[Bridge] KASAN: use-after-free Read in br_mdb_ip_get

[Dmitry Vyukov]

On Wed, May 29, 2019 at 5:27 PM Herbert Xu <herbert at gondor.apana.org.au> wrote:
>
> On Wed, May 29, 2019 at 05:14:17PM +0200, Dmitry Vyukov wrote:
> >
> > > It looks like
> > >
> > > ommit 1515a63fc413f160d20574ab0894e7f1020c7be2
> > > Author: Nikolay Aleksandrov <nikolay at cumulusnetworks.com>
> > > Date:   Wed Apr 3 23:27:24 2019 +0300
> > >
> > >     net: bridge: always clear mcast matching struct on reports and leaves
> > >
> > > may have at least fixed the uninitialised value error.
> >
> >
> > The most up-to-date info is always available here:
> >
> > >> dashboard link: https://syzkaller.appspot.com/bug?extid=bc5ab0af2dbf3b0ae897
> >
> > It says no new crashes happened besides the original one.
> >
> > We now have the following choices:
> >
> > 1. Invalidate with "#syz invalid"
> > 2. Mark as tentatively fixed by that commit (could it fix it?) with
> > "#syz fix: net: bridge: always clear mcast matching struct on reports
> > and leaves"
> > 3. Do nothing, then syzbot will auto-close it soon (bugs without
> > reproducers that did not happen in the past 180 days)
>
> I'm still not quite sure how this could cause the use-after-free,
> but it certainly seems to be the cause for the second issue of
> uninit-value:
>
> https://syzkaller.appspot.com/bug?extid=8dfe5ee27aa6d2e396c2
>
> And this one does seem to have occured again recently (two months
> ago).

I've closed the KMSAN bug report with this commit.

And since the uninit value was used inside of the rhashtable (as
hash?) it could lead to any kind of inconsistencies, I guess we can
do:

#syz fix:
net: bridge: always clear mcast matching struct on reports and leaves

here too.

Thanks for bringing this up!