Core Security Patterns: Best Practices and Strategies for J2EE, Web Services, and Identity Management

SAML was originally developed to meet single sign-on security requirements. It can enable a Web user who is visiting an Internet online store via a Web browser to pass identity information to a second Web site in the same security session without a second login. SAML can also support remote authorization, whereby a service provider asks another service provider to authenticate a user and perhaps retrieves some information or attributes about a user. Another usage of SAML would be authenticating Web services (asserting a user has been authenticated prior to using Web services). The following sections discuss the usage scenarios of SAML and provide examples.

Third-Party Authentication and Authorization

To provide a good user experience, many service providers (trading partners) affiliated with online stores and Internet portals rely on the primary service provider (online stores or Internet portals) for authentication and authorization. A typical scenario involves an Internet portal managed by a telecommunications service provider, where there are many trading partners that provide content and online store services. The benefit of using SAML is that service providers do not need to re-create a new authentication infrastructure. The primary service provider may be using its existing security infrastructure or a third-party service provider (for example, Trust Authority using XKMS) to authenticate the consumer's identity. In such a way, once the primary service provider authenticates the consumer, the affiliated service providers can take the SAML authentication or authorization assertion and allow the consumer to navigate and shop at their online stores or portals seamlessly. This enhances the user experience without requiring multiple logins.

To make use of third-party authentication and authorization, the affiliated service providers must be "federated" with the primary service provider. This denotes that each affiliated service provider trusts and relies on the authentication mechanism provided by the primary service provider. The primary service provider plays the role of the SAML authority, and the affiliated service providers are the SAML-relying parties. The affiliated service providers issue a SAML request for an authentication or authorization assertion. Once they receive the SAML response from the primary service provider, they process any online order or any order request from the consumer.

Global Logout

Corporations that have implemented single sign-on integration for legacy applications and heterogeneous security infrastructures will likely also need global logout capability. However, not all single sign-on implementations are capable of global logout. Single sign-on is usually initiated from a user sign-on action, but global logout can be initiated by a system event such as previous session invalidated or idle session time-out. Many developers have added a session time-out feature (for example, a session that is idle for five minutes will invalidate the previous sign-on session) to their single sign-on infrastructure so that idle user sessions exceeding the time-out limit will trigger a global logout. The global logout capability addresses potential security risks of replay or unauthorized access to resources from invalidated sessions.

In a Web portal that aggregates access to disparate applications, once consumers perform a single sign-on to a primary service provider, they can access any remote resources to which they are entitled with the affiliated service providers. If a consumer decides to sign out of the security session with one particular service provider, the global logout functionality should disconnect from all remaining security sessions with the other service providers. Similarly, if any of the service providers invalidate the user from one of the security sessions, then the primary service provider should also perform a global logout. Typically, a service provider issues a SAML 2.0 global logout request, and the SAML authority processes the global logout request.

Security Threats and Countermeasures

Architects and developers should be aware of the security threats and potential countermeasures associated with exchanging SAML messages. The OASIS SAML Technical Committee (refer to [SAML11Security] and [SAML11Core]) identify some security threats specific to applications that use SAML and discuss countermeasures. These security threats center around the SAML assertions, SAML protocol, SAML protocol bindings, and SAML profiles.

At the level of SAML assertions, once the issuer sends a SAML assertion, the assertion is out of his or her control. The issuer has no control over how other systems persist the SAML assertion or over how parties or trading partners will share the assertions with those systems. Either those systems or their trading partners may be exposed to security risks.

Denial of Service Attack

Security attackers may replay large numbers of SAML messages to launch a Denial of Service (DoS) attack on a SAML service provider. SAML protocol by nature is a request-response protocol model. It is possible that the service provider can check whether the SAML messages are from a valid requester at the origin by using the digital signature in order to filter or discard an influx of invalid incoming requests that may cause DoS. Requiring signed requests and use of XML Signature (for example, using the element <ds:SignatureProperties> with a timestamp to filter influx of the same request from a DoS attack) would help reduce the risk associated with a DoS attack. Requiring client authentication below the SAML protocol level with client-side certificates will help track the source of attacks for diagnosis.

Message Replay and Message Modification

SAML protocol binding using SOAP or HTTP POST is susceptible to eavesdropping, message replay, message insertion, message modification, and man-in-the-middle attack. It is possible for eavesdroppers to take a copy of the real user's SAML responses and included assertions and then compose an HTTP POST message that impersonates the user at the destination site. This is also known as a stolen artifact. For a stolen artifact scenario, hackers may add countermeasures to ensure confidentiality between a site and the user's browser. For example, architects and developers can set a time limit (such as a few minutes) for the time difference between generating a SAML artifact at the source site and placing it on a URL upon receiving the <samlp:Request> message from the destination site, or they can validate the IP address in the element <saml:SubjectLocality>.

Man-in-the-Middle Attack

In addition, SAML messages are also exposed to man-in-the-middle attacks (impersonating the assertion request using an HTML form) and forged assertions (altering an assertion). For a man-in-the-middle attack, architects and developers may want to use the SAML protocol binding that supports bilateral authentication, message integrity, and confidentialityfor example, digital signature. For forged assertion, architects and developers may enforce digital signing of the SAML response that carries the SAML assertions. The destination site can ensure message integrity and authentication by verifying the signature and authenticating the issuer. Man-in-the-middle attacks can also be mitigated by securing the message transport using SSL/TLS. This can ensure point-to-point tamperproof communication.

There are also security risks related to SAML profiles. SAML profiles refer to the rules that depict how to embed SAML assertions into an XML framework and how to extract them from the framework. For the Web Browser Single Sign-on profile, it is possible that hackers can relay service requests, capture the returned SAML assertions or artifacts, and relay back a falsified SAML assertion. To mitigate this security risk, we need to use a number of countermeasures together. First, we need to use a system with strong bilateral authentication. HTTP over TLS/SSL is recommended for use with an appropriate cipher suite (strong encryption for confidentiality and for data integrity) and X.509v3 certificates (for strong authentication). These countermeasures will make man-in-the-middle attacks more difficult.

For the Enhanced Client and Proxy profile (ECP), it is possible for hackers to intercept AuthnRequest and AuthnResponse SOAP messages, which will allow subsequent Principal impersonation. The hackers may then substitute any URL of a responseConsumerServiceURL value in the message header block (PAOS message header) before forwarding the AuthnRequest on to the enhanced client. The inserted URL value may simply point back to itself so that the hackers are able to masquerade as the Principal as the legitimate service provider. To mitigate the security risk, the identity provider can specify to the enhanced client the address to which the enhanced client must send the :AuthnResponse. Thus, the responseConsumerServiceURL in the message header can only be used for error responses from the enhanced client.