https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=281749
--- Comment #41 from Kyle Evans <[email protected]> --- (In reply to Konstantin Belousov from comment #37) FWIW, patch seems to make sense and LGTM. Based on some of these traces the namei must have succeeded because it would appear to be the VOP_ACCESS in zfs' setattr that hits an EBADF, so we would have vref()ed it in zfs_zget() and not released until the end. This would seem to be the race that makes the most sense. The one case I'm still not sure of is the "." branch in zfs_dirlook[0]; the other two will have ref'd the vnode, but that one doesn't. I haven't really found a reason why that's OK yet, but I also haven't really been able to induce misbehavior from it. [0] https://cgit.freebsd.org/src/tree/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_dir.c?h=releng/15.0#n241 -- You are receiving this mail because: You are the assignee for the bug.
