This page describes how to use SpamAssassin to keep junk email under control.

Introduction

You will probably want to set up SpamAssassin to detect junk e-mail for you. SpamAssassin is a program for categorizing e-mail as spam based on a wide range of criteria. It indicates its decisions by adding special headers to messages.

Please note that we will never reject any spam email before it hits your filtering rules. It is up to you to decide how to classify the email that hits your inbox.

Enabling spam detection

We use a custom tool called setsa to determine whether your email should be run through SpamAssassin. To enable SpamAssassin for mail to your UNIX account, run

setsa on

To later disable it, run

setsa off

To check whether you've enabled it or not, run

setsa

You can similarly enable or disable SpamAssassin for a virtual mailbox address by adding it as the first argument to setsa; for example, setsa user@domain.com on enables SpamAssassin for user@domain.com if you have DomTool permissions for domain.com.

Please don't enable this if you can't commit to following the training procedure below when SpamAssassin makes an incorrect classification! SpamAssassin makes small mistakes over time, and this can interact poorly with its automatic learning of which message properties signify spam. If you don't correct its small misclassifications, then these increase the chance of misclassifying future messages, which itself leads to more faulty learning, vicious cycle style.

Moving spam email to a different folder

The above procedure only asks SpamAssassin to examine your mail and add extra headers indicating its verdict, spam or legit. To use these headers to move junk mail to a folder called Spam in your IMAP mailbox, copy the following template to ~/.public/.forward. This is an Exim filter that looks for SpamAssassin headers that indicate spamhood. You need to create a Spam folder manually to use this. You can modify this template to save spam to other places, if you don't use IMAP or prefer another scheme. (If you already have a ~/.public/.forward file because you forward all of your mail to another account elsewhere, then you can ignore this section. You should use that e-mail provider's spam filtering services.)

# Exim filter

logfile $home/.logs/mail/exim.log

if
    "${if def:h_X-Spam-Flag {def}{undef}}" is "def"
then
    save $home/Maildir/.Spam/
    finish
endif

SpamAssassin flags spam with a spamminess level of 5.0 or higher. You can use the X-Spam-Level: header to customize your own filter to your own liking, however. As an example, you can see NathanKennedy's .forward file on the MemberManual/Email/EximFilter page.

Training

One way that SpamAssassin spots spam is by using statistical (Bayesian) analysis. This requires lots of training data to work properly.

Sometimes this analysis will make mistakes, and you'll want to perform the electronic equivalent of slapping it with a newspaper. The way to do that is to deposit misclassified mail in special system-wide IMAP folders, one called SiteSpam for spam that SpamAssassin missed and one called SiteHam for good messages that were erroneously marked as spam.

If you ever run into this situation, here's how you can feed our system-wide trainer:

1. First, this is only going to work if you are using IMAP (our webmail is using IMAP, so that can be used). If you're not, or if you have other sources of spam or ham that you'd like handled specially, place a support request on the portal.

2. Use your IMAP client's "subscribe" feature to subscribe to SiteSpam and/or SiteHam, which should appear in the SpamAssassin mailbox inside the shared tree. If you don't see the shared tree, then file a support request.

3. When you want a message to be used as an example of spam or ham, place a copy of it in the appropriate folder.
4. Every five minutes, our faithful spamhound will sniff these folders, update its data, and clear their contents.

If you would like to automate this process somewhat, check out FeedingSpamAssassin. For the curious and the sysadmins out there, SpamAssassinAdmin gives more details on how we set this up.

CategoryMemberManual CategoryNeedsWork