Re: [PATCH 2/2] mm: Add kvmalloc_ab_c and kvzalloc_struct

From: Kees Cook
Date: Thu May 03 2018 - 20:40:42 EST


On Thu, May 3, 2018 at 5:36 PM, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> On Thu, May 3, 2018 at 4:00 PM, Rasmus Villemoes
> <linux@xxxxxxxxxxxxxxxxxx> wrote:
>> On 2018-05-01 19:00, Kees Cook wrote:
>>> On Mon, Apr 30, 2018 at 2:29 PM, Rasmus Villemoes
>>> <linux@xxxxxxxxxxxxxxxxxx> wrote:
>>>>
>>>> gcc 5.1+ (I think) have the __builtin_OP_overflow checks that should
>>>> generate reasonable code. Too bad there's no completely generic
>>>> check_all_ops_in_this_expression(a+b*c+d/e, or_jump_here). Though it's
>>>> hard to define what they should be checked against - probably would
>>>> require all subexpressions (including the variables themselves) to have
>>>> the same type.
>>>>
>>>> plug: https://lkml.org/lkml/2015/7/19/358
>>>
>>> That's a very nice series. Why did it never get taken?
>>
>> Well, nobody seemed particularly interested, and then
>> https://lkml.org/lkml/2015/10/28/215 happened... but he did later seem
>> to admit that it could be useful for the multiplication checking, and
>> that "the gcc interface for multiplication overflow is fine".
>
> Oh, excellent. Thank you for that pointer! That conversation covered a
> lot of ground. I need to think a little more about how to apply the
> thoughts there with the kmalloc() needs and the GPU driver needs...
>
>> I still think even for unsigned types overflow checking can be subtle. E.g.
>>
>> u32 somevar;
>>
>> if (somevar + sizeof(foo) < somevar)
>> return -EOVERFLOW;
>> somevar += sizeof(this);
>>
>> is broken, because the LHS is promoted to unsigned long/size_t, then so
>> is the RHS for the comparison, and the comparison is thus always false
>> (on 64bit). It gets worse if the two types are more "opaque", and in any
>> case it's not always easy to verify at a glance that the types are the
>> same, or at least that the expression of the widest type is on the RHS.
>
> That's an excellent example, yes. (And likely worth including in the
> commit log somewhere.)
>
>>
>>> It seems to do the right things quite correctly.
>>
>> Yes, I wouldn't suggest it without the test module verifying corner
>> cases, and checking it has the same semantics whether used with old or
>> new gcc.
>>
>> Would you shepherd it through if I updated the patches and resent?
>
> Yes, though we may need reworking if we actually want to do the
> try/catch style (since that was talked about with GPU stuff too...)
>
> Either way, yes, a refresh would be lovely! :)

Whatever the case, I think we need to clean up all the kmalloc() math
anyway. As mentioned earlier, there are a handful of more complex
cases, but the vast majority are just A * B. I've put up a series here
now, and I'll send it out soon. I want to think more about 3-factor
products, addition, etc:

https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git/log/?h=kspp/kmalloc/2-factor-products

The commit logs need more details (i.e. about making constants the
second argument for optimal compiler results, etc), but there's a
Coccinelle-generated first pass.

-Kees

--
Kees Cook
Pixel Security