> On Nov 18, 2023, at 6:36 PM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote: > > On Sat, Nov 18, 2023 at 05:11:39PM -0500, Chuck Lever wrote: > >> We don't hold f_pos_lock in offset_dir_llseek(), though. > > We'd better... Which call chain leads to it without ->f_pos_lock? I based that comment on code audit. It does not appear to be obviously held in this code path on first (or second) reading. However, let me look again, and test with lockdep_assert() to confirm. -- Chuck Lever