[Top] [All Lists]

Re: I-D Action:draft-klensin-rfc2821bis-10.txt

2008-04-17 03:54:14

--On Thursday, 17 April, 2008 11:25 +0100 Tony Finch
<dot(_at_)dotat(_dot_)at> wrote:

On Wed, 16 Apr 2008, Willie Gillespie wrote:
Tony Finch wrote:
On Wed, 16 Apr 2008, Ned Freed wrote:
I doubt that it makes sense to accept email from
test(_at_)ipv6(_dot_)l(_dot_)google(_dot_)com on a system that can only
communicate with IPv4 addresses.

I foresee some company setting up an IPv6 to IPv4 e-mail
relay for individuals or other companies that only have IPv6

I agree with John that this is likely to happen.

For an IPv4-only receiving system, would this appear as an
e-mail from test(_at_)ipv6(_dot_)l(_dot_)google(_dot_)com (even though it 
comes over
the IPv4 link)?  At that point, would it make sense to accept
the message?

The sender's relay has to be two-way, so the IPv6-only site's
MXs would have to refer to the relay's IPv4 address as well as
the site's own IPv6 address. Then the sender's email addresses
can be verified successfully by IPv4-only sites.

I don't think John is right to expect IPv4-only recipient
sites to obtain a 4-to-6 SMTP relay service any time soon.
IPv6 sites will have to deal with the interop burden until v4
is the minority.

For example, there are likely to be problems if
test(_at_)ipv6(_dot_)l(_dot_)google(_dot_)com AAAA-only addresses leak out, 
and these
are likely to be worse than the problems that A-only addresses
like apache(_at_)www(_dot_)example(_dot_)com have.

Whatever I said, I didn't intend it to convey the expectation
that the responsibility would need to lie with the IPv4-only
recipient site.    I have come to believe that it is appropriate
for a receiving site to do at least superficial verification of
the possibility of delivering an NDN before accepting a message
for delivery for which an NDN might be necessary (i.e., for
which delivery cannot be assured while the SMTP connection is
still open).  I think that testing the reverse path and making a
decision to not accept the message if the test fails is entirely
consistent with the "take responsibility" language of 2821 (and

I think that implies that, for the near future at least, if
test(_at_)ipv6(_dot_)l(_dot_)google(_dot_)com wants to have a reasonable 
that mail it sends will be accepted by mail servers running in
IPv4-only  environments, then it (the sender) must expect to
either be dual-stack (and advertise the IPv4 address too) or to
have a lower-priority MX advertised that will accept IPv4

And, unless I misread your note, I think that puts us in violent