mp@mit-vax.UUCP (06/03/83)
I just noticed another example where sites forward netnews but don't forward all types of mail. Our site is one. But it isn't intentional. Now that news 2.10 is here, people are replying to paths with internet addresses in them. We just received a message via uucp that's bound for "eagle!mhuxt!mhuxi!mhuxa!houxm!hocda!spanky!burl!duke!sam@fau.UUCP". Now; our mailer listens to both '@' and '!', unfortunately, it gives '@' precedence and it's going to bomb out because "fau.UUCP" isn't a valid Chaosnet or Arpanet host. So this person's mail is going to be dropped on the floor in a few minutes (maybe it'll even be mailed back to him; who knows?). I don't have time to bring up nmail here, especially since it would need to be integrated with the already hacked-up delivermail we currently have. And I don't think the system administrators would let me install a replacement for delivermail, even if they did think I knew what I was doing. Suggestions will be appreciated. Mark
jim@uw-beaver.UUCP (06/03/83)
The problem of ambiguous mail addresses again rears its ugly head. My solution has been to have /bin/rmail give precedence to the '!'. This way, mail arriving via uucp is parsed as uucp mail. To me this makes the most sense, but I guess I am alone in thinking this way because all other implementations I know of unconditionally give precedence to '@'. Note that mail arriving via Arpanet, ethernet, or generated locally is still parsed with precedence to '@'. On our system, the address given as an example by mp@mit-vax: eagle!mhuxt!mhuxi!mhuxa!houxm!hocda!spanky!burl!duke!sam@fau.UUCP is translated by /bin/rmail into mhuxt!mhuxi!mhuxa!houxm!hocda!spanky!burl!duke!sam@fau.UUCP@eagle.uucp and is then sent on to eagle by delivermail. This requires only one small change to rmail, and none at all to any other mailers, including delivermail. When I first implemented this, it seemed so obvious that I wondered what was wrong with it. Since that time we have had much less dropped mail than before, and I still can't figure out what is wrong with the scheme.
mcg@tekecs.UUCP (06/04/83)
The problem with the scheme suggested by uw-beaver!jim is that
it doesn't always correctly turn the address around! To recap,
the suggested fix was to treat '!' as having higher precedence
that '@'. Unfortunately, in addresses of the form:
site1!site2!site3!person@site4.UUCP
the '@site.UUCP' could have been added in between sites 1 and 2,
or between 2 and 3, as well as before 1 or after 3. Thus, all of
the following are possible correct addresses for the above:
site4!site1!site2!site3!person = ("site1!site2!site3!person")@site4
site1!site4!site2!site3!person = site1!(("site2!site3!person")@site4)
site1!site2!site4!site3!person = site1!site2!(("site3!person")@site4)
site1!site2!site3!site4!person = site1!site2!site3!("person@site4")
You rightly ask "what's going on here?" The answer is: some sites seem
to be converting parts of addresses that come through them into
Internet-style addresses. Since Internet addresses do not recognize
the '!' as anything other than plain text (i.e. part of the destination
address), they simply append their pseudo-Internet "site.UUCP" on the
end. The next site down the road, being a standard old UNIX mailer,
and not realizing that '@' is special, treats the whole thing like
a string, and just adds its own 'site!' at the front.
In degenerate cases (and I have seen quite a few), you get addresses
like:
From: foo!bar!biff!person@DECVAX.UUCP!ucbvax@CBOSG.UUCP
with random sprinklings of '!' and '@' all over the place.
What to do? Pressure all the sites which are currently generating
bogus Internet addresses to generate real UUCP addresses when
talking to UUCP. How? I sure don't know...
S. McGeadyptw@vaxine.UUCP (P. Tucker Withington) (06/07/83)
With regard to mixed up uucp and internet addresses in news from path. Is this a real problem? When I reply to news articles I throw away all the explicit routing and let "pathalias" figure it out for me (actually 2.10 news does this). I believe "pathalias" will find a much better route than the reverse of the semi-random news propagation path. Along these same lines, is anyone working on a dynamic routing algorithm? Something to the effect of uucp files always being labelled with their FINAL destination, and each uucp machine deciding how to best send it on its next hop (with some provision to prevent loops, obviously). Seems ARPA has this pretty well knocked, couldn't IMP routing algorithm be adapted? Or do most really important mail routes occur in one hop anyways? --Tucker (ptw@vaxine.UUCP)
jim@uw-beaver.UUCP (06/08/83)
Something I have considered doing in the way of dynamic routing is to have rmail look at the path by which each message arrives, and if the path reveals the existence of a link previously unknown to the routing table, add it to the table. This isn't really dynamic routing but rather dynamic routing table maintenance. This would be easy to implement in its simplest form, but there are lots of problems with it. For example, it gives no information about the quality of the link, and it makes the assumption that the link is bi-directional. If a bad link gets added to the table, there is no automatic way to remove it.