[syzbot] [reiserfs?] possible deadlock in reiserfs_dirty_inode

[Paul Moore]

On Tue, Nov 7, 2023 at 6:03 AM Roberto Sassu
<roberto.sassu at huaweicloud.com> wrote:
> On Mon, 2023-11-06 at 17:53 -0500, Paul Moore wrote:
> > Hi Roberto,
> >
> > I know you were looking at this over the summer[1], did you ever find
> > a resolution to this?  If not, what do you think of just dropping
> > security xattr support on reiserfs?  Normally that wouldn't be
> > something we could consider, but given the likelihood that this hadn't
> > been working in *years* (if ever), and reiserfs is deprecated, I think
> > this is a viable option if there isn't an obvious fix.
> >
> > [1] https://lore.kernel.org/linux-security-module/CAHC9VhTM0a7jnhxpCyonepcfWbnG-OJbbLpjQi68gL2GVnKSRg@mail.gmail.com/
>
> Hi Paul
>
> at the time, I did some investigation and came with a patch that
> (likely) solves some of the problems:
>
> https://lore.kernel.org/linux-fsdevel/4aa799a0b87d4e2ecf3fa74079402074dc42b3c5.camel@huaweicloud.com/#t

Ah, thanks for the link, it looks like that was swallowed by my inbox.
In general if you feel it is worth adding my email to a patch, you
should probably also CC the LSM list.  If nothing else there is a
patchwork watching the LSM list that I use to make sure I don't
miss/forget about patches.

> I did a more advanced patch (to be validated), trying to fix the root
> cause:
>
> https://lore.kernel.org/linux-fsdevel/ffde7908-be73-cc56-2646-72f4f94cb51b@huaweicloud.com/
>
> However, Jeff Mahoney (that did a lot of work in this area) suggested
> that maybe we should not try invasive changes, as anyway reiserfs will
> be removed from the kernel in 2025.

I tend to agree with Jeff, which is one of the reasons I was
suggesting simply removing LSM xattr support from reiserfs, although
depending on what that involves it might be a big enough change that
we are better off simply leaving it broken.  I think we need to see
what that patch would look like first.

> It wouldn't be a problem to move the first patch forward.

I worry that the first patch you mentioned above doesn't really solve
anything, it only makes it the responsibility of the user to choose
either A) a broken system where LSM xattrs don't work or B) a system
that will likely deadlock/panic.  I think I would rather revert the
original commit and just leave the LSM xattrs broken than ask a user
to make that choice.

-- 
paul-moore.com