[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 10 Jul 2014 23:12:19 +0000
From: Ramon de C Valle <rdecvalle@...are.com>
To: "oss-security@...ts.openwall.com" <oss-security@...ts.openwall.com>
CC: Tomas Hoger <thoger@...hat.com>, Murray McAllister <mmcallis@...hat.com>
Subject: Re: [ruby-core:63604] [ruby-trunk - Bug #10019]
[Open] segmentation fault/buffer overrun in pack.c (encodes)
On Jul 10, 2014, at 6:59 PM, Ramon de C Valle <rdecvalle@...are.com> wrote:
> Hi Thomas, Murray,
>
> On Jul 10, 2014, at 7:43 AM, Tomas Hoger <thoger@...hat.com> wrote:
>
>> On Wed, 9 Jul 2014 15:45:10 +0000 Ramon de C Valle wrote:
>>
>>> I believe this should have a CVE assigned.
>>
>> Can you post more details of your analysis of the issue to clarify what
>> the issue is here?
> From https://urldefense.proofpoint.com/v1/url?u=https://bugs.ruby-lang.org/issues/10019&k=oIvRg1%2BdGAgOoM1BIlLLqw%3D%3D%0A&r=bZpuVimtRQUx3xHFIlu%2BaciWn3GMzM%2FBnwDoBm5jP8U%3D%0A&m=INlQZCdbEhwoZYUL%2FLXSUVHqBdVmMPauC0BH2SHpGhA%3D%0A&s=9d9df67d0535cccd7b8572814df073ac1a01f871aa52ea4d8763008ea023da0c, it seems that you’ve figured it out already. Correct me if I’m wrong but, for Base64, a value of 3072 for len isn’t enough to cause the off-by-one as the while loop will terminate with the value of len being zero (and the value of i being 4092). However, if the value of len is either is 3073* or 3074*, the while loop will terminate with the value of len being 1 or 2 respectively (and the value of i being 4092), with one of the subsequent if/else if conditions evaluating to true, resulting in the off-by-one.
>
> I see you’ve checked the template strings used by aws-sdk gem and its dependencies and they use ‘m0’ only, which rules out the possibility this off-by-one being caused by any of these gems. So, now I’m also not sure what the reporter is referring to.
>
> *It is possible to pass non multiple of 3 values as the len parameter of encodes function by passing a string with length smaller than the count (/ 3 * 3) passed in the template string (see https://urldefense.proofpoint.com/v1/url?u=https://github.com/ruby/ruby/blob/trunk/pack.c%23L839&k=oIvRg1%2BdGAgOoM1BIlLLqw%3D%3D%0A&r=bZpuVimtRQUx3xHFIlu%2BaciWn3GMzM%2FBnwDoBm5jP8U%3D%0A&m=INlQZCdbEhwoZYUL%2FLXSUVHqBdVmMPauC0BH2SHpGhA%3D%0A&s=2eabd2924dd7e3efad12ddf21cba514da7ee0b384de6522507b5bf6af075d3a3).
I just double checked it, and the correct values are 3069, 3070, and 3071. So, a value of 3069 is the value that isn’t enough to cause the off-by-one, and if the value of len is either 3070* or 3071*, the while loop will terminate with the value of len being 1 or 2 respectively (and the value of i being 4092), with one of the subsequent if/else if conditions evaluating to true (incrementing the value of i up to 4096), resulting in the off-by-one (at https://github.com/ruby/ruby/blob/trunk/pack.c#L987), which matches the first case you’ve listed.
Anyway, whatever the reporter is referring to, he mentions it doesn't occur in 1.9.3, and looking at 1.9.3, the only related differences I immediately noticed are the absence of the check at https://github.com/ruby/ruby/blob/trunk/pack.c#L829 in pack_pack function and padding being an int (instead of char) in the encodes function.
>
>>
>> --
>> Tomas Hoger / Red Hat Security Response Team
>
> --
> Ramon de C Valle
> VMware Product Security Engineering
--
Ramon de C Valle
VMware Product Security Engineering
Download attachment "signature.asc" of type "application/pgp-signature" (843 bytes)
Powered by blists - more mailing lists
Please check out the
Open Source Software Security Wiki, which is counterpart to this
mailing list.
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
