Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 14 Aug 2009 20:36:07 +0400
From: Solar Designer <solar@...nwall.com>
To: oss-security@...ts.openwall.com
Subject: Re: CVE id request: groff (pdfroff)

Hi Nico,

Thank you for forwarding this info in here - it has helped.

Have you notified Werner LEMBERG, the upstream maintainer?

I have some comments on "the first bug" and on groff's temporary file
handling in general:

On Sun, Aug 09, 2009 at 03:48:17PM +0200, Nico Golde wrote:
> First one:
> http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=538330
> pdfroff tool of groff is creating files in a insecure manner 
> in the /tmp directory.

There's a mitigating factor: pdfroff will use $TMPDIR (or one of three
other env vars that it checks) if set.  On Owl, pam_mktemp sets $TMPDIR
to point to the user's private temporary file directory upon PAM session
setup, which is invoked on both remote and console logins, and on cron
job invocations.

I found your trivial patch (pdfroff.sh.diff, 389 bytes) a bit unfinished:

-  WRKFILE=${GROFF_TMPDIR=${TMPDIR-${TMP-${TEMP-"."}}}}/pdf$$.tmp
+  WRKFILE=${GROFF_TMPDIR=$(mktemp -t -d groffXXXXXX)}/pdf$$.tmp

- no check for a possible mktemp error;
- will leave the directory around upon pdfroff termination;
- changes semantics, but does not update documentation (the man page is
explicit on what temporary file directories pdfroff uses).

I included more elaborate changes to pdfroff (both the script and the
man page) in groff-1.20.1-owl-tmp.diff available here:

http://cvsweb.openwall.com/cgi/cvsweb.cgi/Owl/packages/groff/

Other issues I found (and patched in groff-1.20.1-owl-tmp.diff) are:

Bugs in eqn2graph.sh, grap2graph.sh, pic2graph.sh (duplicated code)
where these scripts would fail to properly handle the case when all
attempts to create a temporary directory fail.  In that case, they would
proceed to use the last-tried pathname.  Hopefully, this would just fail
in some weird way, but it is also possible that the directory would in
fact exist and be under someone else's control.  Since this was the last
thing I dealt with today and I was already tired, I opted for the
simplest fix, which was to reset the variable at the end of each loop
iteration.  At a later time, I may drop all non-mktemp code from these
scripts, like I did in other places.

gendef.sh and doc/fixinfo.sh created files in $TMPDIR (if defined) or in
the current directory without any precautions.  Not an issue if $TMPDIR
is safe or is not set, and these are only used during groff build, but
it is an issue if someone sets $TMPDIR to /tmp or to another shared
directory.  I've patched these to use mktemp(1).  gendef.sh is on my
list of files accessed during groff build (which means that it could
have been run but this is not certain); fixinfo.sh is not (so it was
certainly not run during my build but this does not mean much for other
builds).

contrib/gdiffmk/tests/runtests.in created files in /tmp in the worst way
possible.  I've patched it to use mktemp(1), although maybe it should
just use the current directory since it does so in other places anyway.
It is on my list of files accessed during groff build.

contrib/groffer/perl/groffer.pl and contrib/groffer/perl/roff2.pl used
only four X'es in filename patterns.  The reasonable minimum is
considered 6; we typically use 10.  Patched to use 10.

doc/groff.texinfo (source) and doc/groff.info-2 (pre-compiled) contained
an example for invoking an external program with ".sy", saving its
output to a file under /tmp.  Patched to use a file in the current
directory instead, even though this is not perfect.

Finally, we're patching configure and config.guess to use the
mktemp-only code just to be fail-close, even though I understand the
rationale behind the original portable code.  I am also adding a #error
comment to src/roff/groff/pipeline.c, just to save a few minutes next
time I or someone else looks at this code.

That's all for the patch.  However, my grep's also identified potential
issues in install-sh and contrib/groffer/shell/*.  Since these files
were not used during our groff package builds, I opted to remove rather
than patch them.  So I added:

# Remove/disable unused files with temporary file handling issues in them to
# make sure that these are in fact unused.
rm -r contrib/groffer/shell
echo -e '#!/bin/sh\nexit 1' > install-sh

to our groff.spec, and I also had to add
groff-1.20.1-owl-groffer-Makefile.diff to allow groff to build without
the contrib/groffer/shell/* files (their presence was checked by make
even though they were not read).  On systems with Perl, groff uses the
version of groffer written in Perl instead of this shell version, but
there's some risk of it falling back to the shell version if Perl can't
be run for whatever reason.

A related detail is that we force the configure script to detect the
presence of mkstemp(3), we don't take chances:

export ac_cv_func_mkstemp=yes \
%configure

In the case of groff, the fallback code (when mkstemp(3) is not present)
is not that bad, though.

Speaking of my grep's, I used:

fgrep -rl /tmp .
fgrep -rl mktemp .
fgrep -rl tmpnam .
fgrep -rl tempnam .
fgrep -rl TMPDIR .

Then I briefly reviewed the files identified in this way.  Of course, I
could have missed something, yet this would have caught the pdfroff
issue.  We updated Owl-current to this version of groff with pdfroff in
it just 8 days ago, and we were not as careful right away...

Alexander

Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Powered by Openwall GNU/*/Linux - Powered by OpenVZ