Hi Joe,

On 6/12/25 04:22, Brian Dickson wrote:
The core advice:

    This document therefore specifies that parent-side entities MUST
    ensure that the updates indicated by CDS/CDNSKEY and CSYNC record
    sets are consistent across all of the child's authoritative
    nameservers, before taking any action based on these records.

is, in general, not actionable. The full set of authority servers for a zone are 
frequently not available from a single vantage point, since a single nameserver address 
often maps to many different individual servers which may or may not serve consistent 
information (e.g. when an individual NS target is deployed using anycast in one or more 
address families). Depending on how you count them, most nameservers are anycast, so I 
think it could be said that this is not a niche observation. The revised advice might 
boil down to "instead of just looking for an answer as a stub resolver would, look 
at a random two out of 100,000 possible authoritative servers" and I'm not convinced 
that is much of an improvement.


I think a balanced reading of the advice, which might depend on favorable 
interpretation of semantics and definitions, can be actionable and an 
improvement.

I think this boils down to looking at signatures as “proof of possession” and 
also as an authoritative indication of intent.

The (IMHO) reasonable expectation regarding these specific record types is that 
they SHOULD be consistent across an anycast set. This reading of the 
recommendations reduces the required queries to one per unique server 
name/address, without weakening the logic or the security model(s) involved.

The idea is really to do a plausibility check, not to rule out things that 
can't be ruled out. In other words, it's not about being sure that all is fine; 
it's about acting reasonably when there is an indication of a problem.

I suspect that there are many cases where errors could happen because the 
synchronization was not properly done (see the cases in the various 
Appendixes), and not because some anycast nodes are lagging.

Of course, that can happen, too. In combination with the advice in 
draft-ietf-dnsop-generalized-notify (in RFC ed queue) that operators not send 
notifications to the parent "until all publicly visible copies of the zone are 
updated", the parent's consistency check actually is suitable to (mostly) trigger 
when an operator actually has made a mistake.

With respect to multi-signer configurations, I think the right way to solve 
that is to sync CDS and CDNSKEY between participating signers just as the 
corresponding DNSKEY RRs are synced. This seems like a solution that is more 
effectively aligned with the problem; to put it another way, multi-signer 
implementations would be more robust if they didn't have to make assumptions 
about whether the polling advice in this document was followed or not.

I completely agree, but it is likely there will be failures around that sync, 
when one provider doesn't not properly honor the state machine and proceeds 
early to the next step. The draft acknowledges such (and other) scenarios in 
A.3.

Historically, there has been no shortage of DNSSEC-induced outages from early 
moves during (manual?) key rollovers and such, and I'm pretty sure that 
multi-provider synchronization implementations will not be immune to such bugs. 
These situations should lead to the update not being applied, instead of the 
wrong update being applied and more outages happening (and another round of 
avoidable DNSSEC reputational loss ensuing).

On 6/12/25 08:10, Joe Abley wrote:
In this case the duct tape is only being applied in places that are feeling the 
consequence of forgetting the contract; the consequences of ignoring the 
contract are largely self-inflicted and not shared with other innocent 
bystanders.

I'm not sure I agree with that: if a domain goes bogus because of a bad DS 
update, then users of that domain are also affected, even though they are 
innocent bystanders.

This is why I think this proposal overall does no great harm and I think it's 
ok to publish.

Ah great! Thank you for reading (and commenting!) anyway. :-p

Best,
Peter

_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to