Postfix errors and backlog from discuss@blu list

[jabr-mNDKBlG2WHs at public.gmane.org (John Abreau)]

We've been having some weird problems since installing SpamAssassin
on the BLU mailserver. This morning I removed the settings from
postfix's master.cf file so that smtpd no longer uses the content filter.
Doing that seems to have fixed the problem for new mail, but
all the old queued mail is still stuck because it has the content
filter hardwired into its queue files, which are in a binary format.

I tried editing one of the files in vi to remove the content filter reference,
but that only resulted in that message being moved from the deferred
director to the corrupt directory in var/spool/postfix.

The errors in maillog show the following for the content filter:


> Jun 12 11:18:24 pegasus postfix/pipe[22783]: fatal: open lock file pid/unix.spamchk: cannot open file: Permission denied
> Jun 12 11:18:25 pegasus postfix/master[12418]: warning: process /usr/libexec/postfix/pipe pid 22783 exit status 1
> Jun 12 11:18:25 pegasus postfix/master[12418]: warning: /usr/libexec/postfix/pipe: bad command startup -- throttling


Given that smtpd no longer uses the filter, I assume that all these are from
attempts to process the backlog of messages from before I edited
master.conf and reloaded postfix.

The "spamchk" script is something I found when I googled how to
integrate spamassassin with postfix. I modified the script to bypass
spamassassin entirely, since I can't find a way to edit the queued
messages to eliminate the content-filter from them, and that
had no effect at first.

The "fatal: open lock file pid/unix.spamchk: cannot open file" seems to
refer to /var/spool/postfix/pid/unix.spamchk. I chmod'ed this to 666,
hoping that doing so might eliminate that part of the problem.
When I did that, all the messages queued in /var/spool/postfix/deferred
quickly moved to /var/spool/postfix/active, but they're still not being
delivered.  New mail is, and the logfile is still getting new entries saying
"fatal: open lock file pid/unix.spamchk: cannot open file".

Meanwhile, it looks like new traffic is being delivered successfully.

The log entry about "warning: /usr/libexec/postfix/pipe: bad command startup"
presumably is telling me that the content-filter definition in master.cf
is somehow invalid, but I'm having no luck figuring out why. Here's how
it's defined:


> spamchk   unix  -       n       n       -       10       pipe
>      flags=Rq user=filter argv=/usr/local/sbin/spamchk -f ${sender} -- ${recipient}

The spamchk script currently looks like this:

> #! /bin/sh
>
> # Variables
> SENDMAIL="/usr/sbin/sendmail.postfix -i"
> TMP=/var/tmp/out.$$
>
> # Clean up when done or when aborting.
> trap "rm -f $TMP" 0 1 2 3 15
>
> #### Temporarily make this a dummy filter, to flush out postfix queue
> cat > $TMP
> $SENDMAIL "$@" < $TMP
> exit $?


Can anyone familiar with postfix see where the problem could be?



-- 
John Abreau / Executive Director, Boston Linux & Unix
AIM abreauj / JABBER jabr-iMZfmuK6BGBxLiRVyXs8+g at public.gmane.org / YAHOO abreauj / SKYPE zusa_it_mgr
Email jabr-mNDKBlG2WHs at public.gmane.org / WWW http://www.abreau.net / PGP-Key-ID 0xD5C7B5D9
PGP-Key-Fingerprint 72 FB 39 4F 3C 3B D6 5B E0 C8 5A 6E F1 2C BE 99