[Captive-portals] AD review of draft-ietf-capport-architecture-07

[Barry Leiba <[e-mail]>]

Thanks for the work in this document.  Here’s my AD review; I think a
revised I-D is needed before we go to last call.

General:
Expect comments during IESG Evaluation about the extensive use of BCP
14 key words in an Informational document.  I don’t object to it here,
though I do find all the “MAY”s odd (example: “A device MAY query this
API at any time”, rather than “A device may query this API at any
time”), but I do expect some ADs to comment on it.

Please check consistency in capitalization of defined terms.  The one
that stands out to me as inconsistent is “User Equipment” / “user
equipment”, but I’ve also noticed others, such as “enforcement
device”.

Please be consistent about using “URI”, and not “URL”.

— Section 1 —

   This document standardizes an architecture for implementing captive
   portals that provides tools for addressing most of those problems.
   We are guided by these principles:

The text that comes before this doesn’t identify anything as problems,
so I don’t see, from what’s written, what “those problems” refers to.
Maybe it would be better to say, “most of the problems that result
from current captive portal mechanisms, guided by these principles:”
or some such?

   *  Solutions MUST operate at the layer of Internet Protocol (IP) or
      above, not being specific to any particular access technology such
      as Cable, WiFi or 3GPP.

3GPP is not an access technology; it’s an SDO (or a consortium of
SDOs).  Maybe “mobile telecom”, or “LTE”, or “5G”?  Or if you want to
retain 3GPP, maybe “protocols developed in 3GPP”?  I think my
preference is “5G” (and it’s only an example anyway).

   *  End-user devices can be notified of captivity with Captive Portal
      Signals in response to traffic.  This notification should work
      with any Internet protocol, not just clear-text HTTP.

I’m not quite sure what this means.  It sounds like it says that the
notification will be in-band with any Internet protocol, which clearly
isn’t possible.  I think it’s talking about a separate protocol to
deliver the notification, which can work alongside any Internet
protocol, rather than having it be in-band to clear-text HTTP.  Maybe
re-wording that sentence will help.

— Section 1.2 —

   hosts (typically the internet).

Nit: capitalize “Internet”.

   Captive Portal API: Also known as API.  An HTTP API allowing User
   Equipment to query its state of captivity within the Captive Portal.

This (and the Abstract) says “an HTTP API”, but there’s nothing at the
architecture level that assumes HTTP, is there?  Instantiation may
well be via HTTP, but why does that need to be in the architecture?  I
don’t see anything else in this document, apart from these two
definitions, binding the API to HTTP at the architecture level.

   If User Equipment supports the Captive Portal API, it MUST validate
   API server TLS certificate (see [RFC2818]).

Nit: “validate the API server’s TLS certificate”

— Section 2.3 —

   The API SHOULD provide evidence
   to the caller that it supports the present architecture.

I don’t understand what this means; can you explain?

— Section 2.4 —

   The Captive Portal Enforcement component restrict the network access

Nit: “restricts”

   *  Allows traffic through for allowed User Equipment that has

“Allows traffic through” reads oddly to me.  Maybe “Allows traffic to
pass through”? Or “Allows general Internet traffic”?

— Section 2.6 —
In Figure 1, shouldn’t “Query captivity status” and “Portal user
interface” both have bidirectional arrows?

— Section 3 —

   these interactions, the components must be able to both identify the
   user equipment from their interactions with it, and be able to agree

Common problem with “both” and “either”: “be able” is already outside
the scope of “both”, so having it inside for the second case is wrong
(and the comma is unnecessary):

NEW
   these interactions, the components must be able to both identify the
   User Equipment from their interactions with it and to agree
END

— Section 3.2.1 —

   In order to uniquely identify the User Equipment, at most one user
   equipment interacting with the other components of the Captive Portal
   MUST have a given value of the identifier.

The way “MUST” is used here feels very odd.  I think you mean the MUST
to be related to the uniqueness, not to what components MUST have.
Maybe something like this works better?:

NEW
   In order that User Equipment be uniquely identified, a given
   identifier MUST NOT represent more than one User Equiment that
   interacts with the other components of the Captive Portal.
END

— Section 3.2.3 —

   Since the API Server will need to perform operations which rely on
   the identity of the user equipment, such as query whether it is
   captive, the API Server needs to be able to relate requests to the
   User Equipment making the request.

The API server doesn’t query whether the UE is captive, it responds to
such queries.

— Section 3.2.4 —

   The Enforcement Device will decide on a per packet basis whether it
   should be permitted to communicate with the external network.

Nit: hyphenate “per-packet”.
“It” should be referring to the packet, not to the ED, so “whether the
packet should”.

— Section 3.3 —
Are we really talking about evaluating individual identifiers here?
Or does this really mean to discuss *methods* of generating or
choosing identifiers?

— Section 3.4.3 —
Is this section talking about using a context-free URI as a UE
identifier?  It should be clearer about that, if so (and if that’s not
what it’s about, the section is misplaced).  There’s nothing in here
that discusses how such identifiers would meet the specified criteria.

— Section 4 —
Please be consistent about the use of “workflow” vs “work-flow”.

— Appendix A —

   method is to attempt to make a HTTP request to a known, vendor hosted

Nit: hyphenate “vendor-hosted”

   DNS queries not using UDP
   may potentially fail this test if operating over TCP or DNS over
   HTTP.

Nit: it’s “DNS over HTTPS”.  But I would say, “DNS queries not using
UDP may potentially fail this test if operating over TCP or HTTPS.”

-- 
Barry