[Openid-specs-ab] Comments on federation draft-08

Roland Hedberg roland at catalogix.se
Fri Aug 16 06:27:31 UTC 2019



> On 13 Aug 2019, at 15:14, Marcos Sanz via Openid-specs-ab <openid-specs-ab at lists.openid.net> wrote:
> 
> Hi all,
> 
> I had a chance to go through the federation draft (though I admit I 
> skipped the appendixes because I found them a bit difficult to read) and 
> have a couple of comments to make. I have separated comments in category 
> "discussion-prone" and simple nits:
> 
> Main things:
> 
> - Section 5: The usage of well-known URIs is different to what I have seen 
> in other OpenID specs: Usually, at the well-known URI you'll find a 
> (static) document that will inform you about configuration parameters, 
> discovery information, URIs of actual endpoints of the service, and so on. 
> However, the federation spec is running the federation API directly under 
> "/.well-known/openid-federation". This is, for starters, a break with the 
> existing practice in the spec family. I then tried to look for standards 
> advice on the topic. In the (now obsolete) RFC 5785 there was the 
> following guidance text:
> 
> "well-known URIs are not intended for general information retrieval or 
> establishment of large URI namespaces on the Web.  Rather, they are 
> designed to facilitate discovery of information on a site when it isn't 
> practical to use other mechanisms". Also from 5785: "the well-known URI 
> space was created with the expectation that it will be used to make 
> site-wide policy information and other metadata available directly (if 
> sufficiently concise), or provide references to other URIs that provide 
> such metadata".
> 
> Both sentences have admittedly disappeared in the younger RFC 8615, but 
> you'll still find text in 8615 speaking of "metadata discovered by 
> dereferencing the well-known URI", which somehow implies what the 
> expectations are when you visit that URI, namely, metadata.
> Now you could claim that fetching entity statements and doing trust 
> negotation is acquiring metadata from the federation, but I think that's 
> executing business logic of the API ("Application _Programmers_ 
> Interface", see? ;-). On top, I see operational problems in front of us if 
> the spec sticks to this well-known URI usage: if it's already difficult to 
> place a static document for a service like IdP discovery under 
> .well-known/webfinger in the generic domain amazon.com (I heard that from 
> an authoritative source at the IIW last year), go figure if you want to 
> run application code there.
> 
> All in all, I defend using the well-known URI of the federation to 
> discover a document where the actual Federation API endpoint is listed. 
> (If you all have already discussed about this and then dropped it, then 
> I'd be thankful for a pointer in the archives).

You’re the first who have commented on this. 
I have no problem with accepting your proposal. In fact I have seen exactly the
DNS problem you refer to,

> - Section 5.4: This looks to me a bit rudimentary. I wonder whether using 
> RFC 7807 and the associated media type application/problem+json wouldn't 
> be more appropriate to carry problems around. One downside of this, 
> though, it's that it hasn't been used so far in the OpenID specs I know 
> of: if I used consistency within the spec family as an argument before, I 
> should be able to accept some pushback here :-)

I definitely wants to follow in the footsteps of the core specs,

> - Section 8.1.3: This is underspecified and "fails-to-establish-trust" is 
> a place where many things can go wrong. I think the spec should take the 
> effort of defining a base set of error codes, the same way for instance 
> that the core spec does in section 3.1.2.6. Further, using a catch-all 
> "invalid_request" code, which originally was thought as for indicating "
> The request is missing a required parameter, includes an invalid parameter 
> value, includes a parameter more than once, or is otherwise malformed" 
> (RFC 6749), is a misuse.

OK, I’ll work on this.

> Nits
> - Section 1.2: Definition of "Trust Chain" uses the expression "trusted 
> chain", which is self-referential
> - Section 4: s/polices/policies/
> - Section 4.5: s/parties that wants/parties that want/
> - Section 4.6: Last example, there's a spurious comma after the last JSON 
> element
> - Section 5: s/multi-teanant/multi-tenant/ in all URIs
> - Section 5: s/requests to the federation API does/requests to the 
> federation API do/
> - Section 5.1.2: the example is kind of useless. I am aware that it must 
> be a JOSE and all of that, but it's also truncated, so it can't even be 
> decoded. I'd rather have an already decoded positive response to show how 
> it should look like (plus a caveat that it will never look like that on 
> the wire)
> - Section 5.2.1: the example uses the parameter "op" while the actual name 
> is "operation"
> - Section 5.2.2: s/polices/policies/
> - Section 5.3.1: the example uses the parameter "op" while the actual name 
> is "operation". Also, the example uses a parameter "type" which isn't 
> defined in the text above.
> - Section 6: The sentence that ends with "...and a list of entity IDs of 
> trusted trust anchors together with the public version of their signing 
> keys". If I get this right the part "together with the public version of 
> their signing keys" is not needed, since federation makes heavy use of 
> discovery and, strictly speaking, that public key information will be 
> dynamically be read from the trust anchors' jwks_uri, right?
> - Section 6.2: Also strictly speaking, the last sentence "no information 
> in the chain of statements should be used before the signature chain has 
> been validated" directly contradicts with the previous paragraph "an 
> implementer MAY therefor chose [sic] to not verify the signature until all 
> the other checks have been done".
> - Section 7.2: For the first time appears the term "trust root", and it's 
> used a couple of times afterwards. If it's the same as "trust anchor", I'd 
> rather consistently stick to the latter.
> - Section 8.1.1: s/authorization request example/authentication request 
> example/ for the sake of consistency
> - Section 8.1.1, example: the parameter "state" is twice in the URI. And I 
> personally find the usage of "localhost" in the client_id kind of 
> distracting.
> - Section 11.1: Replace RFC 5785 with 8615

Good points ! I’ll take care of them.

Thanks Marcos,

> Best regards,
> Marcos
> 
> -- 
> Dipl.-Ing. Marcos Sanz Grossón
> Leiter Software Engineering
> 
> DENIC eG, Kaiserstraße 75 – 77, 60329 Frankfurt am Main, GERMANY
> E-Mail: sanz at denic.de, Fon: +49 69 27235-0, Fax: -235
> https://www.denic.de
> 
> Angaben nach §25a Absatz 1 GenG: DENIC eG (Sitz: Frankfurt am Main)
> Vorstand: Martin Küchenthal, Andreas Musielak, Sebastian Röthler, Dr. Jörg 
> Schweiger
> Vorsitzender des Aufsichtsrats: Thomas Keller
> Eingetragen unter Nr. 770 im Genossenschaftsregister, Amtsgericht 
> Frankfurt am Main
> _______________________________________________
> Openid-specs-ab mailing list
> Openid-specs-ab at lists.openid.net
> http://lists.openid.net/mailman/listinfo/openid-specs-ab

— Roland
Scratch a pessimist and you find often a defender of privilege. -William Beveridge, economist and reformer (5 Mar 1879-1963) 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-specs-ab/attachments/20190816/b3ef5f69/attachment-0001.html>


More information about the Openid-specs-ab mailing list