On Thu, Oct 11, 2018 at 06:48:54PM +0200, Jan Kara wrote: > On Thu 11-10-18 08:29:53, Paul E. McKenney wrote: > > On Thu, Oct 11, 2018 at 12:28:00PM +0200, Jan Kara wrote: > > > On Sat 06-10-18 23:07:06, Theodore Ts'o wrote: > > > > It's possible for ext4_show_quota_options() to try reading > > > > s_qf_names[i] while it is being modified by ext4_remount() --- most > > > > notably, in ext4_remount's error path when the original values of the > > > > quota file name gets restored. > > > > > > > > Reported-by: syzbot+a2872d6feea6918008a9@xxxxxxxxxxxxxxxxxxxxxxxxx > > > > Signed-off-by: Theodore Ts'o <tytso@xxxxxxx> > > > > Cc: stable@xxxxxxxxxx > > > > > > Well, honestly the fact that ->show_options can be called while remount is > > > changing stuff under you looks problematic to me and I bet ext4 is not the > > > only one that would have issues with that. So I believe we might be better > > > off with just synchronizing ->show_options with umount / remount properly. > > > What were the lock dependency problems that made you switch to use RCU? > > > > OK, I will bite... > > > > Ted's patch does in fact just properly synchronize ->show_options with > > umount / remount. Using RCU. ;-) > > Well, it does but only for quota mount options. There may be other mount > options which would need similar treatment and possibly other mount options > in other filesystems. So I think a saner VFS API would be to synchronize > ->show_options with umount / remount so that each filesystem does not have > to do it on its own. On that choice, I must of course defer to the various filesystem and VFS people, yourself included. Thanx, Paul
