Oh, apparently IMAP has a separate IMAPS port so that kinda works. I forgot that. Still DV is meh for non-https usages.
Conversation
STARTTLS services can still be used while enforcing TLS with valid Web PKI certificates. DANE is nice but relying on DNS with DNSSEC has other threats (registrars, etc.) and doesn't give cert transparency. DV certainly sucks but CT is very valuable and Web PKI is being improved.
2
The right way to use DANE is pinning keys. I don't think it makes much sense to use it to pin certificates. As a key pinning mechanism it's perfectly suited to be an additional security check and can live happily alongside Web PKI. CT is a nice check on registrars, not just CAs.
1
If you're thinking about DANE for HTTPS, yes PKIX-EE(1) is reasonably fit for purpose, if/when (some day) the browsers actually implement DANE.
For SMTP, web PKI is not a good fit. See section 1.3 of RFC7672. MTA-STS is a kludge.
WebPKI btw., is just as vulnerable to registrars
1
1
> For SMTP, web PKI is not a good fit. See section 1.3 of RFC7672. MTA-STS is a kludge.
It works fine. MTA-STS is certainly a kludge since it should just be the only option. People who want to run their own mail server and still receive mail can set up a valid certificate.
1
> WebPKI btw., is just as vulnerable to registrars
CT with enforcement of SCTs at least provides an audit trail to detect a compromise or malicious behavior. There isn't a choice between using Web PKI or DANE since they work fine together. Don't see why someone wouldn't do both.
1
CT works much better in theory than in practice. Only the big players have the resources to audit the logs and determine whether any unauthorised certificates have been issued. It is a pretty high cost to avoid forgery of certificates for Google, and at best works after the fact.
1
It hardly takes a lot of resources, especially for a clean / simple setup. I would expect that the complexity of the setup reflects the resources of the organization and if there's an over-engineered sprawling mess that's a whole separate problem. How are they going to use DANE?
1
If someone can set up DANE, they're in a position to audit CT logs for certificates issued with the wrong key. As I mentioned earlier, I don't see why these things are presented as alternatives when they work fine alongside each other. Doing both is the most practical / secure.
3
DANE is utterly trivial to setup. If anything it's harder for big orgs that have large legacy DNS setups not amenable to DNSSEC (ie Google).
1
I don't disagree with you about DANE being a better approach and easy to setup. I just think that Web PKI has both practical value (dealing with anyone not using DANE) and actual value via CT which can also be considered to place a check on registrars too, not just CAs.
I think we've talked about this before and you seem to be misunderstanding my position this time. I don't see a reason not to set up both since it's easy. Can reuse the key when getting a new certificate with LE and now there's actually a verifiable audit trail for certificates.
2
MTA-STS has a stupid design but it's the way of forcing TLS with authentication by several major providers. I reused a web server on another machine anyway. More of them have TLSA records for us to use than the amount verifying DANE but there's still gap that this helps to fill.
1
Show replies
This is addressed in 7672 Section 1.3. For SMTP one can't avoid relying on DNS security, without horrible kludges like dupicating MX RRsets in a policy served via HTTPS (blech...). And with Let's Encrypt, control of DNS gives you control of certificate issuance. So you can't win.
Webpki is worthless for SMTP without MTA-STS or PKIX-EE because otherwise using TLS at all is opportunistic & accepting untrusted key is still far better than sending in clear. And MTA-STS is a standards-hostile Rube Goldberg machine dependent on web infrastructure.
2
TLSA PKIX-EE is an ok way to do both but I'd rather not have legacy CA mess as a possible failure point when you already know client supports DANE and can bypass the need for CAs entirely.