From: Jeff Layton <jlayton@redhat.com>
Date: Wed, 2 Sep 2009 09:05:17 -0400
Subject: [nfs] fix regression in nfs_open_revalidate
Message-id: 1251896717-18023-1-git-send-email-jlayton@redhat.com
O-Subject: [RHEL5.5 PATCH] BZ#511278: nfs: fix regression in nfs_open_revalidate
Bugzilla: 511278
RH-Acked-by: Peter Staubach <staubach@redhat.com>
RH-Acked-by: Steve Dickson <SteveD@redhat.com>

Prior to the patch for BZ#321111, nfs_open_revalidate used to do a
d_drop on the dentry when nfs4_open_revalidate returned false. That
patch changed the function such that it did the d_drop when the function
returned true.

This caused a subtle regression. If a program was being run from an
NFSv4 mount, then the result of every second readlink call on
/proc/self/exe would have the string " (deleted)" appended to it. It
probably also has other effects as well that aren't as easily noticable
or reproducible.

The issue seems to have been a mistake when doing the backport for
BZ#321111. One of the patches that was backported did change a "!ret" to
"ret == 1", but that actually occurred several lines above this spot and
was eventually removed by a later patch in the series.

Tested with the reproducer provided for the problem. Also did some basic
regression testing with connectathon et. al.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c
index 926e26d..6ce4ab9 100644
--- a/fs/nfs/dir.c
+++ b/fs/nfs/dir.c
@@ -1053,7 +1053,7 @@ static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
 	unlock_kernel();
 out:
 	dput(parent);
-	if (ret == 1)
+	if (!ret)
 		d_drop(dentry);
 	return ret;
 no_open: