...
Below are some examples of how to do this using your own script to feed e-mail to SpamAssassin. Aside from the methods below, you can look in the examples folder included with SpamAssassin (located, for example, at /usr/share/doc/spamassassin/examples on Debian installs by default) for other options, like procmail. Detailed information on setting up mail filtering (upon which the examples below are based) can be found in the Postfix FILTER_README documentation.
Note that it is generally good practice to configure Postfix NOT to bounce undeliverable messages, as this can cause backscatter. Either have a 'catch-all' address that all e-mail to unknown senders invalid recipient addresses goes to, or just reject e-mail to invalid recipient addresses. You could also silently discard e-mail to unknown sendersto invalid recipient addresses, although this is a bit hostile to legitimate users trying to send you e-mail. Never bounce e-mail detected as spam; either rewrite it and deliver it, reject it during the SMTP connection, or silently discard it.
bash script to rewrite spam method
This is a simple method to integrate SpamAssassin. All spam e-mail will still come through to your address(es), but it will be clearly marked as spam.
1) Create a bash script to receive e-mail from Postfix and pipe it to SpamAssassin for rewriting. Then forward the rewritten version to Postfix's sendmail implementation:
{{{#!highlight bash
#!/bin/bash
#
# spamfilter.sh
#
# Simple filter to plug SpamAssassin into the Postfix MTA
#
# Modified by Jeremy Morton
#
# This script should probably live at /usr/bin/spamfilter.sh
# ... and have 'chown root:root' and 'chmod 755' applied to it.
#
# Simple filter to plug SpamAssassin into the Postfix MTA
#
# Modified by Jeremy Morton
#
# For use with:
# Postfix 20010228 or later
# SpamAssassin 2.42 or later
#
# Note: Modify the file locations to match suit your particular
# server and installation of SpamAssassin.
# File locations:
# (CHANGE AS REQUIRED TO MATCH SUIT YOUR SERVER) SENDMAIL=/usr/sbin/sendmail SPAMASSASSIN=/usr/bin/spamc
logger <<<"Spam filter piping to SpamAssassin, then to: $SENDMAIL $@"
/bin/cat | ${SPAMASSASSIN} | ${SENDMAIL} "$@"
...
3) Modify the /etc/postfix/master.cf file; first, change the first 'smtp' line of the file to:
| No Format |
|---|
smtp inet n - - - - smtpd -o content_filter=spamfilter
|
Then, add the following (a call to our newly-created spamfilter script) at the end:
| No Format |
|---|
spamfilter unix - n n - - pipe flags=Rq user=spamd argv=/usr/bin/spamfilter.sh -oi -f ${sender} ${recipient} |
...
5) (optional) Tweak SpamAssasin's configuration. Ensure that the below is in /etc/mail/spamassassin/local.cf :and then restart SpamAssassin:
| No Format |
|---|
# Rewrite the subject of suspected spam e-mails #rewrite_header Subject *****SPAM***** rewrite_header Subject *****SPAM***** (_SCORE_) # Just add an X-Spam-Report header to suspected spam, rather than rewriting the content of the e-mail report_safe 0 # Also we want to add a detailed ham report header to even e-mail that ISN'T suspected to be spam add_header ham HAM-Report _REPORT_ # Set the threshold at which a message is considered spam (3 is usually sufficient) required_score 3.0 |
...
advanced bash script to discard
...
'likely spam' method
TODO: describe a similar method to above, but with a Perl script that looks at the rewritten headers and discards spam that is, say, scored higher than 7. That means < 3 is OK, 3-7 is marked as spam but delivered, > 7 is discarded.
...
NOTE: _At the time of writing, the following script will only work with a patched version of spamc from SVN trunk (which can be found at http://svn.apache.org/repos/asf/spamassassin/trunk ). It needs to be patched with the 'Second patch' from the following Bugzilla bug:
https://issues.apache.org/SpamAssassin/show_bug.cgi?id=6890
It introduces a -T (override threshold) parameter which is necessary for this shell script's operation. I'm hoping it will be merged into SpamAssassin trunk one day._
This is a more advanced method to integrate SpamAssassin. The script will set a threshold, and only messages with a score above that threshold will be discarded. Messages between the normal configured SpamAssassin threshold and the threshold set in the script will be considered "grey area" messages, and whilst they will be rewritten as spam by Spamassassin, they will still be delivered to the mailbox. Messages below the normal configured SpamAssassin threshold will be delivered as normal.
1) Create a bash script to receive e-mail from Postfix and pipe it to SpamAssassin for rewriting. Check whether SpamAssassin tells us whether the message is over the threshold or not; if so, discard the message. Otherwise, deliver it through Postfix's sendmail implementation:
{{{#!highlight bash
#!/bin/bash
#
# spamfilter.sh
#
# Advanced filter to plug SpamAssassin into the Postfix MTA
# Only discards messages that are above the threshold that is
# passed in to spamc; other spam messages that are below this
# threshold will be rewritten, but not discarded.
#
# Modified by Jeremy Morton
#
# This script should probably live at /usr/bin/spamfilter.sh
# ... and have 'chown root:root' and 'chmod 755' applied to it.
#
# For use with:
# Postfix 20010228 or later
# ??? SpamAssassin 2.42 or later ??? - should change to the latest version when/if spamc patch gets applied to trunk
# Note: Modify the file locations to suit your particular
# server and installation of SpamAssassin.
# File locations and settings:
# (CHANGE AS REQUIRED TO SUIT YOUR SERVER) THRESHOLD_OFFSET=4.0
# ^ Remember, we're using the threshold offset build, not the absolute value one. Passing in a -T value of 4.0
# will give us an absolute threshold of maybe (base) 3.0 + 4.0 = 7.0. MAX_MESSAGE_SIZE=10485760
# ^ Allow e-mails up to size 10485760 (10MB) - they're not too intensive to process, and any e-mails above that
# size don't get processed, and are assumed to be non-spam. Clearly it's important that we process as many
# messages as possible, and there should be very few messages indeed that exceed 10MB size. SENDMAIL=/usr/sbin/sendmail SPAMASSASSIN=/usr/bin/spamc
# We use -4 to force IPv4 because alas at the time of writing, spamd doesn't listen out for connections on IPv6... logger -s -p mail.notice -t spamfilter <<<"Spam filter piping to SpamAssassin: $SPAMASSASSIN -4 -x -E -s $MAX_MESSAGE_SIZE -T $THRESHOLD_OFFSET" CAUGHT_OUTPUT=${SPAMASSASSIN} -4 -x -E -s $MAX_MESSAGE_SIZE -T $THRESHOLD_OFFSET
SPAMASSASSIN_EXITCODE=$?
| Wiki Markup |
|---|
if \["$SPAMASSASSIN_EXITCODE" -gt 1 \]; then
logger -s -p mail.warning -t spamfilter <<<"Error code $SPAMASSASSIN_EXITCODE processing spam!"
exit $SPAMASSASSIN_EXITCODE elif \["$SPAMASSASSIN_EXITCODE" -eq 1 \]; then
logger -s -p mail.notice -t spamfilter <<<"SpamAssassin says message is likely spam; discarding."
exit 0 fi |
# Message is OK, or grey-area and has been rewritten. Deliver to mailbox. logger -s -p mail.notice -t spamfilter <<<"OK. Piping to sendmail: $SENDMAIL $@"
${SENDMAIL} "$@" <<<"$CAUGHT_OUTPUT" exit $?
}}}
This, of course, assumes you're using the spamd/spamc implementation of SpamAssassin to improve performance - spamc is the command you'll want to invoke to check an e-mail (by setting SPAMASSASSIN to its location).
2) Ensure the newly-created /usr/bin/spamfilter.sh has correct permissions (0755), and is owned by root:root.
3) Modify the /etc/postfix/master.cf file; first, change the first 'smtp' line of the file to:
| No Format |
|---|
smtp inet n - - - - smtpd -o content_filter=spamfilter
|
Then, add the following (a call to our newly-created spamfilter script) at the end:
| No Format |
|---|
spamfilter
unix - n n - - pipe
flags=Rq user=spamd argv=/usr/bin/spamfilter.sh -oi -f ${sender} ${recipient}
|
This setup assumes you have a 'spamd' user for the script to be run as. If you wish to run it as a different user, modify the 'user' argument above.
4) Restart the Postfix service.
Now, Postfix should be sending every incoming e-mail through the spamfilter.sh script (which means it's going through SpamAssassin), with the e-mail headers (and maybe the rest of the e-mail, if it's spam) being rewritten before being delivered to the target mailbox. Optionally, you may wish to tweak SpamAssassin's configuration to cause it to rewrite the e-mails in a more useful way (only rewrite spam mail subject but not whole e-mail, add a 'ham' header to give detailed info on even non-spam, etc.)
5) (optional) Tweak SpamAssasin's configuration. Ensure that the below is in /etc/mail/spamassassin/local.cf and then restart SpamAssassin:
| No Format |
|---|
# Rewrite the subject of suspected spam e-mails
#rewrite_header Subject *****SPAM*****
rewrite_header Subject *****SPAM***** (_SCORE_)
# Just add an X-Spam-Report header to suspected spam, rather than rewriting the content of the e-mail
report_safe 0
# Also we want to add a detailed ham report header to even e-mail that ISN'T suspected to be spam
add_header ham HAM-Report _REPORT_
# Set the threshold at which a message is considered spam (3 is usually sufficient)
required_score 3.0
|
Old alternative method
A major problem with the old method filter (below) is that Postfix attempts to bounce messages flagged as spam. This is a waste of resources as most spams have fake return addresses. Furthermore, if they fake a return address that doesn't belong to them, you may end up bouncing the message to an innocent 3rd party. Enough of this can result in YOU being on an RBL.
...
2) Create filter.sh as follows:
{{{#!highlight sh
#!/bin/sh
# filter.sh
#
# This script redirects mail flagged as spam to a separate account
# You must first create a user account named "spamvac" to hold the flagged mail
...
and create a filter.sh as follows:
{{{#!highlight bash
#!/bin/bash SENDMAIL="/usr/sbin/sendmail -i" SPAMASSASSIN=/usr/bin/spamc
...
Only mail received by SMTP will be scanned with this method, i.e. mail injected with sendmail(1) will not be fed to SpamAssassin (if it were, it too would be fed to the filter script, which would send the e-mail via sendmail, etc. and there would be an infinite loop).