Re: Corrupted messages...

[Dan Nelson]

In the last episode (Jun 09), Chuck Yerkes said:
> We're testing out spamass (2.54 right now) with a group of users. 
> We're starting to see problems and I've turned logging way up to try
> to guess where it's coming from.
> 
> We're marking up Subjects ("Subject: SPAM: orig subj here") and
> including the report in the body of the message.  It's done the MIME
> stuff just fine so far.
> 
> The first symptom was that spams came through with no subject at all
> and the spam was reported within as normal.
> 
> Then we got reports that the attachments within were 'corrupted.'
> 
> One of *US* got a message that was deeply corrupted - filled with
> crap, various headers and a mush of 8-bit and, IMPORTANT, bits of
> mail to OTHER PEOPLE.  (see below)
>
> It looks like a file looks when several processes all write
> to it at once.

I have not seen this particular corruption myself.  Sendmail should do
a good job of locking the mail spool files to prevent multiple writes,
and even if it fails, what you usually wil end up seeing is part of
message A, then the entirety of message B after it.  I don't think
you'd see interleaving like you have.
 
> So memory is corrupted somewhere.  spamd?  spamass-milter?
> The milter code from sendmail?  I dunno.

Most likely the problem is either spamass-milter or libmilter.  both
spamd and sendmail should fork new copies of themselves for each
message, so they shouldn't cause parts of one message to end up in
another one.

If you can stand it, try running spamass-milter with watchmalloc, using
either MALLOC_DEBUG=WATCH (slow) or MALLOC_DEBUG=RW (very slow), which
should cause a coredump on bad memory accesses.  I've run
spamass-milter under dmalloc and valgrind on my systems with no
problems, so if there is a problem it may be OS-dependant.
 
-- 
        Dan Nelson
        address@hidden