This is an example global mail filter that uses an embedded Perl script. "Embedded" means that the Perl interpreter is loaded once, and the same Perl code is repeatedly called to accept or reject incoming messages, one by one. Perl filtering is relatively time consuming (compared to filtering in C or C++), and excessive delays in mail filters result in incoming mail being deferred (rejected with a temporary error code). Therefore the perlfilter wrapper can create multiple perlfilter processes, so that multiple processes are used to filter incoming mail.
perlfilter requires Perl 5.004 or higher. The best way to create a Perl filter is to start with the sample filter, /usr/lib/courier/perlfilter-example.pl. This filter reject messages that contain an excessively long Date: header (designed to crash certain poorly-written mail clients). Use it as a basis for writing your own filter. You can install your filter in any convenient location, then initialize the /etc/courier/filters/perlfilter configuration file, as described below. Run filterctl start perlfilter to activate filtering (if necessary, run courierfilter start to start the mail filtering subsystem).
Most of the ugly details of connecting the Perl script to Courier's mail filtering engine is taken care of by the sample perlfilter-example.pl script. One big no-no: the script MAY NOT change the current directory. Anything else goes, for the most part. Loading other modules and classes, pretty much anything else you can do with Perl, is allowed.
The Perl script, just like any other mail filtering module, receives a pointer to a data file and one or more control files, each time a message is submitted to Courier for delivery. The sample script calls the filterdata() function to process the data file. The data file contains the actual message. The filtercontrol() function is called to process each control file. The control file contains recipient and message metadata. There may be more than one control file for each message. The example script includes an implementation of filterdata() that blocks messages with corrupted headers. The example script doesn't do anything interesting with filtercontrol().
filterdata() and filtercontrol() must return an empty string if no serious objections are raised for this message. Any other return string is interpreted as an SMTP-style error code that is used to reject the message. Care must be taken that any error messages are formatted strictly according to the format of SMTP error messages (even though the message may not actually come in via SMTP).
A lot of the Perl glue code is based on examples from the perlembed manual page, and other sources.
perlfilter uses the following configuration files. Changes to the following files do not take effect until the filter has been stopped and restarted.
If this file exists and contains the word "all", perlfilter will create its socket in /var/lib/courier/allfilters, otherwise the socket will be created in /var/lib/courier/filters, see courierfilter(8) for more information.
This file contains a number that sets how many perlfilter processes are created. The default is 5 processes. There's always an extra perlfilter process that's used to clean up crashed child processes.
This file MUST exist and it must contain a single line of text with the filename of the Perl script to load.
This is a sample Perl script of the kind that /etc/courier/filters/perlfilter points to. Use it as an example of writing your own Perl filters.
Please exercise good judgment in writing Perl-based filters. They should be reasonably fast, and do not allocate megabytes of memory. They should not be very promiscuous in creating global Perl variables, and should clean up after themselves. The current Perl wrapper does not destroy the Perl symbol table after each call to the filter script. However, do not take that for granted. This may change in the future.