Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 24 Apr 2009 19:50:57 -0400 (EDT)
From: "Steven M. Christey" <>
Subject: Re: CVE request? buffer overflow in CIFS in 2.6.*

On Tue, 21 Apr 2009, Eugene Teo wrote:

> > Our maintainer also referenced:
> >
> >
> >
> >
> > They are already in the CIFS git tree:
> >;a=summary
> >;a=commit;h=7b0c8fcff47a885743125dd843db64af41af5a61
> >;a=commit;h=968460ebd8006d55661dec0fb86712b40d71c413
> As discussed with Marcus, these two are unrelated to this issue, so we
> will need new CVE names.
> I spoke to Jeff Layton about this, and it looks like there are some more
> in the pipeline (but unrelated to this issue), so stay tuned.

This ominous statement is proving prophetic.

CVE is not designed to track every intermediate change that's being made,
reviewed, debugged, fixed, re-bugged, re-submitted, re-fixed, and
refactored in multiple branches on a real-time basis.  There's an
extensive list of changes going on at,
many of which don't provide any notions of attack scenarios or
vulnerability (and neither should they - but it puts a severe analytical
burden on CVE to reverse-engineer patches for the bowels of code that only
a handful of people in the world fully understand).  I don't think it
helps most people to have a dozen different CVEs for an extensive set of
issues that are being fixed so dynamically.

One approach might be to "pre-tag" this whole set of changes with a single
CVE, then when they ultimately get merged into a single kernel version or
some other concrete milestone, the "scope" of that CVE ends.

An alternate approach would be for the CVE requesters to provide more
detailed explanations of what exactly is going on and what the scope of
the bug is.  This moves the analytical burden back to the distros, but
theoretically, the requesters are more familiar with the code than CVE
analysts are.

Yet another approach would be for CVE to adopt very generic descriptions
and rely more heavily on the provided references, but that wouldn't cut
down on the number of CVEs being assigned and it would create many
opportunities for incorrect mappings and duplicate entries, not to mention
confusion for consumers (this based on the experience of the past 10 years
of CVE...)

Another more radical approach would be for the Linux community to adopt a
different mechanism for a "universal bug ID" besides CVE, where multiple
people can file multiple bug IDs and share them across distros without
caring about duplicates or inaccuracies or incomplete information.  The
CVE tie-in could occur when something is mature and final enough to make
it into a stable release, and then the CVE becomes a mechanism for the
distros to communicate to their consumers, instead of between themselves.

Any and all thoughts are welcome.  I've thought that a big win for CVE was
as a "universal bug ID" for the Linux community, which was outside what we
had originally envisioned for it.  But these kinds of dynamic,
low-information disclosures really stress the CVE process, so I'm starting
to question whether CVE is really a good fit.

- Steve

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.