On Tue, Apr 21, 2020 at 12:22 AM David Sterba <dsterba@xxxxxxx> wrote: > > On Thu, Apr 16, 2020 at 03:34:30PM +0000, Johannes Thumshirn wrote: > > On 16/04/2020 14:38, Johannes Thumshirn wrote: > > > This fixes a kmemleak complaint from btrfs/074, complete re-run of > > > xfstests is pending, but one down again. > > > > > > Tested-by: Johannes Thumshirn <johannes.thumshirn@xxxxxxx> > > > > > > > I'll take that back, I still see a leak of 'cur_trans' allocated in > > join_transaction() for btrfs/074 on a full xfstests run. The same leak > > is reported for btrfs/072 and generic/127. > > I'm not sure, but the patch "btrfs: drop logs when we've aborted a > transaction" is fixing transaction handle leaks. I've added it to > misc-next, the effects have been observed in test generic/475 so it's > only a weak link, tests btrfs/074 do not stress the transaction cleanup > that much. Hum? That patch fixes a use-after-crash during unmount after a transaction was aborted - it doesn't fix transaction leaks as far as I can see. Perhaps you meant patch "btrfs: fix memory leak of transaction when deleting unused block group", which fixes a regression introduced in 5.7-rc1. btrfs/074 (and tests 060 to 073) often triggers unused block group deletion due to fsstress and other operations in parallel. -- Filipe David Manana, “Whether you think you can, or you think you can't — you're right.”
