Re: kernel BUG at net/core/skbuff.c:LINE! (3)

[Dmitry Vyukov]

On Tue, Dec 3, 2019 at 12:56 PM Neil Horman <nhorman@xxxxxxxxxxxxx> wrote:
>
> On Tue, Dec 03, 2019 at 09:42:14AM +0100, Dmitry Vyukov wrote:
> > On Mon, Dec 2, 2019 at 7:39 PM Marcelo Ricardo Leitner
> > <marcelo.leitner@xxxxxxxxx> wrote:
> > >
> > > On Sat, Nov 30, 2019 at 04:37:56PM +0100, Dmitry Vyukov wrote:
> > > > On Sat, Nov 30, 2019 at 3:50 PM syzbot
> > > > <syzbot+b2bf2652983d23734c5c@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
> > > > >
> > > > > syzbot has bisected this bug to:
> > > > >
> > > > > commit 84e54fe0a5eaed696dee4019c396f8396f5a908b
> > > > > Author: William Tu <u9012063@xxxxxxxxx>
> > > > > Date:   Tue Aug 22 16:40:28 2017 +0000
> > > > >
> > > > >      gre: introduce native tunnel support for ERSPAN
> > > > >
> > > > > bisection log:  https://syzkaller.appspot.com/x/bisect.txt?x=158a2f86e00000
> > > > > start commit:   f9f1e414 Merge tag 'for-linus-4.16-rc1-tag' of git://git.k..
> > > > > git tree:       upstream
> > > > > final crash:    https://syzkaller.appspot.com/x/report.txt?x=178a2f86e00000
> > > > > console output: https://syzkaller.appspot.com/x/log.txt?x=138a2f86e00000
> > > > > kernel config:  https://syzkaller.appspot.com/x/.config?x=34a80ee1ac29767b
> > > > > dashboard link: https://syzkaller.appspot.com/bug?extid=b2bf2652983d23734c5c
> > > > > syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=147bfebd800000
> > > > > C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=13d8d543800000
> > > > >
> > > > > Reported-by: syzbot+b2bf2652983d23734c5c@xxxxxxxxxxxxxxxxxxxxxxxxx
> > > > > Fixes: 84e54fe0a5ea ("gre: introduce native tunnel support for ERSPAN")
> > > > >
> > > > > For information about bisection process see: https://goo.gl/tpsmEJ#bisection
> > > >
> > > > Humm... the repro contains syz_emit_ethernet, wonder if it's
> > > > remote-triggerable...
> > >
> > > The call trace is still from the tx path. Packet never left the system
> > > in this case.
> >
> > My understanding is that this does not necessarily mean that the
> > remote side is not involved. There is enough state on the host for L4
> > protocols, so that the remote side can mess things and then the bad
> > thing will happen with local trigger. But that local trigger can be
> > just anything trivial that everybody does.
> >
> But thats not really helpful.  Unless you see an explicit path from the receive
> side to ip6_append_data, Theres no real way for a received packet to reach this
> code, so we can't really call it remotely triggerable.
>
> My guess is, since this is coming from the rawv6_sendmsg path, that the raw
> protocol is somehow not marshaling its data in a way that ip6_append_data
> expects.

If it's in the local send path and does not use any remotely
controllable data, then this should be good enough estimation of not
being a remote attack vector.