On Tue 21-02-17 12:14:30, Eryu Guan wrote: > On Mon, Feb 20, 2017 at 09:58:19AM -0500, Brian Foster wrote: > > On Mon, Feb 20, 2017 at 03:25:21PM +0100, Michal Hocko wrote: > > > I have discussed that shortly with Jack and he suspects that xfs > > > specific part of the iomap callbacks doesn't cancel reservations. > > > Let's CC more xfs people. The thread starts > > > http://lkml.kernel.org/r/20170210035348.GA7075@xxxxxxxxxxxxxxxxxxxxxxxx > > > > > > > Have you tried commit fa7f138 ("xfs: clear delalloc and cache on buffered > > write failure"), currently in for-next? > > I cherry-picked this commit to 4.10-rc8 tree and xfs/297 xfs/305 all > passed for me without fs corruption. They used to fail for me reliably > with stock 4.10-rc8 kernel. Thanks all for looking at this! Brian, could you make sure that your commit gets to stable (at least) 4.8+ because that is where d1908f52557b ("fs: break out of iomap_file_buffered_write on fatal signals") will be backported to. -- Michal Hocko SUSE Labs
