The OAuth 2.0 Authorization Framework: JWT Pop Token Usage

This document specifies the method for the client to use a proof-of-possestion token against a protected resource. The format of such token is defined in section 3 of .

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119]. Unless otherwise noted, all the protocol parameter names and values are case sensitive.

For the purpose of this document, the terms defined in and are used.

JWT PoP token is a JWS signed JWT whose payload is a JWT Claims Set. The JWT claims set MUST include the following: The issuer identifier of the auhtorization server. The identifier of the resource server. The issuance time of this token. The expiry time of this token. The confirmation method. Their semantics are defined in and .

Following is an example of such.

There are several varieties of sender constrained token. Namely: DN Constrained Token Client ID Constrained Token

DN constrained token is typically used when X.509 client certificate authentication is used at the token endpoint. In this case, the constraint is expressed by including the following member at the top level of cnf claim. The Distinguished Name of the client certificate as a string that the client used in the authorization request. The authorization server finds the relevant DN from the X.509 client certificate authentication that is performed at the token endpoint.

The constraint in the Client ID constrained token is expressed by including the following member at the top level of cnf claim. The client_id of the client that the client used in the authorization request. The combination of the "iss" of the access token and this value forms a globally unique identifier for the client. The authorization server finds the client ID from the client ID used in the client authentication at the token endpoint.

Methods to express key constraints are extensively described in the section 3 of . Such cnf claim is used in the access token described in section 3 to form a key constrained token. defines 4 confirmation methods. JSON Web Key Representing a Public Key Encrypted JSON Web Key Thumbprint of a JWK using the SHA-256 hash function. X.509 Certificate SHA-256 Thumbprint JWK Set URL The client provides the corresponding keys or the pointers to the authorization server as a part of the client configuration. It can be done through out-of-band methods (e.g., developper portal) or through some form of dynamic registration, etc. Following is an example of a JWT payload containing a JWK with a raw key.

Following is an example of a JWT payload containing a jku URI.

Following is an example of a JWT payload containing a x5t#s256 Certificate Thumbprint of a x509 certificate. .

The resource server that supports this specification MUST authenticate the Client by having it demonstrate that it is the holder of the key associated with the access token being used. The confirmation method can be broadly categorized in two forms. A method leveraging on the X.509 client certificate authentication of the TLS connection. cn, x5t#s256, and jku confirmation methods can be used with this access method. (The JWKS referenced by the jku MUST contain JWK with x5c certificate elements for this access method) A method leveraging the signature on the nonce. cid, jku, jwk, jwe, and, jwkt#S256 confirmation methods can be used with this access method.

Under this method, X.509 client certificate authentication at the resource endpoint is being leveraged. The resource endpoint MUST obtain the DN of the client certificate used for the authentication and MUST verify that the value of the dn member in the cnf member matches with it. If it does not match, the process stops here and the resource access MUST be denied. If it is valid, then the resource server MUST verify the access token. If it is valid, the resource SHOULD be returned as HTTP response. Under this method, X.509 client certificate authentication at the resource endpoint is being leveraged. The resource endpoint MUST obtain the client certificate used for the authentication and MUST verify that the base64url-encoded SHA-256 thumprint of the DER encoded X.509 client certificate. The x5t#s256 member in the cnf member MUST exactly match the calculated thumbprint. If it does not match, the process stops here and the resource access MUST be denied. If it is valid, then the resource server MUST verify the access token. If it is valid, the resource SHOULD be returned as HTTP response. Under this method, X.509 client certificate authentication at the resource endpoint is being leveraged. The resource endpoint MUST obtain the client certificate used for the authentication and MUST verify that the certificate matches one of the x5c elements retrieved from the JWKS. Each x5c element may contain a chain of base64-encoded certificates. The client certificate MUST only be compared with the last certificate in the chain. If it does not match, the process stops here and the resource access MUST be denied. If it is valid, then the resource server MUST verify the access token. If it is valid, the resource SHOULD be returned as HTTP response.

For this, the following steps are taken: The client accesses the protected resource and gets an authorization error as in . With it, the client obtains a nonce. The client prepares a client nonce, cnonce, and nounce count, nc as defined in section 3.2.2 of . The client creates JWS compact serialization over the nonce. To obtain it, first create a JSON with a name "nonce" and the value being what was received in the previous step. The JWS MUST contain a kid header element if the client has more than one signing key published via JWKS URI e.g.,

Then, jws-on-nonce is obtained by creating a compact serialization of JWS on this JSON. The client sends the request to the resource server, this time with Authorization Request Header as defined in section 4.2 of with the credential as follows:

In the following example, the access token and the jws-on-nonce are represented as access.token.jwt and jws.of.nonce for the sake of brevity.

The resource server finds the client's public key form the access token through the methods described in . The resource server MUST verify the value of s of the Authorization header. If it fails, the process stops here and the resource access MUST be denied. The resource server MUST verify the access token. If it is valid, the resource SHOULD be returned as HTTP response.

If the client requests the resource without the proper authoization header, the resource server returns a HTTP 401 response with WWW-Authenticate header as defined in section 4.1 of with the challenge as follows:

Following example depicts what the response would look like.

A new scheme has been registered in the HTTP Authentication Scheme Registry as follows: Authentication Scheme Name: Jpop Reference: Section 3 of this specification Notes (optional): The Named Authentication scheme is intended to be used only with OAuth Resource Access, and thus does not support proxy authentication.

Confirmation Method Value: "dn" Confirmation Method Description: DN match with the TLS client auth. Change Controller: IESG Specification Document(s): This document. Confirmation Method Value: "cid" Confirmation Method Description: Client ID Confirmation Change Controller: IESG Specification Document(s): This document.

The "dn" JWT confirmation method relies its security property on the X.509 client certificate authentication. In particular, the validity of the certificate needs to be verified properly. It involves the traversal of all the certificate chain and the certificate validation (e.g., with OCSP).

The client's secret key must be kept securely. Otherwise, the notion of PoP breaks down. It should be noted that JWE confirmation method is significantly weaker form of the PoP, as the resource server and the authorization server can masquerade as the client.

When using the signature method the client must specify to the AS the aud it intends to send the token to, so that it can be included in the AT. A malicious RS could receive a AT with no aud or a logical audience and then replay the AT and jws-on-nonce to the actual server. NOTE another approach would be to include the resource in the jws-on-nonce

When a AS uses dynamic client registration it may accept software statements supplied by a federation operator. Those software statements can contain a JWKS-URI that is hosted by the federation operator or protected by a certificate provisioned from a trusted root. These methods would allow the federation operator to administratively revoke the keys at the JWKS-URI without requiring the JWKS to contain x5c elements with CA issued certificates and having to have the RS perform full certificate validation for each request.

The authors thank the following people for providing valuable feedback to this document. Nov Matake (YAuth).