lkml.org 
[lkml]   [2014]   [Apr]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH] ubi: avoid workqueue format string leak
From
On Tue, Apr 8, 2014 at 7:43 AM, Artem Bityutskiy
<artem.bityutskiy@linux.intel.com> wrote:
> On Tue, 2014-04-08 at 10:57 -0300, Ezequiel Garcia wrote:
>> Hello Kees,
>>
>> Thanks for the patch.
>>
>> On Apr 07, Kees Cook wrote:
>> > When building the name for the workqueue thread, make sure a format
>> > string cannot leak in from the disk name.
>> >
>>
>> Could you enlighten me and explain why you want to avoid the name leak?
>> Is it a security concern?
>>
>> I'd like to understad this better, so I can avoid making such mistakes
>> in the future.
>
> Well, the basics seem to be simple, attacker makes sure gd->disk_name
> contains a bunch of "%s" and other placeholders, and this leads
> "workqueue_alloc()" to read kernel memory and form the workqueue name.

Right. I don't think there is an actual exploitable vulnerability
here, but it's a best-practice to not pass variable strings in as a
potential format string.

> I did not think it through further, though, but that was enough for me
> to apply the patch right away. But yeah, curios parts are:
>
> 1. How attacker could end up with a crafted "gd->disk_name"

At present, the only way I know how to set that is via some special
controls in md, but I assume that would not work via ubi.

> 2. How attacker gets the workqueue name then, I guess there is a sysfs
> file or something, but I do not know off the top of my head.

This I haven't checked, but if there isn't a way to do it now, we can
at least avoid a nasty surprise in the future if one is created. :)

> Yeah, I am interested to get educated on this a too.

Thanks for pulling the fix!

-Kees

--
Kees Cook
Chrome OS Security


\
 
 \ /
  Last update: 2014-04-08 18:21    [W:0.048 / U:3.296 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site