Re: [kernel-hardening] Getting started

[Kees Cook <keescook@chromium.org>]

On Thu, Feb 2, 2017 at 8:28 PM, Vaishali Thakkar
<vaishali.thakkar@oracle.com> wrote:
> On Friday 03 February 2017 08:41 AM, Eddie Kovsky wrote:
>>
>> On 01/30/17, Kees Cook wrote:
>>>
>>> On Mon, Jan 30, 2017 at 5:41 AM, Vaishali Thakkar
>>> <vaishali.thakkar@oracle.com> wrote:
>>>>
>>>> On Monday 30 January 2017 12:13 AM, Eddie Kovsky wrote:
>>>>
>>>>> I'm interested in helping out with this project.
>>>>>
>>>>> I have a few small patches in the kernel. I just finished the Eudyptula
>>>>> Challenge and I'm looking for places where I can continue to
>>>>> contribute.
>>>
>>>
>>> Hi! Welcome to the list. :)
>>>
>>>>> I've been reading the list for several months now. I think I have a
>>>>> general
>>>>> understanding of the development process. Is there a specific TODO item
>>>>> I
>>>>> could start off with?
>>>
>>>
>>> What areas of the kernel are you the most familiar with, and/or what
>>> things are you interested in working on? That could help me tailor
>>> some suggestions.
>>>
>>>> Here, is one TODO list:
>>>>
>>>> https://kernsec.org/wiki/index.php/Kernel_Self_Protection_Project
>>>>
>>>> Although I think few people are already working on some of these things.
>>>> May be you can also check the archives of a mailing list.
>>>
>>>
>>> The list is a bit terse (it's mostly been a brain dump as things come
>>> up), but yeah, if you see something there and want to know more, just
>>> ask. I'm happy to expand on any of them.
>>>
>>
>> I noticed there's been some activity recently with HARDENED_USERCOPY.
>> And I looked over how mm/usercopy.c was merged in from the grsecurity
>> patch. I'm curious about this TODO item:
>>
>>      Identify and extend HARDENED_USERCOPY to other usercopy functions
>>      (e.g. maybe csum_partial_copy_from_user, csum_and_copy_from_user,
>>      csum_and_copy_to_user, csum_partial_copy_nocheck?)
>>
>> It doesn't look like anyone is working on this task right now. But it's
>> not
>> obvious (to me) what needs to happen to make progress with this. Would
>> this
>> be a good task to start off with?
>
>
> Hi,
>
> You may want to read this thread:
> https://patchwork.kernel.org/patch/9409557/
>
> Kees, may be we should remove this item from the TODO list?

Hm, yeah, or rename it to "find any other APIs that look like
copy_to/from_user()". I think Mark already looked through these. Were
there any remaining?

As for a thing to work about how about this:
- provide mechanism to check for ro_after_init memory areas, and
reject structures not marked ro_after_init in vmbus_register()

The idea here would be to provide a mechanism functions can call to
verify that their arguments are const or ro_after_init. I think it'd
look a lot like the stuff in kernel/extable.c like
kernel_text_address(), but it'd need to ask "is this variable in the
rodata section? (Which is complicated by dealing with module rodata
sections.)

Then vmbus_register() could be modified to require that its arguments
are const or ro_after_init.

-Kees

-- 
Kees Cook
Pixel Security