On Mon, Jul 19, 2021 at 07:21:28PM +0100, Christoph Hellwig wrote:

On Mon, Jul 19, 2021 at 11:00:15AM -0700, Darrick J. Wong wrote:

quoted

On Mon, Jul 19, 2021 at 12:20:54PM +0200, Christoph Hellwig wrote:

quoted

Hi all,

xfs/319 keeps crashing for me when running it for a 1k block size
file systems on x86 with debugging enabled.  The problem is that
xfs_bmapi_remap is called on a range that is not a hole.

Hmm.  Dave sent me a warning late last night about some sort of log
recovery problem in -rc2 due to FUA changes or something.  Did this
happen in -rc1?

I've reproduced it all the way back to 5.12 so far.

Ok, so not the regression I introduced in 5.14-rc1 with the journal
flush/FUA changes (caused by flush vs tail_lsn updates racing). THat
smells like the problem both Brian and I have seen that has occurred
occasionally for some time (I see it maybe once a month here) but we
haven't been able to reproduce reliably enough debug it.

Sounds like it's reliable on your setup? What's your test machine
config?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com