glusterfs: zerofill support

Add support for a new ZEROFILL fop. Zerofill writes zeroes to a file in
the specified range. This fop will be useful when a whole file needs to
be initialized with zero (could be useful for zero filled VM disk image
provisioning or  during scrubbing of VM disk images).

Client/application can issue this FOP for zeroing out. Gluster server
will zero out required range of bytes ie server offloaded zeroing. In
the absence of this fop,  client/application has to repetitively issue
write (zero) fop to the server, which is very inefficient method because
of the overheads involved in RPC calls  and acknowledgements.

WRITESAME is a  SCSI T10 command that takes a block of data as input and
writes the same data to other blocks and this write is handled
completely within the storage and hence is known as offload . Linux ,now
has support for SCSI WRITESAME command which is exposed to the user in
the form of BLKZEROOUT ioctl.  BD Xlator can exploit BLKZEROOUT ioctl to
implement this fop. Thus zeroing out operations can be completely
offloaded to the storage device , making it highly efficient.

The fop takes two arguments offset and size. It zeroes out 'size' number
of bytes in an opened file starting from 'offset' position.

This patch adds zerofill support to the following areas:
	- libglusterfs
	- io-stats
	- performance/md-cache,open-behind
	- quota
	- cluster/afr,dht,stripe
	- rpc/xdr
	- protocol/client,server
	- io-threads
	- marker
	- storage/posix
	- libgfapi

Client applications can exloit this fop by using glfs_zerofill introduced in
libgfapi.FUSE support to this fop has not been added as there is no system call
for this fop.

Changes from previous version 3:
* Removed redundant memory failure log messages

Changes from previous version 2:
* Rebased and fixed build error

Changes from previous version 1:
* Rebased for latest master

TODO :
     * Add zerofill support to trace xlator
     * Expose zerofill capability as part of gluster volume info

Here is a performance comparison of server offloaded zeofill vs zeroing
out using repeated writes.

[root@llmvm02 remote]# time ./offloaded aakash-test log 20

real	3m34.155s
user	0m0.018s
sys	0m0.040s
[root@llmvm02 remote]# time ./manually aakash-test log 20

real	4m23.043s
user	0m2.197s
sys	0m14.457s
[root@llmvm02 remote]# time ./offloaded aakash-test log 25;

real	4m28.363s
user	0m0.021s
sys	0m0.025s
[root@llmvm02 remote]# time ./manually aakash-test log 25

real	5m34.278s
user	0m2.957s
sys	0m18.808s

The argument log is a file which we want to set for logging purpose and
the third argument is size in GB .

As we can see there is a performance improvement of around 20% with this
fop.

Change-Id: I081159f5f7edde0ddb78169fb4c21c776ec91a18
BUG: 1028673
Signed-off-by: Aakash Lal Das <aakash@linux.vnet.ibm.com>
Signed-off-by: M. Mohan Kumar <mohan@in.ibm.com>
Reviewed-on: http://review.gluster.org/5327
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>

20 files changed, 1318 insertions, 16 deletions

diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index 5f2d3096f66..c1ec69a5505 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -2609,3 +2609,253 @@ out:
 
         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 883faae6cb5..8e93ca44aaa 100644
--- a/xlators/cluster/afr/src/afr-inode-write.h
+++ b/xlators/cluster/afr/src/afr-inode-write.h
@@ -75,4 +75,8 @@ afr_discard (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
 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.c b/xlators/cluster/afr/src/afr.c
index aa8d002209c..c724eb2ae42 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -499,6 +499,7 @@ struct xlator_fops fops = {
         .fentrylk    = afr_fentrylk,
 	.fallocate   = afr_fallocate,
 	.discard     = afr_discard,
+        .zerofill    = afr_zerofill,
 
         /* inode read */
         .access      = afr_access,
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 12dce541cf3..21064db58d9 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -710,6 +710,14 @@ typedef struct _afr_local {
 			size_t len;
 		} discard;
 
+                struct {
+                        off_t offset;
+                        size_t len;
+                        struct iatt prebuf;
+                        struct iatt postbuf;
+                } zerofill;
+
+
         } cont;
 
         struct {
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index da8923e9b24..8c3449f0b3b 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -695,6 +695,8 @@ 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);
diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c
index 9bcd84ae15a..26db8a533c1 100644
--- a/xlators/cluster/dht/src/dht-inode-write.c
+++ b/xlators/cluster/dht/src/dht-inode-write.c
@@ -21,6 +21,7 @@ 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,
@@ -624,6 +625,141 @@ err:
         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,
diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c
index 0349b63a91f..fc0ca2f7735 100644
--- a/xlators/cluster/dht/src/dht.c
+++ b/xlators/cluster/dht/src/dht.c
@@ -72,6 +72,7 @@ struct xlator_fops fops = {
         .fsetattr    = dht_fsetattr,
 	.fallocate   = dht_fallocate,
 	.discard     = dht_discard,
+        .zerofill    = dht_zerofill,
 };
 
 struct xlator_dumpops dumpops = {
diff --git a/xlators/cluster/stripe/src/stripe.c b/xlators/cluster/stripe/src/stripe.c
index 8faaec19673..d366b352c5b 100644
--- a/xlators/cluster/stripe/src/stripe.c
+++ b/xlators/cluster/stripe/src/stripe.c
@@ -4107,6 +4107,191 @@ err:
 }
 
 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;
@@ -5558,6 +5743,7 @@ struct xlator_fops fops = {
         .readdirp       = stripe_readdirp,
 	.fallocate	= stripe_fallocate,
 	.discard	= stripe_discard,
+        .zerofill       = stripe_zerofill,
 };
 
 struct xlator_cbks cbks = {
diff --git a/xlators/debug/io-stats/src/io-stats.c b/xlators/debug/io-stats/src/io-stats.c
index 4f27a2e418f..7fb697ae45d 100644
--- a/xlators/debug/io-stats/src/io-stats.c
+++ b/xlators/debug/io-stats/src/io-stats.c
@@ -1746,6 +1746,16 @@ io_stats_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
 	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,
@@ -2441,6 +2451,18 @@ io_stats_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
 	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,
@@ -2870,6 +2892,7 @@ struct xlator_fops fops = {
         .fsetattr    = io_stats_fsetattr,
 	.fallocate   = io_stats_fallocate,
 	.discard     = io_stats_discard,
+        .zerofill    = io_stats_zerofill,
 };
 
 struct xlator_cbks cbks = {
diff --git a/xlators/features/marker/src/marker.c b/xlators/features/marker/src/marker.c
index 59152db43c7..6a2c856913e 100644
--- a/xlators/features/marker/src/marker.c
+++ b/xlators/features/marker/src/marker.c
@@ -1953,6 +1953,73 @@ err:
         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"
@@ -2778,6 +2845,7 @@ struct xlator_fops fops = {
         .readdirp    = marker_readdirp,
 	.fallocate   = marker_fallocate,
 	.discard     = marker_discard,
+        .zerofill    = marker_zerofill,
 };
 
 struct xlator_cbks cbks = {
diff --git a/xlators/performance/io-cache/src/io-cache.c b/xlators/performance/io-cache/src/io-cache.c
index 68ca4c56515..201777b380e 100644
--- a/xlators/performance/io-cache/src/io-cache.c
+++ b/xlators/performance/io-cache/src/io-cache.c
@@ -1449,6 +1449,31 @@ ioc_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
        return 0;
 }
 
+static int32_t
+ioc_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *pre,
+                struct iatt *post, dict_t *xdata)
+{
+        STACK_UNWIND_STRICT(zerofill, frame, op_ret,
+                            op_errno, pre, post, xdata);
+        return 0;
+}
+
+static int32_t
+ioc_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata)
+{
+        uint64_t ioc_inode = 0;
+
+        inode_ctx_get (fd->inode, this, &ioc_inode);
+
+        if (ioc_inode)
+                ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode);
+
+        STACK_WIND(frame, ioc_zerofill_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+       return 0;
+}
 
 
 int32_t
@@ -2077,6 +2102,7 @@ struct xlator_fops fops = {
 
         .readdirp    = ioc_readdirp,
 	.discard     = ioc_discard,
+        .zerofill    = ioc_zerofill,
 };
 
 
diff --git a/xlators/performance/io-threads/src/io-threads.c b/xlators/performance/io-threads/src/io-threads.c
index a5fcd0300d0..bbcf4ed26ca 100644
--- a/xlators/performance/io-threads/src/io-threads.c
+++ b/xlators/performance/io-threads/src/io-threads.c
@@ -309,6 +309,7 @@ iot_schedule (call_frame_t *frame, xlator_t *this, call_stub_t *stub)
         case GF_FOP_RCHECKSUM:
 	case GF_FOP_FALLOCATE:
 	case GF_FOP_DISCARD:
+        case GF_FOP_ZEROFILL:
                 pri = IOT_PRI_LO;
                 break;
 
@@ -2510,6 +2511,55 @@ out:
 }
 
 int
+iot_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno,
+                  struct iatt *preop, struct iatt *postop, dict_t *xdata)
+{
+        STACK_UNWIND_STRICT (zerofill, frame, op_ret, op_errno, preop, postop,
+                             xdata);
+        return 0;
+}
+
+int
+iot_zerofill_wrapper(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                     off_t offset, size_t len, dict_t *xdata)
+{
+        STACK_WIND (frame, iot_zerofill_cbk, FIRST_CHILD (this),
+                    FIRST_CHILD (this)->fops->zerofill, fd, offset, len, xdata);
+        return 0;
+}
+
+int
+iot_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata)
+{
+        call_stub_t     *stub     = NULL;
+        int              ret      = -1;
+
+        stub = fop_zerofill_stub(frame, iot_zerofill_wrapper, fd,
+                                 offset, len, xdata);
+        if (!stub) {
+                gf_log (this->name, GF_LOG_ERROR, "cannot create zerofill stub"
+                        "(out of memory)");
+                ret = -ENOMEM;
+                goto out;
+        }
+
+        ret = iot_schedule (frame, this, stub);
+
+out:
+        if (ret < 0) {
+                STACK_UNWIND_STRICT (zerofill, frame, -1, -ret, NULL, NULL,
+                                     NULL);
+                if (stub != NULL) {
+                        call_stub_destroy (stub);
+                }
+        }
+        return 0;
+}
+
+
+int
 __iot_workers_scale (iot_conf_t *conf)
 {
         int       scale = 0;
@@ -2840,6 +2890,7 @@ struct xlator_fops fops = {
         .rchecksum   = iot_rchecksum,
 	.fallocate   = iot_fallocate,
 	.discard     = iot_discard,
+        .zerofill    = iot_zerofill,
 };
 
 struct xlator_cbks cbks;
diff --git a/xlators/performance/md-cache/src/md-cache.c b/xlators/performance/md-cache/src/md-cache.c
index 36d81887c7b..3a5b7a5d1ae 100644
--- a/xlators/performance/md-cache/src/md-cache.c
+++ b/xlators/performance/md-cache/src/md-cache.c
@@ -2098,6 +2098,46 @@ int mdc_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
 }
 
 int
+mdc_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)
+{
+        mdc_local_t  *local = NULL;
+
+        local = frame->local;
+
+        if (op_ret != 0)
+                goto out;
+
+        if (!local)
+                goto out;
+
+        mdc_inode_iatt_set_validate(this, local->fd->inode, prebuf, postbuf);
+
+out:
+        MDC_STACK_UNWIND(zerofill, frame, op_ret, op_errno, prebuf, postbuf,
+                         xdata);
+
+        return 0;
+}
+
+int mdc_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                size_t len, dict_t *xdata)
+{
+        mdc_local_t *local;
+
+        local = mdc_local_get(frame);
+        local->fd = fd_ref(fd);
+
+        STACK_WIND(frame, mdc_zerofill_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->zerofill, fd, offset, len,
+                   xdata);
+
+        return 0;
+}
+
+
+int
 mdc_forget (xlator_t *this, inode_t *inode)
 {
         mdc_inode_wipe (this, inode);
@@ -2229,6 +2269,7 @@ struct xlator_fops fops = {
 	.readdir     = mdc_readdir,
 	.fallocate   = mdc_fallocate,
 	.discard     = mdc_discard,
+        .zerofill    = mdc_zerofill,
 };
 
 
diff --git a/xlators/performance/open-behind/src/open-behind.c b/xlators/performance/open-behind/src/open-behind.c
index df4027509a9..7e5b5727872 100644
--- a/xlators/performance/open-behind/src/open-behind.c
+++ b/xlators/performance/open-behind/src/open-behind.c
@@ -720,6 +720,26 @@ err:
 }
 
 int
+ob_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           size_t len, dict_t *xdata)
+{
+        call_stub_t *stub;
+
+        stub = fop_zerofill_stub(frame, default_zerofill_resume, fd,
+                                 offset, len, xdata);
+        if (!stub)
+                goto err;
+
+        open_and_resume(this, fd, stub);
+
+        return 0;
+err:
+        STACK_UNWIND_STRICT(zerofill, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+}
+
+
+int
 ob_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
 	   dict_t *xdata)
 {
@@ -946,6 +966,7 @@ struct xlator_fops fops = {
 	.fsetattr    = ob_fsetattr,
 	.fallocate   = ob_fallocate,
 	.discard     = ob_discard,
+        .zerofill    = ob_zerofill,
 	.unlink      = ob_unlink,
 	.rename      = ob_rename,
 	.lk          = ob_lk,
diff --git a/xlators/performance/read-ahead/src/read-ahead.c b/xlators/performance/read-ahead/src/read-ahead.c
index 241fa477fda..069ab1f1a91 100644
--- a/xlators/performance/read-ahead/src/read-ahead.c
+++ b/xlators/performance/read-ahead/src/read-ahead.c
@@ -993,6 +993,57 @@ unwind:
 }
 
 int
+ra_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)
+{
+        GF_ASSERT (frame);
+
+        STACK_UNWIND_STRICT (zerofill, frame, op_ret, op_errno, prebuf,
+                             postbuf, xdata);
+        return 0;
+}
+
+static int
+ra_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             size_t len, dict_t *xdata)
+{
+        ra_file_t *file    = NULL;
+        fd_t      *iter_fd = NULL;
+        inode_t   *inode   = NULL;
+        uint64_t  tmp_file = 0;
+        int32_t   op_errno = EINVAL;
+
+        GF_ASSERT (frame);
+        GF_VALIDATE_OR_GOTO (frame->this->name, this, unwind);
+        GF_VALIDATE_OR_GOTO (frame->this->name, fd, unwind);
+
+        inode = fd->inode;
+
+        LOCK (&inode->lock);
+        {
+                list_for_each_entry (iter_fd, &inode->fd_list, inode_list) {
+                        fd_ctx_get (iter_fd, this, &tmp_file);
+                        file = (ra_file_t *)(long)tmp_file;
+                        if (!file)
+                                continue;
+
+                        flush_region(frame, file, offset, len, 1);
+                }
+        }
+        UNLOCK (&inode->lock);
+
+        STACK_WIND (frame, ra_zerofill_cbk, FIRST_CHILD (this),
+                    FIRST_CHILD (this)->fops->zerofill, fd,
+                    offset, len, xdata);
+        return 0;
+
+unwind:
+        STACK_UNWIND_STRICT (zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
+int
 ra_priv_dump (xlator_t *this)
 {
         ra_conf_t       *conf                           = NULL;
@@ -1173,6 +1224,7 @@ struct xlator_fops fops = {
         .ftruncate   = ra_ftruncate,
         .fstat       = ra_fstat,
 	.discard     = ra_discard,
+        .zerofill    = ra_zerofill,
 };
 
 struct xlator_cbks cbks = {
diff --git a/xlators/protocol/client/src/client-rpc-fops.c b/xlators/protocol/client/src/client-rpc-fops.c
index 38f4391145d..6355450c393 100644
--- a/xlators/protocol/client/src/client-rpc-fops.c
+++ b/xlators/protocol/client/src/client-rpc-fops.c
@@ -2048,6 +2048,62 @@ out:
 }
 
 int
+client3_3_zerofill_cbk(struct rpc_req *req, struct iovec *iov, int count,
+                      void *myframe)
+{
+        call_frame_t    *frame         = NULL;
+        gfs3_zerofill_rsp rsp          = {0,};
+        struct iatt      prestat       = {0,};
+        struct iatt      poststat      = {0,};
+        int              ret           = 0;
+        xlator_t *this                 = NULL;
+        dict_t  *xdata                 = NULL;
+
+        this = THIS;
+
+        frame = myframe;
+
+        if (-1 == req->rpc_status) {
+                rsp.op_ret   = -1;
+                rsp.op_errno = ENOTCONN;
+                goto out;
+        }
+        ret = xdr_to_generic(*iov, &rsp, (xdrproc_t) xdr_gfs3_zerofill_rsp);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR, "XDR decoding failed");
+                rsp.op_ret   = -1;
+                rsp.op_errno = EINVAL;
+                goto out;
+        }
+
+        if (-1 != rsp.op_ret) {
+                gf_stat_to_iatt (&rsp.statpre, &prestat);
+                gf_stat_to_iatt (&rsp.statpost, &poststat);
+        }
+
+        GF_PROTOCOL_DICT_UNSERIALIZE (this, xdata, (rsp.xdata.xdata_val),
+                                      (rsp.xdata.xdata_len), ret,
+                                      rsp.op_errno, out);
+
+out:
+        if (rsp.op_ret == -1) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "remote operation failed: %s",
+                        strerror (gf_error_to_errno (rsp.op_errno)));
+        }
+        CLIENT_STACK_UNWIND (zerofill, frame, rsp.op_ret,
+                             gf_error_to_errno (rsp.op_errno), &prestat,
+                             &poststat, xdata);
+
+        free (rsp.xdata.xdata_val);
+
+        if (xdata)
+                dict_unref (xdata);
+
+        return 0;
+}
+
+int
 client3_3_setattr_cbk (struct rpc_req *req, struct iovec *iov, int count,
                        void *myframe)
 {
@@ -5987,6 +6043,50 @@ unwind:
         return 0;
 }
 
+int32_t
+client3_3_zerofill(call_frame_t *frame, xlator_t *this, void *data)
+{
+        clnt_args_t       *args        = NULL;
+        int64_t            remote_fd   = -1;
+        clnt_conf_t       *conf        = NULL;
+        gfs3_zerofill_req   req        = {{0},};
+        int                op_errno    = ESTALE;
+        int                ret         = 0;
+
+        if (!frame || !this || !data)
+                goto unwind;
+
+        args = data;
+        conf = this->private;
+
+        CLIENT_GET_REMOTE_FD (this, args->fd, DEFAULT_REMOTE_FD,
+                              remote_fd, op_errno, unwind);
+
+        req.fd = remote_fd;
+        req.offset = args->offset;
+        req.size = args->size;
+        memcpy(req.gfid, args->fd->inode->gfid, 16);
+
+        GF_PROTOCOL_DICT_SERIALIZE (this, args->xdata, (&req.xdata.xdata_val),
+                                    req.xdata.xdata_len, op_errno, unwind);
+
+        ret = client_submit_request(this, &req, frame, conf->fops,
+                                    GFS3_OP_ZEROFILL, client3_3_zerofill_cbk,
+                                    NULL, NULL, 0, NULL, 0, NULL,
+                                    (xdrproc_t) xdr_gfs3_zerofill_req);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING, "failed to send the fop");
+
+        GF_FREE (req.xdata.xdata_val);
+
+        return 0;
+unwind:
+        CLIENT_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+        GF_FREE (req.xdata.xdata_val);
+
+        return 0;
+}
+
 /* Table Specific to FOPS */
 
 
@@ -6034,6 +6134,7 @@ rpc_clnt_procedure_t clnt3_3_fop_actors[GF_FOP_MAXVALUE] = {
         [GF_FOP_READDIRP]    = { "READDIRP",    client3_3_readdirp },
 	[GF_FOP_FALLOCATE]   = { "FALLOCATE",	client3_3_fallocate },
 	[GF_FOP_DISCARD]     = { "DISCARD",	client3_3_discard },
+        [GF_FOP_ZEROFILL]    = { "ZEROFILL",    client3_3_zerofill},
         [GF_FOP_RELEASE]     = { "RELEASE",     client3_3_release },
         [GF_FOP_RELEASEDIR]  = { "RELEASEDIR",  client3_3_releasedir },
         [GF_FOP_GETSPEC]     = { "GETSPEC",     client3_getspec },
@@ -6088,6 +6189,8 @@ char *clnt3_3_fop_names[GFS3_OP_MAXVALUE] = {
         [GFS3_OP_FREMOVEXATTR] = "FREMOVEXATTR",
 	[GFS3_OP_FALLOCATE]   = "FALLOCATE",
 	[GFS3_OP_DISCARD]     = "DISCARD",
+        [GFS3_OP_ZEROFILL]    = "ZEROFILL",
+
 };
 
 rpc_clnt_prog_t clnt3_3_fop_prog = {
diff --git a/xlators/protocol/client/src/client.c b/xlators/protocol/client/src/client.c
index 7703c6e8fba..1f7d13ea452 100644
--- a/xlators/protocol/client/src/client.c
+++ b/xlators/protocol/client/src/client.c
@@ -2031,6 +2031,42 @@ out:
 }
 
 int32_t
+client_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               size_t len, dict_t *xdata)
+{
+        int          ret              = -1;
+        clnt_conf_t *conf             = NULL;
+        rpc_clnt_procedure_t *proc    = NULL;
+        clnt_args_t  args             = {0,};
+
+        conf = this->private;
+        if (!conf || !conf->fops)
+                goto out;
+
+        args.fd = fd;
+        args.offset = offset;
+        args.size = len;
+        args.xdata = xdata;
+
+        proc = &conf->fops->proctable[GF_FOP_ZEROFILL];
+        if (!proc) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "rpc procedure not found for %s",
+                        gf_fop_list[GF_FOP_ZEROFILL]);
+                goto out;
+        }
+        if (proc->fn)
+                ret = proc->fn (frame, this, &args);
+out:
+        if (ret)
+                STACK_UNWIND_STRICT(zerofill, frame, -1, ENOTCONN,
+                                    NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+int32_t
 client_getspec (call_frame_t *frame, xlator_t *this, const char *key,
                 int32_t flags)
 {
@@ -2749,6 +2785,7 @@ struct xlator_fops fops = {
         .fsetattr    = client_fsetattr,
 	.fallocate   = client_fallocate,
 	.discard     = client_discard,
+        .zerofill    = client_zerofill,
         .getspec     = client_getspec,
 };
 
diff --git a/xlators/protocol/server/src/server-rpc-fops.c b/xlators/protocol/server/src/server-rpc-fops.c
index 59e808b2f14..d2da9aa769c 100644
--- a/xlators/protocol/server/src/server-rpc-fops.c
+++ b/xlators/protocol/server/src/server-rpc-fops.c
@@ -1993,6 +1993,46 @@ out:
         return 0;
 }
 
+int
+server_zerofill_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)
+{
+        gfs3_zerofill_rsp  rsp    = {0,};
+        server_state_t    *state  = NULL;
+        rpcsvc_request_t  *req    = NULL;
+
+        req = frame->local;
+        state  = CALL_STATE (frame);
+
+        GF_PROTOCOL_DICT_SERIALIZE (this, xdata, (&rsp.xdata.xdata_val),
+                                    rsp.xdata.xdata_len, op_errno, out);
+
+        if (op_ret) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "%"PRId64": ZEROFILL%"PRId64" (%s) ==> (%s)",
+                        frame->root->unique, state->resolve.fd_no,
+                        uuid_utoa (state->resolve.gfid),
+                        strerror (op_errno));
+                goto out;
+        }
+
+        gf_stat_from_iatt (&rsp.statpre, statpre);
+        gf_stat_from_iatt (&rsp.statpost, statpost);
+
+out:
+        rsp.op_ret    = op_ret;
+        rsp.op_errno  = gf_errno_to_error (op_errno);
+
+        server_submit_reply(frame, req, &rsp, NULL, 0, NULL,
+                            (xdrproc_t) xdr_gfs3_zerofill_rsp);
+
+        GF_FREE (rsp.xdata.xdata_val);
+
+        return 0;
+}
+
+
 /* Resume function section */
 
 int
@@ -3019,6 +3059,28 @@ err:
         return 0;
 }
 
+int
+server_zerofill_resume (call_frame_t *frame, xlator_t *bound_xl)
+{
+        server_state_t *state = NULL;
+
+        state = CALL_STATE (frame);
+
+        if (state->resolve.op_ret != 0)
+                goto err;
+
+        STACK_WIND (frame, server_zerofill_cbk,
+                    bound_xl, bound_xl->fops->zerofill,
+                    state->fd, state->offset, state->size, state->xdata);
+        return 0;
+err:
+        server_zerofill_cbk(frame, NULL, frame->this, state->resolve.op_ret,
+                           state->resolve.op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
 
 /* Fop section */
 
@@ -3322,6 +3384,65 @@ out:
 
 
 int
+server3_3_zerofill(rpcsvc_request_t *req)
+{
+        server_state_t       *state      = NULL;
+        call_frame_t         *frame      = NULL;
+        gfs3_zerofill_req     args       = {{0},};
+        int                   ret        = -1;
+        int                   op_errno   = 0;
+
+        if (!req)
+                return ret;
+
+        ret = xdr_to_generic (req->msg[0], &args,
+                              (xdrproc_t)xdr_gfs3_zerofill_req);
+        if (ret < 0) {
+                /*failed to decode msg*/;
+                req->rpc_err = GARBAGE_ARGS;
+                goto out;
+        }
+
+        frame = get_frame_from_request (req);
+        if (!frame) {
+                /* something wrong, mostly insufficient memory*/
+                req->rpc_err = GARBAGE_ARGS; /* TODO */
+                goto out;
+        }
+        frame->root->op = GF_FOP_ZEROFILL;
+
+        state = CALL_STATE (frame);
+        if (!frame->root->client->bound_xl) {
+                /* auth failure, request on subvolume without setvolume */
+                req->rpc_err = GARBAGE_ARGS;
+                goto out;
+        }
+
+        state->resolve.type   = RESOLVE_MUST;
+        state->resolve.fd_no  = args.fd;
+
+        state->offset = args.offset;
+        state->size = args.size;
+        memcpy(state->resolve.gfid, args.gfid, 16);
+
+        GF_PROTOCOL_DICT_UNSERIALIZE (frame->root->client->bound_xl, state->xdata,
+                                      (args.xdata.xdata_val),
+                                      (args.xdata.xdata_len), ret,
+                                      op_errno, out);
+
+        ret = 0;
+        resolve_and_resume (frame, server_zerofill_resume);
+
+out:
+        free (args.xdata.xdata_val);
+
+        if (op_errno)
+                req->rpc_err = GARBAGE_ARGS;
+
+        return ret;
+}
+
+int
 server3_3_readlink (rpcsvc_request_t *req)
 {
         server_state_t    *state                 = NULL;
@@ -6040,6 +6161,7 @@ rpcsvc_actor_t glusterfs3_3_fop_actors[] = {
         [GFS3_OP_FREMOVEXATTR] = {"FREMOVEXATTR", GFS3_OP_FREMOVEXATTR, server3_3_fremovexattr, NULL, 0, DRC_NA},
         [GFS3_OP_FALLOCATE]    = {"FALLOCATE",    GFS3_OP_FALLOCATE,    server3_3_fallocate,    NULL, 0, DRC_NA},
         [GFS3_OP_DISCARD]      = {"DISCARD",      GFS3_OP_DISCARD,      server3_3_discard,      NULL, 0, DRC_NA},
+        [GFS3_OP_ZEROFILL]    =  {"ZEROFILL",     GFS3_OP_ZEROFILL,     server3_3_zerofill,     NULL, 0, DRC_NA},
 };
 
 
diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix.c
index 93ece2474f9..fb45c7a6746 100644
--- a/xlators/storage/posix/src/posix.c
+++ b/xlators/storage/posix/src/posix.c
@@ -24,6 +24,7 @@
 #include <ftw.h>
 #include <sys/stat.h>
 #include <signal.h>
+#include <sys/uio.h>
 
 #ifndef GF_BSD_HOST_OS
 #include <alloca.h>
@@ -616,6 +617,166 @@ out:
         return ret;
 }
 
+char*
+_page_aligned_alloc (size_t size, char **aligned_buf)
+{
+        char            *alloc_buf = NULL;
+        char            *buf = NULL;
+
+        alloc_buf = GF_CALLOC (1, (size + ALIGN_SIZE), gf_posix_mt_char);
+        if (!alloc_buf)
+                goto out;
+        /* page aligned buffer */
+        buf = GF_ALIGN_BUF (alloc_buf, ALIGN_SIZE);
+        *aligned_buf = buf;
+out:
+        return alloc_buf;
+}
+
+static int32_t
+_posix_do_zerofill(int fd, off_t offset, size_t len, int o_direct)
+{
+        size_t              num_vect            = 0;
+        int32_t             num_loop            = 1;
+        int32_t             idx                 = 0;
+        int32_t             op_ret              = -1;
+        int32_t             vect_size           = VECTOR_SIZE;
+        size_t              remain              = 0;
+        size_t              extra               = 0;
+        struct iovec       *vector              = NULL;
+        char               *iov_base            = NULL;
+        char               *alloc_buf           = NULL;
+
+        if (len == 0)
+                return 0;
+        if (len < VECTOR_SIZE)
+                vect_size = len;
+
+        num_vect = len / (vect_size);
+        remain = len % vect_size ;
+        if (num_vect > MAX_NO_VECT) {
+                extra = num_vect % MAX_NO_VECT;
+                num_loop = num_vect / MAX_NO_VECT;
+                num_vect = MAX_NO_VECT;
+        }
+
+        vector = GF_CALLOC (num_vect, sizeof(struct iovec),
+                             gf_common_mt_iovec);
+        if (!vector)
+                  return -1;
+        if (o_direct) {
+                alloc_buf = _page_aligned_alloc(vect_size, &iov_base);
+                if (!alloc_buf) {
+                        gf_log ("_posix_do_zerofill", GF_LOG_DEBUG,
+                                 "memory alloc failed, vect_size %d: %s",
+                                  vect_size, strerror(errno));
+                        GF_FREE(vector);
+                        return -1;
+                }
+        } else {
+                iov_base = GF_CALLOC (vect_size, sizeof(char),
+                                        gf_common_mt_char);
+                if (!iov_base) {
+                        GF_FREE(vector);
+                        return -1;
+                 }
+        }
+
+        for (idx = 0; idx < num_vect; idx++) {
+                vector[idx].iov_base = iov_base;
+                vector[idx].iov_len  = vect_size;
+        }
+        lseek(fd, offset, SEEK_SET);
+        for (idx = 0; idx < num_loop; idx++) {
+                op_ret = writev(fd, vector, num_vect);
+                if (op_ret < 0)
+                        goto err;
+        }
+        if (extra) {
+                op_ret = writev(fd, vector, extra);
+                if (op_ret < 0)
+                        goto err;
+        }
+        if (remain) {
+                vector[0].iov_len = remain;
+                op_ret = writev(fd, vector , 1);
+                if (op_ret < 0)
+                        goto err;
+        }
+err:
+        if (o_direct)
+                GF_FREE(alloc_buf);
+        else
+                GF_FREE(iov_base);
+        GF_FREE(vector);
+        return op_ret;
+}
+
+static int32_t
+posix_do_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  off_t offset, size_t len, struct iatt *statpre,
+                  struct iatt *statpost)
+{
+        struct posix_fd *pfd       = NULL;
+        int32_t          ret       = -1;
+
+        DECLARE_OLD_FS_ID_VAR;
+
+        SET_FS_ID (frame->root->uid, frame->root->gid);
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (fd, out);
+
+        ret = posix_fd_ctx_get (fd, this, &pfd);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "pfd is NULL from fd=%p", fd);
+                goto out;
+        }
+
+        ret = posix_fdstat (this, pfd->fd, statpre);
+        if (ret == -1) {
+                ret = -errno;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "pre-operation fstat failed on fd = %p: %s", fd,
+                        strerror (errno));
+                goto out;
+        }
+        ret = _posix_do_zerofill(pfd->fd, offset, len, pfd->flags & O_DIRECT);
+        if (ret < 0) {
+                ret = -errno;
+                gf_log(this->name, GF_LOG_ERROR,
+                       "zerofill failed on fd %d length %ld %s",
+                        pfd->fd, len, strerror(errno));
+                goto out;
+        }
+        if (pfd->flags & (O_SYNC|O_DSYNC)) {
+                ret = fsync (pfd->fd);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "fsync() in writev on fd %d failed: %s",
+                        pfd->fd, strerror (errno));
+                        ret = -errno;
+                        goto out;
+                }
+        }
+
+        ret = posix_fdstat (this, pfd->fd, statpost);
+        if (ret == -1) {
+                ret = -errno;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "post operation fstat failed on fd=%p: %s", fd,
+                        strerror (errno));
+                goto out;
+        }
+
+out:
+        SET_TO_OLD_FS_ID ();
+
+        return ret;
+}
+
 static int32_t
 _posix_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t keep_size,
 		off_t offset, size_t len, dict_t *xdata)
@@ -664,6 +825,28 @@ err:
 
 }
 
+static int32_t
+posix_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                size_t len, dict_t *xdata)
+{
+        int32_t ret                      =  0;
+        struct  iatt statpre             = {0,};
+        struct  iatt statpost            = {0,};
+
+        ret = posix_do_zerofill(frame, this, fd, offset, len,
+                                 &statpre, &statpost);
+        if (ret < 0)
+                goto err;
+
+        STACK_UNWIND_STRICT(zerofill, frame, 0, 0, &statpre, &statpost, NULL);
+        return 0;
+
+err:
+        STACK_UNWIND_STRICT(zerofill, frame, -1, -ret, NULL, NULL, NULL);
+        return 0;
+
+}
+
 int32_t
 posix_opendir (call_frame_t *frame, xlator_t *this,
                loc_t *loc, fd_t *fd, dict_t *xdata)
@@ -2117,22 +2300,6 @@ err:
         return op_ret;
 }
 
-char*
-_page_aligned_alloc (size_t size, char **aligned_buf)
-{
-        char            *alloc_buf = NULL;
-        char            *buf = NULL;
-
-        alloc_buf = GF_CALLOC (1, (size + ALIGN_SIZE), gf_posix_mt_char);
-        if (!alloc_buf)
-                goto out;
-        /* page aligned buffer */
-        buf = GF_ALIGN_BUF (alloc_buf, ALIGN_SIZE);
-        *aligned_buf = buf;
-out:
-        return alloc_buf;
-}
-
 int32_t
 __posix_writev (int fd, struct iovec *vector, int count, off_t startoff,
                 int odirect)
@@ -4938,6 +5105,7 @@ struct xlator_fops fops = {
         .fsetattr    = posix_fsetattr,
 	.fallocate   = _posix_fallocate,
 	.discard     = posix_discard,
+        .zerofill    = posix_zerofill,
 };
 
 struct xlator_cbks cbks = {
diff --git a/xlators/storage/posix/src/posix.h b/xlators/storage/posix/src/posix.h
index a957e18768c..3121db2717e 100644
--- a/xlators/storage/posix/src/posix.h
+++ b/xlators/storage/posix/src/posix.h
@@ -50,6 +50,8 @@
 #include "posix-aio.h"
 #endif
 
+#define VECTOR_SIZE 64 * 1024 /* vector size 64KB*/
+#define MAX_NO_VECT 1024
 /**
  * posix_fd - internal structure common to file and directory fd's
  */