[Openid-specs-ab] First Release Candidates for final OpenID Connect specifications
Michael.Jones at microsoft.com
Mon Oct 21 04:28:33 UTC 2013
Replies are inline. Some of these are questions to the working group, so everyone, please read them.
From: Torsten Lodderstedt [mailto:torsten at lodderstedt.net]
Sent: Friday, October 18, 2013 1:16 PM
To: Mike Jones
Cc: openid-specs-ab at lists.openid.net; board at openid.net
Subject: Re: [Openid-specs-ab] First Release Candidates for final OpenID Connect specifications
@Mike: first of all let me thank you for taking the burden and rework the whole document structure. I think the new structure is a major leap forward for OpenID Connect.
I focused my review on the new core specification. Please find my comments below.
I won't attend IIW but will attend IETF-88. So, if needed, we can talk through my comments there.
I would suggest to a reference to RFC 6749 in the first sentence. It probably also makes sense to explicitly point out that the reader is expected to be familiar with RFC 6749 and RFC 6750 as well as other IETF I-Ds (notably JOSE, JWT and JWT Assertion Profile).
The flow description is a good starting point for readers. I would suggest to add the following information in this section:
- OpenID Connect authentication is basically an extension to the standard OAuth authorization process. This extension is defined for most OAuth grant types.
- Clients wishing to acquire identity information indicate this by sending the scope value "openid" as part of the authorization request parameters. (There are much more parameters used to control the process but this is the "main switch".)
- Such a client is also called relying party (RP). An authorization server also supporting OpenID Connect is called OpenID Provider (OP).
Adding this information will help the reader to understand the way connect utilizes/integrates into OAuth.
I would also suggest to move the definition (syntax and contents) of the ID Token here and make it section 1.4 because this is THE core concepts used throughout the specification. It's introduction in section 220.127.116.11 is to late (in my opinion) because it is cited roughly 20 times in previous sections.
What do people think of defining the ID Token earlier, and in a higher-level section? I think Torsten is probably right about this.
"The Authorization Code Flow is suitable for Clients that can securely maintain a Client Secret between themselves and the Authorization Server ..." - this is confusing since public clients can use the code as well. The key benefits of this grant type in my opinion are:
- AS _can_ authenticate clients
- AS _can_ return refresh tokens
- simplest way for web application backends to acquire tokens
That's why is best suited for web applications and native apps.
"The Authorization Code Flow is appropriate for web applications and native apps as it allows to authenticate clients and obtain refresh tokens whereas the implicit flow does not support these features."
Or just remove the assessment of OAuth grant types and leave it to the implenentors to carry out their assessment.
Torsten, I'm curious what your objection to the statement "The Authorization Code Flow is suitable for Clients that can securely maintain a Client Secret between themselves and the Authorization Server ..." is. I say that, because unless the client can keep the client_secret a secret, client impersonation is possible.
2.1. Authentication using the Authorization Code Flow
OLD: "This provides the benefit of not exposing the Access Token to the Resource Owner ..."
The same indeed holds for the ID Token, which is more important from a security perspective.
NEW: "This provides the benefit of not exposing the Access Token and the ID Token to the Resource Owner ..."
NEW (alternative): "This provides the benefit of not exposing any Token to the Resource Owner ..."
2.1.1. Authorization Code Flow Steps
OLD: "8. Client validates the tokens and retrieves the End-User's subject identifier."
I assume the client is supposed to validate the ID token, only?
NEW: "8. Client validates the ID token and retrieves the End-User's subject identifier."
18.104.22.168. Authorization Request
"When the Client wishes to access a Protected Resource and the End-User Authorization has not yet been obtained, the Client prepares an Authorization Request to the Authorization Endpoint" - Why is this relevent in this context? I suggest to remove this sentence.
"An Authorization Request is a message sent from an RP to the OP's Authorization Endpoint. It is an extended OAuth 2.0 [RFC6749] Authorization Request. Section 4.1.1 and 4.2.1 of OAuth 2.0 [RFC6749] define the OAuth 2.0 Authorization Request parameters." - Why Authorization Request? Shouldn't this be called an "Authentication Request"?
"Communication with the Authorization Endpoint MUST utilize TLS. See Section 15.17 for more information on using TLS.
Authorization Servers MUST support the use of the HTTP GET and POST methods defined in RFC 2616 [RFC2616] at the Authorization Endpoint.Clients MAY use the HTTP GET or POST methods to send the Authorization Request to the Authorization Server. If using the HTTP GET method, the request parameters are serialized using URI Query String Serialization, perSection 12.1. If using the HTTP POST method, the request parameters are serialized using Form Serialization, per Section 12.2."
Seems to be standard OAuth stuff, I suggest to remove it.
Yes, some of the spec repeats standard OAuth stuff. Particularly for security related information, such as the requirement to use TLS, etc., it was thought to be better to repeat it. Repeating the HTTP POST stuff, etc. is just a convenience to the reader. Obviously if we say something that conflicts with OAuth, that's a problem...
- redirect_uri Parameter
"This URI MUST exactly match one of the redirect_uris registered for the Client" - Why is the redirect_uri validation more restrictive than http://tools.ietf.org/html/rfc6749#section-22.214.171.124?
The redirect_uri validation is more restrictive than http://tools.ietf.org/html/rfc6749#section-126.96.36.199 to make it simpler and thereby increase the chance of it being done correctly. (Doing this wrong has been the biggest source of OAuth security problems that I'm aware of.)
"When using this flow, the redirection URI MAY use the http scheme, provided that the Client Type is confidential, as defined in Section 2.1 of OAuth 2.0; otherwise, it MUST use the https scheme" - This, surprisingly enough, is relaxed in comparison to http://tools.ietf.org/html/rfc6749#section-10.5.
RFC 6749 states: "Authorization codes operate as plaintext bearer credentials, used to verify that the resource owner who granted authorization at the authorization server is the same resource owner returning to the client to complete the process. Therefore, if the client relies on the authorization code for its own resource owner authentication, the client redirection endpoint MUST require the use of TLS."
Why is Connect, in this particular case, less restrictive than OAuth?
John, can you speak to why we're allowing http redirect_uri values when apparently OAuth doesn't?
- nonce Parameter
"One method to achieve this is to store a random value as a signed session cookie, and pass the value in the nonce parameter. In that case, the nonce in the returned ID Token can be compared to the signed session cookie to detect ID Token replay by third parties." - I would recommend to move this text into an "implementation note" section
id_token_hint Parameter - "Previously issued ID Token passed to the Authorization Server .." issued by the AS being requested? or any AS? I assume by the same AS
NEW: "ID Token previously issued by this Authorization server to the client ..."
"... it SHOULD return a login_required error." - Does this mean the OP shall try to authenticate the user account identified by the ID token and refuses authentication otherwise? This sounds more like a requirement than a hint.
"When possible, an id_token_hint SHOULD be present when prompt=none is used and an invalid_request error MAY be returned if it is not; however, the server SHOULD respond successfully when possible, even if it is not present." - Why is the login hint recommended for this prompt value? checkid_immediate in OpenID 2.0 worked very well w/o a hint?
Can someone please answer this one?
188.8.131.52. Authorization Request Validation
"3. If the sub (subject) Claim is requested with a specific value for the ID Token ...." The meaning of the text is unclear to me. How is a specific sub value requested? by the login_hint or the id_token_hint?
"As specified in OAuth 2.0 [RFC6749], Authorization Servers SHOULD ignore unrecognized request parameters.
If the Authorization Server encounters any error, it MUST return an error response."
Standard OAuth stuff, I recommend to remove it.
184.108.40.206. Authorization Server Obtains End-User Consent/Authorization
"When permitted by the request parameters used, this MAY be done through an interactive dialogue with the End-User ..." - What if the parameters do not allow for an interactive dialogue, e.g. prompt==none? I assume an error response with return code consent_required or interaction_required is appropriate. I would prefer interaction_required because to RP does not need to know more.
220.127.116.11. Authorization Successful Response
This is a vanilla OAuth 2.0 response, right? I would suggest to just say so.
BTW: This piece of text is not applicable to the code grant type: "This specification only describes OAuth 2.0 Bearer Token Usage [RFC6750]. The OAuth 2.0 response parameter token_type MUST be set to Bearer unless another Token Type has been negotiated with the Client."
I don't understand this comment, since a token_type value is returned from the Token Endpoint.
2.1.3. Tokens Endpoint
"Clients MUST use the HTTP POST method to make requests to the Token Endpoint. Request parameters are added using Form Serialization, per Section 12.2. The Token Endpoint MUST support the use of the HTTP POST method defined in RFC 2616 [RFC2616] at the Token Endpoint.
Communication with the Token Endpoint MUST utilize TLS. See Section 15.17 for more information on using TLS.
All Token Endpoint responses that contain tokens, secrets, or other sensitive information MUST include the following HTTP response header fields and values: ..."
This seems to be standard OAuth stuff. I recommend to remove it.
18.104.22.168. Token Request
"To obtain an ID Token, Access Token, or Refresh Token, the Client MUST authenticate to the Token Endpoint using the authentication method registered for its client_id, as described in Section 8 ..." - At this point the reader is not familiar with the different authentication methods supported by an OpenID OP. I therefore suggest to move the client authentication section before the authentication section (e.g. make it a section 1.5).
What do people think of this suggestion? I understand that it makes logical sense, but if you read Section 8, it introduces a bunch of stuff that would interrupt the flow of just understanding how the Code Flow works.
22.214.171.124. Token Request Validation
The whole sections seems to re-phrase standard OAuth stuff. I recommend to remove it.
126.96.36.199. Token Successful Response
"Servers SHOULD support OAuth 2.0 Bearer Token Usage [RFC6750] for interoperability" - I think this topic is better covered in the MTI section.
"the following parameters MUST be included in the response if the grant_type value is authorization_code and the Authorization Request scope parameter contains openid:" - Seems to be redundant since this whole section is about exactly this use case. I recommend to remove this text. Same holds true for the following text
"An id_token MUST be returned when the grant_type value is authorization_code and MAY be returned when other grant types are used."
The sentence above is definitely not "standard OAuth stuff", as it adds the requirement to return the ID Token.
188.8.131.52. ID Token
That's the key elements of OpenID Connect! As already stated, I recommend to move the (full) description of its content and syntax to section 1 (section 1.4). I think this will facilitate readability. The validation rules should stay in the sections of the respective grant types.
I was thinking of making it 2.1.2 - putting it before the Authorization Endpoint stuff. What do others think? It's not really introduction or overview stuff - it's part of the meat of the specification.
184.108.40.206. ID Token Validation
"1. If the Client has provided an id_token_encrypted_response_alg parameter during Registration, decrypt the ID Token using the key pair specified during Registration." - text depends on dynamic registration and should therefore be generalized.
NEW: "1. If the ID Token is encrypted, the Client first decrypts it using the key agreed upon with the authorization server."
"5. If the id_token is received via direct communication between the Client and the Token Endpoint, the TLS server validation MAY be used to validate the issuer in place of checking the token signature. The Client MUST validate the signature of all other ID Tokens according to JWS [JWS] using the algorithm specified in the alg parameter of the JWT header." - text seems rather generic. As this is about the code flow, the ID token is received via direct communication, so the text can be simplified.
NEW: "5. Since the ID token is received via direct communication, the TLS server validation MUST be used to validate the issuer in place of checking the token signature."
Steps 6-8 can be removed.
220.127.116.11. Access Token Validation
I would recommend to add the text from section 18.104.22.168. because this is the first point where the concept is used.
a_hash validation is mentioned. What's about c_hash validation?
As I understand it, the c_hash is only needed for hybrid flows, in which the ID Token is returned in a fragment. That's why it's introduced there. Do others agree with this decision?
2.2. Authentication using the Implicit Flow
"...which may expose them to the Resource Owner and other applications that have access to the Resource Owner's User-Agent." - I recommend to add: "In contrast to the authorization code flow, this requires additional security mechanisms (described below) to detect falsified ID tokens."
22.214.171.124. Authorization Request
see comments regarding redirect_uri and nonce for section 126.96.36.199
188.8.131.52 - 184.108.40.206 - I recommend to state in section 2.2.2 that those steps are performed in the same manner as for the code flow and to remove these section?
These are here to keep the structure of the sections parallel with 2.1.
220.127.116.11. Authorization Successful Response
access_token Parameter - "Access Token for the UserInfo Endpoint." - In contrast to section 18.104.22.168, this section also described standard OAuth response parameters. I don't think this is needed. Moreover, the term "User Info" is used before it is really introduced. In my opinion, authentication should not talk about user info. The access token returned as part of the authentication result might suited for interactions with any protected resource, including user info.
22.214.171.124. Redirect URI Fragment Handling
This section needs a bit more of context and description. While the introduction of 2.2 states: "The Implicit Flow is mainly used by Clients implemented in a browser using a scripting language", this section suddenly _requires_ the client to send data to a web server ("The Client MUST provide ...").
First of all, I don't understand the MUST.
This could become "The Client needs to implement...".
Second, a description is needed of the different patterns, scripting clients vs. scripted frontends of server-based web-application (here the implicit grant offers better performance and scalability at the cost of a more complex client implementation. Therefore, implementation advices are given.
I also think this section should be part of an "implementation note" section.
126.96.36.199. ID Token Validation
Text around signature validation should be moved from 188.8.131.52 to this section as it is required for the implicit grant (in contrast to code).
2.3. Authentication using the Hybrid Flow
184.108.40.206.-220.127.116.11. can be removed
18.104.22.168. Authorization Successful Response
Again, I would recommend to focus on additional response parameters, as in section 22.214.171.124
126.96.36.199. Redirect URI Fragment Handling
see comment at section 188.8.131.52
I think 184.108.40.206.-220.127.116.11 can be removed - Instead it should be stated that for the hybrid flow client and AS must conform to ALL requirements for code and implicit.
These are here to keep the structure of the sections parallel with 2.1.
3. Initiating Login from a Third Party
I assume this is mainly intended to support OP initiated logins? I don't think it deserves a top-level section. I would recommend to make it part of the Authentication section.
"The Client MAY optionally register [OpenID.Registration] an initiate_login_uri that can be used by the Authorization Server or another party to initiate a login for an End-User at the Client." I assume this feature shall also be supported by OPs w/o dynamic registration? I therefore suggest to move this text to the dynamic registration spec. Instead one could state: "The approach utilized by the 3rd party or the OP to determine the client's respective URL is out-of-scope of this document."
Generally, I don't think any meta data element registered via dynamic registration or discovered via OpenID Discovery should be specified/mentioned in the core spec. I think the core function must work w/o both functions in place.
I agree that it shouldn't be mentioned in a way that implies or states that Discovery or Registration are required. That being said, we've tried to point people to the relevant Discovery & Registration parameters where they're relevant to help implementers more easily grasp the bigger picture.
"This section specifies how the Client can obtain Claims about the End-User ..." claims about the authentication process are supported as well.
NEW: "This section specifies how the Client can obtain Claims about the End-User and the authentication process ...
4.1. Requesting Claims using Scope Values
This is an extension to the authentication part, it should be specified that way. For example, there is no need to specify the use of the scope value "openid" again. IMHO it suffices to say that clients may request access to user data by adding more scope values in conjunction with "openid".
4.2. Standard Claims
I think this section should be the first section as it describes standard claims on a conceptual level and which ways exist for a client to retrieve them.
This seems reasonable. Do others agree with this reordering?
4.3. UserInfo Endpoint
I think this section should be merged with Section 4.1 as the scope values defined there control access to this resource, only.
Disagree, since 4.5 (Requesting Claims using the "claims" Request Parameter) also controls the access to this resource.
4.4. Requesting Claims Locales with the "claims_locales" Request Parameter
I would suggest to move this either before 4.1. or after 4.5. as it seems to be orthogonal to the functions described there.
If we move the Standard Claims to 4.1, I agree that this could logically become 4.2
4.6. Claim Types
This seems to be out of order because after a description of how a client may allocate claims to ID token and user info, this section again exclusively talks about UserInfo. Is it really the intention to support aggregated and distributed claims at the User Info endpoint, only? If so I recommend to move this section before 4.4. Otherwise, please state that such claims can be provided in the ID Token as well.
Fair point. While I understand the general argument that this might apply to the ID Token as well, from a practical matter, they're only likely to ever be used with UserInfo claims.
How is a aggregated or distributed claim requested by a client?
It's up to the OP when and whether to provide them - not up to the RP to ask. I'll try to make that clearer.
5. Passing Request Parameters as JWTs
I would suggest to move this topic into another document. The features described here allow implementors to achieve higher security levels and may reduce the request size but I consider them beyond the scope of a core document.
We considered moving lots of things out to separate documents, but in the end, decided that things would be easier for developers to understand if we avoided a proliferation of documents.
6. Self-Issued OpenID Provider
How mature is the concept? How many implementations exist? Is this really part of a core specification or rather an extension for mobile/personal devices? I would opt for making this section a separate document.
Same response as for 5. And yes, there are implementations.
6.2. Self-Issued OpenID Provider Registration
"When using a Self-Issued OP, the Client is deemed to have registered with the OP and obtained following Client Registration Response." - Does this mean dynamic registration is required for self-issued ID providing?
7. Subject Identifier Types
This section is completely about discovery and registration - it should go there, consequently.
No, it's about core functionality - the meaning of the subject claim in the ID Token.
8. Client Authentication
This section provides fundamental information about the client authentication methods supported/introduced by OpenID Connect. I would suggest to move it into section 1 (e.g. section 1.5).
Section 1 is for introductory material - not specifying requirements. We could move it earlier, if you feel that it's warranted. Where would like you like it to go, other than Section 1?
"During Client Registration, the RP (Client) MAY register an authentication method. " I assume all client authentication methods shall work even if the OP does not support dynamic registration. Consequently, this text must be replaced by a general description of what methods are supported and how the OP is supposed to determine the right method for a particular client.
OLD: "During Client Registration, the RP (Client) MAY register an authentication method. If no method is registered, the default method of client_secret_basic MUST be used."
NEW: "OpenID connect supports the authentication methods listed below. The authorization server determines the authentication method to be used in a particular authorization transaction based on the client_id. The way client and authorization server negotiate the authentication method is out of scope of this specification."
Actually, the Authorization Server advertises what methods it supports and the Client chooses the method it uses from among that set.
9. Signatures and Encryption
Most of this section talks about discovery and dynamic registration - it should go there, consequently.
Things like key rollover are Core functionality. The references to the discovery and registration parameters are there to aid developers in understanding the bigger pictures. The use of them is optional. (If you believe that the text isn't clear on this, please propose specific language changes to clarify this.)
I think the core spec needs MTI algorithms and clear definition when signatures are required. In my opinion, there are two use cases:
- Login via implicit grant
- azp - Login at another AS via ID token
In both cases, Signing ID Tokens with RSA SHA-256 could be good baseline. This could be documented in the ID Tokens section or the MTI section.
14.1 (Mandatory to Implement Features for All OpenID Providers) already does specify RS256 as MTI. Is there something else you want us to say in this regard?
9.3.1. Rotation of Asymmetric Signing Keys
Isn't this an implementation note?
No, it's normative
10. Offline Access
Given the description (" that grants access to the End-User's UserInfo Endpoint ..."), I would say this text can go to the User Info section.
11. Using Refresh Tokens
I think this should go to the authentication section (2.4?), as it describes usage of the refresh token grant type in the Connect context.
It's optional functionality, whereas Authentication is not. Connect implementations do not need to support refresh tokens.
11.2. Refresh Successful Response
"If an ID Token is returned as a result of a token refresh request ..." - Can't we specify the conditions under which an ID token is returned? Otherwise, an interoperable client does not know what to expect or how to control the outcome of this request. For the standard authentication flow, presence of the scope value "openid" is the trigger. I would suggest to use the same logic for the refresh token grant type.
The refresh functionality is outside the scope of the spec. The current language is there to place constraints on what it needs to do, if supported.
14. Implementation Considerations
I would suggest to move MTI requirements to another document as it refers to nearly every document of the OpenID Connect suite. For the core document, I would suggest to (at most) specify the MTI requirements for closed systems, only:
- code flow extension
The logic behind the current organization is that specifying both the closed and open MTI requirements in one place makes them easier for developers to understand.
14.1. Mandatory to Implement Features for All OpenID Providers
"Signing ID Tokens with RSA SHA-256" - I think, in Berlin, we agreed to not require signatures for simple, static setups using authz code. We agreed to move this to the MTI section for dynamic OPs.
Actually, what we agreed was to allow signing to be optional for the Code flow, provided the client registers saying that it wants "none". This is now in the spec.
14.5. Compatibility Notes
Isn't such information better covered on the working group page. Such information typically change more often than the specification itself.
Developers are more likely to read the spec than other working group pages. Again this is there for developer convenience, to help them understand the whole context as easily as possible.
14.6. Related Specifications and Implementer's Guides
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Openid-specs-ab