Re: CVE-2014-6271: remote code execution through bash

[Florian Weimer <fweimer () redhat com>]

On 09/24/2014 08:54 PM, Michal Zalewski wrote:
> > My main concern with the current patch is that still exposes the bash parser
> > and function definition printer to attacks from the network. Bugs in those
> > fairly large components could cause another critical issue.
>
> Yup, that surprised me when testing the patch, too - I can still get a
> function called HTTP_COOKIE, for example. I worry about potential side
> effects of parsing even in absence of parser bugs. In most
> object-oriented languages, such side effects are practically
> guaranteed. Bash may be saved by simplicity, but not sure how robust
> that assumption is.

The parser does make an effort to properly stage all operations for 
later execution, without executing them immediately.

There is certainly the question of incomplete state recovery on parse 
errors.

> I've written more code in bash than I should have and never used
> function exports, or even realized that they exist. I wonder if they
> can be made optional (e.g., gated by a flag on the subprocess) without
> breakage.

I've been told that there are users.  From what I can see, exported 
functions seem somewhat popular in test harnesses:

  <http://codesearch.debian.net/search?q=export\+-f>

Reportedly, some users even create the function definitions outside 
bash, so they rely function name and variable name being identical.  But 
I honestly cannot see a way to preserve such an assumption.

A subprocess flag is unlikely to be present when it is needed.  One 
common use case is to define and export functions in bash.profile, and 
expect them to exist in interactive shells created as grandchildren. 
The fix is to use bashrc instead of bash.profile and non-exported functions.

> Another option may be to export them through specially prefixed
> variables, which should be transparent but minimize the risk of
> interfering with web servers and such.

I added suffixes as well, see the attached patch.  This patch has seen 
some testing, but it certainly needs more.  There are some possibilities 
for simplification if it's acceptable to use asprintf.

What do you think about this approach?

(Chet, this patch is identical to the patch I sent to you a couple of 
minutes ago.)

--
Florian Weimer / Red Hat Product Security

Attachment: variables-affix.patch
Description: