dkim-filter(8) DKIM filter for sendmail


dkim-filter -p socketspec [-a peerlist] [-A] [-b modes] [-c canon] [-C config] [-d domains] [-D] [-f] [-F time] [-i ilist] [-I eilist] [-h] [-k keyfile] [-K] [-l] [-L min] [-m mtas] [-M macro[=value][,...]] [-n] [-o hdrlist] [-P pidfile] [-q] [-r] [-R] [-s selector] [-S signalg] [-t testfile] [-T secs] [-u userid[:group]] [-U popdb] [-v] [-V] [-W] [-x configfile]


dkim-filter implements the DKIM standard for signing and verifying e-mail messages on a per-domain basis.

dkim-filter uses the milter interface, originally distributed as part of version 8.11 of sendmail(8), to provide DKIM signing and/or verifying service for mail transiting a milter-aware MTA.

Most, if not all, of the command line options listed below can also be set using a configuration file. See the -x option for details.


-a peerlist
Identifies a file of "peers" which identifies clients whose connections should be accepted without processing by this filter. The peerlist should contain on each line a hostname, domain name (e.g. ""), IP address, an IPv6 address (including an IPv4 mapped address), or a CIDR-style IP specification (e.g. ""). An entry beginning with a bang ("!") character means "not", allowing exclusions of specific hosts that are otherwise members of larger sets. The order of entries in this file is therefore significant.
Automatically re-start on failures. Use with caution; if the filter fails instantly after it starts, this can cause a tight fork(2) loop. This can be mitigated using some values in the configuration file to limit restarting. See dkim-filter.conf(5).
-b modes
Selects operating modes. modes is a concatenation of characters that indicate which mode(s) of operation are desired. Valid modes are s (signer) and v (verifier). The default is sv except in test mode (see -t below) in which case the default is v.
-c canon
Selects the canonicalization method(s) to be used when signing messages. When verifying, the message's DKIM-Signature: header specifies the canonicalization method. The recognized values are relaxed and simple as defined by the DKIM specification. The default is simple. The value may include two different canonicalizations separated by a slash ("/") character, in which case the first will be applied to the headers and the second to the body.
-C config
Configuration control. See the CONFIGURATION section for details.
-d domain [,...]
A comma-separated list of domains whose mail should be signed by this filter. Mail from other domains will be verified rather than being signed.

The value of this parameter may also be a filename from which domain names will be read. The "#" character in such a file is assumed to indicate a comment. An absolute path must be used (i.e. the first character must be a "/").

In either case, the domain name(s) may contain the special character "*" which is treated as a wildcard character matching zero or more characters in a domain name.

Matching is case-insensitive.

This parameter is not required if -K is in use; in that case, the list of signed domains is implied by the lines in the key file.

Sign subdomains of those listed by the -d option as well as the actual domains.
Normally dkim-filter forks and exits immediately, leaving the service running in the background. This flag suppresses that behaviour so that it runs in the foreground.
-F time
Specifies a fixed time to use when generating signatures. Ignored unless also used in conjunction with -t (see below). The time must be expressed in the usual UNIX time_t (seconds since epoch) format.
Causes dkim-filter to add a header indicating the presence of this filter in the path of the message from injection to delivery. The product's name, version, and the job ID are included in the header's contents.
-i ilist
Identifies a file of internal hosts whose mail should be signed rather than verified. Entries in this file follow the same form as those of the -a option above. If not specified, the default of "" is applied. Naturally, providing a value here overrides the default, so if mail from should be signed, the list provided here should include that address explicitly.
-I eilist
Identifies a file of "external" hosts which may send mail through the server as one of the signing domains without credentials as such. Basically suppresses the "external host (hostname) tried to send mail as (domain)" log messages. Entries in the eilist file should be of the same form as those of the -a option above. The list is empty by default.
Requests multiple-key processing. See also -k below.
-k keyfile
Without -K, gives the location of a PEM-formatted private key to be used for signing all messages. With -K, gives the location of a file listing rules for signing with multiple keys.

In the latter mode, the keyfile should contain a set of lines of the form sender-pattern:signing-domain:keypath where sender-pattern is a pattern to match against message senders (with the special character "*" interpreted as "zero or more characters"), signing-domain is the domain to announce as the signing domain when generating signatures, and keypath is the path to the PEM-formatted private key to be used for signing messages which match the sender-pattern. The selector used in the signature will be the filename portion of keypath.

If the file referenced by keypath cannot be opened, the filter will try again by appending ".pem" and then ".private" before giving up.

Log via calls to syslog(3) any interesting activity.
-L min[%+]
Instructs the verification code to fail messages for which a partial signature was received. There are three possible formats: min indicating at least min bytes of the message must be signed (or if the message is smaller than min then all of it must be signed); min% requiring that at least min percent of the received message must be signed; and min+ meaning there may be no more than min bytes of unsigned data appended to the message for it to be considered valid.
-m mta[,...]
A comma-separated list of MTA names (a la the sendmail(8) DaemonPortOptions Name parameter) whose mail should be signed by this filter. If not set, the MTA name is not used when deciding whether or not a message should be signed.
-M macro[=value][,...]
Defines a set of MTA-provided macros which should be checked to see if the sender has been determined to be a local user and therefore whether or not the message should be signed. If a value is specified, the value of the macro must match the value specified (matching is case-insensitive), otherwise the macro must be defined but may contain any value. Multiple tests may be specified, separated by commas. The set is empty by default, meaning macros are not used when deciding whether or not a message should be signed.

The general format of the string is test1[,test2[,...]] where a "test" is of the form macro[=value1[|value2[|...]]]; if one or more value is defined then the macro must be set to one of the listed values, otherwise the macro must be set but can contain any value.

Parse the configuration file and command line arguments, reporting any errors found, and then exit. The exit value will be 0 if the filter would start up without complaint, or non-zero otherwise.
-o hdrlist
Specifies a list of headers which should be omitted when generating signatures. hdrlist should be a comma-separated list of header names. If an entry in the list names any header which is mandated by the DKIM specification, the entry is ignored. A set of headers is listed in the DKIM specification as "SHOULD NOT" be signed; the default list for this parameter contains those headers (Return-Path, Received, Comments, Keywords, Bcc, Resent-Bcc and DKIM-Signature). To omit no headers, simply use the string "-" (or any string which will match no headers).
-p socketspec
Specifies the socket that should be established by the filter to receive connections from sendmail(8) in order to provide service. socketspec is in one of two forms: local:path which creates a UNIX domain socket at the specified path, or inet:port[@host] which creates a TCP socket on the specified port. If the host is not given as either a hostname or an IP address, the socket will be listening on all interfaces. If neither socket type is specified, local is assumed, meaning the parameter is interpreted as a path at which the socket should be created. This parameter is mandatory.
-P pidfile
Writes the process ID of the filter, once started, to the filename given.
Requests that messages which fail verification be quarantined by the MTA. (Requires a sufficiently recent version of the milter library.)
Checks all messages for compliance with RFC2822 header count requirements. Non-compliant messages are rejected.
When a signature verification fails and the signing site advertises a reporting address (i.e. [email protected] in its policy record), send a structured report to that address containing details needed to reproduce the problem.
-s selector
Defines the name of the selector to be used when signing messages. See the DKIM specification for details.
-S signalg
Selects the signing algorithm to use when generating signatures. If the filter was compiled against version 0.9.8 or later of OpenSSL then both rsa-sha1 and rsa-sha256 are available and the latter is the default. Otherwise, only the former is available and it is (obviously) the default.
-t testfile
Evaluates (verifies) an RFC2822-formatted message found in testfile and exits. The value of testfile may be "-" if the message should be read from standard input.
-T secs
Sets the DNS timeout in seconds. A value of 0 causes an infinite wait. The default is 5. Ignored if not using the asynchronous resolver package. See also the NOTES section below.
-u userid[:group]
Attempts to be come the specified userid before starting operations. The process will be assigned all of the groups and primary group ID of the named userid unless an alternate group is specified.
-U popdb
Requests that the filter consult a POP authentication database for IP addresses that should be allowed for signing. The filter must be specially compiled to enable this feature, since it adds a library dependency.
Increase verbose output during test mode (see -t above). May be specified more than once to request increasing amounts of output.
Print the version number and supported canonicalization and signature algorithms, and then exit without doing anything else.
If logging is enabled (see -l above), issues very detailed logging about the logic behind the filter's decision to either sign a message or verify it. The "W" stands for "Why?!" since the logic behind the decision is non-trivial and can be confusing to administrators not familiar with its operation. A description of how the decision is made can be found in the OPERATION section of this document. This causes a large increase in the amount of log data generated for each message, so it should be limited to debugging use and not enabled for general operation.
-x configfile
Read the named configuration file. See the dkim-filter.conf(5) man page for details. Values in the configuration file are overridden when their equivalents are provided on the command line until a configuration reload occurs. The OPERATION section describes how reloads are triggered.


The value of the -C switch is a comma-separated list of settings of the form result=action which defines what the filter should do with messages that produce certain results. Each result and each action has a full name and an abbreviated name. Either is accepted. Below, the abbreviated name appears in parentheses.
badsignature (bad) the signature found in the message did not verify successfully against the message; dnserror (dns) an error was encountered attempting to retrieve a public key from the nameserver; internal (int) an internal error occurred; nosignature (no) no signature was present on the message; security (sec) the message tripped internal security concerns (e.g. unusually large header blocks). There is also a special result called default (def) whose action is copied onto all of the other results.
accept (a) accept the message; discard (d) discard the message; tempfail (t) temp-fail the message; reject (r) reject the message.

In the interests of minimal initial impact, the defaults for badsignature and nosignature are accept, and the default for the others is tempfail.

Results and actions are processed in order, so use of the default action can be overridden by later specifications. For example, using "def=a,int=t" sets all result actions to "accept" except for internal errors which will generate a temporary failure.


A message will be verified unless it conforms to the signing criteria, which are: (1) the domain on the From: address or Sender: address (if present) must be listed by the -d command line switch or the Domain configuration file setting, and (2) (a) the client connecting to the MTA must have authenticated, or (b) the client connecting to the MTA must be listed in the file referenced by the -i command line switch (or be in the default list for that option), or (c) the client must be connected to a daemon port named by the -m command line switch, or (d) the MTA must have set one or more macros matching the criteria set by the -M command line switch.

When signing a message, a DKIM-Signature: header will be prepended to the message. The signature is computed using the private key provided. You must be running a version of sendmail(8) recent enough to be able to do header prepend operations (8.13.0 or later).

When verifying a message, an Authentication-Results: header will be prepended to indicate the presence of a signature and whether or not it could be validated against the body of the message using the public key advertised by the sender's nameserver. The value of this header can be used by mail user agents to sort or discard messages that were not signed or could not be verified.

Upon receiving SIGUSR1, if the filter was started with a configuration file, it will be re-read and the new values used. Note that any command line overrides provided at startup time will be lost when this is done. Also, the following configuration file values (and their corresponding command line items, if any) are not reloaded through this process: AutoRestart (-A), AutoRestartCount, AutoRestartRate, Background, MilterDebug, PidFile (-P), POPDBFile, Quarantine (-q), QueryCache, Socket (-p), StrictTestMode, TestPublicKeys, UMask, UserID (-u). The filter does not automatically check the configuration file for changes and reload.


The following environment variable(s) can be used to adjust the behaviour of this filter:
The directory to use when creating temporary files. The default is /var/tmp.


When using DNS timeouts (see the -T option above), be sure not to use a timeout that is larger than the timeout being used for interaction between sendmail and the filter. Otherwise, the MTA could abort a message while waiting for a reply from the filter, which in turn is still waiting for a DNS reply.

The POP authentication database is expected to be a Sleepycat DB file (formerly known as a Berkeley DB) in hash format with keys containing the IP address in text form without a terminating NULL. The values of these records are not checked; only the existence of such records is of interest. The filter will attempt to establish a shared lock on the database before reading from it, so any programs which write to the database should keep their lock use to a minimum or else this filter will appear to hang while waiting for the lock operation to complete.

Features that involve specification of IPv4 addresses or CIDR blocks will use the inet_addr(3) function to parse that information. Users should be familiar with the way that function handles the non-trivial cases (for example, "1.2.3/24" and "" are not the same thing).


DKIM is an amalgam of Yahoo!'s DomainKeys proposal, and Cisco's Internet Identified Mail (IIM) proposal.


This man page covers version 2.8.0 of dkim-filter.


Copyright (c) 2005-2008, Sendmail, Inc. and its suppliers. All rights reserved.