-        if (!dict)
+        if (!dict) {
+                gf_log ("", GF_LOG_ERROR, "Not able to create dict.");
                 return -1;
+        }
 
         priv = THIS->private;
-        ret = glusterd_gsync_get_config (master, slave, priv->workdir,
+
+        ret = glusterd_gsync_get_config (master, slave, conf_path,
                                          confd);
         if (ret) {
                 gf_log ("", GF_LOG_ERROR, "Unable to get configuration data"
                         "for %s(master), %s(slave)", master, slave);
-                goto out;
+                goto done;
 
         }
 
-        ret = gsync_status (master, slave, &status);
-        if (ret == 0 && status == -1) {
-                strncpy (buf, "defunct", sizeof (buf));
-                goto done;
-        } else if (ret == -1)
-                goto out;
-
         ret = dict_get_param (confd, "state_file", &statefile);
-        if (ret)
-                goto out;
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "Unable to get state_file's name "
+                        "for %s(master), %s(slave). Please check gsync "
+                        "config file.", master, slave);
+                goto done;
+        }
         ret = glusterd_gsync_read_frm_status (statefile, buf, sizeof (buf));
         if (ret) {
                 gf_log ("", GF_LOG_ERROR, "Unable to read the status"
@@ -1545,12 +2679,27 @@ glusterd_read_status_file (char *master, char *slave,
                 strncpy (buf, "defunct", sizeof (buf));
                 goto done;
         }
-        if (strcmp (buf, "OK") != 0)
+
+        ret = gsync_status (master, slave, conf_path, &status);
+        if (ret == 0 && status == -1) {
+                if ((strcmp (buf, "Not Started")) &&
+                    (strcmp (buf, "Stopped")))
+                    strncpy (buf, "defunct", sizeof (buf));
+                goto done;
+        } else if (ret == -1) {
+                gf_log ("", GF_LOG_ERROR, "Unable to get gsync status");
+                goto done;
+        }
+
+        if (strcmp (buf, "Stable") != 0)
                 goto done;
 
         ret = dict_get_param (confd, "state_socket_unencoded", &statefile);
-        if (ret)
-                goto out;
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "Unable to get state_socket_unencoded"
+                        " filepath. Please check gsync config file.");
+                goto done;
+        }
         ret = glusterd_gsync_fetch_status_extra (statefile, buf, sizeof (buf));
         if (ret) {
                 gf_log ("", GF_LOG_ERROR, "Unable to fetch extra status"
@@ -1571,6 +2720,19 @@ glusterd_read_status_file (char *master, char *slave,
         }
 
  done:
+        if ((!strcmp (buf, "defunct")) ||
+            (!strcmp (buf, "Not Started")) ||
+            (!strcmp (buf, "Stopped"))) {
+                ret = glusterd_get_local_brickpaths (volinfo, &path_list);
+                if (!path_list) {
+                        gf_log ("", GF_LOG_DEBUG, "This node not being part of"
+                                " volume should not be running gsyncd. Hence"
+                                " shouldn't display status for this node.");
+                        ret = 0;
+                        goto out;
+                }
+        }
+
         ret = dict_get_int32 (dict, "gsync-count", &gsync_count);
 
         if (ret)
@@ -1621,23 +2783,22 @@ glusterd_read_status_file (char *master, char *slave,
         if (ret)
                 goto out;
 
-        ret = 0;
  out:
         dict_destroy (confd);
 
-        gf_log ("", GF_LOG_DEBUG, "Returning %d ", ret);
-        return ret;
+        return 0;
 }
 
-static int
+int
 glusterd_check_restart_gsync_session (glusterd_volinfo_t *volinfo, char *slave,
-                                      dict_t *resp_dict)
+                                      dict_t *resp_dict, char *path_list,
+                                      char *conf_path, gf_boolean_t is_force)
 {
 
         int                    ret = 0;
-        uuid_t                 uuid = {0, };
         glusterd_conf_t        *priv = NULL;
         char                   *status_msg = NULL;
+        gf_boolean_t           is_running = _gf_false;
 
         GF_ASSERT (volinfo);
         GF_ASSERT (slave);
@@ -1646,18 +2807,22 @@ glusterd_check_restart_gsync_session (glusterd_volinfo_t *volinfo, char *slave,
 
         priv = THIS->private;
 
-        if (glusterd_gsync_get_uuid (slave, volinfo, uuid))
-                /* session does not exist, nothing to do */
+        ret = glusterd_check_gsync_running_local (volinfo->volname,
+                                                  slave, conf_path,
+                                                  &is_running);
+        if (!ret && (_gf_true != is_running))
+                /* gsynd not running, nothing to do */
                 goto out;
-        if (uuid_compare (MY_UUID, uuid) == 0) {
-                ret = stop_gsync (volinfo->volname, slave, &status_msg);
-                if (ret == 0 && status_msg)
-                        ret = dict_set_str (resp_dict, "gsync-status",
-                                            status_msg);
-                if (ret == 0)
-                        ret = glusterd_start_gsync (volinfo, slave,
-                                                    uuid_utoa(MY_UUID), NULL);
-        }
+
+        ret = stop_gsync (volinfo->volname, slave, &status_msg,
+                          conf_path, is_force);
+        if (ret == 0 && status_msg)
+                ret = dict_set_str (resp_dict, "gsync-status",
+                                    status_msg);
+        if (ret == 0)
+                ret = glusterd_start_gsync (volinfo, slave, path_list,
+                                            conf_path, uuid_utoa(MY_UUID),
+                                            NULL);
 
  out:
         gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
@@ -1665,7 +2830,7 @@ glusterd_check_restart_gsync_session (glusterd_volinfo_t *volinfo, char *slave,
 }
 
 static int32_t
-glusterd_marker_create_volfile (glusterd_volinfo_t *volinfo)
+glusterd_marker_changelog_create_volfile (glusterd_volinfo_t *volinfo)
 {
         int32_t          ret     = 0;
 
@@ -1689,58 +2854,82 @@ out:
 }
 
 static int
-glusterd_set_marker_gsync (glusterd_volinfo_t *volinfo)
+glusterd_set_gsync_knob (glusterd_volinfo_t *volinfo, char *key, int *vc)
 {
-        int                      ret     = -1;
-        int                      marker_set = _gf_false;
-        char                    *gsync_status = NULL;
+        int   ret          = -1;
+        int   conf_enabled = _gf_false;
+        char *knob_on      = NULL;
 
         GF_ASSERT (THIS);
         GF_ASSERT (THIS->private);
 
-        marker_set = glusterd_volinfo_get_boolean (volinfo, VKEY_MARKER_XTIME);
-        if (marker_set == -1) {
-                gf_log ("", GF_LOG_ERROR, "failed to get the marker status");
-                ret = -1;
+        conf_enabled = glusterd_volinfo_get_boolean (volinfo, key);
+        if (conf_enabled == -1) {
+                gf_log ("", GF_LOG_ERROR,
+                        "failed to get key %s from volinfo", key);
                 goto out;
         }
 
-        if (marker_set == _gf_false) {
-                gsync_status = gf_strdup ("on");
-                if (gsync_status == NULL) {
+        ret = 0;
+        if (conf_enabled == _gf_false) {
+                *vc = 1;
+                knob_on = gf_strdup ("on");
+                if (knob_on == NULL) {
                         ret = -1;
                         goto out;
                 }
 
                 ret = glusterd_gsync_volinfo_dict_set (volinfo,
-                                                       VKEY_MARKER_XTIME, gsync_status);
-                if (ret < 0)
-                        goto out;
-
-                ret = glusterd_marker_create_volfile (volinfo);
-                if (ret) {
-                        gf_log ("", GF_LOG_ERROR, "Setting dict failed");
-                        goto out;
-                }
+                                                       key, knob_on);
         }
-        ret = 0;
 
-out:
+ out:
         gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
         return ret;
 }
 
+static int
+glusterd_set_gsync_confs (glusterd_volinfo_t *volinfo)
+{
+        int ret             = -1;
+        int volfile_changed = 0;
+
+        ret = glusterd_set_gsync_knob (volinfo,
+                                       VKEY_MARKER_XTIME, &volfile_changed);
+        if (ret)
+                goto out;
 
+        /**
+         * enable ignore-pid-check blindly as it could be needed for
+         * cascading setups.
+         */
+        ret = glusterd_set_gsync_knob (volinfo, VKEY_MARKER_XTIME_FORCE,
+                                       &volfile_changed);
+        if (ret)
+                goto out;
+
+        ret = glusterd_set_gsync_knob (volinfo,
+                                       VKEY_CHANGELOG, &volfile_changed);
+        if (ret)
+                goto out;
 
+        if (volfile_changed)
+                ret = glusterd_marker_changelog_create_volfile (volinfo);
+
+ out:
+        return ret;
+}
 
 static int
 glusterd_get_gsync_status_mst_slv (glusterd_volinfo_t *volinfo,
-                                   char *slave, dict_t *rsp_dict,
-                                   char *node)
+                                   char *slave, char *conf_path,
+                                   dict_t *rsp_dict, char *node)
 {
+        char              *statefile = NULL;
         uuid_t             uuid = {0, };
         glusterd_conf_t    *priv = NULL;
         int                ret = 0;
+        struct stat        stbuf = {0, };
 
         GF_ASSERT (volinfo);
         GF_ASSERT (slave);
@@ -1750,19 +2939,38 @@ glusterd_get_gsync_status_mst_slv (glusterd_volinfo_t *volinfo,
         priv = THIS->private;
 
         ret = glusterd_gsync_get_uuid (slave, volinfo, uuid);
-        if ((ret == 0) && (uuid_compare (MY_UUID, uuid) != 0))
-                goto out;
-
         if (ret) {
-                ret = 0;
                 gf_log ("", GF_LOG_INFO, "geo-replication status %s %s :"
                         "session is not active", volinfo->volname, slave);
-                goto out;
+
+                ret = glusterd_get_statefile_name (volinfo, slave,
+                                                   conf_path, &statefile);
+                if (ret) {
+                        if (!strstr(slave, "::"))
+                                gf_log ("", GF_LOG_INFO,
+                                        "%s is not a valid slave url.", slave);
+                        else
+                                gf_log ("", GF_LOG_INFO, "Unable to get"
+                                        " statefile's name");
+                        ret = 0;
+                        goto out;
+                }
+
+                ret = lstat (statefile, &stbuf);
+                if (ret) {
+                        gf_log ("", GF_LOG_INFO, "%s statefile not present.",
+                                statefile);
+                        ret = 0;
+                        goto out;
+                }
         }
 
-        ret = glusterd_read_status_file (volinfo->volname,
-                                         slave, rsp_dict, node);
- out:
+        ret = glusterd_read_status_file (volinfo, slave, conf_path,
+                                         rsp_dict, node);
+out:
+        if (statefile)
+                GF_FREE (statefile);
+
         gf_log ("", GF_LOG_DEBUG, "Returning with %d", ret);
         return ret;
 }
@@ -1813,6 +3021,7 @@ glusterd_get_gsync_status (dict_t *dict, char **op_errstr, dict_t *rsp_dict)
 {
         char               *slave  = NULL;
         char               *volname = NULL;
+        char               *conf_path = NULL;
         char               errmsg[PATH_MAX] = {0, };
         gf_boolean_t       exists = _gf_false;
         glusterd_volinfo_t *volinfo = NULL;
@@ -1850,7 +3059,14 @@ glusterd_get_gsync_status (dict_t *dict, char **op_errstr, dict_t *rsp_dict)
                 goto out;
         }
 
-        ret = glusterd_get_gsync_status_mst_slv (volinfo, slave,
+        ret = dict_get_str (dict, "conf_path", &conf_path);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to fetch conf file path.");
+                goto out;
+        }
+
+        ret = glusterd_get_gsync_status_mst_slv (volinfo, slave, conf_path,
                                                  rsp_dict, my_hostname);
 
  out:
@@ -1859,300 +3075,406 @@ glusterd_get_gsync_status (dict_t *dict, char **op_errstr, dict_t *rsp_dict)
 }
 
 static int
-glusterd_send_sigstop (pid_t pid)
-{
-        int ret = 0;
-        ret = kill (pid, SIGSTOP);
-        if (ret)
-                gf_log ("", GF_LOG_ERROR, GEOREP"failed to send SIGSTOP signal");
-        return ret;
-}
-
-static int
-glusterd_send_sigcont (pid_t pid)
-{
-        int ret = 0;
-        ret = kill (pid, SIGCONT);
-        if (ret)
-                gf_log ("", GF_LOG_ERROR, GEOREP"failed to send SIGCONT signal");
-        return ret;
-}
-
-/*
- * Log rotations flow is something like this:
- * - Send SIGSTOP to process group (this will stop monitor/worker process
- *   and also the slave if it's local)
- * - Rotate log file for monitor/worker
- * - Rotate log file for slave if it's local
- * - Send SIGCONT to the process group. Monitor wakes up, kills the worker
- *   (this is done in the SIGCONT handler), which results in the termination
- *   of the slave (local/remote). After returning from signal handler,
- *   monitor detects absence of worker and starts it again, which in-turn
- *   starts the slave.
- */
-static int
-glusterd_send_log_rotate_signal (pid_t pid, char *logfile1, char *logfile2)
+glusterd_gsync_delete (glusterd_volinfo_t *volinfo, char *slave, char *slave_ip,
+                       char *slave_vol, char *path_list, dict_t *dict,
+                       dict_t *resp_dict, char **op_errstr)
 {
-        int         ret         = 0;
-        char rlogfile[PATH_MAX] = {0,};
-        time_t      rottime     = 0;
-
-        ret = glusterd_send_sigstop (-pid);
-        rottime = time (NULL);
+        int32_t         ret     = -1;
+        runner_t        runner    = {0,};
+        glusterd_conf_t *priv   = NULL;
+        char            *master = NULL;
+        char            *gl_workdir = NULL;
+        char             geo_rep_dir[PATH_MAX] = "";
+        char            *conf_path = NULL;
 
-        snprintf (rlogfile, sizeof (rlogfile), "%s.%"PRIu64, logfile1,
-                  (uint64_t) rottime);
-        ret = rename (logfile1, rlogfile);
-        if (ret)
-                gf_log ("", GF_LOG_ERROR, "rename failed for geo-rep log file");
+        GF_ASSERT (slave);
+        GF_ASSERT (slave_ip);
+        GF_ASSERT (slave_vol);
+        GF_ASSERT (op_errstr);
+        GF_ASSERT (dict);
+        GF_ASSERT (resp_dict);
 
-        if (!*logfile2) {
-                gf_log ("", GF_LOG_DEBUG, "Slave is not local,"
-                        " skipping rotation");
-                ret = 0;
+        if (THIS)
+                priv = THIS->private;
+        if (priv == NULL) {
+                gf_log ("", GF_LOG_ERROR, "priv of glusterd not present");
+                *op_errstr = gf_strdup ("glusterd defunct");
                 goto out;
         }
 
-        (void) snprintf (rlogfile, sizeof (rlogfile), "%s.%"PRIu64, logfile2,
-                         (uint64_t) rottime);
-        ret = rename (logfile2, rlogfile);
-        if (ret)
-                gf_log ("", GF_LOG_ERROR, "rename failed for geo-rep slave"
-                        " log file");
+        ret = dict_get_str (dict, "conf_path", &conf_path);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to fetch conf file path.");
+                goto out;
+        }
 
- out:
-        ret = glusterd_send_sigcont (-pid);
+        gl_workdir = priv->workdir;
+        master = "";
+        runinit (&runner);
+        runner_add_args  (&runner, GSYNCD_PREFIX"/gsyncd",
+                          "--delete", "-c", NULL);
+        runner_argprintf (&runner, "%s", conf_path);
 
-        return ret;
-}
+        if (volinfo) {
+                master = volinfo->volname;
+                runner_argprintf (&runner, ":%s", master);
+        }
+        runner_add_arg (&runner, slave);
+        runner_redir (&runner, STDOUT_FILENO, RUN_PIPE);
+        synclock_unlock (&priv->big_lock);
+        ret = runner_run (&runner);
+        synclock_lock (&priv->big_lock);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "gsyncd failed to "
+                        "delete session info for %s and %s peers",
+                        master, slave);
 
-static int
-glusterd_get_pid_from_file (char *master, char *slave, pid_t *pid)
-{
-        int ret                =