Merge remote-tracking branch 'origin/release-3.8' into merge-3.8

Change-Id: Ib336c2ada491c2d2fcbbbe6865f9eb975a405b36

8 files changed, 141 insertions, 108 deletions

diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c
index bc3a3ee5ca1..549c3c1ba71 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.c
+++ b/xlators/cluster/afr/src/afr-self-heal-common.c
@@ -26,6 +26,8 @@ afr_selfheal_post_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 	local = frame->local;
 
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
 	syncbarrier_wake (&local->barrier);
 
 	return 0;
@@ -39,6 +41,7 @@ afr_selfheal_post_op (call_frame_t *frame, xlator_t *this, inode_t *inode,
 	afr_private_t *priv = NULL;
 	afr_local_t *local = NULL;
 	loc_t loc = {0, };
+        int ret = 0;
 
 	priv = this->private;
 	local = frame->local;
@@ -46,15 +49,20 @@ afr_selfheal_post_op (call_frame_t *frame, xlator_t *this, inode_t *inode,
 	loc.inode = inode_ref (inode);
 	gf_uuid_copy (loc.gfid, inode->gfid);
 
+        local->op_ret = 0;
+
 	STACK_WIND (frame, afr_selfheal_post_op_cbk, priv->children[subvol],
 		    priv->children[subvol]->fops->xattrop, &loc,
 		    GF_XATTROP_ADD_ARRAY, xattr, xdata);
 
 	syncbarrier_wait (&local->barrier, 1);
+        if (local->op_ret < 0)
+                ret = -local->op_errno;
 
         loc_wipe (&loc);
+        local->op_ret = 0;
 
-	return 0;
+	return ret;
 }
 
 int
@@ -78,6 +86,49 @@ afr_check_stale_error (struct afr_reply *replies, afr_private_t *priv)
                 return -op_errno;
 }
 
+int
+afr_sh_generic_fop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+          int op_ret, int op_errno, struct iatt *pre, struct iatt *post,
+          dict_t *xdata)
+{
+        int i = (long) cookie;
+        afr_local_t *local = NULL;
+
+        local = frame->local;
+
+        local->replies[i].valid = 1;
+        local->replies[i].op_ret = op_ret;
+        local->replies[i].op_errno = op_errno;
+        if (pre)
+                local->replies[i].prestat = *pre;
+        if (post)
+                local->replies[i].poststat = *post;
+        if (xdata)
+                local->replies[i].xdata = dict_ref (xdata);
+
+        syncbarrier_wake (&local->barrier);
+
+        return 0;
+}
+
+int
+afr_selfheal_restore_time (call_frame_t *frame, xlator_t *this, inode_t *inode,
+                           int source, unsigned char *healed_sinks,
+                           struct afr_reply *replies)
+{
+        loc_t loc = {0, };
+
+        loc.inode = inode_ref (inode);
+        gf_uuid_copy (loc.gfid, inode->gfid);
+
+        AFR_ONLIST (healed_sinks, frame, afr_sh_generic_fop_cbk, setattr, &loc,
+                    &replies[source].poststat,
+                    (GF_SET_ATTR_ATIME|GF_SET_ATTR_MTIME), NULL);
+
+        loc_wipe (&loc);
+
+        return 0;
+}
 
 dict_t *
 afr_selfheal_output_xattr (xlator_t *this, gf_boolean_t is_full_crawl,
@@ -1937,8 +1988,10 @@ afr_selfheal_newentry_mark (call_frame_t *frame, xlator_t *this, inode_t *inode,
         changelog = afr_mark_pending_changelog (priv, newentry, xattr,
                                             replies[source].poststat.ia_type);
 
-        if (!changelog)
+        if (!changelog) {
+                ret = -ENOMEM;
                 goto out;
+        }
 
         /* Pre-compute how many sources we have, if we made it in here
          * without any sources defined, we are doing a conservative
@@ -1958,7 +2011,8 @@ afr_selfheal_newentry_mark (call_frame_t *frame, xlator_t *this, inode_t *inode,
                 if (!sources[i] && source_count) {
                         continue;
                 }
-                afr_selfheal_post_op (frame, this, inode, i, xattr, NULL);
+                ret |= afr_selfheal_post_op (frame, this, inode, i, xattr,
+                                             NULL);
             }
 out:
         if (changelog)
diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c
index 9c12e433097..98ee8059640 100644
--- a/xlators/cluster/afr/src/afr-self-heal-data.c
+++ b/xlators/cluster/afr/src/afr-self-heal-data.c
@@ -47,33 +47,6 @@ __checksum_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 	return 0;
 }
 
-
-static int
-attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	  int op_ret, int op_errno, struct iatt *pre, struct iatt *post,
-	  dict_t *xdata)
-{
-	int i = (long) cookie;
-	afr_local_t *local = NULL;
-
-	local = frame->local;
-
-	local->replies[i].valid = 1;
-	local->replies[i].op_ret = op_ret;
-	local->replies[i].op_errno = op_errno;
-	if (pre)
-		local->replies[i].prestat = *pre;
-	if (post)
-		local->replies[i].poststat = *post;
-	if (xdata)
-		local->replies[i].xdata = dict_ref (xdata);
-
-	syncbarrier_wake (&local->barrier);
-
-	return 0;
-}
-
-
 static gf_boolean_t
 __afr_can_skip_data_block_heal (call_frame_t *frame, xlator_t *this, fd_t *fd,
                                 int source, unsigned char *healed_sinks,
@@ -307,7 +280,8 @@ afr_selfheal_data_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd,
         if (!priv->ensure_durability)
                 return 0;
 
-	AFR_ONLIST (healed_sinks, frame, attr_cbk, fsync, fd, 0, NULL);
+	AFR_ONLIST (healed_sinks, frame, afr_sh_generic_fop_cbk, fsync, fd, 0,
+                    NULL);
 
 	for (i = 0; i < priv->child_count; i++)
 		if (healed_sinks[i] && local->replies[i].op_ret != 0)
@@ -318,27 +292,6 @@ afr_selfheal_data_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd,
 	return 0;
 }
 
-
-static int
-afr_selfheal_data_restore_time (call_frame_t *frame, xlator_t *this,
-				inode_t *inode, int source,
-				unsigned char *healed_sinks,
-				struct afr_reply *replies)
-{
-	loc_t loc = {0, };
-
-	loc.inode = inode_ref (inode);
-	gf_uuid_copy (loc.gfid, inode->gfid);
-
-	AFR_ONLIST (healed_sinks, frame, attr_cbk, setattr, &loc,
-		    &replies[source].poststat,
-		    (GF_SET_ATTR_ATIME|GF_SET_ATTR_MTIME), NULL);
-
-	loc_wipe (&loc);
-
-	return 0;
-}
-
 static int
 afr_data_self_heal_type_get (afr_private_t *priv, unsigned char *healed_sinks,
                              int source, struct afr_reply *replies)
@@ -443,7 +396,8 @@ __afr_selfheal_truncate_sinks (call_frame_t *frame, xlator_t *this,
                 healed_sinks[ARBITER_BRICK_INDEX] = 0;
         }
 
-	AFR_ONLIST (healed_sinks, frame, attr_cbk, ftruncate, fd, size, NULL);
+	AFR_ONLIST (healed_sinks, frame, afr_sh_generic_fop_cbk, ftruncate, fd,
+                    size, NULL);
 
 	for (i = 0; i < priv->child_count; i++)
 		if (healed_sinks[i] && local->replies[i].op_ret == -1)
@@ -780,7 +734,7 @@ unlock:
 	if (ret)
                 goto out;
 restore_time:
-	afr_selfheal_data_restore_time (frame, this, fd->inode, source,
+	afr_selfheal_restore_time (frame, this, fd->inode, source,
 					healed_sinks, locked_replies);
 
         if (!is_arbiter_the_only_sink) {
diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c
index 12a5940e7cd..09f26690588 100644
--- a/xlators/cluster/afr/src/afr-self-heal-entry.c
+++ b/xlators/cluster/afr/src/afr-self-heal-entry.c
@@ -75,26 +75,29 @@ afr_selfheal_entry_delete (xlator_t *this, inode_t *dir, const char *name,
 
 
 int
-afr_selfheal_recreate_entry (xlator_t *this, int dst, int source, inode_t *dir,
+afr_selfheal_recreate_entry (call_frame_t *frame, int dst, int source,
+                             unsigned char *sources, inode_t *dir,
                              const char *name, inode_t *inode,
-                             struct afr_reply *replies,
-                             unsigned char *newentry)
+                             struct afr_reply *replies)
 {
 	int ret = 0;
 	loc_t loc = {0,};
 	loc_t srcloc = {0,};
+        xlator_t *this = frame->this;
 	afr_private_t *priv = NULL;
 	dict_t *xdata = NULL;
 	struct iatt *iatt = NULL;
 	char *linkname = NULL;
 	mode_t mode = 0;
 	struct iatt newent = {0,};
-	priv = this->private;
+        unsigned char *newentry = NULL;
 
+        priv = this->private;
 	xdata = dict_new();
 	if (!xdata)
 		return -ENOMEM;
 
+        newentry = alloca0 (priv->child_count);
 	loc.parent = inode_ref (dir);
 	gf_uuid_copy (loc.pargfid, dir->gfid);
 	loc.name = name;
@@ -113,6 +116,15 @@ afr_selfheal_recreate_entry (xlator_t *this, int dst, int source, inode_t *dir,
 
 	srcloc.inode = inode_ref (inode);
 	gf_uuid_copy (srcloc.gfid, iatt->ia_gfid);
+        if (iatt->ia_type != IA_IFDIR)
+                ret = syncop_lookup (priv->children[dst], &srcloc, 0, 0, 0, 0);
+        if (iatt->ia_type == IA_IFDIR || ret == -ENOENT || ret == -ESTALE) {
+                newentry[dst] = 1;
+                ret = afr_selfheal_newentry_mark (frame, this, inode, source,
+                                                  replies, sources, newentry);
+                if (ret)
+                        goto out;
+        }
 
 	mode = st_mode_from_ia (iatt->ia_prot, iatt->ia_type);
 
@@ -120,12 +132,9 @@ afr_selfheal_recreate_entry (xlator_t *this, int dst, int source, inode_t *dir,
 	case IA_IFDIR:
 		ret = syncop_mkdir (priv->children[dst], &loc, mode, 0,
                                     xdata, NULL);
-                if (ret == 0)
-                        newentry[dst] = 1;
 		break;
 	case IA_IFLNK:
-		ret = syncop_lookup (priv->children[dst], &srcloc, 0, 0, 0, 0);
-		if (ret == 0) {
+		if (!newentry[dst]) {
 			ret = syncop_link (priv->children[dst], &srcloc, &loc,
 					   &newent, NULL, NULL);
 		} else {
@@ -135,8 +144,6 @@ afr_selfheal_recreate_entry (xlator_t *this, int dst, int source, inode_t *dir,
 				goto out;
 			ret = syncop_symlink (priv->children[dst], &loc,
                                               linkname, NULL, xdata, NULL);
-                        if (ret == 0)
-                                newentry[dst] = 1;
                 }
 		break;
 	default:
@@ -146,10 +153,6 @@ afr_selfheal_recreate_entry (xlator_t *this, int dst, int source, inode_t *dir,
 		ret = syncop_mknod (priv->children[dst], &loc, mode,
                     makedev (ia_major(iatt->ia_rdev), ia_minor (iatt->ia_rdev)),
                     &newent, xdata, NULL);
-		if (ret == 0 && newent.ia_nlink == 1) {
-			/* New entry created. Mark @dst pending on all sources */
-                        newentry[dst] = 1;
-		}
 		break;
 	}
 
@@ -172,12 +175,9 @@ __afr_selfheal_heal_dirent (call_frame_t *frame, xlator_t *this, fd_t *fd,
 	int ret = 0;
 	afr_private_t *priv = NULL;
 	int i = 0;
-	unsigned char *newentry = NULL;
 
 	priv = this->private;
 
-	newentry = alloca0 (priv->child_count);
-
 	if (!replies[source].valid)
 		return -EIO;
 
@@ -200,17 +200,15 @@ __afr_selfheal_heal_dirent (call_frame_t *frame, xlator_t *this, fd_t *fd,
 					   replies[source].poststat.ia_gfid))
 				continue;
 
-			ret = afr_selfheal_recreate_entry (this, i, source,
-							   fd->inode, name, inode,
-							   replies, newentry);
+			ret = afr_selfheal_recreate_entry (frame, i, source,
+                                                           sources, fd->inode,
+                                                           name, inode,
+							   replies);
                 }
 		if (ret < 0)
 			break;
 	}
 
-	if (AFR_COUNT (newentry, priv->child_count))
-		afr_selfheal_newentry_mark (frame, this, inode, source, replies,
-					    sources, newentry);
 	return ret;
 }
 
@@ -278,13 +276,10 @@ __afr_selfheal_merge_dirent (call_frame_t *frame, xlator_t *this, fd_t *fd,
         int             ret       = 0;
         int             i         = 0;
         int             source    = -1;
-        unsigned char  *newentry  = NULL;
         afr_private_t  *priv      = NULL;
 
 	priv = this->private;
 
-	newentry = alloca0 (priv->child_count);
-
 	for (i = 0; i < priv->child_count; i++) {
 		if (replies[i].valid && replies[i].op_ret == 0) {
 			source = i;
@@ -319,14 +314,11 @@ __afr_selfheal_merge_dirent (call_frame_t *frame, xlator_t *this, fd_t *fd,
 		if (replies[i].op_errno != ENOENT)
 			continue;
 
-		ret = afr_selfheal_recreate_entry (this, i, source, fd->inode,
-                                                   name, inode, replies,
-                                                   newentry);
+		ret = afr_selfheal_recreate_entry (frame, i, source, sources,
+                                                   fd->inode, name, inode,
+                                                   replies);
 	}
 
-	if (AFR_COUNT (newentry, priv->child_count))
-		afr_selfheal_newentry_mark (frame, this, inode, source, replies,
-					    sources, newentry);
 	return ret;
 }
 
diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c
index a67717bf9d4..d5d95ef8852 100644
--- a/xlators/cluster/afr/src/afr-self-heal-metadata.c
+++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c
@@ -427,6 +427,13 @@ afr_selfheal_metadata (call_frame_t *frame, xlator_t *this, inode_t *inode)
                 if (ret)
                         goto unlock;
 
+                /* Restore atime/mtime for files that don't need data heal as
+                 * restoring timestamps happens only as a part of data-heal.
+                 */
+                if (!IA_ISREG (locked_replies[source].poststat.ia_type))
+                        afr_selfheal_restore_time (frame, this, inode, source,
+                                                  healed_sinks, locked_replies);
+
                 ret = afr_selfheal_undo_pending (frame, this, inode, sources,
                                                  sinks, healed_sinks,
                                                  undid_pending,
diff --git a/xlators/cluster/afr/src/afr-self-heal-name.c b/xlators/cluster/afr/src/afr-self-heal-name.c
index 3445ecccf9c..b28ce4170f1 100644
--- a/xlators/cluster/afr/src/afr-self-heal-name.c
+++ b/xlators/cluster/afr/src/afr-self-heal-name.c
@@ -109,12 +109,10 @@ __afr_selfheal_name_impunge (call_frame_t *frame, xlator_t *this,
 	int i = 0;
 	afr_private_t *priv = NULL;
         int ret = 0;
-        unsigned char *newentry = NULL;
         unsigned char *sources = NULL;
 
         priv = this->private;
 
-	newentry = alloca0 (priv->child_count);
 	sources = alloca0 (priv->child_count);
 
 	gf_uuid_copy (parent->gfid, pargfid);
@@ -128,15 +126,17 @@ __afr_selfheal_name_impunge (call_frame_t *frame, xlator_t *this,
                         sources[i] = 1;
 			continue;
                 }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (sources[i])
+                        continue;
 
-		ret |= afr_selfheal_recreate_entry (this, i, gfid_idx, parent,
-                                                    bname, inode, replies,
-                                                    newentry);
+		ret |= afr_selfheal_recreate_entry (frame, i, gfid_idx, sources,
+                                                    parent, bname, inode,
+                                                    replies);
 	}
 
-	if (AFR_COUNT (newentry, priv->child_count))
-		afr_selfheal_newentry_mark (frame, this, inode, gfid_idx, replies,
-					    sources, newentry);
 	return ret;
 }
 
@@ -457,8 +457,7 @@ __afr_selfheal_name_do (call_frame_t *frame, xlator_t *this, inode_t *parent,
         }
 
 	ret = __afr_selfheal_name_impunge (frame, this, parent, pargfid,
-                                           bname, inode,
-                                           replies, gfid_idx);
+                                           bname, inode, replies, gfid_idx);
         if (ret == -EIO)
                 ret = -1;
 
diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h
index a33905033cc..0a3d6482ca3 100644
--- a/xlators/cluster/afr/src/afr-self-heal.h
+++ b/xlators/cluster/afr/src/afr-self-heal.h
@@ -166,6 +166,15 @@ afr_selfheal_extract_xattr (xlator_t *this, struct afr_reply *replies,
 			    afr_transaction_type type, int *dirty, int **matrix);
 
 int
+afr_sh_generic_fop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, struct iatt *pre,
+                        struct iatt *post, dict_t *xdata);
+
+int
+afr_selfheal_restore_time (call_frame_t *frame, xlator_t *this, inode_t *inode,
+                           int source, unsigned char *healed_sinks,
+                           struct afr_reply *replies);
+int
 afr_selfheal_undo_pending (call_frame_t *frame, xlator_t *this, inode_t *inode,
 			   unsigned char *sources, unsigned char *sinks,
 			   unsigned char *healed_sinks,
@@ -174,10 +183,10 @@ afr_selfheal_undo_pending (call_frame_t *frame, xlator_t *this, inode_t *inode,
                            unsigned char *locked_on);
 
 int
-afr_selfheal_recreate_entry (xlator_t *this, int dst, int source, inode_t *dir,
-                             const char *name, inode_t *inode,
-                             struct afr_reply *replies,
-                             unsigned char *newentry);
+afr_selfheal_recreate_entry (call_frame_t *frame, int dst, int source,
+                             unsigned char *sources,
+                             inode_t *dir, const char *name, inode_t *inode,
+                             struct afr_reply *replies);
 
 int
 afr_selfheal_post_op (call_frame_t *frame, xlator_t *this, inode_t *inode,
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index a97d03bb055..81f7a8a40e6 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -3929,6 +3929,7 @@ dht_setxattr2 (xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
                 goto err;
 
         local = frame->local;
+        op_errno = local->op_errno;
 
         if (we_are_not_migrating (ret)) {
                 /* This dht xlator is not migrating the file. Unwind and
@@ -3943,7 +3944,6 @@ dht_setxattr2 (xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
         if (subvol == NULL)
                 goto err;
 
-        op_errno = local->op_errno;
 
         local->call_cnt = 2; /* This is the second attempt */
 
@@ -3968,6 +3968,15 @@ err:
 }
 
 int
+dht_nuke_dir_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)
+{
+        STACK_UNWIND_STRICT (setxattr, frame, op_ret, op_errno, NULL);
+        return 0;
+}
+
+int
 dht_nuke_dir (call_frame_t *frame, xlator_t *this, loc_t *loc, data_t *tmp)
 {
         if (!IA_ISDIR(loc->inode->ia_type)) {
@@ -3998,7 +4007,7 @@ dht_nuke_dir (call_frame_t *frame, xlator_t *this, loc_t *loc, data_t *tmp)
          * obscure the fact that we came in via this path instead of a genuine
          * rmdir.  That makes debugging just a tiny bit easier.
          */
-        STACK_WIND (frame, default_rmdir_cbk, this, this->fops->rmdir,
+        STACK_WIND (frame, dht_nuke_dir_cbk, this, this->fops->rmdir,
                     loc, 1, NULL);
 
         return 0;
@@ -4351,10 +4360,11 @@ dht_removexattr2 (xlator_t *this, xlator_t *subvol, call_frame_t *frame,
         dht_local_t *local    = NULL;
         int          op_errno = EINVAL;
 
-        if (!frame || !frame->local || !subvol)
+        if (!frame || !frame->local)
                 goto err;
 
         local = frame->local;
+        op_errno = local->op_errno;
 
         local->call_cnt = 2; /* This is the second attempt */
 
@@ -4369,6 +4379,9 @@ dht_removexattr2 (xlator_t *this, xlator_t *subvol, call_frame_t *frame,
                 return 0;
         }
 
+        if (subvol == NULL)
+                goto err;
+
         if (local->fop == GF_FOP_REMOVEXATTR) {
                 STACK_WIND (frame, dht_file_removexattr_cbk, subvol,
                             subvol->fops->removexattr, &local->loc,
diff --git a/xlators/cluster/ec/src/ec-common.c b/xlators/cluster/ec/src/ec-common.c
index 2b4cec406a9..8a5d0eb74e8 100644
--- a/xlators/cluster/ec/src/ec-common.c
+++ b/xlators/cluster/ec/src/ec-common.c
@@ -93,6 +93,7 @@ void ec_check_status(ec_fop_data_t * fop)
 {
     ec_t * ec = fop->xl->private;
     int32_t partial = 0;
+    char str1[32], str2[32], str3[32], str4[32], str5[32];
 
     if (!ec_fop_needs_heal(fop)) {
         return;
@@ -109,11 +110,15 @@ void ec_check_status(ec_fop_data_t * fop)
 
     gf_msg (fop->xl->name, GF_LOG_WARNING, 0,
             EC_MSG_OP_FAIL_ON_SUBVOLS,
-            "Operation failed on some "
-            "subvolumes (up=%lX, mask=%lX, "
-            "remaining=%lX, good=%lX, bad=%lX)",
-            ec->xl_up, fop->mask, fop->remaining, fop->good,
-            ec->xl_up & ~(fop->remaining | fop->good));
+            "Operation failed on %d of %d subvolumes.(up=%s, mask=%s, "
+            "remaining=%s, good=%s, bad=%s)",
+            gf_bits_count(ec->xl_up & ~(fop->remaining | fop->good)), ec->nodes,
+            ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes),
+            ec_bin(str2, sizeof(str2), fop->mask, ec->nodes),
+            ec_bin(str3, sizeof(str3), fop->remaining, ec->nodes),
+            ec_bin(str4, sizeof(str4), fop->good, ec->nodes),
+            ec_bin(str5, sizeof(str5),
+            ec->xl_up & ~(fop->remaining | fop->good), ec->nodes));
 
     if (fop->use_fd)
     {