Can you articulate precisely the problem you believe this will solve? From my perspective it seems like it’s just making the system more fragile and harder to fix since DNSSEC requires OS updates to improve, while not meaningfully preventing state-level attacks.
> The current number of CAs does not prevent state-level attacks either.
Right, so the question is why we should put a huge amount of effort into implementing and operating a system which doesn't make significant improvements.
> DNSSEC works I don’t really get that point.
It's mostly a layering question: if a new cryptographic algorithm is released or a problem with an old one comes out, browsers can update very quickly. Updating the operating systems and network hardware which implement DNSSEC takes considerably longer. DNSSEC lingered on 90s crypto for ages, key rotations were put off for years, etc. because everyone in this space has to be extremely conservative. That has security implications as well as delaying most attempts to improve performance or usability.
Similarly, browsers can have extensive UI and custom validation logic for HTTPS. A lot of that information isn't present if you use DNSSEC without implementing your own resolver, so you get generic error messages and you don't get control over the policies set by your network administrator. This is especially interesting both as a risk if you don't trust your ISP or for dealing with compromises — if I compromise your DNS server and publish DNSSEC records with a long TTL, your users are at risk until you can get every ISP with a copy to purge the cached records ahead of schedule.
All of those issues can be improved but it's not clear that there's enough benefit to be worthwhile.
> The “root of trust” problem is hard to solve. I kinda hear the DANE guys’ argument, I’d rather trust one authority than a thousand.
This is the best argument for DNSSEC but it's not clear to me how much difference it makes in practice when you're comparing the still nascent DNSSEC adoption to modern TLS + certificate transparency which also catches spoofing and is far more widely implemented.
Kinda off-topic, but I wonder if the adoption of DNS-over-HTTPS will eventually solve the ossification problem you're referring to by moving DNS resolution to the application level.
It can definitely help since you're removing the network operator from the critical path. Large enterprises and ISPs are, not without reason, very conservative about breaking legacy clients but a browser vendor only has to worry about their own software in the release they ship DoH in (with some caveats they've addressed about internal split-view DNS, etc.) so they don't have to deal with complaints if, say, including an extra header breaks 5% of old IoT devices which haven't had an update in a decade.
