Exim supports a number of different ways of using the SMTP protocol, and its LMTP variant, which is an interactive protocol for transferring messages into a closed mail store application. This chapter contains details of how SMTP is processed. For incoming mail, the following are available:
SMTP over TCP/IP (Exim daemon or inetd);
SMTP over the standard input and output (the -bs option);
Batched SMTP on the standard input (the -bS option).
For mail delivery, the following are available:
SMTP over TCP/IP (the smtp transport);
LMTP over TCP/IP (the smtp transport with the protocol option set to ``lmtp'');
LMTP over a pipe to a process running in the local host (the lmtp transport);
Batched SMTP to a file or pipe (the appendfile and pipe transports with the use_bsmtp option set).
Batched SMTP is the name for a process in which batches of messages are stored in or read from files (or pipes), in a format in which SMTP commands are used to contain the envelope information.
Outgoing SMTP and LMTP over TCP/IP is implemented by the smtp transport. The protocol option selects which protocol is to be used, but the actual processing is the same in both cases.
If, in response to its EHLO command, Exim is told that the SIZE parameter is supported, it adds SIZE=<n> to each subsequent MAIL command. The value of <n> is the message size plus the value of the size_addition option (default 1024) to allow for additions to the message such as per-transport header lines, or changes made in a transport filter. If size_addition is set negative, the use of SIZE is suppressed.
If the remote server advertises support for PIPELINING, Exim uses the pipelining extension to SMTP (RFC 2197) to reduce the number of TCP/IP packets required for the transaction.
If the remote server advertises support for the STARTTLS command, and Exim was built to support TLS encryption, it tries to start a TLS session unless the server matches hosts_avoid_tls. See chapter 36 for more details.
If the remote server advertises support for the AUTH command, Exim scans the authenticators configuration for any suitable client settings, as described in chapter 32.
Responses from the remote host are supposed to be terminated by CR followed by LF. However, there are known to be hosts that do not send CR characters, so in order to be able to interwork with such hosts, Exim treats LF on its own as a line terminator.
If a message contains a number of different addresses, all those with the same characteristics (for example, the same envelope sender) that resolve to the same set of hosts, in the same order, are sent in a single SMTP transaction, even if they are for different domains, unless there are more than the setting of the max_rcpts option in the smtp transport allows, in which case they are split into groups containing no more than max_rcpts addresses each. If remote_max_parallel is greater than one, such groups may be sent in parallel sessions. The order of hosts with identical MX values is not significant when checking whether addresses can be batched in this way.
When the smtp transport suffers a temporary failure that is not message-related, Exim updates its transport-specific database, which contains records indexed by host name that remember which messages are waiting for each particular host. It also updates the retry database with new retry times. Exim's retry hints are based on host name plus IP address, so if one address of a multi-homed host is broken, it will soon be skipped most of the time. See the next section for more detail about error handling.
When a message is successfully delivered over a TCP/IP SMTP connection, Exim looks in the hints database for the transport to see if there are any queued messages waiting for the host to which it is connected. If it finds one, it creates a new Exim process using the -MC option (which can only be used by a process running as root or the Exim user) and passes the TCP/IP socket to it so that it can deliver another message using the same socket. The new process does only those deliveries that are routed to the connected host, and may in turn pass the socket on to a third process, and so on.
The connection_max_messages option of the smtp transport can be used to limit the number of messages sent down a single TCP/IP connection. The second and subsequent messages delivered down an existing connection are identified in the main log by the addition of an asterisk after the closing square bracket of the IP address.
Three different kinds of error are recognized for outgoing SMTP: host errors, message errors, and recipient errors.
A host error is not associated with a particular message or with a particular recipient of a message. The host errors are:
Connection refused or timed out,
Any error response code on connection,
Any error response code to EHLO or HELO,
Loss of connection at any time, except after ``.'',
I/O errors at any time,
Timeouts during the session, other than in response to MAIL, RCPT or the ``.'' at the end of the data.
For a host error, a permanent error response on connection, or in response to EHLO, causes all addresses routed to the host to be failed. Any other host error causes all addresses to be deferred, and retry data to be created for the host. It is not tried again, for any message, until its retry time arrives. If the current set of addresses are not all delivered in this run (to some alternative host), the message is added to the list of those waiting for this host, so if it is still undelivered when a subsequent successful delivery is made to the host, it will be sent down the same SMTP connection.
A message error is associated with a particular message when sent to a particular host, but not with a particular recipient of the message. The message errors are:
Any error response code to MAIL, DATA, or the ``.'' that terminates the data,
Timeout after MAIL,
Timeout or loss of connection after the ``.'' that terminates the data. A timeout after the DATA command itself is treated as a host error, as is loss of connection at any other time.
For a message error, a permanent error response (5xx) causes all addresses to be failed, and a delivery error report to be returned to the sender. A temporary error response (4xx), or one of the timeouts, causes all addresses to be deferred. Retry data is not created for the host, but instead, a retry record for the combination of host plus message id is created. The message is not added to the list of those waiting for this host. This ensures that the failing message will not be sent to this host again until the retry time arrives. However, other messages that are routed to the host are not affected, so if it is some property of the message that is causing the error, it will not stop the delivery of other mail.
If the remote host specified support for the SIZE parameter in its response to EHLO, Exim adds SIZE=nnn to the MAIL command, so an over-large message will cause a message error because the error arrives as a response to MAIL.
A recipient error is associated with a particular recipient of a message. The recipient errors are:
Any error response to RCPT,
Timeout after RCPT.
For a recipient error, a permanent error response (5xx) causes the recipient address to be failed, and a bounce message to be returned to the sender. A temporary error response (4xx) or a timeout causes the failing address to be deferred, and routing retry data to be created for it. This is used to delay processing of the address in subsequent queue runs, until its routing retry time arrives. This applies to all messages, but because it operates only in queue runs, one attempt will be made to deliver a new message to the failing address before the delay starts to operate. This ensures that, if the failure is really related to the message rather than the recipient (``message too big for this recipient'' is a possible example), other messages have a chance of getting delivered. If a delivery to the address does succeed, the retry information gets cleared, so all stuck messages get tried again, and the retry clock is reset.
The message is not added to the list of those waiting for this host. Use of the host for other messages is unaffected, and except in the case of a timeout, other recipients are processed independently, and may be successfully delivered in the current SMTP session. After a timeout it is of course impossible to proceed with the session, so all addresses get deferred. However, those other than the one that failed do not suffer any subsequent retry delays. Therefore, if one recipient is causing trouble, the others have a chance of getting through when a subsequent delivery attempt occurs before the failing recipient's retry time.
In all cases, if there are other hosts (or IP addresses) available for the current set of addresses (for example, from multiple MX records), they are tried in this run for any undelivered addresses, subject of course to their own retry data. In other words, recipient error retry data does not take effect until the next delivery attempt.
Some hosts have been observed to give temporary error responses to every MAIL command at certain times (``insufficient space'' has been seen). It would be nice if such circumstances could be recognized, and defer data for the host itself created, but this is not possible within the current Exim design. What actually happens is that retry data for every (host, message) combination is created.
The reason that timeouts after MAIL and RCPT are treated specially is that these can sometimes arise as a result of the remote host's verification procedures. Exim makes this assumption, and treats them as if a temporary error response had been received. A timeout after ``.'' is treated specially because it is known that some broken implementations fail to recognize the end of the message if the last character of the last line is a binary zero. Thus, it is helpful to treat this case as a message error.
Timeouts at other times are treated as host errors, assuming a problem with the host, or the connection to it. If a timeout after MAIL, RCPT, or ``.'' is really a connection problem, the assumption is that at the next try the timeout is likely to occur at some other point in the dialogue, causing it then to be treated as a host error.
There is experimental evidence that some MTAs drop the connection after the terminating ``.'' if they do not like the contents of the message for some reason, in contravention of the RFC, which indicates that a 5xx response should be given. That is why Exim treats this case as a message rather than a host error, in order not to delay other messages to the same host.
Variable Envelope Return Paths - see ftp://koobera.math.uic.edu/www/proto/verp.txt - can be supported in Exim by using the return_path generic transport option to rewrite the return path at transport time. For example, the following could be used on an smtp transport:
return_path = \
${if match {$return_path}{^(.+?)-request@your.dom.example\$}\
{$1-request=$local_part%$domain@your.dom.example}fail}
This has the effect of rewriting the return path (envelope sender) on all outgoing SMTP messages, if the local part of the original return path ends in ``-request'', and the domain is your.dom.example. The rewriting inserts the local part and domain of the recipient into the return path. Suppose, for example, that a message whose return path has been set to somelist-request@your.dom.example is sent to subscriber@other.dom.example. In the transport, the return path is rewritten as
somelist-request=subscriber%other.dom.example@your.dom.example
For this to work, you must arrange for outgoing messages that have ``-request'' in their return paths to have just a single recipient. This can be done by setting
max_rcpt = 1
in the smtp transport. Otherwise a single copy of a message might be addressed to several different recipients in the same domain, in which case $local_part is not available (because it is not unique). Of course, if you do start sending out messages with this kind of return path, you must also configure Exim to accept the bounce messages that come back to those paths. Typically this would be done by setting an local_part_suffix option for a suitable router.
The overhead incurred in using VERP depends very much on the size of the message, the number of recipient addresses that resolve to the same remote host, and the speed of the connection over which the message is being sent. If a lot of addresses resolve to the same host and the connection is slow, sending a separate copy of the message for each address may take substantially longer than sending a single copy with many recipients (for which VERP cannot be used).
Incoming SMTP messages can be accepted in one of two ways: by running a listening daemon, or by using inetd. In the latter case, the entry in /etc/inetd.conf should be like this:
smtp stream tcp nowait exim /opt/exim/bin/exim in.exim -bs
Exim distinguishes between this case and the case of a locally running user agent using the -bs option by checking whether or not the standard input is a socket. When it is, either the port must be privileged (less than 1024), or the caller must be root or the Exim user. If any other user passes a socket with an unprivileged port number, Exim prints a message on the standard error stream and exits with an error code.
By default, Exim does not make a log entry when a remote host connects or disconnects (either via the daemon or inetd), unless the disconnection is unexpected. It can be made to write such log entries by setting the smtp_connection log selector.
Commands from the remote host are supposed to be terminated by CR followed by LF. However, there are known to be hosts that do not send CR characters, so in order to be able to interwork with such hosts, Exim treats LF on its own as a line terminator.
One area that sometimes gives rise to problems concerns the EHLO or HELO commands. Some clients send syntactically invalid versions of these commands, which Exim rejects by default. (This is nothing to do with verifying the data that is sent, so helo_verify_hosts is not relevant.) You can tell Exim not to apply a syntax check by setting helo_accept_junk_hosts to match the broken hosts that send invalid commands.
The amount of disc space available is checked whenever SIZE is received on a MAIL command, independently of whether message_size_limit or check_spool_space is configured, unless smtp_check_spool_space is set false. A temporary error is given if there is not enough space. If check_spool_space is set, the check is for that amount of space plus the value given with SIZE, that is, it checks that the addition of the incoming message will not reduce the space below the threshold.
When a message is successfully received, Exim includes the local message id in its response to the final ``.'' that terminates the data. If the remote host logs this text it can help with tracing what has happened to a message.
The Exim daemon can limit the number of simultaneous incoming connections it is prepared to handle (see the smtp_accept_max option). It can also limit the number of simultaneous incoming connections from a single remote host (see the smtp_accept_max_per_host option). Additional connection attempts are rejected using the SMTP temporary error code 421.
The Exim daemon does not rely on the SIGCHLD signal to detect when a subprocess has finished, as this can get lost at busy times. Instead, it looks for completed subprocesses every time it wakes up. Provided there are other things happening (new incoming calls, starts of queue runs), completed processes will be noticed and tidied away. On very quiet systems you may sometimes see a ``defunct'' Exim process hanging about. This is not a problem; it will be noticed when the daemon next wakes up.
When running as a daemon, Exim can reserve some SMTP slots for specific hosts, and can also be set up to reject SMTP calls from non-reserved hosts at times of high system load - for details see the smtp_accept_reserve, smtp_load_reserve, and smtp_reserve_hosts options. The load check applies in both the daemon and inetd cases.
Exim normally starts a delivery process for each message received, though this can be varied by means of the -odq command line option and the queue_only, queue_only_file, and queue_only_load options. The number of simultaneously running delivery processes started in this way from SMTP input can be limited by the smtp_accept_queue and smtp_accept_queue_per_connection options. When either limit is reached, subsequently received messages are just put on the input queue without starting a delivery process.
The controls that involve counts of incoming SMTP calls (smtp_accept_max, smtp_accept_queue, smtp_accept_reserve) are not available when Exim is started up from the inetd daemon, because in that case each connection is handled by an entirely independent Exim process. Control by load average is, however, available with inetd.
Exim can be configured to verify addresses in incoming SMTP commands as they are received. See chapter 37 for details. It can also be configured to rewrite addresses at this time - before any syntax checking is done. See section 30.7.
Exim can also be configured to limit the rate at which a client host submits MAIL and RCPT commands in a single SMTP session. See the smtp_ratelimit_hosts option.
When Exim receives a VRFY or EXPN command on a TCP/IP connection, it runs the ACL specified by acl_smtp_vrfy or acl_smtp_expn (as appropriate) in order to decide whether the command should be accepted or not. If no ACL is defined, the command is rejected.
When VRFY is accepted, it runs exactly the same code as when Exim is called with the -bv option. When EXPN is accepted, a single-level expansion of the address is done. EXPN is treated as an ``address test'' (similar to the -bt option) rather than a verification (the -bv option). If an unqualified local part is given as the argument to EXPN, it is qualified with qualify_domain. Rejections of VRFY and EXPN commands are logged on the main and reject logs, and VRFY verification failures are logged on the main log for consistency with RCPT failures.
RFC 1985 describes an SMTP command called ETRN that is designed to overcome the security problems of the TURN command (which has fallen into disuse). When Exim receives an ETRN command on a TCP/IP connection, it runs the ACL specified by acl_smtp_etrn in order to decide whether the command should be accepted or not. If no ACL is defined, the command is rejected.
The ETRN command is concerned with ``releasing'' messages that are awaiting delivery to certain hosts. As Exim does not organize its message queue by host, the only form of ETRN that is supported by default is the one where the text starts with the ``#'' prefix, in which case the remainder of the text is specific to the SMTP server. A valid ETRN command causes a run of Exim with the -R option to happen, with the remainder of the ETRN text as its argument. For example,
ETRN #brigadoon
runs the command
exim -R brigadoon
which causes a delivery attempt on all messages with undelivered addresses containing the text ``brigadoon''. When smtp_etrn_serialize is set (the default), Exim prevents the simultaneous execution of more than one queue run for the same argument string as a result of an ETRN command. This stops a misbehaving client from starting more than one queue-runner at once.
Exim implements the serialization by means of a hints database in which a record is written whenever a process is started by ETRN, and deleted when the process completes. However, Exim does not keep the SMTP session waiting for the ETRN process to complete. Once ETRN is accepted, the client is sent a ``success'' return code. Obviously there is scope for hints records to get left lying around if there is a system or program crash. To guard against this, Exim ignores any records that are more than six hours old.
For more control over what ETRN does, the smtp_etrn_command option can used. This specifies a command that is run whenever ETRN is received, whatever the form of its argument. For example:
smtp_etrn_command = /etc/etrn_command $domain $sender_host_address
The string is split up into arguments which are independently expanded. The expansion variable $domain is set to the argument of the ETRN command, and no syntax checking is done on the contents of this argument. Exim does not wait for the command to complete, so its status code is not checked. Exim runs under its own uid and gid when receiving incoming SMTP, so it is not possible for it to change them before running the command.
Some user agents use SMTP to pass messages to their local MTA using the standard input and output, as opposed to passing the envelope on the command line and writing the message to the standard input. This is supported by the -bs option. This form of SMTP is handled in the same way as incoming messages over TCP/IP (including the use of ACLs), except that the envelope sender given in a MAIL command is ignored unless the caller is trusted. In an ACL you can detect this form of SMTP input by testing for an empty host identification. It is common to have this as the first line in the ACL that runs for RCPT commands:
accept hosts = :
This accepts SMTP messages from local processes without doing any other tests.
Both the appendfile and pipe transports can be used for handling batched SMTP. Each has an option called use_bsmtp which causes messages to be output in BSMTP format. No SMTP responses are possible for this form of delivery. All it is doing is using SMTP commands as a way of transmitting the envelope along with the message.
The message is written to the file or pipe preceded by the SMTP commands MAIL and RCPT, and followed by a line containing a single dot. Lines in the message that start with a dot have an extra dot added. The SMTP command HELO is not normally used. If it is required, the message_prefix option can be used to specify it.
Because appendfile and pipe are both local transports, they accept only one recipient address at a time by default. However, you can arrange for them to handle several addresses at once by setting the batch_max option. When this is done for BSMTP, messages may contain multiple RCPT commands. See chapter 24 for more details.
When one or more addresses are routed to a BSMTP transport by a router that sets up a host list, the name of the first host on the list is available to the transport in the variable $host. Here is an example of such a transport and router:
begin routers route_append: driver = manualroute transport = smtp_appendfile route_list = domain.example batch.host.example begin transports smtp_appendfile: driver = appendfile directory = /var/bsmtp/$host batch_max = 1000 use_bsmtp user = exim
This causes messages addressed to domain.example to be written in BSMTP format to /var/bsmtp/batch.host.example, with only a single copy of each message (unless there are more than 1000 recipients).
The -bS command line option causes Exim to accept one or more messages by reading SMTP on the standard input, but to generate no responses. If the caller is trusted, the senders in the MAIL commands are believed; otherwise the sender is always the caller of Exim. Unqualified senders and receivers are not rejected (there seems little point) but instead just get qualified. HELO and EHLO act as RSET; VRFY, EXPN, ETRN and HELP, act as NOOP; QUIT quits.
No policy checking is done for BSMTP input. That is, no ACL is run at anytime. In this respect it is like non-SMTP local input.
If an error is detected while reading a message, including a missing ``.'' at the end, Exim gives up immediately. It writes details of the error to the standard output in a stylized way that the calling program should be able to make some use of automatically, for example:
554 Unexpected end of file Transaction started in line 10 Error detected in line 14
It writes a more verbose version, for human consumption, to the standard error file, for example:
An error was detected while processing a file of BSMTP input. The error message was: 501 '>' missing at end of address The SMTP transaction started in line 10. The error was detected in line 12. The SMTP command at fault was: rcpt to:<malformed@in.com.plete 1 previous message was successfully processed. The rest of the batch was abandoned.
The return code from Exim is zero only if there were no errors. It is 1 if some messages were accepted before an error was detected, and 2 if no messages were accepted.