Bug 181941 - mail-filter/spamassassin Symlink DoS issue (CVE-2007-2873)
Bug#: 181941 Product:  Gentoo Security Version: unspecified Platform: All
OS/Version: Linux Status: RESOLVED Severity: minor Priority: P2
Resolution: FIXED Assigned To: security@gentoo.org Reported By: jaervosz@gentoo.org
Component: Vulnerabilities
URL:  http://spamassassin.apache.org/advisories/cve-2007-2873.txt
Summary: mail-filter/spamassassin Symlink DoS issue (CVE-2007-2873)
Keywords:  
Status Whiteboard: C3? [noglsa] jaervosz
Opened: 2007-06-13 19:36 0000
Description:   Opened: 2007-06-13 19:36 0000
CVE reference: CVE-2007-2873

Description:

A local user symlink-attack DoS vulnerability in SpamAssassin has been found,
affecting versions 3.1.x, 3.2.0, and SVN trunk.  It has been assigned
CVE-2007-2873. Details:

- It only affects systems where spamd is run as root, is used with vpopmail or
  virtual users via the "-v"/"--vpopmail" OR "--virtual-config-dir" switch, AND
  with the "-x"/"--no-user-config AND WITHOUT the "-u"/"--username" switch AND
  with the "-l"/"--allow-tell" switch.

  This is not default on any distro package, and is not a common configuration.

- It is a local exploit that requires the attacker to have a local account
  whose mail is being processed by spamd.

- The effect of the exploit is to allow overwriting of arbitrary files that are
  accessible by the spamd process (running as root), with data that is not
  under the control of the attacker.  Hence it is a DoS vulnerability that does
  not allow remote execution nor escalation of local privileges.


Workaround: If you are running spamd using a vulnerable combination of
switches,
add the "-u" / "--username" switch to specify a non-root user that spamd child
processes will run as.

Note that in a mixed real/virtual user environment you will now have to run two
separate instances of spamd on different ports, with the instance that
specifies "-v"/"--vpopmail" or "--virtual-config-dir" also specifying
"-u"/"--username".

Fix: The vulnerability is fixed in SpamAssassin version 3.2.1 by, among other
fixes, no longer allowing the use of "-v"/"--vpopmail" or
"--virtual-config-dir" without the "-u"/"--username" switch.  Thus, the
configuration change described in the above workaround is still necessary when
upgrading to 3.2.1.

Further info: mail <security at SpamAssassin.apache.org>
Announced: Jun 11 2007
Corrected: Jun 11 2007
Affects: all versions before the correction date, after and including 3.1.0
Credit: discovery of this vulnerability credited to Martin F. Krafft 
  <madduck@debian.org>

------- Comment #1 From Jakub Moc (RETIRED) 2007-06-15 14:15:10 0000 -------
*** Bug 182149 has been marked as a duplicate of this bug. ***

------- Comment #2 From Sune Kloppenborg Jeppesen 2007-06-16 06:18:29 0000 -------
perl please advise and bump as necessary.

------- Comment #3 From Coran 2007-06-16 14:46:37 0000 -------
3.2.1 is a major bug-fix release, including a potential local DoS.  The
major highlights are:

- bug 5480: fix for CVE-2007-2873: a local user symlink-attack DoS
  vulnerability. It only affects systems where spamd is run as root, is used
  with vpopmail or virtual users via the "-v"/"--vpopmail" OR
  "--virtual-config-dir" switch, AND with the "-x"/"--no-user-config AND
  WITHOUT the "-u"/"--username" switch AND with the "-l"/"--allow-tell" switch.
  This is not default on any distro package, and is not a common configuration.
  More details of the vulnerability can be read at
  <http://spamassassin.apache.org/advisories/cve-2007-2873.txt>.

<list of other bugs truncated as they aren't related to this bug>

------- Comment #4 From Christian Hartmann 2007-06-17 12:49:53 0000 -------
It's failing in tests for me. So no bump from me currently.

------- Comment #5 From Michael Cummings (RETIRED) 2007-06-18 16:08:01 0000 -------
hmmm...passed all tests over here (hence my bump when i saw the update notice
from apache.org, irrespective of this bug). where's it failing ian?

------- Comment #6 From Michael Cummings (RETIRED) 2007-06-18 22:46:36 0000 -------
http://www.gossamer-threads.com/lists/spamassassin/users/102895

tests seem broken for root. running as a regular user produces success

t/utf8......................ok                                               
t/util_wrap.................ok                                               
t/whitelist_addrs...........ok                                               
t/whitelist_from............ok                                               
t/whitelist_subject.........ok                                               
t/whitelist_to..............ok                                               
t/zz_cleanup................ok                                               
All tests successful, 16 tests skipped.
Files=129, Tests=1985, 582 wallclock secs (288.68 cusr + 22.90 csys = 311.58
CPU)


Move to disable tests for this release (especially if we're looking at a gsla
related keywording)

------- Comment #7 From Sune Kloppenborg Jeppesen 2007-06-23 18:02:30 0000 -------
Thx Micheal.

Arches please test and mark stable. Target keywords are:

spamassassin-3.2.1-r1.ebuild:KEYWORDS="alpha amd64 hppa ia64 mips ppc ppc64
sparc x86"

------- Comment #8 From Christoph Mende 2007-06-23 18:26:15 0000 -------
amd64 done

------- Comment #9 From Jeroen Roovers 2007-06-23 23:57:21 0000 -------
Marked stable for HPPA:
 dev-util/re2c-0.12.0
 mail-filter/spamassassin-3.2.1-r1

------- Comment #10 From Raúl Porcel 2007-06-24 15:30:57 0000 -------
alpha/ia64/x86 stable

------- Comment #11 From Tobias Scherbaum 2007-06-24 19:51:52 0000 -------
ppc stable

------- Comment #12 From Gustavo Zacarias (RETIRED) 2007-06-25 13:28:18 0000 -------
sparc stable.

------- Comment #13 From Markus Rothe 2007-06-26 05:47:26 0000 -------
ppc64 stable

------- Comment #14 From Sune Kloppenborg Jeppesen 2007-07-01 02:23:07 0000 -------
This one is ready for GLSA decision. I vote NO.

------- Comment #15 From Pierre-Yves Rofes 2007-07-01 09:51:03 0000 -------
voting NO too.

------- Comment #16 From Matt Drew 2007-07-02 21:47:14 0000 -------
vote no, too specific and extremely unlikely.

------- Comment #17 From Stefan Cornelius (RETIRED) 2007-07-13 16:19:26 0000 -------
yet another no and closing