On Wed, May 29, 2019 at 04:33:00AM -0700, Greg Kroah-Hartman wrote: > On Wed, May 29, 2019 at 01:23:14PM +0200, David Sterba wrote: > > Hi, > > > > upon closer inspection we found a problem with the patch > > > > "btrfs: Honour FITRIM range constraints during free space trim" > > > > that has been merged to 5.1.4. This could happen with ranged FITRIM > > where the range is not 'honoured' as it was supposed to. > > > > Please revert it and push in the next stable release so the buggy > > version is not in the wild for too long. > > > > Affected trees: > > > > 5.0.18 > > 5.1.4 > > 4.9.179 > > 4.19.45 > > 4.14.122 > > > > Master branch will have the revert too. Thanks. > > What is the git commit id of the revert in Linus's tree? Due to further changes in the code, a revert that will be in Linus' branch can't be backported and stable-specific patches would have to be provided anyway. There's a slight change in logic and handling of the trimmed ranges, the upstream revert removes code and updates calls to functions that are not in the stable branches. So I'm going to send you patches for all the affected branches. After analyzing the situation, the conclusion is that the patch should have never been tagged for stable. The patch is in 5.2-rc ie. an unreleased kernel and the bug would be handled as a regression before 5.20 final. The backport to stable happened before we knew that so the reverts are IMO the best solution we have now. I hope you understand that and sorry for the trouble.
