[Openid-specs-heart] HEART Scopes & Resource Sets

[Josh Mandel]

I continue to find these discussions fascinating and high-yield, so please
don't take my "status quo" reasoning in the wrong light. I'm pushing back
on what's different because it's the only way I know to get at the
important questions!

Responses inline (and I've snipped a bunch to try to focus on key points).

Best,

  -Josh


> The question is about how this works under the hood.  I think we were
>> discussing two models:
>>
>> *Model 1: The "UMA-First" approach*
>> *We have a resource set like "Alice's Medications", with scopes like
>> "view" and "prescribe". And we'd have a resource set like "Alice's Step
>> Counts" with scopes like "view" and "delete".*
>>
>> *Model 2: The "OAuth-First" approach*
>> *We have a resource set like "Alice's FHIR Endpoint", with scopes like
>> "Medications.view", "Medications.prescribe", "Steps.view", and
>> "Steps.delete".*
>>
>>
> Talking about an "OAuth-first" approach for setting policies is making me
> confused.
>

I was trying to describe an approach to designing scopes that would be
amendable to OAuth workflows and UMA workflows. I just wanted to be clear
that the UMA policy-setting piece is meant to have a strong, helpful UX, no
matter what decisions we make about scope design. So yes: the
policy-setting is something that happens in the UMA world, and perhaps that
was the source of confusion.


> I know what it looks like to enable OAuth-like flows in UMA when Alice is
> both the requesting party and the owner of the resource. And I know what it
> looks like to enable Alice to set policies at an UMA authorization server
> (which might hold the results of a previous OAuth-like flow done in UMA).
> But I don't know what "setting policies in OAuth" means because the OAuth
> experience is about consenting at run time
>

I think you introduced the phrase "setting policies in OAuth". I was trying
to describe how the policy-setting plays out in UMA, and then thinking
through two different ways in which the world might be broken down into
UMA-friendly (Resource Set, scope) tuples.

So the closest UX analog would probably be the wording displayed in an
> OAuth consent dialog, maybe something like:
>
>    - View [and prescribe] your medications
>    - View [and delete] your steps
>
> Yes, that's right. Or a runtime OAuth consenting flow like "first pick
which record(s) you want to share; then pick which pieces of it to share"
(a process that could be repeated as needed to share different pieces of
different records).

If the *types* of Resource Sets and the allowed scopes are standardized in
>> advance (which UMA supports), then a mapping between Model 1 and "vanilla"
>> OAuth could be as simple as: "concatenate the UMA resource set type
>> followed by ':' followed by the UMA scope name" -- so for example, you
>> might derive an OAuth scope like "
>> https://openid.net/heart/resource-types/StepCounts:https://openid.net/heart/scopes/view
>> <https://urldefense.proofpoint.com/v2/url?u=https-3A__openid.net_heart_resource-2Dtypes_StepCounts-3Ahttps-3A__openid.net_heart_scopes_view&d=BQMFaQ&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=c7b1QeR755-dBx2b0xnlepDTylromoEzcLl-6ixmBL3TpXSxSvtAvT553fzSgLpm&m=Yvi0iXWjH-HuDwj067Ier3mSZrAAoByMQ55RWTEncls&s=3eZ-ycJwrGeqj21YuaP-LIULSjdLS_pIcorkjW-UcZs&e=>".
>> Or under Model 2, the scopes could be reused directly (no mapping
>> required).
>>
>
> In what sense is "reuse" meant here?
>

In the sense of: UMA and OAuth both use something called "scopes". To what
extent can the same scope values be used in both contexts? Or if not
literally the same values, to what extent can we define a deterministic
mapping between them, so we only have to define the actual concepts once?


> A coding model, or an architectural model, or a semantic model?...
>

I'm not sure what these mean, but I think I've answered your question above.

>
> Let's say (totally making this up) the FHIR has two endpoints, with one
> endpoint for medication records and one for fitness steps. There's an
> UMA-standardized resource type for each. There's "*https://www.hl7.org/fhir/rsrc/med.json
> <https://urldefense.proofpoint.com/v2/url?u=https-3A__www.hl7.org_fhir_rsrc_med.json&d=BQMFaQ&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=c7b1QeR755-dBx2b0xnlepDTylromoEzcLl-6ixmBL3TpXSxSvtAvT553fzSgLpm&m=Yvi0iXWjH-HuDwj067Ier3mSZrAAoByMQ55RWTEncls&s=XQmQhCmLFsJ8xmKSmiMAjy9ALdeP13Bkn1oF30MKlSc&e=>*",
> with instances of it registered with scopes "*view*", "*download*", "
> *transmit*", and "*add*" (so some clients can insert new entries).
> Alice's medications might be in a resource something like "*/alice/meds*".
> (What's displayed in her AS dashboard might look a lot nicer than that.)
> And there's "*https://www.hl7.org/fhir/rsrc/step.json
> <https://urldefense.proofpoint.com/v2/url?u=https-3A__www.hl7.org_fhir_rsrc_step.json&d=BQMFaQ&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=c7b1QeR755-dBx2b0xnlepDTylromoEzcLl-6ixmBL3TpXSxSvtAvT553fzSgLpm&m=Yvi0iXWjH-HuDwj067Ier3mSZrAAoByMQ55RWTEncls&s=NRFrrL6qQ_Do9SoJ561B8i1oz9_4FoW3I2ZN6_L-mRI&e=>*",
> with instances of it registered with scopes "*view*", "*download*", "
> *transmit*", and "*chart*". Alice's steps might be in a resource like
> "/alice/steps".
>
> (If the scopes are in the form of URIs, they could be standardized to a
> further degree, in that a bunch of metadata could be pulled by the
> authorization server and used to present standard labels and icons, and
> other semantics could be linked to them.)
>
> If the very same API were OAuth-protected, with the very same resource
> endpoints, there might still be the same resource endpoints, with the same
> scopes, where three of them work on both resource types, "*add*" only
> works on "*med*", and "*chart*" only works on "*step*". These resources
> could still have a standardized meaning in terms of both resource naming
> and schema/format; there just would be nowhere to "hook" a standardized
> resource type URI into.
>

This approach is totally possible. I haven't seen anything quite like it in
the wild (but please point me to good examples!), perhaps because the UX of
asking someone to consent N times in a row to N different OAuth "authorize"
calls could become painful. In typical OAuth implementations, at least, the
model is one consent flow per authorization request, and one authorization
request per service provider. So we typically see an app asking to connect
"to my google account," and then I can specify which pieces. In places
where I've seen more fine-grained access, they've either been implicit
("allow this PDF viewer app to read all documents that you open from
GDrive *with
this PDF viewer app*" -- so the binding to actual resources happens at
runtime), or non-standardized (e.g. Google Docs lets you generate share
links, but there's no automated OAuth-y way to do it. It's just a custom
in-app interface for generating/revoking these links).
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-specs-heart/attachments/20150616/b25bf400/attachment.html>