Exim Internet Mailer

1. Specifying a system filter

The name of the file that contains the system filter must be specified by setting system_filter. If you want the filter to run under a uid and gid other than root, you must also set system_filter_user and system_filter_group as appropriate. For example:

system_filter = /etc/mail/exim.filter
system_filter_user = exim

If a system filter generates any deliveries directly to files or pipes (via the save or pipe commands), transports to handle these deliveries must be specified by setting system_filter_file_transport and system_filter_pipe_transport, respectively. Similarly, system_filter_reply_transport must be set to handle any messages generated by the reply command.

2. Testing a system filter

You can run simple tests of a system filter in the same way as for a user filter, but you should use -bF rather than -bf, so that features that are permitted only in system filters are recognized.

If you want to test the combined effect of a system filter and a user filter, you can use both -bF and -bf on the same command line.

3. Contents of a system filter

The language used to specify system filters is the same as for users’ filter files. It is described in the separate end-user document Exim’s interface to mail filtering. However, there are some additional features that are available only in system filters; these are described in subsequent sections. If they are encountered in a user’s filter file or when testing with -bf, they cause errors.

There are two special conditions which, though available in users’ filter files, are designed for use in system filters. The condition first_delivery is true only for the first attempt at delivering a message, and manually_thawed is true only if the message has been frozen, and subsequently thawed by an admin user. An explicit forced delivery counts as a manual thaw, but thawing as a result of the auto_thaw setting does not.

Warning: If a system filter uses the first_delivery condition to specify an “unseen” (non-significant) delivery, and that delivery does not succeed, it will not be tried again. If you want Exim to retry an unseen delivery until it succeeds, you should arrange to set it up every time the filter runs.

When a system filter finishes running, the values of the variables $n0 – $n9 are copied into $sn0 – $sn9 and are thereby made available to users’ filter files. Thus a system filter can, for example, set up “scores” to which users’ filter files can refer.

4. Additional variable for system filters

The expansion variable $recipients, containing a list of all the recipients of the message (separated by commas and white space), is available in system filters. It is not available in users’ filters for privacy reasons.

5. Defer, freeze, and fail commands for system filters

There are three extra commands (defer, freeze and fail) which are always available in system filters, but are not normally enabled in users’ filters. (See the allow_defer, allow_freeze and allow_fail options for the redirect router.) These commands can optionally be followed by the word text and a string containing an error message, for example:

fail text "this message looks like spam to me"

The keyword text is optional if the next character is a double quote.

The defer command defers delivery of the original recipients of the message. The fail command causes all the original recipients to be failed, and a bounce message to be created. The freeze command suspends all delivery attempts for the original recipients. In all cases, any new deliveries that are specified by the filter are attempted as normal after the filter has run.

The freeze command is ignored if the message has been manually unfrozen and not manually frozen since. This means that automatic freezing by a system filter can be used as a way of checking out suspicious messages. If a message is found to be all right, manually unfreezing it allows it to be delivered.

The text given with a fail command is used as part of the bounce message as well as being written to the log. If the message is quite long, this can fill up a lot of log space when such failures are common. To reduce the size of the log message, Exim interprets the text in a special way if it starts with the two characters << and contains >> later. The text between these two strings is written to the log, and the rest of the text is used in the bounce message. For example:

fail "<<filter test 1>>Your message is rejected \
     because it contains attachments that we are \
     not prepared to receive."

Take great care with the fail command when basing the decision to fail on the contents of the message, because the bounce message will of course include the contents of the original message and will therefore trigger the fail command again (causing a mail loop) unless steps are taken to prevent this. Testing the error_message condition is one way to prevent this. You could use, for example

if $message_body contains "this is spam" and not error_message
then fail text "spam is not wanted here" endif

though of course that might let through unwanted bounce messages. The alternative is clever checking of the body and/or headers to detect bounces generated by the filter.

The interpretation of a system filter file ceases after a defer, freeze, or fail command is obeyed. However, any deliveries that were set up earlier in the filter file are honoured, so you can use a sequence such as

mail ...
freeze

to send a specified message when the system filter is freezing (or deferring or failing) a message. The normal deliveries for the message do not, of course, take place.

6. Adding and removing headers in a system filter

Two filter commands that are available only in system filters are:

headers add <string>
headers remove <string>

The argument for the headers add is a string that is expanded and then added to the end of the message’s headers. It is the responsibility of the filter maintainer to make sure it conforms to RFC 2822 syntax. Leading white space is ignored, and if the string is otherwise empty, or if the expansion is forced to fail, the command has no effect.

You can use “\n” within the string, followed by white space, to specify continued header lines. More than one header may be added in one command by including “\n” within the string without any following white space. For example:

headers add "X-header-1: ....\n  \
             continuation of X-header-1 ...\n\
             X-header-2: ...."

Note that the header line continuation white space after the first newline must be placed before the backslash that continues the input string, because white space after input continuations is ignored.

The argument for headers remove is a colon-separated list of header names. This command applies only to those headers that are stored with the message; those that are added at delivery time (such as Envelope-To: and Return-Path:) cannot be removed by this means. If there is more than one header with the same name, they are all removed.

The headers command in a system filter makes an immediate change to the set of header lines that was received with the message (with possible additions from ACL processing). Subsequent commands in the system filter operate on the modified set, which also forms the basis for subsequent message delivery. Unless further modified during routing or transporting, this set of headers is used for all recipients of the message.

During routing and transporting, the variables that refer to the contents of header lines refer only to those lines that are in this set. Thus, header lines that are added by a system filter are visible to users’ filter files and to all routers and transports. This contrasts with the manipulation of header lines by routers and transports, which is not immediate, but which instead is saved up until the message is actually being written (see section 48.6).

If the message is not delivered at the first attempt, header lines that were added by the system filter are stored with the message, and so are still present at the next delivery attempt. Header lines that were removed are still present, but marked “deleted” so that they are not transported with the message. For this reason, it is usual to make the headers command conditional on first_delivery so that the set of header lines is not modified more than once.

Because header modification in a system filter acts immediately, you have to use an indirect approach if you want to modify the contents of a header line. For example:

headers add "Old-Subject: $h_subject:"
headers remove "Subject"
headers add "Subject: new subject (was: $h_old-subject:)"
headers remove "Old-Subject"

7. Setting an errors address in a system filter

In a system filter, if a deliver command is followed by

errors_to <some address>

in order to change the envelope sender (and hence the error reporting) for that delivery, any address may be specified. (In a user filter, only the current user’s address can be set.) For example, if some mail is being monitored, you might use

unseen deliver monitor@spying.example errors_to root@local.example

to take a copy which would not be sent back to the normal error reporting address if its delivery failed.

8. Per-address filtering

In contrast to the system filter, which is run just once per message for each delivery attempt, it is also possible to set up a system-wide filtering operation that runs once for each recipient address. In this case, variables such as $local_part_data and $domain_data can be used, and indeed, the choice of filter file could be made dependent on them. This is an example of a router which implements such a filter:

central_filter:
  check_local_user
  driver = redirect
  domains = +local_domains
  file = /central/filters/$local_part_data
  no_verify
  allow_filter
  allow_freeze

The filter is run in a separate process under its own uid. Therefore, either check_local_user must be set (as above), in which case the filter is run as the local user, or the user option must be used to specify which user to use. If both are set, user overrides.

Care should be taken to ensure that none of the commands in the filter file specify a significant delivery if the message is to go on to be delivered to its intended recipient. The router will not then claim to have dealt with the address, so it will be passed on to subsequent routers to be delivered in the normal way.