Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Low] dm-push-backup: pb->alive KCSAN race only partly fixed, plain accesses remain --
commit 9d9907f2e7f049fda97985539a4920b9e2e49764 Author: Andrey Zhadchenko <[email protected]> drivers/md/push_backup: clear pb->alive under pb->lock This patch makes pb_message() take pb->lock around the store that clears pb->alive when handling push_backup_stop. The aim is to avoid a possible KCSAN report and to match other users of pb->alive. Link: https://virtuozzo.atlassian.net/browse/VSTOR-152912 > diff --git a/drivers/md/dm-push-backup.c b/drivers/md/dm-push-backup.c > index 9a6252f2b21e..1ae4ccc208ff 100644 > --- a/drivers/md/dm-push-backup.c > +++ b/drivers/md/dm-push-backup.c > @@ -581,7 +581,9 @@ static int pb_message(struct dm_target *ti, unsigned int > argc, char **argv, > if (!pb->map) > return -EBADF; > > + spin_lock_irq(&pb->lock); > pb->alive = false; > + spin_unlock_irq(&pb->lock); > wake_up_interruptible(&pb->waitq); /* pb->alive = false */ > } [Severity: Low] Does this fully avoid the KCSAN report described in the commit message ("it may potentially trigger KCSAN")? This store in pb_message() runs before ctl_rwsem is taken, so it holds pb->lock but not ctl_rwsem. A concurrent push_backup_start() holds ctl_rwsem but not pb->lock, so the two do not exclude each other: push_backup_start() if (pb->alive) return -EEXIST; That read is a plain read with no pb->lock and no READ_ONCE(), so it can still race with the store above. The wait_event condition used by push_backup_read() is also evaluated without pb->lock. It reads the flag through pb_has_event(): !READ_ONCE(pb->alive) || !READ_ONCE(pb->map_bits) || ... Here the reader is marked, but none of the writers use WRITE_ONCE(). The stores in pb_message() (this patch), push_backup_stop(), and the pb->alive = true store in the push_backup_start() setup path are all still plain stores. KCSAN can report a plain store racing with a lockless read, so taking pb->lock around only this one store may not remove the report. Would it make sense to use WRITE_ONCE() on the stores to pb->alive and READ_ONCE() on the read in push_backup_start() (or take pb->lock for that read)? The commit message could also be adjusted if the intent is only to make the locking consistent. -- Sashiko AI review ยท https://sashiko.vzint.dev/#/patchset/[email protected]?part=2 _______________________________________________ Devel mailing list [email protected] https://lists.openvz.org/mailman/listinfo/devel
