[policy-charter] Authorization API Proposal

Wed Jun 28 18:40:05 UTC 2023

Please add me to the repository as well: davidjbrossard

Good feedback overall. I would argue that everything can be described in
terms of attributes. Historically, in XACML an attribute has a type,
identifier, category, and optionally issuer. For instance, userRole of type
string and category subject.

In retrospect, we realize that specifying the category is overkill.
Oftentimes the identifier will convey the grammatical meaning of the
attribute therefore nulling the need for a category.

Also the original core spec allows for something like 15 plus datatypes
including URI email and so on. In reality most deployments use string,
boolean, number, and date/time. The rest is too much.

I would steer away from terms like principal. In my mind it makes the spec
harder to understand because we do not all agree on what principal is. This
maps nicely to XACML categories.

Alex, I did not see hasFriend as an action but an attribute of the user.

Regarding API authentication, I think it is totally orthogonal to this
spec. If the customer wants to allow anonymous access, let them be. If they
want to use OAuth, so be it...

More to come

On Wed, Jun 28, 2023, 10:51 AM Alex Babeanu via policy-charter <
policy-charter at lists.openid.net> wrote:

> Thanks for sharing this Atul,
>
> Some additional comments:
>
> > Re: Oauth2, I agree with Omri: I think this spec should just state that
> authentication is required, but out of scope of the doc. We could suggest
> Oauth2 that being said...
>
> > I would also prefer to stick to "Subject" and "Resource"
>
> 3.5 - Principals ("Subjects")
> > An ID should indeed suffice. Now it's probably a good idea to also
> *optionally* add some Subject claims here that the PDP can use (thinking
> JWT claims coming into the PEP, or environmental values for example). But
> in that case it should not be just "IP" and "DeviceID", but rather an array
> of "key"="Value" claim pairs, which may be completely custom and
> use-case-specific.
>
> > It would be good to also have a Subject Type - make it optional if not
> needed (but we would need it for example).
>
> 3.6 - Assets
> > ID should be mandatory I think.
>
> 3.7 - Actions
> --> Omri, if "Action" is enough to represent relationships in your world,
> it isn't in mine. Just think of "HAS_FRIEND" relationships for example...
> Anyway, there's probably no point in adding any graph-modelling to this
> doc...
>
> 3.12.2 - Evaluation response
> > I don't think we need an `exp` value: we can't "predict" how long an
> access rule would be true for: that's a business decision! Besides, in the
> spirit of Zero Trust, a decision is just made at 1 point in time, the next
> access request may yield a completely different response. Up to the PDP to
> do caching or whatever to speed-up processing.
>
> > I don't think it's a good idea to reply with all language strings, this
> could be a huge response and it would probably slow-down the PDP too.
> Instead, have the client PEP request the language it wants its responses
> in, and just return that 1 reason string.
>
> 3.13 - Search API
> > We also need the reverse query, i.e., who can access this given Resource
> ?
> > Again, I don't think we need `exp` here.
>
> Regards,
>
> ./\.
>
> On Tue, Jun 27, 2023 at 10:48 PM Omri Gazitt via policy-charter <
> policy-charter at lists.openid.net> wrote:
>
>> Thanks Atul - it's clear you and the SGNL team put some thought into this.
>>
>> I'm still looking through this but I like that you're specifying both a
>> "check" function (Access Evaluation) and an "expand" function (Search
>> API).  Both are important, but perhaps only the Access Evaluation API ought
>> to be required.
>>
>> I'm not sure what is the best way to provide feedback. Github issues?
>> PRs? put it in a google doc and use comments?
>>
>> Here are some thoughts (sorry, no particular order):
>>
>> > Policy Distribution Point
>> Nit - this should be "Policy Decision Point" (which you name correctly in
>> other places in the doc)
>>
>> > The Authorization API is itself authorized using OAuth 2.0 ([RFC6749
>> <https://sgnl-ai.github.io/authzapi/#RFC6749>])
>> I think this is overspecified. I can certainly see clients that call an
>> Authorization API using an API key.
>>
>> > Terminology
>> A bit of a quibble, but perhaps we tip the hat to XACML and use the terms
>> "subject" and "resource" instead of "principal" and "asset". I'm not
>> religious about these terms, but "asset" in particular seems less common
>> than "object" or "resource".
>>
>> > Principals
>> In many systems, subjects will be extracted from a JWT "sub" claim, which
>> is a string. I'm not sure that specifying that this must be a JSON
>> structure, and further specifying two optional fields (ipAddress and
>> deviceId) is necessary, and feels overspecified.
>>
>> > Assets
>> Having a type and id makes sense, but could these be specified in a
>> single string?  For example, zanzibar defines these as type:id
>>
>> I do like having the ability to pass in properties in addition to the
>> object identifier - but it seems like this should be a json object
>> (key:value pairs), not just an array of attribute names.
>>
>> > Actions
>> It's not clear to me that separating actions into "standard" and "custom"
>> is useful. I think the types of actions you list (CRUD) are common examples
>> but should not be normative.
>>
>> I do think it's possible to use the generic concept of "Action" to
>> encompass permissions and relations.
>>
>> > Queries (as array)
>> I think that allowing a set of decisions to be requested at once is
>> valuable, but IMO the spec should not mandate that implementations must
>> support more than one query at a time. Some PDPs don't support that
>> semantic and it should be considered optional.
>>
>>
>>
>>
>> On Tue, Jun 27, 2023 at 4:38 PM Atul Tulshibagwale via policy-charter <
>> policy-charter at lists.openid.net> wrote:
>>
>>> Hi all,
>>>
>>> Here's a proposal of an Authorization API that we would like to
>>> contribute to this group (in its current form or if / when we find a home
>>> in a standards body). This is similar to at least a few vendors' current
>>> offerings, so I hope everyone finds this helpful, and we can accelerate our
>>> standardization efforts as a result.
>>>
>>> You can read the proposal in HTML format here:
>>> https://sgnl-ai.github.io/authzapi/
>>>
>>> The sources (under the MIT License) are here:
>>> https://github.com/SGNL-ai/authzapi
>>>
>>> - Atul Tulshibagwale, Erik Gustavson and Marc Jordan
>>>
>>> --
>>>
>>> <https://sgnl.ai>
>>>
>>> Atul Tulshibagwale
>>>
>>> CTO
>>>
>>> <https://linkedin.com/in/tulshi> <https://twitter.com/zirotrust>
>>> <atul at sgnl.ai>
>>> --
>>> policy-charter mailing list
>>> policy-charter at lists.openid.net
>>> https://lists.openid.net/mailman/listinfo/policy-charter
>>>
>> --
>> policy-charter mailing list
>> policy-charter at lists.openid.net
>> https://lists.openid.net/mailman/listinfo/policy-charter
>>
>
>
> --
> [image: This is Alexandre Babeanu's card. Their email is alex at 3edges.com.
> Their phone number is +1 604 728 8130.]
> <https://hihello.me/p/cda689b1-0378-4b9c-88cf-33a9bc8ef0c5>
>
> CONFIDENTIALITY NOTICE: This e-mail message, including any attachments
> hereto, is for the sole use of the intended recipient(s) and may contain
> confidential and/or proprietary information.
> --
> policy-charter mailing list
> policy-charter at lists.openid.net
> https://lists.openid.net/mailman/listinfo/policy-charter
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/policy-charter/attachments/20230628/1dfbb564/attachment.html>