[dm-devel] xfs_btree_cur leak with 3.5-rc4

[Dave Chinner]

On Mon, Jul 02, 2012 at 09:26:28PM -0400, Mike Snitzer wrote:
> On Mon, Jul 02 2012 at  8:26pm -0400,
> Dave Chinner <david at fromorbit.com> wrote:
> 
> > On Mon, Jul 02, 2012 at 03:14:12PM -0400, Mike Snitzer wrote:
> > > I get kmemleak reports against XFS when I run the thinp-test-suite
> > > against 3.5-rc4.  I didn't see any XFS leaks with 3.4.
> > 
> > Already fixed in 3.5-rc5:
> > 
> > 76d0953 xfs: fix allocbt cursor leak in xfs_alloc_ag_vextent_near
> 
> Odd, I already have that fix (as it was included in 3.5-rc4).

I misread the git describe as saying it was included after -rc4. My
mistake. That indicates that the above fix wasn't quite correct...

Try the patch below.

Cheers,

Dave.
-- 
Dave Chinner
david at fromorbit.com


xfs: really first cursor leak in xfs_alloc_ag_vextent_near

From: Dave Chinner <dchinner at redhat.com>

The current cursor is reallocated when retrying the allocation, so
the exist cursor needs to be destroyed in both the restart and the
failure case.

Signed-off-by: Dave Chinner <dchinner at redhat.com>
---
 fs/xfs/xfs_alloc.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/xfs/xfs_alloc.c b/fs/xfs/xfs_alloc.c
index 9d1aeb7..f654f51 100644
--- a/fs/xfs/xfs_alloc.c
+++ b/fs/xfs/xfs_alloc.c
@@ -1074,13 +1074,13 @@ restart:
 	 * If we couldn't get anything, give up.
 	 */
 	if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
+		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
+
 		if (!forced++) {
 			trace_xfs_alloc_near_busy(args);
 			xfs_log_force(args->mp, XFS_LOG_SYNC);
 			goto restart;
 		}
-
-		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
 		trace_xfs_alloc_size_neither(args);
 		args->agbno = NULLAGBLOCK;
 		return 0;