[PATCH V2 2/2] fs/super.c: don't fool lockdep in freeze_super() and thaw_super() paths

From: Oleg Nesterov
Date: Mon Sep 26 2016 - 12:55:33 EST

On 09/26, Jan Kara wrote:
> On Mon 26-09-16 18:08:06, Oleg Nesterov wrote:
> > +/*
> > + * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
> > + */
> > +static void sb_freeze_acquire(struct super_block *sb)
> Can we call this lockdep_sb_freeze_acquire() or something like that so that
> it is clear this is only about lockdep annotations? Similarly with
> sb_freeze_unlock()...

OK, thanks, done. See V2 below.

> and I hope you really tested
> there are no more lockdep false positives ;).

Heh ;) if only I knew how to test this... I ran the following script under qemu

mkfs.xfs -f /dev/vda
mkfs.xfs -f /dev/vdb


./check `grep -il freeze tests/*/???`

this time.

And yes, I'm afraid this change can uncover some false positives later. But at
the same time potentially it can find the real problems.

It would be nice to remove another hack in __sb_start_write under ifdef(CONFIG_LOCKDEP),
but iirc XFS actually takes the same rw_sem twice for reading, so we can't do this.