gfapi: glfs_subvol_done should NOT wait for graph migration.

In graph_setup function glfs_subvol_done is called which
is executed in an epoll thread. glfs_lock waits on other
thread to finish graph migration. This can lead to dead lock
if we consume all the epoll threads.

In general any call-back function executed in epoll thread
should not call any blocking call which waits on a network
reply either directly or indirectly, e.g. syncop functions
should not be called in these threads.

As a fix we should not wait for migration in the call-back path.

Change-Id: If96d0689fe1b4d74631e383048cdc30b01690dc2
BUG: 1397754
Signed-off-by: Rajesh Joseph <rjoseph@redhat.com>
Reviewed-on: http://review.gluster.org/15913
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
Smoke: Gluster Build System <jenkins@build.gluster.org>
Reviewed-by: Niels de Vos <ndevos@redhat.com>
CentOS-regression: Gluster Build System <jenkins@build.gluster.org>

1 files changed, 12 insertions, 3 deletions

diff --git a/api/src/glfs-internal.h b/api/src/glfs-internal.h
index fc62bbf027d..a42822420ea 100644
--- a/api/src/glfs-internal.h
+++ b/api/src/glfs-internal.h
@@ -305,17 +305,26 @@ do {                                                                \
   we can give up the mutex during syncop calls so
   that bottom up calls (particularly CHILD_UP notify)
   can do a mutex_lock() on @glfs without deadlocking
-  the filesystem
+  the filesystem.
+
+  All the fops should wait for graph migration to finish
+  before starting the fops. Therefore these functions should
+  call glfs_lock with wait_for_migration as true. But waiting
+  for migration to finish in call-back path can result thread
+  dead-locks. The reason for this is we only have finite
+  number of epoll threads. so if we wait on epoll threads
+  there will not be any thread left to handle outstanding
+  rpc replies.
 */
 static inline int
-glfs_lock (struct glfs *fs)
+glfs_lock (struct glfs *fs, gf_boolean_t wait_for_migration)
 {
 	pthread_mutex_lock (&fs->mutex);
 
 	while (!fs->init)
 		pthread_cond_wait (&fs->cond, &fs->mutex);
 
-	while (fs->migration_in_progress)
+        while (wait_for_migration && fs->migration_in_progress)
 		pthread_cond_wait (&fs->cond, &fs->mutex);
 
 	return 0;