Re: drivers: random: Shift out-of-bounds in _mix_pool_bytes

[Andrey Ryabinin]

On 10/24/2014 02:01 PM, Peter Zijlstra wrote:
> On Mon, Oct 20, 2014 at 10:16:35AM -0400, Theodore Ts'o wrote:
>> On Mon, Oct 20, 2014 at 04:09:30PM +0200, Daniel Borkmann wrote:
>>>>
>>>> It's triggering when input_rotate == 0, so UBSan complains about right shift in rol32()
>>>>
>>>> static inline __u32 rol32(__u32 word, unsigned int shift)
>>>> {
>>>> 	return (word << shift) | (word >> (32 - shift));
>>>> }
>>>
>>> So that would be the case when the entropy store's input_rotate calls
>>> _mix_pool_bytes() for the very first time ... I don't think it's an
>>> issue though.
>>
>> I'm sure it's not an issue, but it's still true that 
>>
>> 	return (word << 0) | (word >> 32);
>>
>> is technically not undefined, and while it would be unfortunate (and
>> highly unlikely) if gcc were to say, start nethack, it's technically
>> allowed by the C spec.  :-)
> 
> In fact, n >> 32 == n.
> 
> #include <stdio.h>
> 
> int main(int argc, char **argv)
> {
> 	int i = atoi(argv[1]);
> 	int shift = atoi(argv[2]);
> 	printf("%x\n", i >> shift);
> 	return 0;
> }
> 
> $ ./shift 5 32
> 5
> 
> On x86 at least the shift ops simply mask out the upper bits and
> therefore the 32 == 0.
> 
> So you end up OR-ing the same value twice, which is harmless.
> 
> So no misbehaviour on the rol32() function.
> 

E.g. on arm (i >> 32) == 0, so rol32() will also work as expected.
But what about other architectures?

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/