We were talking about resolvers setting the DO bit, not end nodes setting the DO bit.
As for the armies of dumb IoT devices, we actually require them to do diffie-hellman with decent-sized keys these days, and there are _really_ good reasons to do that. I don't know of any that currently do DNSSEC validation, but I don't think it would be quite as onerous as you suggest, for two reasons. First, they wouldn't do it very often: they only need to do it when looking up a host they want to talk to. Second, even when they are looking up a host they want to talk to, mostly the answer will be the same as last time. If nothing has expired, then we don't have to re-verify the signatures. This might open up some small opportunities for attack in cases of key revocation, and maybe I'm missing other opportunities, but although it's worth making it "not free" to attack constrained IoT devices, they are not generally extremely high-value targets either. So best here is the enemy of good enough. The biggest problem with these devices is that a lot of them don't know what time it is, so they have to rely on some external source of truth to determine whether the valid result they have is also timely. I'm not seriously suggesting that we need IoT devices to start validating—at the moment I think that we generally address the authentication/authorization problem at the application layer, so a cache-poisoning attack on such a device mostly acts as a DoS. But I'm saying that if for some reason in the future we decide that we need to do this, it's not at all clear to me that it would be prohibitively expensive. As for my misinterpretation of the meaning of the DO bit, my understanding and experience is that if you set it, you get back a bunch more stuff. Which is what I thought we were talking about. I do not claim to be an expert on this topic. :) > On 3 Nov 2026, at 13:21, Jim Reid <[email protected]> wrote: > > > >> On 31 Oct 2026, at 12:30, Ted Lemon <[email protected]> wrote: >> >> As for the DO bit, it should always be set, because we should always be >> validating. > > No. There are plenty of use cases where DNSSEC validation isn't needed or > wanted: armies of dumb IoT devices that have feeble CPUs and die when their > batteries run flat, stub resolvers using encrypted transports that are > considered "good enough", enterprise networks that can't/won't drink the > DNSSEC Kool-Aid, etc. Foisting (unwelcome) validation that's doomed to fail > in these scenarios isn't going to have a happy ending or lead to the > universal deployment of DNSSEC. > > You're also redefining the meaning of the DO bit. It was introduced to signal > to a server that the (stub) resolver understood DNSSEC related RRtypes => it > wouldn't fall over if those RRtypes were included in a response. RFC3225 says > nothing about validation at all. If this matters, I suppose we can always > burn another EDNS header bit to say "I [don't] do DNSSEC validation". > > BTW early versions of BIND9 always set the DO bit. Even when the name server > wasn't configured with a trust anchor or wasn't linked with a crypto library > to do validation. This was unwise IMO. Let's not continue setting the DO bit > no matter what. Just use it for its intended purpose. > >> Asking people not to set it if they aren't validating is planning for >> failure. > > Maybe, maybe not. If someone isn't validating - for whatever reason - why > bother setting the DO bit on their outbound queries? What's gained by doing > that? How would flooding resolvers with unwanted RRSIGs and so on encourage > DNS admins to turn on DNSSEC? > _______________________________________________ DNSOP mailing list -- [email protected] To unsubscribe send an email to [email protected]
