From: Josef Bacik <jbacik@redhat.com>
Date: Mon, 20 Apr 2009 16:06:30 -0400
Subject: [fs] keep eventpoll from locking up the box
Message-id: 1240257990-1823-1-git-send-email-jbacik@redhat.com
O-Subject: [PATCH] [RHEL5.4] keep eventpoll from locking up the box
Bugzilla: 487585
RH-Acked-by: Peter Staubach <staubach@redhat.com>
RH-Acked-by: Dave Anderson <anderson@redhat.com>

This is in reference to bz 487585.  This was fixed the same way in RHEL4, in bz
446409, I just forgot to do it for RHEL5.  The reason this problem goes away
upstream is because there have been several locking changes, but I'd rather go
with this tried and true method instead of pulling back these potentially
disruptive changes.  If we'd rather go with the locking reworking patches I can
pull those back instead, but I think this will be safer and less disruptive.
Thank you,

Josef

Signed-off-by: Josef Bacik <jbacik@redhat.com>

diff --git a/fs/eventpoll.c b/fs/eventpoll.c
index 8d54433..be06bf1 100644
--- a/fs/eventpoll.c
+++ b/fs/eventpoll.c
@@ -1475,7 +1475,7 @@ static int ep_events_transfer(struct eventpoll *ep,
 	 * We need to lock this because we could be hit by
 	 * eventpoll_release_file() and epoll_ctl(EPOLL_CTL_DEL).
 	 */
-	down_read(&ep->sem);
+	down_write(&ep->sem);
 
 	/* Collect/extract ready items */
 	if (ep_collect_ready_items(ep, &txlist, maxevents) > 0) {
@@ -1486,7 +1486,7 @@ static int ep_events_transfer(struct eventpoll *ep,
 		ep_reinject_items(ep, &txlist);
 	}
 
-	up_read(&ep->sem);
+	up_write(&ep->sem);
 
 	return eventcnt;
 }