%$Id: milter-protocol.txt,v 1.6 2004/08/04 16:27:50 tvierling Exp $
%
%
##THE SENDMAIL MILTER PROTOCOL, VERSION 2
The Sendmail and "libmilter" implementations of the protocol described
herein are:
> Copyright (c) 1999-2002 [Sendmail, Inc.](http://www.sendmail.com) and its suppliers.
> All rights reserved.
This document is:
> Copyright (c) 2002-2003, Todd Vierling
> All rights reserved.
Permission is granted to copy or reproduce this document in its entirety
in any medium without charge, provided that the copy or reproduction is
without modification and includes the above copyright notice(s).
###OVERVIEW
The date of this document is contained within the "Id" symbolic CVS/RCS
tag present at the top of this document.
This document describes the Sendmail "milter" mail filtering and
MTA-level mail manipulation protocol, version 2, based on the publicly
available C-language source code to Sendmail, version 8.11.6.
As of this writing, this protocol document is based on the
implementation of milter in Sendmail 8.11, but has been verified
compatible with Sendmail 8.12. Some Sendmail 8.12 extensions,
determined by flags sent with the SMFIC_OPTNEG command, are not yet
described here.
Technical terms describing mail transport are used throughout. A reader
should have ample understanding of RFCs 821, 822, 2821, and their
successors, and (for Sendmail MTAs) a cursory understanding of Sendmail
configuration procedures.
###LEGEND
All integers are assumed to be in network (big-endian) byte order.
Data items are aligned to a byte boundary, and are not forced to any
larger alignment.
This document makes use of a mnemonic representation of data structures
as transmitted over a communications endpoint to and from a milter
program. A structure may be represented like the following:
'W' |SMFIC_HWORLD |Hello world packet
--------|---------------|------------------
uint16 len ||Length of string
char str[len] ||Text value
This structure contains a single byte with the ASCII representation 'W',
a 16-bit network byte order integer, and a character array with the
length given by the "len" integer. Character arrays described in this
fashion are an exact number of bytes, and are not assumed to be NUL
terminated.
A special data type representation is used here to indicate strings and
arrays of strings using C-language semantics of NUL termination.
char str[] |String, NUL terminated
------------------------|--------------------------------
char array[][] |Array of strings, NUL terminated
Here, "str" is a NUL-terminated string, and subsequent data items are
assumed to be located immediately following the NUL byte. "array" is a
stream of NUL-terminated strings, located immediately following each
other in the stream, leading up to the end of the data structure
(determined by the data packet's size).
###LINK/PACKET PROTOCOL
The MTA makes a connection to a milter by connecting to an IPC endpoint
(socket), via a stream-based protocol. TCPv4, TCPv6, and "Unix
filesystem" sockets can be used for connection to a milter.
(Configuration of Sendmail to make use of these different endpoint
addressing methods is not described here.)
Data is transmitted in both directions using a structured packet
protocol. Each packets is comprised of:
uint32 len |Size of data to follow
------------------------|--------------------------------
char cmd |Command/response code
char data[len-1] |Code-specific data (may be empty)
The connection can be closed at any time by either side. If closed by
the MTA, the milter program should release all state information for the
previously established connection. If closed by the milter program
without first sending an accept or reject action message, the MTA will
take the default action for any message in progress (configurable to
ignore the milter program, or reject with a 4xx or 5xx error).
###A TYPICAL MILTER CONVERSATION
The MTA drives the milter conversation. The milter program sends
responses when (and only when) specified by the particular command code
sent by the MTA. It is an error for a milter either to send a response
packet when not requested, or fail to send a response packet when
requested. The MTA may have limits on the time allowed for a response
packet to be sent.
The typical lifetime of a milter connection can be viewed as follows:
MTA |Milter
------------------------|----------------------
SMFIC_OPTNEG |
|SMFIC_OPTNEG
SMFIC_MACRO:'C' |
SMFIC_CONNECT |
|Accept/reject action
SMFIC_MACRO:'H' |
SMFIC_HELO |
|Accept/reject action
SMFIC_MACRO:'M' |
SMFIC_MAIL |
|Accept/reject action
SMFIC_MACRO:'R' |
SMFIC_RCPT |
|Accept/reject action
SMFIC_HEADER (multiple) |
|Accept/reject action (per SMFIC_HEADER)
SMFIC_EOH |
|Accept/reject action
SMFIC_BODY (multiple) |
|Accept/reject action (per SMFIC_BODY)
SMFIC_BODYEOB |
|Modification action (multiple, may be none)
|Accept/reject action
|
|(Reset state to before SMFIC_MAIL and continue,
| unless connection is dropped by MTA)
Several of these MTA/milter steps can be skipped if requested by the
SMFIC_OPTNEG response packet; see below.
###PROTOCOL NEGOTIATION
Milters can perform several actions on a SMTP transaction. The following is
a bitmask of possible actions, which may be set by the milter in the
"actions" field of the SMFIC_OPTNEG response packet. (Any action which MAY
be performed by the milter MUST be included in this field.)
0x01 |SMFIF_ADDHDRS |Add headers (SMFIR_ADDHEADER)
--------|-----------------------|-----------------------------------
0x02 |SMFIF_CHGBODY |Change body chunks (SMFIR_REPLBODY)
0x04 |SMFIF_ADDRCPT |Add recipients (SMFIR_ADDRCPT)
0x08 |SMFIF_DELRCPT |Remove recipients (SMFIR_DELRCPT)
0x10 |SMFIF_CHGHDRS |Change or delete headers (SMFIR_CHGHEADER)
0x20 |SMFIF_QUARANTINE |Quarantine message (SMFIR_QUARANTINE)
>XXX: SMFIF_DELRCPT has an impact on how address rewriting affects
addresses sent in the SMFIC_RCPT phase. This will be described in a
future revision of this document.
Protocol content can contain only selected parts of the SMTP
transaction. To mask out unwanted parts (saving on "over-the-wire" data
churn), the following can be set in the "protocol" field of the
SMFIC_OPTNEG response packet.
0x01 |SMFIP_NOCONNECT |Skip SMFIC_CONNECT
--------|-----------------------|-----------------
0x02 |SMFIP_NOHELO |Skip SMFIC_HELO
0x04 |SMFIP_NOMAIL |Skip SMFIC_MAIL
0x08 |SMFIP_NORCPT |Skip SMFIC_RCPT
0x10 |SMFIP_NOBODY |Skip SMFIC_BODY
0x20 |SMFIP_NOHDRS |Skip SMFIC_HEADER
0x40 |SMFIP_NOEOH |Skip SMFIC_EOH
For backwards-compatible milters, the milter should pay attention to the
"actions" and "protocol" fields of the SMFIC_OPTNEG packet, and mask out
any bits that are not part of the offered protocol content. The MTA may
reject the milter program if any action or protocol bit appears outside
the MTA's offered bitmask.
####COMMAND CODES
The following are commands transmitted from the MTA to the milter
program. The data structures represented occupy the "cmd" and "data"
fields of the packets described above in LINK/PACKET PROTOCOL. (In
other words, the data structures below take up exactly "len" bytes,
including the "cmd" byte.)
'A' |SMFIC_ABORT |Abort current filter checks
--------|---------------|---------------------------
| |Expected response: NONE
>Resets internal state of milter program to before SMFIC_HELO, but keeps
the connection open.
'B' |SMFIC_BODY |Body chunk
--------|---------------|---------------------------
||Expected response: Accept/reject action
char buf[] ||Up to MILTER_CHUNK_SIZE (65535) bytes
>These body chunks can be buffered by the milter for later replacement
via SMFIR_REPLBODY during the SMFIC_BODYEOB phase.
'C' |SMFIC_CONNECT |SMTP connection information
--------|---------------|---------------------------
||Expected response: Accept/reject action
char hostname[] ||Hostname, NUL terminated
char family ||Protocol family (see below)
uint16 port ||Port number (SMFIA_INET or SMFIA_INET6 only)
char address[] ||IP address (ASCII) or unix socket path, NUL terminated
>Sendmail invoked via the command line or via "-bs" will report the
connection as the "Unknown" protocol family.
>Protocol families used with SMFIC_CONNECT in the "family" field:
>'U' |SMFIA_UNKNOWN |Unknown (NOTE: Omits "port" and "host" fields entirely)
>--------|---------------|---------------------------
>'L' |SMFIA_UNIX |Unix (AF_UNIX/AF_LOCAL) socket ("port" is 0)
>'4' |SMFIA_INET |TCPv4 connection
>'6' |SMFIA_INET6 |TCPv6 connection
'D' |SMFIC_MACRO |Define macros
--------|---------------|---------------------------
||Expected response: NONE
char cmdcode ||Command for which these macros apply
char nameval[][] ||Array of NUL-terminated strings, alternating
||between name of macro and value of macro.
>SMFIC_MACRO appears as a packet just before the corresponding "cmdcode"
(here), which is the same identifier as the following command. The
names correspond to Sendmail macros, omitting the "$" identifier
character.
>Types of macros, and some commonly supplied macro names, used with
SMFIC_MACRO are as follows, organized by "cmdcode" value.
Implementations SHOULD NOT assume that any of these macros will be
present on a given connection. In particular, communications protocol
information may not be present on the "Unknown" protocol type.
>'C' |SMFIC_CONNECT |$_ $j ${daemon_name} ${if_name} ${if_addr}
--------|---------------|---------------------------
'H' |SMFIC_HELO |${tls_version} ${cipher} ${cipher_bits}
||${cert_subject} ${cert_issuer}
'M' |SMFIC_MAIL |$i ${auth_type} ${auth_authen} ${auth_ssf}
||${auth_author} ${mail_mailer} ${mail_host}
||${mail_addr}
'R' |SMFIC_RCPT |${rcpt_mailer} ${rcpt_host} ${rcpt_addr}
>For future compatibility, implementations MUST allow SMFIC_MACRO at any
time, but the handling of unspecified command codes, or SMFIC_MACRO not
appearing before its specified command, is currently undefined.
'E' |SMFIC_BODYEOB |Final body chunk
--------|---------------|---------------------------
||Expected response: Zero or more modification
||actions, then accept/reject action
'H' |SMFIC_HELO |HELO/EHLO name
--------|---------------|---------------------------
||Expected response: Accept/reject action
char helo[] ||HELO string, NUL terminated
'L' |SMFIC_HEADER |Mail header
--------|---------------|---------------------------
||Expected response: Accept/reject action
char name[] ||Name of header, NUL terminated
char value[] ||Value of header, NUL terminated
'M' |SMFIC_MAIL |MAIL FROM: information
--------|---------------|---------------------------
||Expected response: Accept/reject action
char args[][] ||Array of strings, NUL terminated (address at index 0).
||args[0] is sender, with <> qualification.
||args[1] and beyond are ESMTP arguments, if any.
'N' |SMFIC_EOH |End of headers marker
--------|---------------|---------------------------
||Expected response: Accept/reject action
'O' |SMFIC_OPTNEG |Option negotiation
--------|---------------|---------------------------
||Expected response: SMFIC_OPTNEG packet
uint32 version |SMFI_VERSION|(2)
uint32 actions ||Bitmask of allowed actions from SMFIF_*
uint32 protocol ||Bitmask of possible protocol content from SMFIP_*
'R' |SMFIC_RCPT |RCPT TO: information
--------|---------------|---------------------------
||Expected response: Accept/reject action
char args[][] ||Array of strings, NUL terminated (address at index 0).
||args[0] is recipient, with <> qualification.
||args[1] and beyond are ESMTP arguments, if any.
'Q' |SMFIC_QUIT |Quit milter communication
--------|---------------|---------------------------
||Expected response: Close milter connection
####RESPONSE CODES
The following are commands transmitted from the milter program to the
MTA, in response to the appropriate type of command packet. The data
structures represented occupy the "cmd" and "data" fields of the packets
described above in LINK/PACKET PROTOCOL. (In other words, the data
structures below take up exactly "len" bytes, including the "cmd" byte.)
Response codes:
'+' |SMFIR_ADDRCPT |Add recipient (modification action)
--------|---------------|---------------------------
char rcpt[] ||New recipient, NUL terminated
'-' |SMFIR_DELRCPT |Remove recipient (modification action)
--------|---------------|---------------------------
char rcpt[] ||Recipient to remove, NUL terminated (string must match the one in SMFIC_RCPT exactly)
'a' |SMFIR_ACCEPT |Accept message completely (accept/reject action)
--------|---------------|---------------------------
>This will skip to the end of the milter sequence, and recycle back to
the state before SMFIC_MAIL. The MTA may, instead, close the connection
at that point.
'b' |SMFIR_REPLBODY |Replace body (modification action)
--------|---------------|---------------------------
char buf[] ||Full body, as a single packet
'c' |SMFIR_CONTINUE |Accept and keep processing (accept/reject action)
--------|---------------|---------------------------
>If issued at the end of the milter conversation, functions the same as
SMFIR_ACCEPT.
'd' |SMFIR_DISCARD |Set discard flag for entire message (accept/reject action)
--------|---------------|---------------------------
>Note that message processing MAY continue afterwards, but the mail will
not be delivered even if accepted with SMFIR_ACCEPT.
'h' |SMFIR_ADDHEADER|Add header (modification action)
--------|---------------|---------------------------
char name[] ||Name of header, NUL terminated
char value[] ||Value of header, NUL terminated
'm' |SMFIR_CHGHEADER|Change header (modification action)
--------|---------------|---------------------------
uint32 index ||Index of the occurrence of this header
char name[] ||Name of header, NUL terminated
char value[] ||Value of header, NUL terminated
>Note that the "index" above is per-name--i.e. a 3 in this field
indicates that the modification is to be applied to the third such
header matching the supplied "name" field. A zero length string for
"value", leaving only a single NUL byte, indicates that the header
should be deleted entirely.
'p' |SMFIR_PROGRESS |Progress (asynchronous action)
--------|---------------|---------------------------
This is an asynchronous response which is sent to the MTA to reset the
communications timer during long operations. The MTA should consume
as many of these responses as are sent, waiting for the real response
for the issued command.
'q' |SMFIR_QUARANTINE|Quarantine message (modification action)
--------|---------------|---------------------------
char reason[] ||Reason for quarantine, NUL terminated
This quarantines the message into a holding pool defined by the MTA.
(First implemented in Sendmail in version 8.13; offered to the milter by
the SMFIF_QUARANTINE flag in "actions" of SMFIC_OPTNEG.)
'r' |SMFIR_REJECT |Reject command/recipient with a 5xx (accept/reject action)
--------|---------------|---------------------------
't' |SMFIR_TEMPFAIL |Reject command/recipient with a 4xx (accept/reject action)
--------|---------------|---------------------------
'y' |SMFIR_REPLYCODE|Send specific Nxx reply message (accept/reject action)
--------|---------------|---------------------------
char smtpcode[3] ||Nxx code (ASCII), not NUL terminated
char space ||' '
char text[] ||Text of reply message, NUL terminated
>'%' characters present in "text" must be doubled to prevent problems
with printf-style formatting that may be used by the MTA.
'O' |SMFIC_OPTNEG |Option negotiation (in response to SMFIC_OPTNEG)
--------|---------------|---------------------------
uint32 version ||SMFI_VERSION (2)
uint32 actions ||Bitmask of requested actions from SMFIF_*
uint32 protocol ||Bitmask of undesired protocol content from SMFIP_*
###CREDITS
Sendmail, Inc. - for the Sendmail program itself
The anti-spam community - for making e-mail a usable medium again
The spam community - for convincing me that it's time to really do
somthing to quell the inflow of their crap