SAML sender-vouches use case

In my previous blog here I described the different SAML subject confirmation methods (SCM) and how they integrate with an STS. This blog describes the use case for the Sender-Vouches (SV) SCM. As this use case can be solved with an STS also, I compare the two approaches finally.


Use case for SAML Sender-Vouches



The SAML token profile states that the attesting entity is different from the subject. You have such kind of a scenario where a web service proxy/gateway (intermediary) is deployed between your service consumer and service provider. The proxy is the attesting entity and vouches for verification (or authentication) of the subject (service consumer). The proxy signs the SAML token and the soap body to protect against modification by another party.

The following picture illustrate the deployment diagram of this intermediary pattern:

(This kind of intermediary pattern is often referred to as an ESB)

The characteristic of the intermediary is that TCP connections (HTTP) are terminated and new connections are established between the intermediary and the service providers (the term proxy is widely used for settings in your OS or browser which support HTTP tunneling).


The service provider must not accept a SAML SV token unless the token and the SOAP message content being vouched for are protected by the intermediary (attesting entity) who is trusted by the service provider. The protection can be achieved by using XML signature in the WS-Security header by signing the SAML token and the SOAP body. A direct trust is established between service provider and intermediary (service consumer).


The usage of XML Signature has a significant performance impact as most XML security libraries are DOM dependent which means you need an full in-memory representation of the SOAP message. This breaks streaming (StAX parser) and increases memory usage by factors of the original message size. Another issue to consider is the potential lack of support for the STR Dereference transform algorithm  to build the signature (this was not supported two years ago in .NET). This algorithm is described in section 3.4.3 in the SAML Token Profile specification.
(Apache CXF/WSS4J 2.0 is going to support streaming based XML signature and encryption in the near future)


The OASIS WS-SecurityPolicy Examples document describes in 2.3.1.2 a more performant approach. Instead of signing the SOAP message on the XML level it uses a mutual SSL handshake where the service provider requires that the intermediary has configured a client certificate for the HTTPS communication he trusts. The side effect is that this won't be supported by all web service stacks and so you can run into interoperability issues.


SAML SV works fine if the service provider, the intermediary and service consumer are in the same security domain which means that the principal name "jdoe" is processable in all of them. 

• If the service providers are in different security domains how shall the intermediary map/federate the identity? 
• If the service provider should support more than one security domain how should the intermediary encode the security domain information into the SAML token? 
• Do the different web service stacks of your service providers support the same kind of mechanism to decode security domain information or do you have to write code for all your stacks to support that? 
• If the service provider requires role information in the SAML token how should the intermediary retrieve that and encode it within the SAML token?
• Do the web service stacks of your service providers support the same kind of mechanism to decode role information thus you can use container managed role based access control?
• Are the issued security tokens/retrieved authorization information cached?
• If the service provider requires specific claims information in the SAML token ...

As you can see, you have to consider a few things before deciding to go with the SAML Sender-Vouches subject confirmation method.

If your enviroment can fulfill the following requirements you can consider using SAML Sender-Vouches:

• there is only one security domain
• you don't have to authorize your web services and thus don't need additional claims in the token
• you're using only one Java or .NET web service stack
• soap message size is small
• performance is not critical or business logic related processing time is  comparatively higher

By now it should be apparent that  SAML Sender-Vouches is not my favourite subject confirmation method. The main reasons are the underlying pattern of a direct trust (resulting in more administration effort, comparable to the usage of self-signed certificates without a certificate authority) and lots of functionality that should instead be handled by an IDM solution which can be integrated into different kind of contexts (Web Services, Web SSO, B2B gateway, ...).
Is there another solution to address the intermediary case?



Use case "Intermediary" for WS-Trust Security Token Service

The answer to the question in my last paragraph is yes.


The WS-Trust specification defines two ways to address it:

• OnBehalfOf
• ActAs
  

The following blog post provides more information about the differences of OnBehalfOf and ActAs. In both cases, the security token received by the intermediary is sent within one of these elements in the request (RST) to the STS. The STS validates the security token and issues a new one with the information of the original principal.


Is there a standard mechanism to tell the intermediary that the service provider requires a token issued by an STS?
Yes, the service provider's WS-SecurityPolicy defines an IssuedToken assertion.

Is there a standard mechanism to tell the intermediary what kind of security token (saml1, saml2, other) the service provider requires?
Yes, the service provider's WS-SecurityPolicy defines the TokenType in the RequestSecurityTokenTemplate of the IssuedToken assertion.

Is there a standard mechanism to tell the intermediary from which security domain the service provider requires a token issued?
Yes, the service provider's WS-SecurityPolicy defines the Issuer element in the IssuedToken assertion.

Is there a standard mechanism to tell the intermediary which claims information the service provider requires in the issued token?
Yes, the service provider's WS-SecurityPolicy defines the Claims element in the IssuedToken assertion.

This logic doesn't have to be coded and managed statically within the intermediary in a non-standard compliant way. It's very flexible by just changing the policy information of the service provider.

In both intermediary solutions (with and without STS), if the intermediary has to communicate with a remote component (ID system or STS) it's recommended to cache issued security tokens for the duration of their lifetimes. This avoids needing to obain a new security token for each web service request issued by the intermediary on behalf of a particular subject.

How does the intermediary know how long should a token be cached?
How should the intermediary know if a security token has been issued for a specific service only?
This is fairly easy, the response of an STS usually returns a Lifetime and AppliesTo element. The cache can be built around the incoming security token id, the subject the token is for, the lifetime and AppliesTo information to build a flexible and token agnostic cache.

Last but not least, is there a web services stack which supports this functionality without having to write a single line of code?
Yes, it's Apache CXF. The intermediary use case is described in the above mentioned blog post. A system test and its configuration for the CXF STS is available too.


CXF is not the only web services stack supporting the intermediary case with an STS as the following article shows:

• Using WS-Trust for token transformation
  

One question remains: How can the service provider process claims information like roles, etc. which are encoded as an AttributeStatement in the SAML token? Do I have to implement and maintain this for Websphere, Weblogic, Tomcat, JBoss, etc.

This answer is no.
In one of my next posts I will explain how to reduce the complexity and increase interoperability of your web services deployment even you've got different application server stacks in house.