loudermilk.org

Spam

Everyone hates spam. I'm doing several things to cut down on spam:

• Using sendmail's anti-relaying rules. This doesn't cut down on my spam, but keeps spammers from using my system to send out more spam. I've noticed that RoadRunner does an open relay test on me about once a month, so at least they're trying to make sure their customers run secure mail servers.
• Greylisting using milter-greylist. This causes mail from servers that haven't talked to my mail server recently to be rejected with a temporary failure. The theory is that real mail servers will try again, but spam software won't bother. I've been running this since mid-2006, and it's cut way down on the spam reaches me.
• Filtering locally-deliverable mail through SpamAssassin, which tags probable spam messages by adding some headers and altering the subject line and body. It decides what's spam by giving points to each message for things common in spam (phrases like "make money fast," or the use of large fonts), and if the score is above a configurable threshold, it marks the message as spam.
• Enabling SpamAssassin's Bayesian classifier, which can be trained to learn what spam looks like.
• Enabling Vipul's Razor checks in SpamAssassin. This checks the "fingerprint" of each message against Razor's servers to see if anyone has reported the message as spam.
• Enabling DCC checks in SpamAssassin. This checks each message's fingerprint against the DCC servers, much like Razor. But with DCC, people don't report whether the message is spam or not, they just report that they received that message. The DCC check tells me how many people have received that message before. If the number of recipients is very large, it's probably spam.

Here's a breakdown of how many of the messages my mail server has processed this week were rejected or tagged as spam:

And here's a graph of how much of that spam is addressed to me:

Navigation