3 ms·
I agree with all points here. Most anti-DNSSEC talk is just silly, especially if you ignore DANE. The two are separate protocols with separate use cases. Some
by jeroenhd 5y ago
I agree with all points here. Most anti-DNSSEC talk is just silly, especially if you ignore DANE. The two are separate protocols with separate use cases.
Some of the arguments brought up ("but Libya could pown bit.ly") are complete nonsense because as the TLD owners, they already control EVERYTHING, including who the rightful owner is. With DV certificates like Let's Encrypt the whole idea of CA's somehow stopping the regime from a takeover becomes completely silly. If you want control not to fall in the hands of a foreign military regime, don't host your domain with said military regime.
All this talk reminds me about the strong resistance against HTTPS in browsers. Certificate validation was broken in many browsers just like validation of DNSSEC responses is broken today.
The outages due to DNSSEC Slack reported were all because of buggy software, with buggy software written by Amazon in their last failed rollout. Route 53 was incredibly late to the party, only introducing DNSSEC less than a year ago, so it makes sense that their implementation still contains bugs. I fail to see why this would be DNSSEC's fault.
European TLDs (.no, .cz, .nl, .se) seem to be ahead of the curve on DNSSEC with adoption rates already above 50% and most American TLDs have basically no DNSSEC adoption at all. It'll take a while for DNS to be verifiable with no interest from the major TLDs and no suggested alternatives.
- tptacek 5y agoVirtually every outage is because of buggy software. When we evaluate the resilience of protocols, what we're really talking about is how difficult it is to build reliable implementations of those protocols. DNSSEC has a particularly clumsy design (it's heavily encumbered by decisions made by government contractors in the mid-1990s, when it was widely believed that DNS servers simply wouldn't be powerful enough to "do" cryptography). I'd have a problem with DNSSEC even without the wretched design it has now (it would remain a government-controlled PKI), but the 1990s design is most of what animates me about it.
- jeroenhd 5y agoOur current TLS PKI is already government controlled PKI. Several governments control their own certificate authorities. Perhaps it's better this way because at least it's not just the American government, bit it's hardly an argument against the change. I don't really see what the "1990s design" argument is all about. If we'd design such a system today, we'd get the same system with maybe some more JSON or a blockchain of sorts, centralised around a few profit-oriented companies. Look at DoH (the version with and without the intermediate proxy server) to see what I mean. You're right that the idea of assumptions about CPU power for DNS servers are outdated, but with modern software that's hardly a problem. DNS servers already solve this problem in every existing DNSSEC server I know.
- tptacek 5y agoYes, we agree, DNS servers are fully capable of implementing modern cryptography. The problem is that the protocol is designed to assume they can't, and it's a series of creaky workarounds for that fact --- much of DNSSEC's jankiness comes from the protocol requirement for offline-only signers.
- jeroenhd 5y agoBut that's the thing, common server software already collapses that part of the protocol into a single operation for the query itself. To enable DNSSEC for most systems, you just tell your nameserver to sign the zone and copy the necessary keys over to your registrar. That's two commands in PowerDNS, one to sign the records and one to fetch the keys. In BIND it requires some manual key generation and config, but that's still hardly a challenge for someone who can manage their own DNS zones. CDS/CDNSKEY can automate the communication with the TLD provider, but that's completely optional. The real jankiness is hidden away in the bespoke implementations on the backend that are designed to have their own complex DNS logic. Those systems don't need to abide by the weird offline-signing DNSSEC standards at all, because there's no reason for them to be compatible with other servers unless you're glueing your own code to an existing server. The offline signing quirk also actually helps you if you use a master/slave DNS setup because you don't need to set up a complicated multi-keyed DNS system to have many resolvers serve your records; one place signs it, and the other servers just forward the responses. The client does nothing more complex than validating the chain of trust like we're already doing for every other part of our modern connections, and all the server needs to do is respond with the right records (be it signed responses or signed errors).