On Fri, May 31, 2013 at 12:59 PM, Jakub Zawadzki
<darkjames-ws@xxxxxxxxxxxx> wrote:
> Hi,
>
> On Thu, May 30, 2013 at 10:23:05PM -0400, Evan Huus wrote:
>> On Thu, May 30, 2013 at 10:15 PM, Evan Huus <eapache@xxxxxxxxx> wrote:
>> > On Thu, May 30, 2013 at 9:46 PM, <morriss@xxxxxxxxxxxxx> wrote:
>> >> http://anonsvn.wireshark.org/viewvc/viewvc.cgi?view=rev&revision=49644
>> >>
>> >> User: morriss
>> >> Date: 2013/05/30 06:46 PM
>> >>
>> >> Log:
>> >> (Finally!) check in part of Didier's patch to fix
>> >> https://bugs.wireshark.org/bugzilla/show_bug.cgi?id=3290
>> >> (TRY_TO_FAKE_THIS_ITEM disables bounds errors):
>> >>
>> >> Before calling TRY_TO_FAKE_THIS_ITEM() check if the length given (or, in
>> >> the case of FT_UINT_{STRING,BYTES}, the length we retrieve from the TVB)
>> >> exceeds what's left in the TVB.
>> >>
>> >> Do this only for proto_tree_add_item() for now (it's the most commonly used
>> >> and thus the biggest trouble maker in this area).
>> >>
>> >> Similar changes for other APIs will come later (if nothing blows up). Despite
>> >> the fuzz failures this bug has caused I'm not sure about back-porting it...
>> >>
>> >> Directory: /trunk/epan/
>> >> Changes Path Action
>> >> +28 -3 proto.c Modified
>> >
>> > Thank you for this!
>> >
>> > If we get through a round or two of fuzz-testing without any failures
>> > I would really like to see this backported to every stable branch
>> > (even 1.6). It closes an entire class of security vulnerabilities, and
>> > while it is a fairly non-trivial behavioural change in a hot code
>> > path, it is relatively short and clearly not doing anything too odd.
>> >
>> > Fingers crossed for no unexpected side-effects...
>> > Evan
>>
>> P.S. I don't know when we're expecting 1.10 final but I think delaying
>> it (if necessary) in order to include this fix is fully justified.
>
> Has anyone performed some benchmarks before/after this patch? :)
I've been too afraid :P
However, security > performance. Protocols that want performance can
still manually put blocks of proto_tree_ calls under if (tree), they
just have to be sure they're not breaking anything when they do so.
Still worth backporting :)