On Tue, Jan 21, 2020 at 11:51:41AM -0500, Josef Bacik wrote: > This series is to address a few issues with the fixup worker we hit in > production. > > The first of this is a resend of > > Btrfs: keep pages dirty when using > > I've cleaned this up based on the feedback and added a bunch more comments to > make it clear what is happening and why we're doing it. > > The next patch is a cleanup that is made possible by the previous patch, again > to clear up the fixup workers job. > > btrfs: drop the -EBUSY case in __extent_writepage_io > > And finally the deadlock fix that I submitted earlier. I noticed while trying > to backport this onto our kernel that we had changed the error case with the > above patch from Chris, and actually we really, really need Chris's fix as well. > There is also a change in the error handling from v1 where we now set the page > error properly but only once we've locked the page and verified we're still > responsible for COW'ing the page. Thanks, Reviewed-by: David Sterba <dsterba@xxxxxxxx> Very tricky stuff that fixup worker, it's like the worst present a filesystem can get from memory management. Estimated Merge target is 5.6, post rc1 so we have enough time for testing. It'll appear either in misc-next or for-next.
