locks: close potential race in lease_get_mtime
authorJeff Layton <jlayton@primarydata.com>
Fri, 22 Aug 2014 14:18:44 +0000 (10:18 -0400)
committerJeff Layton <jlayton@primarydata.com>
Tue, 7 Oct 2014 18:06:12 +0000 (14:06 -0400)
lease_get_mtime is called without the i_lock held, so there's no
guarantee about the stability of the list. Between the time when we
assign "flock" and then dereference it to check whether it's a lease
and for write, the lease could be freed.

Ensure that that doesn't occur by taking the i_lock before trying
to check the lease.

Cc: J. Bruce Fields <bfields@fieldses.org>
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
fs/locks.c

index f5f648e..def1ac2 100644 (file)
@@ -1456,8 +1456,18 @@ EXPORT_SYMBOL(__break_lease);
  */
 void lease_get_mtime(struct inode *inode, struct timespec *time)
 {
-       struct file_lock *flock = inode->i_flock;
-       if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
+       bool has_lease = false;
+       struct file_lock *flock;
+
+       if (inode->i_flock) {
+               spin_lock(&inode->i_lock);
+               flock = inode->i_flock;
+               if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
+                       has_lease = true;
+               spin_unlock(&inode->i_lock);
+       }
+
+       if (has_lease)
                *time = current_fs_time(inode->i_sb);
        else
                *time = inode->i_mtime;