btrfs/004 with my new fsstress was sometimes failing with a transaction abort while trying to remove a root ref. This turned out to be because of how we stub out subvol links in snapshot'ed subvolumes. The specific steps are detailed in "btrfs: fix invalid removal of root ref", but they are relatively straightforward. A xfstest is forthcoming, I want to get an overnight run of fsstress with these patches in place to make sure there are no other rename+removal shenanigans left. With these patches we pass my basic test and no longer abort the transaction. Thanks, Josef
