Bruce Barnes wrote:
Mailservers are technically required by RFC1123 5.2.17, to
accept mail to domain literals, for any of their assigned IP
addresses.
Yes, that's the RCPT TO case, but for SPF the relevant problem
is MAIL FROM an address literal. The RfC 2821 syntax allows
this (4.1.1.2): A reverse path minus routing info is simply a
mailbox in angle brackets.
A mailbox is local part @ domain. A domain is either an FQDN
or an address literal:
| Domain = (sub-domain 1*("." sub-domain)) / address-literal
But as Raymond said, address literals in MAIL FROM are very
unusual, and maybe they are even "verboten". Unfortunately I
don't find anything in the RfC to support this interpretation.
Receivers are generally free to reject any MAIL FROM if they
don't like it. SPF uses this to reject a forged MAIL FROM if
the result is FAIL. It's also possible to reject a MAIL FROM
for other reasons (RHSBLs, syntax errors, etc.)
But normally receivers have to accept legitimate mail to their
postmaster mailbox. Otherwise they are "RfC-ignorants". Now
if a mailer with IP 1,2.3.4 sends MAiL
FROM:<user(_at_)[1(_dot_)2(_dot_)3(_dot_)4]>
to the postmaster of a system supporting SPF, then this will
always FAIL with a reason "malformed domain".
But [1.2.3.4] isn't a "malformed domain", it's a "Domain" as
specified in RfC 2821. Are systems supporting SPF potential
"RfC-ignorants" ? Is this a bug in the actual SPF draft ?
Bye, Frank