[ih] MX and WKS records

[Craig Partridge]

You rang? :-)

Pulling on memories here, and let's hope they are accurate.

On Mon, Feb 3, 2025 at 1:45 PM John Levine via Internet-history <
internet-history at elists.isoc.org> wrote:

> Over on another list we're having yet another rerun of the argument about
> what
> to do when you want to send mail to a domain and it has no MX records or
> it has
> dodgy MX records and when you fall back to A or AAAA records. RFC 974
> answered
> these questions thirty-nine years ago but a few mysteries remain. I figure
> that
> since the author of 974 is on this list, it's a good a place to ask as any.
>
> If there's no MX records, you pretend there's an MX pointing at the domain
> itself and use the A record. Was that intended as a temporary transition
> until
> everyone published MX, or a permanent part of the spec? (As we all know,
> there
> is nothing so permanent as a temporary transition hack.)
>

That was intended as permanent.  The notion was that to force everyone to
have an MX RR to start getting email was overreach at the time (and
presumably into the future).  MX RRs were to make email routing better, and
we wanted folks to use them, but we didn't want to force it. As I recall,
and here memory is a bit hazy, Jon Postel also liked this approach as it
met his robustness vision (maximized chances that email would go through)
and Rick Adams (who at the time was running one of the biggest email
servers, the gateway to USENET) felt it ought to be a transitional hack.
Another hazy recollection is that, at the moment I was writing RFC 974, it
was not clear exactly how useful wildcards would be in the DNS --
wildcards, of course, made it much easier to avoid having to create an MX
RR for every hostname.

Note RFC 974 does say that IF an MX RR exists but none of the RRs can be
used (long side discussion), you may NOT just connect to the host.  (I.e.
if you use MX RRs, they are definitive.  But if you haven't set them up,
we'll just try to directly connect).


> It also says to do a WKS lookup to see what services a host provides, and
> in particular whether it does SMTP.  I started running mail servers in the
> early 1990s and I don't ever remember publishing or looking for WKS.  Did
> anyone ever use them?


The issue WKS solved was, at the time I was writing RFC974, it was pre-MIME
and there were several multimedia email prototypes on the Internet. If you
saw the demos (as I had, and Jon P., and Paul Mockapetris and Rudi Nedved
had*), you realized that multimedia mail was coming and we wanted MX RRs to
just work (as they seemed like the right general architecture) rather than
forcing the creation and deployment of say MMX RRs (DNS ossification was
already beginning to be a visible risk due to issues in updating the list
of root servers).  So the notion was, use WKS.  Someone with a better
crystal ball might have envisioned the transition to MIME (which was
derived from the CMU Andrew multimedia system already operating) combined
with 8-bit SMTP (which was politically painful, but turned out to be easy
in practice), but we missed.  In practice, everyone used SMTP (and skipped
WKS) and since we never developed a parallel multimedia mail parallel to
SMTP, WKS was never needed.

Hoping that helps!

Craig

*PS: While I wrote RFC 974 (and there's a backstory, see the PPS), I was
very green at the time and so Jon Postel put in place a team to support
me.  As memory serves it was Paul Mockapetris, Jon Crowcrowft and Rudi
Nedved and I'm probably forgetting a couple of key folks (apologies!).  All
bad ideas in RFC 974 are mine --  some of the good ones are likely theirs.

PPS: I ended up writing RFC 974 because I submitted an RFC draft on how to
use MD and MF RRs correctly (the DNS spec being unclear and I had been
charged with making CSNET domain name compliant).  As part of that process,
I discovered a race condition that could only be solved by forcing
email-related DNS queries to use TCP.  Jon DID NOT WANT the Internet's
major app being forced to use TCP rather than UDP for DNS queries.
Accordingly, he said approximately "you found this bug, you fix it, and
here are some folks to help you find a fix."


>
> Fallback to A turned out to have the unfortunate side effect that mail to
> domains with no mail service hammer on the host(s) at the A record until
> they
> give up, typically days later. In 2015 we finally published RFC 7505, the
> Null
> MX band-aid, the reverse of WKS saying a host does not accept mail as
> opposed
> to WKS which said that it did.
>
> R's,
> John
> --
> Internet-history mailing list
> Internet-history at elists.isoc.org
> https://elists.isoc.org/mailman/listinfo/internet-history
>


-- 
*****
Craig Partridge's email account for professional society activities and
mailing lists.