linux-kernel - Re: WARNING in can

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Wed, 17 Jan 2018 09:22:47 +0100
From:   Oliver Hartkopp <socketcan@...tkopp.net>
To:     Dmitry Vyukov <dvyukov@...gle.com>,
        Eric Biggers <ebiggers3@...il.com>
Cc:     Marc Kleine-Budde <mkl@...gutronix.de>,
        syzbot <syzbot+4386709c0c1284dca827@...kaller.appspotmail.com>,
        David Miller <davem@...emloft.net>, linux-can@...r.kernel.org,
        LKML <linux-kernel@...r.kernel.org>,
        netdev <netdev@...r.kernel.org>, syzkaller-bugs@...glegroups.com
Subject: Re: WARNING in can_rcv



On 01/17/2018 08:39 AM, Dmitry Vyukov wrote:
> On Wed, Jan 17, 2018 at 8:12 AM, Eric Biggers <ebiggers3@...il.com> wrote:
>> On Wed, Jan 17, 2018 at 07:39:24AM +0100, Oliver Hartkopp wrote:
>>>
>>>
>>> On 01/16/2018 07:11 PM, Dmitry Vyukov wrote:
>>>> On Tue, Jan 16, 2018 at 7:07 PM, Marc Kleine-Budde <mkl@...gutronix.de> wrote:
>>>>> On 01/16/2018 06:58 PM, syzbot wrote:
>>>>>> Hello,
>>>>>>
>>>>>> syzkaller hit the following crash on
>>>>>> a8750ddca918032d6349adbf9a4b6555e7db20da
>>>>>> git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/master
>>>>>> compiler: gcc (GCC) 7.1.1 20170620
>>>>>> .config is attached
>>>>>> Raw console output is attached.
>>>>>> C reproducer is attached
>>>>>> syzkaller reproducer is attached. See https://goo.gl/kgGztJ
>>>>>> for information about syzkaller reproducers
>>>>>>
>>>>>>
>>>>>> IMPORTANT: if you fix the bug, please add the following tag to the commit:
>>>>>> Reported-by: syzbot+4386709c0c1284dca827@...kaller.appspotmail.com
>>>>>> It will help syzbot understand when the bug is fixed. See footer for
>>>>>> details.
>>>>>> If you forward the report, please keep this part and the footer.
>>>>>>
>>>>>> device eql entered promiscuous mode
>>>>>> ------------[ cut here ]------------
>>>>>> PF_CAN: dropped non conform CAN skbuf: dev type 65534, len 42, datalen 0
>>>>>> WARNING: CPU: 0 PID: 3650 at net/can/af_can.c:729 can_rcv+0x1c5/0x200
>>>>>> net/can/af_can.c:724
>>>>>> Kernel panic - not syncing: panic_on_warn set ...
>>>>>
>>>>> Invalid packages generate a warning (WARN_ONCE()), and you have
>>>>> panic_on_warn active. Should we better silently drop these CAN packages?
>>>>
>>>> Hi,
>>>>
>>>> pr_warn_once() will be more appropriate. It prints a single line.
>>>>
>>>
>>> The idea behind this WARN() is to detect really bad things that might have
>>> happen on network driver level:
>>>
>>> The CAN subsystem registers with dev_add_pack() for ETH_P_CAN and
>>> ETH_P_CANFD only. These ETH_P_ types are only allowed to be created by CAN
>>> network devices (like vcan, vxcan, and real CAN drivers).
>>>
>>> I don't have any strong opinion on using WARN() or pr_warn_once().
>>> Is this detected violation worth using WARN(), as something already must
>>> have gone really wrong to trigger this issue?
>>>
>>
>> WARN() indicates a kernel bug.  If it's instead "userspace did something
>> stupid", or "someone sent some unexpected network packet", it needs to be
>> pr_warn_once(), pr_warn_ratelimited(), or removed entirely.
> 
> 
> The packet comes from tun device. We could change tun to filter out
> such packages earlier. However, in the context of "syzkaller support
> for AF_CAN" discussion, it would actually be useful for fuzzer to be
> able emit can packets for testing purposes. 

Yes - definitely! It's a safer process to check the reception side 
instead of maintaining thousands of potential transmitters.

> For example, for tcp it
> can not just emit random packets, it can build complex user<->network
> interactions, for example, open a listening socket, connect to it
> "from outside", accept the connection, and then exchange some data
> over the active connection. It could do the same for can.

Yes.

> Is it possible to allow can packets via tun?

Hm - didn't even think about it.
CAN frames have a fixed data structure (struct can_frame) so the tunnel 
would need to be capable to process SOCK_SEQPACKET (?!?) traffic.

Right now there has been no work to 'tunnel' CAN traffic.

> Then we could leave this
> WARNING in place.

Yes.

> tun/vcan are contained within a net namespace, so
> this should not be a security problem, right?

vcan can be created in or moved into a namespace. vxcan can bridge 
namespaces similar to veth. This is all local traffic then.

What kind of security problem would you have in mind there?

> Or is there a way to do the same with vcan? If yes, then fuzzer could
> use vcan.

Yes. This would be my idea too. Unfortunately I'm very busy @work this 
week - so I would like to dig deeper into your mail some days ago at the 
beginning of next week.

> But then we need some fix for this WARNING: either change it
> to pr_warn or change tun (I don't have strong preference which one).

 From the discussions (also with Eric) I think going with pr_warn is the 
right way for now.

Tnx & best regards,
Oliver