> On 28 Oct 2026, at 14:05, David Blacka via Datatracker <[email protected]> 
> wrote:
> 
> Document: draft-ietf-dnsop-delegation-mgmt-via-ddns
> Title: Automating DNS Delegation Management via DDNS
> Reviewer: David Blacka
> Review result: Ready with Issues

David,

Thanks for a careful review. We will address all of it in -03 and
plan to publish that shortly after the current IETF silent period.

Replies below in the order of your review.

On the companion drafts: yes, they are meant to travel together, but
keystate, announce-scanner and signaling-zone-owner-intent are only
informative references here, and this draft describes what it needs
from them itself.

* Introduction / RFC 9859. 

Agreed, "borrows heavily" understates it.

When Peter and I (Johan) were working on the early versions of what
became RFC9859 I think we had examples of both NOTIFY and UPDATE
"schemes" but we decided to focus only on NOTIFY to keep that document
simpler.

But the consequences are becoming clear: to avoid repeating a lot of
the content of RFC9859 the very tight relationship between the two
documents must be made very explicit. Several of your observations and
questions further down are explained and described fully in RFC9859
and that absolutely makes reading this document without its
precursor... challenging.

Suggested text for-03: "The reader is also expected to be very
familiar with [RFC9859], in particular the DSYNC resource record and
the way the target is located (its Section 3), which this document
reuses unchanged for UPDATE."

* Section 3.

Fair, it reads as a tutorial and we agree that it may be
removed.

* Section 4.

Agreed. The protocol is the existing one; the receiver is
not. Suggested text for -03: "This document does not define a new
update protocol. The child's update is an ordinary DNS UPDATE
[RFC2136], SIG(0)-signed [RFC3007][RFC2931], exactly as specified
there on the wire. What differs from a traditional dynamic-update
deployment is the receiver, which applies policy and need not apply
the change to a running zone."

That said, of all the things that seem to confuse readers the primary
one is the concept of using the DNS protocol to communicate with
something that may be, but does not have to be, a nameserver. We
struggled with that throughout the entire process with the DSYNC
protocol that resulted in RFC9859 and the struggle is likely not over.

* Section 5.

Agreed, your wording is better. We will use this verbatim
in -03: "This document defines a new DSYNC scheme, "UPDATE" (rather
than "NOTIFY"), but otherwise uses the same method as generalized
NOTIFY to determine the QNAME to look up (e.g. {child}._dsync.{parent}
and the variants as described in Section 3 of [RFC9859])."

The old sentence also covered the RRtype; the examples use ANY and we
need to keep one clause saying so.

* Section 7.1.

Agreed, simpler. Suggested text for -03: "A single UPDATE
MUST contain either delegation changes (NS, glue, DS) or KEY changes,
not both. A bootstrap UPDATE is signed by the key it introduces, which
the receiver does not yet trust, so it cannot also authorize
delegation changes."  [JOHAN: confirm tdns never combines them, and
the RCODE for a mixed UPDATE; REFUSED fits.]

* Section 9.1.

"Multi-query convergence" was shorthand for the
unsigned-child bootstrap method (Section 9.4.1.3): the receiver
fetches the child KEY RRset over TCP from several vantage points and
times and accepts it only if consistent, as in RFC 8078. It is not
defined where first used and we realize that term doesn't help at all
and should be replaced. Suggested text for -03: "or, for an unsigned
child, the KEY RRset checked as described in Section 9.4.1.3".

* Section 9.4.1.1.

Yes, your reading is the intended one. -03 will say so
here and in 9.4.1.2: the receiver validates the KEY RRset with DNSSEC;
if the validated RRset contains a KEY RR identical to the key in the
bootstrap UPDATE, that key is validated and may be promoted to trusted;
otherwise it is not.

* Section 9.4.1.3.

A child finds its registrar's receiver through the
per-child DSYNC form in Example 2 of Section 5: the registry publishes
{child}._dsync.{parent} with the registrar's UPDATE Receiver as target,
and the child looks up only its own name. That is not obvious from
9.4.1.3, and you are right that the registrar/EPP model applies to the
whole process, not just unsigned bootstrap. -03 will move the paragraph
to Section 7, cross-reference Example 2, and state plainly that how the
registry maintains those records, and the EPP mapping, are out of scope.

* Section 9.5.1.

Good question. The apex copy is not needed for
validation; the receiver validates the _sig0key copy. The apex KEY is
needed for the nameserver operator to learn the key to publish;
"HSYNCPARAM pubkey" is a mechanism to instruct the nameserver operator
to "please re-publish the KEY found at the zone apex under the
sig0key... name if possible". I.e. the same intent as used for CDS
re-publication in RFC9615 but made explicit.

-03 will define at-ns as publication under _sig0key... and describe
the apex copy and HSYNCPARAM as ways of conveying the key to the
nameserver operators. That also removes the apparent overlap with
at-apex in the example.

* Section 9.6.

Yes, the only difference is which key signs. A bootstrap
UPDATE is signed by the key it introduces (not yet trusted); any later
UPDATE, including a key roll, is signed by a key the receiver already
trusts. We will make this more explicit in -03.

* Nits: will fix the comma in Section 2; the Section 3.1 sentence goes
with Section 3; the "Operator coordination via HSYNCPARAM" paragraph in
9.4.1.2 becomes a proper subsection heading.

Again, thanks for a very complete and thoughtful review.

Regards,
Johan

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

Reply via email to