Re: [PATCH v2 1/2] mm: mmap: Add new /proc tunable for mmap_base ASLR.

From: Kees Cook
Date: Wed Nov 04 2015 - 17:37:25 EST


On Wed, Nov 4, 2015 at 2:10 PM, Eric W. Biederman <ebiederm@xxxxxxxxxxxx> wrote:
> Daniel Cashman <dcashman@xxxxxxxxxxx> writes:
>
>> On 11/3/15 5:31 PM, Andrew Morton wrote:
>>> On Tue, 03 Nov 2015 18:40:31 -0600 ebiederm@xxxxxxxxxxxx (Eric W. Biederman) wrote:
>>>
>>>> Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> writes:
>>>>
>>>>> On Tue, 3 Nov 2015 10:10:03 -0800 Daniel Cashman <dcashman@xxxxxxxxxxx> wrote:
>>>>>
>>>>>> ASLR currently only uses 8 bits to generate the random offset for the
>>>>>> mmap base address on 32 bit architectures. This value was chosen to
>>>>>> prevent a poorly chosen value from dividing the address space in such
>>>>>> a way as to prevent large allocations. This may not be an issue on all
>>>>>> platforms. Allow the specification of a minimum number of bits so that
>>>>>> platforms desiring greater ASLR protection may determine where to place
>>>>>> the trade-off.
>>>>>
>>>>> Can we please include a very good description of the motivation for this
>>>>> change? What is inadequate about the current code, what value does the
>>>>> enhancement have to our users, what real-world problems are being solved,
>>>>> etc.
>>>>>
>>>>> Because all we have at present is "greater ASLR protection", which doesn't
>>>>> really tell anyone anything.
>>>>
>>>> The description seemed clear to me.
>>>>
>>>> More random bits, more entropy, more work needed to brute force.
>>>>
>>>> 8 bits only requires 256 tries (or a 1 in 256) chance to brute force
>>>> something.
>>>
>>> Of course, but that's not really very useful.
>>>
>>>> We have seen in the last couple of months on Android how only having 8 bits
>>>> doesn't help much.
>>>
>>> Now THAT is important. What happened here and how well does the
>>> proposed fix improve things? How much longer will a brute-force attack
>>> take to succeed, with a particular set of kernel parameters? Is the
>>> new duration considered to be sufficiently long and if not, are there
>>> alternative fixes we should be looking at?
>>>
>>> Stuff like this.
>>>
>>>> Each additional bit doubles the protection (and unfortunately also
>>>> increases fragmentation of the userspace address space).
>>>
>>> OK, so the benefit comes with a cost and people who are configuring
>>> systems (and the people who are reviewing this patchset!) need to
>>> understand the tradeoffs. Please.
>>
>> The direct motivation here was in response to the libstagefright
>> vulnerabilities that affected Android, specifically to information
>> provided by Google's project zero at:
>>
>> http://googleprojectzero.blogspot.com/2015/09/stagefrightened.html
>>
>> The attack there specifically used the limited randomness used in
>> generating the mmap base address as part of a brute-force-based exploit.
>> In this particular case, the attack was against the mediaserver process
>> on Android, which was limited to respawning every 5 seconds, giving the
>> attacker an average expected success rate of defeating the mmap ASLR
>> after over 10 minutes (128 tries at 5 seconds each). With change to the
>> maximum proposed value of 16 bits, this would change to over 45 hours
>> (32768 tries), which would make the user of such a system much more
>> likely to notice such an attack.
>>
>> I understand the desire for this clarification, and will happily try to
>> improve the explanation for this change, especially so that those
>> considering use of this option understand the tradeoffs, but I also view
>> this as one particular hardening change which is a component of making
>> attacks such as these harder, rather than the only solution. As for the
>> clarification itself, where would you like it? I could include a cover
>> letter for this patch-set, elaborate more in the commit message itself,
>> add more to the Kconfig help description, or some combination of the above.
>
> Unless I am mistaken this there is no cross over between different
> processes of this randomization. Would it make sense to have this as
> an rlimit so that if you have processes on the system that are affected
> by the tradeoff differently this setting can be changed per process?

I think that could be a good future bit of work, but I'd want to get
this in for all architectures first, so we have a more common base to
work from before introducing a new rlimit.

-Kees

--
Kees Cook
Chrome OS Security
--
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/