Re: Last Call: <draft-ietf-weirds-rdap-query-15.txt> (Registration Data

A few “comebacks” in-line.

From:  "Hollenbeck, Scott" <shollenbeck(_at_)verisign(_dot_)com>
Date:  Tuesday, October 21, 2014 at 8:51
To:  Edward Lewis <edward(_dot_)lewis(_at_)icann(_dot_)org>, 
"ietf(_at_)ietf(_dot_)org" <ietf(_at_)ietf(_dot_)org>,
"iesg(_at_)ietf(_dot_)org" <iesg(_at_)ietf(_dot_)org>
Cc:  "weirds(_at_)ietf(_dot_)org" <weirds(_at_)ietf(_dot_)org>, Andy Newton 
<andy(_at_)arin(_dot_)net>
Subject:  RE: Last Call: <draft-ietf-weirds-rdap-query-15.txt> (Registration
Data Access Protocol Query Format) to Proposed Standard

> >  -----Original Message-----
> >  From: Edward Lewis [mailto:edward(_dot_)lewis(_at_)icann(_dot_)org]
> >  Sent: Friday, October 17, 2014 12:02 PM
> >  To: ietf(_at_)ietf(_dot_)org; iesg(_at_)ietf(_dot_)org
> >  Cc: weirds(_at_)ietf(_dot_)org; Hollenbeck, Scott; Andy Newton
> >  Subject: Re: Last Call: <draft-ietf-weirds-rdap-query-15.txt>
> >  (Registration Data Access Protocol Query Format) to Proposed Standard
> Thanks for the feedback, Ed.
>
> >  A couple of comments
> >  
> >  Comment 1
> >  
> >  Section 1.1
> >  
> >  Nit: the detail on REST and RESTful seems out of place in a list of
> >  acronyms and abbreviations.
> The "detail" is no more than a sentence or two. I'm inclined to leave that nit
> alone.
>
> >  Comment 2
> >  
> >  Section 2
> >  
> >  ##   Additionally, resource management, provisioning and update
> >  functions
> >  ##   are out of scope for this document.  Registries have various and
> >  ##   divergent methods covering these functions, and it is unlikely a
> >  ##   uniform approach for these functions will ever be possible.
> >  
> >  Nit: should be "it is unlikely that a uniform approach ... is needed
> >  for
> >  interoperability."
> Agreed.
>
> >  Comment 3
> >  
> >  
> >  Section 3.2
> >  
> >  Confusion: domain/domains and entity/entities plural forms are defined
> >  externally to the list, but nameserver/nameservers plural form is
> >  not.  I think that is just an inconstency.
> I don't understand this comment. The text I'm looking at in Section 3.2
> includes search path segment definitions for "domains", "nameservers", and
> "entities". The singular forms used for lookups are defined in Section 3.1.
> What am I missing?
That in the bullets you have:
   o  'domains': ...
   o  'nameservers': ...
   o  'entities': …
And in prose you have:
For the domain and entity objects described in this document the plural
object forms are "domains" and "entities”.
Merely the mismatch in the “number of parameters” (- the prose doesn’t
mention nameservers)
> >  The word ''match'' is used referring to a FQDN.  Does this mean to
> >  imply
> >  ''match'' as defined in RFC 1034 (where '*' matches anything)?
> Matching is described in Section 4.1.
The you have a “forward reference here - using “match” in 3.2 (and in the
context of a “fully-qualified domain name), then later in the document
defining it in 4.1.

I’ve seen this problem when implementers read the RFCs.  They might read 3.2
and assume match as defined in the DNS.  Match is such an overloaded term it
deserves qualification (or an adjective).

> >  And what is throwing me is that this begins talking about testing to
> >  see
> >  if something exists and then seems to start talking about finding
> >  inexact
> >  matches (the plurals).
> >  
> >  It reads as if the first paragraph doesn't relate to the rest of the
> >  section.
> Hmm, OK. That first paragraph (which describes using the HTTP HEAD method to
> check for object existence) could be moved to Section 3.1.
>
> >  Comment 4
> >  
> >  Section 3.2.1 and 3.2.2
> >  
> >  Confusion: Both 3.2.1 and 3.2.2 cover searching for name servers.  Why
> >  are
> >  two approaches presented?
> >  
> >     http://example.com/rdap/nameservers?name=ns1.example*.com
> >  and
> >     http://example.com/rdap/domains?nsLdhName=ns1.example*.com
> >  
> >  I suppose this isn't a major issue, but seems awkward.
> 3.2.1 describes searching for *domains*. 3.2.2 describes searching for *name
> servers*.
Then maybe the example “ns1.example*.com” ought to be just "example*.com.”?

> >  Comment 5
> >  
> >  Section 4.2
> >  
> >  Needed work: What is a "name-record" ?  I can imagine a few different
> >  meanings, which is the problem.  Is this a "registration" , i.e., the
> >  mapping of the network object to an entity (set)?  Are associated
> >  "name-records" like variants of IDNs?  Are they related in a IP
> >  hierarchy,
> >  or somehow jointly registered objects?
> I believe this text came from John Klensin. My understanding of the meaning is
> that the term refers to a row in a relational database. Would it be clearer to
> just say "record" or "entry" instead of "name record"?
I’ll go with what was meant.  As a reader, my note is that the term may be a
little too abstract.

Perhaps:

Conceptually, a search-query-matching record in a database may be linked to
an
   associated record, which may also be linked to another such
   record, and so on.
That is, if “linked to” is the right term.  Perhaps “associated with”
(knowing that term is in the title).  Of perhaps this

Conceptually, a search-query-matching record in a database may be a member
of a set of one or more records associated with a registration event [if
that’s
the appropriate grouping].

(Sorry about the formatting - I’ve lost control of how this mail reader does
formatting.)

I have a new comment coming from re-reading this section:
If an implementation is to return more than one
   name-record in response to a query, information from the records
   thereby identified is returned.
> From that I would assume the draft is saying all associated records would be
too - and later we talk about issues with authorization to release
information.  I’d change the “is returned” to “is subject to be included in
a response (pending other considerations [and maybe cite the Security
Considerations]).”

Oh, wait, that’s in the next paragraph.

Perhaps using the word “set” might make 4.2 work better.

Uh, see below...
> >  Comment 6
> >  
> >  
> >  Section 4.2 - later on
> >  
> >  Confusion: It sounds here that a client cannot set up an expectation
> >  that
> >  it's query will return something specific or return a list, which is
> >  bad
> >  (meaning the client can’t set expectation).  A client ought to be able
> >  to
> >  know what it will get back, even if it is an error.  What I mean is
> >  'returning information that was not explicitly selected by an exact-
> >  match
> >  lookup' is ... is bothering me.
> >  
> >  Maybe these terms need to be ironed out better: search vs. lookup and
> >  what
> >  does match mean.  I do see match defined - after it is used a few
> >  times.
> >  And in section 3.2, I first ran into the confusion of search v. lookup.
> >  
> >  (In the DNS world, we distunguish this by saying DNS does not search,
> >  it
> >  does a lookup.)
> Can you suggest appropriate text?
4.2.  Associated Records

Conceptually, any query-matching record in the registry's database might be
a member of a set of related records, related in some fashion as defined by
the registry. (E.g., variants of an IDN.)  When constructing the response,
the entire set ought to be considered as candidates for inclusion.  However,
the construction of the final response needs to be mindful of privacy and
other data-releasing policies when assembling the RDAP response set.
Note too that due to the nature of searching, there may be a list of
query-matching records.  Each one of those is subject to being a member of a
set as described in the previous paragraph.  What is ultimately returned in
a response will be the union of all the sets and be filtered by whatever
policies are in place.
> >  Comment 7
> >  
> >  Section 5
> >  
> >  Suggestion: this seems to be begging for a "standard" definition of an
> >  non-standard extension prefix - or everyone will use "custom_" because
> >  that is the example here.  (Name collisions.)  And if a custom one
> >  becomes
> >  a standard, can the transition be eased?  (TXT to SPF issue.)  And how
> >  do
> >  we encourage more standard and fewer uniques (the EPP[EXT] issue)?
> Note the reference to Section 6 of the using-http document. It describes an
> IANA registry for these prefixes.
Then I’d change this:

Custom path segments can be created by prefixing the segment with a
   unique identifier followed by an underscore character (0x5F).
To:
Custom path segments SHOULD be created by prefixing the segment with a
   unique identifier followed by an underscore character (0x5F).
(A one word change, “can” to “SHOULD”.)

> >  Comment 8
> >  
> >  Section 6
> >  
> >  Suggestion: Shouldn't RDAP applications be considered to be IDNA2008
> >  aware? The argument for this is in RFC 6912.  Or is that pressing a
> >  policy
> >  into the protocol needlessly?
> One of the authors of RFC 6912 (Andrew Sullivan) was very helpful in
> developing the text for this section. I would defer to his guidance.
>
> >  Comment 9
> >  
> >  
> >  Section 8
> >  
> >  ##   Search functionality also increases the privacy risk of disclosing
> >  ##   object relationships that might not otherwise be obvious.  For
> >  ##   example, a search that returns IDN variants [RFC6927] that do not
> >  ##   explicitly match a client-provided search pattern can disclose
> >  ##   information about registered domain names that might not be
> >  otherwise
> >  ##   available.  Implementers need to consider the policy and privacy
> >  ##   implications of returning information that was not explicitly
> >  ##   requested.
> >  
> >  I think this can improved by mentioning that implementers need to
> >  “consider policy and privacy implications, e.g., by checking for
> >  authorization to release information to the requestor, in all cases and
> >  especially here, when returning information that was not explicitly
> >  requested.”
> >  
> >  
> >  In general, I’ve always been a little suspect on RDAP’s search ability
> >  and
> >  now carefully reading the last call documents am still a little shaky
> >  on
> >  it.  Perhaps the word “search” to me has a wider definition than is
> >  being
> >  applied here, where RDAP just looks for a common pre-amble (prefix,
> >  starting string, whatever you want to call it).  I am aware of the
> >  perils
> >  of unicode, sigh, but I wish there was a chance to have a search
> >  definition that was free of ASCII assumptions and was always mindful of
> >  the authorization to “release” information.
> >  
> >  I am aware I’m saying this very late in the game.  I’m willing to
> >  accept
> >  the definition going forward as is.  But I wanted to drop the comment
> >  somewhere.  But please add in, where implementers “consider”, mention
> >  of
> >  the authorization model.  This is because I can see operators pulling
> >  open
> >  source tools for this from the shelf, they won’t be handcrafting the
> >  code,
> >  and need some place to “configure this”.
> How about this?
>
> OLD:
> "Implementers need to consider the policy and privacy implications of
> returning information that was not explicitly requested."
>
> NEW:
> "Implementers need to consider the policy and privacy implications of
> returning information that was not explicitly requested. Clients should only
> receive information that they are explicitly authorized to receive."
Almost…”Servers should only send information that clients are explicitly
authorized to receive.”

The way it is worded is impossible to "enforce."

> Scott

smime.p7s
Description: S/MIME cryptographic signature