Re: [OAUTH-WG] Call for Adoption

[Nat Sakimura <[e-mail]>]

So, is there a lot of cases that the authority section of the Good AS's
Authorization Endpoint and the Token Endpoints are different?
If not, then requiring that they are the same seems to virtually remove the
attack surface for the mix-up related attacks. It does not introduce new
parameter nor discovery. If it can be done, it probably is not worth adding
a new wire protocol element to mitigate the mix-up variants.



2016年1月27日(水) 4:44 Brian Campbell <[e-mail]>:

> I understand what you're saying but disagree with the premise.
>
> On Tue, Jan 26, 2016 at 12:07 PM, George Fletcher <[e-mail]>
> wrote:
>
>> So I'm fine with not requiring discovery in the simple case of an RS
>> supporting a single AS. However, once the RS moves to supporting multiple
>> authorization servers then I believe that discovery based on the 'iss'
>> string is required. The discovery results can be cached so a discovery
>> lookup per transaction is not required.
>>
>> Thanks,
>> George
>>
>> On 1/26/16 1:58 PM, Brian Campbell wrote:
>>
> I'm staying that it's sufficiently unlikely that it shouldn't be part of
>> the threat model and that a discovery lookup on iss isn't necessary.
>>
>>
>> On Tue, Jan 26, 2016 at 8:21 AM, George Fletcher <[e-mail]>
>> wrote:
>>
>> While it's a different way of getting the endpoints mixed up, I don't
>>> think that makes it invalid. If we are going to address the attack vector I
>>> believe we should solve it for "all" cases not just the ones that seem most
>>> likely.
>>>
>>> Thanks,
>>> George
>>>
>>> On 1/26/16 8:37 AM, Brian Campbell wrote:
>>>
>> I'm not sure what's described in the blog post is actually a variant of
>>> an attack that anyone is really concerned about? A client developer would
>>> have to change a production system to use an unfamiliar value for the token
>>> endpoint based solely on a an email without even looking at service
>>> documentation or metadata.
>>>
>>> On Fri, Jan 22, 2016 at 6:29 PM, Nat Sakimura < <[e-mail]>
>>> [e-mail]> wrote:
>>>
>>> I wrote a blog on why the current mix-up draft approach does not solve
>>>> the issue.
>>>>
>>>>
>>>> http://nat.sakimura.org/2016/01/22/code-phishing-attack-on-oauth-2-0-rfc6749/
>>>>
>>>> Hopefully, the point I am making is clearer by this.
>>>>
>>>> Best,
>>>>
>>>> Nat
>>>>
>>>> 2016年1月23日(土) 8:32 Mike Jones <[e-mail]>:
>>>>
>>> I like the “from which the authorization server's configuration
>>>>> information location can be derived” language.  Thanks.  I’ll plan to
>>>>> incorporate that the next time the draft is revised.
>>>>>
>>>>>
>>>>>
>>>>>                                                                 -- Mike
>>>>>
>>>>>
>>>>>
>>>>> *From:* Brian Campbell [mailto: <[e-mail]>
>>>>> [e-mail]]
>>>>> *Sent:* Friday, January 22, 2016 3:26 PM
>>>>> *To:* Nat Sakimura < <[e-mail]>[e-mail]>
>>>>> *Cc:* Mike Jones < <[e-mail]>
>>>>> [e-mail]>; Justin Richer < <[e-mail]>
>>>>> [e-mail]>; <[e-mail]> <[e-mail]>
>>>>>
>>>>>
>>>>> *Subject:* Re: [OAUTH-WG] Call for Adoption
>>>>>
>>>>
>>>>>
>>>>> I agree that the language describing OAuth issuer could and should be
>>>>> improved. The text now reads like it is the exact and full URL where the
>>>>> metadata/discovery document is located. Perhaps something more like "the
>>>>> URL from which the authorization server's configuration information
>>>>> location can be derived" and explain that adding the .well-known bits to
>>>>> the issuer is where the configuration information can actually be found.
>>>>>
>>>>> On Thu, Jan 21, 2016 at 7:07 PM, Nat Sakimura < <[e-mail]>
>>>>> [e-mail]> wrote:
>>>>>
>>>> Re: iss. I discussed this a bit with Nov in more details. It probably
>>>>> is a sloppy language of the specs that is making it difficult to read what
>>>>> you wanted to achieve.
>>>>>
>>>>>
>>>>>
>>>>> In mix-up-mitigation draft, OAuth issuer is "the URL of the
>>>>> authorization server's configuration information location". In OAuth
>>>>> discovery draft, there is something called "OAuth 2.0 Configuration
>>>>> Information Location URL", which is equal to "OpenID Connect Issuer".
>>>>>
>>>>> When I wrote the statement, I thought you were pointing to the URL
>>>>> that you can actually pull the configuration information by "the URL of the
>>>>> authorizaiton server's configuration information location" since otherwise
>>>>> you would have used the term "OAuth 2.0 Configuration Information Location
>>>>> URL". But after all, you probably meant these two are the same. Then I
>>>>> would strongly recommend to fix the language.
>>>>>
>>>>> Now, even If that is the case, the relationship like
>>>>>
>>>>> ·         iss + .well-know/openid-configuration = Connect OP config
>>>>> endoint
>>>>>
>>>>> ·         OAuth config endpoint - .well-known/openid-configuration =
>>>>> OAuth iss
>>>>>
>>>>> is very confusing.
>>>>>
>>>>>
>>>>>
>>>>> You also claim that your approach is simpler, but to me, your approach
>>>>> seem to be overly complex. It requires discovery and the check for the
>>>>> value of the discovered config information to work as the mitigation.
>>>>> (Right. Draft -01 does not have it, so it does not solve the mix-up issue.)
>>>>> With oauth-meta, you do not need it.
>>>>>
>>>>>
>>>>>
>>>>> Finally, your point that HATEOAS reminds you of WSDL, it is not. If
>>>>> you want to have something similar to WSDL in REST API area, it is Swagger.
>>>>> (Actually, it is gaining a lot of momentum recently, but that's beside the
>>>>> fact ;-). And the point here is not the format but the fact that we need to
>>>>> have a way to associate metadata to the data. The root cause of this mix-up
>>>>> attack is that the metadata and data is not associated properly. We have a
>>>>> standard way of associating the data and metadata with link-rel such as
>>>>> RFC5988 so why not use it? Link-rel-href pattern is used a lot now. Most
>>>>> modern web pages actually have it. Using a proper way to associate metadata
>>>>> with data will save you from a lot of other attacks in the future. Instead
>>>>> of doing patch works, we should solve it architecturally.
>>>>>
>>>>>
>>>>>
>>>>> Nat
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> 2016年1月22日(金) 10:34 Mike Jones < <[e-mail]>
>>>>> [e-mail]>:
>>>>>
>>>>> Nat, I’m confused by this statement in the message you reference “Unfortunately,
>>>>> this does not match the value of OAuth issuer defined in Section 2
>>>>> of draft-jones-oauth-mix-up-mitigation-01 nor the 'iss' returned as
>>>>> specified in 3.1. As such, validation as specified in bullet 1 of Section 4
>>>>> fails in Google's case -- OR it means that the document is internally
>>>>> inconsistent.”.  The issuer definition in draft-jones-oauth-discovery
>>>>> is 100% compatible with the one in OpenID Connect Discovery, by design.  In
>>>>> the Google example, both the OpenID issuer and the OAuth issuer values
>>>>> would be the string “ <https://accounts.google.com>
>>>>> https://accounts.google.com”.  What is the inconsistency that you
>>>>> perceive?
>>>>>
>>>>>
>>>>>
>>>>> The discussion of the duplication problem happened in the private
>>>>> meetings in Yokohama.
>>>>>
>>>>>
>>>>>
>>>>> I will admit, in Yokohama, I didn’t speak up in the public meetings to
>>>>> point out that a simpler alternative to oauth-meta was already being
>>>>> discussed there in private, because then I would have had to talk about the
>>>>> security issues, which we’d decided not to publicly do at that point.  So I
>>>>> stayed silent during the poll, out of politeness.  Perhaps I should have
>>>>> found a way to say something then anyway, but that’s water under the bridge
>>>>> now.
>>>>>
>>>>>
>>>>>
>>>>> Finally, for what it’s worth, the HATEOAS stuff reminds me far too
>>>>> much of Web Services Description Language (WSDL) – part of the complexity
>>>>> baggage that helped sink Web Services.  The use of “link rel” to define an
>>>>> interaction vocabulary and publish endpoints for that vocabulary seems
>>>>> pretty baroque and reminiscent of “microformats” – another cute “Webby”
>>>>> innovation that never caught on.  The industry has pretty much voted with
>>>>> its feet and is using JSON for publishing discovery data structures – not
>>>>> “link rel”.  I am against us reverting to the “link rel” proposal from 2008
>>>>> that never caught on when JSON is simpler and does a better job.
>>>>>
>>>>>
>>>>>
>>>>>                                                                 -- Mike
>>>>>
>>>>>
>>>>>
>>>>> *From:* Nat Sakimura [mailto: <[e-mail]>[e-mail]]
>>>>> *Sent:* Thursday, January 21, 2016 6:24 AM
>>>>> *To:* Justin Richer < <[e-mail]>[e-mail]>; Mike Jones <
>>>>> <[e-mail]>[e-mail]>
>>>>> *Cc:* William Denniss < <[e-mail]>[e-mail]>; <
>>>>> <[e-mail]>[e-mail]> < <[e-mail]>[e-mail]>
>>>>>
>>>>>
>>>>> *Subject:* Re: [OAUTH-WG] Call for Adoption
>>>>>
>>>>>
>>>>>
>>>>> Mike,
>>>>>
>>>>>
>>>>>
>>>>> You just criticize my draft. That's ok, but I would really like to get
>>>>> some response to my questions stated in
>>>>> <https://www.ietf.org/mail-archive/web/oauth/current/msg15483.html>
>>>>> https://www.ietf.org/mail-archive/web/oauth/current/msg15483.html .
>>>>> To me, it just does not seem to work, and the combination of the oauth-meta
>>>>> and PKCE seems to be much more elegan, nicer, and much simpler to
>>>>> implement. If you just give up the dynamic response at the authorization
>>>>> endpoint, then you even do not have to touch the code but just change a
>>>>> config file.
>>>>>
>>>>>
>>>>>
>>>>> Please convince me by answering to my questions.
>>>>>
>>>>>
>>>>>
>>>>> For the record of Yokohama, I do not recall much about duplication in
>>>>> OAuth session. The poll in the room was 19 for / zero against / 4
>>>>> persons need more information. Indeed, it is not duplicating much. And if
>>>>> you move to a new model without pre-configured discovery, it is not
>>>>> duplicating any but the resource endpoint URI, which is optional and is for
>>>>> the cases where the client did not know the concrete resource endpoint to
>>>>> start with.
>>>>>
>>>>>
>>>>>
>>>>> I understand your usecases always start from a concrete endpoint
>>>>> location. Mine do not. In a four party model, it is likely not. The user
>>>>> just want to have the service to fetch his data from some resource
>>>>> endpoint. He just hits the service. He does not hit the resource endpoint
>>>>> directly. For example, in the case of a consumer using a personal finance
>>>>> platform (PFP)to manage his pension fund, he hits the PFP and not the
>>>>> Pension fund. Assuming that the pension fund has delegated the
>>>>> authorization control to the authorization server, then, the authorization
>>>>> server should return both the access token and the endpoint of the pension
>>>>> fund so that the PFP can pull the data using them. A similar model holds
>>>>> for personal health service and health care providers.
>>>>>
>>>>>
>>>>>
>>>>> Best,
>>>>>
>>>>>
>>>>>
>>>>> Nat
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> 2016年1月21日(木) 21:18 Justin Richer < <[e-mail]>[e-mail]>:
>>>>>
>>>>> Convergence is exactly what I’m arguing for, though. These things
>>>>> ought to work together.
>>>>>
>>>>>
>>>>>
>>>>>  — Justin
>>>>>
>>>>> On Jan 21, 2016, at 2:55 AM, Mike Jones <
>>>>> <[e-mail]>[e-mail]> wrote:
>>>>>
>>>>>
>>>>>
>>>>> My memory of the discussions of the oauth-meta draft in Yokohama were
>>>>> that many people felt that it was unnecessarily dynamically duplicating a
>>>>> lot of information that the client already had.  Most of us that were aware
>>>>> of the attacks then were in favor of a more targeted, minimal approach.
>>>>> You were listened to in Yokohama, but that didn’t necessarily mean that
>>>>> people agreed with the approach.  Participants were already aware of the
>>>>> oauth-meta proposal in Darmstadt but no one spoke up in favor of it that I
>>>>> can recall.  Rather, I think people were thinking that “less is more”.
>>>>>
>>>>>
>>>>>
>>>>> There have also been discussions in the last day about how dynamically
>>>>> returning a resource URL, which oauth-meta does, is both unnecessary (since
>>>>> the client initiated the resource authorization already knowing what
>>>>> resource it wants to access) and often problematic, since many
>>>>> authorization servers can authorize access to multiple resources.  If
>>>>> anything, the client should be telling the authorization server what
>>>>> resource it wants to access – not the other way around.
>>>>>
>>>>>
>>>>>
>>>>> I’m not saying that there aren’t some good ideas in the oauth-meta
>>>>> draft – I’m sure there are, just as there are in the approach designed by
>>>>> the participants in Darmstadt.  While I volunteered to write the first
>>>>> draft of the mix-up-mitigation approach, it really reflects something a lot
>>>>> of people have already bought into – as evidenced in the passion in the
>>>>> high-volume “Mix-Up About The Mix-Up Mitigation” thread, and not just my
>>>>> personal project.
>>>>>
>>>>>
>>>>>
>>>>> If you think there are things missing or wrong in the
>>>>> mix-up-mitigation draft, please say what they are.  That will help us
>>>>> quickly converge on a solution that will work for everyone.
>>>>>
>>>>>
>>>>>
>>>>>                                                           Sincerely,
>>>>>
>>>>>                                                           -- Mike
>>>>>
>>>>>
>>>>>
>>>>> *From:* Nat Sakimura [ <[e-mail]>mailto:[e-mail]
>>>>> <[e-mail]>]
>>>>> *Sent:* Wednesday, January 20, 2016 11:17 PM
>>>>> *To:* Mike Jones < <[e-mail]>
>>>>> [e-mail]>; William Denniss < <[e-mail]>
>>>>> [e-mail]>; Justin Richer < <[e-mail]>[e-mail]
>>>>> >
>>>>> *Cc:* <[e-mail]>[e-mail]
>>>>> *Subject:* Re: [OAUTH-WG] Call for Adoption
>>>>>
>>>>>
>>>>>
>>>>> Hi Mike.
>>>>>
>>>>>
>>>>>
>>>>> Conversely, I would like to ask why this approach does not work for
>>>>> Mix-up attack. As Nov stated, we in fact have discussed the approach in
>>>>> quite a length back in Yokohama. I really would like to know why it does
>>>>> not work.
>>>>>
>>>>>
>>>>>
>>>>> Besides, for oauth-meta approach, mix-up attack is only one of the
>>>>> thing it solves.
>>>>>
>>>>>
>>>>>
>>>>> Nat Sakimura
>>>>>
>>>>>
>>>>>
>>>>> 2016年1月21日(木) 16:02 Mike Jones < <[e-mail]>
>>>>> [e-mail]>:
>>>>>
>>>>> Not to be negative, but I disagree with adopting
>>>>> draft-sakimura-oauth-meta.  We should define and promote one mitigation
>>>>> approach to the mix-up attacks.  Having two would confuse implementers and
>>>>> cause compatibility problems – reducing overall security.
>>>>>
>>>>>
>>>>>
>>>>> The approach defined in draft-jones-oauth-mix-up-mitigation was
>>>>> created in collaboration with the security researchers who identified the
>>>>> problems in the first place, was vigorously discussed in the security
>>>>> meeting Hannes and Torsten held in Darmstadt, and has been since refined
>>>>> based on substantial input from the working group.  And at least three
>>>>> implementers have already stated that they’ve implemented it.  I’m not
>>>>> saying that it’s, but if there are things missing or things that need to be
>>>>> improved in our approach, we should do it there, rather introducing a
>>>>> competing approach.
>>>>>
>>>>>
>>>>>
>>>>> Also, standard OAuth deployments register the client and then use the
>>>>> information gathered at registration time for subsequent protocol
>>>>> interactions.  They do not need all the configuration information for the
>>>>> authorization server to be retransmitted at runtime.  The oauth-meta draft
>>>>> goes too far in that direction, at least as I see it.  Returning things two
>>>>> ways creates its own problems, as discussed in the Duplicate Information
>>>>> Attacks security considerations section (7.2) of the mix-up-mitigation
>>>>> draft.
>>>>>
>>>>>
>>>>>
>>>>> I’ll note that the mix-up-mitigation approach is compatible with
>>>>> existing practice in both static and dynamic metadata discovery.  Replying
>>>>> to Justin’s comment that “It's the pre-configured discovery document
>>>>> that's at the root of the mix-up attack in the first place” – this is
>>>>> not the case.  The attacks can be performed without either discovery or
>>>>> dynamic registration.
>>>>>
>>>>>
>>>>>
>>>>> I would be interested in hearing a technical discussion on whether
>>>>> there are aspects of the oauth-meta approach that mitigate aspects of the
>>>>> attacks that the mix-up-mitigation approach does not.  That could help
>>>>> inform whether there are additional things we should add to or change in
>>>>> the mix-up draft.
>>>>>
>>>>>
>>>>>
>>>>>                                                           -- Mike
>>>>>
>>>>>
>>>>>
>>>>> *From:* OAuth [mailto: <[e-mail]>[e-mail]]
>>>>> *On Behalf Of *William Denniss
>>>>> *Sent:* Wednesday, January 20, 2016 10:37 PM
>>>>> *To:* Justin Richer < <[e-mail]>[e-mail]>
>>>>> *Cc:* <[e-mail]>[e-mail]
>>>>> *Subject:* Re: [OAUTH-WG] Call for Adoption
>>>>>
>>>>> +1 to adopt this, and I agree with Justin's comments.
>>>>>
>>>>> On Wed, Jan 20, 2016 at 9:53 PM, Justin Richer < <[e-mail]>
>>>>> [e-mail]> wrote:
>>>>>
>>>>> +1
>>>>>
>>>>> Inline discovery and pre-configured discovery (ie, .well-known) should
>>>>> at the very least be compatible and developed together. It's the
>>>>> pre-configured discovery document that's at the root of the mix-up attack
>>>>> in the first place.
>>>>>
>>>>>  -- Justin
>>>>>
>>>>>
>>>>>
>>>>> On 1/19/2016 10:30 PM, Nat Sakimura wrote:
>>>>>
>>>>> Just to give more context, at IETF 94, I have done a presentation on
>>>>> discovery.
>>>>>
>>>>>
>>>>>
>>>>> According to the minutes,
>>>>>
>>>>>
>>>>>
>>>>>     (f) Discovery (Nat)
>>>>>
>>>>>
>>>>>
>>>>>              Nat explains his document as an example of the work that has to be done
>>>>>
>>>>>              in the area of discovery, which is a topic that has been identified
>>>>>
>>>>>              as necessary for interoperability since many years but so far there
>>>>>
>>>>>              was not time to work on it. Mike, John and Nat are working on a new
>>>>>
>>>>>              document that describes additional discovery-relevant components.
>>>>>
>>>>>
>>>>>
>>>>>              Poll: 19 for / zero against / 4 persons need more information.
>>>>>
>>>>>
>>>>>
>>>>> The document discussed there was
>>>>> <https://tools.ietf.org/html/draft-sakimura-oauth-meta-05>
>>>>> <https://tools.ietf.org/html/draft-sakimura-oauth-meta-05>
>>>>> https://tools.ietf.org/html/draft-sakimura-oauth-meta-05. This is a
>>>>> simple (only 1-page!) but a very powerful document that nudges towards
>>>>> HATEOAS which is at the core of RESTful-ness. It also mitigates the Mix-up
>>>>> attack without introducing the concept of issuer which is not in RFC6749.
>>>>> It is also good for selecting different endpoints depending on the user
>>>>> authentication and authorization results and more privacy sensitive than
>>>>> pre-announced Discovery document. It also allows you to find to which
>>>>> protected resource endpoint you can use the access token against.
>>>>>
>>>>>
>>>>>
>>>>> In the last sentence of the minutes, it talks about "a new document
>>>>> that describes additional discovery-relevant components". This is
>>>>> <https://tools.ietf.org/html/draft-jones-oauth-discovery-00>
>>>>> https://tools.ietf.org/html/draft-jones-oauth-discovery-00.  It went
>>>>> for the call for adoption. However, it is only a half of the story. I
>>>>> believe  <https://tools.ietf.org/html/draft-sakimura-oauth-meta-05>
>>>>> https://tools.ietf.org/html/draft-sakimura-oauth-meta-05 that was
>>>>> discussed at IETF 94 and had support there should be adopted as well.
>>>>>
>>>>>
>>>>>
>>>>> Nat Sakimura
>>>>>
>>>>>
>>>>>
>>>>>