From: David Jeffery <djeffery@redhat.com>
Date: Tue, 3 Mar 2009 11:23:46 -0500
Subject: [nfs] fix hung clients from deadlock in flush_workqueue
Message-id: 49AD5992.9000107@redhat.com
O-Subject: [PATCH RHEL 5.4 BZ483627] Deadlock from flush_workqueue() results in hung nfs clients
Bugzilla: 483627
RH-Acked-by: Jeff Layton <jlayton@redhat.com>
RH-Acked-by: Peter Staubach <staubach@redhat.com>

RH BZ: https://bugzilla.redhat.com/show_bug.cgi?id=483627

Description:
During umount, nfs clients can deadlock by attempting to flush
workqueues from within a workqueue.  When nfs_release_automount_timer()
calls flush_scheduled_work() from within a workqueue, if another
processor has already acquired the workqueue mutex and started another
flush operation, the two will deadlock and never complete.

This patch is a backport of changes from
3d39c691ff486142dd9aaeac12f553f4476b7a62 and
560aef74503e928f44ddbf481b8b02d9cef37dbf upstream.  The
flush_scheduled_work() is removed from nfs_release_automount_timer() as
it does not protect anything and only creates deadlocks.

Found:
RHEL 5.2

kABI status:
No change

Backport of the nfs3 portions of commits
3d39c691ff486142dd9aaeac12f553f4476b7a62 and
560aef74503e928f44ddbf481b8b02d9cef37dbf from upstream

Fixes a deadlock with nfs's use of workqueues.

diff --git a/fs/nfs/namespace.c b/fs/nfs/namespace.c
index f69d82a..e9dbefd 100644
--- a/fs/nfs/namespace.c
+++ b/fs/nfs/namespace.c
@@ -173,10 +173,8 @@ static void nfs_expire_automounts(void *data)
 
 void nfs_release_automount_timer(void)
 {
-	if (list_empty(&nfs_automount_list)) {
+	if (list_empty(&nfs_automount_list))
 		cancel_delayed_work(&nfs_automount_task);
-		flush_scheduled_work();
-	}
 }
 
 /*