The manualroute router is so-called because it provides a way of manually routing an address according to its domain. It is mainly used when you want to route addresses to remote hosts according to your own rules, bypassing the normal DNS routing that looks up MX records. However, manualroute can also route to local transports, a facility that may be useful if you want to save messages for dial-in hosts in local files.
The manualroute router compares a list of domain patterns with the domain it is trying to route. If there is no match, the router declines. Each pattern has associated with it a list of hosts and some other optional data, which may include a transport. The combination of a pattern and its data is called a ``routing rule''. For patterns that do not have an associated transport, the generic transport option must specify a transport, unless the router is being used purely for verification (see verify_only).
In the case of verification, matching the domain pattern is sufficient for the router to accept the address. When actually routing an address for delivery, an address that matches a domain pattern is queued for the associated transport. If the transport is not a local one, a host list must be associated with the pattern; IP addresses are looked up for the hosts, and these are passed to the transport along with the mail address. For local transports, a host list is optional. If it is present, it is passed in $host as a single text string.
The list of routing rules can be provided as an inline string in route_list, or the data can be obtained by looking up the domain in a file or database by setting route_data. Only one of these settings may appear in any one instance of manualroute. The format of routing rules is described below, following the list of private options.
The private options for the manualroute router are as follows:
This option controls what happens when manualroute tries to find an IP address for a host, and the host does not exist. The option can be set to one of
freeze decline defer pass fail
The default assumes that this state is a serious configuration error. The difference between ``pass'' and ``decline'' is that the former forces the address to be passed to the next router (or the router defined by pass_router), overriding no_more, whereas the latter passes the address to the next router only if more is true.
This option applies only to a definite ``does not exist'' state; if a host lookup gets a temporary error, delivery is deferred unless the generic pass_on_timeout option is set.
If this option is set, the order of the items in a host list in a routing rule is randomized each time it is used. This can be used to do crude load sharing. However, there is a complication when a message has more than one address that is routed by the same rule. Without randomization, each such address ends up with an identical host list, and so they are all eligible for batching and sending in a single SMTP transaction. When the host order is randomized by the router, the addresses won't all end up with the same host list, and so they will not be batched in the same way.
This may be desirable if you want a number of different delivery processes to be used. However, if you do want all addresses that route to the same hosts to be batched together, but still want to use the hosts in a random order, you should not set hosts_randomize on this router. Instead, arrange to use a special smtp transport, and set hosts_randomize on the transport.
If this option is set, it must expand to yield the data part of a routing rule. Typically, the expansion string includes a lookup based on the domain. For example:
route_data = ${lookup{$domain}dbm{/etc/routes/}}
If the expansion is forced to fail, or the result is an empty string, the router declines. Other kinds of expansion failure cause delivery to be deferred.
This string is a list of routing rules, in the form defined below. Note that, unlike most string lists, the items are separated by semicolons. This is so that they may contain colon-separated host lists.
Addresses with the same domain are normally routed by the manualroute router to the same list of hosts. However, this cannot be presumed, because the router options and pre-conditions may refer to the local part of the address. By default, therefore, Exim routes each address in a message independently. DNS servers run caches, so repeated DNS lookups are not normally expensive, and in any case, personal messages rarely have more than a few recipients.
If you are running mailing lists with large numbers of subscribers at the same domain, and you are using a manualroute router which is independent of the local part, you can set same_domain_copy_routing to bypass repeated DNS lookups for identical domains in one message. In this case, when manualroute routes an address to a remote transport, any other unrouted addresses in the message that have the same domain are automatically given the same routing without processing them independently. However, this is only done if headers_add and headers_remove are unset.
The value of route_list is a string consisting of a sequence of routing rules, separated by semicolons. If a semicolon is needed in a rule, it can be entered as two semicolons. Empty rules are ignored. The format of each rule is
<domain pattern> <host list> <options>
The following example contains two rules, each with a simple domain pattern and no options:
route_list = \ dict.ref.example mail-1.ref.example:mail-2.ref.example ; \ thes.ref.example mail-3.ref.example:mail-4.ref.example
The three parts of a rule are separated by white space. The pattern and host list can be enclosed in quotes if necessary, and if they are, the usual quoting rules apply. Each rule in a route_list must start with a single domain pattern, which is the only mandatory item in the rule. The pattern is in the same format as one item in a domain list (see section 10.6), that is, it may be wildcarded or a regular expression, or a file or database lookup (with semicolons doubled, because of the use of semicolon as a separator in a route_list).
The rules in route_list are searched in order until one of the patterns matches the domain that is being routed. The host list and options are then used as described below. If there is no match, the router declines. When route_list is set, route_data must not be set.
The use of route_list is convenient when there are only a small number of routing rules. For larger numbers, it is easier to use a file or database to hold the routing information, and use the route_data option instead. Most commonly, the value of route_data is a string that contains an expansion lookup. For example, suppose we place two routing rules in a file like this:
dict.ref.example: mail-1.ref.example:mail-2.ref.example thes.ref.example: mail-3.ref.example:mail-4.ref.example
This data can be accessed by setting
route_data = ${lookup{$domain}lsearch{/the/file/name}}
Failure of the lookup results in an empty string, causing the router to decline. However, you do not have to use a lookup in route_data. The only requirement is that the result of expanding the string is a single host list, possibly followed by options, separated by white space. The host list can be enclosed in quotes if necessary.
A host list, whether obtained via route_data or route_list, is always separately expanded before use. If the expansion fails, the router declines. The result of the expansion must be a colon-separated list of host names and/or IP addresses. IP addresses in host lists are not enclosed in brackets.
If the host list was obtained from a route_list item, the following variables are set during the expansion of the host list:
If the domain was matched against a regular expression, the numeric variables $1, $2, etc. may be set.
$0 is always set to the entire domain.
$1 is also set when partial matching is done in a file lookup.
If the pattern that matched the domain was a lookup item, the data that was looked up is available in the expansion variable $value.
Options are a sequence of words, but in practice no more than two are ever present. One of the words can be the name of a transport, and this overrides the transport option on the router for this particular routing rule only. The other word (if present) specifies how the IP addresses of the hosts named in the host list are to be found when routing to a remote transport:
byname: use gethostbyname(). This is the default way of finding host addresses if no options are given.
bydns: look up address records for the host(s) in the DNS; fail if there are none.
If no IP address for a host can be found, what happens is controlled by the host_find_failed option.
When an address is routed to a local transport, IP addresses are not looked up. The host list is passed to the transport in the $host variable.
In some of the examples that follow, the presence of the remote_smtp transport, as defined in the default configuration file, is assumed:
The manualroute router can be used to forward all external mail to a smart host. If you have set up, in the main part of the configuration, a named domain list that contains your local domains, for example,
domainlist local_domains = my.domain.example
you can arrange for all other domains to be routed to a smart host by making your first router something like this:
smart_route: driver = manualroute domains = !+local_domains transport = remote_smtp route_list = * smarthost.ref.example
This causes all non-local addresses to be sent to the single host smarthost.ref.example. If a colon-separated list of smart hosts is given, they are tried in order. Another way of configuring the same thing is this:
smart_route: driver = manualroute transport = remote_smtp route_list = !+local_domains smarthost.ref.example
There is no difference in behaviour between these two routers as they stand. However, they behave differently if no_more is added to them. In the first example, the router is skipped if the domain does not match the domains pre-condition; the following router is always tried. If the router runs, it always matches the domain and so can never decline. Therefore, no_more would have no effect. In the second case, the router is never skipped; it always runs. However, if it doesn't match the domain, it declines. In this case no_more would prevent subsequent routers from running.
A mail hub is a host which receives mail for a number of domains via MX records in the DNS and delivers it via its own private routing mechanism. Often the final destinations are behind a firewall, with the mail hub being the one machine that can connect to machines both inside and outside the firewall. The manualroute router is usually used on a mail hub to route incoming messages to the correct hosts. For a small number of domains, the routing can be inline, using the route_list option, but for a larger number a file or database lookup is easier to manage.
If the domain names are in fact the names of the machines to which the mail is to be sent by the mail hub, the configuration can be quite simple. For example,
hub_route: driver = manualroute transport = remote_smtp route_list = *.rhodes.tvs.example $domain
This configuration routes domains that match *.rhodes.tvs.example to hosts whose names are the same as the mail domains. A similar approach can be taken if the host name can be obtained from the domain name by a string manipulation that the expansion facilities can handle. Otherwise, a lookup based on the domain can be used to find the host:
through_firewall:
driver = manualroute
transport = remote_smtp
route_data = ${lookup {$domain} cdb {/internal/host/routes}}
The result of the lookup must be the name or IP address of the host (or hosts) to which the address is to be routed. If the lookup fails, the route data is empty, causing the router to decline. The address then passes to the next router.
You can use manualroute to deliver messages to pipes or files in batched SMTP format for onward transportation by some other means. This is one way of storing mail for a dial-up host when it is not connected. The route list entry can be as simple as a single domain name in a configuration like this:
save_in_file: driver = manualroute transport = batchsmtp_appendfile route_list = saved.domain.example
though often a pattern is used to pick up more than one domain. If there are several domains or groups of domains with different transport requirements, different transports can be listed in the routing information:
save_in_file:
driver = manualroute
route_list = \
*.saved.domain1.example $domain batch_appendfile; \
*.saved.domain2.example \
${lookup{$domain}dbm{/domain2/hosts}{$value}fail} \
batch_pipe
The first of these just passes the domain in the $host variable, which doesn't achieve much (since it is also in $domain), but the second does a file lookup to find a value to pass, causing the router to decline to handle the address if the lookup fails.
Routing mail directly to UUCP software is a specific case of the use of manualroute in a gateway to another mail environment. This is an example of one way it can be done:
# Transport
uucp:
driver = pipe
user = nobody
command = /usr/local/bin/uux -r - \
${substr_-5:$host}!rmail ${local_part}
return_fail_output = true
# Router
uucphost:
transport = uucp
driver = manualroute
route_data = \
${lookup{$domain}lsearch{/usr/local/exim/uucphosts}}
The file /usr/local/exim/uucphosts contains entries like
darksite.ethereal.example: darksite.UUCP
It can be set up more simply without adding and removing ``.UUCP'' but this way makes clear the distinction between the domain name darksite.ethereal.example and the UUCP host name darksite.