Re: [DNSOP] New Version Notification for draft-adpkja-dnsop-special-names-problem-00.txt

[Stuart Cheshire <[e-mail]>]

Hi Joe, Peter, Alain,

I have some feedback on draft-adpkja-dnsop-special-names-problem-00.

There seems to have been lots of recent discussion regarding confusion surrounding RFC 6761. I’m a little surprised by this, since I didn’t think RFC 6761 was unclear. But then as co-author, that’s my failing. I thought it clearly stated our intention, and I thought the IETF Last Call scrutiny should have identified any ambiguity, but apparently not.

>    In recent years, using the last label of a domain name (aka TLD) as
>    switch to indicate how to treat name resolution has been experimented
>    using the framework of [RFC6761].

Not only the last label. The last _n_ labels are also commonly used as the algorithmic method for determining what names qualify as special. Some examples:

10.in-addr.arpa.
168.192.in-addr.arpa.
example.com.
example.net.
example.org.

>    This switch practice is not explicitly documented anywhere.

That was the intention of the seven-question list: to have protocol specifications explicitly document their “switch practice” -- i.e. how their special names are to be unambiguously recognized, and what software should do having recognized one of them.

>    Answers to these seven questions provide guidance to the
>    corresponding seven audiences on how to handle a special-use domain
>    name once it has been reserved by inclusion in the Registry.
>    However, they are inadequate for making the determination whether a
>    particular domain name qualifies as being special in the first place.

You have it backwards. The seven questions were not “what to expect *after* this special-use registration is done”; they were the justifications *why* the special-use registration should be granted, required in a document demonstrating that it (a) provides a result that the community judges to be good, and (b) the aforementioned good result cannot reasonably be achieved better another way.

>    The lack of a more elegant way to specify a name resolution protocol
>    in (for example) a URI amounts to an architectural oversight.

I’m not sure I agree that there *is* a more elegant way to achieve the desired effect. Unless you intend to actually describe some hypothetical “more elegant way” of doing this, I suggest simply removing this unsupported implication from the document.

>    It should also be noted that there are other choice than using the
>    most significant label for a protocol switch.  In particular, a
>    proposal to move those protocol switches under a specific top level
>    domain has been discussed (.ALT).  If that architecture choice is
>    made, some of the questions listed in the sections bellow (sic) would
>    become moot.

RFC 6761 is not, and has never been, about TLDs. It is about special behaviours, and how to trigger them. As such, whether the protocol switch is a TLD, or SLD, or 3LD, or whatever, has no bearing at all on “the questions listed in the sections below”. The same concerns would apply equally, regardless of how the algorithmic determination of “specialness” is to be performed. RFC 6761 states that the algorithmic determination is *typically* done by looking for a certain suffix (i.e., one or more labels). I deliberately did not want RFC 6761 to over-specify this, so as to leave the door open for someone smarter than me who might come along in the future and propose some new and valid algorithmic determinant that I had not anticipated. Clearly a document stating that any name containing any label starting with an “x” is now “special” ought to be thrown out by the IESG as unreasonable, but there could be other valid ideas beyond my current imagination.

>    Note: [RFC6761] mentions the reserved names could be any label in any
>    random string, not just the rightmost one (or ones).

Can you cite the actual text that says that? I couldn’t find what particular text you’re paraphrasing there.

>    What does it mean to have a "non-DNS" entry in the registry
>    described above?

The answers to the seven questions, and other text in the associated protocol specification, are supposed to tell you what it means.

>    Are applications supposed to check that registry to know what to do?

No.

>    Can/Should applications do this check dynamically?

No.

>    What if an application makes this dynamic check and realizes the
>    name contains a switch it does not know how to treat?

Not applicable. The closest equivalent to the run-time mechanism you’re hinting at is something that already exists: The global DNS. Putting in authoritative NXDOMAIN results for all appropriate special-use names is the best way to achieve the result you want.

>    it is not clear who is responsible for carrying out an
>    evaluation.  A document which requests additions to the Registry
>    might be performed by the IESG, by the IAB, by the DNSOP working
>    group, by an ad-hoc working group, by expert review or any
>    combination of those approaches.  [RFC6761] provides no direction.

RFC 6761 says:

   When IANA receives a request to record a new "Special-Use Domain
   Name", it should verify, in consultation with the IESG, that the IETF
   "Standards Action" or "IESG Approval" document [RFC5226] includes the
   required "Domain Name Reservation Considerations" section stating how
   the special meaning of this name affects the behavior of hardware,
   software, and humans in the seven categories.  If IANA and the IESG
   determine that special handling of this "Special-Use Domain Name" is
   appropriate, IANA should record the Special-Use Domain Name, and a
   reference to the specification that documents it, in the registry.

I confess I has assumed that IANA would designate some person with the expertise and experience to evaluate whether “... special handling of this "Special-Use Domain Name" is appropriate ...”, much as requests for TCP and UDP ports are evaluated. That was my mistake. I should have been explicit about the intended review process.

>    For example, is large scale prior deployment an acceptable criteria?

Large scale prior deployment should *not* be required -- that would just make squatting a necessary prerequisite for getting a special use name assigned. RFC 6761 was intended to provide a legitimate path for proposals to be evaluated on technical merit, rather than de facto squatting.

>    Going back to the previous point of prior usage of the protocol, in
>    the case of LOCALHOST, LOCAL and ONION, those particular domain names
>    were already in use by a substantial population of end-users at the
>    time they were requested to be added to the Registry.  Rightly or
>    not, the practical cost of a transition was argued as a justification
>    for their inclusion in the registry.

No. The justification for having an entry in the registry was to facilitate the desired special behaviour. The particular choice of what name went in the registry was influenced by existing usage, but the justification for having the entry itself was motivated by the desire for special behaviour. If the IETF process went a bit more smoothly, the IETF would have more participation in the choice of name *before* it becomes too late to easily change that.

As I read it, draft-adpkja-dnsop-special-names-problem seems to focus far too intently on the issue names. (But then, that is what ICANN is in the business of selling, so maybe it’s not surprising.) The substance of RFC 6761 is about enabling special behaviours, and using special names to trigger those special behaviours is merely a means to the end. What is interesting and important, and worthwhile for the IETF to support, is the special behaviours, not the names.

Stuart Cheshire