Hi Ben,

Thank you for reading the draft and for these comments. Peter
Thomassen has raised related points on this thread, and I am answering
him separately; I have tried to keep the two answers consistent.

> I haven't attempted to fully understand this specification, with all
> of its roles.  However, I do not think the proposed usage of SVCB is
> compliant with RFC 9460.
>
> The SVCB record type is for identifying hosts ("endpoints") that
> provide a service via IP.  It seems like the SVCB type is being used
> here to encode a different kind of relational graph between names.
> That should be accomplished using a different RR type.

I agree. The record at _op.<originator> names no endpoint: it names
the CP issuers that the originator accepts, and the draft already has
to depart from the client behavior of RFC 9460 to use SVCB that way
(Section 8.1). That departure is a sign that SVCB is the wrong fit,
and the next revision will move this record off SVCB.

The candidates I see are a dedicated RR type, as you suggest; PTR,
which Peter suggests; TXT; and URI (RFC 7553). What weighs most in
the choice is who publishes the record. It is the originator, a news
site or a mail domain for instance, which typically uses a commercial
DNS hosting provider, and a new type may not be provisionable there
for a long time. This is why the draft avoided a new type for this
record. I would welcome the group's views on which of the candidates
fits best.

For the records that the issuers publish, the draft does request a
new type, OPKEY, with the TLSA format. We separated it from TLSA
following the precedent of SMIMEA (RFC 8162) and OPENPGPKEY
(RFC 7929), which have types of their own. Peter has asked why not
simply stay with TLSA, and if TLSA under _op is acceptable to the
group, that would be fine with me as well. The next revision will
record this as an open question.

> I also doubt the wisdom of relying on DNSSEC for security in this
> system.  Virtually no end-user systems perform DNSSEC validation
> today, and most domains are not signed.  Given OP's stated goal of
> proving provenance to end-users, this seems like a major problem.

On validation: the verifier here is OP verification software, not the
operating system's stub resolver. The draft requires it either to
validate by itself or to obtain the validation state from a validating
resolver over a secure channel (Section 6.1), and the web media use
case has a verifier that runs in a browser carry a validator of its
own or such a channel (Section 11.1). One form we are looking at is a
browser-based verifier that fetches the records over DNS over HTTPS,
with the DO and CD bits set so that the RRSIGs come along, and
validates them itself. DoH there only carries the data; the verifier
does not rely on the AD bit of the resolver. Whether this is practical
in browsers is one of the things the experiment is meant to observe
(Appendix B).

On signing: DNSSEC is required of the zones of the issuers, not of
every domain. The CP issuers and PA issuers are expected to be few,
and we assume they are flexible operators able to run DNSSEC; theirs
are the zones that hold the records a verifier depends on. An
originator is not asked to sign its zone for the web media use case;
its own zone matters only in particular cases, such as when it pins
its CP issuers or presents its content on hosts outside its own name.
The email use case is currently the exception: it requires the zone
of the sending domain to be signed.

(A small point of terms: the Originator Profile, as this draft uses
it, is about attribution, that is, who originated a piece of content,
rather than its full provenance.)

> Reliance on DNSSEC also seems to be unnecessary, since OP proposes to
> operate its own cryptographic root program.

The root program and the DNS answer different questions. The chain of
signed documents, anchored at a root that the verifier holds, says who
is accredited, and it carries every key that the verifier uses: keys
are never taken from the DNS. The DNS answers only whether a key, a
profile or an annotation is still current, from a zone that its issuer
controls, with an authenticated denial when it is not. A key is
accepted only if it is both in the signed documents and listed in the
DNS.

What the draft does not say is why that signal of current state comes
from the DNS rather than from the root program itself, for instance
through status responses signed by the issuers, or through documents
short-lived enough not to need revocation. That is a fair question,
and the next revision will discuss it. Peter's point on replay bears
on the same comparison.

> Finally, the vast majority of DNSSEC-signed domains are signed by a
> third party DNS operator, not the domain holder per se.  Such a DNS
> operator would have the ability to impersonate their customers in OP,
> in this model.

Because keys are never taken from the DNS, the operator of an issuer's
zone cannot impersonate the issuer or an originator. A digest that it
adds for a key of its own matches no key in a document signed under
the chain, and it cannot produce such a document without the issuer's
private key.

What such an operator can do is narrower, but real. It can hide or
undo a revocation, by restoring a removed Originator Key Record,
removing a Revocation Record, or keeping a withdrawn issuer key
listed; it can make a profile or an annotation appear revoked, by
removing an Originator Key Record or adding a Revocation Record; and
it can make verification fail, by breaking records. The operator of
an originator's zone can likewise remove or change the originator's
pin. The next revision will spell this out in Section 11.1.

best regards,

shigeya

On Sat, Nov 7, 2026, at 05:57, Ben Schwartz wrote:
> I haven't attempted to fully understand this specification, with all
> of its roles.  However, I do not think the proposed usage of SVCB is
> compliant with RFC 9460.
>
> The SVCB record type is for identifying hosts ("endpoints") that
> provide a service via IP.  It seems like the SVCB type is being used
> here to encode a different kind of relational graph between names.
> That should be accomplished using a different RR type.
>
> I also doubt the wisdom of relying on DNSSEC for security in this
> system.  Virtually no end-user systems perform DNSSEC validation
> today, and most domains are not signed.  Given OP's stated goal of
> proving provenance to end-users, this seems like a major problem.
> Reliance on DNSSEC also seems to be unnecessary, since OP proposes to
> operate its own cryptographic root program.  Finally, the vast
> majority of DNSSEC-signed domains are signed by a third party DNS
> operator, not the domain holder per se.  Such a DNS operator would
> have the ability to impersonate their customers in OP, in this model.
>
> --Ben
>
> On Thu, Nov 5, 2026 at 4:38 PM Shigeya Suzuki <[email protected]> wrote:
>>
>> >
>> Hi DNSOP,
>>
>> I submitted draft-shigeya-dns-opkey-00, "Using Secure DNS for Originator 
>> Profile", and would like this group's feedback, both on the DNS design and 
>> on where the work should go.
>>
>> https://urldefense.com/v3/__https://datatracker.ietf.org/doc/draft-shigeya-dns-opkey/__;!!Bt8RZUm9aw!6Elsu0KnurJElB3GWu5ZBuWGK4OpoGeYNqnbkIWXsqmTaLKk9jMkn39sei2fsJVwNsS_wYENnO3Xevg$
>>
>> Originator Profile (OP) lets a verifier of web content, advertising, or 
>> email check who originated it, through signed profiles that issuers attach 
>> to the originator. A verifier needs to know, through an authenticated 
>> channel, that the originator's key is current, that its profile has not been 
>> revoked by its issuer, and that the issuers' own keys are current. The draft 
>> uses DNSSEC-signed DNS for these signals only. The profiles themselves are 
>> not distributed through the DNS, and keys are never taken from the DNS.
>>
>> What the draft defines, all under a new `_op` label:
>>
>> - OPKEY, a new RR type with the TLSA RDATA format, restricted to usage 3, 
>> selector 1, and matching type 1 or 2 (a digest of the SubjectPublicKeyInfo). 
>> It is published at `_issuer-key._op.<issuer>` for issuer keys, and at 
>> `<label>._originator-key._op.<cp-issuer>` for each originator profile. The 
>> presence of a Secure RRset means the profile is not revoked; a Secure 
>> NXDOMAIN or NODATA means it is revoked. Until a type value is allocated, the 
>> same RDATA is published as TLSA at the same names.
>> - An SVCB RRset at `_op.<originator>`, naming the issuers the originator 
>> accepts. An AliasMode record with a TargetName of "." states that the 
>> originator has no profile.
>> - A TXT RRset at `<label>._revocation._op.<pa-issuer>`, whose presence 
>> revokes an annotation on a profile.
>>
>> Only Secure answers are used. Bogus, Insecure, and indeterminate results are 
>> failures, never absence, and wildcards are not allowed at these names. The 
>> draft is intended as Experimental. The experiment is whether a new RR type 
>> can be put into service by the hosting providers and verifiers involved, so 
>> that the TLSA form and the fallback can be dropped.
>>
>> Questions on which I would especially welcome comments:
>>
>> 1. A new RR type with the TLSA format, following SMIMEA and OPENPGPKEY, 
>> rather than TLSA itself at `_op` names. The draft does this to keep the 
>> semantics of the two types apart.
>> 2. Using a Secure NXDOMAIN or NODATA as the revocation signal for a profile.
>> 3. The use of SVCB, including AliasMode with ".", to state an originator's 
>> accepted issuers, or that it has none.
>> 4. TXT for annotation revocation.
>>
>> I would also like advice on the right home for this work. I understand that 
>> the call for DNSOP agenda items for IETF 127 closed on October 31, so DNS 
>> DISPATCH will most likely have to wait until the next meeting. If a slot 
>> does open up at IETF 127, though, I would be glad to take it.
>>
>> Related documents (published on October 27th):
>>
>> - Originator Profile Framework - Architectural Overview: 
>> https://urldefense.com/v3/__https://docs.originator-profile.org/en/aov/__;!!Bt8RZUm9aw!6Elsu0KnurJElB3GWu5ZBuWGK4OpoGeYNqnbkIWXsqmTaLKk9jMkn39sei2fsJVwNsS_wYENQDfi6Fg$
>> - Originator Profile Framework - Core Specification: 
>> https://urldefense.com/v3/__https://docs.originator-profile.org/en/core/__;!!Bt8RZUm9aw!6Elsu0KnurJElB3GWu5ZBuWGK4OpoGeYNqnbkIWXsqmTaLKk9jMkn39sei2fsJVwNsS_wYENwxcciNg$
>> - Originator Profile Framework - Web Media Use Case Specification: 
>> https://urldefense.com/v3/__https://docs.originator-profile.org/en/web-media/__;!!Bt8RZUm9aw!6Elsu0KnurJElB3GWu5ZBuWGK4OpoGeYNqnbkIWXsqmTaLKk9jMkn39sei2fsJVwNsS_wYENu1jLz2M$
>> - Originator Profile Framework - Email Use Case Specification: 
>> https://urldefense.com/v3/__https://docs.originator-profile.org/en/email/__;!!Bt8RZUm9aw!6Elsu0KnurJElB3GWu5ZBuWGK4OpoGeYNqnbkIWXsqmTaLKk9jMkn39sei2fsJVwNsS_wYENa0liGHI$
>>
>> thanks,
>>
>> Shigeya Suzuki, Ph.D
>> Keio University / WIDE Project / Originator Profile CIP
>> _______________________________________________
>> DNSOP mailing list -- [email protected]
>> To unsubscribe send an email to [email protected]
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to