moore@betelgeuse.cs.utk.edu (Keith Moore) (11/03/89)
In article <43@looney.twinsun.com> eggert@twinsun.com (Paul Eggert) writes: >We're a UUCP site connected to the Internet via MX forwarding. We often get >mail with addresses that have long combinations of domain names and UUCP names >separated by !, e.g. something like: > > ucla-cs!usc!cs.utexas.edu!purdue!tut.cis.ohio-state.edu!dec!olsen > >typically in response to Usenet postings. Is it OK to remove all leading >components of such a path up to the last component that is a domain name, e.g. >rewrite the above as the following? > > tut.cis.ohio-state.edu!dec!olsen > This isn't a good idea, because (a) the legal syntax for UUCP addresses is rather limited, and (b) the email world is bigger than just Internet or UUCP. Hypothetical example: Say someone on our DECnet sends a piece of mail through cs.utk.edu (utkcs in the UUCP world) to a UUCP site. The return address would look something like ...!utkcs!foo.decnet!user . Now, foo.decnet is not an Internet domain, but it looks like one to sendmail rewrite rules. (One could argue that the address should be ...!cs.utk.edu!foo.decnet!user instead, but the former seems more consistent with UUCP syntax, and therefore hopefully more compatible with mail software in the UUCP world.) Now of course it would be much better if cs.utk.edu had used foo.decnet's domain name instead, and our DECnet gateways do this if the decnet node actually has a domain name. However, many of the machines connected to our DECnet are not under our administrative control, and we are not at liberty to assign domain names for them. (A "trapdoor" MX record for every machine in our DECnet is a possibility, but still under discussion.) The point is, with mail systems which use relative addressing such as UUCP (or DECnet mail, for that matter), each node in the path is free to interpret the next hop in the address however it chooses. Normally this is handled consistently from one node to another, but special hacks for gatewayed mail are all too common. Such hacks depend on the notion that any replies or error messages should traverse the inverse path that the original message took. In networks with relative addressing, this is a reasonable assumption. (For robustness, error messages should *always* take the inverse path of the original message, RFC1123 notwithstanding.) Back to the question of short-circuit routing within sendmail. It might be reasonable to short-circuit a UUCP-style route to a domain if one were certain that the domain was in the nameservers with either an A or an MX record. But even this can cause problems, because the mail exchanger for foo.bar might rewrite mail destined for user@foo.bar as path!to!foo!dot!bar!foo.bar!user. If one of the intervening nodes tried to short-circuit the path back to foo.bar!user, the mail would get in a loop between that node and foo.bar's mail exchanger. So in general it seems wisest not to rewrite UUCP paths for reasons of efficiency. Keith Moore Internet: moore@cs.utk.edu University of Tenn. CS Dept. BITNET: moore@utkvx 107 Ayres Hall, UT Campus UT Decnet: utkcs::moore Knoxville Tennessee 37996-1301 Telephone: +1 615 974 0822
lear@GENBANK.BIO.NET (Eliot Lear) (11/12/89)
The solution to Keith Moore's objection to short circuiting is to keep a list of valid top level domains, and only short circuit on those. -- Eliot Lear [lear@net.bio.net]