spamassassin-milter 0.0.3

Milter for spam filtering with SpamAssassin

SpamAssassin Milter

SpamAssassin Milter is a milter application that filters email through SpamAssassin server using the spamc client. It is a light-weight component that serves to integrate Apache SpamAssassin with a milter-capable MTA (mail server) such as Postfix. Its task is thus helping combat spam on email sites.

SpamAssassin Milter operates as a milter hooked into the MTA’s SMTP protocol handler. It passes incoming messages to SpamAssassin for analysis, and then interprets the response from SpamAssassin and applies suggested changes to the message.

By default, the following modifications are made:

  • Always: Add SpamAssassin headers to the message (headers starting with X-Spam-)
  • Spam only: Rewrite headers Subject From To, if requested
  • Spam only: Replace message body (and rewrite related headers MIME-Version Content-Type, if requested)

Alternatively, messages flagged as spam may be rejected at the SMTP level with an SMTP error reply code.

Both SpamAssassin and SpamAssassin Milter provide various configuration options to alter the default behaviour. See below for a discussion of several configuration and integration approaches.

This application can be used as a replacement for spamass-milt; it has a reduced feature set, but it should be satisfactory for a personal mail server setup. SpamAssassin Milter has been used in such a setup together with Postfix, SpamAssassin, and for delivery Dovecot with LMTP and the Sieve plugin.


This project is a Rust package. Build it with Cargo as usual.

As a milter, this package requires the libmilter C library to be available. Be sure to install the libmilter shared library and header files.

If your distribution does not install a pkg-config metadata file for libmilter, you can use the provided milter.pc file. Put it on the pkg-config path as follows:

PKG_CONFIG_PATH=. cargo build

The integration tests rely on the miltertest utility. Make sure miltertest is available and can be executed when running the integration tests.

Note that until recently miltertest had a bad bug that prevents most integration tests in this package from completing. Make sure you use an up-to-date version of miltertest.


Once installed, SpamAssassin Milter can be by invoked as spamassassin-milter. spamassassin-milter takes one mandatory argument, namely the listening socket of the milter (the socket to which the MTA will connect). The socket spec can be in one of the formats inet:port@host or inet6:port@host (IPv6), or unix:path, for a TCP or UNIX domain socket, respectively.

For example, the following invocation starts SpamAssassin Milter on port 3000:

spamassassin-milter inet:3000@localhost

The available options and flags can be glimpsed by passing the -h flag:

spamassassin-milter -h

More detailed information can be found in the provided man page spamassassin-milter(8).

Setting up SpamAssassin Milter as a system service is easiest by using the provided systemd service file: Edit spamassassin-milter.service with the desired port, install it in /etc/systemd/system, then enable and start the service.


SpamAssassin Milter is designed as a light-weight ‘glue’ application with just a few configuration options; this is intentional, as the SpamAssassin components are themselves already highly configurable.

SpamAssassin Milter is configured by setting command-line options. All options have reasonable defaults that work well with a stock installation of SpamAssassin server (spamd) and client (spamc). You can get started with just picking a socket and things should just work. Some integration options are discussed in subsequent sections.

First-time users may wish to run SpamAssassin Milter with the --dry-run option before ‘going live’. Combined with the --verbose option, this gives accurate insight into the changes that SpamAssassin Milter would apply.

spamassassin-milter --dry-run --verbose inet:3000@localhost

Integration with SpamAssassin

SpamAssassin Milter must be integrated with two SpamAssassin components: the SpamAssassin server itself, called spamd, which does the actual work, and the SpamAssassin client spamc, which serves as an intermediary between the milter and the server.

SpamAssassin configuration

The main SpamAssassin configuration file is /etc/spamassassin/local.conf. See perldoc Mail::SpamAssassin::Conf for detailed information.

The phrase ‘flagged as spam’, which appears throughout this documentation, refers to whether SpamAssassin has marked a message as being spam by adding the header X-Spam-Flag: YES. A message being spam or ham (not spam) is a binary property. The classification threshold can be configured by setting required_score as follows:

required_score 9.0

By default, SpamAssassin creates ‘reports’ for messages it recognises as spam. These reports replace the message body, that is, the message body is rewritten to present a report instead of the original message, and the original message is relegated to a MIME attachment. SpamAssassin Milter by default applies reports.

Reports are controlled with the report_safe configuration parameter. Disable reports as follows:

report_safe 0

In addition, body rewriting can also be suppressed on the SpamAssassin Milter side with the --preserve-body flag.

SpamAssassin may also rewrite the Subject and other headers, for example, adding a prefix ‘***Spam***’. This is not done by default, but may be enabled with a setting like the following:

rewrite_header Subject ***SPAM***

SpamAssassin Milter by default applies header rewriting. Header rewriting can be suppressed on the SpamAssassin Milter side with the --preserve-headers flag.

spamc configuration

spamc can be configured by passing it command-line options, or preferably, by setting the command-line options in the configuration file /etc/spamassassin/spamc.conf.

By default, spamc will try to reach SpamAssassin server on the dedicated port 783, so that a stock installation of SpamAssassin should work with SpamAssassin Milter without further configuration.

If SpamAssassin server listens on a different port or on a UNIX domain socket instead, set the --socket option as appropriate in spamc.conf:


When reports are disabled, it is recommended to use the --headers option.


This option is just a shortcut that causes spamd not to transmit the message body back to spamc. (This option obviously only makes sense and should only be used when SpamAssassin reports are disabled.)

Finally, a pitfall of spamc deserves highlighting: spamc by default tries to resist failure to an extent that it will not indicate failure even if it cannot connect to SpamAssassin server at all (apart from warnings logged to syslog)! If it cannot connect to the server, it simply echoes what it received, and so masks the error condition. This behaviour is labelled ‘safe fallback’ and is perhaps best disabled once the system is set up. Set the following flag:


Integration with Postfix

To integrate with Postfix, ensure the SpamAssassin Milter service is up and running, and then configure its listening socket in /etc/postfix/ as follows:

smtpd_milters = inet:localhost:3000
non_smtpd_milters = $smtpd_milters

After reloading the Postfix configuration, mail will be processed by SpamAssassin Milter.

By default, SpamAssassin Milter will accept, that is, won’t check messages coming from local connections (for example, mail sent locally from the command-line), and messages from authenticated senders (for example, mail submitted via a SASL-authenticated channel).

Integration with mail delivery

A further component that can be useful with SpamAssassin Milter is a Sieve-capable mail delivery agent. A Sieve script can for example look at the X-Spam- SpamAssassin headers of the incoming message, and take action based on those.

As an example, in case Dovecot does mail delivery with LMTP, enable the Sieve plugin for the LMTP protocol, then setup a global Sieve script that files messages flagged as spam into the ‘Junk’ folder:

require "fileinto";

if header :contains "X-Spam-Flag" "YES" {
  fileinto "Junk";


Copyright © 2020 David Bürgin

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.