Looking for What’s Not There
potaroo.net
potaroo.net
Aren't most TLS certs domain validated anyways? The current TLS practice already puts full trust in registrars AND CAs.
Also, complaints that registrars can quietly change out sites without anyone noticing, is pretending that we cannot watch the state of the DNS over time. Don't we have certificate transparency already to solve a similar problem in the WebPKI?
But I'm not sure certificate transparency will work as well for DNS as it does for TLS. At least not without some changes (that are a good idea for many other reasons) like permitting SRV for HTTP.
Huston suggests a new reason to deploy DNSSEC: it might stop a specific DNS denial of service attack.
(The attack is trivial: you can just have a botnet make up random DNS names and query them, saturating the authority servers.)
DNSSEC is designed around "offline signers", meaning that as originally designed, it can't synthesize new DNS responses for queries. But it also wants "authenticated denial", which just means you can cryptographically trust "no such record" responses. That's problematic, because you can't predict the nonexistent names people will ask you about, and you can't sign responses on demand.
The way DNSSEC originally solved this problem was with "NSEC", which is a record that states that no valid name exists between a lexicographically ordered pair of names. You pre-generate the covering NSEC ranges for your zone, and your servers can now authenticate the non-existence of any record.
(This addresses Huston's DOS attack by allowing DNS caches to NXDOMAIN queries they've never seen before, because they've cached NSEC responses from authority servers that state that there's no point in forwarding the queries).
There's an obvious problem here: attackers can also now enumerate precisely all the hosts in your zone, which is an old feature of DNS (AXFR) that network security teams have been frantically disabling for 2 decades. There's a whole line of DNSSEC revisions that try to account for this fact, most notably NSEC3, which uses a weak password hash to scramble the ranges (pretty unsuccessfully). But the best-known mode of deployment for DNSSEC now involves something called "whitelies", in which the DNSSEC server now simply becomes an online signer and generates new ranges on the fly.
Which breaks Huston's proposal.
Mostly I think it's funny that anyone would try to motivate DNSSEC deployment as a mitigation to a single species of DOS attack, especially when DNSSEC itself is (fairly or not) itself notorious as a DOS amplifier (DNSSEC responses are huge, so there's a pretty big packet magnification effect in its deployment).
Why doesn't DNS simply accept the fact that either:
1) your hosts are public
2) DNS is a stateful protocol and different clients _may_ see different responses by server? (e.g. return false denials for unauthorized clients)
Caching is what makes DNS usable. If you do this in a way that doesn't break caching, it's not so simple.
For the record, the HN submission guidelines encourage changing the title if it's misleading, but not if it's vague.