Neither TLS provide in such case. Attacker can request ACME cert.
Depends whose DNS you can mess with, but yes! It may be possible to poison DNS records for one e-mail server, but ACME certificate providers like letsencrypt (supposedly) try to do DNS lookups from multiple locations (so hopefully a simple man-in-the-middle attack will not be sufficient), and they do lookups directly from the authoritative DNS servers. This is, of course, not perfect and theoretically suffers from all of the same mitm problems, but it’s more thorough than most mail servers will be and would potentially limit who would be in the position to perform these attacks and get a bogus certificate issued.
With DNSSEC and DANE you are even able to specify which TLS certificate should be used for a service in a TLSA record, and you can protect your A records and your CAA record which should make it much harder to get bogus certificates issued. Of course you need to trust the TLDs in order to trust DNSSEC, but you already do implicitly (as you point out, if you control the TLD you can get whatever certificate you want issued through ACME). The reality right now is that all trust on the web ultimately stems from the TLDs and DNS, but the current situation with CAs introduces several potential attack vectors. The internet is certainly a lot more secure than it used to be even 10 years ago, but I think there’s still a lot of work to be done. DNSSEC, or something like it, would go a long way to solving some of the remaining issues.
Neither TLS provide in such case. Attacker can request ACME cert.
Depends whose DNS you can mess with, but yes! It may be possible to poison DNS records for one e-mail server, but ACME certificate providers like letsencrypt (supposedly) try to do DNS lookups from multiple locations (so hopefully a simple man-in-the-middle attack will not be sufficient), and they do lookups directly from the authoritative DNS servers. This is, of course, not perfect and theoretically suffers from all of the same mitm problems, but it’s more thorough than most mail servers will be and would potentially limit who would be in the position to perform these attacks and get a bogus certificate issued.
With DNSSEC and DANE you are even able to specify which TLS certificate should be used for a service in a TLSA record, and you can protect your A records and your CAA record which should make it much harder to get bogus certificates issued. Of course you need to trust the TLDs in order to trust DNSSEC, but you already do implicitly (as you point out, if you control the TLD you can get whatever certificate you want issued through ACME). The reality right now is that all trust on the web ultimately stems from the TLDs and DNS, but the current situation with CAs introduces several potential attack vectors. The internet is certainly a lot more secure than it used to be even 10 years ago, but I think there’s still a lot of work to be done. DNSSEC, or something like it, would go a long way to solving some of the remaining issues.