On Thu, Mar 21, 2019 at 07:27:44AM -0400, Jeff Layton wrote: > Andreas reported that he was seeing the tdbtorture test fail in > some cases with -EDEADLCK when it wasn't before. Some debugging > showed that deadlock detection was sometimes discovering the > caller's lock request itself in a dependency chain. > > If posix_locks_deadlock() fails to find a deadlock, the caller_fl > will be passed to __locks_insert_block(), and this wakes up all > locks that are blocked on caller_fl, clearing the fl_blocker link. > > So if posix_locks_deadlock() finds caller_fl while searching for > a deadlock, I'm feeling dense. Could you step me through the scenario in a little more detail? Also, how do we know this catches every such case? And why aren't we unhashing blocks when we wake them up? --b. > it can be sure that link in the cycle is about to be > broken and it need not treat it as the cause of a deadlock. > > More details here: https://bugzilla.kernel.org/show_bug.cgi?id=202975 > > Cc: stable@xxxxxxxxxxxxxxx > Fixes: 5946c4319ebb ("fs/locks: allow a lock request to block other requests.") > Reported-by: Andreas Schneider <asn@xxxxxxxxxx> > Signed-off-by: Neil Brown <neilb@xxxxxxxx> > Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx> > --- > fs/locks.c | 4 ++++ > 1 file changed, 4 insertions(+) > > diff --git a/fs/locks.c b/fs/locks.c > index eaa1cfaf73b0..b074f6d7fd2d 100644 > --- a/fs/locks.c > +++ b/fs/locks.c > @@ -1023,6 +1023,10 @@ static int posix_locks_deadlock(struct file_lock *caller_fl, > while ((block_fl = what_owner_is_waiting_for(block_fl))) { > if (i++ > MAX_DEADLK_ITERATIONS) > return 0; > + > + if (caller_fl == block_fl) > + return 0; > + > if (posix_same_owner(caller_fl, block_fl)) > return 1; > } > -- > 2.20.1
