Re: xfstests 073 regression

On Tue, Aug 02, 2011 at 07:44:28PM +0800, Wu Fengguang wrote:

On Tue, Aug 02, 2011 at 12:52:42AM +0800, Christoph Hellwig wrote:

quoted

wb_check_background_flush is indeed what we're hitting.

That means s_umount is NOT held by another queued writeback work.

Right. We already kind of knew that was ocurring because there's
a remount,ro going on.

quoted

See the trace output using a patch inspired by Curt's below:

# tracer: nop
#
#           TASK-PID    CPU#    TIMESTAMP  FUNCTION
#              | |       |          |         |
           <...>-4279  [000]   113.034052: writeback_grab_super_failed: bdi 7:0: sb_dev 0:0 nr_pages=9223372036854775807 sync_mode=0 kupdate=0 range_cyclic=1 background=1 reason=wb_check_background_flush
           <...>-4279  [000]   113.034052: writeback_grab_super_failed: bdi 7:0: sb_dev 0:0 nr_pages=9223372036854775807 sync_mode=0 kupdate=0 range_cyclic=1 background=1 reason=wb_check_background_flush
           <...>-4279  [000]   113.034052: writeback_grab_super_failed: bdi 7:0: sb_dev 0:0 nr_pages=9223372036854775807 sync_mode=0 kupdate=0 range_cyclic=1 background=1 reason=wb_check_background_flush

What's that bdi 7:0? And sb_dev=0:0, nr_pages=9223372036854775807=0x7fffffffffffffff.

All are indicating some special bdi/inode.

#define LOOP_MAJOR 7

It's a loop device. xfstests uses them quite a lot.

Maybe it would be a good idea to run xfstests on an xfs filesystem
in your regular writeback testing cycle to get decent coverage of
this case?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com