the cache is re-populated while waiting for i_mutex, it may find that a
d_lookup() subsequently succeeds (see the "Uhhuh!  Nasty case" comment).

Previously, real_lookup() would drop i_mutex and do_revalidate() again. 
If revalidate failed _again_, however, it would give up with -ENOENT.  The
problem here that network file systems may be invalidating dentries via
server callbacks, e.g.  due to concurrent access from another client, and
-ENOENT is frequently the wrong answer.

This problem has been seen with both Lustre and Ceph.  It seems possible
to hit this case with NFS as well if the cache lifetime is very short.

Instead, we should do_revalidate() while i_mutex is still held.  If
revalidation fails, we can move on to a ->lookup() and ensure a correct
result without worrying about any subsequent races.

Note that do_revalidate() is called with i_mutex held elsewhere.  For
example, do_filp_open(), lookup_create(), do_unlinkat(), do_rmdir(), and
possibly others all take the directory i_mutex, and then

-> lookup_hash
        -> __lookup_hash
                -> cached_lookup
                        -> do_revalidate

so this does not introduce any new locking rules for d_revalidate
implementations.

Yes, the goto is ugly.  A cleanup patch follows.

Cc: Ian Kent <raven@themaw.net>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger@sun.com>
Signed-off-by: Yehuda Sadeh <yehuda@newdream.net>
Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>