From 17d10b42fc4041442e6cd0bfda45944edea498c6 Mon Sep 17 00:00:00 2001 From: Rajesh Joseph Date: Tue, 22 Nov 2016 22:25:42 +0530 Subject: 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 Reviewed-on: http://review.gluster.org/15913 NetBSD-regression: NetBSD Build System Smoke: Gluster Build System Reviewed-by: Niels de Vos CentOS-regression: Gluster Build System --- api/src/glfs-internal.h | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) (limited to 'api/src/glfs-internal.h') 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; -- cgit