xfs: skip stale inodes in xfs_iflush_cluster

commit 7d3aa7f upstream.

We don't write back stale inodes so we should skip them in
xfs_iflush_cluster, too.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

fs/xfs/xfs_inode.c

@@ -2925,6 +2925,7 @@ xfs_iflush_cluster(
 		 */
 		spin_lock(&iq->i_flags_lock);
 		if (!iq->i_ino ||
+		    __xfs_iflags_test(iq, XFS_ISTALE) ||
 		    (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
 			spin_unlock(&iq->i_flags_lock);
 			continue;