Exim Internet Mailer

1. Main configuration settings

The main (global) configuration option settings must always come first in the file. The first thing you’ll see in the file, after some initial comments, is the line

# primary_hostname =

This is a commented-out setting of the primary_hostname option. Exim needs to know the official, fully qualified name of your host, and this is where you can specify it. However, in most cases you do not need to set this option. When it is unset, Exim uses the uname() system function to obtain the host name.

The first three non-comment configuration lines are as follows:

domainlist local_domains = @
domainlist relay_to_domains =
hostlist   relay_from_hosts = 127.0.0.1

These are not, in fact, option settings. They are definitions of two named domain lists and one named host list. Exim allows you to give names to lists of domains, hosts, and email addresses, in order to make it easier to manage the configuration file (see section 10.5).

The first line defines a domain list called local_domains; this is used later in the configuration to identify domains that are to be delivered on the local host.

There is just one item in this list, the string “@”. This is a special form of entry which means “the name of the local host”. Thus, if the local host is called a.host.example, mail to any.user@a.host.example is expected to be delivered locally. Because the local host’s name is referenced indirectly, the same configuration file can be used on different hosts.

The second line defines a domain list called relay_to_domains, but the list itself is empty. Later in the configuration we will come to the part that controls mail relaying through the local host; it allows relaying to any domains in this list. By default, therefore, no relaying on the basis of a mail domain is permitted.

The third line defines a host list called relay_from_hosts. This list is used later in the configuration to permit relaying from any host or IP address that matches the list. The default contains just the IP address of the IPv4 loopback interface, which means that processes on the local host are able to submit mail for relaying by sending it over TCP/IP to that interface. No other hosts are permitted to submit messages for relaying.

Just to be sure there’s no misunderstanding: at this point in the configuration we aren’t actually setting up any controls. We are just defining some domains and hosts that will be used in the controls that are specified later.

The next two configuration lines are genuine option settings:

acl_smtp_rcpt = acl_check_rcpt
acl_smtp_data = acl_check_data

These options specify Access Control Lists (ACLs) that are to be used during an incoming SMTP session for every recipient of a message (every RCPT command), and after the contents of the message have been received, respectively. The names of the lists are acl_check_rcpt and acl_check_data, and we will come to their definitions below, in the ACL section of the configuration. The RCPT ACL controls which recipients are accepted for an incoming message – if a configuration does not provide an ACL to check recipients, no SMTP mail can be accepted. The DATA ACL allows the contents of a message to be checked.

Two commented-out option settings are next:

# av_scanner = clamd:/tmp/clamd
# spamd_address = 127.0.0.1 783

These are example settings that can be used when Exim is compiled with the content-scanning extension. The first specifies the interface to the virus scanner, and the second specifies the interface to SpamAssassin. Further details are given in chapter 40.

Two more commented-out options settings follow:

# qualify_domain =
# qualify_recipient =

The first of these specifies a domain that Exim uses when it constructs a complete email address from a local login name. This is often needed when Exim receives a message from a local process. If you do not set qualify_domain, the value of primary_hostname is used. If you set both of these options, you can have different qualification domains for sender and recipient addresses. If you set only the first one, its value is used in both cases.

The following line must be uncommented if you want Exim to recognize addresses of the form user@[10.11.12.13] that is, with a “domain literal” (an IP address within square brackets) instead of a named domain.

# allow_domain_literals

The RFCs still require this form, but many people think that in the modern Internet it makes little sense to permit mail to be sent to specific hosts by quoting their IP addresses. This ancient format has been used by people who try to abuse hosts by using them for unwanted relaying. However, some people believe there are circumstances (for example, messages addressed to postmaster) where domain literals are still useful.

The next configuration line is a kind of trigger guard:

never_users = root

It specifies that no delivery must ever be run as the root user. The normal convention is to set up root as an alias for the system administrator. This setting is a guard against slips in the configuration. The list of users specified by never_users is not, however, the complete list; the build-time configuration in Local/Makefile has an option called FIXED_NEVER_USERS specifying a list that cannot be overridden. The contents of never_users are added to this list. By default FIXED_NEVER_USERS also specifies root.

When a remote host connects to Exim in order to send mail, the only information Exim has about the host’s identity is its IP address. The next configuration line,

host_lookup = *

specifies that Exim should do a reverse DNS lookup on all incoming connections, in order to get a host name. This improves the quality of the logging information, but if you feel it is too expensive, you can remove it entirely, or restrict the lookup to hosts on “nearby” networks. Note that it is not always possible to find a host name from an IP address, because not all DNS reverse zones are maintained, and sometimes DNS servers are unreachable.

The next two lines are concerned with ident callbacks, as defined by RFC 1413 (hence their names):

rfc1413_hosts = *
rfc1413_query_timeout = 5s

These settings cause Exim to make ident callbacks for all incoming SMTP calls. You can limit the hosts to which these calls are made, or change the timeout that is used. If you set the timeout to zero, all ident calls are disabled. Although they are cheap and can provide useful information for tracing problem messages, some hosts and firewalls have problems with ident calls. This can result in a timeout instead of an immediate refused connection, leading to delays on starting up an incoming SMTP session.

When Exim receives messages over SMTP connections, it expects all addresses to be fully qualified with a domain, as required by the SMTP definition. However, if you are running a server to which simple clients submit messages, you may find that they send unqualified addresses. The two commented-out options:

# sender_unqualified_hosts =
# recipient_unqualified_hosts =

show how you can specify hosts that are permitted to send unqualified sender and recipient addresses, respectively.

The percent_hack_domains option is also commented out:

# percent_hack_domains =

It provides a list of domains for which the “percent hack” is to operate. This is an almost obsolete form of explicit email routing. If you do not know anything about it, you can safely ignore this topic.

The last two settings in the main part of the default configuration are concerned with messages that have been “frozen” on Exim’s queue. When a message is frozen, Exim no longer continues to try to deliver it. Freezing occurs when a bounce message encounters a permanent failure because the sender address of the original message that caused the bounce is invalid, so the bounce cannot be delivered. This is probably the most common case, but there are also other conditions that cause freezing, and frozen messages are not always bounce messages.

ignore_bounce_errors_after = 2d
timeout_frozen_after = 7d

The first of these options specifies that failing bounce messages are to be discarded after 2 days on the queue. The second specifies that any frozen message (whether a bounce message or not) is to be timed out (and discarded) after a week. In this configuration, the first setting ensures that no failing bounce message ever lasts a week.

2. ACL configuration

In the default configuration, the ACL section follows the main configuration. It starts with the line

begin acl

and it contains the definitions of two ACLs, called acl_check_rcpt and acl_check_data, that were referenced in the settings of acl_smtp_rcpt and acl_smtp_data above.

The first ACL is used for every RCPT command in an incoming SMTP message. Each RCPT command specifies one of the message’s recipients. The ACL statements are considered in order, until the recipient address is either accepted or rejected. The RCPT command is then accepted or rejected, according to the result of the ACL processing.

acl_check_rcpt:

This line, consisting of a name terminated by a colon, marks the start of the ACL, and names it.

accept  hosts = :

This ACL statement accepts the recipient if the sending host matches the list. But what does that strange list mean? It doesn’t actually contain any host names or IP addresses. The presence of the colon puts an empty item in the list; Exim matches this only if the incoming message did not come from a remote host, because in that case, the remote hostname is empty. The colon is important. Without it, the list itself is empty, and can never match anything.

What this statement is doing is to accept unconditionally all recipients in messages that are submitted by SMTP from local processes using the standard input and output (that is, not using TCP/IP). A number of MUAs operate in this manner.

deny    message       = Restricted characters in address
        domains       = +local_domains
        local_parts   = ^[.] : ^.*[@%!/|]

deny    message       = Restricted characters in address
        domains       = !+local_domains
        local_parts   = ^[./|] : ^.*[@%!] : ^.*/\\.\\./

These statements are concerned with local parts that contain any of the characters “@”, “%”, “!”, “/”, “|”, or dots in unusual places. Although these characters are entirely legal in local parts (in the case of “@” and leading dots, only if correctly quoted), they do not commonly occur in Internet mail addresses.

The first three have in the past been associated with explicitly routed addresses (percent is still sometimes used – see the percent_hack_domains option). Addresses containing these characters are regularly tried by spammers in an attempt to bypass relaying restrictions, and also by open relay testing programs. Unless you really need them it is safest to reject these characters at this early stage. This configuration is heavy-handed in rejecting these characters for all messages it accepts from remote hosts. This is a deliberate policy of being as safe as possible.

The first rule above is stricter, and is applied to messages that are addressed to one of the local domains handled by this host. This is implemented by the first condition, which restricts it to domains that are listed in the local_domains domain list. The “+” character is used to indicate a reference to a named list. In this configuration, there is just one domain in local_domains, but in general there may be many.

The second condition on the first statement uses two regular expressions to block local parts that begin with a dot or contain “@”, “%”, “!”, “/”, or “|”. If you have local accounts that include these characters, you will have to modify this rule.

Empty components (two dots in a row) are not valid in RFC 2822, but Exim allows them because they have been encountered in practice. (Consider the common convention of local parts constructed as “first-initial.second-initial.family-name” when applied to someone like the author of Exim, who has no second initial.) However, a local part starting with a dot or containing “/../” can cause trouble if it is used as part of a file name (for example, for a mailing list). This is also true for local parts that contain slashes. A pipe symbol can also be troublesome if the local part is incorporated unthinkingly into a shell command line.

The second rule above applies to all other domains, and is less strict. This allows your own users to send outgoing messages to sites that use slashes and vertical bars in their local parts. It blocks local parts that begin with a dot, slash, or vertical bar, but allows these characters within the local part. However, the sequence “/../” is barred. The use of “@”, “%”, and “!” is blocked, as before. The motivation here is to prevent your users (or your users’ viruses) from mounting certain kinds of attack on remote sites.

accept  local_parts   = postmaster
        domains       = +local_domains

This statement, which has two conditions, accepts an incoming address if the local part is postmaster and the domain is one of those listed in the local_domains domain list. The “+” character is used to indicate a reference to a named list. In this configuration, there is just one domain in local_domains, but in general there may be many.

The presence of this statement means that mail to postmaster is never blocked by any of the subsequent tests. This can be helpful while sorting out problems in cases where the subsequent tests are incorrectly denying access.

require verify        = sender

This statement requires the sender address to be verified before any subsequent ACL statement can be used. If verification fails, the incoming recipient address is refused. Verification consists of trying to route the address, to see if a bounce message could be delivered to it. In the case of remote addresses, basic verification checks only the domain, but callouts can be used for more verification if required. Section 39.31 discusses the details of address verification.

accept  hosts         = +relay_from_hosts
        control       = submission

This statement accepts the address if the message is coming from one of the hosts that are defined as being allowed to relay through this host. Recipient verification is omitted here, because in many cases the clients are dumb MUAs that do not cope well with SMTP error responses. For the same reason, the second line specifies “submission mode” for messages that are accepted. This is described in detail in section 43.1; it causes Exim to fix messages that are deficient in some way, for example, because they lack a Date: header line. If you are actually relaying out from MTAs, you should probably add recipient verification here, and disable submission mode.

accept  authenticated = *
        control       = submission

This statement accepts the address if the client host has authenticated itself. Submission mode is again specified, on the grounds that such messages are most likely to come from MUAs. The default configuration does not define any authenticators, which means that no client can in fact authenticate. You will need to add authenticator definitions if you want to make use of this ACL statement.

# deny    message     = rejected because $sender_host_address \
#                       is in a black list at $dnslist_domain\n\
#                       $dnslist_text
#         dnslists    = black.list.example
#
# warn    message     = X-Warning: $sender_host_address is \
#                       in a black list at $dnslist_domain
#         log_message = found in $dnslist_domain
#         dnslists    = black.list.example

These commented-out lines are examples of how you could configure Exim to check sending hosts against a DNS black list. The first statement rejects messages from blacklisted hosts, whereas the second merely inserts a warning header line.

accept  domains       = +local_domains
        endpass
        verify        = recipient

This statement accepts the incoming recipient address if its domain is one of the local domains, but only if the address can be verified. Verification of local addresses normally checks both the local part and the domain. The endpass line needs some explanation: if the condition above endpass fails, that is, if the address is not in a local domain, control is passed to the next ACL statement. However, if the condition below endpass fails, that is, if a recipient in a local domain cannot be verified, access is denied and the recipient is rejected.

accept  domains       = +relay_to_domains
        endpass
        verify        = recipient

This statement accepts the incoming recipient address if its domain is one of the domains for which this host is a relay, but again, only if the address can be verified.

deny    message       = relay not permitted

The final statement denies access, giving a specific error message. Reaching the end of the ACL also causes access to be denied, but with the generic message “administrative prohibition”.

acl_check_data:

This line marks the start of the second ACL, and names it. Most of the contents of this ACL are commented out:

# deny    malware   = *
#         message   = This message contains a virus \
#                     ($malware_name).

These lines are examples of how to arrange for messages to be scanned for viruses when Exim has been compiled with the content-scanning extension, and a suitable virus scanner is installed. If the message is found to contain a virus, it is rejected with the given custom error message.

# warn    spam      = nobody
#         message   = X-Spam_score: $spam_score\n\
#                     X-Spam_score_int: $spam_score_int\n\
#                     X-Spam_bar: $spam_bar\n\
#                     X-Spam_report: $spam_report

These lines are an example of how to arrange for messages to be scanned by SpamAssassin when Exim has been compiled with the content-scanning extension, and SpamAssassin has been installed. The SpamAssassin check is run with nobody as its user parameter, and the results are added to the message as a series of extra header line. In this case, the message is not rejected, whatever the spam score.

accept

This final line in the DATA ACL accepts the message unconditionally.

3. Router configuration

The router configuration comes next in the default configuration, introduced by the line

begin routers

Routers are the modules in Exim that make decisions about where to send messages. An address is passed to each router in turn, until it is either accepted, or failed. This means that the order in which you define the routers matters. Each router is fully described in its own chapter later in this manual. Here we give only brief overviews.

# domain_literal:
#   driver = ipliteral
#   domains = !+local_domains
#   transport = remote_smtp

This router is commented out because the majority of sites do not want to support domain literal addresses (those of the form user@[10.9.8.7]). If you uncomment this router, you also need to uncomment the setting of allow_domain_literals in the main part of the configuration.

dnslookup:
  driver = dnslookup
  domains = ! +local_domains
  transport = remote_smtp
  ignore_target_hosts = 0.0.0.0 : 127.0.0.0/8
  no_more

The first uncommented router handles addresses that do not involve any local domains. This is specified by the line

domains = ! +local_domains

The domains option lists the domains to which this router applies, but the exclamation mark is a negation sign, so the router is used only for domains that are not in the domain list called local_domains (which was defined at the start of the configuration). The plus sign before local_domains indicates that it is referring to a named list. Addresses in other domains are passed on to the following routers.

The name of the router driver is dnslookup, and is specified by the driver option. Do not be confused by the fact that the name of this router instance is the same as the name of the driver. The instance name is arbitrary, but the name set in the driver option must be one of the driver modules that is in the Exim binary.

The dnslookup router routes addresses by looking up their domains in the DNS in order to obtain a list of hosts to which the address is routed. If the router succeeds, the address is queued for the remote_smtp transport, as specified by the transport option. If the router does not find the domain in the DNS, no further routers are tried because of the no_more setting, so the address fails and is bounced.

The ignore_target_hosts option specifies a list of IP addresses that are to be entirely ignored. This option is present because a number of cases have been encountered where MX records in the DNS point to host names whose IP addresses are 0.0.0.0 or are in the 127 subnet (typically 127.0.0.1). Completely ignoring these IP addresses causes Exim to fail to route the email address, so it bounces. Otherwise, Exim would log a routing problem, and continue to try to deliver the message periodically until the address timed out.

system_aliases:
  driver = redirect
  allow_fail
  allow_defer
  data = ${lookup{$local_part}lsearch{/etc/aliases}}
# user = exim
  file_transport = address_file
  pipe_transport = address_pipe

Control reaches this and subsequent routers only for addresses in the local domains. This router checks to see whether the local part is defined as an alias in the /etc/aliases file, and if so, redirects it according to the data that it looks up from that file. If no data is found for the local part, the value of the data option is empty, causing the address to be passed to the next router.

/etc/aliases is a conventional name for the system aliases file that is often used. That is why it is referenced by from the default configuration file. However, you can change this by setting SYSTEM_ALIASES_FILE in Local/Makefile before building Exim.

userforward:
  driver = redirect
  check_local_user
# local_part_suffix = +* : -*
# local_part_suffix_optional
  file = $home/.forward
# allow_filter
  no_verify
  no_expn
  check_ancestor
  file_transport = address_file
  pipe_transport = address_pipe
  reply_transport = address_reply

This is the most complicated router in the default configuration. It is another redirection router, but this time it is looking for forwarding data set up by individual users. The check_local_user setting specifies a check that the local part of the address is the login name of a local user. If it is not, the router is skipped. The two commented options that follow check_local_user, namely:

# local_part_suffix = +* : -*
# local_part_suffix_optional

show how you can specify the recognition of local part suffixes. If the first is uncommented, a suffix beginning with either a plus or a minus sign, followed by any sequence of characters, is removed from the local part and placed in the variable $local_part_suffix. The second suffix option specifies that the presence of a suffix in the local part is optional. When a suffix is present, the check for a local login uses the local part with the suffix removed.

When a local user account is found, the file called .forward in the user’s home directory is consulted. If it does not exist, or is empty, the router declines. Otherwise, the contents of .forward are interpreted as redirection data (see chapter 22 for more details).

Traditional .forward files contain just a list of addresses, pipes, or files. Exim supports this by default. However, if allow_filter is set (it is commented out by default), the contents of the file are interpreted as a set of Exim or Sieve filtering instructions, provided the file begins with “#Exim filter” or “#Sieve filter”, respectively. User filtering is discussed in the separate document entitled Exim’s interfaces to mail filtering.

The no_verify and no_expn options mean that this router is skipped when verifying addresses, or when running as a consequence of an SMTP EXPN command. There are two reasons for doing this:

1. Whether or not a local user has a .forward file is not really relevant when checking an address for validity; it makes sense not to waste resources doing unnecessary work.

2. More importantly, when Exim is verifying addresses or handling an EXPN command during an SMTP session, it is running as the Exim user, not as root. The group is the Exim group, and no additional groups are set up. It may therefore not be possible for Exim to read users’ .forward files at this time.

The setting of check_ancestor prevents the router from generating a new address that is the same as any previous address that was redirected. (This works round a problem concerning a bad interaction between aliasing and forwarding – see section 22.5).

The final three option settings specify the transports that are to be used when forwarding generates a direct delivery to a file, or to a pipe, or sets up an auto-reply, respectively. For example, if a .forward file contains

a.nother@elsewhere.example, /home/spqr/archive

the delivery to /home/spqr/archive is done by running the address_file transport.

localuser:
  driver = accept
  check_local_user
# local_part_suffix = +* : -*
# local_part_suffix_optional
  transport = local_delivery

The final router sets up delivery into local mailboxes, provided that the local part is the name of a local login, by accepting the address and assigning it to the local_delivery transport. Otherwise, we have reached the end of the routers, so the address is bounced. The commented suffix settings fulfil the same purpose as they do for the userforward router.

4. Transport configuration

Transports define mechanisms for actually delivering messages. They operate only when referenced from routers, so the order in which they are defined does not matter. The transports section of the configuration starts with

begin transports

One remote transport and four local transports are defined.

remote_smtp:
  driver = smtp

This transport is used for delivering messages over SMTP connections. All its options are defaulted. The list of remote hosts comes from the router.

local_delivery:
  driver = appendfile
  file = /var/mail/$local_part
  delivery_date_add
  envelope_to_add
  return_path_add
# group = mail
# mode = 0660

This appendfile transport is used for local delivery to user mailboxes in traditional BSD mailbox format. By default it runs under the uid and gid of the local user, which requires the sticky bit to be set on the /var/mail directory. Some systems use the alternative approach of running mail deliveries under a particular group instead of using the sticky bit. The commented options show how this can be done.

Exim adds three headers to the message as it delivers it: Delivery-date:, Envelope-to: and Return-path:. This action is requested by the three similarly-named options above.

address_pipe:
  driver = pipe
  return_output

This transport is used for handling deliveries to pipes that are generated by redirection (aliasing or users’ .forward files). The return_output option specifies that any output generated by the pipe is to be returned to the sender.

address_file:
  driver = appendfile
  delivery_date_add
  envelope_to_add
  return_path_add

This transport is used for handling deliveries to files that are generated by redirection. The name of the file is not specified in this instance of appendfile, because it comes from the redirect router.

address_reply:
  driver = autoreply

This transport is used for handling automatic replies generated by users’ filter files.

5. Default retry rule

The retry section of the configuration file contains rules which affect the way Exim retries deliveries that cannot be completed at the first attempt. It is introduced by the line

begin retry

In the default configuration, there is just one rule, which applies to all errors:

*   *   F,2h,15m; G,16h,1h,1.5; F,4d,6h

This causes any temporarily failing address to be retried every 15 minutes for 2 hours, then at intervals starting at one hour and increasing by a factor of 1.5 until 16 hours have passed, then every 6 hours up to 4 days. If an address is not delivered after 4 days of temporary failure, it is bounced.

6. Rewriting configuration

The rewriting section of the configuration, introduced by

begin rewrite

contains rules for rewriting addresses in messages as they arrive. There are no rewriting rules in the default configuration file.

7. Authenticators configuration

The authenticators section of the configuration, introduced by

begin authenticators

defines mechanisms for the use of the SMTP AUTH command. No authenticators are specified in the default configuration file.