cluster/dht: assign layout onto missing directories too

The current self-healing algorithm is ignoring missing directories
for assigning new layout. When lookup() is racing against mkdir()
or when self-healing a half-done mkdir(), the layout assignment split
must happen based on the final number of directories, and not the
currently existing number of directories (because we finish mkdir()
of missing directories before hash layout assignment).

Without this fix, concurrent mkdir() and lookup() will step on
each others feet, create a messed up layout on disk, and end up
with different in-memory layouts.

Once two clients have different in-memory layouts, creation of
subdirectory will not arbitrate on the same hashed subvolume and will
result in GFID mismatch of the sub-directory.

Change-Id: Ia47acad67c265060405984c822b4d37512b9dbb3
BUG: 907072
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.org/5849
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Amar Tumballi <amarts@redhat.com>
Reviewed-by: Peter Portante <pportant@redhat.com>
Tested-by: Peter Portante <pportant@redhat.com>

1 files changed, 28 insertions, 4 deletions

diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c
index b220a0e25ac..76ed26e1a72 100644
--- a/xlators/cluster/dht/src/dht-selfheal.c
+++ b/xlators/cluster/dht/src/dht-selfheal.c
@@ -564,9 +564,33 @@ 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 +800,7 @@ 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 +813,7 @@ 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) {