Linux Network Administrator Guide, Second Edition

It is inevitable that when you bring together a number of different systems and a number of clever people, they will seek ways to interconnect the differing systems so they are capable of internetworking. Consequently, there are a number of different mail gateways that are able to link two different email systems together so that mail may be forwarded from one to another. Addressing is the critical question when linking two systems. We won't look at the gateways themselves in any detail, but let's take a look at some of the addressing complications that may arise when gateways of this sort are used.

Consider mixing the UUCP style bang-path notation and RFC-822. These two types of addressing don't mix too well. Assume there is an address of domainA!user@domainB. It is not clear whether the @ sign takes precedence over the path, or vice versa: do we have to send the message to domainB, which mails it to domainA!user, or should it be sent to domainA, which forwards it to user@domainB?

Addresses that mix different types of address operators are called hybrid addresses. The most common type, which we just illustrated, is usually resolved by giving the @ sign precedence over the path. In domainA!user@domainB, this means sending the message to domainB first.

However, there is a way to specify routes in RFC-822 conformant ways: ‹@domainA,@domainB:user@domainC› denotes the address of user on domainC, where domainC is to be reached through domainA and domainB (in that order). This type of address is frequently called a source routed address. It's not a good idea to rely on this behavior, as revisions to the RFCs describing mail routing recommend that source routing in a mail address be ignored and instead an attempt should be made to deliver directly to the remote destination.

Then there is the % address operator: user%domainB@domainA is first sent to domainA, which expands the rightmost (in this case, the only) percent sign to an @ sign. The address is now user@domainB, and the mailer happily forwards your message to domainB, which delivers it to user. This type of address is sometimes referred to as 'Ye Olde ARPAnet Kludge,' and its use is discouraged.

There are some implications to using these different types of addressing that will be described throughout the following sections. In an RFC-822 environment, you should avoid using anything other than absolute addresses, such as [email protected].

How Does Mail Routing Work?

The process of directing a message to the recipient's host is called routing. Apart from finding a path from the sending site to the destination, it involves error checking and may involve speed and cost optimization.

There is a big difference between the way a UUCP site handles routing and the way an Internet site does. On the Internet, the main job of directing data to the recipient host (once it is known by its IP address) is done by the IP networking layer, while in the UUCP zone, the route has to be supplied by the user or generated by the mail transfer agent.

Mail Routing on the Internet

On the Internet, the destination host's configuration determines whether any specific mail routing is performed. The default is to deliver the message to the destination by first determining what host the message should be sent to and then delivering it directly to that host. Most Internet sites want to direct all inbound mail to a highly available mail server that is capable of handling all this traffic and have it distribute the mail locally. To announce this service, the site publishes a so-called MX record for its local domain in its DNS database. MX stands for Mail Exchanger and basically states that the server host is willing to act as a mail forwarder for all mail addresses in the domain. MX records can also be used to handle traffic for hosts that are not connected to the Internet themselves, like UUCP networks or FidoNet hosts that must have their mail passed through a gateway.

MX records are always assigned a preference. This is a positive integer. If several mail exchangers exist for one host, the mail transport agent will try to transfer the message to the exchanger with the lowest preference value, and only if this fails will it try a host with a higher value. If the local host is itself a mail exchanger for the destination address, it is allowed to forward messages only to MX hosts with a lower preference than its own; this is a safe way of avoiding mail loops. If there is no MX record for a domain, or no MX records left that are suitable, the mail transport agent is permitted to see if the domain has an IP address associated with it and attempt delivery directly to that host.

Suppose that an organization, say Foobar, Inc., wants all its mail handled by its machine mailhub. It will then have MX records like this in the DNS database:

green.foobar.com. IN MX 5 mailhub.foobar.com.

This announces mailhub.foobar.com as a mail exchanger for green.foobar.com with a preference of 5. A host that wishes to deliver a message to [email protected] checks DNS and finds the MX record pointing at mailhub. If there's no MX with a preference smaller than 5, the message is delivered to mailhub, which then dispatches it to green.

This is a very simple description of how MX records work. For more information on mail routing on the Internet, refer to RFC-821, RFC-974, and RFC-1123.

Mail Routing in the UUCP World

Mail routing on UUCP networks is much more complicated than on the Internet because the transport software does not perform any routing itself. In earlier times, all mail had to be addressed using bang paths. Bang paths specified a list of hosts through which to forward the message, separated by exclamation marks and followed by the user's name. To address a letter to a user called Janet on a machine named moria, you would use the path eek!swim!moria!janet. This would send the mail from your host to eek, from there on to swim, and finally to moria.

The obvious drawback of this technique is that it requires you to remember much more about network topology, fast links, etc. than Internet routing requires. Much worse than that, changes in the network topology - like links being deleted or hosts being removed - may cause messages to fail simply because you aren't aware of the change. And finally, in case you move to a different place, you will most likely have to update all these routes.

One thing, however, that made the use of source routing necessary was the presence of ambiguous hostnames. For instance, assume there are two sites named moria, one in the U.S. and one in France. Which site does moria!janet refer to now? This can be made clear by specifying what path to reach moria through.

The first step in disambiguating hostnames was the founding of the UUCP Mapping Project. It is located at Rutgers University and registers all official UUCP hostnames, along with information on their UUCP neighbors and their geographic location, making sure no hostname is used twice. The information gathered by the Mapping Project is published as the Usenet Maps, which are distributed regularly through Usenet. A typical system entry in a map (after removing the comments) looks like this: