On Tue 29-11-11 06:06:21, Mikulas Patocka wrote: > On Tue 29-11-11 11:19:01, Jan Kara wrote: > > Hi, > > > > On Mon 28-11-11 18:32:18, Mikulas Patocka wrote: > > > > Hi > > > > > > > > Where can I get that patch set? > > > > > > > > We are experiencing other similar deadlocks on RHEL-6, caused by sync or > > > > background writeback (these code paths take s_umount and wait trying to do > > > > I/O), but I wasn't able to reproduce these deadlocks on upstream kernel? > > > > Are there other known deadlock possibilities? > > > > > > I found some patch named "[RFC PATCH 1/3] VFS: Fix s_umount thaw/write > > > deadlock" (I couldn't find the next two parts of the patch in the > > > archives). And the patch looks wrong: > > Yes, that seems to be the series. I generally agree with you that the > > last iteration still had some problems and some changes were requested. > > That's why it's not merged yet after all... > > > > > - down_read_trylock(&sb->s_umount) doesn't fix anything. The lock is not > > > held when the filesystem is frozen and it is taken for write when thawing. > > > Consequently, any task can succeed with down_read_trylock(&sb->s_umount) > > > on a frozen filesystem and if this tasks attempts to do an I/O that is > > > waiting for thaw, it may still deadlock. > > Agreed. > > > > > - skipping sync on frozen filesystem violates sync semantics. > > > Applications, such as databases, assume that when sync finishes, data were > > > written to stable storage. If we skip sync when the filesystem is frozen, > > > we can cause data corruption in these applications (if the system crashes > > > after we skipped a sync). > > Here I don't agree. Filesystem must guarantee there are no dirty data on > > a frozen filesystem. > > This is technically impossible to achieve on ext2, fat or other > non-transactional filesystems. These filesystems have no locks around code > paths that set data or inodes dirty. And you still need working sync for > ext2. So the best thing to do in sync is to wait until the filesystem is > unfrozen. Then suspend is effectively unsupported on the filesystem and should return EOPNOTSUPP? At least that's what I'd expect... Honza -- Jan Kara <jack@xxxxxxx> SUSE Labs, CR -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
