Bugs item #1179487, was opened at 2005-04-08 14:46 Message generated for change (Comment added) made by mrcrispin You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=100103&aid=1179487&group_id=103 Category: mail delivery Group: 2.1 (stable) Status: Open Resolution: None Priority: 5 Submitted By: Mark Crispin (mrcrispin) Assigned to: Tokio Kikuchi (tkikuchi) Summary: denial of service security bug Initial Comment: We've had multiple incidents of this problem. If a digest gets a message containing an attachment using an RFC 2231 encoded parameter has a character set that is unknown to Python (in this case, "X- UNKNOWN"), then routine get_filename() in email/Message.py (not to be confused with Mailman/Message.py) calls unicode() without any error trap. The result is that digest delivery for that entire mailing list is suspended until that message is manually removed. It appears that passing an "ignore" as the errors parameter to unicode() won't stop Python from generating this error. I'm not sure as to the best way to fix this. I haven't worked much with Python at all, and Mailman support was just dumped on my lap. I can see that there are lots of unicode() calls throughout the Mailman source that don't have any error protection. I don't know which ones are also vulnerable to this attack. Traceback (most recent call last): File "/usr/local/mailman/cron/senddigests", line 94, in ? main() File "/usr/local/mailman/cron/senddigests", line 86, in main mlist.send_digest_now() File "/usr/local/mailman/Mailman/Digester.py", line 60, in send_digest_n ow ToDigest.send_digests(self, mboxfp) File "/usr/local/mailman/Mailman/Handlers/ToDigest.py", line 132, in sen d_digests send_i18n_digests(mlist, mboxfp) File "/usr/local/mailman/Mailman/Handlers/ToDigest.py", line 306, in sen d_i18n_digests msg = scrubber(mlist, msg) File "/usr/local/mailman/Mailman/Handlers/Scrubber.py", line 268, in pro cess url = save_attachment(mlist, part, dir) File "/usr/local/mailman/Mailman/Handlers/Scrubber.py", line 362, in sav e_attachment fnext = os.path.splitext(msg.get_filename(''))[1] File "/usr/local/mailman/pythonlib/email/Message.py", line 731, in get_f ilename return unicode(newvalue[2], newvalue[0] or 'us-ascii') LookupError: unknown encoding: X-UNKNOWN ----------------------------------------------------------------------
Comment By: Mark Crispin (mrcrispin) Date: 2005-04-09 18:48
Message: Logged In: YES user_id=1255784 Our version of mailman is 2.1.5, the current release version, along with customizations made at UW by my predecessor for use with our web pubcookie authentication system. However, the fault occurs in unmodified Mailman code, and he insists that nothing he did would affect this. I call it a security issue because anyone can send a message to a mailman mailing list that will cause digests to fail and be stuck, just by using a bogus character set name in an attachment filename. Not only isn't the message in question sent, but all subsequent messages are also held because of the trap. A denial of service problem *is* a security problem. I don't know how extensive the problem is in Mailman, but I see numerous unicode() calls in the Mailman source that have no protection from error traps. So maybe more than just digests are affected. If you can't reproduce the problem, I'll be happy to provide some of the messages which hung our digests. The problem definitely happens with charset names in encoded- parameters in MIME (attachment filenames). Thank you in advance for your rapid attention. ---------------------------------------------------------------------- Comment By: Tokio Kikuchi (tkikuchi) Date: 2005-04-09 13:57 Message: Logged In: YES user_id=67709 What is your mailman version? I believe i18n charset issues are greatly improved in 2.1.6 beta. BTW, I don't like to call this a security issue. ---------------------------------------------------------------------- Comment By: Mark Crispin (mrcrispin) Date: 2005-04-09 09:38 Message: Logged In: YES user_id=1255784 We've kept copies of the messages which caused the problem in the most recent incident, so if you need help in reproducing/testing we'll be happy to supply them as test data. ---------------------------------------------------------------------- You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=100103&aid=1179487&group_id=103