--On Sunday, March 30, 2008 7:05 PM -0700 "Carl S. Gutekunst"
<csg(_at_)alameth(_dot_)org> wrote:
Requiring an MX record for every mail host certainly would be
a paradigm shift. The historical use of the MX record is to
identify which hosts are responsible for the incoming E-mail
for a domain, not identifying the specific hosts that are able
to accept E-mail. There are many situations where a sending
MTA coerces the RHS of the bounce address to its own hostname
to ensure that bounces are returned specifically to the
sending host. This might be done for list management, capacity
management, internal corporate organization, security, or a
number of other reasons. If MX records were required, then
they'd have to be added for every one one of these sending
hosts. I know of many organizations for which this would be a
significant administrative burden.
Hmm. This example, which I agree with but hadn't though to
introduce into this discussion, is probably another example of
why we don't want to try to tamper with things now by simply
changing the defaults (or by distinguishing IPv6 from IPv4). If
one were designing the architecture de novo today, in today's
environment, there would be a strong case for "no MX default",
but the sort of situation Carl mentions would also make a strong
case for separate "MX-forward-path" and "MX-reverse-path"
(bounces) records.
That issue is similar to Frank Ellermann's oft-repeated
complaint that (paraphrasing) we weakened the architecture
vis-a-vis today's needs when reverse-paths stopped looking like
<@nearest-relay: user(_at_)presumed-origin> rather than depending on
MX records to determine the mail routing for rejection/bounce
messages and thereby treating them the same way forward-path
messages are treated.
But those sorts of considerations are important if and when we
decide to go design a replacement email system. In the
environment we have, it is yet another reason why the notion of
an address-based MX default has to apply to address records of
all types.
john