Re: [PATCH v6 2/4] x86/modules: Increase randomization for modules

[Kees Cook]

On Mon, Sep 24, 2018 at 2:27 PM, Edgecombe, Rick P
<rick.p.edgecombe@xxxxxxxxx> wrote:
> On Mon, 2018-09-24 at 12:58 -0700, Kees Cook wrote:
>> On Mon, Sep 24, 2018 at 11:57 AM, Edgecombe, Rick P
>> <rick.p.edgecombe@xxxxxxxxx> wrote:
>> > > Instead of having two open-coded __vmalloc_node_range() calls left in
>> > > this after the change, can this be done in terms of a call to
>> > > try_module_alloc() instead? I see they're slightly different, but it
>> > > might be nice for making the two paths share more code.
>> > Not sure what you mean. Across the whole change, there is one call
>> > to __vmalloc_node_range, and one to __vmalloc_node_try_addr.
>> I guess I meant the vmalloc calls -- one for node_range and one for
>> node_try_addr. I was wondering if the logic could be combined in some
>> way so that the __vmalloc_node_range() could be made in terms of the
>> the helper that try_module_randomize_each() uses. But this could just
>> be me hoping for nice-to-read changes. ;)
>>
>> -Kees
> One thing I had been considering was to move the whole "try random locations,
> then use backup" logic to vmalloc.c, and just have parameters for random area
> size, number of tries, etc. This way it could be possibly be re-used for other
> architectures for modules. Also on our list is to look at randomizing vmalloc
> space (especially stacks), which may or may not involve using a similar method.
>
> So maybe bit pre-mature refactoring, but would also clean up the code in
> module.c. Do you think it would be worth it?

I'd love to hear thoughts from -mm folks. Andrew, Matthew?

-Kees

-- 
Kees Cook
Pixel Security