Re: [PATCH 01/11] crypto: shash: Remove VLA usage

[Kees Cook]

On Wed, Jun 20, 2018 at 1:39 PM, Christophe LEROY
<christophe.leroy@xxxxxx> wrote:
>
>
> Le 20/06/2018 Ã 22:36, Kees Cook a Ãcrit :
>>
>> On Wed, Jun 20, 2018 at 12:30 PM, Christophe Leroy
>> <christophe.leroy@xxxxxx> wrote:
>>>
>>>
>>>
>>> On 06/20/2018 07:03 PM, Kees Cook wrote:
>>>>
>>>>
>>>> In the quest to remove all stack VLA usage from the kernel[1], this
>>>> removes the VLAs in SHASH_DESC_ON_STACK (via crypto_shash_descsize())
>>>> by using the maximum allowable size (which is now more clearly captured
>>>> in a macro). Similar limits are turned into macros as well.
>>>>
>>>> [1]
>>>>
>>>> https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@xxxxxxxxxxxxxx
>>>>
>>>> Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
>>>
>>>
>>>
>>> Got the following warnings:
>>>
>>> crypto/testmgr.c: In function âalg_test_crc32c.part.4â:
>>> crypto/testmgr.c:1896:1: warning: the frame size of 2088 bytes is larger
>>> than 1024 bytes [-Wframe-larger-than=]
>>> crypto/hmac.c: In function âhmac_setkeyâ:
>>> crypto/hmac.c:88:1: warning: the frame size of 2088 bytes is larger than
>>> 1024 bytes [-Wframe-larger-than=]
>>
>>
>> Ah yes. I didn't do 32-bit builds. So, here's the issue: this uncovers
>> the frame size problems that were hidden by in being a VLA before. It
>> was always possible for the frame to get this big, it's just that the
>> compiler couldn't see it.
>>
>> For qat, I raised the -Wframe-larger-than flag. It seems we'll need to
>> do this in some other places too.
>
>
> Maybe the issue is because I have selected 16k pages ?

That would do it! And that's exactly the problem Arnd mentioned. For
v2, I will switch all the PAGE_SIZE-related limits to an explicit 512.
(And I'll do some 32-bit builds too to see if any other cases pop up
that I need to mask out.)

-Kees

-- 
Kees Cook
Pixel Security