On 20 Jul 2018, at 2:15 (-0400), dln wrote:
The advice to use the MX record to 'redirect' email for
client-domain.net to
mail.server.com (for example) will work happily.
Indeed, as it has worked for decades.
However (referring to the OP's use case), won't the client (say a
Thunderbird user)
No, MUAs do not generally chase MX records. MX records exist to instruct
MTAs on where to pass mail for non-local domains. MUAs use explicit
names for MSAs, either by manual configuration or via automated config
schemes like ACAP.
be presented with the LE certificate for server.com and
not one from his own "client-domain"?
This is not a problem. In the MTA-MTA case the sender has resolved a MX
record to a name with an A record, which should be a valid subject name
for the receiver's certificate. In the MUA-MSA case the client has an
explicitly configured name which should be a valid subject name for the
server's certificate OR should resolve to such a name via a CNAME
record.
Such an appearance may cause confusion/distrust?
I've been managing SSL/TLS-capable mail systems for as long as that has
been possible, using multiple versions of multiple MTA/MSA packages with
clients of all sorts and I have never seen a client complain about a
valid MSA certificate whose names include the one that the client has
been configured to use. I have never seen a sending MTA allow
certificate name or trust chain issues to persistently impede delivery
and only rarely to persistently force fallback to sending in the clear.
The reason these theoretical problems are not actual problems in the
real world is that for many years the most common mode of using TLS for
SMTP was with self-signed certificates (because that is in fact adequate
without trustworthy DNS) and mail submission was commonly done over port
25 just like transport. Because of that history, interoperability
concerns have kept most mail software from being strict about
certificate validity by default.
(and perhaps it should!)
Probably not in most cases for mail transport (i.e. governed by MX
records) but it is already true that many client TLS implementations are
not tolerant of some sorts of subject name mismatch for MSAs, which is
as it should be.
--
Bill Cole
b...@scconsult.com or billc...@apache.org
(AKA @grumpybozo and many *@billmail.scconsult.com addresses)
Currently Seeking Steadier Work: https://linkedin.com/in/billcole