On Tue 10-10-23 11:43:59, Christian Brauner wrote: > On Mon, Oct 9, 2023, 18:34 Jan Kara <jack@xxxxxxx> wrote: > > > On Mon 09-10-23 18:16:45, Christian Brauner wrote: > > > On Mon, Oct 09, 2023 at 02:33:41PM +0200, Christian Brauner wrote: > > > > We can't open devices with s_umount held without risking deadlocks. > > > > So drop s_umount and reacquire it when opening the journal device. > > > > > > > > Reported-by: syzbot+062317ea1d0a6d5e29e7@xxxxxxxxxxxxxxxxxxxxxxxxx > > > > Signed-off-by: Christian Brauner <brauner@xxxxxxxxxx> > > > > --- > > > > > > Groan, I added a dumb bug in here. > > > > Which one? I went through the patch again but I still don't see it... > > > > (Sorry, from phone.) > > I'm dropping s_umount across a lot of work > instead of just over device opening which is really the wrong way of doing > this. > I should just drop it over journal_dev_init(). So I was kind of suspecting that but I couldn't figure out how it would exactly matter when SB_BORN still is not set... Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR