On Mon, Oct 13, 2025 at 02:17:56PM +1100, Viktor Dukhovni via mailop wrote:
[...]
> With STARTTLS, message transmission is fully *encrypted*, but the remote
> end is often not *authenticated*. That said, your key observation is that
> SMTP is often not end-to-end, and the *client* is then not necessarily
> related to the sender.

The client's IP address is already authentication. After all, it is commonly
used as part of authorisation. It's a pretty strong signal in its own right.

> Therefore, in the context of MTA-to-MTA (port 25) email relayng, a client
> certificate could perhaps be used as a lookup key for client reputation,
> that could be more robust than an IP address. [...]

"Could perhaps" and then "could" again, all in the same sentence? This is
the kind of weasel wording found on homeopathic remedies. Talking of things
which are expensive, full of bafflegab, and don't work...

You gave a link to an Internet-Draft. It proposes adding a new domain name
field to the TLS client certificate to be used by DANE. DANE is little more
than a mechanism to store certificate identities in DNS and sidestep
certificate authorities, i.e. it's a workaround to avoid paying hundreds of
dollars for SSL certificates in the bad old days before Let's Encrypt came
along.

So let's look at how this would work in practice.

Imagine that a spammer has just bought canter-and-siegel.ai with a stolen
credit card. They self-sign a certificate (or maybe even get one from Let's
Encrypt to lend a bit more credibility) and put its identity in the
DNSSEC-signed zone, which of course they can do because they control the
domain. This is all hosted on the AWS free tier or wherever else will accept
the stolen credit card for long enough to do a decent spam run.

Meanwhile, you receive a connection from the spammer's server, which is
successfully authenticated as being from canter-and-siegel.ai. How does this
bring things any closer to *authorisation*?

Sure, you could refuse to accept mail not from that domain, but this
disadvantages non-spammy providers such as Google -- stop laughing at the
back -- who send mail on behalf of many domains. You'd have to fall back on
some other mechanism such as authenticating the sender domain of individual
messages, and oh look, DKIM already exists.

In other words, the client certificate has not contributed anything useful
to the eventual accept/reject decision by the server, and was a complete
waste of time.

This is a FUSSP and should be treated as such.

_______________________________________________
mailop mailing list
[email protected]
https://list.mailop.org/listinfo/mailop

Reply via email to