On Fri, Aug 04, 2023 at 03:20:45PM +0200, Christian Brauner wrote:
> On Fri, Aug 04, 2023 at 12:14:08PM +0200, Christoph Hellwig wrote:
> > FYI, I can reproduce this trivially locally, but even after spending a
> > significant time with the trace I'm still puzzled at what is going
> > on. I've started trying to make sense of the lockdep report about
> > returning to userspace with s_umount held, originall locked in
> > get_tree_bdev and am still missing how it could happen.
>
> So in the old scheme:
>
> s = alloc_super()
> -> down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
>
> and assume you're not finding an old one immediately afterwards you'd
>
> -> spin_lock(&sb_lock)
>
> static int set_bdev_super(struct super_block *s, void *data)
> {
> s->s_bdev = data;
> s->s_dev = s->s_bdev->bd_dev;
> s->s_bdi = bdi_get(s->s_bdev->bd_disk->bdi);
>
> if (bdev_stable_writes(s->s_bdev))
> s->s_iflags |= SB_I_STABLE_WRITES;
> return 0;
> }
>
> -> spin_unlock(&sb_lock)
>
> in the new scheme you're doing:
>
> s = alloc_super()
> -> down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
>
> and assume you're not finding an old one immediately afterwards you'd
>
> up_write(&s->s_umount);
>
> error = setup_bdev_super(s, fc->sb_flags, fc);
> -> spin_lock(&sb_lock);
> sb->s_bdev = bdev;
> sb->s_bdi = bdi_get(bdev->bd_disk->bdi);
> if (bdev_stable_writes(bdev))
> sb->s_iflags |= SB_I_STABLE_WRITES;
> -> spin_unlock(&sb_lock);
>
> down_write(&s->s_umount);
>
> Which looks like the lock ordering here is changed?
Yes, that none only should be safe, but more importantly should not
lead to a return to userspace with s_umount held.
Anyway, debugging a regression in mainline right now so I'm taking a
break from this one.
