Re: [syzbot] WARNING: refcount bug in p9_client_walk

Re: [syzbot] WARNING: refcount bug in p9_client_walk

[asmadeus]

Albert Guo wrote on Wed, Nov 15, 2023 at 03:50:48AM +0000:
> Hi Asemadeus,

Please always cc the subsystem's mailing list when asking such
questions; there's no point in singling out maintainers.
(And it'd really help if you wouldn't mispell names; although I guess
you could also say it does help since rilled up folks are more likely to
answer...)

> I’m Albert Guo from VMware.  Sorry to disturb you. I’m writing to you to seek some help for Linux 9p driver.
> We use 9p driver in our product.  We seem to encounter similar refcount issue in our test:
> [syzbot] WARNING: refcount bug in p9_client_walk
> https://groups.google.com/g/syzkaller-bugs/c/UrvjxOLiykQ/m/LlVAtCzQBQAJ?pli=1
> 
> syzbot
> 2023年3月23日 06:02:37
> 
> 收件人 syzkall...@googlegroups.com<mailto:syzkall...@googlegroups.com>
> Auto-closing this bug as obsolete.
> No recent activity, existing reproducers are no longer triggering the issue.
> 
> 
> Looks like the issue isn’t reproducible from syzbot? Is it fixed? Do
> you know which commit fix the issue?

It might have been fixed in 26273ade77f54716e30dfd40ac6e85ceb54ac0f9 as
uninitialized use could be causing that, otherwise I don't know --
there's been no other fix around refcounting since that report...

> Our kernel version is  6.1.56: https://elixir.bootlin.com/linux/v6.1.56/source

But it's long been fixed (included in 6.2 and 6.1.2), so there must be
another problem.

Please give your full stack trace (and not just a link to syzbot); the
warn message might help.
Also, how often do you hit this? (Can we consider it "reproductible"
under test where you would be able to run with tracepoints, or
simplifying the load etc?)

-- 
Dominique Martinet | Asmadeus