Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Sun, 25 May 2014 13:56:33 +0200
From: Guillem Jover <>
To: Raphael Geissert <>
Subject: Re: Re: CVE request: directory traversal in DSA-2915-1-patched dpkg
 in Debian squeeze


On Sun, 2014-05-25 at 11:17:52 +0200, Raphael Geissert wrote:
> On Thursday 01 May 2014 20:52:37 wrote:
> [...]
> > However, there is another question that could possibly result in a
> > second CVE ID. On wheezy, there's apparently a security problem in, at
> > least, these two cases:
> > 
> >   1. the installed patch program is an older version of patch that's
> >      identical to squeeze's supported version of patch. Possibly, this
> >      can happen on a correctly maintained/supported Debian system
> >      because sometimes a system is in a partially upgraded state. In
> >      particular, the vulnerability affects root's use of dpkg, which
> >      may be a completely expected activity because the administrator
> >      might want to unpack a source package even though an upgrade is
> >      unfinished.
> > 
> >   2. the patch program in root's path is not something obtained from
> >      Debian, e.g., the administrator intentionally decided to install
> >      a non-GNU patch program

Actually, dpkg-source relies pretty much on the patch program being
GNU patch, or at least an implementation with a compatible command line
interface, which I don't know exists. Still, yes, the patch program
could be locally installed say directly into /usr/local, or from a
local binary package.

> > We're not sure what can be done about case 1. The general issue is
> > that, when doing an upgrade of an arbitrary OS, the system might
> > intermittently be in a state in which incompatibility of
> > old-OS-version programs and new-OS-version programs has a major
> > security risk. Maybe the right answer is a policy that nobody is
> > allowed to do anything while an upgrade is in progress. In other
> > words, regular users may not be logged in, and root should only be
> > running the upgrade program and nothing else. ("nothing else" is
> > impractical; realistically, risk is reduced by running as little else
> > as possible.)

In the end I fixed the subsequent issue by outright rejecting C-style
formatted pathnames, which covers all cases and any combination of patch
program versions. I guess it could be considered that this was supposed
to be the proper fix from the start, and the intial one was just a
partial fix. So I'm not sure if this then merits a CVE id though.

> > Case 2 possibly violates the expectations of the concept of a Linux
> > distribution OS or other OS. In other words, the dpkg program requires
> > the patch program, and therefore the administrator must ensure that
> > the patch program (at least, in root's path) is the one provided by
> > the OS vendor. If the administrator decided to install any other patch
> > program, the resulting security problem would typically be considered
> > a site-specific problem and thus outside the scope of CVE.

Rigth, from the Debian prespective, I guess this could easily be
considered unsupported. Not from dpkg upstream PoV though, as it is
also used on other distributions or could be built from source, with
unknown combinations of other component versions installed.

> > However, a
> > CVE ID could be assigned of any of these is true:
> > 
> >   - on wheezy, dpkg doesn't have any explicit requirements about the
> >     patch program

Yes, dpkg does not impose any kind of versioned dependency on neither
squeeze or wheezy. So it is susceptible to at least the problem on partial
upgrades, which the distribution tends to support.

> >   - the dpkg documentation states that a non-GNU patch program can be
> >     used if desired

It's not stated, and I doubt that would work, but from dpkg PoV, it
should be considered supported as long as the command line is GNU
patch compatible, and the patch program can be invoked without error.

> >   - Debian has a general policy that a supported program (such as
> >     dpkg) must be robust in the face of non-standard but reasonable
> >     site-specific software changes. In other words, if Debian program
> >     1 allows arbitrary code execution in situations where the
> >     administrator replaces Debian program 2 with an
> >     often-considered-equivalent non-Debian program, this is supposed
> >     to be treated as a vulnerability in Debian program 1.
> Debian doesn't have a policy such as the one in the third point. I'm CC'ing 
> the dpkg maintainer to allow him to answer from his POV.

Rigth, not Debian, but I'd consider that a supported setup from dpkg
PoV, with the aforementioned caveats of GNU patch compatibility.

Hope that clarifies.


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.