Sendmail X README

Claus Aßmann


Introduction to sendmail X

Sendmail X is a Message Transfer System

This distribution contains the source code for sendmail1.1 X which implements a message transfer system (MTS)1.2. It supports the Simple Mail Transfer Protocol (SMTP) as specified by RFC 2821 [Kle01] and various extensions, e.g., STARTTLS [Hof99], AUTH [Mye99], PIPELINING [Fre00], as well as other protocols, e.g., LMTP [Mye96].

sendmail X.0 is intended to be used as a secure and efficient mail gateway. It does not provide any mail content modification capabilities, e.g., masquerading of addresses or changing (addition, removal) of headers. Later versions will probably add such capabilities.

Main Components of sendmail X

sendmail X is a modularized message transfer system consisting of five (or more) persistent processes, four of which are multi-threaded. A central queue manager (QMGR) controls SMTP servers (SMTPS) and SMTP clients (SMTPC) to receive and send e-mails, an address resolver (SMAR) provides lookups in various maps including DNS for mail routing, and a main control program (MCP) starts the others processes and watches over their execution. The queue manager organizes the flow of messages through the system and provides measures to avoid overloading the local or remote systems by implementing a central control instance.

More information about each component will be given in the appropriate sections. Complete documentation and background information can be found in [Aßmb]. Section 7.1 describes the data flow in sendmail X, the following is a brief summary. Figure 1.1 shows the interaction of the various components and databases1.3. Incoming messages are accepted by an SMTP server (SMTPS) which stores the messages in the content database (CDB). The envelope information, i.e., sender and recipients, is stored by the queue manager in an incoming queue (IQDB) and written to disk to the incoming queue backup database (IBDB). For a delivery, the envelope information must be transferred into the active queue (AQ). The scheduler in QMGR takes recipient envelopes from AQ and creates transactions which are given to an SMTP client (SMTPC) for delivery. An SMTP client takes the transaction information and tries to send a message whose content is read from CDB. After a successful delivery attempt a record is written to IBDB that logs this information. The deferred envelope database (DEFEDB) is only used if a message cannot be delivered during the first attempt.

Figure 1.1: Sendmail X: Overall Structure

\begin{picture}(120, 120)\epsfxsize 120mm

This version of sendmail X does not come with a local delivery agent nor a mail submission program. See Sections 4.3.1 and 4.2 which programs can be used to achieve the desired functionality.


The document ``Sendmail X: Requirements, Architecture, and Functional Specification'' [Aßmb] provides the background about the sendmail X design, its architecture, as well as the functional specification, and details about the implementation.

Typographical Conventions

In this documentation, a command written as

$ command

should be executed as an unprivileged user. Only a command written as

# command

should be executed as the superuser.

If a command contains components that need to be replaced by values that depend on the environment or the local configuration, then it is usually written as a macro, e.g., $LOGFILE.

A number in parentheses behind a command or function refers to the manual section, e.g., syslog(3), as it is usual for Unix documents.


This document has been written for sendmail X., see also the greeting of the SMTP server and the version output of the main components. See Section 6.8 for information about version naming.

Current State

There are still some error conditions which may not be handled gracefully, i.e., in case of some resource problems (e.g., out of memory or out of disk space) the system may abort; however, this is common for most open source MTAs which simply abort if they are running out of memory. See Section 4.10.1 how to deal with those conditions. The software is running since 2004-01-01 as MTS on the main machine of the author without any significant problem, i.e., it never lost any mail.

Providing Feedback

Please report bugs and provide feedback either to the developers list[Aßma] (if you are subscribed) or directly to1.4:

< smx + feedback (at) sendmailx . org >

or in the usenet newsgroup comp.mail.sendmail.

Feedback about the code, the documentation (including typographical, syntactical, and grammatical errors, pointing out parts that are not well enough explained, etc.), as well as patches and enhancements are highly appreciated.

For the Impatient

For those who do not want to read the entire documentation, it is adviced to read at least sections 2.2 and 2.4, and the appropriate section of Chapter 4.

Building, Testing, and Installing sendmail X

Verifying the Source Code Distribution

The source code is distributed as a (compressed) tar file and is accompanied by a PGP signature file which has the same name as the tar file plus the ending .sig. To verify the integrity of the source code PGP [PGP] or GPG [Gnu] are required as well as the sendmail PGP signing keys [Sen]:

$ gpg -verify smX-$VERSION.tar.gz.sig
$ pgp smX-$VERSION.tar.gz.sig smX-$VERSION.tar.gz

Further information, especially about warnings or possible errors, can be found in the documentation for PGP or GPG.

Building sendmail X

sendmail X uses a configure file generated by GNU autoconf for configuration. Hence you can build it (after verifying and unpacking the distribution) as follows:

$ mkdir obj.$OS && cd obj.$OS && $PATHTO/smX-$VERSION/configure $OPTIONS

Obviously you have to replace $OS and $VERSION as well as $PATHTO. It is also possible to build sendmail X in the source tree, however, this is discouraged:

$ ./configure && make && make check

Notes: do not run this as root; this is not just a basic security measure (only use a privileged account if it is really required), but most of the programs refuse to run with root privileges. It might be useful to save the output of these commands2.1for later inspection.

Compile-Time Configuration Options

Beside the usual configure options like -prefix a few sendmail X specific configuration options are available:

Enable check for STARTTLS support. The default is yes, i.e., configure tries to determine whether OpenSSL is available on the machine. Requires OpenSSL 0.9.6 or newer [Ope]. Note: check the OpenSSL website [Ope] for security announcement and be aware that due to the complexity of the software it may cause (security) problems.

Enable check for AUTH support. The default is yes, i.e., configure tries to determine whether Cyrus SASL v2 is available on the machine. Requires Cyrus SASL version 2.1.18 or newer [Cyr]. Notes:
  1. check and for security announcement and be aware that due to the complexity of the software it may cause (security) problems.
  2. If Cyrus SASL uses Berkeley DB then it is necessary that the version which has been used during compilation matches the version that it is linked against.

Path to directory containing Cyrus SASL v2 library.

Path to directory containing Cyrus SASL v2 include files.

sendmail X ships with a modified version of Berkeley DB 4.3.28 which is built and used by default. To use a different version of Berkeley DB (it must be 4.3, 4.2, or 4.1), e.g., one that is part of the host OS, specify -disable-included-bdb.


  1. If you do not use the Berkeley DB version that comes with sendmail X, make sure you run all the tests. For example, with Berkeley DB 4.2.50 on OpenBSD 3.2/i386 at least one of the test programs fails and hence this combination must not be used. Moreover, if you encounter a problem using some other BDB version then you must try to reproduce the problem with the shipped version before reporting a possible bug.
  2. Do not use Berkeley DB 4.3.27/28 in 64 bit mode on Solaris 5.8/9 as it crashes at least in those configurations2.2. This bug is fixed in the version that is distributed with sendmail X.

Path to directory containing Berkeley DB library. This option is only needed if -disable-included-bdb is used and Berkeley DB is not installed in a location that the compiler or linker use by default.

Path to directory containing Berkeley DB include files. This option is only needed if -disable-included-bdb is used and Berkeley DB is not installed in a location that the compiler uses by default. Example:
$ B=/usr/local/BerkeleyDB.4.3
$ $PATHTO/smX-$VERSION/configure --with-bdb-libdir=$B/lib \

Enable policy milter protocol, see Chapter 5.

To get the current list of configuration options, use configure -help.

Test Programs

$ make check

will run all test programs; currently those tests take about eighty minutes to run on a standard workstation. For each of the test programs one line is printed to denote whether the test succeeded, i.e., the output consists of lines with the marker PASS: or FAIL: and the name of the test program program. Additional output might be generated by the test programs themselves, e.g.,

2 of 2 tests completed successfully,

or some debug output. The debug output may even indicate an error, but only a final FAIL: indicates a test failure. Some tests depend on compilation options and are only conditionally enabled; others may depend on environment variables, see 2.3.1. For disabled tests SKIP is shown.

Since some of the tests may fail (see Section 2.3.2) and make will usually stop after encountering an error, it might be required to use

$ make -i check >check.out 2>&1

to perform all tests.

Environment Variables used by Test Programs

Environment variables can be used to disable some test programs if required or change the behavior of some test programs. These environment variables and their effects are:

In this example the DNS timeout is set to 60 seconds and tests that take a very long time are disabled:

$ make -i check

Known Test Program Problems

For more information about possible test program problems see Section 12.3.2. For problems with programs in the contrib/ directory, see contrib/README.

Installing sendmail X

sendmail X needs several users to provide separation of privileges and to enhance security. Currently there are four required accounts (the numbers for uid and gid are examples only); the last one listed below (smx) is not really required:

smxs:*:260:260:Sendmail X SMTPS:/nonexistent:/sbin/nologin
smxq:*:261:261:Sendmail X QMGR:/nonexistent:/sbin/nologin
smxc:*:262:262:Sendmail X SMTPC:/nonexistent:/sbin/nologin
smxm:*:263:263:Sendmail X misc:/nonexistent:/sbin/nologin
smx:*:264:264:Sendmail X other:/nonexistent:/sbin/nologin

with the corresponding groups:


To check whether the required users and groups exist, run

$ ./misc/ -p

(in the build directory); see below how to override the default values for the user and group names.

A shell script to setup the directories, files, etc. as described below is available in misc/ This script is modified by configure to create misc/ (in the build directory) which is invoked when

# make install

is called. Most defaults in the installation script misc/ can be overridden with environment variables (default is listed in square brackets):

Important Notes:

  1. The users and groups must be created before make install is invoked.

  2. misc/ will not overwrite existing files or directories, hence it does not work for upgrading a system if configuration files or directory/file owners need to changed.

Directories, Files, and Permissions

make install (i.e., misc/ will create all the required directories and files with the correct permissions provided the users and groups have been set up properly. This section shows explains what the created structure looks like.

The CDB directories (0-9, A-F) must be owned by smxs and have group smxq with the permissions 0771:

drwxrwx--x  2 smxs  smxq        0/

Note: this means that everyone with access to the machine can guess the name of content files (see Section 10.1 for the format; the names can also be read from the logfiles if those are world-readable) and list (ls(1)) them, however, they cannot access the content files as those are owned by smxs with mode 0640 and group smxc, e.g.,

-rw-r-----  1 smxs  smxc  1993 Jul  9 21:19 2/S000000000006B1D200

The main (DEFEDB) and incoming queues (IBDB) must belong to smxq and should not accessible by anyone else:

drwx------  2 smxq  smxq        defedb/
drwx------  2 smxq  smxq        ibdb/
drwx------  2 smxq  smxq        ibdb/ibdb/

Mailertable, aliases map, and other maps for SMAR (see Section 3.9.3) should belong to smxm and can be readable as local conventions require:

-rw-r--r--  1 smxm  smxm         mt
-rw-r--r--  1 smxm  smxm         aliases.db

In general, maps should be owned by the user id of the program that uses them, e.g., smxq owns the QMGR configuration map qmgr_conf.db (see Section 3.8.1).

The sendmail X configuration file can either belong to root or the generic sendmail X user:

-rw-r--r--  1 smx   smx          smx.conf

The directories in which the communication sockets between QMGR and the other programs are located must belong to smxq and be group accessible for the corresponding program:

drwxrws---  2 smxq  smxm        qmsmar/
drwxrws---  2 smxq  smxc        qmsmtpc/
drwxrws---  2 smxq  smxs        qmsmtps/

The directory in which the communication socket between MCP and SMTPS is located must belong to smxs:

drwxr-x---  2 smxs  smxs        smtps/

The logfiles must be owned by the corresponding user and may have relaxed group (or even world) read permissions:

-rw-r-----  1 smxq  operator   qmgr.log
-rw-r-----  1 smxm  operator   smar.log
-rw-r-----  1 smxc  operator   smtpc.log
-rw-r-----  1 smxs  operator   smtps.log

To check whether an installation was successful, run

# ./misc/ -P

(in the build directory).

Run-Time Configuration of sendmail X


Configuration of sendmail X can be done via command line parameters or via a configuration file (the latter is preferred, the former offers only a small subset of the available configuration options). If a configuration file and command line options are specified, then the options are currently processed in order, i.e., later settings override earlier ones for the same options. Information about the former is available by invoking a program with the option -h (MCP currently uses syslog(3) instead of stderr), it will show the usage as well as the default values. The syntax of the configuration files is specified in the following sections. To actually use a configuration file, the option -f $CONFIGFILE must be used, otherwise the programs use only the built-in default values, but not a configuration file. Option '-V' can be used to show version information, specifying '-V' multiple times shows more detail, e.g., '-VVVVV' will show the configuration data including the default value for (almost) every option, and '-VVVVVV' will also show all available flags.

Some configuration options can be set via Berkeley DB hash maps, these maps are: qmgr_conf for QMGR (see Section 3.8.1) and access for SMTPS (indirectly via the address resolver, see Section 3.9.3).

Configuration File Syntax

The grammar for a sendmail X configuration file is very simple:

conf ::= entries
entries ::= entry *
entry ::= option $\vert$ section
section ::= keyword [name ] "{" entries "}" [";"]
option ::= option-name "=" rhs
rhs ::= value ";" $\vert$ "{" value-list "}" [";"]

A configuration file consists of entries, each entry is either an option or a section. An option has a name, an equal sign, and a value terminated by a semicolon or a (bracketed) list of values separated by comma. A section consists of a keyword, an optional name, and a (bracketed) sequence of entries. Keywords and options are not case sensitive. The layout of a configuration file does not matter, i.e., indentation and line breaks are irrelevant (in general, but see below for strings).

Configuration File Values

Values in a configuration file are usually strings or numbers. If a string is used, then it should be quoted, unless it contains no special characters which are treated specially by the grammar. If a string is very long it can be broken into substrings spread out over several lines (just like strings in ANSI C), e.g.,

  somemessage = "this is a very long string which is spread "
     "out over several lines because otherwise it is too "
     "hard too read.";

Numeric values can have the usual prefixes (known from the programming language C) of 0x for hexadecimal (with digits 0 to 9, A to F, and a to f) and 0 for octal (with digits 0 to 7). Valid boolean values are 0, false, off for false, and 1, true, on for true (case insensitive).

In some cases it is possible to have units for values. Currently time and size values make use of this feature. Valid time units are w for weeks, d for days, h for hours, m for minutes, and s for seconds. Valid units for size are B for bytes, KB for kilo bytes, MB for mega bytes, and GB for giga bytes. It is allowed to specify a sequence of numbers and units, e.g., 1h 5m 12s. Unless otherwise specified, the default units for times and sizes in a configuration file are s and B, respectively; for those values these units can be used.

Example Configuration File

The installation script creates the file smx.conf in the configuration directory (/etc/smx, see Section 2.4). Check the comments in the file and edit it if required. A configuration file for sendmail X contains several sections: a global section which specifies the locations of sockets and directories that are used by multiple components, and one section each for QMGR, SMAR, SMTP server, and SMTP client. Other sections may define services that are started by MCP, e.g., a local mailer.

CDB_base_directory = "/var/spool/smx/";

qmgr {
  AQ_max_entries = 8192;
  smtpc { initial_connections = 19; max_connections = 101; }
  smtps { max_connections = 5; max_connection_rate=160; }
  wait_for_server = 4; wait_for_client = 4;
  start_action = wait; user = smxq;
  restart_dependencies = { smtps, smtpc, smar };
  path = "/usr/libexec/qmgr"; arguments = "qmgr -f /etc/smx/smx.conf";

smtps { flags = {8bitmime}; CDB_gid = 261; IO_timeout = 5m3s;
  listen_socket { type = inet; port = 25; }
  start_action = pass; pass_fd_socket = smtps/smtpsfd;
  user = smxs; path = /usr/libexec/smtps;
  arguments = "smtps -f /etc/smx/smx.conf"; }

smtpc {
  Log_Level = 12; IO_timeout = 6m; wait_for_server = 4;
  start_action = wait; user = smxc; path = "/usr/libexec/smtpc";
  arguments = "smtpc -f /etc/smx/smx.conf"; }

smar {
  Log_Level = 12;
  nameserver = {,};
  DNS_timeout = 6;
  start_action = wait; user = smxm; restart_dependencies = { smtps, qmgr };
  path = "/usr/libexec/smar"; arguments = "smar -f /etc/smx/smx.conf";

Common Global Configuration

All of the following options have defaults and should only be changed if necessary.

  1. hostname: set the hostname to use for the various components. This can be set if gethostbyname(3) does not return a valid (fully qualified) hostname (format: string).
  2. CDB_base_directory: base directory of CDB (format: string); this should either be empty (which is the default) or a path to a directory including a trailing slash; the CDB library currently simply appends the directory names (see Section 2.4.1) to it. It might be useful to move some subdirectories to different disks (by creating (symbolic) links (ln(1))) to spread the I/O load.
  3. SMAR_socket: socket created by the address resolver over which clients (SMTPS, QMGR) can send requests (format: string).
  4. SMTPC_socket: communication socket between SMTPC and QMGR (format: string).
  5. SMTPS_socket: communication socket between SMTPS and QMGR (format: string).

The sockets are currently Unix domain sockets only, hence the value is simply the pathname of the socket.

Common Configuration Options

There is currently one configuration option which is the same across all modules but is not specified in the global section because it is specific to the individual modules.

  1. log: this is a section with the following options:
    1. facility: see syslog(3) for valid facilities, here are some valid options provided the OS offers them: daemon, mail, auth, local0, etc.
    2. ident: identification string for openlog(3), defaults to name of the modules. It might be useful to chose other identifiers, e.g., smXmta or smxQMGR.
    3. options: options for openlog(3) (without the leading LOG_) as provided by the OS, e.g., pid or ndelay.


    qmgr { log { facility = daemon; ident=smX-qmgr; } }
    smtps { log { facility = mail; ident=smX-MTA; } }

    Note: debug output is currently sent to stdout; syslog(3) is not used for debugging.

All modules have an option to set the amount of logging (log_level) that should be done. The larger the value the more information is logged. For normal operation a value of 9 is recommended. During testing values of 12 to 14 are useful.

Pathnames for Files, Directories, and Maps

Most names of files (including maps) and directories in the configuration file have a default name (compiled into the binary) without an absolute path, e.g., aliases.db. If a pathname is not explicitly set in the configuration file or does not use a absolute path (i.e., begins with a slash), then the default is relative to either

  1. the configuration directory: maps and configuration files, e.g., aliases.db and cert_file.
  2. the main queue directory: pathnames of sockets, and databases to store envelope information (IBDB, DEFEDB) or message contents (CDB).

The paths for files mentioned in case 1 are taken relative to the path of the configuration file which is passed via the -f option to the various modules. For example: if SMAR is started as

/usr/libexec/smar -f /etc/smx/smx.conf

then the pathname used for the aliases map is /etc/smx/aliases.db. This applies to the SMAR maps aliases, mailertable, and access (3.9.2), the QMGR qmgr_conf map (3.8.1), and the STARTTLS related files and directories used by the SMTP server (3.10) and client (3.11).

The paths for files mentioned in case 2 are taken relative to the execution directory. All sendmail X modules should be started (via MCP) in the main queue directory (default: /var/spool/smx, see Section 2.4).

See the various configuration options explained below how to override the defaults. Note: relative pathnames specified in the configuration file are (currently) always relative to the main queue directory.

Configuration for MCP

Every section in a sendmail X configuration file that refers to one of its four main components (QMGR, SMTPS, SMTPC, and SMAR; see Section 1.1.1) has some options that are relevant for MCP. These MCP options are:

  1. start_action: one of nostartaccept, accept, pass, wait (required).

  2. listen_socket: this is a subsection that specifies the socket on which a process should listen. It must be specified for any start_action except wait. There are two different socket types available:

    1. type = inet
      1. port: port number on which process should listen (format: numeric).
      2. address: IP address on which process should listen, if none is specified the process listens on all local (IPv4) addresses (format: IPv4 address).

    2. type = unix
      1. path: pathname of Unix Domain socket on which process should listen (format: string).
      2. umask: umask for socket (format: numeric).
      3. user: owner of socket (format: string).
      4. group: group of socket (format: string).

  3. pass_fd_socket: pathname of Unix Domain socket to pass a file descriptor to the process.
  4. user: user name to run process.
  5. group: group name to run process.
  6. restart_dependencies: list of other sendmail X components that need to be restarted when this one is restarted (or crashes).
  7. path: path to program to execute (required).
  8. arguments: arguments (argv), must start with name of program, see execv(2) (required).
  9. pass_id: option to use to pass a unique, numeric identifier to the spawned process via the command line. The option will be inserted as first argument. Example:
    smtpc { pass_id = "-i"; min_processes = 4; max_processes = 4;
      path = /usr/libexec/smtpc; arguments = "smtpc -f smx.conf"; }
    will cause MCP to start four smtpc processes, each with the options -i $ID$ -f smx.conf where $ID$ is replaced with a unique identifier.

Notes about start_action:

MCP is currently a generic control program that does not have any builtin knowledge about the various sendmail X modules. Hence the MCP options for each sendmail X component must be specified properly, there are no builtin defaults that could be associated with the functionality of the various sendmail X modules. The default configuration file created by the installation program contains the correct defaults. These should only be changed if really necessary.

Configuration for QMGR

The following configuration options are valid for QMGR:

  1. AQ_max_entries: maximum number of entries in AQ (active queue) (unit: entries). Note: this value must be larger than the largest number of recipients accepted by a single transaction.
  2. conf: name of configuration map (including extension), see Section 3.8.1 for details. See also Section 3.6 about relative pathnames.
  3. control_socket: specify pathname of ``control'' socket (for querying and making requests). This socket can be used by the query/control program qmgrctl, see Section 4.6.3.

  4. subsection DEFEDB:
    1. base_directory: home directory for DEFEDB.
    2. log_directory: log directory for DEFEDB. For better performance, this directory can be set to point to a different disk than the base directory of DEFEDB.

  5. subsection DSN_handling:
    1. merge_delay_max: maximum time to wait for merging multiple DSNs into one (unit: s).
    2. flags: configuration flags:
      1. header_only: include only the headers in a DSN; by default the first bounce includes the entire message and subsequent ones include only the headers.
      2. MIME_Format: use MIME to structure a DSN. Note: this is not (yet) a DSN in the format specified by RFC 3464 [MV03].
    3. max_errors_per_DSN: maximum number of error messages (failed recipients) in a bounce (DSN) (unit: entries).

  6. subsection IBDB:
    1. max_commit_delay: maximum time between commits to IBDB (unit: $\mu$s)
    2. size: maximum size of each IBDB file (unit: B).
    3. max_open_TAs: maximum number of open transactions in IBDB before a commit is performed (unit: entries).

    Note: the configuration file offers no way to specify a base directory for IBDB, however, the directory can be easily moved elsewhere and a (symbolic) link (ln(1)) can be added.

  7. subsection IQDB:
    1. max_cache_entries: maximum number of entries in IQDB cache (unit: entries). This must be larger than the sum of all recipients in open transactions.
    2. hash_table_entries: size of hash table for IQDB (unit: entries). This must be larger than max_cache_entries.

  8. log_level: logging level.
  9. min_disk_space: minimum amount of free disk space (unit: KB). This value should be significantly larger than the maximum size of a message to be accepted by the SMTP server, it should be as large as the maximum message size multiplied by the maximum number of incoming connections.
  10. OCC_max_entries: size of outgoing (SMTPC) connection cache (unit: entries). This should be large enough to keep track of outgoing connections over a time span that is at least as long as the maximum retry time.
  11. ok_disk_space: amount of free disk space at which normal operation continues (unit: KB). Must be larger than min_disk_space.
  12. queue_return_timeout: maximum time in queue (unit: s).
  13. queue_delay_timeout: send delay warning (``delayed DSN'') if the mail is still in the queue after at least this duration (unit: s). To turn off delayed DSNs set this to a value bigger than queue_return_timeout. Note: based on the retry schedule the delayed DSN might be sent later than the option specifies.
  14. retry_max_delay: maximum time for retrying a delivery (unit: s).
  15. retry_min_delay: minimum time for retrying a delivery (unit: s).

  16. subsection smtpc:
    1. initial_connections: maximum initial number of outgoing connections to a single host (unit: entries). The sliding window for the slow start algorithm (see Section 3.8.1) is initialized with this value.
    2. max_connections: maximum number of outgoing connections to a single host (unit: entries).
    3. lmtp_max_rcpts_per_transaction: maximum number of recipients per transaction for mail sent via LMTP (unit: entries).
    4. smtp_max_rcpts_per_transaction: maximum number of recipients per transaction for mail sent via (E)SMTP (unit: entries).

  17. subsection smtps:
    1. max_connection_rate: maximum incoming connection rate from a single host (unit: connections/60s).
    2. max_connections: maximum number of open incoming connection from a single host (unit: entries).

  18. wait_for_client: maximum amount of time to wait for a client to become available (unit: s)
  19. wait_for_server: maximum amount of time to wait for a server to become available (unit: s)

Configuration Map for QMGR

QMGR implements a ``slow start'' algorithm to control the number of concurrent connections to one IP address. Initially, it will at most create a (small) number of open connections up to a specified initial limit. For each successful delivery, the allowed number is increased up to specified maximum limit.

For incoming connections, QMGR establishes two limits: the connection rate and the number of open connections.

The Berkeley DB hash map qmgr_conf.db (the file should be owned by smxq) can have the following entries:

  1. oci: this key specifies the initial number of concurrent outgoing connection to an IP address.
  2. ocm: this key specifies the maximum number of concurrent outgoing connection to an IP address.
  3. octo: specify the timeout for an entry in the outgoing connection cache.
  4. icr: this key specifies the maximum rate for incoming connections (per 60s).
  5. icm: this key specifies the maximum number of concurrently open incoming sessions.

oci:, ocm:, icr:, and icm: take an IP address/net as parameter such that the limits can be imposed per IP address/net. For example:

oci:      5
ocm:     10
oci:10            10
ocm:10            50
oci:               1
ocm:               4
icr:10             5
icr:    100
icm:    120

Note, however, that the limits apply only to single IP addresses, they are not aggregated for nets. That is, for the example every single host in the IP net 10.x.y.z can have a maximum incoming connection rate of 5 messages per minute.

The default values for these configuration options are set in the binary and can be changed via command line options or the configuration file (see Section 3.8):

  1. -C n maximum number of concurrent connections to one IP address [default: 100]
  2. -c n initial number of concurrent connections to one IP address [default: 10]
  3. -O R=n maximum connection rate per 60s (SMTPS) [default: 100]
  4. -O O=n maximum number of open connections (SMTPS) [default: 100]

Configuration for SMAR

Declaring Maps for SMAR

In general, maps must be declared before they can get used. Each map declaration in a configuration file is a named subsection - the name is used for later references - map in the smar section with the following options:

  1. type: type of the map; currently one of hash (Berkeley DB hash), sequence, socket, and passwd.
  2. file: the filename of the db file (including the extension) (for type hash).
  3. mapname: name of the map used in the protocol (type socket only).
  4. address: IPv4 address of inet socket. (type socket only).
  5. path: the pathname of the Unix domain socket (for type socket).
  6. port: port for inet socket (type socket only).
  7. maps: list of map names to use in the map (type sequence only).

Note: for socket maps either a Unix domain socket (path) or an inet socket (address and port) must be specified.


map localusers { type = hash; file = "/etc/smx/localusr.db"; }
map otherusers { type = hash; file = "/etc/smx/otherusr.db"; }
map password { type = passwd; }
map seq1 { type = sequence; maps = { localusers, otherusers }; }
map seq2 { type = sequence; maps = { password, otherusers }; }

Configuration Options for SMAR

The following configuration options are valid for SMAR:

  1. access_map: this is a subsection that specifies the access control map, currently it has only one valid entry: file: filename of access map (including extension) [default: access.db]. See Section 3.9.3 for details.
  2. address_delimiter: delimiter (one character) for address extensions in local part, [default: '+'].
  3. aliases: this is a subsection that specifies the parameters for aliases:
    1. file: filename of aliases map (including extension) [default: aliases.db].
    2. flags:
      1. localpart: the aliases map contains only localparts of addresses and those are only looked up for local addresses.
      2. local_domains: the aliases map contains fully qualified addresses which are only looked up for local addresses. This can be used similar to virtual users in sendmail 8, e.g.,
        vuser1@virt1.tld: user1
        vuser2@virt1.tld: user2
        vuser3@virt2.tld: user3

      3. all_domains: the aliases map contains fully qualified addresses which are only looked up for any domain.
      4. implicitly_match_detail: the items are looked up according to the algorithm specified in Section 3.12.1. and additionally +detail is implicitly matched when the pattern is ``user@hostname''. That is, it overrides the default matching explained in case 1e in Section 3.12.1.
      5. replace_macros: replace macros in the RHS of the map entries by the appropriate value, see Section 3.12.3.
      6. preserve_domain: if the RHS of an entry is an unqualified address, do not append the local hostname to it but the domain of the original address, i.e., preserve the original domain.
  4. dnsbl: specify a DNS based blacklist3.1. This section can be specified multiple times3.2; it has the following required options: The client IPv4 address A.B.C.D is looked up via DNS as D.C.B.A.domain querying for an A record. If an A record W.X.Y.Z is found, then it is looked up in the access map as tag:W.X.Y.Z. for temporary and permanent DNS lookup failures the entries that will be checked in the access map are tag:temp and tag:perm, respectively.


    The access map entry should have one of the usual rejection RHSs as explained in 3.9.3. Example: configuration file:
    smar { dnsbl { domain = dnsbl.tld; tag = dnsbltld; } }

    access map:

    dnsbltld:  error:550 5.7.1 listed at dnsbl.tld as open relay
    dnsbltld:  error:550 5.7.1 listed at dnsbl.tld as spam source
    dnsbltld:  error:451 4.7.1 listed at dnsbl.tld as suspicious
    dnsbltld:temp       error:451 4.7.1 temporary lookup failure at dnsbl.tld

    If multiple DNS based blacklists are specified, the DNS queries are made concurrently but the lookups in the access map are performed in the order in which the blacklists are given; the first successful lookup is used as result, no further priorization is performed.

  5. DNS_timeout: timeout for DNS requests (unit: s).

  6. greylisting: specify greylisting options, see Section 3.9.4 for details.
    1. grey_wait: how long before greylisted can be confirmed.
    2. grey_expire: timeout for greylisted entries (did not confirm within that time).
    3. white_expire: expire whitelisted entries after this time if necessary.
    4. white_timeout: force whitelisted entries to reconfirm after this time.
    5. main_DB_name: name of main database (including .db extension).
    6. secondary_DB_name: name of secondary database (including .db extension).
    7. expire_limit: try to expire entries when this limit is reached.
    8. netmask: by default the entire IPv4 address is used as a key, however, by specifying a netmask, e.g., 0xFFFFFF00, the least significant bits can be cut off. This can be used to deal with server farms, see Section 3.9.4, e.g., if those are in the same class C subnet.

  7. local_user_map: this is a subsection that specifies a map of valid local addresses.
    1. name: Name of the map of valid local addresses; the map must have been declared as explained in Section 3.9.1.
    2. flags:
      1. implicitly_match_detail: +detail is implicitly matched when the pattern is ``user@hostname''. That is, it overrides the default matching explained in case 1e in Section 3.12.1.

  8. log_level: logging level.

  9. mailertable: this is a subsection that specifies a mailertable, currently you can specify exactly one of the following two options:
    1. file: filename of mailertable [default: mt]. In this case a plain text file is read during startup and placed in an internal hash table.
    2. name: name of a mailertable map that has been declared before (see Section 3.9.1).
    The format of entries in the map is explained in Section 3.9.3. Note: reloading mailertable (Section 4.7) while SMAR is running can be done only if it is declared as Berkeley DB (case 9b with the proper map)

  10. nameserver: list of up to four IPv4 addresses3.3of nameservers.

Configuration Maps for SMAR

SMAR requires a mailertable, and it can make use of an alias map as well as an access map, all of which are described in the subsequent sections.

Access Map

To activate the access map the flag access (see Section 3.10, item 3h) (or the option -a) must be given to the SMTP servers. All entries consist of a left hand side (LHS, key) which in turn has a tag and a (partial) address and a right hand side (RHS, value). Valid tags are:

Tag refers to
from: envelope sender address (MAIL)
to: envelope recipient address (RCPT)
cltaddr: client IPv4 address
cltname: client host name
cltresolve: result of forward and reverse client lookup
mxbadip: IPv4 addresses that are not allowed for MX - A records
certissuer: DN of CA cert that signed that presented cert
certsubject: DN of presented cert
protectedrcpt: restrictions for recipient address (see Section 3.10.2)

Valid addresses for from: and to: are RFC 2821 addresses without the angle backets (localpart@domain) as well as partial addresses in the form localpart and @domain, i.e., domains must be preceeded with an at (@) sign. Valid addresses for cltaddr: and mxbadip: are IPv4 addresses and (sub)nets, and for cltname: host names. The client host name is determined by performing a reverse lookup (PTR record) for its IP address. The resulting names are looked up as A records. Only if one of the A records matches the client IP address, the host name is set. The result of these lookups can be used for cltresolve: where the following keys are valid:

ok reverse and forward lookup match
no reverse and forward lookup do not match
tempptr reverse lookup (PTR) caused a temporary error
tempa forward lookup (A) caused a temporary error

Valid values for RHS are

relay allow relaying; currently only for to:, cltaddr:,
  cltname:, certissuer:, and certsubject:
ok accept command
error:XYZ A.B.C.D text return an error consisting of SMTP reply code XYZ,
  enhanced status code A.B.C.D, and text,
  i.e., the part after error: is returned to the client.
reject same as error:550 5.7.0 Rejected.
discard accept command but silently discard its effects.
cont stop current check (e.g., map lookup), but continue others.

Some tags may allow for other RHS values, these are explained when those tags are discussed in more detail.

Optionally a RHS can be preceeded by the modifier quick:. For an error: entry it causes an immediate rejection when the entry matches. Otherwise rejections can be delayed to the RCPT stage - if SMTPS is configured appropriately, see Section 3.10, item 3b - and can be overridden using the modifier quick: together with ok or relay in the access map for the recipient address with the to: tag. Using the modifier quick: together with relay for an entry with the cltaddr: tag causes it to override all other access map checks. quick:ok for an entry with the cltaddr: tag causes it to override other access map checks unless they are necessary to allow relaying.

Domain names (@domain) must have an exact match, subdomain matching can be specified with a leading dot, i.e., @.domain, see Section 3.12.1.


cltresolve:tempptr error:451 4.7.1 reverse lookup failed
mxbadip: error:551 5.7.1 Bad IP address in MX/A list
mxbadip: error:551 5.7.1 Bad IP address in MX/A list
from:@spammer.domain error:551 5.7.1 No spammers
from:@.spammer.domain error:551 5.7.1 No spammers in subdomains either
to:root error:551 5.7.1 No mail to root
to:abuse quick:ok
cltaddr:10 error:551 5.7.1 No direct mail from 10.x.y.z
cltname:spammer.domain quick:error:551 5.7.1 No mail from spammers
to:@primary.domain relay
cltaddr:10 relay
cltaddr: quick:relay


The effect of discard depends on the protocol stage in which it is returned. If it is returned for a session, e.g., when a client connects, all transactions in the session are discarded. If it is returned for MAIL only that transaction is discarded. If it is returned for RCPT only that recipient is discarded; however, if no valid recipients are left, the entire transaction is discarded. Moreover, if quick:discard is returned for one recipient the entire transaction is discarded too.


The address resolver implements an asynchronous DNS resolver and by default it uses a file called mt (mailertable) (see Section 3.9.2, item 9) which consists of domain parts of e-mail addresses and corresponding IP addresses (in square brackets) or domain/host names. An entry consists (as usual in a map) of a LHS and a RHS; in the case of a flat text file, i.e., case 9a of Section 3.9.2, those are separated by one or more whitespace characters.

LHS ::= [ "." ] hostname $\vert$ "."
RHS ::= [[ port "^" ] ["esmtp:"] hostlist $\vert$ "lmtp:" $\vert$ port "^"lmtp:" hostlist
port ::= integer
hostlist ::= host [ " " hostlist ]
host ::= "[" IPv4-address "]" $\vert$ hostname

The key (LHS) is a hostname or a dot (denoting the default entry), the value (RHS) consists of an optional port number, an optional (esmtp) mailer and a list of hosts which are separated by spaces. If LMTP should be used, then the lmtp mailer must be selected. There are two cases: just lmtp: by itself means the delivery agent will use the Unix domain socket specified in the configuration file (see Section 3.11, item 2), if an inet socket should be used then a port and a host must be specified. A host is either a hostname (which is subject to MX lookups) or an IPv4 address in square brackets.


localhost lmtp:
SPAM.FILTER.DOMAIN 2525^esmtp:[]
MY.DOMAIN esmtp:[]
. esmtp:SMART.HOST

Note: currently this file must exist, even if there are no entries (it is created during installation).


To specify aliases for local addresses the Berkeley DB hash map aliases.db (Section 3.9.2, item 3a) is used. The key in the map must be

based on the flags of the aliases option (see Section 3.9.2, 3b). The value (RHS) for an alias entry is a list of one or more RFC 2821 addresses (including the angle brackets) separated by spaces (not commas). If the RHS has only a single address which does not have an '@' sign, then it is converted into an RFC 2821 address by SMAR, i.e., SMAR will append the hostname of the machine and put angle brackets around the string. Example:

myalias: localuser
mylist: <user1@my.dom> <user2@my.dom> <>
owner-mylist: someuser

For mailing lists, the owner- notation is supported, i.e., if there are aliases list and owner-list then mail sent to list will use owner-list as envelope sender address; the original domain will be preserved.

Example for the flag local_domains (see 3.9.2, 3(b)ii). Let two domains be local, i.e., in mailertable:

first.dom lmtp:
second.dom lmtp:

and these entries be in aliases:

myalias@first.dom: user1
another@second.dom: user2

Then mail to <myalias@second.dom> and <another@first.dom> would be rejected while mail to <myalias@first.dom> or <another@second.dom> would be accepted.

Aliases can be nested (currently up to 5 levels, see smar/rcpts.c).


sendmail X supports a very simple form of greylisting [Hara] which only uses the client IP address as key [Posa] instead of a tuple consisting of client IP address, envelope sender, and envelope recipient. The idea behind greylisting is simple: do not accept mail from an unknown source on the first connection, but reject it with a temporary error. Any MTA that conforms to RFC 2821 [Kle01] will try to send the mail later on, however, spamming systems often do not do that. An IP address can be in three different states: unknown: the client has not connected before or the entry is expired from the database, greylisted: the client has connected before but it did not yet connect again within the configured time interval, whitelisted: the client has connected before and it connected again within the configured time interval. The time interval is specified by its lower limit grey_wait and its upper limit grey_expire. A lower limit is used to prevent system from getting accepted that just send a single message within a few seconds again and again. The upper limit is used to avoid filling up the database. If an entry has made it to the whitelisted state, it will stay there for (at least) up to the timeout specified by white_expire. The greylisting algorithm implemented in sendmail X uses another timeout white_timeout after which a whitelisted entry is considered stale and must go through the greylisting stages again, i.e., it is considered to be in state unknown. Each time a mail is sent from a whitelisted host, the entry is updated, to avoid that systems which regularly sent mail become greylisted again.

Greylisting is performed at the RCPT stage of the SMTP dialogue. It is only done when a valid recipient is specified, i.e., all other checks must have been successful. Hence clients that do not try to send mail or just try invalid recipient addresses will not be added to the greylisting database. If a transaction is subject to greylisting then the session is aborted with an 421 error. If a server uses callbacks to verify the sender address, then the option delay_greylisting_error_until_DATA (see Section 3.10, item 3c) is useful to avoid unnecessary delays. Here is an example: host A is the main MX server for domain and it uses greylisting, host B is the main MX server for domain and it uses sender callbacks. If a mail is sent to host B for <> with the sender address <> then host B will connect to host A to test whether <> is a valid recipient. However, if host A does not have host B in its whitelist, it will return a 421 error after the RCPT To:<> command, which (depending on the implementation of the sender callback) will cause host B to temporarily reject the mail for <>. By specifying the option delay_greylisting_error_until_DATA on host A the RCPT command will succeed and the original mail to <> will go through without delay.

The greylisting implementation uses two persistent databases (specified by main_DB_name and secondary_DB_name), where the second DB is just a secondary index (by expiration time) for the main DB. These databases should be on a filesystem with sufficient free disk space depending on how many connections from different clients the MTA receives. Entries are only removed from the DB if there are more than expire_limit elements. However, if none of the entries are expired yet, then the number of elements can exceed that limit.

Greylisting: Whitelisting

Greylisting can be disabled for selected hosts by adding them to the access map (see Section 3.9.3), e.g.,

cltaddr:10 relay
cltaddr: quick:relay

Possible Problems with Greylisting

Some legitimate mailers do not behave properly and will not retry a mail that had a temporary error. This can cause mail loss in various situations, e.g., because the receiving system is currently out of some resources. However, to minimize the impact of greylisting on these misbehaving mailers it might be useful to explicitly whitelist them as:

cltaddr: ok
cltaddr:64.12.137 ok

A list of such broken mailers can be found at [Harb]. A related problem are server farms where a mail might be resent from a different IP address. These should probably be whitelisted too; some of these can be found at the URL given before. However, entries in that file which have the comment ``unique sender per attempt'' do not need to be whitelisted as this implementation does not use the sender address.

Note: if a client authenticates via STARTTLS or AUTH such that relaying is allowed then greylisting is disabled for that client.

Configuration for SMTP Server

The following configuration options are valid for SMTPS:

  1. auth: this is a subsection that specifies the parameters for AUTH support. It is only available if the system has been configured with the option -enable-SASL, see Section 2.2.1.
    1. flags: flags for SMTP AUTH

      See the Cyrus SASL documentation for the meaning of these flags: noplaintext, noactive, nodictionary, forward_secrecy, noanonymous, pass_credentials, mutual_auth.

    2. trusted_mechs: list of SASL mechanisms for which relaying is allowed if a client successfully authenticated using one of those

    Note: the name for the Cyrus-SASL configuration file is currently sendmail.conf (note the lower case 's' compared to the name for the sendmail 8 Cyrus-SASL configuration file). That file can be used to adjust the list of mechanisms that should be advertised (besides many other things), hence this option is not in the SMTP server itself.

  2. CDB_gid: (numeric) group id for CDB, i.e., the group id of smxq, see Section 2.4.1.
  3. flags:
    1. 8bitmime: offer 8BITMIME: sendmail X is 8 bit transparent, but it does not perform any conversion, so this option should only be used if all communication partners can deal with 8 bit data.
    2. delay_checks: delay acceptance check until RCPT stage (unless explicitly overridden, see Section 3.9.3).
    3. delay_greylisting_error_until_DATA: if greylisting (3d) is enabled then wait until the DATA command to return an error; see Section 3.9.4 for details.
    4. greylisting: enable greylisting (which must also be enabled in SMAR, see Section 3.9.2 item 6), see Section 3.9.4 for details.
    5. lmtp_does_not_imply_relaying: even if a domain in the mailertable has lmtp: as RHS do not implicitly allow relaying to it, i.e., do not consider the domain as ``local'' with respect to relaying. This is useful for an MSA to avoid external mail to local domains without authentication.
    6. soft_bounce: change permanent (5xy) SMTP error replies into temporary (4xy) errors. This is a useful feature for testing to avoid bounces due to misconfigurations.
    7. strict_ehlo_checks: perform a strict syntax check on the argument for EHLO (or HELO).
    8. access: use access map (in SMAR). Note: currently this flag is required to perform a reverse lookup for a client IP address to get the hostname of the client which then can be used for logging and the Received: header.
  4. id: unique identifier for SMTP server (0); see Section 3.10.1.
  5. io_timeout: timeout for SMTP operations.
  6. max_threads: maximum number of threads.
  7. max_bad_commands_per_session: maximum number of bad, i.e., unknown, SMTP commands per session accepted by server. After this limit is reached the connection is terminated with an 421 error.
  8. max_invalid_addresses_per_session maximum number of invalid, e.g., unknown, RCPT addresses per session accepted by server. After this limit is reached the connection is terminated with an 421 error.
  9. max_nop_commands_between_transactions: maximum number of NOOP, RSET, and related SMTP commands between two successful transactions accepted by server. After this limit is reached the connection is terminated with an 421 error.
  10. max_bad_commands_per_transaction: maximum number of bad, i.e., unknown, SMTP commands per transaction accepted by server. After this limit is reached the connection is terminated with an 421 error.
  11. max_nop_commands_in_transaction: maximum number of NOOP and related SMTP commands in a single transaction accepted by server. After this limit is reached the connection is terminated with an 421 error.
  12. max_invalid_addresses_per_transaction maximum number of invalid, e.g., unknown, RCPT addresses per transaction accepted by server. After this limit is reached the connection is terminated with an 421 error.
  13. max_recipients_per_session: maximum number of recipients per session.
  14. max_recipients_per_transaction: maximum number of recipients per transaction.
  15. max_hops: maximum number of hops (Received: headers). If this value is exceeded the incoming mail is rejected because it is considered a possible mail loop.
  16. max_message_size: maximum message size (unit: KB).
  17. policy_milter: this is a subsection that specifies the parameters for pmilter support (see Section 5). It is only available if it has been enabled during configure (-enable-pmilter, see Section 2.2.1).
    1. socket: this is a subsection that specifies the socket to communicate with policy milter. The type (option type) of the socket must be either inet or unix.
      1. type = inet
        1. port: port number for connection.
        2. address: IP address for connection.

      2. type = unix
        1. path: pathname of Unix Domain socket.

    2. timeout: maximum amount of time to wait for a reply from a policy milter.
    3. flags: policy milter flags. If the connection to pmilter fails then SMTPS will ignore pmilter by default. This behavior can be changed by setting one of following two flags:
      1. abort: if the connection to pmilter fails then abort the current session with a 421 error.
      2. accept_but_reconnect: if the connection to pmilter fails then continue the current session but try to reconnect for the next session.

  18. processes: number of processes to start.
  19. protected_recipients: this is a subsection which provides a few simple options to protect recipients by restricting who can send mail to them.
    1. allow_by: this is a required subsection which has two possible flags (at least one must be specified).
      1. sender: allow sending mail based on the envelope sender (MAIL) address. Even though this address can be forged it provides some basic protection.
      2. client_ip: allow sending mail based on the client IP address.

    2. match_type: this specifies what type of matching should be done. By default, exact matches are required. Alternatively, one of the following two options can be selected:
      1. generic_lookup: the items are looked up according to the algorithm specified in Section 3.12.1.
      2. implicitly_match_detail: the items are looked up according to the algorithm specified in Section 3.12.1. and additionally +detail is implicitly matched when the pattern is ``user@hostname''. That is, it overrides the default matching explained in case 1e in Section 3.12.1.

    See Section 3.10.2 for details.

  20. max_transactions: maximum number of transactions per session.

  21. tls: this is a subsection that specifies the parameters for STARTTLS support. It is only available if the system been configured with the option -enable-TLS, see Section 2.2.1. See Section 11.1 for some background information about these options.

    1. cert_file: file with certificate in PEM format.
    2. key_file: file with private key for certificate in PEM format.
    3. CAcert_file: file with CA certificate in PEM format.
    4. CAcert_directory: directory with (symbolic links for) CA certificates in PEM format.

    5. flags: some flags are available to influence the behavior of the SMTP server with respect to STARTTLS.
      1. allow_relaying_if_verified: if the client presented a certificate that can be verified by the CA certificates that are available to the server (see above: CAcert_file and CAcert_directory), then relaying is allowed for the SMTP session.

      2. check_access_map_for_relaying: if this flag is set then the access map (which must be activated, see 3h) is checked to see whether relaying should be allowed for a client which presented a certificate that has been verified (see above). For this purpose, the DN of the cert issuer is looked up in the access map using the tag certissuer:. If the resulting value is relay, relaying is allowed. If it is cont, the DN of the cert subject is looked up next in the access map using the tag certsubject:. If the value is relay, relaying is allowed; every other value is currently ignored.

        To avoid problems with the DN names in map lookups, they are modified as follows: each non-printable character and the characters '<', '>', '(', ')', '"', '+', ' ' are replaced by their hexadecimal ASCII value with a leading '+'. For example:

        Darth Mail (Cert)/

        is encoded as:



        To allow relaying for everyone who can present a cert signed by


        simply use:

        Darth+20Mail+20+28Cert+29/ relay

        To allow relaying only for a subset of machines that have a cert signed by



        Darth+20Mail+20+28Cert+29/ cont
        DeathStar/ relay


        • line breaks have been inserted after CN= for readability, each tagged entry must be one (long) line in the access map.
        • if OpenSSL 0.9.6 is used then the emailAddress= part of a DN is replaced by Email=.

Multiple SMTP Servers with different Configurations

The normal way to run multiple SMTP servers is to let MCP start several SMTP servers. Each SMTP server must given a unique identifier (see Section 3.10, item 4) and each SMTP server section in smx.conf must have a unique name (e.g., MTA and MSA), which is passed via the option -N name to smtps. Example: smx.conf:

smtps MTA {
  listen_socket { type=inet; port = 25; }
  start_action = pass; pass_fd_socket = smtps/mtafd;
  user = smxs;
  path = /usr/libexec/smtps;
  arguments = "smtps -N MTA -f /etc/smx/smx.conf";
  log { facility = mail; ident=smX-MTA; }

smtps MSA {
  listen_socket { type=inet; port = 587; }
  start_action = pass; pass_fd_socket = smtps/msafd;
  user = smxs;
  path = /usr/libexec/smtps;
  arguments = "smtps -N MSA -f /etc/smx/smx.conf";
  log { facility = mail; ident=smX-MSA; }
  auth { trusted_mechs = { CRAM-MD5, DIGEST-MD5 };
         flags = { noplaintext }; } }

For tests it is also possible to let MCP start only one SMTP server which creates several copies of itself if multiple daemon addresses are specified (see Section 3.10, item 1). Note: this only works for unprivileged ports because the SMTP server does not run as root.

Protecting Recipients

A few simple features are available to protect recipients by restricting who can send mail to them. To do this the configuration section protected_recipients must be turned on and at least one of the two flags allow_by_sender and allow_by_client_ip must be selected. If this is done, then every recipient is looked up in the access map (which must be activated, see Section 3.10, item 3h), using the tag protectedrcpt:. If a matching entry is found, it must have a list of (one or more) restrictions, each of which must be one of the following:

restriction required flag
from:sender allow_by_sender
cltaddr:IPv4-address allow_by_client_ip
list:alias allow_by_sender

The meaning of the first two restriction types should be obvious, the third one is interesting: it refers to an alias (in the aliases map, see Section 3.9.3) and requires that the sender address matches one of the entries to which the alias expands. This can be used to allow only subscribed members of a mailing list to send mail to it.

The restrictions are evaluated sequentially, if there is a match, the recipient is accepted (sequential OR). If none of them matches, the recipient is rejected.

By default exact matches are required. However, if the flag generic_lookup is set, the items are looked up as specified in Section 3.12.1. The flag implicitly_match_detail is useful for the list: restriction if a sender uses +detail without having that specified during subscription.

Examples: consider the following aliases map:

list1: <user1-1@l1-1.dom> <user2-1@l1-1.dom> <list2@local.dom>
list2: <user1-2@l2-1.dom> <user2-2@l2-2.dom>
list3: <user1-3@l3-1.dom> <user2-3@l3-2.dom>

together with this access map:

protectedrcpt:list1@local.dom list:<list1@local.dom>
protectedrcpt:list3 from:<moderator3@local.dom> cltaddr: cltaddr:10

The mails to <list1@local.dom> are only accepted from <user1-1@l1-1.dom> and <user2-1@l1-1.dom>. Note: the list is not recursively expanded, i.e., members of list2 are not allowed, that restriction must be listed in the access map. Mails to <list3@local.dom> are only accepted from <moderator3@local.dom>, the client with the IPv4 address, or clients in the IPv4 net 10. The latter requires that the flag generic_lookup is turned on too.

Configuration for SMTP Client

The following configuration options are valid for SMTPC:

  1. io_timeout: timeout for SMTP operations (unit: s).
  2. LMTP_socket: Unix domain socket to use for LMTP [default: lmtpsock].
  3. log_level: logging level.

  4. tls: this is a subsection that specifies the parameters for STARTTLS support. It is only available if the system been configured with the option -enable-TLS, see Section 2.2.1. See Section 11.1 for some background information about these options.
    1. cert_file: file with certificate in PEM format.
    2. key_file: file with private key for certificate in PEM format.
    3. CAcert_file: file with CA certificate in PEM format.
    4. CAcert_directory: directory with (symbolic links for) CA certificates in PEM format.

  5. wait_for_server: maximum amount of time to wait for a server (QMGR) to become available (unit: s).

Lookup Orders

Lookup Orders in Maps

In many cases an item is not just looked up verbatim in a map, but it may be split into logical parts and then less significant parts are iteratively removed and the remaining data is looked up until either a match is found or the data is empty; in the latter case a default key may be looked up depending on the map.

For domain names of the form ``sub2.sub1.tld'' the lookup order is ``sub2.sub1.tld'', ``.sub1.tld'', ``.tld'', and ``.'' (without the quotes), the last lookup is only done if the map type requests it, e.g., mailertable. Obviously this schema is extended if more components are specified. As the sequence shows there is no implicit ``match all subdomains'' lookup, instead entries in a map must have a leading dot for subdomains matches. To reiterate: ``sub2.sub1.tld'' does neither match the entry ``sub1.tld'' nor ``tld''.

For IPv4 addresses of the form ``A.B.C.D'', the lookup order is ``A.B.C.D'', ``A.B.C'', ``A.B'', and ``A'' (without the quotes). In contrast to domain lookups, no trailing dots are required (nor checked) to denote subnet matches, because the number of components of an IPv4 address is fixed (and known) in contrast to the number of components in a host name or domain name.

For RFC 2821 addresses of the form ``$<$user+detail@domain$>$'' (where ``+detail'' is optional and ``+'' is the address_delimiter, see Section 3.9.2) the lookups are done according to the following sequence:

  1. Repeat the following lookups for each subdomain of domain (as explained above):
    1. ``user+detail@subdomain'' if ``+detail'' exists; this is a verbatim match.
    2. ``user++@subdomain'' if ``+detail'' exists and ``detail'' is not empty; this matches any non-empty ``+detail''. Note: the second ``+'' character is a fixed metacharacter, it does not depend on address_delimiter; it is a modelled after the ``+'' operator in regular expressions etc to denote a non-empty sequence of items.
    3. ``user+*@subdomain'' if ``+detail'' exists; this matches any ``+detail'' (including just ``+'').
    4. ``user*@subdomain''; this matches ``user@subdomain'' as well as ``user+detail@subdomain'' (``detail'' can be empty). Note: ``*'' is not a generic metacharacter here, it matches only a token beginning with address_delimiter or an empty sequence, it does not match any other character sequence. For example: the input ``user1@subdomain'' does not match the LHS ``user*@subdomain''.
    5. ``user@subdomain''; this does not match if ``+detail'' exists, unless the option implicitly_match_detail is selected for the map to implicitly match a detail even if there is no wildcard in the pattern.
    6. ``@subdomain''.

  2. If nothing has been found and the map type requests it, then try localpart only (with the same meaning as above):
    1. ``user+detail'' if detail exists
    2. ``user++'' if detail exists and is not empty,
    3. ``user+*'' if detail exists,
    4. ``user*'',
    5. ``user''

Lookup Orders for Anti-Spam Measures

Map lookups for anti-spam measures are performed according to the SMTP dialogue, i.e., connection information (cltaddr: and cltname:), MAIL command (from:), and RCPT command (to:). Whether a rejection has an immediate effect depends on the result of the lookup, e.g., the quick: modifier, and whether the option delay_checks is set. If multiple checks are performed during a single stage of the SMTP dialogue then they are done sequentially until one of them returns something else than cont.

Note: in the description of the algorithms below some items are marked as check:. Only those can change the result value, other steps perform just operations that may be needed later on but have no immediate effect on the outcome of the checks.


During connect the following operations are performed if the access flag (see Section 3.10, item 3h) is enabled:

  1. check: lookup client IP address using tag cltaddr: (as explained in 3.12.1)
  2. start DNS blacklist queries,
  3. map client IP address to client hostname returning the tuple (clientresolve, clientname).
  4. check: lookup cltresolve:clientresolve
  5. check: lookup cltname:clientname (as explained in 3.12.1)
  6. check: lookup results of DNS blacklists in access map.


After a MAIL command has been received the following checks are performed unless the address is <> or a session check resulted in quick:relay or quick:ok:

  1. check: is the address routeable? That is, if the sender address would be used as a recipient address (as it would be necessary if a DSN must be sent) is it possible to find a host that will deal with the address? This means that the domain part must have a valid MX or A record or that routing is specified via mailertable.
  2. check: if the domain of sender address is local: is the local part valid?
  3. check: lookup the address in the access map (provided it is enabled) with the tag from:.
  4. check: lookup the IP addresses that were found when trying to determine whether the address is routeable with the tag mxbadip: in the access map.


A RCPT command causes different checks:

  1. is this a relaying attempt and if so, is it authorized? Relaying can be allowed for the entire session, e.g., due to the client IP address or other authorization based on some authentication (STARTTLS, AUTH).
  2. are there any other restrictions for the recipient address?

For case 1 the following tests are performed:

  1. check: is the recipient local and does the address exist? If yes, it is not a relaying attempt and hence allowed.
  2. check: lookup the recipient address with the tag to: and check whether the RHS is relay, otherwise reject the RCPT command as unauthorized relaying attempt.

For case 2 the following steps are taken:

  1. lookup address with tag protectedrcpt:, if found perform all the necessary checks as explained in Section 3.10.2.
  2. check: lookup the address using the tag to: if all of the following three conditions are met:
    1. the access map is enabled.
    2. a session check did not result in quick:relay.
    3. a session check did not result in quick:ok and relaying is allowed by other means.
  3. check: perform greylisting unless the session is marked as ok or relay.

Macro Replacements in RHS

The alias map allows the use of macro in the right hand side of map entries. Macros have the form ``${name}'' (without the quotes). Available macros are: user, detail, domain, tag, delimiter, subdomain, extension. They have the obvious meaning; subdomain refers to the part of the domain before the dot, i.e., if the pattern is @.domain and the input is user@host.domain then subdomain refers to host, extension is the delimiter and the detail together (provided the address contains them).


alias*@.domain     user${extension}@${subdomain}.domain

provides the following mappings:

alias@host.domain             user@host.domain
alias+detail@host2.domain     user+detail@host2.domain

Running sendmail X

Starting sendmail X

All components of sendmail X are under control of the MCP which must be started as root in the directory /var/spool/smx (i.e., the main queue directory, see Section 2.4: SMXQDIR) using

# ./ start

The script contains the runtime path for MCP based on the data used by configure as well as a reference to the sendmail X configuration file.

To stop the entire sendmail X MTS use

# ./ stop
or simply terminate the MCP, it will forward the signal to all processes it started.

The MCP provides some restart functionality: if a process terminates (e.g., crashes), it will restart it unless the exit code indicates that a restart is useless, e.g., EX_USAGE. Moreover, the processes listed in the restart dependencies will be stopped and started too.

Using sendmail X only for Outgoing Mail

sendmail X can be used in combination with a MUA that speaks (E)SMTP directly or with the sendmail 8 MSP (Mail Submission Program) for outgoing mail. For the latter add this to your sendmail 8 file (see also misc/

R$* + X<@$*>    $#smx $@ localhost $: $1 <@$2>

R$+                     $: $>PseudoToReal $1            sender/recipient common
R$+                     $: $>MasqSMTP $1                qualify unqual'ed names
R$* + X<@$*>            $: $1 < @ $2 >
R$* < @ *LOCAL* > $*    $: $1 < @ $j . > $2

Msmx,   P=[IPC], F=kmDFMuXa, S=EnvFromSMTP/HdrFromSMTP, R=EnvToSMTP/HdrToSMTP,
        E=\r\n, L=990, T=DNS/RFC822/SMTP,
        A=TCP $h 2009

and run the SMTP server of sendmail X as listener on localhost:2009. Then mail to $<$user+X@domain$>$ will be sent via sendmail X, i.e., by adding the sequence +X to the address $<$user@domain$>$ the mail will be redirected to sendmail X (and +X will be removed). After initial testing the relay mailer can be changed to use port 2009 by default hence the local additions shown above can be removed.

There are also other programs available as substitute for the command line invocation of sendmail as mail submission program, e.g., mini_sendmail [Posb].

Using sendmail X for Incoming Mail

Local Delivery and Specifying Local Domains

If the domain of a recipient address matches an entry in mailertable (see Section 3.9.3) with the right hand side lmtp:4.1then SMTPC talks LMTP over the local socket lmtpsock (see 3.11). If you have an LDA that runs as daemon and can talk LMTP over a local socket you can use it for local delivery. It is also possible to use procmail [vdBG] in LMTP mode and start it from mcp, see smx.conf. See contrib/procmail.lmtp.p0 for a patch for procmail 3.22 to allow handling of addresses with extensions (+detail) in LMTP mode. A mailertable for local delivery via LMTP should look like this:

localhost lmtp:
MY.DOM lmtp:

By default mail to addresses whose domain part is listed in mailertable with RHS lmtp: is allowed, i.e., those domains are considered local and hence relaying (even though technically this might not be called relaying) to them is allowed. This behavior can be turned off (see Section 3.10, item 3e) in which case it is necessary to also allow relaying to these domains which can be done either via the access map (see Section 3.10, 3h), or the command line option -T for SMTPS. This allows for treating (some of) these domains as private by not allowing relaying to them, hence they will be only reachable from systems from which relaying is allowed.

Specifying Valid Local Addresses

To validate addresses for local domains, SMAR uses the Berkeley DB hash map aliases.db, which can be created using createmap, or a map specified by the option local_user_map (see Section 3.9.2, item 7). The key in the map must be the local part of a valid (local) e-mail address. If the local part cannot be found in either map, the address is rejected.

To list valid local addresses in the alias map the right hand side must be the string ``local:'', e.g.,

postmaster: $<$user@host.domain$>$
abuse: user+abuse
user++: local:
user: local:

Note: local addresses are checked for the envelope recipient and sender.

Using sendmail X as Gateway

sendmail X can easily be used as an internet gateway. To override routing, mailertable entries (see Section 3.9.3) can be specified. A list of valid addresses can be made available via the access map by allowing relaying to those addresses instead of entire domains, e.g.,

to:user1@my.domain relay
to:user2@my.domain relay
to:postmaster@my.domain relay
cltaddr:10.12 relay

Using sendmail X as Backup MX Server

The previous section showed how to specify valid remote addresses if all of them are known. However, for systems that act as backup MX servers it might not be simple to always keep such a list up to date. In that case, a default entry for a domain should be made, e.g.,

to:user1@other.domain relay
to:user2@other.domain relay
to:postmaster@other.domain relay
to:@other.domain error:451 4.3.3 Try main MX server
cltaddr:10.12 quick:relay
cltaddr: quick:relay

The last two entries allow local systems to send mail to any user at other.domain; without those entries mail to unlisted users will be (temporarily) rejected and hence cannot be delivered via this system.

Note about Backup MX Servers

It is not a good idea to run a backup MX server B for a host A that has stronger anti-spam measures; if mails are sent to A via B, then B may accept them for delivery, but A may reject them and hence B has to sent bounces, which, in case of spam, are most likely to forged addresses, hence those bounces will only cause additional problems. The opposite case (B has stronger anti-spam measures than A) can cause the rejection of mail that A actually wanted to receive. Hence B and A should have the same anti-spam measures; i.e., a system that acts as backup MX server for another one should perform the same anti-spam checks as the main MX server(s).

Miscellaneous Programs

Do not run programs as root User

Almost all sendmail X programs (except for MCP) refuse to run with root privileges. To run a program as a different user the utility misc/runas can be used, e.g., after installation in /usr/local/bin/

# /usr/local/bin/runas smxq mailq -V
(specify -h to see the usage).

Displaying Content of Mail Queues

The program mailq displays the content of the mail queues (defedb and ibdb). Currently its output is in a similar format as the sendmail 8 version. The option -h shows how to use the program; see the previous section about using runas for mailq. Note: the output of this program might not be accurate due to internal buffering by QMGR. Moreover, this program reads DEFEDB in such a way that only entries that have been checkpointed (see Section 8.2 about options for checkpoints) are shown. This is done to avoid interference with the operation of QMGR.

Interacting with QMGR

The program qmgrctl allows to interact with the QMGR via the control socket (see Section 3.8, item 3). Invoke qmgrctl -h to see the available options. By default the program will show the current status of QMGR. If QMGR has been compiled with the option QMGR_STATS (see Section 12.1) then additional statistics is available, e.g., the number of transactions and recipients that has been handled.

Enhancement to this program are welcome to provide more functionality.

Reloading Maps

Maps (for SMAR and QMGR) can be reloaded by moving the old db file out of the way, creating a new file and then sending a USR1 signal to the appropriate process to reopen the map.

# mv $MAP.db $MAP.old.db
# /usr/local/bin/runas smxq createmap -F $MAP.db < $MAP
# kill -USR1 $PID

Note: for QMGR it is also possible to use qmgrctl -r instead, see Section 4.6.3.


Logging is done via syslog(3) (see Section 3.5, 1) or to stdout/stderr, which is redirected by the default MCP configuration to PROG.log. The logging format is not yet completely consistent across programs. Moreover, the logging entries might not be easy to understand because they contain some details which are not interesting to a potential postmaster, but to developers. Nevertheless, the logging entries should show the flow of mail through the system. See Section 10.2 for an explanation of the format of logfile entries.

Note: logfiles must exist with the proper owner and permissions to be used. Neither MCP nor the modules will currently create logfiles. This is done by make install, i.e., misc/, which parses smx.conf to extract the section titles/names and user entries to create the logfiles with the correct name and owner. This does not (yet) properly work if unique logfile names are created, see Section 8.2, 3.

Logfile Rotation

Unless syslog(3) is used (see Section 3.5, 1), logfile rotation can be achieved by copying the existing logfile to a backup file, e.g.,

# cp qmgr.log qmgr.log.0

and sending a USR2 signal which will cause the processes to rewind the logfile. Note: the author is aware that this is not an optimal solution, however, using syslog(3) will usually provide a better way.

Regular Checks

There are at least two things that should be done regularly:

  1. Check the logfile for errors4.2:

    $ egrep 'sev=(ALERT|CRIT|ERR|FAIL)|\<assertion\>' $LOGFILE

  2. Keep track of the size of the processes, e.g,

    $ date >> $SMXPROCS
    $ ps axuww | grep '^smx' | sort >> $SMXPROCS

    If one of the processes continuously grows then sendmail X should be compiled with -DSM_HEAP_CHECK (see Section 12.1.1) and a heap dump should be taken regularly by sending the USR1 signal to the process. By comparing subsequent heap dumps it should be possible to locate a possible memory leak.

Please report problems that cannot be resolved locally, see Section 1.4.1.

Dealing with Errors

Resource Problems

Resource problems in certain parts of the code can lead to a stop of the involved program. In such a case it will be restarted automatically but if the resource problem has not been taken care of the MTS may stop again. In that case manual interaction is required. The simple solution to a resource problem is of course to add more resources (RAM/disk) or to free up some resources, e.g., stopping programs that do not need to run or deleting unused files. There are also ways to control resource usage within sendmail X:

Database Problems

See Section 7.1 for some background information about the usage of the various databases before trying to fix any possible problems.

If the deferred database is corrupted then the Berkeley DB utilities to deal with such situations should be tried [Sleb], e.g., db_recover.

Currently messages stored in CDB have the transaction identifier (ss_ta, see Section 10.2) as filename. In the worst case, i.e., if IBDB or DEFEDB are destroyed, this allows to reconstruct the envelope data together with the logfile entries. See the script misc/ for an example, here is a description of its operation. First, check which messages are still in CDB: in the CDB directory (3.4: CDB_base_directory) issue:

# ls -1 [0-9A-F]/S*

Then search for each of those transaction ids ($TAID) in the logfile ($LOG):

$ egrep "ss_ta=$TAID, (mail|rcpt)=" $LOG | \
  sed -e 's;^.*\(mail=<.*>\), .*;\1;' -e 's;^.*\(rcpt=<.*>\), .*;\1;'

will show the sender (mail=) and the recipients (rcpt=). Based on this data it is possible to resend the messages.

Note: contributions in this area are welcome, e.g., better scripts that perform more checks and maybe allow for completely automatic recovery.

Writing Core Dumps

By default, all sendmail X processes are executed in the main queue directory. As those processes are running with different user and group ids not all of them can write a core dump into that directory if a fatal error occurs. Some operating systems have commands (e.g., coreadm(1M) on SunOS 5.x) to specify a different directory in which a core dump is written. On operating systems where such a command is not available, the option working_directory can be used (see Section 8.2, item 4).

Replacements for Features available in other MTAs

sendmail X.0 does not offer some of the features that are available in other MTAs. This section describes replacements or workarounds for some of those features.

  1. Address Masquerading: The best way to use the correct e-mail addresses is to properly configure your MUA. Some MUAs offer more flexibility for this than the default masquerading features of sendmail 8, e.g., mutt [mut] allows to select sender addresses based on recipient addresses.

    Alternatively a mail submission program (MSP) can be used which offers address rewriting capabilities, e.g., the MSP from sendmail 8.

  2. .forward: procmail can be used as LDA (see Section 4.3.1) and its configuration file .procmailrc allows to implement the same functionality as a .forward from sendmail 8 and some other MTAs.

  3. Sending mail to programs: see previous item 2: this can be done with the help of procmail.

  4. Appending mails to files: see item 2: this can be done with the help of procmail.

Policy Milter

Policy Milter Overview

Sendmail X has support for a policy milter which is similar to a milter in sendmail 8. The most important difference is that a policy milter in sendmail X can (currently) not modify any part of an e-mail, it can only decide whether to accept or reject an e-mail.

Native Policy Milter API

Note: this API may evolve over time.

Naming conventions: A policy milter (also called pmilter) is a program that uses the API provided by libpmilter. The latter interacts with the SMTP servers via an internal protocol, i.e., this protocol can be changed without changing the visible API and should not directly be accessed by a user application.

Data Structures

libpmilter itself uses three context structures all of which must be treated by a milter as opaque.

  1. pmg_ctx: ``global'' libpmilter context (only one per process).
  2. pmss_ctx: libpmilter context per SMTP server that connects to this instance.
  3. pmse_ctx: libpmilter context per SMTP session.

Any of the libpmilter functions takes one of these contexts as parameter; e.g., all SMTP session oriented functions have a parameter of type pmse_ctx_P.

A milter can have its own contexts for each of these three environments, see Section 5.2.5.

Start and Stop

The functions in this section return SM_SUCCESS (0) on success and a negative value in case of an error.

First libpmilter must be initialized; a pmilter must specify a variable pmg_ctx_P pmg_ctx; which is passed per reference to the initialization function:

sm_ret_T sm_pmfi_init(pmg_ctx_P *pmg_ctx)

The pmilter global context must be treated as opaque data structure, it is passed to subsequent libpmilter function calls.

Next pmilter starts libpmilter by handing control over to the library; the pmilter passes a description of its requirements and functionality:

sm_ret_T sm_pmfi_start(pmg_ctx_P pmg_ctx, pmilter_P pmilter)

A milter can stop by calling:

sm_ret_T sm_pmfi_stop(pmg_ctx_P pmg_ctx)

There are various functions to set some options which can be called after libpmilter is initialized but before it is started. To set the path of the Unix domain socket over which the SMTP servers (see Section 3.10, item 17) and libpmilter communicate:

sm_ret_T sm_pmfi_setconn(pmg_ctx_P pmg_ctx, const char *path)

The backlog parameter of the listen(2) function can be set:

sm_ret_T sm_pmfi_setbacklog(pmg_ctx_P pmg_ctx, int backlog)

The debug level of libpmilter might be set via (this requires knowledge of the internals of the library which can be acquired by looking at the source code):

sm_ret_T sm_pmfi_setdbg(pmg_ctx_P pmg_ctx, int debuglevel)

To set the communication timeout:

sm_ret_T sm_pmfi_settimeout(pmg_ctx_P pmg_ctx, int timeout)

New SMTP Server

Whenever an SMTP server connects to a milter an option negotiation is performed (similar to ESMTP itself). A pmilter can check whether server capabilities are acceptable and return the options that it wants:

sm_ret_T pmfi_negotiate(pmss_ctx_P pmss_ctx, uint32_t srv_cap, uint32_t srv_fct, uint32_t srv_feat, uint32_t srv_misc, uint32_t *pm_cap, uint32_t *pm_fct, uint32_t *pm_feat, uint32_t *pm_misc)

Currently only the capabilities field is used: srv_cap is set by the SMTP server to a list (implemented as bit field) of phases of the ESMTP dialogue that can be passed to a pmilter. In turn the pmilter must set *pm_cap to includes those phases of the ESMTP dialogue that it wants to receive. For details, see include/sm/pmilter.h. For each of those phases a callback is invoked (see Section 5.2.4) which must be set by the pmilter in its description structure struct pmilter_S (see include/sm/pmfapi.h).

SMTP Session and Transaction

The protocol steps from ESMTP are forwarded to the policy milter which can decide to accept or reject them.

Set and Get pmilter Contexts

As explained in Section 5.2.1 a milter can have a ``global'' context pmilter_g_ctx, a context per SMTP server pmilter_ss_ctx, and a context per SMTP session pmilter_se_ctx. The following functions are provided to set and get these contexts.

Set the ``global'' context pmilter_g_ctx:

sm_ret_T sm_pmfi_set_ctx_g(pmg_ctx_P pmg_ctx, void *pmilter_g_ctx).

This must be done after libpmilter has been initialized but before control is transferred to it.

To retrieve the ``global'' context invoke:

void *sm_pmfi_get_ctx_g(pmg_ctx_P pmg_ctx)

Note: this requires the ``global'' libpmilter context which is not usually passed to pmilter functions in callbacks. See below how to access the ``global'' context pmilter_g_ctx from other places.

To set the pmilter context per SMTP server pmilter_ss_ctx use:

sm_ret_T sm_pmfi_set_ctx_ss(pmss_ctx_P pmss_ctx, void *pmilter_ss_ctx);

to retrieve it call:

void *sm_pmfi_get_ctx_ss(pmss_ctx_P pmss_ctx)

The ``global'' pmilter context pmilter_g_ctx can be retrieved from the libpmilter context per SMTP server:

void *sm_pmfi_get_ctx_g_ss(pmss_ctx_P pmss_ctx)

At the lowest level a context per SMTP session pmilter_se_ctx can be set via:

sm_ret_T sm_pmfi_set_ctx_se(pmse_ctx_P pmse_ctx, void *pmilter_se_ctx)

and retrieved by:

void *sm_pmfi_get_ctx_se(pmse_ctx_P pmse_ctx).

Just as before there is a function to retrieve the pmilter context per SMTP server pmilter_ss_ctx from the libpmilter context per SMTP session:

void *sm_pmfi_get_ctx_ss_se(pmse_ctx_P pmse_ctx)

Note: if a pmilter uses these contexts, then it is useful that each ``lower level'' context contains a link to its ``higher level'' context. That is, each pmilter context per SMTP session pmilter_se_ctx should have a pointer to its pmilter context per SMTP server pmilter_ss_ctx which in turn should have a pointer to the ``global'' pmilter context pmilter_g_ctx. This allows access from a function that is specific to a SMTP session to each relevant context.

Accessing MTA Symbols

A pmilter can set a list of symbols it wants to receive from the MTA by calling

sm_pmfi_setmaclist(pmss_ctx_P pmss_ctx, uint where, ...)

during the option negotiation, i.e., in pmfi_negotiate(). The parameter where denotes the stage of the ESMTP dialogue when the value of the symbol should be sent. It must be one of

PM_SMST_CONNECT Session start
PM_SMST_DOT Final dot of mail body

A sequence of up to PM_MAX_MACROS macros can be requested which must end with PMM_END. Valid values are:

PMM_SRVHOSTNAME hostname of SMTP server  
PMM_SEID session id  
PMM_MAIL_TAID transaction id  
PMM_DOT_MSGID Message-Id  

PMM_MAIL_TAID cannot be requested before PM_SMST_MAIL and PMM_DOT_MSGID can only be requested at stage PM_SMST_DOT.

To retrieve the value of a symbol the function

sm_pmfi_getmac(pmse_ctx_P pmse_ctx, uint32_t macro, char **pvalue)

can be used in the various callback functions of the ESMTP dialogue. If the macro was not in the request list, an error will be returned. If the macro has not yet been received, *pvalue will be NULL. Otherwise *pvalue will point to the value of the macro. Note: the string to which *pvalue points must not be changed.

Further Capabilities

In addition to selecting which SMTP commands to send to pmilter (see Section 5.2.3), there are some more capabilities available:

  1. SM_SCAP_PM_RCPT_ST causes the MTA to send RCPT information even if the command has been rejected, e.g., because the recipient is unknown, the recipient has been rejected due to access map checks, or relaying has been denied. Note: RCPT commands that are rejected for other reasons, e.g., because the address is syntactically invalid, or some limit (maximum number of recipients) is exceeded, will not be sent to pmilter.
  2. SM_SCAP_PM_MSG_RC allows a pmilter to return a reply code as specified in 5.2.9 from pmfi_msg(). This is useful if a pmilter can make a decision about the mail without having to read the entire message. If this capability is turned on, pmfi_msg() must return SMTP_R_CONT for each message chunk by default to receive subsequent parts. Otherwise pmfi_eom() will not be called but the return code from pmfi_msg() will be used at the end of the message (in response to the final dot).

The function

sm_ret_T sm_pmfi_getstatus(pmse_ctx_P pmse_ctx, sfsistat_T *pstatus)

should be used in that case to access the current SMTP reply code for the command. This functionality is useful for a pmilter that wants to keep track of all recipients, not just those which are accepted, e.g., to deal with dictionary attacks.

Miscellaneous Functions

To set a reply text in an SMTP session or transaction oriented callback in addition to the reply code use:

sm_ret_T sm_pmfi_setreply(pmse_ctx_P pmse_ctx, const char *reply)

Note: the reply string must contain the full SMTP reply, i.e., it must be of the form

XYZ D.S.N text\r\n

where XYZ is a valid SMTP reply code (see RFC 2821 [Kle01]) which must match the return code of the function from which sm_pmfi_setreply() is called, D.S.N is an enhanced status code as defined in RFC 3463 [Vau03] and the rest is an explanation of the status including CRLF (\r\n).

Return version number of libpmilter:

sm_ret_T sm_pmfi_version(pmg_ctx_P pmg_ctx, uint32_t *major, uint32_t *minor, uint32_t *patchlevel)

This can be used to compare the version number of the library against which pmilter is linked with the version number against which pmilter is compiled. The major version numbers must match otherwise the program will not run.

Signal handler function:

sm_ret_T pmfi_signal(pmg_ctx_P pmg_ctx, int sig)

This will be called when a USR1 or USR2 signal is received; it is not called within a signal handler, i.e., the code does not have to be signal-safe. Note: this is not yet implemented.

Return Values

SMTP Session and transaction oriented functions use sfsistat_T as return type. Allowed values for this type are (as defined in include/sm/smreplycodes.h):

Additionally return values can be modified by using SMTP_R_SET_QUICK(returnvalue). See Section 3.9.3 for the effects of this.

For functions that use sm_ret_T as return type a successful call returns SM_SUCCESS (0) and a negative value in case of an error.

Implementation Notes

As libpmilter currently does not keep track of the status of a transaction or session, the functions pmfi_abort() and pmfi_close() may be called even if no transaction or session is currently active. This can happen if an SMTP server unexpectedly aborts the connection to a policy milter. An application must be aware of this and keep track of its state properly.

Policy Milter Examples

The program libpmilter/example-pmilter-0.c is a simple example how to write a policy milter. It might be useful as a template for other milters.

Also available is a policy milter contrib/milter-spamd.c that offers an interface to spamd(1) which is a daemonized version of spamassassin(1). milter-spamd.c is written by Daniel Hartmeier [Harc] (see the file itself for the Copyright) for sendmail 8 and modified to work with the policy milter API of sendmail X.



If something goes wrong then the component which fails usually logs an error message. Depending on the configuration, an error is either logged via syslog(3) or printed into a logfile (as explained in Section 4.8). Note: even if the system is configured to use syslog(3) (Section 3.4, item 1) errors at startup are printed to the logfile if those errors occur before the configuration is read, hence those files need to be checked too.

If sendmail X fails to start properly the reason should be logged as explained before. Some possible reasons are

  1. invalid configuration file: use misc/smconf to check the syntax before deploying a new configuration file.

  2. wrong permissions: check the permissions as explained in Section 2.4:
    $ ./misc/

  3. missing or invalid maps: make sure maps are created properly with createmap(8).


The following problems exist in this version of sendmail X.0:

Syntax and Semantics Checks in SMTP Server

The SMTP server has some builtin checks which are explained in the following.

The EHLO parameter is checked against the local hostname unless the connection comes from localhost (IP address or the access map returned quick:ok or quick:relay, see Section 3.9.3.

Strict RFC Compliance

The SMTP server currently enforces fairly strict RFC 2821 compliance. For example, a MAIL command must be given in the following format

MAIL From:<user@some.domain>

i.e., the angle brackets are required, there must be no space after ":", etc. This has the useful side effect of catching some spam programs:

5.5.0 Syntax error., input=MAIL FROM: <>

Moreover, the server requires that lines end in CRLF (\r\n), it will not accept command input without the correct line ending, i.e., trying to do that will cause a read error.

Another requirement is that MX records must point to hostnames, not IP addresses [Moc87]. This applies to receiving mail - a MAIL address using a domain whose MX record points to an IP address will be rejected (553 5.1.8 Sender address does not exist) - as well as to sending mail - a RCPT address with a domain whose MX record points to an IP address is not resolved by SMAR.

Security Checks

There are currently no additional security checks when creating/accessing files or directories besides those provided by the operating system. This could be a problem if MCP is misconfigured because it runs as root. Hence it will simply overwrite existing files if those are specified in the configuration file. The other modules run as non-privileged users, hence the OS provides sufficient access checks - unless the system is misconfigured and the sendmail X accounts are misused for other purposes too.


Besides the obviously missing functionality there are some other things that may restrict the use of sendmail X in certain environments. Here is an incomplete list:

Everything that is not described in the documentation does either not exist in the current version of sendmail X, or is unlikely to work. However, there may be omissions in the documentation, please inform the author of such bugs.

Code Review, Enhancements, Patches

Source code inspection as well as patches and suggestions are very welcome.

Enhancements and extensions are very welcome too, especially to extend the basic functionality of the current sendmail X release.


Porting to currently unsupported platforms including non-Unix systems is encouraged. Note that the destination system must support statethreads [SGI01] and Berkeley DB 4.x. It might be necessary to port those first.

Version Naming

Each sendmail X version has a name in the following format:

sendmail X.major.minor.[qualifier]qualifier-version.patchlevel

The major number changes between releases when new features are introduced (major changes, but see below about the development phases). The minor number changes when no new features are introduced, but bugfixes and (portability) enhancements are made. That is, no configuration changes are needed when going from one minor version to the next. The patchlevel number is used for intermediate patches between releases, e.g., if something is broken but it is not important enough for a new release because it is barely used or encountered.

There are several different qualifiers:

  1. PreAlpha: This means the software is not feature complete and hence might be missing some functionality that is considered important by different users. Additionally, there is most likely no compatibility in data structures stored on disk between different pre-alpha versions, e.g., when upgrading from PreAlpha16 to PreAlpha17 the main queue format may have changed without checks in the software for this. Hence old queues must be drained before upgrading. Moreover, the protocols used for communication between sendmail modules may have changed without providing backward compatibility, therefore modules from different releases must not be used together. Such incompatibilities are usually stated in the list of changes ChangeLog.

    Do not run this on a production server unless you are aware of the possible consequences. The software is still under development and not fully functional. Moreover, it may not be sufficiently tested.

  2. Alpha: In this state the software is ready for public testing but its features may still change.

  3. Beta: Feature changes are unlikely, but still possible if required. Usually only bugfixes occur between beta versions.

  4. Gamma: This is a release candidate. Usually only critical bugfixes occur between gamma versions. There might be no gamma versions at all if beta testing was considered successful and sufficient.

  5. A release version does not have an explicit qualifier.

The qualifier-version is used to distinguish between different version of the same qualifier, e.g., PreAlpha16 and PreAlpha17. It is 0 for a release version.

Examples for version names: sendmail X.0.0.PreAlpha19.0, sendmail X. (this is the name of the first release).

See the file include/sm/version.h how the version string is converted into a 32 bit number that denotes the version number.


From time to time snapshots may be made available. Those are marked with a date in the distribution file name, e.g., smX- The name indicates that it is a snapshot of what will become version smX-, i.e., the next release will have the given version number (without the date). The only other indication in the distribution is the inclusion of an s in the version number that is shown in the version output of the main components. A snapshot did not go through the usual release cycle and is made available as technology preview.

Data Flow in Sendmail X

Data Flow in Sendmail X

This Section explains how Sendmail X stores information about messages that are transferred. It gives some background information which is useful for troubleshooting. Details about the operation of sendmail X can be found in [Aßmb].

Sendmail X uses two different databases on disk to store envelope information (sender and recipients): IBDB: incoming backup database, DEFEDB: deferred envelope database, and one database to store message contents: CDB: content database. See Section 2.4.1 about the location and layout of these databases7.1. The queue manager additionally uses two internal envelope databases: IQDB (Incoming Queue DataBase) and AQ (Active Queue).

Incoming mails are accepted by the SMTP servers which store the content in the CDB (complete messages including headers in the format as received). The envelope information, i.e., sender (MAIL) and recipients (RCPT), is stored by the queue manager in IQDB and written to IBDB which is just a log of envelope data and what happened to it. That is, the files in IBDB are written sequentially and are continuously growing. If a file reaches its size limit (see Section 3.8: IBDB), then it is closed and a new file is opened. For a delivery, the envelope information must be transferred into AQ. For incoming mail this happens as soon as a transaction is accepted, in which case the data is moved from IQDB to AQ. A transaction is only accepted if the message is safely written to CDB and the envelope information has been committed to IBDB, i.e., all information is committed to persistent storage7.2.

The scheduler in QMGR takes recipient envelopes from AQ and creates transactions which are given to the SMTP clients for delivery. An SMTP client takes the transaction information and tries to send a message whose content is read from CDB. After a successful delivery attempt a record is written to IBDB that logs this information. A cleanup task removes periodically old IBDB files which contain only data that is no longer referenced.

The deferred envelope database is only used if a message cannot be delivered during the first attempt. In that case the appropriate envelope data is added to DEFEDB and a record is written to IBDB stating that the data has been transferred to DEFEDB. Entries in DEFEDB contain a timestamp called next-time-to-try at which QMGR reads them from the database into AQ and the scheduler tries another delivery attempt. If that succeeds, the entries are removed from DEFEDB, otherwise they are either requeued with a new next-time-to-try (in case of a temporary error) or a DSN (bounce message) is generated (in case of a permanent error).

Advanced Configuration Options


Some configuration options are only needed in special situations and may require background knowledge of the involved systems. Those advanced configuration options are explained in the subsequent sections.


Usually flags are not set and hence a configuration file only needs to turn on flags (if required). However, in some cases flags are set by default and under some rare circumstances need to be disabled. To achieve this, the name of the flag can be prefixed with one of not_, dont_, no_, -, !, or ~, e.g., ~remove_unused_logfiles or dont_remove_unused_logfiles.

Advanced Configuration for MCP

  1. max_processes: maximum number of processes to start [default: 1].
  2. min_processes: minimum number of processes to start [default: 1].
  3. use_id_in_logfile_name: if more than one process can be started then it might be useful to have unique logfiles unless the processes use syslog(3). If set, this (boolean) option causes MCP to include a unique identifier (the same as for pass_id, which must be used too) in the logfile name. By default the logfile has the name of the section (or the section keyword if no section name is given), preceeded by the log directory (option -L for MCP), and .log appended. If use_id_in_logfile_name is turned on, then the numeric id is added before the extension, e.g., /var/log/smx/mailer0.log for -L /var/log/smx/ and a section with the name mailer.
  4. working_directory: perform a chdir(2) to the specified directory before executing the process. Note: this option essentially requires that all relevant pathnames in the configuration file are absolute, otherwise it is very easy to misconfigure some pathnames, especially those shared between different processes.

Note: the number of processes for almost all sendmail X modules should be 1. It must be 1 for QMGR and SMAR, it can be larger than 1 for SMTPC. For SMTPS it should be 1 in the default setup as the file descriptor to which MCP binds on behalf of SMTPS can be passed to only one process.

Advanced Configuration for QMGR

  1. connection_control_hash_table_size: size of the hash table used for connection control, i.e., number of incoming connections and connection rate (see Section 3.8, 17a and 17b)
  2. debug_level: debug level (only if compiled with QMGR_DEBUG).

  3. subsection DEFEDB: Note: The Berkeley DB documentation [Sleb] should be consulted before modifying any of these options.
    1. page_size: DB page size (this can only be set when the DB is initially created).
    2. cache_size: DB cache size.
    3. KBytes_written_for_checkpointing: If non-zero, a checkpoint will be done if more than the amount of KBytes of log data have been written since the last checkpoint (unit: KB).
    4. delay_between_2_checkpoints: Minimum delay between two checkpoints (unit: s).
    5. flags: flags for DEFEDB:
      1. remove_unused_logfiles: this is on by default, hence to turn it off one of the forms explained in Section 8.1.1, e.g., dont_remove_unused_logfiles, can be used. This should only be done if the Berkeley DB logfiles are removed some other way, e.g., after archiving.

  4. delivery_timeout: timeout for a single delivery attempt (unit: s). This value should be large enough that even big mails can be delivered over a slow link before the QMGR considers the delivery attempt a failure because the delivery agent did not return a result yet.

  5. flags: configuration flags:
    1. reuse_connection: try to reuse open SMTP connections for delivery. Note: this feature is still experimental.

  6. max_fds: maximum number of file descriptors. This sets an upper limit on the number of clients that can connect to QMGR.

  7. max_threads: maximum number of threads.
  8. min_threads: minimum number of threads.

  9. scheduler_timeout: as a safety measure against unforseen problems an item is removed from AQ after the specified timeout. This timeout must be large enough to allow for scheduling delays if all delivery agents are busy which can happen if deliveries are slow or if there are fewer delivery agents available than entries in the active queue.

  10. SMAR_timeout: timeout in address resolver, i.e., how long to wait for a result from SMAR (unit: s). Note: this value must be larger than the total DNS timeout and it must take alias expansion into account.

  11. subsection smtps:
    1. connection_control_cache_size: size of connection control hash table.

  12. tests: testing only (available if QMGR is compiled with -DQMGR_TEST). See the source code for details.

Advanced Configuration for SMAR

  1. DNS_flags: valid flags are:
    1. use_TCP: use TCP instead of UDP for connections to a nameserver. Note: currently the DNS resolver does not automatically fall back to a TCP connection if the reply was too big. This may be added in a later version.

    2. max_fds: maximum number of file descriptors. This sets an upper limit on the number of clients that can connect to SMAR.

    3. max_threads: maximum number of threads.
    4. min_threads: minimum number of threads.

    5. use_connect: use connect(2) even if using UDP. This is required on systems like FreeBSD jail(8).

Advanced Configuration for SMTP Server

  1. daemon_address: address for daemon to listen on; this should not be used in normal operation. Current (preliminary) format is: host:port, :port (listen on host (port defaults to 8000). Up to 16 addresses8.1can be specified. See the notes below.

  2. flags:
    1. background: fork(2) after start; this should not be used in normal operation.
    2. serialize_accept: serialize accept(2) calls, see the statethreads documentation [SGI01] for details.

  3. listen_queue: length of listen(2) queue; this must not be used in normal operation, i.e., if MCP is used.

  4. max_wait_threads: maximum number of waiting threads.
  5. min_wait_threads: minimum number of waiting threads.

  6. processes: number of processes to start.

  7. tls:
    1. DSA_cert_file: file with DSA certificate in PEM format.
    2. DSA_key_file: file with private key for DSA certificate in PEM format.

  8. wait_for_smar: maximum amount of time to wait for a reply from SMAR.

Notes: only one of daemon_address and pass_fd_socket must be specified. In normal operation it is almost always pass_fd_socket because the SMTP server cannot bind to privileged ports, hence the file descriptor must be passed from MCP.

Advanced Configuration for SMTP Client

  1. connect_only_to: Specify an IP address to which all outgoing mail is sent. This can be used for testing with otherwise real data, i.e., addresses, by running an SMTP sink8.2on a computer and specifying its IP address. Then all mails that should be sent via SMTP will go to that host instead of the addresses determined by SMAR. Note: it is nevertheless a good idea to use firewall rules to prevent mail going out to the internet, i.e., prohibit connections to port 25 to external hosts.

  2. debug_level: debug level (only if compiled with SMTPC_DEBUG).

  3. max_wait_threads: maximum number of waiting threads.
  4. min_wait_threads: minimum number of waiting threads.

  5. remote_port: port to which connections should be made. Note: if multiple SMTP clients are specified, all of them must use the same value for remote_port. Currently the scheduler requires that all SMTP clients behave the same. If different ports are required, then those must be listed in mailertable entries.

  6. tls:
    1. DSA_cert_file: file with DSA certificate in PEM format.
    2. DSA_key_file: file with private key for DSA certificate in PEM format.


Size of Queues, Caches, and Databases

All data structures in QMGR have some maximum size. This is not just done to avoid resource exhaustion in high load situations but also to provide a feedback loop between SMTP servers (producers) and SMTP clients (consumers). This feedback loop helps to avoid flooding the system with mails that it cannot deliver fast enough. The incoming queue (IQDB) and the active queue (AQ) implement this feedback loop. As explained in Section 7.1 the data from the SMTP servers is stored in the incoming queue first which has a fixed size. If more data is produced than taken out (by the scheduler into the active queue) the queue will fill up and the QMGR will throttle the SMTP servers by dynamically reducing the number of available threads. Throttling the SMTP servers is done based on various resources, e.g., IQDB, AQ, available disk space, and much more. Hence by limiting the size of IQDB (see Section 3.8, item 7a) and of course the maximum number of threads in the SMTP servers the incoming flow of messages can be controlled. The size of IQDB should be greater than the maximum number of threads in the SMTP servers multiplied by the average number of recipients, otherwise transaction will be rejected before all threads are busy.

The active queue should be large enough to provide enough work for all SMTP clients (threads) and it must be larger than the largest number of recipients accepted by a single transaction (see Section 3.8, item 1).

Disk I/O

In most MTAs disk I/O is the limiting factor unless special hardware is used which employs battery backed RAM cache to achieve high I/O rates (IOP: I/O operations). If multiple disks are available, they can be used to spread the load. Disk files (see Section 2.4.1) are used for:

  1. IBDB: the directory can be linked to a different disk.
  2. DEFEDB: the base directory can be changed via an option (Section 3.8, item 4a), as well as the directory for logfiles (Section 3.8, item 4b).
  3. CDB: the base directory can be changed via an option (see Section 3.4, item 2). Individual subdirectories (see Section 2.4.1) can be linked to different disks.

Processes and Threads

The main sendmail X processes are multi-threaded. However, two different threading implementations are used: POSIX threads (pthreads) for QMGR and SMAR and statethreads [SGI01] for SMTP server and client. Statethreads only switch between threads on network I/O operations as it is a threading implementation in user space without kernel support. Hence operations that can take a long time, e.g., computations for asymmetric cryptography (as required during the STARTTLS handshake) or in some cases even synchronous disk I/O, will not just stop a single thread but the entire process. If this happens it is possible to start multiple SMTP servers, see Section 8.5, item 6. If it becomes necessary to start multiple SMTP clients, then the MCP can be instructed to do so, see Section 8.2, item 1.

Format Specifications

Format of Session/Transaction Identifiers

The format of session and transaction identifiers is specified in include/sm/mta.h. For the SMTP server it consists of a leading 'S', a 64 bit counter and an 8 bit ``process'' identifier, both of which are printed in hexadecimal format. For the SMTP client it consists of a leading 'C', an 8 bit ``process'' identifier, a 32 bit counter, and a 32 bit thread index, all of which are printed in hexadecimal format.

Examples: S00000000407CE49200, C010000137D00000000.

SMTP server session/transaction identifiers are unique until the 64 bit counter wraps around, SMTP client session/transaction identifiers are unique only within a single invocation of QMGR.

Note: the format may change between different release of sendmail X, hence the identifiers should be considered opaque.

Logfile Format

The general format of entries in a logfile is a sequence of named field which are separated by commas. Each field consists of a name, an equal sign, and a value. If the value is a text field that is received from an external (untrusted) source, then all non-printable characters, commas, and percent signs are shown as their two digit hexadecimal ASCII representation with a leading percent sign. For example, the text

550 5.7.1 no, not now, 99% usage
is encoded as
550 5.7.1 no%2C not now%2C 99%25 usage

This encoding allows a logfile analyzer to use the comma symbol as a delimiter of fields without having to perform complicated parsing, e.g, the Unix awk utility can be used with comma as field separator. Note: suggestions for a better encoding or different solution for the problem are welcome (more details can be found in [Aßmb]).

Logfiles use the identifiers described earlier such that transactions and sessions can be easily recognized. For the following examples logfile entries have been slightly edited and line breaks have been inserted.

Here is one example of a session in an SMTP server:

ss_sess=S00000000407EAE3800, client_ipv4=,
ss_sess=S00000000407EAE3800, where=connection, starttls=successful
ss_sess=S00000000407EAE3800, ss_ta=S00000000407EAE4E00,
  mail=<>, stat=0
ss_sess=S00000000407EAE3800, ss_ta=S00000000407EAE4E00,
  rcpt=<>, idx=0, stat=0
ss_sess=S00000000407EAE3800, ss_ta=S00000000407EAE4E00,
  rcpt=<SOMEONE@SOME.DOMAIN>, idx=1, stat=0
ss_sess=S00000000407EAE3800, ss_ta=S00000000407EAE4E00,
  msgid=<>, size=1177, stat=0

The first entry shows a successful session creation including the IPv4 address and the hostname of the client. The second entry indicates that STARTTLS has been used. A new transaction is shown in the third entry and two recipients are given thereafter (along with the index idx). The last entry shows that the transaction was successful (status=0; 0 is used instead of 250 or other SMTP reply codes that indicate success) and the size of the received mail (in bytes) as well as its Message-Id.

Here is one example of a session in an SMTP client:

da_sess=C01000006C800000002, status=connected, port=25, addr=
da_sess=C01000006C800000002, where=connection, starttls=successful
da_sess=C01000006C800000002, da_ta=C01000006C900000002,
  ss_ta=S00000000407EAE4E00, mail=<>, stat=0,
  reply=250 2.5.0 MAIL command succeeded
da_sess=C01000006C800000002, da_ta=C01000006C900000002,
  ss_ta=S00000000407EAE4E00, rcpt=<>, stat=0,
  reply=250 2.1.5 RCPT ok
da_sess=C01000006C800000002, da_ta=C01000006C900000002,
  ss_ta=S00000000407EAE4E00, where=final_dot, size=1177, stat=0

This is very similar to the format of the entries entries in the SMTP server and should not require an explanation. In addition to the delivery agent session and transaction ids (da_sess and da_ta) the SMTP server transaction id (ss_ta) is logged too. This makes it simple to track a message through the MTS. Obviously ss_ta can be used for multiple outgoing messages if the incoming message has been sent to multiple recipients (maybe indirectly via an alias), hence this is not a unique identifier in the SMTP client log.

QMGR can also log the delay time for each recipient, e.g.,

func=q_upd_rcpt_ok, rcpt_id=S00000000407EAE4E00-000000,
  rcpt=<>, xdelay=0, delay=1
where xdelay is the time for this delivery attempt, and delay is the total delivery time.

Format of Received Header

The format of the Received: header added by the SMTP server is specified in smtps/smtps.c.

        id SMTP-TA-ID; DATE

where PROTOCOL is one of ESMTP, ESMTPS, ESMTPA, ESMTPSA, or SMTP [New04]. If STARTTLS is active, then (TLS=TLSVERSION, cipher=CIPHERSUITE, bits=CIPHERBITS, verify=VERIFYRESULT) is placed before id, where TLSVERSION is the TLS protocol version, e.g., TLSv1, SSLv3, SSLv2; CIPHERSUITE is the cipher suite that was in use, e.g., AES256-SHA, EDH-DSS-DES-CBC3-SHA, EDH-RSA-DES-CBC-SHA, CIPHERBITS denotes the effective keylength (in bits) of the symmetric encryption algorithm of the TLS connection, and VERIFYRESULT is one of the following:

OK verification succeeded.
NO no cert presented.
NOT no cert requested.
FAIL cert presented but could not be verified, e.g., the signing CA cert is missing.

Note: the name of the client is only shown if the access map feature is activated (see Section 3.10, 3h), otherwise the time-consuming DNS lookups (PTR and A records) are not performed.

Format of DSNs

DSNs (bounces) are currently not compliant to RFC 1891ff, however, a configuration option (Section 3.8, item 5(b)ii) can be set to send DSNs in MIME format, i.e., Content-Type: multipart/mixed with Content-Type: message/rfc822 for the original message. The format looks like this:

From: Mailer-Daemon@HOST.NAME
Subject: Undeliverable mail

Hi! This is the sendmail X MTA. I'm sorry to inform you that a mail
from you could not be delivered. See below for details.

and then a list of recipients and the reasons for the failure, e.g.,

550 5.7.1 <>... Access denied
during RCPT

Setup for STARTTLS

Certificates for STARTTLS

When acting as a server, sendmail X requires X.509 certificates to support STARTTLS: one as certificate for the server, at least one root CA (CAcert_file), i.e., a certificate that is used to sign other certificates, and a path to a directory which contains certs of other CAs (CAcert_directory). The file specified via CAcert_file can contain several certificates of CAs. The DNs of these certificates are sent to the client during the TLS handshake (as part of the CertificateRequest) as the list of acceptable CAs. However, do not list too many root CAs in that file, otherwise the TLS handshake may fail; e.g.,

error:14094417:SSL routines:SSL3_READ_BYTES:
sslv3 alert illegal parameter:s3_pkt.c:964:SSL alert number 47
You should probably put only the CA cert into that file that signed your own cert(s), or at least only those you trust. The directory specified via CAcert_directory must contain the hashes of each CA certificate as filenames (or as links to them). Symbolic links can be generated with the following two (Bourne) shell commands:
ln -s $C `openssl x509 -noout -hash < $C`.0
An X.509 certificate is also required for authentication in client mode, however, sendmail X will always use STARTTLS when offered by a server. The client and server certificates can be identical. Certificates can be obtained from a certificate authority or created with the help of OpenSSL. The required format for certificates and private keys is PEM. To allow for automatic startup of sendmail X, private keys must be stored unencrypted. The keys are only protected by the permissions of the file system, hence they should not be readable by anyone but the owner. If server and client share the same key it is ok to make the key group readable however. Never make a private key available to a third party.

More About Configuration, Compilation, Debugging, and Testing

Compile Time Options

There are several compile time parameters which might be useful in some situations that are listed below. Compile time options to turn on additional debugging are listed in the next section.

To enable QMGR statistics, e.g., number of transactions and recipients that have been handled, set QMGR_STATS.

The address resolver imposes limits on the number of MX and A records that it accepts when it does routing lookups. These macros are: SM_DNS_MX_MAX: maximum number of MX records for a domain, SM_DNS_A_PER_MX_MAX: maximum number of A records for one MX record, SM_DNS_A_MAX: maximum number of A records for a domain (after performing MX lookups).

Debugging Compile Time Options

There are several compile time parameters to support debugging. An option that applies to all modules (as they use the same libraries) is SM_HEAP_CHECK which turns on various heap checks and keep track of memory usage.

Other options are specific to a module and can be used to turn on debugging output. Since currently no logging abstraction is in use, the output is done on a per-module basis (whatever is simplest for the individual module). These compile time options are:

SC_DEBUG SMTPC debugging
SSQ_DEBUG SMTPS - QMGR communication debugging
SS_DATA_DEBUG SMTPS DATA stage debugging
SM_LIBDNS_DEBUG libdns debugging

For details see the source code.

Note: it is possible to set different debug levels for different debug categories in QMGR. For a list of categories see include/sm/qmgrdbg.h. To set a debug level n for a category c use the option -xc.n. The general syntax for the parameters is:

debugoptions ::= debugoption [ "," debugoptions ]
debugoption ::= range [ "." level ]
range ::= first [ "-" last ]

If level is omitted, it defaults to 1. Example: -x1-3.4,5.3,9-11

A simple way to set compile time options is to use:


A more complicated example is:

$ CFLAGS="-O -g -DSM_HEAP_CHECK -I/usr/local/include" \
 LDFLAGS="-L/usr/local/lib" \

Hint: it is useful to write the command line into a local file that can be reused for subsequent builds and versions.

Note: if configure has problems with OpenSSL because you do not have KerberosV installed, add


Possible Compilation Problems

If gcc is used as C compiler and full checking is turned on warnings like these are produced:

warning: unknown conversion type character `N' in format
warning: unsigned int format, sm_str_P arg (arg 3)
warning: too many arguments for format

Unfortunately gcc cannot be told about additional format specifiers and hence the misleading warning might be generated.

More About Test Programs

More Environment Variables used by Test Programs

The environment variable SM_NAMESERVER can be used to set a specific nameserver (IPv4 address) in case the simple script which extracts the first line beginning with nameserver from the file /etc/resolv.conf does not give the desired result.

If the example policy milter milter-regex is compiled, then tests for it can be run by setting SM_PMILTER_REGEX_TEST before using make check in chkmts/. Note: this requires that make check is executed in contrib/ before make check in chkmts/.

Other Potential Problems with Test Programs

Some of the test programs may generate warnings, e.g., most of the tree related programs cause compilers on 32 bit systems to emit a warning integer constant too large which can be ignored.

Known Test Program Problems specific to an OS/setup

FreeBSD systems when running in a jail(8) exhibit the following problems:

MacOS 10.3.4 has a problem with sigwait(3), see Apple's bug 3675391; hence sendmail X does not work on this OS (and other versions that have the same bug).


The license for sendmail X can be found in the file LICENSE distributed with the source code. Additionally, sendmail X contains code from other projects whose licenses can be either found in the respective source files or in statethreads/README for the statethreads library and db-4.3.28.NC/LICENSE for Berkeley DB. Some source code is licensed under a BSD license which can be found at the begin of those files.


Claus Aßmann.
Sendmail X.

Claus Aßmann.
Sendmail X: Requirements, Architecture, Functional Specification, Implementation, and Performance.

Project Cyrus.,

N. Freed.
SMTP service extension for command pipelining.
RFC 2920, Internet Engineering Task Force, 2000.

GNU Privacy Guard.

Evan Harris.
The next step in the spam control war: Greylisting.

Evan Harris.

Daniel Hartmeier. - milter-spamd.

P. Hoffman.
SMTP service extension for secure SMTP over TLS.
RFC 2487, Internet Engineering Task Force, 1999.

Simple mail transfer protocol.
RFC 2821, Internet Engineering Task Force, 2001.

P.V. Mockapetris.
Domain names - implementation and specification.
RFC 1035, Internet Engineering Task Force, 1987.


K. Moore and G. Vaudreuil.
An Extensible Message Format for Delivery Status Notifications.
RFC 3464, Internet Engineering Task Force, 2003.

John Myers.
Local mail transfer protocol.
RFC 2033, Internet Engineering Task Force, 1996.

J. Myers.
SMTP service extension for authentication.
RFC 2554, Internet Engineering Task Force, 1999.

Chris Newman.
ESMTP and LMTP Transmission Types Registration.
RFC 3848, Internet Engineering Task Force, 2004.



Jef Poskanzer.
graymilter - simple graylist mail filter module.

Jef Poskanzer.
mini_sendmail - accept email on behalf of real sendmail.

PGP keys.

State threads for internet applications, 2001.

Berkeley DB 4.4.XX Change Log.

Berkeley DB Tutorial and Reference Guide, version 4.3.28.

G. Vaudreuil.
Enhanced mail system status codes.
RFC 3463, Internet Engineering Task Force, 2003.

S.R. van den Berg and Philip Guenther.


Configuration for SMTP Server
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for SMTP Server
Configuration Options for SMAR
Configuration for MCP | Declaring Maps for SMAR | Configuration for SMTP Server
Configuration Options for SMAR
Configuration Options for SMAR
Configuration Options for SMAR
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for QMGR
Configuration for MCP
Configuration for SMTP Server
Advanced Configuration for SMTP
Configuration for QMGR
Configuration for SMTP Server | Configuration for SMTP Client
Configuration for SMTP Server | Configuration for SMTP Client
Advanced Configuration for QMGR
Common Global Configuration
Configuration for SMTP Server
Configuration for SMTP Server | Configuration for SMTP Client
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for QMGR
Advanced Configuration for SMTP
Advanced Configuration for QMGR
Advanced Configuration for QMGR
Configuration for QMGR
Advanced Configuration for SMTP
Advanced Configuration for QMGR | Advanced Configuration for SMTP
Advanced Configuration for QMGR
Configuration for SMTP Server
Configuration for SMTP Server
Advanced Configuration for QMGR
Advanced Configuration for SMAR
Configuration Options for SMAR
Configuration Options for SMAR
Configuration Options for SMAR
Advanced Configuration for SMTP | Advanced Configuration for SMTP
Advanced Configuration for SMTP | Advanced Configuration for SMTP
Configuration Options for SMAR
Common Configuration Options
Declaring Maps for SMAR | Configuration Options for SMAR | Configuration Options for SMAR | Configuration Options for SMAR
Configuration for QMGR | Configuration Options for SMAR | Configuration Options for SMAR | Configuration for SMTP Server | Configuration for SMTP Server | Configuration for SMTP Server | Advanced Configuration for QMGR | Advanced Configuration for QMGR | Advanced Configuration for SMTP
Configuration for SMTP Server
Configuration Options for SMAR
Configuration Options for SMAR
Configuration Options for SMAR | Configuration for SMTP Server
Configuration for MCP | Configuration for MCP
Configuration for QMGR
Configuration for QMGR
Common Global Configuration
Configuration for SMTP Server
Common Configuration Options
Configuration Options for SMAR | Configuration Options for SMAR | Configuration for SMTP Server
Configuration for QMGR
Configuration for SMTP Server | Configuration for SMTP Client
Advanced Configuration for QMGR
Configuration for SMTP Server | Configuration for SMTP Client
Advanced Configuration for SMTP
Configuration for MCP
Configuration for SMTP Server
Configuration for QMGR
Configuration for SMTP Client
Configuration Options for SMAR
Configuration Options for SMAR
Configuration Options for SMAR
Common Configuration Options
Configuration for QMGR
Configuration for QMGR | Configuration Options for SMAR | Configuration for SMTP Client
Configuration Options for SMAR
Configuration Options for SMAR
Declaring Maps for SMAR
Declaring Maps for SMAR
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for QMGR
Configuration for QMGR
Configuration for QMGR
Configuration for QMGR | Configuration for QMGR
Configuration for QMGR
Advanced Configuration for QMGR | Advanced Configuration for SMAR
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for QMGR
Advanced Configuration for MCP
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for SMTP Server | Advanced Configuration for QMGR | Advanced Configuration for SMAR
Configuration for SMTP Server
Advanced Configuration for SMTP | Advanced Configuration for SMTP
Configuration for QMGR
Configuration for QMGR
Configuration for QMGR
Advanced Configuration for MCP
Advanced Configuration for QMGR | Advanced Configuration for SMAR
Advanced Configuration for SMTP | Advanced Configuration for SMTP
Configuration Options for SMAR | Configuration Options for SMAR
Configuration Options for SMAR
Configuration Options for SMAR
Configuration for QMGR
Configuration for QMGR
Common Configuration Options
Advanced Configuration for QMGR
Configuration for MCP
Configuration for MCP
Configuration for MCP | Configuration for MCP | Declaring Maps for SMAR | Configuration for SMTP Server
Configuration for SMTP Server
Configuration for MCP | Declaring Maps for SMAR | Configuration for SMTP Server
Configuration Options for SMAR
Configuration for SMTP Server | Advanced Configuration for SMTP
Configuration for SMTP Server
Configuration for QMGR
Configuration for QMGR
Advanced Configuration for SMTP
Advanced Configuration for QMGR
Configuration Options for SMAR
Configuration for MCP
Configuration for QMGR
Configuration for QMGR
Advanced Configuration for QMGR
Advanced Configuration for QMGR
Configuration Options for SMAR
Configuration for SMTP Server
Advanced Configuration for SMTP
Configuration for QMGR
Environment Variables used by
Environment Variables used by
Environment Variables used by
Environment Variables used by
Environment Variables used by
Common Global Configuration
Advanced Configuration for QMGR
Configuration for QMGR
Common Global Configuration
Common Global Configuration
Installing sendmail X
Installing sendmail X
Installing sendmail X
Installing sendmail X
Installing sendmail X
Installing sendmail X
Installing sendmail X
Installing sendmail X
Installing sendmail X
Configuration for SMTP Server
Configuration for SMTP Server
Configuration for MCP
Configuration for SMTP Server
Configuration Options for SMAR
Advanced Configuration for QMGR
Configuration for SMTP Server
Configuration for SMTP Server | Configuration for SMTP Client | Advanced Configuration for SMTP | Advanced Configuration for SMTP
Configuration for SMTP Server
Declaring Maps for SMAR
Configuration for MCP
Advanced Configuration for SMAR
Advanced Configuration for MCP
Advanced Configuration for SMAR
Configuration for MCP | Configuration for MCP
Configuration for QMGR
Configuration for QMGR | Configuration for SMTP Client
Advanced Configuration for SMTP
Configuration Options for SMAR
Configuration Options for SMAR
Advanced Configuration for MCP


... sendmail1.1
sendmail is a trademark of Sendmail, Inc.
... (MTS)1.2
often also called MTA: message transfer agent
... databases1.3
the term database is used loosely here
... to1.4
Sorry for the obfuscation, replace (at) with @ and remove the spaces, but not the plus sign.
... commands2.1
using script(1) or redirecting it to some file.
... configurations2.2
``Private database environments on 64-bit machines no longer drop core because of 64-bit address truncation. [11983]'' [Slea]
... blacklist3.1
This option is modelled after dnsblaccess written by Neil Rickert for sendmail 8.
... times3.2
Compile time option SM_MAX_DNSBL: currently set to 8.
... addresses3.3
4 is the default value for the compile time option SM_DNS_MAX_TSKS
... lmtp:4.1
currently internally encoded as
... errors4.2
See egrep(1) for the correct syntax for word delimiters on your OS.
... databases7.1
the term database is used loosely here, only DEFEDB is a real database, the others are just ways to store some information and access them in some way.
... storage7.2
If non-persistent storage is used for these databases mail can of course be lost.
... addresses8.1
Compile time option SS_MAX_BIND_ADDRS
... sink8.2
For example, statethreads/examples/smtps2

Claus Assmann 2005-10-28