Re: [OAUTH-WG] Followup on draft-ietf-oauth-token-exchange-12.txt

[Brian Campbell <[e-mail]>]

Hi Eric,

Apologies for my somewhat slow response. I've honestly been unsure of how
else to try and address the comment/question. But will continue trying...

My expectation would be that access control decisions would be made based
on the subject of the token itself or on the current actor. And maybe a
combination of both in some situations (like, for example, the actor is an
administrator and the token allows admin level access to the stuff the
token subject would normally have access to).  However, I don't believe
that nested prior actors would or should be considered in access control
decisions. The nesting is more just to express what has happened for
auditing or tracking or the like. To be honest, the nesting was added in
the draft largely because the structure naturally and easily allowed for it
and it seemed like it might be useful information to convey in some cases.

So in that A->B->C case (the claims of such a token would, I think, look
like the JSON below), B *is not* giving C his authority. B is just noted in
the token as having been involved previously.  While A is identified as the
subject of the token and C is the current actor.

    {
      "aud":"... ,"iss":... , "exp":..., etc. etc. ...
      "sub":"A",
      "act":
      {
        "sub":"C",
        "act":
        {
          "sub":"B"
        }
      }
    }


Would some text explicitly saying that only the token subject (top level
sub and claims) and the party identified by the outermost "act" claim (the
current actor) are to be considered in access control decisions address
your concern?


On Tue, May 29, 2018 at 4:19 PM, Eric Rescorla <[e-mail]> wrote:

> Hi Brian,
>
> To be clear, I'm not opposing Delegation. My concern here is that we have
> a chain of signed assertions and I'm trying to understand how I as a
> consumer of those assertions am supposed to evaluate it.
>
> I don't think it's sufficient to just say that that the access control
> rules are local policy, because then the entity generating the signature
> has no way of knowing how its signature will be used.
>
> To go back to the case I gave in my initial e-mail, say we have a chain
> A->B->C and a resource that A and C could ordinarily not access, but B can.
> If C has this delegation, can C access the resource? I.e., is B giving C
> his authority or just passing on A's authority? It seems pretty important
> for B to know that before he gives the token to C.
>
> -Ekr
>
>
> On Thu, May 17, 2018 at 11:06 AM, Brian Campbell <
> [e-mail]> wrote:
>
>> Delegation has been in the document since its inception and throughout
>> the three and a half years as a working group document.
>>
>> From a process point of view, the document is now in AD Evaluation. I
>> worked through a number of questions and clarifications with Eric (said
>> AD), however he raised the particular questions that started this thread on
>> the WG list. And I responded with an attempt at addressing those questions.
>> That was about a month ago.
>>
>> Eric, was my explanation helpful in clarify anything for you? Is there
>> some text that you'd like to see added? Something else? I'm unsure how to
>> proceed but would like to move things forward.
>>
>>
>> On Thu, May 17, 2018 at 8:03 AM, Bill Burke <[e-mail]> wrote:
>>
>>> This is an honest question: How important is the actor stuff to the
>>> players involved?  Are people going to use it?  IMO, its an edge case
>>> and I think more important areas, like external token exchange (realm
>>> to realm, domain to domain) are being neglected.  I'm quite unfamiliar
>>> how consensus is reached in this WG or the IETF, so I hope I'm not
>>> sounding rude.  Just trying to provide some constructive feedback.
>>>
>>>
>>>
>>> On Thu, May 17, 2018 at 9:26 AM, Mike Jones <[e-mail]>
>>> wrote:
>>> > Moving the actor claim to a separate specification would only make
>>> things more complicated for developers.  There already plenty of OAuth
>>> specs.  Needlessly adding another one will only make related things harder
>>> to find.
>>> >
>>> > Just like in the JWT [RFC 7519] spec itself in which use of all the
>>> claims is optional, use of the actor claim in this spec.  If you don't need
>>> it, don't use it.  Just because some won't use it is no better an argument
>>> for moving it to a different spec than the argument that JWT should have
>>> defined each of its claims in different specs.  That would have made things
>>> harder, not easier.
>>> >
>>> >                                 -- Mike
>>> >
>>> > -----Original Message-----
>>> > From: OAuth <[e-mail]> On Behalf Of Bill Burke
>>> > Sent: Thursday, May 17, 2018 2:11 PM
>>> > To: Brian Campbell <[e-mail]>
>>> > Cc: oauth <[e-mail]>
>>> > Subject: Re: [OAUTH-WG] Followup on draft-ietf-oauth-token-exchang
>>> e-12.txt
>>> >
>>> > My personal opinion is that I'm glad this actor stuff is optional.
>>> > For one, none of our users have asked for it and really only do simple
>>> exchanges.  Secondly, the rules for who can exchange what for what is
>>> controlled and defined within our AS.  Makes things a lot simpler on the
>>> client.  I kind of wish the actor stuff would be defined in a separate
>>> specification.  I don't see us implementing it unless users start asking us
>>> to.
>>> >
>>> > On Wed, May 16, 2018 at 6:11 PM, Brian Campbell <
>>> [e-mail]> wrote:
>>> >> Well, it's already called the "actor claim" so the claimed part is
>>> >> kind of implied. And "claimed actor claim" is a rather awkward.
>>> >> Really, all JWT claims are "claimed something" but they don't include
>>> >> the "claimed" bit in the name. RFC 7519, for example, defines the
>>> >> subject claim but not the claimed subject claim.
>>> >>
>>> >> On Fri, Apr 20, 2018 at 11:38 AM, Denis <[e-mail]> wrote:
>>> >>>
>>> >>> Brian,
>>> >>>
>>> >>> Eric said: "what is the RP supposed to do when they encounter it?
>>> >>> This seems kind of under specified".
>>> >>>
>>> >>> After reading your explanations below, it looks like the RP can do
>>> >>> anything he wants with the "actor".
>>> >>> It is a "claimed actor" and, if we keep the concept, it should be
>>> >>> called as such. Such a claim cannot be verified.
>>> >>> A RP could copy and paste that claim in an audit log. No standard
>>> >>> action related to the content of such a claim can be specified in the
>>> >>> spec. If the content of a "claimed actor" is used by the RP, it
>>> >>> should be only used as an hint and thus be subject to other
>>> >>> verifications which are not specified in this specification.
>>> >>>
>>> >>> Denis
>>> >>>
>>> >>> Eric, I realize you weren't particularly impressed by my prior
>>> >>> statements about the actor claim but, for lack of knowing what else
>>> >>> to say, I'm going to kind of repeat what I said about it over in the
>>> >>> Phabricator tool and add a little color.
>>> >>>
>>> >>> The actor claim is intended as a way to express that delegation has
>>> >>> happened and identify the entities involved. Access control or other
>>> >>> decisions based on it are at the discretion of the consumer of the
>>> >>> token based on whatever policy might be in place.
>>> >>>
>>> >>> There are JWT claims that have concise processing rules with respect
>>> >>> to whether or not the JWT can be accepted as valid. Some examples
>>> are "aud"
>>> >>> (Audience), "exp" (Expiration Time), and "nbf" (Not Before) from RFC
>>> 7519.
>>> >>> E.g. if the token is expired or was intended for someone or something
>>> >>> else, reject it.
>>> >>>
>>> >>> And there are JWT claims that appropriately don't specify such
>>> >>> processing rules and are solely statements of fact or circumstance.
>>> >>> Also from RFC 7519, the "sub" (Subject) and "iat" (Issued At) claims
>>> are good examples of such.
>>> >>> There might be application or policy specific rules applied to the
>>> >>> content of those kinds of claims (e.g. only subjects from a
>>> >>> particular organization are able to access tenant specific data or,
>>> >>> less realistic but still possible, disallow access for tokens issued
>>> >>> outside of regular business
>>> >>> hours) but that's all outside the scope of a specification's
>>> >>> definition of the claim.
>>> >>>
>>> >>> The actor claim falls into the latter category. It's a way for the
>>> >>> issuer of the token to tell the consumer of the token what is going
>>> >>> on. But any action to take (or not) based on that information is at
>>> >>> the discretion of the token consumer. I honestly don't know it could
>>> >>> be anything more. And don't think it should be.
>>> >>>
>>> >>> There are two main expected uses of the actor claim (that I'm aware
>>> >>> of
>>> >>> anyway) that describing here might help. Maybe. One is a human to
>>> >>> human delegation case like a customer service rep doing something on
>>> >>> behalf of an end user. The subject would be that user and the actor
>>> >>> would be the customer service rep. And there wouldn't be any chaining
>>> >>> or nesting of the actor. The other case is so called service chaining
>>> >>> where a system might exchange a token it receives for a new token
>>> >>> that it can use to call a downstream service. And that service in
>>> >>> turn might do another exchange to get a new token suitable to call
>>> >>> yet another downstream service. And again and so on and turtles all
>>> >>> the way. I'm not necessarily endorsing that level of granularity in
>>> >>> chaining but it's bound to happen somewhere/sometime. The nested
>>> >>> actor claim is able to express that all that has happened with the
>>> >>> top level or outermost one being the system currently using the token
>>> >>> and prior systems being nested.. What actually gets done with that
>>> >>> information is up to the respective systems involved. There might be
>>> >>> policy about what system is allowed to call what other system that is
>>> >>> enforced. Or maybe the info is just written to an audit log
>>> >>> somewhere. Or something else. I don't know. But whatever it is
>>> application/deployment/policy dependent and not specifiable by a spec.
>>> >>>
>>> >>>
>>> >>>
>>> >>>
>>> >>>
>>> >>>
>>> >>> On Fri, Apr 13, 2018 at 6:38 PM, Eric Rescorla <[e-mail]> wrote:
>>> >>>>
>>> >>>> Hi folks,
>>> >>>>
>>> >>>> I've gone over draft-ietf-oauth-token-exchange-12 and things seem
>>> >>>> generally OK. I do still have one remaining concern, which is about
>>> >>>> the actor claim. Specifically, what is the RP supposed to do when
>>> >>>> they encounter it? This seems kind of underspecified.
>>> >>>>
>>> >>>> In particular:
>>> >>>>
>>> >>>> 1. What facts am I supposed to know here? Merely that everyone in
>>> >>>>    the chain signed off on the next person in the chain acting as
>>> them?
>>> >>>>
>>> >>>> 2. Am I just supposed to pretend that the person presenting the
>>> token
>>> >>>>    is the identity at the top of the chain? Say I have the
>>> >>>>    delegation A -> B -> C, and there is some resource which
>>> >>>>    B can access but A and C cannot, should I give access?
>>> >>>>
>>> >>>> I think the first question definitely needs an answer. The second
>>> >>>> question I guess we could make not answer, but it's pretty hard to
>>> >>>> know how to make a system with this left open..
>>> >>>>
>>> >>>> -Ekr
>>> >>>>
>>> >>>>
>>> >>>> _______________________________________________
>>> >>>> OAuth mailing list
>>> >>>> [e-mail]
>>> >>>> https://www.ietf.org/mailman/listinfo/oauth
>>> >>>>
>>> >>>
>>> >>>
>>> >>> CONFIDENTIALITY NOTICE: This email may contain confidential and
>>> >>> privileged material for the sole use of the intended recipient(s).
>>> >>> Any review, use, distribution or disclosure by others is strictly
>>> >>> prohibited..  If you have received this communication in error,
>>> >>> please notify the sender immediately by e-mail and delete the message
>>> >>> and any file attachments from your computer. Thank you.
>>> >>>
>>> >>> _______________________________________________
>>> >>> OAuth mailing list
>>> >>> [e-mail]
>>> >>> https://www.ietf.org/mailman/listinfo/oauth
>>> >>>
>>> >>>
>>> >>>
>>> >>> _______________________________________________
>>> >>> OAuth mailing list
>>> >>> [e-mail]
>>> >>> https://www.ietf.org/mailman/listinfo/oauth
>>> >>>
>>> >>
>>> >>
>>> >> CONFIDENTIALITY NOTICE: This email may contain confidential and
>>> >> privileged material for the sole use of the intended recipient(s). Any
>>> >> review, use, distribution or disclosure by others is strictly
>>> >> prohibited..  If you have received this communication in error, please
>>> >> notify the sender immediately by e-mail and delete the message and any
>>> >> file attachments from your computer. Thank you.
>>> >> _______________________________________________
>>> >> OAuth mailing list
>>> >> [e-mail]
>>> >> https://www.ietf.org/mailman/listinfo/oauth
>>> >>
>>> >
>>> >
>>> >
>>> > --
>>> > Bill Burke
>>> > Red Hat
>>> >
>>> > _______________________________________________
>>> > OAuth mailing list
>>> > [e-mail]
>>> > https://www.ietf.org/mailman/listinfo/oauth
>>>
>>>
>>>
>>> --
>>> Bill Burke
>>> Red Hat
>>>
>>
>>
>> *CONFIDENTIALITY NOTICE: This email may contain confidential and
>> privileged material for the sole use of the intended recipient(s). Any
>> review, use, distribution or disclosure by others is strictly prohibited.
>> If you have received this communication in error, please notify the sender
>> immediately by e-mail and delete the message and any file attachments from
>> your computer. Thank you.*
>
>
>

-- 
_CONFIDENTIALITY NOTICE: This email may contain confidential and privileged 
material for the sole use of the intended recipient(s). Any review, use, 
distribution or disclosure by others is strictly prohibited.  If you have 
received this communication in error, please notify the sender immediately 
by e-mail and delete the message and any file attachments from your 
computer. Thank you._