Re: [net-rfc 04/16] can: dev: can_get_len(): add a helper function to get the correct length of Classical frames

From: Oliver Hartkopp <socketcan@hartkopp.net>
To: Vincent Mailhol <mailhol.vincent@wanadoo.fr>,
	Marc Kleine-Budde <mkl@pengutronix.de>,
	linux-can@vger.kernel.org
Cc: kernel@pengutronix.de, "Stéphane Grosjean" <s.grosjean@peak-system.com>
Subject: Re: [net-rfc 04/16] can: dev: can_get_len(): add a helper function to get the correct length of Classical frames
Date: Tue, 20 Oct 2020 19:04:04 +0200	[thread overview]
Message-ID: <a9605011-2674-dc73-111c-8ebf724a13ac@hartkopp.net> (raw)
In-Reply-To: <20201020160739.104686-1-mailhol.vincent@wanadoo.fr>

On 20.10.20 18:07, Vincent Mailhol wrote:

> I also did the test. I can send a CAN with a DLC of 13 on one
> controller and the other ones correctly received a frame of 8 bytes
> with a DLC of 13.

o_O

You see me perplexed ...

> After, I am not saying that absolutely all the controllers will allow
> DLC greater than 8. I would not be surprised to see some controllers
> attempting to do some sanitization (which would violate the ISO) and
> maybe you did your testing on such controllers. Only thing I can tell
> is that all the controllers I studied allowed it (I can give more
> examples upon request).

I believe you.

> As for security testing, I worked as a security consultant in the
> automotive industry for the last three years and with our colleagues,
> we witnessed some ECUs that would completely stop responding after
> receiving some DLCs greater than 8 due to some buffer overflow. This
> is a real thing which can be found in production, I think it would be
> great to be able to test that using socket CAN.

Yes. That's a valid use-case. Many people are testing CAN setups based 
on SocketCAN. So getting every aspect of CAN available is needed to be 
able to provide a real OSS solution.

> Some professional tools such as the CAN testing suite of Defensics by
> Synopsys also include these kind of tests. Because Socket CAN does not
> support this, Synopsys actually recommends to use the proprietary
> drivers from the Peak controller which do allow this (unfortunately,
> the Defensics documentation is not available publicly so I can not
> give you a link to support my claim on that last example).

Stephane from PEAK is working on the Linux driver (Mainline Linux & PEAK 
chardev), so I put him on CC. Or are you referring to the Windows driver?

> I hope that I could highlight in this answer that I am more than just
> a hobbyist who got exited after ready the ISO and that I know this
> subject. What I explain here is well known in the niche community of
> automotive security researcher but outside of it I just think that
> people are not aware of it.

Well I have done a lot in automotive CAN security too - with message 
authentication and with CAN IDS - but this DLC thing was still new to me ...

 From a first thought I would see a new flag CAN_CTRLMODE_RAW_DLC in the 
netlink interface of IFLA_CAN_CTRLMODE for the CAN controller driver.

This could switch the sanitizing AND the CAN controller can properly 
expose its ability to support this mode.

I think I have to pick a beer and look at some code ... :-)

Best regards,
Oliver