On Mon, Apr 29, 2024 at 08:26:52AM -0700, Darrick J. Wong wrote: > I think this is actually a bug fix. If an xfs_iext_count_upgrade > dirties the transaction and then xfs_iext_lookup_extent cancels the > transaction due to the overlapping AIO race, the _trans_cancel shuts > down the fs, right? True. Pretty unlikely to hit, but still a bug.
