Message-ID: <44BA240C.12130.55988FE@stuart.cyberdelix.net>
Date: Sun Jul 16 11:33:45 2006
From: stuart at cyberdelix.net (lsi)
Subject: throwing the book at spam

http://www.cyberdelix.net/tech/kaboom.htm


This page is to help you kill spammers, err, I mean spam, here's the 
blueprints to my silver bullet, which (when combined with my other 
filters) kills 99.75% of my spam. That 0.25% corresponds to about 1 
message per day (and it is the target of further work). 

where this filter fits

This is the Filter of Last Resort. Reason being, it's very 
aggressive. You'll see why below. For this reason, this filter should 
be used last, after all the other filters. This way, this filter will 
only ever deal with the dregs, which means it's not so dangerous. 

As this is the Filter of Last Resort, it does NOT include filtering 
for all kinds of spam. Rather, it is designed to kill the spams that 
the other filters miss. 

It is dangerous. Most filters mark the spam and let it be. This 
filter kills it. No Deleted Items, no Recycle Bin, no undo, no 'are 
you sure'. Bang bang, dead dead. All that is left is a logfile entry. 

automatic whitelist maintenance

To minimise the chances of a legit mail being terminated, this filter 
includes a "make whitelist" command. This command tells the filter to 
collect all email addresses from inside a number of other files (my 
address books), eliminate the duplicates and save the list to disk. 
This list is then used by the "move whitelisted" command, which moves 
any message containing whitelisted strings to a separate folder (the 
"whitebox"). 

This filter also supports three other whitelists, these are 
good_senders, good_recipients and good_subjects. Any mail containing 
a whitelisted string in the correct location is automatically 
"whiteboxed" (moved to the whitebox). 

how it works

This filter is actually built of 11 special-purpose filters. Any mail 
matching one of these tests is deleted. The filters are as follows: 

missing_addressee (missing 'To:' or 'for' field) 
missing_sender (missing 'From:' field) 
unlikely_chars (non-alphabetic subject or sender) 
unlikely_dates (message date too old, or in future) 
bounces (mail delivery failure, etc) 
blacklisted (bad_senders/bad_recipients/bad_subjects) 
gifs_attached (message has an attached GIF image) 
X-RBL (message contains X-RBL-Warning: headerline) 
X-DNS (message contains X-DNS-Warning: headerline) 
X-SVF (message contains X-Sender-Verification-Failed: headerline) 
analyse_received (Received: line invalid - see below) 

These tests are fairly self-explanatory, with the exception of the 
analyse_received test. This test analyses the significant Received: 
headerline inside each mail (there are usually several Received: 
lines, but only one is relevant for our purpose). Any mail with an 
invalid Received: line is deleted. The tests for validity are as 
follows: 

IP_missing 
IP_obfuscation 
IP_unreversible 
by-line_not_present 
sending_SMTP_server_unresolvable 
sending_hostname_not_provided 

If these tests all pass, the message is then tested for a mismatch 
between the sender's hostname and the hostname of the sender recorded 
by the receiver. Again, a fail results in the message being deleted. 

why it works

Spammers try and get their messages through by hiding, disguising or 
armouring their spams. This filter spends most of its time looking 
for evidence of armour. It assumes that an attempt at armouring means 
the mail is spam. 

Note that this approach is very unforgiving toward badly configured, 
but legitimate systems, or systems using non-standard data formats. 
Another reason to run this filter last. Mistakes can be minimised by 
keeping the whitelists up-to-date, and encouraging all to run RFC-
compliant nodes. 

And no, I'm not worried about posting my blueprints. Spammers are 
welcome to use less obfuscation - this will send them straight into 
the jaws of standard spam filters, but life's a bitch eh. They can 
use more obfuscation, be my guest cos I need some extra handles to 
kill that last 0.25%. 

caveats

These notes are posted ahead of any software release, so as to 
maximise the damage they can cause. 

This is developmental software. It works, but only in the development 
environment. In particular, it supports Pegasus Mail ONLY. 

Addressbooks must be TEXT files, or they will not be processed by the 
whitelister. 

The whitebox must currently be processed manually. 

greetz

Thanks have got to go to the twits out there sending me 1000+ spams a 
day. Without your contribution, I would never have had the sample 
size I needed.

---
Stuart Udall
stuart at@...erdelix.dot net - http://www.cyberdelix.net/

--- 
 * Origin: lsi: revolution through evolution (192:168/0.2)

Powered by blists - more mailing lists