MeTA1 README Claus Assmann Contents * Introduction to MeTA1 + MeTA1 is a Message Transfer Agent o Main Components of MeTA1 + Documentation o Typographical Conventions + Version + Current State o Providing Feedback + For the Impatient * Building, Testing, and Installing MeTA1 + Verifying the Source Code Distribution + Building MeTA1 o Compile-Time Configuration Options + Test Programs o Environment Variables used by Test Programs o Known Test Program Problems + Installing MeTA1 o Directories, Files, and Permissions o Upgrading from earlier MeTA1 Versions * Run-Time Configuration of MeTA1 + Overview + Configuration File Syntax o Configuration File Values + Example Configuration File + Common Global Configuration + Common Configuration Options + Pathnames for Files, Directories, and Maps + Configuration for MCP + Configuration for QMGR o Configuration Map for QMGR + Configuration for SMAR o Declaring Maps for SMAR o Configuration Options for SMAR o Configuration Maps for SMAR o Greylisting + Configuration for SMTP Server o SMTP Server Session Configuration o Multiple SMTP Servers with different Configurations o Protecting Recipients + Configuration for SMTP Client o SMTP Client Session/Recipient Configuration + Lookup Orders o Lookup Orders in Maps o Lookup Orders for Anti-Spam Measures o Macro Replacements in RHS + STARTTLS Restrictions + VERP * Running MeTA1 + Starting MeTA1 + Using MeTA1 only for Outgoing Mail + Using MeTA1 for Incoming Mail o Local Delivery and Specifying Local Domains o Specifying Valid Local Addresses + Using MeTA1 as Gateway + Using MeTA1 as Backup MX Server o Note about Backup MX Servers + Miscellaneous Programs o Do not run programs as root User o Displaying Content of Mail Queues o Interacting with QMGR + Reloading Maps + Logging o Logfile Rotation + Regular Checks + Dealing with Errors o Resource Problems o Database Problems o Writing Core Dumps + Replacements for Features available in other MTAs * Policy Milter + Policy Milter Overview + Native Policy Milter API o Data Structures o Start and Stop o New SMTP Server o SMTP Session and Transaction o Set and Get pmilter Contexts o Accessing MTA Symbols o Sender Modification o Recipient Modifications o Header Modifications o Message Replacement o Further Capabilities o Miscellaneous Functions o Return Values o Implementation Notes + Policy Milter Examples * Miscellaneous + Troubleshooting o Startup Problems o Logfile Entries + Caveats + Checks in SMTP Server o Strict RFC Compliance o Various Checks + Security Checks + Restrictions + Code Review, Enhancements, Patches + Porting + Version Naming o Snapshots * Data Flow in MeTA1 + Data Flow in MeTA1 * Advanced Configuration Options + Overview o Flags + Advanced Configuration for MCP + Advanced Configuration for QMGR + Advanced Configuration for SMAR + Advanced Configuration for SMTP Server + Advanced Configuration for SMTP Client * Tuning + Size of Queues, Caches, and Databases + Disk I/O + Processes and Threads * Format Specifications + Socket Map + Format of Session/Transaction Identifiers + Logfile Format + Format of Received Header + Format of DSNs * Setup for STARTTLS + Certificates for STARTTLS * More About Configuration, Compilation, Debugging, and Testing + Compile Time Options o Generic o QMGR o SMAR o SMTPS o Debugging Compile Time Options + Possible Compilation Problems or Warnings + More About Test Programs o More Environment Variables used by Test Programs o Other Potential Problems with Test Programs * Licenses * Bibliography * Index Introduction to MeTA1 MeTA1 is a Message Transfer Agent This distribution contains the source code for MeTA1 which implements a message transfer agent (MTA). 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]. MeTA1 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 MeTA1 MeTA1 is a modularized message transfer agent 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 MeTA1, the following is a brief summary. Figure 1.1 shows the interaction of the various components and databases^1.1. 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: MeTA1: Overall Structure \begin{picture}(120, 120)\ epsfxsize 120mm\leavevmode\ epsffile{overview1.eps}\end {picture} This version of MeTA1 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. Documentation The document ``sendmail X: Requirements, Architecture, and Functional Specification'' [Assmb] provides the background about the MeTA1 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. Version This document has been written for MeTA1-1.0.PreAlpha24.0, 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 MTA 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 to^1.2: < MeTA1 + feedback (at) MeTA1 . org > 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 MeTA1 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 MeTA1 PGP signing key [MeT ]: $ gpg -verify meta1-$VERSION.tar.gz.sig or: $ pgp meta1-$VERSION.tar.gz.sig meta1-$VERSION.tar.gz Further information, especially about warnings or possible errors, can be found in the documentation for PGP or GPG. Building MeTA1 MeTA1 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 && $PATHTO/meta1-$VERSION/configure $OPTIONS \ && make && make check Obviously you have to replace $OS, $VERSION, $OPTIONS, as well as $PATHTO. It is also possible to build MeTA1 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 commands^2.1for later inspection. Compile-Time Configuration Options Beside the usual configure options like --prefix a few MeTA1 specific configuration options are available: --enable-TLS 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-SASL 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 http://asg.web.cmu.edu/cyrus/ and http://asg.web.cmu.edu/ sasl/ 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. --with-sasl-libdir=path Path to directory containing Cyrus SASL v2 library. --with-sasl-incdir=path Path to directory containing Cyrus SASL v2 include files. --disable-included-bdb MeTA1 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. Notes: 1. If you do not use the Berkeley DB version that comes with MeTA1, 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 configurations^2.2. This bug is fixed in the version that is distributed with MeTA1. --with-bdb-libdir=path 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. Note: configure currently checks only for a static library. --with-bdb-incdir=path 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 $ $PATHTO/meta1-$VERSION/configure --with-bdb-libdir=$B/lib \ --with-bdb-incdir=$B/include --disable-included-bdb --enable-pmilter Enable policy milter protocol, see Chapter 5. --enable-msp Enable a simple mail submission program (MSP) that is currently not supported (located in contrib/). This is just a helper program for those who do not want to install a different MSP but need only some basic functionality (which does not include a queueing mechanism). Note: this will install the MSP as sendmail thus overriding any existing program of that name (as well as a man page). --enable-tinycdb Enable support for cdb map type, based on tinycdb 0.75 [Tok]. 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: * MTA_NO_DNS_TEST: disable tests that perform DNS lookups. These lookups may use domains that are under control of the MeTA1 author. * MTA_DNS_TIMEOUT: can be used to set a different timeout than the default, however, it may not be obeyed by all DNS test programs. * MTA_TEST_DNS_TIMING: run DNS tests that are timing dependent and may fail under certain conditions (e.g., network too slow). * MTA_TIMING: run MTA tests that are timing dependent and may fail under certain conditions. * MTA_STOPONERROR: causes most test scripts that perform multiple checks to stop on the first error that occurs instead of performing all checks. * MTA_NO_LOG_TEST: disables some tests that use syslog(3). * MTA_NO_SLOW_TEST: disables some tests that take a very long time. In this example the DNS timeout is set to 60 seconds and tests that take a very long time are disabled: $ MTA_DNS_TIMEOUT=60 $ MTA_NO_SLOW_TEST=1 $ export MTA_DNS_TIMEOUT MTA_NO_SLOW_TEST $ make -i check Known Test Program Problems * connctl.sh will fail on systems that have neither inet_pton(3) nor inet_aton(3). Fix: upgrade your OS or write a replacement function and put it into librepl/. * t-evthr-slp can fail in some circumstances if the OS is busy with other tasks as it depends on the OS scheduler. Fix: just rerun the test. * t-evthr-sig.sh fails on Linux systems that use a thread implementation that is not POSIX compliant. The test is currently disabled on all Linux versions. Note: if you know a simple way to figure out whether the OS actually provides POSIX compliant pthreads, please let me know. * t-hostname fails on systems where gethostname() does not return any FQHN at all (e.g., default SunOS 4/5 installations). Add the FQHN as alias to /etc/hosts (see hosts(5) to solve this problem, e.g., 10.1.2.3 myname myname.my.domain or 10.1.2.3 myname.my.domain myname * t-parsesockstr fails on systems like AIX which treat an empty string as a valid IP address in inet_addr(3). * t-mts-icr.sh and t-mts-ocr-?.sh try to test incoming/outgoing rate control. They rely on the time it takes to send/receive mails which may not work on machines that are significantly slower or faster than the machines available to the author. * t-smar-0.sh, t-smar-3.sh, and t-dns-1.sh may fail sometimes due to DNS timeouts. Run the tests again or increase the DNS timeout, see Section 2.3.1. Note: DNS related test programs may fail if the first nameserver entry in /etc/resolv.conf does not respond properly (and reasonably fast) to DNS queries. See Section 12.3.1 how to override the default nameserver selection: MTA_NAMESERVER. 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 MeTA1 MeTA1 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 (meta1) is not really required: meta1s:*:260:260:meta1 SMTPS:/nonexistent:/sbin/nologin meta1q:*:261:261:meta1 QMGR:/nonexistent:/sbin/nologin meta1c:*:262:262:meta1 SMTPC:/nonexistent:/sbin/nologin meta1m:*:263:263:meta1 misc:/nonexistent:/sbin/nologin meta1:*:264:264:meta1 other:/nonexistent:/sbin/nologin with the corresponding groups: meta1s:*:260: meta1q:*:261: meta1c:*:262:meta1s meta1m:*:263:meta1s,meta1q meta1:*:264: Note: on some operating systems the star character is not a valid value for the password field. Check passwd(5)^2.3to determine which value to use to disable the password. To check whether the required users and groups exist, run $ ./misc/sm.check.sh -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/sm.setup.sh.in. This script is modified by configure to create misc/sm.setup.sh (in the build directory) which is invoked when # make install is called. Most defaults in the installation script misc/sm.setup.sh can be overridden with environment variables (default is listed in square brackets): * MTACONFDIR: [/etc/meta1] configuration directory. * MTAQDIR: [/var/spool/meta1] queue directory; communication sockets are created in this directory by default too. * MTALOGDIR: [.] logging directory (relative to MTAQDIR). If logging is done via syslog(3) then this directory is not used. * MTAS [meta1s] SMTP Server user and group. * MTAC [meta1c] SMTP Client user and group. * MTAQ [meta1q] QMGR user and group. * MTAM [meta1m] address resolver (misc) user and group. * MTA [meta1] generic (configuration etc) user and group. * MTALG group for logfiles; the install program tries operator, sysadmin, and root. Important Notes: 1. The users and groups must be created before make install is invoked. 2. misc/sm.setup.sh 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/sm.setup.sh) will create all the required directories and files with the correct permissions provided the users and groups have been set up properly. This section explains what the created structure looks like. The CDB directories (0-9, A-F) must be owned by meta1s and have group meta1q with the permissions 0771: drwxrwx--x 2 meta1s meta1q 0/ Note: this means that everyone with access to the machine can guess the name of content files (see Section 10.2 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 meta1s with mode 0640 and group meta1c, e.g., -rw-r----- 1 meta1s meta1c 1993 Jul 9 21:19 2/S000000000006B1D200 The main (DEFEDB) and incoming queues (IBDB) must belong to meta1q and should not accessible by anyone else: drwx------ 2 meta1q meta1q defedb/ drwx------ 2 meta1q meta1q ibdb/ drwx------ 2 meta1q meta1q ibdb/ibdb/ Mailertable, aliases map, and other maps for SMAR (see Section 3.9.3) should belong to meta1m and can be readable as local conventions require: -rw-r--r-- 1 meta1m meta1m mt -rw-r--r-- 1 meta1m meta1m aliases.db In general, maps should be owned by the user id of the program that uses them, e.g., meta1q owns the QMGR configuration map qmgr_conf.db (see Section 3.8.1). The meta1 configuration file can either belong to root or the generic meta1 user: -rw-r--r-- 1 meta1 meta1 meta1.conf The directories in which the communication sockets between QMGR and the other programs are located must belong to meta1q and be group accessible for the corresponding program: drwxrws--- 2 meta1q meta1m qmsmar/ drwxrws--- 2 meta1q meta1c qmsmtpc/ drwxrws--- 2 meta1q meta1s qmsmtps/ The directory in which the communication socket between MCP and SMTPS is located must belong to meta1s: drwxr-x--- 2 meta1s meta1s smtps/ The logfiles must be owned by the corresponding user and may have relaxed group (or even world) read permissions: -rw-r----- 1 meta1q operator qmgr.log -rw-r----- 1 meta1m operator smar.log -rw-r----- 1 meta1c operator smtpc.log -rw-r----- 1 meta1s operator smtps.log To check whether an installation was successful, run # ./misc/sm.check.sh -P (in the build directory). Note: this script uses the same environment variables as the installation script. Upgrading from earlier MeTA1 Versions Currently there is no support for automated upgrades. If you have an earlier version of MeTA1 installed and want to upgrade, here are some tips (note: all programs should be run from the build directory unless mentioned otherwise): * To check whether an installation was successful, run # ./misc/sm.check.sh -P * To check whether the configuration file needs changes, run $ ./misc/smconf /etc/meta1/meta1.conf If the file is syntactically invalid for this version of MeTA1 the program will show those errors. Use -h as argument to see the available option, e.g., -u might be useful. Run-Time Configuration of MeTA1 Overview Configuration of MeTA1 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 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 MeTA1 configuration file is very simple: conf ::= entries entries ::= entry * entry ::= option | section section ::= keyword [name ] "{" entries "}" [";"] option ::= option-name "=" rhs rhs ::= value ";" | "{" 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^3.1. 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 meta1.conf in the configuration directory (/etc/meta1, see Section 2.4). Check the comments in the file and edit it if required. A configuration file for meta1 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/meta1/"; qmgr { AQ_max_entries = 8192; smtpc { initial_connections = 19; max_connections = 101; } smtps { max_connections = 5; max_connection_rate=160; } max_errors_per_DSN=16; wait_for_server = 4; wait_for_client = 4; start_action = wait; user = meta1q; restart_dependencies = { smtps, smtpc, smar }; path = "/usr/libexec/qmgr"; arguments = "qmgr -f /etc/meta1/meta1.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 = meta1s; path = /usr/libexec/smtps; arguments = "smtps -f /etc/meta1/meta1.conf"; } smtpc { Log_Level = 12; IO_timeout = 6m; wait_for_server = 4; start_action = wait; user = meta1c; path = "/usr/libexec/smtpc"; arguments = "smtpc -f /etc/meta1/meta1.conf"; } smar { Log_Level = 12; nameserver = {10.10.10.9, 127.0.0.1}; DNS_timeout = 6; start_action = wait; user = meta1m; restart_dependencies = { smtps, qmgr }; path = "/usr/libexec/smar"; arguments = "smar -f /etc/meta1/meta1.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., MeTA1 or MeTA1QMGR. 3. options: options for openlog(3) (without the leading LOG_) as provided by the OS, e.g., pid or ndelay. Example: qmgr { log { facility = daemon; ident=meta1-qmgr; } } smtps { log { facility = mail; ident=meta1-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/meta1/meta1.conf then the pathname used for the aliases map is /etc/meta1/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 MeTA1 modules should be started (via MCP) in the main queue directory (default: /var/spool/meta1, 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 MeTA1 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 MeTA1 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 meta1.conf"; } will cause MCP to start four smtpc processes, each with the options -i $ID$ -f meta1.conf where $ID$ is replaced with a unique identifier. Notes about start_action: * For start_action = pass the option pass_fd_socket must be specified; in this case MCP binds to the specified socket (listen_socket) and passes it via the Unix domain socket (pass_fd_socket) to the started process. * For start_action = nostartaccept MCP waits for incoming connections, and then starts a process to handle a single connection. * For start_action = accept MCP binds to the socket and then starts a process to handle the connections without waiting for an actual request. * For start_action = wait MCP simply starts the requested number of processes without passing them any open connections. This is intended for processes that do not communicate with external clients. MCP is currently a generic control program that does not have any builtin knowledge about the various MeTA1 modules. Hence the MCP options for each MeTA1 component must be specified properly, there are no builtin defaults that could be associated with the functionality of the various MeTA1 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. double_bounce_address: RFC 2821 address for double bounces; defaults to . 7. 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. 8. 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. 9. log_level: logging level. 10. 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. 11. 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. 12. ok_disk_space: amount of free disk space at which normal operation continues (unit: KB). Must be larger than min_disk_space. 13. queue_return_timeout: maximum time in queue (unit: s). 14. 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. 15. retry_max_delay: maximum time for retrying a delivery (unit: s). 16. retry_min_delay: minimum time for retrying a delivery (unit: s). 17. 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). 5. flags: configuration flags: 1. lookup_rcpt_conf: Look up recipient configuration data (see Section 3.11.1) in the access map (see Section 3.9.3) 2. lookup_session_conf: Look up session configuration data (see Section 3.11.1) in the configuration map (see item 2). 6. rcpt_conf_lookup_flags: If recipient configuration data (see item 17(e)i) is looked up in the access map, then these flags determine which kind of lookups should be performed. 1. full_adress: use the full address as key. 2. detail_plus: lookup also ``user++@subdomain''. 3. detail_star: lookup also ``user+*@subdomain''. 4. star: lookup also ``user*@subdomain''. 5. domain: lookup domain part. 6. dotsubdomain: iterate through subdomains. 7. dot: lookup also ``.''. The default is to perform all lookups. 18. 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). 19. wait_for_client: maximum amount of time to wait for a client to become available (unit: s) 20. 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 meta1q) 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. 6. smtpc_session_conf: see Section 3.11.1 oci:, ocm:, icr:, icm:, and smtpc_session_conf: take an IP address/net as parameter such that the limits can be imposed per IP address/net. For example: oci:127.0.0.1 5 ocm:127.0.0.1 10 oci:10 10 ocm:10 50 oci: 1 ocm: 4 icr:10 5 icr:127.0.0.1 100 icm:127.0.0.1 120 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), cdb (tinycdb), sequence, socket, and passwd. 2. file: the filename of the db file (including the extension) (for type hash, cdb). 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. Example: map localusers { type = hash; file = "/etc/meta1/localusr.db"; } map otherusers { type = cdb; file = "/etc/meta1/otherusr.cdb"; } 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. See Section 3.9.3 for details. Note: only one of file (1a) and name (1b) must be specified. 1. file: filename of access map (including extension) [default: access.db]. 2. name: name of access map . This can be used if a different map type should be used, in which case the map must be declared as explained in Section 3.9.1. 2. address_delimiter: list of delimiters (specified as string) for address extensions in local part, [default: "+"]. Note: if address_delimiter has more than one character, the first one that is found in the local part of an address is used as delimiter in map lookups (see Section 3.12). For example: if the following option is used in the configuration file: address_delimiter = "/_-"; then for the address ``'', the delimiter for map lookups is ``/'' and the address detail is ``ext-list'', while for the address ``'', the delimiter for map lookups is ``-'' and the address detail is ``ext_list''. 3. aliases: this is a subsection that specifies the parameters for aliases. Note: only one of file (3a) and name (3b) must be specified. 1. file: filename of aliases map (including extension) [default: aliases.db]. 2. name: name of aliases map . This can be used if a different map type should be used, in which case the map must be declared as explained in Section 3.9.1. 3. 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 looked up for all domains. 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. DNS: this subsection contains DNS related options. 1. nameservers: list of up to four IPv4 addresses^3.2of nameservers. Note: it is important that all of these nameserves work properly. Currently they are used ``round robin'' without excluding nameservers that do not answer^3.3. 2. retries: maximum number of retries. A value of 0 means one query only, i.e., no retry. 3. timeout: the default timeout for a single DNS query (unit: s). Notes: o the timeout for a DNS request is the product of the number of tries and the individual timeout, i.e., (retries + 1) * timeout. o this value is only the default timeout which can be overridden by an application. For example, QMGR dynamically increases the timeout for addresses which did not resolve in earlier tries. 4. flags: The flag use_resolvconf causes the list of nameservers (see 4a) to be read from /etc/resolv.conf. This flag is set by default unless the nameservers option is used. Note: the list of nameservers is not updated when /etc/resolv.conf is changed, smar needs to be restarted to achieve that. More flags can be found in Section 8.4. 5. dnsbl: specify a DNS based blacklist^3.4. This section can be specified multiple times^3.5; it has the following required options: + domain: specify the domain to use for DNS lookups, e.g., dnsbl.tld. + tag: specify the tag to use for lookups in the access map (which must be enabled, see Section 3.10, 3b). 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. Notes: + DNS lookups in blacklists can be disabled via entries in the access map using the tag cltaddr, see Section 3.9.3. + Some DNS blacklists return multiple A records. For those the A records are checked (in the order returned by the DNS server) until an access map entry is found. This is an intermediate solution as it may cause random results if multiple access map entries for a DNS blacklist exist. However, as long as all of them have the same return code type (i.e., temporary or permanent), this behavior is sufficient. + currently a colon is added as delimiter after tag, this may be removed in later versions to allow for more flexibility; e.g., the configuration option itself can include a delimiter. 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:127.0.0.1 error:550 5.7.1 listed at dnsbl.tld as open relay dnsbltld:127.0.0.2 error:550 5.7.1 listed at dnsbl.tld as spam source dnsbltld:127.0.0.9 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. 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: i. 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). 3. flags: these flags can be used to select a subset of the matching described in Section 3.12.1. 1. full_adress: use the full address as key. 2. detail_plus: lookup also ``user++@subdomain''. 3. detail_star: lookup also ``user+*@subdomain''. 4. star: lookup also ``user*@subdomain''. 5. domain: lookup domain part. 6. dotsubdomain: iterate through subdomains. 7. dot: lookup also ``.''. The default is domain, dotsubdomain, dot. 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) 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 3b) (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 3.10.3) smtps_session_conf: configuration options for a session in the SMTP server (see Section 3.10.1) smtpc_rcpt_conf: configuration options for recipient in the SMTP client (see Section 3.11.1) 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. Note: the host name has a trailing dot after DNS resolution, this dot must be included in the entry. 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 return an error consisting of SMTP reply code XYZ, text 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 3c - 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. Examples: cltresolve:tempptr error:451 4.7.1 reverse lookup failed mxbadip:127.0.0.1 error:551 5.7.1 Bad IP address 127.0.0.1 in MX/A list mxbadip:192.168.255.255 error:551 5.7.1 Bad IP address 192.168.255.255 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:127.0.0.1 quick:relay Discard 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. Mailertable 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 :: [ local "@" ] [ "." ] hostname | "." = RHS :: [[ port "^" ] ["esmtp:"]] hosts | "lmtp:" | port " = ^"lmtp:" hosts port :: integer = hosts :: hostname | iplist = iplist :: "[" IPv4-address "]" [ " " iplist ] = The key (LHS) is an address (without angle brackets), a hostname, or a dot (denoting the default entry), the value (RHS) consists of an optional port number, an optional (esmtp) mailer and a hostname or a list of IPv4 addreses (in square brackets) 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 hostname is subject to MX lookups. Example: localhost lmtp: SPAM.FILTER.DOMAIN 2525^esmtp:[127.0.0.1] LMTPHOST.MY.DOMAIN 525^lmtp:[10.11.12.13] MY.DOMAIN esmtp:[10.1.2.3] ANOTHER.DOMAIN esmtp:MTA.SERVER .TLD esmtp:GATE.WAY . esmtp:SMART.HOST Note: currently this file must exist, even if there are no entries (it is created during installation). Aliases To specify aliases for local addresses the map aliases.db (Section 3.9.2, item 3a) is used. The key in the map must be * the local part of a valid (local) e-mail address, * or a complete local e-mail address, * or any e-mail address, based on the flags of the aliases option (see Section 3.9.2, 3c). 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 start with an angle bracket, 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: 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(c)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 and would be rejected while mail to or would be accepted. Aliases can be nested (currently up to 5 levels, see smar/rcpts.c). Greylisting MeTA1 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 MeTA1 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 3d) is useful to avoid unnecessary delays. Here is an example: host A is the main MX server for domain example.com and it uses greylisting, host B is the main MX server for domain example.net 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:127.0.0.1 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:12.107.209.244 ok cltaddr:64.12.137 ok A list of such broken mailers can be found at http://cvs.puremagic.com/ viewcvs/greylisting/schema/whitelist_ip.txt [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_mechanisms: 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 meta1.conf. That file can be used to adjust the list of mechanisms (option mech_list) 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 files, i.e., the group id of meta1c, see Section 2.4.1. 3. flags: 1. 8bitmime: offer 8BITMIME: MeTA1 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. 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. 3. delay_checks: delay acceptance check until RCPT stage (unless explicitly overridden, see Section 3.9.3). 4. delay_greylisting_error_until_DATA: if greylisting (3e) is enabled then wait until the DATA command to return an error; see Section 3.9.4 for details. 5. greylisting: enable greylisting (which must also be enabled in SMAR, see Section 3.9.2 item 6), see Section 3.9.4 for details. 6. 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. 7. 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. 8. require_EHLO_before_MAIL: require EHLO (or HELO) before a MAIL command. 9. strict_EHLO_checks: perform a strict syntax check on the argument for EHLO (or HELO). 10. xverp: offer XVERP extension to turn on VERP [Ber97] support for a transaction. This is a parameter for the MAIL command, e.g. MAIL From: XVERP 11. allow_data_before_greeting: allow a client to send data before the initial 220 greeting. 12. prdr: PRDR is an SMTP extension [Hal07] to return individual RCPT status after the end of data (similar to LMTP). This feature is turned on if a policy milter is configured (see item 17). 13. lookup_session_conf: Look up session configuration data (see Section 3.10.1) in the access map (which must be activated, see 3b ). 4. id: unique identifier for SMTP server (0); see Section 3.10.2. 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 A. 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 3.10.3 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. cache_size: size of TLS session session cache (0: disable cache). 2. cache_timeout: timeout for entries in TLS session session cache. 3. cert_file: file with certificate in PEM format. 4. key_file: file with private key for certificate in PEM format. 5. CAcert_file: file with CA certificate in PEM format. 6. CAcert_directory: directory with (symbolic links for) CA certificates in PEM format. 7. 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 3b) 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: /C=US/ST=California/O=endmail.org/OU=private/CN= Darth Mail (Cert)/emailAddress=darth+cert@endmail.org is encoded as: /C=US/ST=California/O=endmail.org/OU=private/CN= Darth+20Mail+20+28Cert+29/emailAddress= darth+2Bcert@endmail.org Examples: To allow relaying for everyone who can present a cert signed by /C=US/ST=California/O=endmail.org/OU=private/CN= Darth+20Mail+20+28Cert+29/emailAddress= darth+2Bcert@endmail.org simply use: certissuer:/C=US/ST=California/O=endmail.org/OU=private /CN= Darth+20Mail+20+28Cert+29/emailAddress= relay darth+2Bcert@endmail.org To allow relaying only for a subset of machines that have a cert signed by /C=US/ST=California/O=endmail.org/OU=private/CN= Darth+20Mail+20+28Cert+29/emailAddress= darth+2Bcert@endmail.org use: certissuer:/C=US/ST=California/O=endmail.org/OU=private /CN= Darth+20Mail+20+28Cert+29/emailAddress= cont darth+2Bcert@endmail.org CertSubject:/C=US/ST=California/O=endmail.org/OU= private/CN= DeathStar/emailAddress=deathstar@endmail.org relay Notes: # 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=. 8. requirements_violation: This option decides how to (globally) handle TLS requirement violations (see Section 3.10.1). Possible values are: permfail: generate permanent failure, tempfail: generate temporary failure, abort: abort session. These option take effect when a MAIL command is issued by the client. 22. session_features: This section can be used to define classes of features for an SMTP server session which then can be referenced via the access map. Available flags are a subset of those listed in item 3 : starttls, auth, delay_checks, allow_data_before_greeting, require_EHLO_before_MAIL, strict_EHLO_checks, check_EHLO. Example: to turn off STARTTLS for some clients, declare a notls class in the configuration file: session_features notls { flags = { -starttls }} and reference it in the access map for those clients: smtps_session_conf:217.126.135.148 session_feature=notls; SMTP Server Session Configuration Some options can be set via the access map (see Section 3.9.3, tag smtps_session_conf) because they apply to a session, not globally. Currently available are: STARTTLS requirements (tls_requirements) (see Section 3.13) and session_features (see Section 3.10, item 22). 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 meta1.conf must have a unique name (e.g., MTA and MSA), which is passed via the option -N name to smtps. Example: meta1.conf: smtps MTA { id = 0; listen_socket { type=inet; port = 25; } start_action = pass; pass_fd_socket = smtps/mtafd; user = meta1s; path = /usr/libexec/smtps; arguments = "smtps -N MTA -f /etc/meta1/meta1.conf"; log { facility = mail; ident=meta1-MTA; } } smtps MSA { id = 1; listen_socket { type=inet; port = 587; } start_action = pass; pass_fd_socket = smtps/msafd; user = meta1s; path = /usr/libexec/smtps; arguments = "smtps -N MSA -f /etc/meta1/meta1.conf"; log { facility = mail; ident=meta1-MSA; } auth { trusted_mechanisms = "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 sender and client_ip must be selected (see Section 3.10, item 19a). If this is done, then every recipient is looked up in the access map (which must be activated, see Section 3.10, item 3b), 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 allow_by flag | |----------------------+------------------------| | from:sender | sender | |----------------------+------------------------| | cltaddr:IPv4-address | client_ip | |----------------------+------------------------| | list:alias | 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 (see Section 3.10, item 19b), 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: list2: list3: together with this access map: protectedrcpt:list1@local.dom list: protectedrcpt:list3 from: cltaddr:1.2.3.4 cltaddr:10 The mails to are only accepted from and . 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 are only accepted from , the client with the IPv4 address 1.2.3.4, 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). SMTP Client Session/Recipient Configuration The following options can only be set via the access map (see Section 3.9.3, tag smtpc_rcpt_conf) or the configuration map for QMGR (see 3.8.1, tag smtpc_session_conf) not in the SMTP client configuration section itself, as they apply to a session or recipient, not globally. Currently only STARTTLS requirements are available which are documented in Section 3.13, more option might be added later on. Note: as more options might be added, the structure of this may change a bit. 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. These steps can be controlled by flags that are specified in the configuration file for the map. These flags are listed below for the various lookup steps. For domain names of the form ``sub2.sub1.tld'' the lookup order is ``sub2.sub1.tld'', ``.sub1.tld'', ``.tld'', and ``.'' (without the quotes). The subdomains are tried if the flag dotsubdomain is set. The last lookup (``.'') is only done if the flag dot is set, as it is the default for 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 ``'', where ``+detail'' is optional and ``+'' is the first delimiter in the localpart that was found in the address_delimiter string (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. Flag: full_adress. 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. Flag: detail_plus 3. ``user+*@subdomain'' if ``+detail'' exists; this matches any ``+detail'' (including just ``+''). Flag: detail_star 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''. Flag: star 5. ``user@subdomain''; this does not match if ``+detail'' exists, unless the flag 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. Connect During connect the following operations are performed if the access flag (see Section 3.10, item 3b) is enabled: 1. check: look up 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: look up cltresolve:clientresolve 5. check: look up cltname:clientname (as explained in 3.12.1) 6. check: look up results of DNS blacklists in access map. MAIL 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: look up the address in the access map (provided it is enabled) with the tag from:. 4. check: look up the IP addresses that were found when trying to determine whether the address is routeable with the tag mxbadip: in the access map. RCPT A RCPT command causes different checks (note: these are sequential checks, not exclusive, i.e., if the first one applies and it does not return a decision, the second one is performed): 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: look up the recipient address with the tag to: (provided the access map is enabled) (provided the access map is enabled) and check whether the RHS is relay, otherwise reject the RCPT command as unauthorized relaying attempt unless the RHS is an error entry which is then used as reply. For case 2 the following steps are taken: 1. look up address with tag protectedrcpt:, if found perform all the necessary checks as explained in Section 3.10.3. 2. check: look up 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 or the RCPT lookup returned quick:ok or quick: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 ``${name}'' (without the quotes). Available macros are: user, detail, domain, tag, delimiter, subdomain, extension. They have the obvious meaning; subdomain refers to the part of the domain before the dot, i.e., if the pattern is @.domain and the input is user@host.domain then subdomain refers to host, extension is the delimiter and the detail together (provided the address contains them). Example: alias*@.domain user${extension}@${subdomain}.domain provides the following mappings: alias@host.domain user@host.domain alias+detail@host2.domain user+detail@host2.domain STARTTLS Restrictions STARTTLS requirements can be set for the SMTP server and SMTP client via map entries (see Sections 3.10.1 and 3.11.1). These restrictions are listed in a section called tls_requirements. 1. common_name: require that the CN (Common Name) of the presented TLS certificate matches the specified value. 2. cert_subject require that the CS (CERT subject) of the presented TLS certificate matches the specified value. 3. cert_issuer require that the DN (Distinguised Name) of the issuer of the presented TLS certificate matches the specified value. 4. min_cipher_bits require that the effective keylength (in bits) of the symmetric encryption algorithm used for a TLS connection is at least as big as the specified value. 5. flags: available flags are currently: 1. verified: the TLS connection must be verified, i.e., the cert issuer must be listed in CAcert_file or CAcert_directory (see Section 3.10, item 21). 2. encrypted the TLS connection must be encrypted (i.e., the effective keylength (in bits) of the symmetric encryption algorithm used for a TLS connection is greater than zero. 6. requirements_violation See Section 3.10, item 21h, for possible values. Example (note: the entry must be currently written in a single line to be properly processed by createmap(8)): smtpc_rcpt_conf:@meta.org tls_requirements { cert_issuer="/C=US/ST=Berkeley/L=Endmail+20Org/O=MTA/CN=Claus+20Assmann+20CA+20RSA+202007/emailAddress=ca+ca-rsa2007@esmtp.org"; } VERP There are two ways to enable VERP [Ber97]: 1. Via the XVERP extension in the SMTP server (see Section 3.10, item 3j ). 2. Via an entry in aliases marking a list with verp-: otherlist: verp-otherlist: something Running MeTA1 Starting MeTA1 All components of MeTA1 are under control of the MCP which must be started as root in the directory /var/spool/meta1 (i.e., the main queue directory, see Section 2.4: MTAQDIR) using # ./mcp.sh start The script contains the runtime path for MCP based on the data used by configure as well as a reference to the MeTA1 configuration file. To stop all MeTA1 components use # ./mcp.sh 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 MeTA1 only for Outgoing Mail MeTA1 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 submit.mc file (see also misc/ sm8.submit.mc): LOCAL_RULE_0 R$* + X<@$*> $#meta1 $@ localhost $: $1 <@$2> LOCAL_RULESETS SHdrToSMTP R$+ $: $>PseudoToReal $1 sender/recipient common R$+ $: $>MasqSMTP $1 qualify unqual'ed names R$* + X<@$*> $: $1 < @ $2 > R$* < @ *LOCAL* > $* $: $1 < @ $j . > $2 MAILER_DEFINITIONS Mmeta1, 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 MeTA1 as listener on localhost:2009. Then mail to will be sent via MeTA1, i.e., by adding the sequence +X to the address the mail will be redirected to MeTA1 (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 MeTA1 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 meta1.conf. See contrib/procmail.lmtp.p0 for a patch^4.2for 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: HOST.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 3f) 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, 3b ), 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 map aliases.db (Section 3.9.2, item 3a), 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: abuse: user+abuse user++: local: user: local: Note: local addresses are checked for the envelope recipient and sender. Using MeTA1 as Gateway MeTA1 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 MeTA1 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:127.0.0.1 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 MeTA1 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 meta1q 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. A simple shell script wrapper mailq.sh is available that invokes mailq via runas. 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 meta1q 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 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 10.3 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/sm.setup.sh, which parses meta1.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 errors^4.3: $ egrep 'sev=(ALERT|CRIT|ERR|FAIL)|\' $LOGFILE 2. Keep track of the size of the processes, e.g, $ date >> $MTAPROCS $ ps axuww | grep '^meta1' | sort >> $MTAPROCS If one of the processes continuously grows then MeTA1 should be compiled with -DSM_HEAP_CHECK (see Section 12.1.5) 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 MTA 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 MeTA1: * memory shortage: the memory usage of MeTA1 can be controlled by restricting the size of various data structures, see Section 3.8, e.g., AQ_max_entries, IQDB, and OCC_max_entries. However, setting these values too low will result in a very slow MTA that may operate in a degraded state which is not acceptable. * disk shortage: MeTA1 has options that let it decide how much free disk space is needed for operation, see Section 3.8: ok_disk_space and min_disk_space. However, if there is not enough space to store the envelope databases (DEFEDB and IBDB) then the system cannot work, hence sufficiently free disk space is essential for proper operation. 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 10.3) 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/rcvrenvfromlog.sh 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 MeTA1 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 MeTA1 does currently 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 MeTA1 has support for a policy milter which is similar to a milter in sendmail 8. The API is slightly different, however, it should be possible to write an compatibility layer to emulate the sendmail 8 API. Another difference is that MeTA1 itself only talks to a single pmilter. Support for multiple pmilters should be implemented in a multiplexor that connects to multiple pmilters and coordinates their responses. 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. There can be multiple SMTP servers connecting to one libpmilter 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. * New SMTP session: sfsistat_T pmfi_connect(pmse_ctx_P pmse_ctx, const char *hostname, sm_sock_addr_T *hostaddr) hostname: host name, as determined by a reverse lookup on the host IP address; hostaddr: host address, as determined by a getpeername(2) call on the SMTP socket. * SMTP HELO/EHLO command: sfsistat_T pmfi_helo(pmse_ctx_P pmse_ctx, const char *helohost, bool ehlo) helohost: Value passed to HELO/EHLO command, which should be the domain name of the sending host. ehlo: true iff EHLO was used. * MAIL (envelope sender): sfsistat_T pmfi_mail(pmse_ctx_P pmse_ctx, const char *mail, char **argv) mail: envelope mail address; argv: null-terminated MAIL command arguments. * RCPT (envelope recipient): sfsistat_T pmfi_rcpt(pmse_ctx_P pmse_ctx, const char *rcpt, char **argv) rcpt: envelope recipient address; argv: null-terminated RCPT command arguments. * DATA: sfsistat_T pmfi_data(pmse_ctx_P pmse_ctx) * unknown/not implemented SMTP command: sfsistat_T pmfi_unknown(pmse_ctx_P pmse_ctx, const char *cmd) cmd: SMTP command. Note: this is not yet implemented. * For each chunk of a message: sfsistat_T pmfi_msg(pmse_ctx_P pmse_ctx, unsigned char *msgp, size_t msglen) msgp: pointer to message data; msglen: length of message data. There may be multiple message chunks passed to the filter. End-of-lines are represented as received from SMTP (normally Carriage-Return/Line-Feed; CRLF). Notes: + the last message chunk contains the final dot of the SMTP transmission, i.e., ``CRLF.CRLF'' + the message is not modified in any form, i.e., dots at the begin of a line are duplicated (by the SMTP client) as specified in section 4.5.2 of RFC 2821 [Kle01] which must be undone by the application if so desired. + the message is streamed while being received. That is, the mail is not first stored on disk and then sent to the filter, but each part received from the client is sent directly to the filter (at the same speed as received from the network which might be slow). This may mean that the filter does not receive the entire message as the transmission may get interrupted or the SMTP server may decide to skip the rest of the message because it exceeds the maximum size. + the return code is ignored unless SM_SCAP_PM_MSG_RC is set, see Section 5.2.11, item 2. * End of message (final dot of message has been received): sfsistat_T pmfi_eom(pmse_ctx_P pmse_ctx) * Message is aborted outside of the control of the filter, for example, if the SMTP client issues an RSET command. sm_ret_T pmfi_abort(pmse_ctx_P pmse_ctx) * QUIT (end of an SMTP session): sm_ret_T pmfi_close(pmse_ctx_P pmse_ctx) This is called when an SMTP session ends. 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_EHLO EHLO or HELO command PM_SMST_MAIL MAIL command PM_SMST_RCPT RCPT command PM_SMST_DATA DATA command 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: 1. PMM_SRVHOSTNAME hostname of SMTP server. 2. PMM_SEID session id. 3. PMM_CLIENT_RESOLVE result of client lookups. 4. PMM_MAIL_TAID transaction id. 5. PMM_DOT_MSGID Message-Id. 6. PMM_TLS_VERSION TLS/SSL version used. 7. PMM_TLS_CIPHER_SUITE cipher suite used. 8. PMM_TLS_CIPHER_BITS effective key length of the symmetric encryption algorithm. 9. PMM_TLS_ALG_BITS maximum key length of the symmetric encryption algorithm. This may be less than the effective key length for export controlled algorithms. 10. PMM_TLS_VRFY the result of the verification of the presented cert. 11. PMM_TLS_CERT_SUBJECT the DN (distinguished name) of the presented certificate. 12. PMM_TLS_CERT_ISSUER the DN (distinguished name) of the CA (certifcate authority) that signed the presented certificate (the cert issuer). 13. PMM_TLS_CN_SUBJECT the CN (common name) of the presented certificate. 14. PMM_TLS_CN_ISSUER the CN (common name) of the CA that signed the presented certificate. Notes: * PMM_MAIL_TAID cannot be requested before PM_SMST_MAIL and PMM_DOT_MSGID can only be requested at stage PM_SMST_DOT. * All macros beginning with PMM_TLS are only valid after a STARTTLS command. 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. Sender Modification The sender address (MAIL) can be replaced: sm_ret_T sm_pmfi_mail_rplc(pmse_ctx_P pmse_ctx, const char *mail_pa, char **argv) This function must only be called during pmfi_eom(). The address mail_pa must be in RFC 2821 format. The argument argv can be used to specify SMTP parameters for the sender address, however, this is currently not implemented, hence it must be set to NULL for now. Recipient Modifications Recipients can be added: sm_ret_T sm_pmfi_rcpt_add(pmse_ctx_P pmse_ctx, const char *rcpt_pa, char **argv) or deleted: sm_ret_T sm_pmfi_rcpt_del(pmse_ctx_P pmse_ctx, const char *rcpt_pa, rcpt_idx_T rcpt_idx) These functions must only be called during pmfi_eom(). The addresses rcpt_pa must be in RFC 2821 format. The argument argv can be used to specify SMTP parameters for the recipient address, however, this is currently not implemented, hence it must be set to NULL for now. As the MTA does not remove identical recipient addresses, the address itself is not sufficient to identify one RCPT, but its index must be specified too. This index can be retrieved during a RCPT command (pmfi_rcpt()) using sm_ret_T sm_pmfi_getrcpt_idx(pmse_ctx_P pmse_ctx, rcpt_idx_T *prcpt_idx) Note: it is invalid to remove all recipients of a transaction. To discard a transaction, return SMTP_R_DISCARD as a result of one of the transaction oriented callbacks, e.g., pmfi_eom(). Header Modifications To request modifications of the header of a mail being sent, the function sm_pmfi_hdr_mod(pmse_ctx_P pmse_ctx, uint type, uint pos, const unsigned char *header) can be used. This function must only be called during pmfi_eom(). The argument type specifies which kind of modification is requested, legitimate values are defined in include/sm/hdrmoddef.h; these are: SM_HDRMOD_T_PREPEND, SM_HDRMOD_T_INSERT, SM_HDRMOD_T_REPLACE, SM_HDRMOD_T_REMOVE, and SM_HDRMOD_T_APPEND. The argument header must be a complete header line including the proper line ending (CRLF). The argument pos specifies the position for the types SM_HDRMOD_T_INSERT, SM_HDRMOD_T_REPLACE, and SM_HDRMOD_T_REMOVE. The first header of the original mail has position one; zero is the Received: header added by the SMTP server. Message Replacement If a pmilter wants to replace the entire message, the function pmfi_eom() must return the value SMTP_R_RPLCMSG. This will cause the invocation of the callback sfsistat_T pmfi_msg_rplc(pmse_ctx_P pmse_ctx, const unsigned char **pmsgchunk, size_t *pmsglen) which then must set a pointer to a message chunk and its length. Multiple chunks can be sent by returning SMTP_R_CONT as result of the callback. For the last chunk, SMTP_R_OK should be returned. The size of each chunk (pmsglen) must be less than PMILTER_CHUNK_SIZE as defined in include/sm/ pmfapi.h. libpmilter will thereafter invoke the callback sfsistat_T pmfi_msg_rplc_stat(pmse_ctx_P pmse_ctx, sm_ret_T status) to give the pmilter a chance to clean up after the transaction, and to let it know whether the message replacement was successful. Notes: * as the entire message is replaced and by default only the data that is received from the SMTP client is sent to a pmilter, the Received: header field that smtps generates is lost. To avoid this, a pmilter can request to receive that header field by setting SM_SCAP_PM_SND_RCVD and sending it as first chunk of the message replacement. * the message must be in SMTP format, i.e., lines must end in CRLF and the final chunk should have the usual SMTP end of message indication: CRLF dot CRLF (\r\n.\r\n), however, it can also just end in CRLF. * if pmfi_msg_rplc() encounters an error, e.g., due to an API violation or due to a communication error with smtps, then it will invoke pmfi_msg_rplc_stat() directly without waiting for the entire message even if it consists of more chunks. 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. 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. 2. SM_SCAP_PM_MSG_RC allows a pmilter to return a reply code as specified in 5.2.13 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). 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). The text can be a multi-line reply in the form: XYZ-D.S.N text1 XYZ-D.S.N text2 XYZ D.S.N text3 which must be specified in the format required by SMTP: XYZ-D.S.N text1\r\nXYZ-D.S.N text2\r\nXYZ D.S.N text3\r\n To set reply codes for commands that need multiple reply values the function: sm_ret_T sm_pmfi_setreplies(pmse_ctx_P pmse_ctx, uint nreplies, int *rcodes, const char **rtexts) must be used. Currently this function makes only sense if PRDR is available in the SMTP server and actually used by the client. A pmilter can determine the latter by parsing the arguments of the MAIL command (see pmfi_mail()). Note: currently the argument rtexts is ignored, only the array of reply codes (rcodes) is used. The size of this array must be nreplies which must match the number of valid RCPTs for this transaction. The reply codes in that array must be in the same order in which the RCPTs have been received. 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 callback: sm_ret_T pmfi_signal(pmg_ctx_P pmg_ctx, int sig) This function 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): * SMTP_R_OK: accept command. * SMTP_R_ACCEPT: accept entire transaction or session depending on where this value is returned. Note: this is just a shortcut for SMTP_R_SET_QUICK(SMTP_R_OK). * SMTP_R_DISCARD: discard effect of command. * SMTP_R_CONT: continue other checks. * SMTP_R_SSD: shut down SMTP session. * SMTP_R_TEMP: reject command with a temporary error. * SMTP_R_SYNTAX: syntax error. * SMTP_R_PERM: reject command with a permanent error. * other valid SMTP reply codes [Kle01]. 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 MeTA1. Miscellaneous Troubleshooting 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. Startup Problems If MeTA1 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/sm.check.sh 3. missing or invalid maps: make sure maps are created properly with createmap(8). Logfile Entries Most logfile entries should be self-explaining. However, some are more subtle and indicate only indirectly what might be wrong. Example: smtps: ... client_name=Hostname_Not_Determined indicates that the access map (Section 3.10, 3b) is