Escape lookup deferral error message when logging.

[exim.git] / doc / doc-docbook / filter.xfpt

. $Cambridge: exim/doc/doc-docbook/filter.xfpt,v 1.11 2010/05/29 19:26:31 nm4 Exp $

. /////////////////////////////////////////////////////////////////////////////
. This is the primary source of the document that describes Exim's filtering
. facilities. It is an xfpt document that is converted into DocBook XML for
. subsequent conversion into printing and online formats. The markup used
. herein is "standard" xfpt markup, with some extras. The markup is summarized
. in a file called Markup.txt.
. /////////////////////////////////////////////////////////////////////////////

.include stdflags
.include stdmacs
.docbook

. /////////////////////////////////////////////////////////////////////////////
. These lines are processing instructions for the Simple DocBook Processor that
. Philip Hazel has developed as a less cumbersome way of making PostScript and
. PDFs than using xmlto and fop. They will be ignored by all other XML
. processors.
. /////////////////////////////////////////////////////////////////////////////

.literal xml
<?sdop
  foot_right_recto="&chaptertitle;"
  foot_right_verso="&chaptertitle;"
  table_warn_soft_overflow="no"
  toc_chapter_blanks="yes,yes"
  toc_title="Exim's interfaces to mail filtering"
?>
.literal off

.book

. ===========================================================================
. Additional xfpt markup used by this document, over and above the default
. provided in the xfpt library.

. Override the &$ flag to automatically insert a $ with the variable name

.flag &$  $&   "<varname>$"  "</varname>"

. A macro for the common 2-column tables

.macro table2 100pt 300pt
.itable none 0 0 2 $1 left $2 left
.endmacro
. ===========================================================================


. /////////////////////////////////////////////////////////////////////////////
. /////////////////////////////////////////////////////////////////////////////

. This preliminary stuff creates a <bookinfo> entry in the XML. This is removed
. when creating the PostScript/PDF output, because we do not want a full-blown
. title page created for those versions. When fop is being used to create
. PS/PDF, the stylesheet fudges up a title line to replace the text "Table of
. contents". When SDoP is being used, a processing instruction does this job.
. For the other forms of output, the <bookinfo> element is retained and used.

.literal xml
<bookinfo>
<title>Exim's interfaces to mail filtering</title>
<titleabbrev>Exim filtering</titleabbrev>
<date>23 November 2009</date>
<author><firstname>Philip</firstname><surname>Hazel</surname></author>
<authorinitials>PH</authorinitials>
<revhistory><revision>
  <revnumber>4.74</revnumber>
  <date>21 Jan 2011</date>
  <authorinitials>PH</authorinitials>
</revision></revhistory>
<copyright><year>2010</year><holder>University of Cambridge</holder></copyright>
</bookinfo>
.literal off

. /////////////////////////////////////////////////////////////////////////////
. /////////////////////////////////////////////////////////////////////////////


.chapter "Forwarding and filtering in Exim" "CHAPforandfilt"
This document describes the user interfaces to Exim's in-built mail filtering
facilities, and is copyright &copy; University of Cambridge 2007. It
corresponds to Exim version 4.74.



.section "Introduction" "SEC00"
Most Unix mail transfer agents (programs that deliver mail) permit individual
users to specify automatic forwarding of their mail, usually by placing a list
of forwarding addresses in a file called &_.forward_& in their home
directories. Exim extends this facility by allowing the forwarding instructions
to be a set of rules rather than just a list of addresses, in effect providing
&"&_.forward_& with conditions"&. Operating the set of rules is called
&'filtering'&, and the file that contains them is called a &'filter file'&.

Exim supports two different kinds of filter file. An &'Exim filter'& contains
instructions in a format that is unique to Exim. A &'Sieve filter'& contains
instructions in the Sieve format that is defined by RFC 3028. As this is a
standard format, Sieve filter files may already be familiar to some users.
Sieve files should also be portable between different environments. However,
the Exim filtering facility contains more features (such as variable
expansion), and better integration with the host environment (such as the use
of external processes and pipes).

The choice of which kind of filter to use can be left to the end-user, provided
that the system administrator has configured Exim appropriately for both kinds
of filter. However, if interoperability is important, Sieve is the only
choice.

The ability to use filtering or traditional forwarding has to be enabled by the
system administrator, and some of the individual facilities can be separately
enabled or disabled. A local document should be provided to describe exactly
what has been enabled. In the absence of this, consult your system
administrator.

This document describes how to use a filter file and the format of its
contents. It is intended for use by end-users. Both Sieve filters and Exim
filters are covered. However, for Sieve filters, only issues that relate to the
Exim implementation are discussed, since Sieve itself is described elsewhere.

The contents of traditional &_.forward_& files are not described here. They
normally contain just a list of addresses, file names, or pipe commands,
separated by commas or newlines, but other types of item are also available.
The full details can be found in the chapter on the &(redirect)& router in the
Exim specification, which also describes how the system administrator can set
up and control the use of filtering.



.section "Filter operation" "SEC01"
It is important to realize that, in Exim, no deliveries are actually made while
a filter or traditional &_.forward_& file is being processed. Running a filter
or processing a traditional &_.forward_& file sets up future delivery
operations, but does not carry them out.

The result of filter or &_.forward_& file processing is a list of destinations
to which a message should be delivered. The deliveries themselves take place
later, along with all other deliveries for the message. This means that it is
not possible to test for successful deliveries while filtering. It also means
that any duplicate addresses that are generated are dropped, because Exim never
delivers the same message to the same address more than once.




.section "Testing a new filter file" "SECTtesting"
Filter files, especially the more complicated ones, should always be tested, as
it is easy to make mistakes. Exim provides a facility for preliminary testing
of a filter file before installing it. This tests the syntax of the file and
its basic operation, and can also be used with traditional &_.forward_& files.

Because a filter can do tests on the content of messages, a test message is
required. Suppose you have a new filter file called &_myfilter_& and a test
message in a file called &_test-message_&. Assuming that Exim is installed with
the conventional path name &_/usr/sbin/sendmail_& (some operating systems use
&_/usr/lib/sendmail_&), the following command can be used:
.code
/usr/sbin/sendmail -bf myfilter <test-message
.endd
The &%-bf%& option tells Exim that the following item on the command line is
the name of a filter file that is to be tested. There is also a &%-bF%& option,
which is similar, but which is used for testing system filter files, as opposed
to user filter files, and which is therefore of use only to the system
administrator.

The test message is supplied on the standard input. If there are no
message-dependent tests in the filter, an empty file (&_/dev/null_&) can be
used. A supplied message must start with header lines or the &"From&~"& message
separator line that is found in many multi-message folder files. Note that
blank lines at the start terminate the header lines. A warning is given if no
header lines are read.

The result of running this command, provided no errors are detected in the
filter file, is a list of the actions that Exim would try to take if presented
with the message for real. For example, for an Exim filter, the output
.code
Deliver message to: gulliver@lilliput.fict.example
Save message to: /home/lemuel/mail/archive
.endd
means that one copy of the message would be sent to
&'gulliver@lilliput.fict.example'&, and another would be added to the file
&_/home/lemuel/mail/archive_&, if all went well.

The actions themselves are not attempted while testing a filter file in this
way; there is no check, for example, that any forwarding addresses are valid.
For an Exim filter, if you want to know why a particular action is being taken,
add the &%-v%& option to the command. This causes Exim to output the results of
any conditional tests and to indent its output according to the depth of
nesting of &(if)& commands. Further additional output from a filter test can be
generated by the &(testprint)& command, which is described below.

When Exim is outputting a list of the actions it would take, if any text
strings are included in the output, non-printing characters therein are
converted to escape sequences. In particular, if any text string contains a
newline character, this is shown as &"\n"& in the testing output.

When testing a filter in this way, Exim makes up an &"envelope"& for the
message. The recipient is by default the user running the command, and so is
the sender, but the command can be run with the &%-f%& option to supply a
different sender. For example,
.code
/usr/sbin/sendmail -bf myfilter \
   -f islington@never.where <test-message
.endd
Alternatively, if the &%-f%& option is not used, but the first line of the
supplied message is a &"From&~"& separator from a message folder file (not the
same thing as a &'From:'& header line), the sender is taken from there. If
&%-f%& is present, the contents of any &"From&~"& line are ignored.

The &"return path"& is the same as the envelope sender, unless the message
contains a &'Return-path:'& header, in which case it is taken from there. You
need not worry about any of this unless you want to test out features of a
filter file that rely on the sender address or the return path.

It is possible to change the envelope recipient by specifying further options.
The &%-bfd%& option changes the domain of the recipient address, while the
&%-bfl%& option changes the &"local part"&, that is, the part before the @
sign. An adviser could make use of these to test someone else's filter file.

The &%-bfp%& and &%-bfs%& options specify the prefix or suffix for the local
part. These are relevant only when support for multiple personal mailboxes is
implemented; see the description in section &<<SECTmbox>>& below.


.section "Installing a filter file" "SEC02"
A filter file is normally installed under the name &_.forward_& in your home
directory &-- it is distinguished from a conventional &_.forward_& file by its
first line (described below). However, the file name is configurable, and some
system administrators may choose to use some different name or location for
filter files.


.section "Testing an installed filter file" "SEC03"
Testing a filter file before installation cannot find every potential problem;
for example, it does not actually run commands to which messages are piped.
Some &"live"& tests should therefore also be done once a filter is installed.

If at all possible, test your filter file by sending messages from some other
account. If you send a message to yourself from the filtered account, and
delivery fails, the error message will be sent back to the same account, which
may cause another delivery failure. It won't cause an infinite sequence of such
messages, because delivery failure messages do not themselves generate further
messages. However, it does mean that the failure won't be returned to you, and
also that the postmaster will have to investigate the stuck message.

If you have to test an Exim filter from the same account, a sensible precaution
is to include the line
.code
if error_message then finish endif
.endd
as the first filter command, at least while testing. This causes filtering to
be abandoned for a delivery failure message, and since no destinations are
generated, the message goes on to be delivered to the original address. Unless
there is a good reason for not doing so, it is recommended that the above test
be left in all Exim filter files. (This does not apply to Sieve files.)



.section "Details of filtering commands" "SEC04"
The filtering commands for Sieve and Exim filters are completely different in
syntax and semantics. The Sieve mechanism is defined in RFC 3028; in the next
chapter we describe how it is integrated into Exim. The subsequent chapter
covers Exim filtering commands in detail.



.chapter "Sieve filter files" "CHAPsievefilter"
The code for Sieve filtering in Exim was contributed by Michael Haardt, and
most of the content of this chapter is taken from the notes he provided. Since
Sieve is an extensible language, it is important to understand &"Sieve"& in
this context as &"the specific implementation of Sieve for Exim"&.

This chapter does not contain a description of Sieve, since that can be found
in RFC 3028, which should be read in conjunction with these notes.

The Exim Sieve implementation offers the core as defined by RFC 3028,
comparison tests, the subaddress parameter, the &*copy*&, &*envelope*&,
&*fileinto*&, &*notify*&, and &*vacation*& extensions, but not the &*reject*&
extension. Exim does not support message delivery notifications (MDNs), so
adding it just to the Sieve filter (as required for &*reject*&) makes little
sense.

In order for Sieve to work properly in Exim, the system administrator needs to
make some adjustments to the Exim configuration. These are described in the
chapter on the &(redirect)& router in the full Exim specification.


.section "Recognition of Sieve filters" "SEC05"
A filter file is interpreted as a Sieve filter if its first line is
.code
# Sieve filter
.endd
This is what distinguishes it from a conventional &_.forward_& file or an Exim
filter file.



.section "Saving to specified folders" "SEC06"
If the system administrator has set things up as suggested in the Exim
specification, and you use &(keep)& or &(fileinto)& to save a mail into a
folder, absolute files are stored where specified, relative files are stored
relative to &$home$&, and &_inbox_& goes to the standard mailbox location.



.section "Strings containing header names" "SEC07"
RFC 3028 does not specify what happens if a string denoting a header field does
not contain a valid header name, for example, it contains a colon. This
implementation generates an error instead of ignoring the header field in order
to ease script debugging, which fits in with the common picture of Sieve.



.section "Exists test with empty list of headers" "SEC08"
The &*exists*& test succeeds only if all the specified headers exist. RFC 3028
does not explicitly specify what happens on an empty list of headers. This
implementation evaluates that condition as true, interpreting the RFC in a
strict sense.



.section "Header test with invalid MIME encoding in header" "SEC09"
Some MUAs process invalid base64 encoded data, generating junk. Others ignore
junk after seeing an equal sign in base64 encoded data. RFC 2047 does not
specify how to react in this case, other than stating that a client must not
forbid to process a message for that reason. RFC 2045 specifies that invalid
data should be ignored (apparently looking at end of line characters). It also
specifies that invalid data may lead to rejecting messages containing them (and
there it appears to talk about true encoding violations), which is a clear
contradiction to ignoring them.

RFC 3028 does not specify how to process incorrect MIME words. This
implementation treats them literally, as it does if the word is correct but its
character set cannot be converted to UTF-8.



.section "Address test for multiple addresses per header" "SEC10"
A header may contain multiple addresses. RFC 3028 does not explicitly specify
how to deal with them, but since the address test checks if anything matches
anything else, matching one address suffices to satisfy the condition. That
makes it impossible to test if a header contains a certain set of addresses and
no more, but it is more logical than letting the test fail if the header
contains an additional address besides the one the test checks for.



.section "Semantics of keep" "SEC11"
The &(keep)& command is equivalent to
.code
fileinto "inbox";
.endd
It saves the message and resets the implicit keep flag. It does not set the
implicit keep flag; there is no command to set it once it has been reset.



.section "Semantics of fileinto" "SEC12"
RFC 3028 does not specify whether &(fileinto)& should try to create a mail
folder if it does not exist. This implementation allows the sysadmin to
configure that aspect using the &(appendfile)& transport options
&%create_directory%&, &%create_file%&, and &%file_must_exist%&. See the
&(appendfile)& transport in the Exim specification for details.



.section "Semantics of redirect" "SEC13"
Sieve scripts are supposed to be interoperable between servers, so this
implementation does not allow mail to be redirected to unqualified addresses,
because the domain would depend on the system being used. On systems with
virtual mail domains, the default domain is probably not what the user expects
it to be.



.section "String arguments" "SEC14"
There has been confusion if the string arguments to &(require)& are to be
matched case-sensitively or not. This implementation matches them with the
match type &(:is)& (default, see section 2.7.1 of the RFC) and the comparator
&(i;ascii-casemap)& (default, see section 2.7.3 of the RFC). The RFC defines
the command defaults clearly, so any different implementations violate RFC
3028. The same is valid for comparator names, also specified as strings.



.section "Number units" "SEC15"
There is a mistake in RFC 3028: the suffix G denotes gibi-, not tebibyte.
The mistake is obvious, because RFC 3028 specifies G to denote 2^30
(which is gibi, not tebi), and that is what this implementation uses as
the scaling factor for the suffix G.



.section "RFC compliance" "SEC16"
Exim requires the first line of a Sieve filter to be
.code
# Sieve filter
.endd
Of course the RFC does not specify that line. Do not expect examples to work
without adding it, though.

RFC 3028 requires the use of CRLF to terminate a line. The rationale was that
CRLF is universally used in network protocols to mark the end of the line. This
implementation does not embed Sieve in a network protocol, but uses Sieve
scripts as part of the Exim MTA. Since all parts of Exim use LF as the newline
character, this implementation does, too, by default, though the system
administrator may choose (at Exim compile time) to use CRLF instead.

Exim violates RFC 2822, section 3.6.8, by accepting 8-bit header names, so this
implementation repeats this violation to stay consistent with Exim. This is in
preparation for UTF-8 data.

Sieve scripts cannot contain NUL characters in strings, but mail headers could
contain MIME encoded NUL characters, which could never be matched by Sieve
scripts using exact comparisons. For that reason, this implementation extends
the Sieve quoted string syntax with \0 to describe a NUL character, violating
\0 being the same as 0 in RFC 3028. Even without using \0, the following tests
are all true in this implementation. Implementations that use C-style strings
will only evaluate the first test as true.
.code
Subject: =?iso-8859-1?q?abc=00def

header :contains "Subject" ["abc"]
header :contains "Subject" ["def"]
header :matches "Subject" ["abc?def"]
.endd
Note that by considering Sieve to be an MUA, RFC 2047 can be interpreted in a
way that NUL characters truncating strings is allowed for Sieve
implementations, although not recommended. It is further allowed to use encoded
NUL characters in headers, but that's not recommended either. The above example
shows why.

RFC 3028 states that if an implementation fails to convert a character set to
UTF-8, two strings cannot be equal if one contains octets greater than 127.
Assuming that all unknown character sets are one-byte character sets with the
lower 128 octets being US-ASCII is not sound, so this implementation violates
RFC 3028 and treats such MIME words literally. That way at least something
could be matched.

The folder specified by &(fileinto)& must not contain the character sequence
&".."& to avoid security problems. RFC 3028 does not specify the syntax of
folders apart from &(keep)& being equivalent to
.code
fileinto "INBOX";
.endd
This implementation uses &_inbox_& instead.

Sieve script errors currently cause messages to be silently filed into
&_inbox_&.  RFC 3028 requires that the user is notified of that condition.
This may be implemented in the future by adding a header line to mails that
are filed into &_inbox_& due to an error in the filter.



.chapter "Exim filter files" "CHAPeximfilter"
This chapter contains a full description of the contents of Exim filter files.


.section "Format of Exim filter files" "SEC17"
Apart from leading white space, the first text in an Exim filter file must be
.code
# Exim filter
.endd
This is what distinguishes it from a conventional &_.forward_& file or a Sieve
filter file. If the file does not have this initial line (or the equivalent for
a Sieve filter), it is treated as a conventional &_.forward_& file, both when
delivering mail and when using the &%-bf%& testing mechanism. The white space
in the line is optional, and any capitalization may be used. Further text on
the same line is treated as a comment. For example, you could have
.code
#   Exim filter   <<== do not edit or remove this line!
.endd
The remainder of the file is a sequence of filtering commands, which consist of
keywords and data values. For example, in the command
.code
deliver gulliver@lilliput.fict.example
.endd
the keyword is &`deliver`& and the data value is
&`gulliver@lilliput.fict.example`&. White space or line breaks separate the
components of a command, except in the case of conditions for the &(if)&
command, where round brackets (parentheses) also act as separators. Complete
commands are separated from each other by white space or line breaks; there are
no special terminators. Thus, several commands may appear on one line, or one
command may be spread over a number of lines.

If the character # follows a separator anywhere in a command, everything from
# up to the next newline is ignored. This provides a way of including comments
in a filter file.


.section "Data values in filter commands" "SEC18"
There are two ways in which a data value can be input:

.ilist
If the text contains no white space, it can be typed verbatim. However, if it
is part of a condition, it must also be free of round brackets (parentheses),
as these are used for grouping in conditions.
.next
Otherwise, text must be enclosed in double quotation marks. In this case, the
character \ (backslash) is treated as an &"escape character"& within the
string, causing the following character or characters to be treated specially:
.display
&`\n`&   is replaced by a newline
&`\r`&   is replaced by a carriage return
&`\t`&   is replaced by a tab
.endd
.endlist

Backslash followed by up to three octal digits is replaced by the character
specified by those digits, and &`\x`& followed by up to two hexadecimal digits
is treated similarly. Backslash followed by any other character is replaced by
the second character, so that in particular, &`\"`& becomes &`"`& and &`\\`&
becomes &`\`&. A data item enclosed in double quotes can be continued onto the
next line by ending the first line with a backslash. Any leading white space at
the start of the continuation line is ignored.

In addition to the escape character processing that occurs when strings are
enclosed in quotes, most data values are also subject to &'string expansion'&
(as described in the next section), in which case the characters &`$`& and
&`\`& are also significant. This means that if a single backslash is actually
required in such a string, and the string is also quoted, &`\\\\`& has to be
entered.

The maximum permitted length of a data string, before expansion, is 1024
characters.


.section "String expansion" "SECTfilterstringexpansion"
Most data values are expanded before use. Expansion consists of replacing
substrings beginning with &`$`& with other text. The full expansion facilities
available in Exim are extensive. If you want to know everything that Exim can
do with strings, you should consult the chapter on string expansion in the Exim
documentation.

In filter files, by far the most common use of string expansion is the
substitution of the contents of a variable. For example, the substring
.code
$reply_address
.endd
is replaced by the address to which replies to the message should be sent. If
such a variable name is followed by a letter or digit or underscore, it must be
enclosed in curly brackets (braces), for example,
.code
${reply_address}
.endd
If a &`$`& character is actually required in an expanded string, it must be
escaped with a backslash, and because backslash is also an escape character in
quoted input strings, it must be doubled in that case. The following two
examples illustrate two different ways of testing for a &`$`& character in a
message:
.code
if $message_body contains \$ then ...
if $message_body contains "\\$" then ...
.endd
You can prevent part of a string from being expanded by enclosing it between
two occurrences of &`\N`&. For example,
.code
if $message_body contains \N$$$$\N then ...
.endd
tests for a run of four dollar characters.


.section "Some useful general variables" "SEC19"
A complete list of the available variables is given in the Exim documentation.
This shortened list contains the ones that are most likely to be useful in
personal filter files:

&$body_linecount$&: The number of lines in the body of the message.

&$body_zerocount$&: The number of binary zero characters in the body of the
message.

&$home$&: In conventional configurations, this variable normally contains the
user's home directory. The system administrator can, however, change this.

&$local_part$&: The part of the email address that precedes the @ sign &--
normally the user's login name. If support for multiple personal mailboxes is
enabled (see section &<<SECTmbox>>& below) and a prefix or suffix for the local
part was recognized, it is removed from the string in this variable.

&$local_part_prefix$&: If support for multiple personal mailboxes is enabled
(see section &<<SECTmbox>>& below), and a local part prefix was recognized,
this variable contains the prefix. Otherwise it contains an empty string.

&$local_part_suffix$&: If support for multiple personal mailboxes is enabled
(see section &<<SECTmbox>>& below), and a local part suffix was recognized,
this variable contains the suffix. Otherwise it contains an empty string.

&$message_body$&: The initial portion of the body of the message. By default,
up to 500 characters are read into this variable, but the system administrator
can configure this to some other value. Newlines in the body are converted into
single spaces.

&$message_body_end$&: The final portion of the body of the message, formatted
and limited in the same way as &$message_body$&.

&$message_body_size$&: The size of the body of the message, in bytes.

&$message_exim_id$&: The message's local identification string, which is unique
for each message handled by a single host.

&$message_headers$&: The header lines of the message, concatenated into a
single string, with newline characters between them.

&$message_size$&: The size of the entire message, in bytes.

&$original_local_part$&: When an address that arrived with the message is
being processed, this contains the same value as the variable &$local_part$&.
However, if an address generated by an alias, forward, or filter file is being
processed, this variable contains the local part of the original address.

&$reply_address$&: The contents of the &'Reply-to:'& header, if the message
has one; otherwise the contents of the &'From:'& header. It is the address to
which normal replies to the message should be sent.

&$return_path$&: The return path &-- that is, the sender field that will be
transmitted as part of the message's envelope if the message is sent to another
host. This is the address to which delivery errors are sent. In many cases,
this variable has the same value as &$sender_address$&, but if, for example,
an incoming message to a mailing list has been expanded, &$return_path$& may
have been changed to contain the address of the list maintainer.

&$sender_address$&: The sender address that was received in the envelope of
the message. This is not necessarily the same as the contents of the &'From:'&
or &'Sender:'& header lines. For delivery error messages (&"bounce messages"&)
there is no sender address, and this variable is empty.

&$tod_full$&: A full version of the time and date, for example: Wed, 18 Oct
1995 09:51:40 +0100. The timezone is always given as a numerical offset from
GMT.

&$tod_log$&: The time and date in the format used for writing Exim's log files,
without the timezone, for example: 1995-10-12 15:32:29.

&$tod_zone$&: The local timezone offset, for example: +0100.



.section "Header variables" "SECTheadervariables"
There is a special set of expansion variables containing the header lines of
the message being processed. These variables have names beginning with
&$header_$& followed by the name of the header line, terminated by a colon.
For example,
.code
$header_from:
$header_subject:
.endd
The whole item, including the terminating colon, is replaced by the contents of
the message header line. If there is more than one header line with the same
name, their contents are concatenated. For header lines whose data consists of
a list of addresses (for example, &'From:'& and &'To:'&), a comma and newline
is inserted between each set of data. For all other header lines, just a
newline is used.

Leading and trailing white space is removed from header line data, and if there
are any MIME &"words"& that are encoded as defined by RFC 2047 (because they
contain non-ASCII characters), they are decoded and translated, if possible, to
a local character set. Translation is attempted only on operating systems that
have the &[iconv()]& function. This makes the header line look the same as it
would when displayed by an MUA. The default character set is ISO-8859-1, but
this can be changed by means of the &(headers)& command (see below).

If you want to see the actual characters that make up a header line, you can
specify &$rheader_$& instead of &$header_$&. This inserts the &"raw"&
header line, unmodified.

There is also an intermediate form, requested by &$bheader_$&, which removes
leading and trailing space and decodes MIME &"words"&, but does not do any
character translation. If an attempt to decode what looks superficially like a
MIME &"word"& fails, the raw string is returned. If decoding produces a binary
zero character, it is replaced by a question mark.

The capitalization of the name following &$header_$& is not significant.
Because any printing character except colon may appear in the name of a
message's header (this is a requirement of RFC 2822, the document that
describes the format of a mail message) curly brackets must &'not'& be used in
this case, as they will be taken as part of the header name. Two shortcuts are
allowed in naming header variables:

.ilist
The initiating &$header_$&, &$rheader_$&, or &$bheader_$& can be
abbreviated to &$h_$&, &$rh_$&, or &$bh_$&, respectively.
.next
The terminating colon can be omitted if the next character is white space. The
white space character is retained in the expanded string. However, this is not
recommended, because it makes it easy to forget the colon when it really is
needed.
.endlist

If the message does not contain a header of the given name, an empty string is
substituted. Thus it is important to spell the names of headers correctly. Do
not use &$header_Reply_to$& when you really mean &$header_Reply-to$&.


.section "User variables" "SEC20"
There are ten user variables with names &$n0$& &-- &$n9$& that can be
incremented by the &(add)& command (see section &<<SECTadd>>&). These can be
used for &"scoring"& messages in various ways. If Exim is configured to run a
&"system filter"& on every message, the values left in these variables are
copied into the variables &$sn0$& &-- &$sn9$& at the end of the system filter,
thus making them available to users' filter files. How these values are used is
entirely up to the individual installation.


.section "Current directory" "SEC21"
The contents of your filter file should not make any assumptions about the
current directory. It is best to use absolute paths for file names; you can
normally make use of the &$home$& variable to refer to your home directory. The
&(save)& command automatically inserts &$home$& at the start of non-absolute
paths.




.section "Significant deliveries" "SECTsigdel"
When in the course of delivery a message is processed by a filter file, what
happens next, that is, after the filter file has been processed, depends on
whether or not the filter sets up any &'significant deliveries'&. If at least
one significant delivery is set up, the filter is considered to have handled
the entire delivery arrangements for the current address, and no further
processing of the address takes place. If, however, no significant deliveries
are set up, Exim continues processing the current address as if there were no
filter file, and typically sets up a delivery of a copy of the message into a
local mailbox. In particular, this happens in the special case of a filter file
containing only comments.

The delivery commands &(deliver)&, &(save)&, and &(pipe)& are by default
significant. However, if such a command is preceded by the word &"unseen"&, its
delivery is not considered to be significant. In contrast, other commands such
as &(mail)& and &(vacation)& do not set up significant deliveries unless
preceded by the word &"seen"&. The following example commands set up
significant deliveries:
.code
deliver jack@beanstalk.example
pipe $home/bin/mymailscript
seen mail subject "message discarded"
seen finish
.endd
The following example commands do not set up significant deliveries:
.code
unseen deliver jack@beanstalk.example
unseen pipe $home/bin/mymailscript
mail subject "message discarded"
finish
.endd



.section "Filter commands" "SEC222"
The filter commands that are described in subsequent sections are listed
below, with the section in which they are described in brackets:

.table2
.row &(add)&        "&~&~increment a user variable (section &<<SECTadd>>&)"
.row &(deliver)&    "&~&~deliver to an email address (section &<<SECTdeliver>>&)"
.row &(fail)&       "&~&~force delivery failure (sysadmin use) (section &<<SECTfail>>&)"
.row &(finish)&     "&~&~end processing (section &<<SECTfinish>>&)"
.row &(freeze)&     "&~&~freeze message (sysadmin use) (section &<<SECTfreeze>>&)"
.row &(headers)&    "&~&~set the header character set (section &<<SECTheaders>>&)"
.row &(if)&         "&~&~test condition(s) (section &<<SECTif>>&)"
.row &(logfile)&    "&~&~define log file (section &<<SECTlog>>&)"
.row &(logwrite)&   "&~&~write to log file (section &<<SECTlog>>&)"
.row &(mail)&       "&~&~send a reply message (section &<<SECTmail>>&)"
.row &(pipe)&       "&~&~pipe to a command (section &<<SECTpipe>>&)"
.row &(save)&       "&~&~save to a file (section &<<SECTsave>>&)"
.row &(testprint)&  "&~&~print while testing (section &<<SECTtestprint>>&)"
.row &(vacation)&   "&~&~tailored form of &(mail)& (section &<<SECTmail>>&)"
.endtable

The &(headers)& command has additional parameters that can be used only in a
system filter. The &(fail)& and &(freeze)& commands are available only when
Exim's filtering facilities are being used as a system filter, and are
therefore usable only by the system administrator and not by ordinary users.
They are mentioned only briefly in this document; for more information, see the
main Exim specification.



.section "The add command" "SECTadd"
.display
&`     add `&<&'number'&>&` to `&<&'user variable'&>
&`e.g. add 2 to n3`&
.endd

There are 10 user variables of this type, with names &$n0$& &-- &$n9$&. Their
values can be obtained by the normal expansion syntax (for example &$n3$&) in
other commands. At the start of filtering, these variables all contain zero.
Both arguments of the &(add)& command are expanded before use, making it
possible to add variables to each other. Subtraction can be obtained by adding
negative numbers.



.section "The deliver command" "SECTdeliver"
.display
&`     deliver`& <&'mail address'&>
&`e.g. deliver "Dr Livingstone <David@somewhere.africa.example>"`&
.endd

This command provides a forwarding operation. The delivery that it sets up is
significant unless the command is preceded by &"unseen"& (see section
&<<SECTsigdel>>&). The message is sent on to the given address, exactly as
happens if the address had appeared in a traditional &_.forward_& file. If you
want to deliver the message to a number of different addresses, you can use
more than one &(deliver)& command (each one may have only one address).
However, duplicate addresses are discarded.

To deliver a copy of the message to your normal mailbox, your login name can be
given as the address. Once an address has been processed by the filtering
mechanism, an identical generated address will not be so processed again, so
doing this does not cause a loop.

However, if you have a mail alias, you should &'not'& refer to it here. For
example, if the mail address &'L.Gulliver'& is aliased to &'lg303'& then all
references in Gulliver's &_.forward_& file should be to &'lg303'&. A reference
to the alias will not work for messages that are addressed to that alias,
since, like &_.forward_& file processing, aliasing is performed only once on an
address, in order to avoid looping.

Following the new address, an optional second address, preceded by
&"errors_to"& may appear. This changes the address to which delivery errors on
the forwarded message will be sent. Instead of going to the message's original
sender, they go to this new address. For ordinary users, the only value that is
permitted for this address is the user whose filter file is being processed.
For example, the user &'lg303'& whose mailbox is in the domain
&'lilliput.example'& could have a filter file that contains
.code
deliver jon@elsewhere.example errors_to lg303@lilliput.example
.endd
Clearly, using this feature makes sense only in situations where not all
messages are being forwarded. In particular, bounce messages must not be
forwarded in this way, as this is likely to create a mail loop if something
goes wrong.



.section "The save command" "SECTsave"
.display
&`     save `&<&'file name'&>
&`e.g. save $home/mail/bookfolder`&
.endd

This command specifies that a copy of the message is to be appended to the
given file (that is, the file is to be used as a mail folder). The delivery
that &(save)& sets up is significant unless the command is preceded by
&"unseen"& (see section &<<SECTsigdel>>&).

More than one &(save)& command may be obeyed; each one causes a copy of the
message to be written to its argument file, provided they are different
(duplicate &(save)& commands are ignored).

If the file name does not start with a / character, the contents of the
&$home$& variable are prepended, unless it is empty, or the system
administrator has disabled this feature. In conventional configurations, this
variable is normally set in a user filter to the user's home directory, but the
system administrator may set it to some other path. In some configurations,
&$home$& may be unset, or prepending may be disabled, in which case a
non-absolute path name may be generated. Such configurations convert this to an
absolute path when the delivery takes place. In a system filter, &$home$& is
never set.

The user must of course have permission to write to the file, and the writing
of the file takes place in a process that is running as the user, under the
user's primary group. Any secondary groups to which the user may belong are not
normally taken into account, though the system administrator can configure Exim
to set them up. In addition, the ability to use this command at all is
controlled by the system administrator &-- it may be forbidden on some systems.

An optional mode value may be given after the file name. The value for the mode
is interpreted as an octal number, even if it does not begin with a zero. For
example:
.code
save /some/folder 640
.endd
This makes it possible for users to override the system-wide mode setting for
file deliveries, which is normally 600. If an existing file does not have the
correct mode, it is changed.

An alternative form of delivery may be enabled on your system, in which each
message is delivered into a new file in a given directory. If this is the case,
this functionality can be requested by giving the directory name terminated by
a slash after the &(save)& command, for example
.code
save separated/messages/
.endd
There are several different formats for such deliveries; check with your system
administrator or local documentation to find out which (if any) are available
on your system. If this functionality is not enabled, the use of a path name
ending in a slash causes an error.



.section "The pipe command" "SECTpipe"
.display
&`     pipe `&<&'command'&>
&`e.g. pipe "$home/bin/countmail $sender_address"`&
.endd

This command specifies that the message is to be delivered to the specified
command using a pipe. The delivery that it sets up is significant unless the
command is preceded by &"unseen"& (see section &<<SECTsigdel>>&). Remember,
however, that no deliveries are done while the filter is being processed. All
deliveries happen later on. Therefore, the result of running the pipe is not
available to the filter.

When the deliveries are done, a separate process is run, and a copy of the
message is passed on its standard input. The process runs as the user, under
the user's primary group. Any secondary groups to which the user may belong are
not normally taken into account, though the system administrator can configure
Exim to set them up. More than one &(pipe)& command may appear; each one causes
a copy of the message to be written to its argument pipe, provided they are
different (duplicate &(pipe)& commands are ignored).

When the time comes to transport the message, the command supplied to &(pipe)&
is split up by Exim into a command name and a number of arguments. These are
delimited by white space except for arguments enclosed in double quotes, in
which case backslash is interpreted as an escape, or in single quotes, in which
case no escaping is recognized. Note that as the whole command is normally
supplied in double quotes, a second level of quoting is required for internal
double quotes. For example:
.code
pipe "$home/myscript \"size is $message_size\""
.endd
String expansion is performed on the separate components after the line has
been split up, and the command is then run directly by Exim; it is not run
under a shell. Therefore, substitution cannot change the number of arguments,
nor can quotes, backslashes or other shell metacharacters in variables cause
confusion.

Documentation for some programs that are normally run via this kind of pipe
often suggest that the command should start with
.code
IFS=" "
.endd
This is a shell command, and should &'not'& be present in Exim filter files,
since it does not normally run the command under a shell.

However, there is an option that the administrator can set to cause a shell to
be used. In this case, the entire command is expanded as a single string and
passed to the shell for interpretation. It is recommended that this be avoided
if at all possible, since it can lead to problems when inserted variables
contain shell metacharacters.

The default PATH set up for the command is determined by the system
administrator, usually containing at least &_/bin_& and &_/usr/bin_& so that
common commands are available without having to specify an absolute file name.
However, it is possible for the system administrator to restrict the pipe
facility so that the command name must not contain any / characters, and must
be found in one of the directories in the configured PATH. It is also possible
for the system administrator to lock out the use of the &(pipe)& command
altogether.

When the command is run, a number of environment variables are set up. The
complete list for pipe deliveries may be found in the Exim reference manual.
Those that may be useful for pipe deliveries from user filter files are:

.display
&`DOMAIN            `&   the domain of the address
&`HOME              `&   your home directory
&`LOCAL_PART        `&   see below
&`LOCAL_PART_PREFIX `&   see below
&`LOCAL_PART_SUFFIX `&   see below
&`LOGNAME           `&   your login name
&`MESSAGE_ID        `&   the unique id of the message
&`PATH              `&   the command search path
&`RECIPIENT         `&   the complete recipient address
&`SENDER            `&   the sender of the message
&`SHELL             `&   &`/bin/sh`&
&`USER              `&   see below
.endd

LOCAL_PART, LOGNAME, and USER are all set to the same value, namely, your login
id. LOCAL_PART_PREFIX and LOCAL_PART_SUFFIX may be set if Exim is configured to
recognize prefixes or suffixes in the local parts of addresses. For example, a
message addressed to &'pat-suf2@domain.example'& may cause the filter for user
&'pat'& to be run. If this sets up a pipe delivery, LOCAL_PART_SUFFIX is
&`-suf2`& when the pipe command runs. The system administrator has to configure
Exim specially for this feature to be available.

If you run a command that is a shell script, be very careful in your use of
data from the incoming message in the commands in your script. RFC 2822 is very
generous in the characters that are permitted to appear in mail addresses, and
in particular, an address may begin with a vertical bar or a slash. For this
reason you should always use quotes round any arguments that involve data from
the message, like this:
.code
/some/command '$SENDER'
.endd
so that inserted shell meta-characters do not cause unwanted effects.

Remember that, as was explained earlier, the pipe command is not run at the
time the filter file is interpreted. The filter just defines what deliveries
are required for one particular addressee of a message. The deliveries
themselves happen later, once Exim has decided everything that needs to be done
for the message.

A consequence of this is that you cannot inspect the return code from the pipe
command from within the filter. Nevertheless, the code returned by the command
is important, because Exim uses it to decide whether the delivery has succeeded
or failed.

The command should return a zero completion code if all has gone well. Most
non-zero codes are treated by Exim as indicating a failure of the pipe. This is
treated as a delivery failure, causing the message to be returned to its
sender. However, there are some completion codes that are treated as temporary
errors. The message remains on Exim's spool disk, and the delivery is tried
again later, though it will ultimately time out if the delivery failures go on
too long. The completion codes to which this applies can be specified by the
system administrator; the default values are 73 and 75.

The pipe command should not normally write anything to its standard output or
standard error file descriptors. If it does, whatever is written is normally
returned to the sender of the message as a delivery error, though this action
can be varied by the system administrator.



.section "Mail commands" "SECTmail"
There are two commands that cause the creation of a new mail message, neither
of which count as a significant delivery unless the command is preceded by the
word &"seen"& (see section &<<SECTsigdel>>&). This is a powerful facility, but
it should be used with care, because of the danger of creating infinite
sequences of messages. The system administrator can forbid the use of these
commands altogether.

To help prevent runaway message sequences, these commands have no effect when
the incoming message is a bounce (delivery error) message, and messages sent by
this means are treated as if they were reporting delivery errors. Thus, they
should never themselves cause a bounce message to be returned. The basic
mail-sending command is
.display
&`mail [to `&<&'address-list'&>&`]`&
&`     [cc `&<&'address-list'&>&`]`&
&`     [bcc `&<&'address-list'&>&`]`&
&`     [from `&<&'address'&>&`]`&
&`     [reply_to `&<&'address'&>&`]`&
&`     [subject `&<&'text'&>&`]`&
&`     [extra_headers `&<&'text'&>&`]`&
&`     [text `&<&'text'&>&`]`&
&`     [[expand] file `&<&'filename'&>&`]`&
&`     [return message]`&
&`     [log `&<&'log file name'&>&`]`&
&`     [once `&<&'note file name'&>&`]`&
&`     [once_repeat `&<&'time interval'&>&`]`&
&`e.g. mail text "Your message about $h_subject: has been received"`&
.endd
Each <&'address-list'&> can contain a number of addresses, separated by commas,
in the format of a &'To:'& or &'Cc:'& header line. In fact, the text you supply
here is copied exactly into the appropriate header line. It may contain
additional information as well as email addresses. For example:
.code
mail to "Julius Caesar <jc@rome.example>, \
         <ma@rome.example> (Mark A.)"
.endd
Similarly, the texts supplied for &%from%& and &%reply_to%& are copied into
their respective header lines.

As a convenience for use in one common case, there is also a command called
&(vacation)&. It behaves in the same way as &(mail)&, except that the defaults
for the &%subject%&, &%file%&, &%log%&, &%once%&, and &%once_repeat%& options
are
.code
subject "On vacation"
expand file .vacation.msg
log  .vacation.log
once .vacation
once_repeat 7d
.endd
respectively. These are the same file names and repeat period used by the
traditional Unix &(vacation)& command. The defaults can be overridden by
explicit settings, but if a file name is given its contents are expanded only
if explicitly requested.

&*Warning*&: The &(vacation)& command should always be used conditionally,
subject to at least the &(personal)& condition (see section &<<SECTpersonal>>&
below) so as not to send automatic replies to non-personal messages from
mailing lists or elsewhere. Sending an automatic response to a mailing list or
a mailing list manager is an Internet Sin.

For both commands, the key/value argument pairs can appear in any order. At
least one of &%text%& or &%file%& must appear (except with &(vacation)&, where
there is a default for &%file%&); if both are present, the text string appears
first in the message. If &%expand%& precedes &%file%&, each line of the file is
subject to string expansion before it is included in the message.

Several lines of text can be supplied to &%text%& by including the escape
sequence &"\n"& in the string wherever a newline is required. If the command is
output during filter file testing, newlines in the text are shown as &"\n"&.

Note that the keyword for creating a &'Reply-To:'& header is &%reply_to%&,
because Exim keywords may contain underscores, but not hyphens. If the &%from%&
keyword is present and the given address does not match the user who owns the
forward file, Exim normally adds a &'Sender:'& header to the message, though it
can be configured not to do this.

The &%extra_headers%& keyword allows you to add custom header lines to the
message. The text supplied must be one or more syntactically valid RFC 2822
header lines. You can use &"\n"& within quoted text to specify newlines between
headers, and also to define continued header lines. For example:
.code
extra_headers "h1: first\nh2: second\n continued\nh3: third"
.endd
No newline should appear at the end of the final header line.

If no &%to%& argument appears, the message is sent to the address in the
&$reply_address$& variable (see section &<<SECTfilterstringexpansion>>& above).
An &'In-Reply-To:'& header is automatically included in the created message,
giving a reference to the message identification of the incoming message.

If &%return message%& is specified, the incoming message that caused the filter
file to be run is added to the end of the message, subject to a maximum size
limitation.

If a log file is specified, a line is added to it for each message sent.

If a &%once%& file is specified, it is used to hold a database for remembering
who has received a message, and no more than one message is ever sent to any
particular address, unless &%once_repeat%& is set. This specifies a time
interval after which another copy of the message is sent. The interval is
specified as a sequence of numbers, each followed by the initial letter of one
of &"seconds"&, &"minutes"&, &"hours"&, &"days"&, or &"weeks"&. For example,
.code
once_repeat 5d4h
.endd
causes a new message to be sent if at least 5 days and 4 hours have elapsed
since the last one was sent. There must be no white space in a time interval.

Commonly, the file name specified for &%once%& is used as the base name for
direct-access (DBM) file operations. There are a number of different DBM
libraries in existence. Some operating systems provide one as a default, but
even in this case a different one may have been used when building Exim. With
some DBM libraries, specifying &%once%& results in two files being created,
with the suffixes &_.dir_& and &_.pag_& being added to the given name. With
some others a single file with the suffix &_.db_& is used, or the name is used
unchanged.

Using a DBM file for implementing the &%once%& feature means that the file
grows as large as necessary. This is not usually a problem, but some system
administrators want to put a limit on it. The facility can be configured not to
use a DBM file, but instead, to use a regular file with a maximum size. The
data in such a file is searched sequentially, and if the file fills up, the
oldest entry is deleted to make way for a new one. This means that some
correspondents may receive a second copy of the message after an unpredictable
interval. Consult your local information to see if your system is configured
this way.

More than one &(mail)& or &(vacation)& command may be obeyed in a single filter
run; they are all honoured, even when they are to the same recipient.



.section "Logging commands" "SECTlog"
A log can be kept of actions taken by a filter file. This facility is normally
available in conventional configurations, but there are some situations where
it might not be. Also, the system administrator may choose to disable it. Check
your local information if in doubt.

Logging takes place while the filter file is being interpreted. It does not
queue up for later like the delivery commands. The reason for this is so that a
log file need be opened only once for several write operations. There are two
commands, neither of which constitutes a significant delivery. The first
defines a file to which logging output is subsequently written:
.display
&`     logfile `&<&'file name'&>
&`e.g. logfile $home/filter.log`&
.endd
The file name must be fully qualified. You can use &$home$&, as in this
example, to refer to your home directory. The file name may optionally be
followed by a mode for the file, which is used if the file has to be created.
For example,
.code
logfile $home/filter.log 0644
.endd
The number is interpreted as octal, even if it does not begin with a zero.
The default for the mode is 600. It is suggested that the &(logfile)& command
normally appear as the first command in a filter file. Once a log file has
been obeyed, the &(logwrite)& command can be used to write to it:
.display
&`     logwrite "`&<&'some text string'&>&`"`&
&`e.g. logwrite "$tod_log $message_id processed"`&
.endd
It is possible to have more than one &(logfile)& command, to specify writing to
different log files in different circumstances. Writing takes place at the end
of the file, and a newline character is added to the end of each string if
there isn't one already there. Newlines can be put in the middle of the string
by using the &"\n"& escape sequence. Lines from simultaneous deliveries may get
interleaved in the file, as there is no interlocking, so you should plan your
logging with this in mind. However, data should not get lost.



.section "The finish command" "SECTfinish"
The command &(finish)&, which has no arguments, causes Exim to stop
interpreting the filter file. This is not a significant action unless preceded
by &"seen"&. A filter file containing only &"seen finish"& is a black hole.


.section "The testprint command" "SECTtestprint"
It is sometimes helpful to be able to print out the values of variables when
testing filter files. The command
.display
&`     testprint `&<&'text'&>
&`e.g. testprint "home=$home reply_address=$reply_address"`&
.endd
does nothing when mail is being delivered. However, when the filtering code is
being tested by means of the &%-bf%& option (see section &<<SECTtesting>>&
above), the value of the string is written to the standard output.


.section "The fail command" "SECTfail"
When Exim's filtering facilities are being used as a system filter, the
&(fail)& command is available, to force delivery failure. Because this command
is normally usable only by the system administrator, and not enabled for use by
ordinary users, it is described in more detail in the main Exim specification
rather than in this document.


.section "The freeze command" "SECTfreeze"
When Exim's filtering facilities are being used as a system filter, the
&(freeze)& command is available, to freeze a message on the queue. Because this
command is normally usable only by the system administrator, and not enabled
for use by ordinary users, it is described in more detail in the main Exim
specification rather than in this document.



.section "The headers command" "SECTheaders"
The &(headers)& command can be used to change the target character set that is
used when translating the contents of encoded header lines for insertion by the
&$header_$& mechanism (see section &<<SECTheadervariables>>& above). The
default can be set in the Exim configuration; if not specified, ISO-8859-1 is
used. The only currently supported format for the &(headers)& command in user
filters is as in this example:
.code
headers charset "UTF-8"
.endd
That is, &(headers)& is followed by the word &"charset"& and then the name of a
character set. This particular example would be useful if you wanted to compare
the contents of a header to a UTF-8 string.

In system filter files, the &(headers)& command can be used to add or remove
header lines from the message. These features are described in the main Exim
specification.



.section "Obeying commands conditionally" "SECTif"
Most of the power of filtering comes from the ability to test conditions and
obey different commands depending on the outcome. The &(if)& command is used to
specify conditional execution, and its general form is
.display
&`if    `&<&'condition'&>
&`then  `&<&'commands'&>
&`elif  `&<&'condition'&>
&`then  `&<&'commands'&>
&`else  `&<&'commands'&>
&`endif`&
.endd
There may be any number of &(elif)& and &(then)& sections (including none) and
the &(else)& section is also optional. Any number of commands, including nested
&(if)& commands, may appear in any of the <&'commands'&> sections.

Conditions can be combined by using the words &(and)& and &(or)&, and round
brackets (parentheses) can be used to specify how several conditions are to
combine. Without brackets, &(and)& is more binding than &(or)&. For example:
.code
if
$h_subject: contains "Make money" or
$h_precedence: is "junk" or
($h_sender: matches ^\\d{8}@ and not personal) or
$message_body contains "this is not spam"
then
seen finish
endif
.endd
A condition can be preceded by &(not)& to negate it, and there are also some
negative forms of condition that are more English-like.



.section "String testing conditions" "SEC23"
There are a number of conditions that operate on text strings, using the words
&"begins"&, &"ends"&, &"is"&, &"contains"& and &"matches"&. If you want to
apply the same test to more than one header line, you can easily concatenate
them into a single string for testing, as in this example:
.code
if "$h_to:, $h_cc:" contains me@domain.example then ...
.endd
If a string-testing condition name is written in lower case, the testing
of letters is done without regard to case; if it is written in upper case
(for example, &"CONTAINS"&), the case of letters is taken into account.

.display
&`     `&<&'text1'&>&` begins `&<&'text2'&>
&`     `&<&'text1'&>&` does not begin `&<&'text2'&>
&`e.g. $header_from: begins "Friend@"`&
.endd

A &"begins"& test checks for the presence of the second string at the start of
the first, both strings having been expanded.

.display
&`     `&<&'text1'&>&` ends `&<&'text2'&>
&`     `&<&'text1'&>&` does not end `&<&'text2'&>
&`e.g. $header_from: ends "public.com.example"`&
.endd

An &"ends"& test checks for the presence of the second string at the end of
the first, both strings having been expanded.

.display
&`     `&<&'text1'&>&` is `&<&'text2'&>
&`     `&<&'text1'&>&` is not `&<&'text2'&>
&`e.g. $local_part_suffix is "-foo"`&
.endd

An &"is"& test does an exact match between the strings, having first expanded
both strings.

.display
&`     `&<&'text1'&>&` contains `&<&'text2'&>
&`     `&<&'text1'&>&` does not contain `&<&'text2'&>
&`e.g. $header_subject: contains "evolution"`&
.endd

A &"contains"& test does a partial string match, having expanded both strings.

.display
&`     `&<&'text1'&>&` matches `&<&'text2'&>
&`     `&<&'text1'&>&` does not match `&<&'text2'&>
&`e.g. $sender_address matches "(bill|john)@"`&
.endd

For a &"matches"& test, after expansion of both strings, the second one is
interpreted as a regular expression. Exim uses the PCRE regular expression
library, which provides regular expressions that are compatible with Perl.

The match succeeds if the regular expression matches any part of the first
string. If you want a regular expression to match only at the start or end of
the subject string, you must encode that requirement explicitly, using the
&`^`& or &`$`& metacharacters. The above example, which is not so constrained,
matches all these addresses:
.code
bill@test.example
john@some.example
spoonbill@example.com
littlejohn@example.com
.endd
To match only the first two, you could use this:
.code
if $sender_address matches "^(bill|john)@" then ...
.endd
Care must be taken if you need a backslash in a regular expression, because
backslashes are interpreted as escape characters both by the string expansion
code and by Exim's normal processing of strings in quotes. For example, if you
want to test the sender address for a domain ending in &'.com'& the regular
expression is
.code
\.com$
.endd
The backslash and dollar sign in that expression have to be escaped when used
in a filter command, as otherwise they would be interpreted by the expansion
code. Thus, what you actually write is
.code
if $sender_address matches \\.com\$
.endd
An alternative way of handling this is to make use of the &`\N`& expansion
flag for suppressing expansion:
.code
if $sender_address matches \N\.com$\N
.endd
Everything between the two occurrences of &`\N`& is copied without change by
the string expander (and in fact you do not need the final one, because it is
at the end of the string). If the regular expression is given in quotes
(mandatory only if it contains white space) you have to write either
.code
if $sender_address matches "\\\\.com\\$"
.endd
or
.code
if $sender_address matches "\\N\\.com$\\N"
.endd

If the regular expression contains bracketed sub-expressions, numeric
variable substitutions such as &$1$& can be used in the subsequent actions
after a successful match. If the match fails, the values of the numeric
variables remain unchanged. Previous values are not restored after &(endif)&.
In other words, only one set of values is ever available. If the condition
contains several sub-conditions connected by &(and)& or &(or)&, it is the
strings extracted from the last successful match that are available in
subsequent actions. Numeric variables from any one sub-condition are also
available for use in subsequent sub-conditions, because string expansion of a
condition occurs just before it is tested.


.section "Numeric testing conditions" "SEC24"
The following conditions are available for performing numerical tests:

.display
&`     `&<&'number1'&>&` is above `&<&'number2'&>
&`     `&<&'number1'&>&` is not above `&<&'number2'&>
&`     `&<&'number1'&>&` is below `&<&'number2'&>
&`     `&<&'number1'&>&` is not below `&<&'number2'&>
&`e.g. $message_size is not above 10k`&
.endd

The <&'number'&> arguments must expand to strings of digits, optionally
followed by one of the letters K or M (upper case or lower case) which cause
multiplication by 1024 and 1024x1024 respectively.


.section "Testing for significant deliveries" "SEC25"
You can use the &(delivered)& condition to test whether or not any previously
obeyed filter commands have set up a significant delivery. For example:
.code
if not delivered then save mail/anomalous endif
.endd
&"Delivered"& is perhaps a poor choice of name for this condition, because the
message has not actually been delivered; rather, a delivery has been set up for
later processing.


.section "Testing for error messages" "SEC26"
The condition &(error_message)& is true if the incoming message is a bounce
(mail delivery error) message. Putting the command
.code
if error_message then finish endif
.endd
at the head of your filter file is a useful insurance against things going
wrong in such a way that you cannot receive delivery error reports. &*Note*&:
&(error_message)& is a condition, not an expansion variable, and therefore is
not preceded by &`$`&.


.section "Testing a list of addresses" "SEC27"
There is a facility for looping through a list of addresses and applying a
condition to each of them. It takes the form
.display
&`foranyaddress `&<&'string'&>&` (`&<&'condition'&>&`)`&
.endd
where <&'string'&> is interpreted as a list of RFC 2822 addresses, as in a
typical header line, and <&'condition'&> is any valid filter condition or
combination of conditions. The &"group"& syntax that is defined for certain
header lines that contain addresses is supported.

The parentheses surrounding the condition are mandatory, to delimit it from
possible further sub-conditions of the enclosing &(if)& command. Within the
condition, the expansion variable &$thisaddress$& is set to the non-comment
portion of each of the addresses in the string in turn. For example, if the
string is
.code
B.Simpson <bart@sfld.example>, lisa@sfld.example (his sister)
.endd
then &$thisaddress$& would take on the values &`bart@sfld.example`& and
&`lisa@sfld.example`& in turn.

If there are no valid addresses in the list, the whole condition is false. If
the internal condition is true for any one address, the overall condition is
true and the loop ends. If the internal condition is false for all addresses in
the list, the overall condition is false. This example tests for the presence
of an eight-digit local part in any address in a &'To:'& header:
.code
if foranyaddress $h_to: ( $thisaddress matches ^\\d{8}@ ) then ...
.endd
When the overall condition is true, the value of &$thisaddress$& in the
commands that follow &(then)& is the last value it took on inside the loop. At
the end of the &(if)& command, the value of &$thisaddress$& is reset to what it
was before. It is best to avoid the use of multiple occurrences of
&(foranyaddress)&, nested or otherwise, in a single &(if)& command, if the
value of &$thisaddress$& is to be used afterwards, because it isn't always
clear what the value will be. Nested &(if)& commands should be used instead.

Header lines can be joined together if a check is to be applied to more than
one of them. For example:
.code
if foranyaddress $h_to:,$h_cc: ....
.endd
This scans through the addresses in both the &'To:'& and the &'Cc:'& headers.


.section "Testing for personal mail" "SECTpersonal"
A common requirement is to distinguish between incoming personal mail and mail
from a mailing list, or from a robot or other automatic process (for example, a
bounce message). In particular, this test is normally required for &"vacation
messages"&.

The &(personal)& condition checks that the message is not a bounce message and
that the current user's email address appears in the &'To:'& header. It also
checks that the sender is not the current user or one of a number of common
daemons, and that there are no header lines starting &'List-'& in the message.
Finally, it checks the content of the &'Precedence:'& header line, if there is
one.

You should always use the &(personal)& condition when generating automatic
responses. This example shows the use of &(personal)& in a filter file that is
sending out vacation messages:
.code
if personal then
mail to $reply_address
subject "I am on holiday"
file $home/vacation/message
once $home/vacation/once
once_repeat 10d
endif
.endd
It is tempting, when writing commands like the above, to quote the original
subject in the reply. For example:
.code
subject "Re: $h_subject:"
.endd
There is a danger in doing this, however. It may allow a third party to
subscribe you to an opt-in mailing list, provided that the list accepts bounce
messages as subscription confirmations. (Messages sent from filters are always
sent as bounce messages.) Well-managed lists require a non-bounce message to
confirm a subscription, so the danger is relatively small.

If prefixes or suffixes are in use for local parts &-- something which depends
on the configuration of Exim (see section &<<SECTmbox>>& below) &-- the tests
for the current user are done with the full address (including the prefix and
suffix, if any) as well as with the prefix and suffix removed. If the system is
configured to rewrite local parts of mail addresses, for example, to rewrite
&`dag46`& as &`Dirk.Gently`&, the rewritten form of the address is also used in
the tests.



.section "Alias addresses for the personal condition" "SEC28"
It is quite common for people who have mail accounts on a number of different
systems to forward all their mail to one system, and in this case a check for
personal mail should test all their various mail addresses. To allow for this,
the &(personal)& condition keyword can be followed by
.display
&`alias `&<&'address'&>
.endd
any number of times, for example:
.code
if personal alias smith@else.where.example
            alias jones@other.place.example
then ...
.endd
The alias addresses are treated as alternatives to the current user's email
address when testing the contents of header lines.


.section "Details of the personal condition" "SEC29"
The basic &(personal)& test is roughly equivalent to the following:
.code
not error_message and
$message_headers does not contain "\nList-Id:" and
$message_headers does not contain "\nList-Help:" and
$message_headers does not contain "\nList-Subscribe:" and
$message_headers does not contain "\nList-Unsubscribe:" and
$message_headers does not contain "\nList-Post:" and
$message_headers does not contain "\nList-Owner:" and
$message_headers does not contain "\nList-Archive:" and
(
"${if def:h_auto-submitted:{present}{absent}}" is "absent" or
$header_auto-submitted: is "no"
) and
$header_precedence: does not contain "bulk" and
$header_precedence: does not contain "list" and
$header_precedence: does not contain "junk" and
foranyaddress $header_to:
( $thisaddress contains "$local_part$domain" ) and
not foranyaddress $header_from:
(
$thisaddress contains "$local_part@$domain" or
$thisaddress contains "server@" or
$thisaddress contains "daemon@" or
$thisaddress contains "root@" or
$thisaddress contains "listserv@" or
$thisaddress contains "majordomo@" or
$thisaddress contains "-request@" or
$thisaddress matches  "^owner-[^@]+@"
)
.endd
The variable &$local_part$& contains the local part of the mail address of
the user whose filter file is being run &-- it is normally your login id. The
&$domain$& variable contains the mail domain. As explained above, if aliases
or rewriting are defined, or if prefixes or suffixes are in use, the tests for
the current user are also done with alternative addresses.




.section "Testing delivery status" "SEC30"
There are two conditions that are intended mainly for use in system filter
files, but which are available in users' filter files as well. The condition
&(first_delivery)& is true if this is the first process that is attempting to
deliver the message, and false otherwise. This indicator is not reset until the
first delivery process successfully terminates; if there is a crash or a power
failure (for example), the next delivery attempt is also a &"first delivery"&.

In a user filter file &(first_delivery)& will be false if there was previously
an error in the filter, or if a delivery for the user failed owing to, for
example, a quota error, or if forwarding to a remote address was deferred for
some reason.

The condition &(manually_thawed)& is true if the message was &"frozen"& for
some reason, and was subsequently released by the system administrator. It is
unlikely to be of use in users' filter files.


.section "Multiple personal mailboxes" "SECTmbox" "SEC31"
The system administrator can configure Exim so that users can set up variants
on their email addresses and handle them separately. Consult your system
administrator or local documentation to see if this facility is enabled on your
system, and if so, what the details are.

The facility involves the use of a prefix or a suffix on an email address. For
example, all mail addressed to &'lg303-'&<&'something'&> would be the property
of user &'lg303'&, who could determine how it was to be handled, depending on
the value of <&'something'&>.

There are two possible ways in which this can be set up. The first possibility
is the use of multiple &_.forward_& files. In this case, mail to &'lg303-foo'&,
for example, is handled by looking for a file called &_.forward-foo_& in
&'lg303'&'s home directory. If such a file does not exist, delivery fails
and the message is returned to its sender.

The alternative approach is to pass all messages through a single &_.forward_&
file, which must be a filter file so that it can distinguish between the
different cases by referencing the variables &$local_part_prefix$& or
&$local_part_suffix$&, as in the final example in section &<<SECTex>>& below.

It is possible to configure Exim to support both schemes at once. In this case,
a specific &_.forward-foo_& file is first sought; if it is not found, the basic
&_.forward_& file is used.

The &(personal)& test (see section &<<SECTpersonal>>&) includes prefixes and
suffixes in its checking.



.section "Ignoring delivery errors" "SEC43"
As was explained above, filtering just sets up addresses for delivery &-- no
deliveries are actually done while a filter file is active. If any of the
generated addresses subsequently suffers a delivery failure, an error message
is generated in the normal way. However, if a filter command that sets up a
delivery is preceded by the word &"noerror"&, errors for that delivery,
and any deliveries consequent on it (that is, from alias, forwarding, or
filter files it invokes) are ignored.



.section "Examples of Exim filter commands" "SECTex"
Simple forwarding:

.code
# Exim filter
deliver baggins@rivendell.middle-earth.example
.endd

Vacation handling using traditional means, assuming that the &_.vacation.msg_&
and other files have been set up in your home directory:

.code
# Exim filter
unseen pipe "/usr/ucb/vacation \"$local_part\""
.endd

Vacation handling inside Exim, having first created a file called
&_.vacation.msg_& in your home directory:

.code
# Exim filter
if personal then vacation endif
.endd

File some messages by subject:

.code
# Exim filter
if $header_subject: contains "empire" or
$header_subject: contains "foundation"
then
save $home/mail/f+e
endif
.endd

Save all non-urgent messages by weekday:

.code
# Exim filter
if $header_subject: does not contain "urgent" and
$tod_full matches "^(...),"
then
save $home/mail/$1
endif
.endd

Throw away all mail from one site, except from postmaster:

.code
# Exim filter
if $reply_address contains "@spam.site.example" and
$reply_address does not contain "postmaster@"
then
seen finish
endif
.endd

Handle multiple personal mailboxes:

.code
# Exim filter
if $local_part_suffix is "-foo"
then
save $home/mail/foo
elif $local_part_suffix is "-bar"
then
save $home/mail/bar
endif
.endd