JAX-RS: OAuth2

Introduction

New:

  • CXF 3.0.0-milestone2 makes it simpler to encrypt the token and other OAuth2 server model state
  • OAuthDataProvider has a new revokeToken() method added  to support the client-driven token revocation
  • AccessTokenValidationService has been enhanced to ensure the validation can proceed only if the authenticated Principal is available.   
  • OAuthRequestInterceptor has been added to make it easier for non-JAXRS endpoints to use OAuth2 tokens

CXF provides the implementation of OAuth 2.0. See also the JAX-RS OAuth page for information about OAuth 1.0.

Authorization Code, Implicit, Client Credentials, Resource Owner Password Credentials, Refresh Token and SAML2 Assertions grants are currently supported.

Custom grant handlers can be registered.

OAuth2 is a new protocol which offers a complex yet elegant solution toward helping end users (resource owners) authorize third-party providers to access their resources.

The OAuth2 flow which is closely related to the original OAuth 1.0 3-leg flow is called Authorization Code and involves 3 parties: the end user, the third party service (client) and the resource server which is protected by OAuth2 filters. Typically a client offers a service feature that an end user requests and which requires the former to access one or more protected resources on behalf of this user which are located at the resource server. For example, the client may need to access the end user's photos in order to print them and post to the user or read and possibly update a user's calendar in order to make a booking.

In order to make it happen, the third-party service application/client needs to register itself with the OAuth2 server. This happens out-of-band and after the registration the client gets back a client key and secret pair. Typically the client is expected to provide the name and description of the application, the application logo URI, one or more redirect URIs, and other information that may help the OAuth2 authorization server to identify this client to the end user at the authorization time.

From then on, the authorization code flow works like this:
1. End User requests the third-party service using a browser.

2. The client redirects the end user to OAuth2 Authorization Service, adding its client id, the state, redirect URI and the optional scope to the target URI. The state parameter represents the current end user's request, redirect URI - where the authorization code is expected to be returned to, and the scope is the list of opaque permissions that the client needs in order to access the protected resources.

3. Authorization Service will retrieve the information about the client using its client id, build an HTML form and return it to the end user. The form will ask the user if a given third-party application can be allowed to access some resources on behalf of this user.

4. If the user approves it then Authorization Service will generate an authorization code and redirect the user back to the redirect uri provided by the client, also adding a state parameter to the redirect URI.

5. The client requests an access token from OAuth2 Access Token Service by providing an authorization code grant.

6. After getting an access token token, the service finally proceeds with accessing the current user's resources and completes the user's request.

As you can see the flow can be complex yet it is very effective. A number of issues may need to be taken care along the way such as managing expired tokens, making sure that the OAuth2 security layer is functioning properly and is not interfering with the end user itself trying to access its own resources, etc.

Please check the specification and the Wikipedia article as well as other resources available on the WEB for more information you may need to know about OAuth2.

CXF JAX-RS gives the best effort to making this process as simple as possible and requiring only a minimum effort on behalf of OAuth2 server developers. It also offers the utility code for greatly simplifying the way the third-party application can interact with the OAuth2 service endpoints.

Maven dependencies

Client Registration

Client Registration is typically done out of band (OAuth2 experts are also finalizing the dynamic client registration).
The client registration service will offer an HTML form where the clients will enter their details, see a Client bean for the currently supported properties.
Note CXF may offer an abstract client registration service in the future to minimize the effort to get the custom registration service created from scratch.

Developing OAuth2 Servers

OAuth2 server is the core piece of the complete OAuth2-based solution. Typically it contains 2 services for:
1. Authorizing request tokens by asking the end users to let clients access some of their resources and returning the
grants back to the client (Authorization Service)
2. Exchanging the token grants for access tokens (Access Token Service)

CXF offers several JAX-RS service implementations that can be used to create the OAuth2 servers fast: AuthorizationCodeGrantService and ImplicitGrantService for managing the redirection-based flows, as well as AccessTokenService for exchanging the grants for new tokens.

Note that some grants that do not require the redirection-based support, such as SAML2 one, etc, may only require an Access Token Service be operational.

All of these services rely on the custom OAuthDataProvider which persists the access tokens and converts the opaque scope values to the information that can be presented to the users. Additionally, AuthorizationCodeDataProvider is an OAuthDataProvider which can keep temporary information about the authorization code grants which needs to be removed after the tokens are requested in exchange.

Writing your own OAuthDataProvider implementation is what is needed to get the OAuth2 server up and running. In many cases all you need to do is to persist or remove the Authorization Code Grant data, use one of the available utility classes to create a new access token and also persist it or remove the expired one, and finally convert the optional opaque scope values (if any are supported) to a more view-able information.

Authorization Service

The main responsibility of OAuth2 Authorization Service is to present an end user with a form asking the user to allow or deny the client accessing some of the user resources. CXF offers AuthorizationCodeGrantService and ImplicitGrantService for accepting the redirection requests, challenging the end users with the authorization forms, handling the end user decisions and returning the results back to the clients.

One of the differences between the AuthorizationCode and Implicit flows is that in the latter case the grant is the actual access token which is returned as the URI fragment value. The way the end user is asked to authorize the client request is similar between the two flows. In this section we will assume that the Authorization Code flow is being exercized.

A third-party client redirects the current user to AuthorizationCodeGrantService, for example, here is how a redirection may happen:

The client application asks the current user (the browser) to go to a new address provided by the Location header and the follow-up request to AuthorizationCodeGrantService will look like this:

Note that the end user needs to authenticate. The Request URI includes the client_id, custom scope value, response_type set to 'code', the current request state and the redirect uri. Note the scope is optional - the Authorization Service will usually allocate a default scope; however even if the client does include an additional custom scope the end user may still not approve it. The redirect uri is also optional, assuming one or more ones redirect URIs have been provided at the client registration time.

AuthorizationCodeGrantService will report a warning is no secure HTTPS transport is used:

It can also be configured to reject the requests over un-secure HTTP transport.

AuthorizationCodeGrantService will retrieve the information about the client application to populate an instance of OAuthAuthorizationData bean and return it. OAuthAuthorizationData contains application name and URI properties, optional list of Permissions and other properties which can be either presented to the user or kept in the hidden form fields in order to uniquely identify the actual authorization request when the end user returns the decision.

One important OAuthAuthorizationData property is "authenticityToken". It is used for validating that the current session has not been hijacked - AuthorizationCodeGrantService generates a random key, stores it in a Servlet HTTPSession instance and expects the returned authenticityToken value to match it - this is a recommended approach and it also implies that the authenticityToken value is hidden from a user, for example, it's kept in a 'hidden' form field. The other properties which are meant to be hidden are clientId, state, redirectUri, proposedScope.

The helper "replyTo" property is an absolute URI identifying the AuthorizationCodeGrantService handler processing the user decision and can be used by view handlers when building the forms or by other OAuthAuthorizationData handlers.

So the populated OAuthAuthorizationData is finally returned. Note that it's a JAXB XMLRootElement-annotated bean and can be processed by registered JAXB or JSON providers given that AuthorizationCodeGrantService supports producing "application/xml" and "application/json" (See the OAuth Without Browser section below for more). But in this case we have the end user working with a browser so an HTML form is what is really expected back.

AuthorizationCodeGrantService supports producing "text/html" and simply relies on a registered RequestDispatcherProvider to set the OAuthAuthorizationData bean as an HttpServletRequest attribute and redirect the response to a view handler (can be JSP or some other servlet) to actually build the form and return it to the user. Alternatively, registering XSLTJaxbProvider would also be a good option for creating HTML views.

Assuming RequestDispatcherProvider is used, the following example log shows the initial response from AuthorizationCodeGrantService:

Note that a "/forms/oauthAuthorize.jsp" view handler will create an HTML view - this is a custom JSP handler and whatever HTML view is required can be created there, using the OAuthAuthorizationData bean for building the view. Most likely you will want to present a form asking the user to allow or deny the client accessing some of this user's resources. If OAuthAuthorizationData has a list of Permissions set then adding the information about the permissions is needed.

Next the user makes a decision and selects a button allowing or denying the client accessing the resources. The form data are submitted to AuthorizationCodeGrantService:

AuthorizationCodeGrantService will use a 'session_authenticity_token' to validate that the session is valid and will process the user decision next.

If the decision is "allow" then it will check the status of the individual scope values. It relies on the "scopename_status" convention, if the form has offered the user a chance to selectively enable individual scopes then name/value pairs such as "updateCalendar-7_status=allow" are submitted. If none of such pairs is coming back then it means the user has approved all the default and additional (if any) scopes.

Next it will ask OAuthDataProvider to generate an authorization code grant and return it alongside with the state if any by redirecting the current user back to the redirect URI:

which leads to a browser redirecting the user:

If a user decision was set to "deny" then the error will be returned to the client.

Assuming the decision was "allow", the client has now received back the authorization code grant and is ready to exchange it for a new access token.

EndUser Name in Authorization Form

You may want to display a resource owner/end user name in the authorization form this user will be facing, you can get org.apache.cxf.rs.security.oauth2.provider.ResourceOwnerNameProvider registered with either AuthorizationCodeGrantService or ImplicitGrantService.
org.apache.cxf.rs.security.oauth2.provider.DefaultResourceOwnerNameProvider, if registered, will return an actual login name, the custom implementations may choose to return a complete user name instead, etc.

Public Clients (Devices)

CXF 2.7.7 provides an initial support for public clients (such as various mobile devices).

Client can be 'public' if it has been registered as a public client with no client secret the service itself has a "canSupportPublicClients" property enabled. The same property will also have to be enabled on AccessTokenService (described in the next section) for a public client without a secret be able to exchange a code grant for an access token.

OOB Response

If a public client has not registered a redirect URI with the Authorization service then the authorization code can be returned out-of-band (OOB), see OOBAuthorizationResponse bean. By default, it is returned directly to the end user, unless a custom OOBResponseDeliverer is registered with AuthorizationCodeGrantService which may deliver it to the client via some custom back channel.

Having OOB responses supported is useful when a public client (typically a device which can not keep the client secrets and where no redirect URI is supported) needs to get a code grant. What will happen is that a device owner will send a request to Authorization Service which may look like this:

GET
http://localhost:8080/services/social/authorize?client_id=mobileClient&response_type=code   

Assuming the 'mobileClient' has been registered as public one with no secret and the service has been set up to support such clients, the end user will get a chance to authorize this client the same way it can do confidential clients, and after this user gets back a code (delivered directly in the response HTML page by default) the user will enter the code securely into the device which will then replace it for a time-scoped access token by contacting AccessTokenService.

Secure code acquisition with redirect URI

The following extension is supported to help public clients with redirect URIs to accept the code securely staring from CXF 3.0.0-milestone2.

The public (mobile) client will include a 'code_verifier' value when requesting the authorization code and it will be saved by Authorization service, with the help of the registered AuthorizationCodeDataProvider into an instance of ServerAuthorizationCodeGrant. The client will next request a token providing the 'code' and 'code_challenge' - the latter will be compared by AuthorizationCodeGrantHandler with the original 'code_verifier'. By default, the 'code_challenge' is expected to be equal to the original 'code_verifier', but the grant handler can be registered with the custom org.apache.cxf.rs.security.oauth2.grants.code.CodeVerifierTransformer - CXF ships a DigestCodeVerifierTransformer which implements a transformation mentioned in the extension.

AccessTokenService

The role of AccessTokenService is to exchange a token grant for a new access token which will be used by the client to access the end user's resources.
Here is an example request log:

This request contains a client_id and client_secret (Authorization header), the grant_type, the grant value (code)
plus the redirect URI the authorization grant was returned to which is needed for the additional validation.
Note that the alternative client authentication methods are also possible, in this case the token service will expect a mapping between the client credentials and the client_id representing the client registration available.

After validating the request, the service will find a matching AccessTokenGrantHandler and request to create a ServerAccessToken which is a server-side representation of the access token.
The grant handlers, such as AuthorizationCodeGrantHandler may delegate the creation of the actual access token to data providers, which may create Bearer or MAC tokens with the help of utility classes shipped with CXF or depend on other 3rd party token libraries.

The data providers do not strictly required to persist the data such as access tokens, instead the token key may act as an encrypted bag capturing all the relevant information.

Now that the token has been created, it is mapped by the service to a client representation and is returned back as a JSON payload:

The client will use this access token to access the current user's resources in order to complete the original user's request, for example, the request to access a user's calendar may look like this:

Note that the access token key is passed as the Bearer scheme value. Other token types such as MAC ones, etc, can be represented differently.

Access Token Types

As mentioned above, AccessTokenService can work with whatever token is created by a given data provider. This section provides more information on how CXF may help with supporting Bearer and MAC tokens.

Bearer

The following code fragment shows how a BearerAccessToken utility class can be used to create Bearer tokens:

CustomOAuthDataProvider will also be asked by OAuthRequestFilter to validate the incoming Bearer tokens given that they typically act as database key or key alias, if no Bearer token validator is registered.

HAWK

Starting from CXF 3.0.0-milestone2 the Hawk scheme is supported instead of MAC (described in the next section). The way it is supported is identical to the way MAC scheme is supported in earlier CXF versions. The only differences are: 'Hawk' replaces 'Mac' in the Authorization header, the Hawk token returned by the server will have 'secret' and 'algorithm' parameters instead of 'mac_key' and 'mac_algorithm' parameters.

MAC

The text below applies to CXF up to 3.0.0-milestone2. Starting from 3.0.0-milestone2 MAC scheme is not supported, see above about the Hawk scheme. Support for the MAC scheme will be re-introduced once the OAuth2 working group finishes this effort.

CXF 2.6.2 supports MAC tokens as specified in the latest MAC Access Authentication draft created by Eran Hammer and others. MAC tokens offer an option for clients to demonstrate they 'hold' the token secret issued to them by AccessTokenService.
It is recommended that AccessTokenService endpoint issuing MAC tokens enforces a two-way TLS for an extra protection of the MAC token data returned to clients.

The following code fragment shows how a MacAccessToken utility class can be used to create MAC tokens:

One can expect the following response:

Note that 'access_token' is the MAC key identifier.

MacAccessTokenValidator has to be registered with OAuthRequestFilter for validating the incoming MAC tokens. This validator can get a reference to custom NonceVerifier with CXF possibly shipping a default implementation in the future.

The client can use CXF OAuthClientUtils to create Authorization MAC headers. All is needed is to provide references to ClientAccessToken representing the MAC token issued by AccessTokenService and HttpRequestProperties capturing the information about the current request URI:

This code will result in something like:

where 'ts' attribute is used to pass a timestamp value.

Encrypted tokens

Typically, the tokens are persisted in the storage. The alternative approach is to completely encrypt the token state and return the encrypted representation back to a client: the processing time to do with the encryption and decryption might increase but the server wins on avoiding the DB / storage lookups.    

CXF 3.0.0-milestone2 introduces the utility support for encrypting the state of BearerAccessToken and RefreshToken.

The tokens can be encrypted and decrypted with symmetric (secret) keys or certificates (public and private keys) and the combination of certificates and secret keys.

ModelEncryptionSupport can be used to encrypt the tokens using the custom serialization format. EncryptionUtils can be used directly if the alternative serialization format is preferred.  

Note that ServerAuthorizationGrant and Client can also be encrypted.

Using Secret Keys

The simplest strategy is to encrypt and decrypt the tokens with the symmetric/secret keys. Every new token can be encrypted with a unique secret key or all of them can be encrypted with a single secret key. The utilities provide few methods for creating secret keys with the default and advanced properties, in addition there are many examples around on how to create the keys with the specific properties.

For example, see EncryptingDataProvider:

 

 
Using Certificates

Working with the certificates to encrypt the state is similar to working with the symmetric keys. Please check the code examples in EncryptionsUtilsTest.

One needs to load a Certificate, use its public key to encrypt and the private key to decrypt. using the certificate to encrypt the whole serialized token representation might be marginally slower compared to using the symmetric keys, however given that the sequence is about 300+ characters maximum the performance can be reasonable.

Using Certificates and Secret Keys

The other approach is to generate a secret key, use this key to encrypt the token and then use the certificate to encrypt the key. The encrypted token and the actual encrypted secret key can be returned to the client as a token parameter, for example, as a 'key' parameter. This 'key' parameter will need to be returned to the OAuth2 server, via the HTTP header or the custom authorization scheme. The data providers using this mechanism will need to implement AccessTokenValidator and decrypt the encrypted key with the private certificate key, and decrypt the token with the decrypted secret key. Please check the code example in EncryptionsUtilsTest.

Encrypted JWT Tokens

JWT Token can be JWE-encrypted and the encrypted string passed to ServerAccessToken as access token id parameter.

See JSON Web Tokens wiki page for more information on how to sign and encrypt JSON Web Tokens.

Custom tokens

If needed, users can use their own custom token types, with the only restriction that the custom token type implementations have to extend org.apache.cxf.rs.security.oauth2.common.ServerAccessToken.

Simple Tokens and Audience

Starting from CXF 2.7.7 an audience parameter is supported during the client token requests.

AccessTokenValidationService

The AccessTokenValidationService is a CXF specific OAuth2 service for accepting the remote access token validation requests. Typically, OAuthRequestFilter (see on it below) may choose to impersonate itself as a third-party client and will ask AccessTokenValidationService to return the information relevant to the current access token, before setting up a security context. More on it below.

TokenRevocationService

TokenRevocationService is a simple OAuth2 service supporting the clients wishing to revoke the access or refresh tokens they own themselves, please see OAuth2 Token Revocation Draft for more information.

TokenRevocationService and AccessTokenService share the same code which enforces that the clients have been correctly authenticated.

Note, OAuthDataProvider implementations processing a revocation request should simply ignore the invalid tokens as recommended by the specification which will let TokenRevocationService return HTTP 200 which is done to minimize a possible attack surface (specifically for bad clients not to see if their requests failed or succeeded) and throw the exceptions only if the token revocation feature is not currently supported.

Supported Grants

The following subsections briefly describe how the well-known grant types can be supported on the server side. Please also check the "Client Side Support" section on how to use the related AccessTokenGrant implementations to request the access tokens.

Authorization Code

As described above, AuthorizationCodeGrantService service and AuthorizationCodeDataProvider data provider can support a redirection-based Authorization Code flow.

The code that the client receives in the end of the redirection process will need to be exchanged for a new access token with AccessTokenService. CXF-based clients can use a helper AuthorizationCodeGrant bean to request a new access token with OAuthClientUtils.

Implicit

Implicit grant is supported the same way Authorization Code grant is except that the response to the client running within a web browser is formatted differently, using URI fragments.

ImplicitGrantService service and AuthorizationCodeDataProvider data provider can support a redirection-based Implicit flow.

Note the only difference is the use of ImplicitGrantService instead of AuthorizationCodeGrantService.

Also note that when an Implicit grant client (running within a browser) replaces the code grant for a new access token and tries to access the end user's resource, Cross Origin Resource Sharing (CORS) support will most likely need to be enabled on the end user's resource server.
The simplest approach is to register a CXF CORS filter, right before OAuth2 filter (see on it below).

Starting from CXF 2.7.5 it is possible to request ImplicitGrantService to return a registered Client id to the browser-hosted client. This is recommended so that the client can verify that the token is meant to be delivered to this client.

Client Credentials

Register ClientCredentialsGrantHandler handler with AccessTokenService for this grant be supported.

CXF-based clients can use a helper ClientCredentialsGrant bean to request a new access token with OAuthClientUtils.

Resource Owner Password Credentials

Register ResourceOwnerGrantHandler handler with AccessTokenService for this grant be supported.

CXF-based clients can use a helper ResourceOwnerGrant bean to request a new access token with OAuthClientUtils.

Refresh Token

The client can issue a refresh token grant if the current access token it owns has expired or been revoked and the refresh token was issued alongside with the access token which is now invalid and get the new, 'refreshed' access token. This can allow the client to avoid seeking a new authorization approval from the end user.

Register RefreshTokenGrantHandler handler with AccessTokenService for this grant be supported. Note this grant handler is only useful for refreshing the existing access token, so one or more of the other grant handlers (Authorization Code, Implicit, etc) will also have to be registered with AccessTokenService.

CXF-based clients can use a helper RefreshTokenGrant bean to request a new access token with OAuthClientUtils.

Assertions

SAML2 Bearer and JWT assertions can be used as token grants.

Please see JAXRS OAuth2 Assertions section for more information.

Custom Grants

If you need to customize the way the well-known grant requests are handled then consider extending one of the grant handlers listed in the previous sub-sections.

Alternatively create a custom AccessTokenGrantHandler and register it with AccessTokenService. Additionally, consider providing a related AccessTokenGrant implementation for making it easy for the client code to request a new access token with this custom grant.

PreAuthorized access tokens

When working with the flows which require the end users/resource owners explicitly authorizing clients (for example, as in the case of redirection-based flows), using pre-authorized access tokens is one option to minimize the need for the end-user intervention.
OAuthDataProvider is always checked first if the pre-authorized access token for a given Client exists and if yes then it will be returned immediately, without starting the authorization process involving the end user (as required by some flows).

Consider providing a user interface which will let the end users/resource owners to pre-authorize specific clients early. Note, a CXF service for supporting the users pre-authorizing the clients or revoking the tokens for some of the clients may be introduced in the future.

Also note that using a refresh token grant may further help with minimizing the end user involvement, in cases when the current access token has expired.

Pre-registered scopes

Clients can register custom scopes they will be expected to use and then avoid specifying the scopes when requesting the code grants or access tokens.
Alternatively it makes it easier to support so called wild-card scopes. For example, a client pre-registers a scope "update" and actually uses an "update-7" scope: Redirection-based services and access token grants can be configured to do a partial scope match, in this case, validate that "update-7" starts from "update"

Writing OAuthDataProvider

Using CXF OAuth service implementations will help a lot with setting up an OAuth server. As you can see from the above sections, these services rely on a custom OAuthDataProvider implementation.

The main task of OAuthDataProvider is to persist and generate access tokens. Additionally, as noted above, AuthorizationCodeDataProvider needs to persist and remove the code grant registrations. The way it's done is really application-specific. Consider starting with a basic memory based implementation and then move on to keeping the data in some DB.

Note that OAuthDataProvider supports retrieving Client instances but it has no methods for creating or removing Clients. The reason for it is that the process of registering third-party clients is very specific to a particular OAuth2 application, so CXF does not offer a registration support service and hence OAuthDataProvider has no Client create/update methods. You will likely need to do something like this:

CustomOAuthProvider will also remove all tokens associated with a given Client in removeClient(String cliendId).

Finally OAuthDataProvider may need to convert opaque scope values such as "readCalendar" into a list of OAuthPermissions. AuthorizationCodeGrantService and OAuth2 security filters will depend on it (assuming scopes are used in the first place). In the former case AuthorizationCodeGrantService will use this list to populate OAuthAuthorizationData - the reason this bean only sees Permissions is that some of the properties OAuthPermission keeps are of no interest to OAuthAuthorizationData handlers.

OAuth Server JAX-RS endpoints

With CXF offering OAuth service implementations and a custom OAuthDataProvider provider in place, it is time to deploy the OAuth2 server.
Most likely, you'd want to deploy AccessTokenService as an independent JAX-RS endpoint, for example:

AccessTokenService listens on a relative "/token" path. Given that jaxrs:server/@adress is "/oauth" and assuming a context name is "/services", the absolute address of AccessTokenService would be something like "http://localhost:8080/services/oauth/token".

If the remote token validation is supported then have AccessTokenValidationService added too:

The absolute address of AccessTokenValidateService would be something like "http://localhost:8080/services/oauth/validate".

AuthorizationCodeGrantService is easier to put where the application endpoints are. It can be put alongside AccessTokenService, but ideally an SSO based authentication solution will be also be deployed, for the end user to avoid signing in separately several times (see more in it below). Here is an example of AuthorizationCodeGrantService being collocated with the application endpoint:

AuthorizationCodeGrantService listens on a relative "/authorize" path so in this case its absolute address will be something like "http://localhost:8080/services/myapp/authorize". This address and that of AccessTokenService will be used by third-party clients.

Third Party Client Authentication

When a client requests a token from Access Token Service, it needs to get authenticated. Providing its client_id and client secret as part of Basic Authorization scheme or posting them directly as form parameters are typical options, however other authentication schemes can be supported if required.

For example, using client certificates or assertions like SAML2 Bearer or JWT is all acceptable - the only additional requirement in this case is that a given security filter processing a specific authentication scheme maps the client credentials to an actual client_id - CXF Access Token Service will check a "client_id" property on the current message context as the last resort.

Please see JAXRS OAuth2 Assertions section for more information on how it may work.

User Session Authenticity

Redirection-based Authorization Code and Implicit flows depend on end users signing in if needed during the initial redirection, challenged with the client authorization form and returning their decision. By default, CXF will enforce the user session authenticity by keeping the session state in a servlet container's HTTPSession. If the alternative storage is preferred then you can register a new SessionAuthenticityTokenProvider (available from CXF 2.6.4) with either AuthorizationCodeGrantService or ImplicitGrantService beans.

Customizing End User Subject initialization

By default, redirection based authorization services will the the current CXF SecurityContext to initialize a subject representing the authenticated resource owner/end user. If the customization if needed: custom CXF filter can be used to create UserSubject and set it on the message or org.apache.cxf.rs.security.oauth2.provider.SubjectCreator interface implementation can be registered with either AuthorizationCodeGrantService or ImplicitGrantService.

Protecting resources with OAuth filters

OAuthRequestFilter request handler can be used to protect the resource server when processing the requests from the third-party clients. Add it as a jaxrs:provider to the endpoint which deals with the clients requesting the resources.

When checking a request like this:

the filter will do the following:

1. Retrieve a ServerAccessToken by delegating to a matching registered AccessTokenValidator. AccessTokenValidator is expected to check the validity of the incoming token parameters and possibly delegate to OAuthDataProvider to find the token representation - this is what the filter will default to if no matching AccessTokenValidator is found and the Authorization scheme is 'Bearer'.

2. Check the token has not expired

3. AccessToken may have a list of OAuthPermissions. For every permission it will:

  • If it has a uri property set then the current request URI will be checked against it
  • If it has an httpVerb property set then the current HTTP verb will be checked against it

4. Finally, it will create a CXF SecurityContext using this list of OAuthPermissions, the UserSubject representing the client or the end user who authorized the grant used to obtain this token.

This SecurityContext will not necessarily be important for some of OAuth2 applications. Most of the security checks will be done by OAuth2 filters and security filters protecting the main application path the end users themselves use. Only if you would like to share the same JAX-RS resource code and access URIs between end users and clients then it can become handy. More on it below.

Here is one example of how OAuthRequestFilter can be configured:

It will rely on an instance of OAuthDataProvider to get the information about the current access token and will validate it.
This option works OK for when it is easy to get the same OAuthDataProvider shared between this filter, as well as Authorization and AccessToken services. OAuthDataProvider can also be implemented such that it manages the information in the distributed manner so the above configuration option may scale well for more sophisticated deployments.

When one has Authorization and AccessToken service not collocated with the application endpoints, the following may work better:

OAuth2 tokens and SOAP endpoints

 

If you use HTTP Authorization header or WS-Security Binary token to pass OAuth2 tokens to SOAP endpoints then OAuthRequestInterceptor can be used to validate such tokens. It is OAuthRequestFilter running as CXF interceptor which will work OOB for tokens passed with Authorization header and it can be easily extended to support WS-Security binary tokens

How to get the user login name

When one writes a custom server application which needs to participate in OAuth2 flows, the major question which needs to be addressed is
how one can access a user login name that was used during the end-user authorizing the third-party client. This username will help to uniquely identify the resources that the 3rd party client is now attempting to access. The following code shows one way of how this can be done:

The above shows a fragment of the JAX-RS service managing the access to user resources from authorized 3rd-party clients (see the Design Considerations section for more information).

The injected MessageContext provides an access to OAuthContext which has been set by OAuth2 filters described in the previous section. OAuthContext will act as a container of the information which can be useful to the custom application code which do not need to deal with the OAuth2 internals. OAuthContextUtils provides a number of utility methods for retrieving and working with OAuthContext.

Note that starting from CXF 2.7.6 it is also possible to inject OAuthContext as JAX-RS context:

org.apache.cxf.rs.security.oauth2.provider.OAuthContextProvider will have to be registered as jaxrs:provider for it to work.

Client-side support

When developing a third party application which needs to participate in OAuth2 flows one has to write the code that will redirect users to OAuth2 AuthorizationCodeGrantService, interact with AccessTokenService in order to exchange code grants for access tokens as well as correctly build Authorization OAuth2 headers when accessing the end users' resources. JAX-RS makes it straightforward to support the redirection, while OAuthClientUtils class makes it possible to encapsulate most of the complexity away from the client application code.

For example, the following custom code can be used by the third-party application:

The reason such a simple wrapper can be introduced is to minimize the exposure to OAuth2 of the main application code to the bare minimum, this is why in this example OAuthServiceExceptions are caught, presumably logged and null values are returned which will indicate to the main code that the request failed. Obviously, OAuthClientUtils can be used directly as well.

Note that in the above example, an instance of AuthorizationCodeGrant is passed as the last parameter to OAuthClientUtils.getAccessToken() method, alongside the references to the AccessTokenService client and OAuthClientUtils.Consumer bean keeping the client id and secret.

CXF provides the utility grant beans for all the grants it supports, AuthorizationCodeGrant, ClientCredentialsGrant, ResourceOwnerGrant and RefreshTokenGrant. Please use the appropriate grant bean relevant to your application when requesting an access token or create a custom AccessTokenGrant bean implementation.

For example, consider a case where a client who already owns an authorized access token and accessing the end user resource gets HTTP 401 error back and the client also owns a refresh token. Here is one possible way to handle it:

OAuth2 without the Explicit Authorization

Client Credentials is one of OAuth2 grants that does not require the explicit authorization and is currently supported by CXF.

OAuth Without a Browser

When an end user is accessing the 3rd party application and is authorizing it later on, it's usually expected that the user is relying on a browser.
However, supporting other types of end users is easy enough. Writing the client code that processes the redirection requests from the 3rd party application and AuthorizationCodeGrantService is simple with JAX-RS and additionally CXF can be configured to do auto-redirects on the client side.

Also note that AuthorizationCodeGrantService can return XML or JSON OAuthAuthorizationData representations. That makes it easy for a client code to get OAuthAuthorizationData and offer a pop-up window or get the input from the command-line. Authorizing the third-party application might even be automated in this case - which can lead to a complete 3-leg OAuth flow implemented without a human user being involved.

Reporting error details

This section lists all the error properties that can be returned to the client application. CXF OAuth2 services will always report a required 'error' property but will omit the optional error properties by default (for example, in case of access token grant handlers throwing OAuthServiceException initialized with OAuthError which may have the optional properties set).
When reporting the optional error properties is actually needed then setting a 'writeCustomErrors' property to 'true' will help:

Design considerations

This section will talk about various design considerations one need to take into account when deploying OAuth-based solutions.

Controlling the Access to Resource Server

One of the most important issues one need to resolve is how to partition a URI space of the resource server application.

We have two different parties trying to access it, the end users which access the resource server to get to the resources which they own and 3rd party clients which have been authorized by the end users to access some of their resources.

In the former case the way the authentication is managed is completely up to the resource server application: basic authentication, two-way TLS, OpenId (more on it below), you name it.

In the latter case an OAuth filter must enforce that the 3rd party client has been registered using the provided client key and that it has a valid access token which represents the end user's approval.

Letting both parties access the resource server via the same URI(s) complicates the life for the security filters but all the parties are only aware of the single resource server URI which all of them will use.

Providing different access points to end users and clients may significantly simplify the authentication process - the possible downside is that multiple access points need to be maintained by the resource server.

Both options are discussed next.

Sharing the same access path between end users and clients

The first problem which needs to be addressed is how to distinguish end users from third-party clients and get both parties authenticated as required.
Perhaps the simplest option is to extend a CXF OAuth2 filter (JAX-RS or servlet one), check Authorization header, if it is OAuth2 then delegate to the superclass, alternatively - proceed with authenticating the end users:

The next issue is how to enforce that the end users can only access the resources they've been authorized to access.
For example, consider the following JAX-RS resource class:

Let's assume that the 3rd party client has been allowed to read the public user Calendars at "/calendar/{id}" only, how to make sure that the client won't try to:
1. update the calendar available at the same path
2. read the private Calendars available at "/calendar/{id}/private"

As noted above, OAuthPermission has an optional URIs property. Thus one way to solve the problem with the private calendar is to add, say, a uri "/calendar/{id}" or "/calendar/1" (etc) property to OAuthPermission (representing a scope like "readCalendar") and the OAuth filter will make sure no subresources beyond "/calendar/{id}" can be accessed. Note, adding a "*" at the end of a given URI property, for example, "/a*" will let the client access "/a", "/a/b", etc.

Solving the problem with preventing the update can be easily solved by adding an httpVerb property to a given OAuthPermission.

One more option is to rely on the role-based access control and have @RolesAllowed allocated such that only users in roles like "client" or "enduser" can invoke the getCalendar() method and let only those in the "enduser" role access getPrivateCalendar() and updateCalendar(). OAuthPermission can help here too as described in the section on using OAuth fiters.

Providing different access points to end users and clients

Rather than letting both the end users and 3rd party clients use the same URI such as "http://myapp.com/service/calendars/{id}", one may want to introduce two URIs, one for end users and one for third-party clients, for example, "http://myapp.com/service/calendars/{id}" - for endusers, "http://myapp.com/partners/calendars/{id}" - for the 3rd party clients and deploy 2 jaxrs endpoints, where one is protected by the security filter checking the end users, and the one - by OAuth filters.

Additionally the endpoint managing the 3rd party clients will deploy a resource which will offer a resticted URI space support. For example, if the application will only allow 3rd party clients to read calendars then this resource will only have a method supporting @GET and "/calendar/{id}".

Single Sign On

When dealing with authenticating the end users, having an SSO solution in place is very handy. This is because the end user interacts with both the third-party and its resource server web applications and is also redirected from the client application to the resource server and back again. Additionally, the end user may need to authenticate with Authorization service if it is not collocated with the application endpoints. OpenID or say a WebBrowser SSO profile can help.

CXF 2.6.1 provides an initial support for a SAML2 SSO profile. This will make it easier to minimize a number of sign ins to a single attempt and run OAuth2 Authorization servers separately from the application endpoints.