cluster/afr: Make afr_fsync a transaction

Change-Id: I713401feb96393f668efb074f2d5b870d19e6fda
BUG: 1548361
Signed-off-by: karthik-us <ksubrahm@redhat.com>

5 files changed, 117 insertions, 164 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index ea3df12b8bb..0052968e0e4 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -3503,169 +3503,6 @@ out:
         return 0;
 }
 
-/* }}} */
-
-
-/* {{{ 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)
-{
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int i = 0;
-        int call_count = -1;
-        int child_index = (long) cookie;
-	int read_subvol = 0;
-	call_stub_t *stub = NULL;
-
-        local = frame->local;
-        priv = this->private;
-
-        LOCK (&frame->lock);
-        {
-                local->replies[child_index].valid = 1;
-                local->replies[child_index].op_ret = op_ret;
-                local->replies[child_index].op_errno = op_errno;
-                if (op_ret == 0) {
-                        if (prebuf)
-                                local->replies[child_index].prestat = *prebuf;
-                        if (postbuf)
-                                local->replies[child_index].poststat = *postbuf;
-                        if (xdata)
-                                local->replies[child_index].xdata =
-                                        dict_ref (xdata);
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                local->op_ret = -1;
-                local->op_errno = afr_final_errno (local, priv);
-	        read_subvol = afr_data_subvol_get (local->inode, this, NULL,
-                                                   local->readable, NULL, NULL);
-                /* Pick a reply that is valid and readable, with a preference
-                 * given to read_subvol. */
-                for (i = 0; i < priv->child_count; i++) {
-                        if (!local->replies[i].valid)
-                                continue;
-                        if (local->replies[i].op_ret != 0)
-                                continue;
-                        if (!local->readable[i])
-                                continue;
-                        local->op_ret = local->replies[i].op_ret;
-                        local->op_errno = local->replies[i].op_errno;
-                        local->cont.inode_wfop.prebuf =
-                                local->replies[i].prestat;
-                        local->cont.inode_wfop.postbuf =
-                                local->replies[i].poststat;
-                        if (local->replies[i].xdata) {
-                                if (local->xdata_rsp)
-                                        dict_unref (local->xdata_rsp);
-                                local->xdata_rsp =
-                                        dict_ref (local->replies[i].xdata);
-                        }
-                        if (i == read_subvol)
-                                break;
-                }
-
-		/* Make a stub out of the frame, and register it
-		   with the waking up post-op. When the call-stub resumes,
-		   we are guaranteed that there was no post-op pending
-		   (i.e changelogs were unset in the server). This is an
-		   essential "guarantee", that fsync() returns only after
-		   completely finishing EVERYTHING, including the delayed
-		   post-op. This guarantee is expected by FUSE graph switching
-		   for example.
-		*/
-		stub = fop_fsync_cbk_stub (frame, afr_fsync_unwind_cbk,
-                                           local->op_ret, local->op_errno,
-                                           &local->cont.inode_wfop.prebuf,
-                                           &local->cont.inode_wfop.postbuf,
-                                           local->xdata_rsp);
-		if (!stub) {
-			AFR_STACK_UNWIND (fsync, frame, -1, ENOMEM, 0, 0, 0);
-			return 0;
-		}
-
-		/* If no new unstable writes happened between the
-		   time we cleared the unstable write witness flag in afr_fsync
-		   and now, calling afr_delayed_changelog_wake_up() should
-		   wake up and skip over the fsync phase and go straight to
-		   afr_changelog_post_op_now()
-		*/
-		afr_delayed_changelog_wake_resume (this, local->fd, stub);
-        }
-
-        return 0;
-}
-
-
-int
-afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
-	   dict_t *xdata)
-{
-	afr_private_t *priv = NULL;
-        afr_local_t *local = NULL;
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_errno = ENOMEM;
-
-	priv = this->private;
-
-	local = AFR_FRAME_INIT (frame, op_errno);
-	if (!local)
-		goto out;
-
-        local->op = GF_FOP_FSYNC;
-	if (!afr_is_consistent_io_possible (local, priv, &op_errno))
-		goto out;
-
-        local->fd = fd_ref (fd);
-
-	if (afr_fd_has_witnessed_unstable_write (this, fd)) {
-		/* don't care. we only wanted to CLEAR the bit */
-	}
-
-	local->inode = inode_ref (fd->inode);
-
-        call_count = local->call_count;
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_fsync_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fsync,
-                                           fd, datasync, xdata);
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-	return 0;
-out:
-	AFR_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
-
-        return 0;
-}
-
-/* }}} */
-
-/* {{{ fsync */
 
 int
 afr_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index 89ec6ee4d61..60c459d8463 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -2542,3 +2542,111 @@ out:
 	AFR_STACK_UNWIND (fxattrop, frame, -1, op_errno, NULL, NULL);
         return 0;
 }
+
+
+int
+afr_fsync_unwind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        local = frame->local;
+
+        main_frame = afr_transaction_detach_fop_frame (frame);
+        if (!main_frame)
+                return 0;
+
+        AFR_STACK_UNWIND (fsync, main_frame, local->op_ret, local->op_errno,
+                          &local->cont.inode_wfop.prebuf,
+                          &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+
+        return 0;
+}
+
+
+int
+afr_fsync_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
+{
+        return __afr_inode_write_cbk (frame, cookie, this, op_ret, op_errno,
+                                      prebuf, postbuf, NULL, xdata);
+}
+
+
+int
+afr_fsync_wind (call_frame_t *frame, xlator_t *this, int subvol)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+
+        local = frame->local;
+        priv = this->private;
+
+        STACK_WIND_COOKIE (frame, afr_fsync_wind_cbk, (void *)(long) subvol,
+                           priv->children[subvol],
+                           priv->children[subvol]->fops->fsync,
+                           local->fd, local->cont.fsync.datasync,
+                           local->xdata_req);
+        return 0;
+}
+
+int
+afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+	   dict_t *xdata)
+{
+        afr_local_t *local = NULL;
+        call_frame_t *transaction_frame = NULL;
+        int ret = -1;
+        int32_t op_errno = ENOMEM;
+
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame)
+                goto out;
+
+	local = AFR_FRAME_INIT (transaction_frame, op_errno);
+	if (!local)
+		goto out;
+
+        if (xdata)
+                local->xdata_req = dict_copy_with_ref (xdata, NULL);
+        else
+                local->xdata_req = dict_new ();
+
+        if (!local->xdata_req)
+                goto out;
+
+        local->fd = fd_ref (fd);
+        ret = afr_set_inode_local (this, local, fd->inode);
+        if (ret)
+                goto out;
+
+        local->op = GF_FOP_FSYNC;
+        local->cont.fsync.datasync = datasync;
+
+	if (afr_fd_has_witnessed_unstable_write (this, fd)) {
+		/* don't care. we only wanted to CLEAR the bit */
+	}
+
+        local->transaction.wind   = afr_fsync_wind;
+        local->transaction.fop    = __afr_txn_write_fop;
+        local->transaction.done   = __afr_txn_write_done;
+        local->transaction.unwind = afr_fsync_unwind;
+
+        local->transaction.main_frame = frame;
+
+        ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
+
+	return 0;
+out:
+	if (transaction_frame)
+		AFR_STACK_DESTROY (transaction_frame);
+
+	AFR_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
diff --git a/xlators/cluster/afr/src/afr-inode-write.h b/xlators/cluster/afr/src/afr-inode-write.h
index e174cc2d610..1e8bb5c12b3 100644
--- a/xlators/cluster/afr/src/afr-inode-write.h
+++ b/xlators/cluster/afr/src/afr-inode-write.h
@@ -87,4 +87,8 @@ afr_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
 int32_t
 afr_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
               gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata);
+
+int
+afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+	   dict_t *xdata);
 #endif /* __INODE_WRITE_H__ */
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index 10e81138aea..3467729c345 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -636,7 +636,6 @@ struct xlator_fops fops = {
         .lk          = afr_lk,
         .flush       = afr_flush,
         .statfs      = afr_statfs,
-        .fsync       = afr_fsync,
         .fsyncdir    = afr_fsyncdir,
         .inodelk     = afr_inodelk,
         .finodelk    = afr_finodelk,
@@ -668,6 +667,7 @@ struct xlator_fops fops = {
         .zerofill    = afr_zerofill,
         .xattrop     = afr_xattrop,
         .fxattrop    = afr_fxattrop,
+        .fsync       = afr_fsync,
 
         /*inode open*/
         .opendir     = afr_opendir,
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index fdda295718c..88d80f07a3c 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -725,6 +725,10 @@ typedef struct _afr_local {
                         gf_seek_what_t what;
                 } seek;
 
+                struct {
+                        int32_t datasync;
+                } fsync;
+
         } cont;
 
         struct {