Re: [PATCH net 1/2] net: dsa: realtek: fix out-of-bounds access

[Ahmad Fatoum <a.fatoum@pengutronix.de>]

Hi Linus,

On 15.03.23 14:15, Linus Walleij wrote:
> On Wed, Mar 15, 2023 at 2:09 PM Ahmad Fatoum <a.fatoum@pengutronix.de> wrote:
> 
>> The probe function sets priv->chip_data to (void *)priv + sizeof(*priv)
>> with the expectation that priv has enough trailing space.
>>
>> However, only realtek-smi actually allocated this chip_data space.
>> Do likewise in realtek-mdio to fix out-of-bounds accesses.
>>
>> Fixes: aac94001067d ("net: dsa: realtek: add new mdio interface for drivers")
>> Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
> 
> Reviewed-by: Linus Walleij <linus.walleij@linaro.org>

Thanks for the review.
 
> That this worked for so long is kind of scary, and the reason why we run Kasan
> over so much code, I don't know if Kasan would have found this one.

It still worked. I looked into it some more and for some reason, struct realtek_priv
has a char buf[4096] member that's unused. I assume it caused kmalloc to return a 8K
slab, where the out-of-bound writes didn't overwrite anything of value. That buffer
ought to be removed, but that's for net-next.

I just checked with KASAN and it does detect the OOB on ARM64. I first noticed the
bug on barebox though, which has a near verbatim port of the Linux driver, but a
TLSF allocator, which fits allocations more tightly, hence it crashes not long
after driver probe unlike Linux.

> Rewriting the whole world in Rust will fix this problem, but it will
> take a while...

^^'.

Cheers,
Ahmad



> 
> Yours,
> Linus Walleij
> 

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |