Security attribute propagation
With Security attribute propagation, WebSphere Application Server can transport security attributes (authenticated Subject contents and security context information) from one server to another in the configuration. WAS might obtain these security attributes from either an enterprise user registry, which queries static attributes, or a custom login module, which can query static or dynamic attributes. Dynamic security attributes, which are custom in nature, might include the authentication strength used for the connection, the identity of the original caller, the location of the original caller, the IP address of the original caller, and so on.
Security attribute propagation provides propagation services using Java serialization for any objects that are contained in the Subject. However, Java code must be able to serialize and deserialize these objects. The Java programming language specifies the rules for how Java code can serialize an object. Because problems can occur when dealing with different platforms and versions of software, WAS also offers a token framework that enables custom serialization functionality. The token framework has other benefits that include the ability to identify the uniqueness of the token. This uniqueness determines how the Subject gets cached and the purpose of the token. The token framework defines four marker token interfaces that enable the WAS runtime to determine how to propagate the token.
Important: Any custom tokens used in this framework are not used by WAS for authorization or authentication. The framework serves as a way to notify WAS that we want these tokens propagated in a particular way. WAS handles the propagation details, but does not handle serialization or deserialization of custom tokens. Serialization and deserialization of these custom tokens are carried out by the implementation and handled by a custom login module.
With WAS v6.0 and later, a custom Java Authorization Contract for Container (JACC) provider can be configured to enforce access control for Java EE applications. A custom JACC provider can explore the custom security attributes in the caller JAAS subject in making access control decisions.
When a request is being authenticated, a determination is made by the login modules whether this request is an initial login or a propagation login. An initial login is the process of authenticating the user information, typically a user ID and password, and then calling the APIs for the remote user registry to look up secure attributes that represent the user access rights. A propagation login is the process of validating the user information, typically a LTPA token, and then deserializing a series of tokens that constitute both custom objects and token framework objects known to WAS.
The following marker tokens are introduced in the framework:
- Authorization token
- The authorization token contains most of the authorization-related security attributes that are propagated. The default authorization token is used by the WAS authorization engine to make Java EE authorization decisions. Service providers can use custom authorization token implementations to isolate their data in a different token, perform custom serialization and de-serialization, and make custom authorization decisions using the information in their token at the appropriate time. For information on how to use and implement this token type, see Use the default propagation token to propagate security attributes and Implementing a custom propagation token for security attribute propagation.
- Single sign-on (SSO) token
- A custom SingleSignonToken token that is added to the Subject is automatically added to the response as an HTTP cookie and contains the attributes sent back to web browsers. The token
interface getName method with the getVersion method defines the cookie name. WAS defines a default SingleSignonToken token with the LtpaToken name and v2. The cookie name added is LtpaToken2. Do not add sensitive information,
confidential information, or unencrypted data to the response cookie.
It is also recommended any time that we use cookies, use the SSL protocol to protect the request. Using an SSO token, web users can authenticate once when accessing web resources across multiple WAS. A custom SSO token extends this functionality by adding custom processing to the single sign-on scenario. For more information on SSO tokens, see Implement single sign-on to minimize web user authentications. For information on how to use and implement this token type, see Use the default single sign-on token with default or custom token factory to propagate security attributes and Implementing a custom single sign-on token for security attribute propagation.
- Propagation token
- The propagation token is not associated with the authenticated
user so it is not stored in the Subject. Instead, the propagation
token is stored on the thread and follows the invocation wherever
it goes. When a request is sent outbound to another server, the propagation
tokens on that thread are sent with the request and the tokens are
run by the target server. The attributes stored on the thread
are propagated regardless of the Java EE
RunAs user switches.
The default propagation token monitors and logs all user switches and host switches. We can add additional information to the default propagation token using the WSSecurityHelper APIs. To retrieve and set custom implementations of a propagation token, we can use the WSSecurityPropagationHelper class. For information on how to use and implement this token type, see Use the default propagation token to propagate security attributes and Implementing a custom propagation token for security attribute propagation.
- Authentication token
- The authentication token flows to downstream servers and contains
the identity of the user. This token type serves the same function
as the LTPA token in previous
versions. Although this token type is typically reserved for internal WAS purposes, we can add this token to the Subject and the token is propagated using the getBytes method of the token interface.
A custom authentication token is used solely for the purpose of the service provider that adds it to the Subject. WAS does not use it for authentication purposes because a default authentication token exists used for WAS authentication. This token type is available for the service provider to identify how the custom data uses the token to perform custom authentication decisions. For information on how to use and implement this token type, see Default authentication token and Implementing a custom authentication token for security attribute propagation.
- Kerberos authentication token
- The Kerberos authentication token contains Kerberos credentials such as the Kerberos principal name, GSSCredential and Kerberos delegation credential. This token is propagated to the downstream server. Although this token type is typically reserved for internal WAS purposes, if it contains the GSSCredential we can use the getGSSCredential method to extract the GSSCredential. We can then place it in the subject and it can be used for the application. This token is created when you authenticate to WAS with either SPNEGO web or Kerberos authentication.
Horizontal propagation versus downstream propagation
In WAS, both horizontal propagation, which uses single sign-on for web requests, and downstream propagation, which uses Remote Method Invocation over the Internet Inter-ORB Protocol (RMI/IIOP) to access enterprise beans, are available.
Horizontal propagation
In horizontal propagation, security attributes are propagated among front-end servers. The serialized security attributes, which are the Subject contents and the propagation tokens, can contain both static and dynamic attributes. The single sign-on (SSO) token stores additional system-specific information needed for horizontal propagation. The information contained in the SSO token tells the receiving server where the originating server is located and how to communicate with that server. Additionally, the SSO token also contains the key to look up the serialized attributes. To enable horizontal propagation, configure the single sign-on token and the web inbound security attribute propagation features. We can configure both of these features using the administrative console.
When front-end servers are configured and in the same data replication service (DRS) replication domain, the application server automatically propagates the serialized information to all of the servers within the same domain. In figure 1, application 1 is deployed on server 1 and server 2, and both servers are members of the same DRS replication domain. If a request originates from application 1 on server 1 and then gets redirected to application 1 on server 2, the original login attributes are found on server 2 without additional remote requests.
However, if the request originates from application 1 on either server 1 or server 2, but the request is redirected to application 2 on either server 1 or server 2, the serialized information is not found in the DRS cache because the servers are not configured in the same replication domain. As a result, a remote JMX request is sent back to the originating server that hosts application 1 to obtain the serialized information so that original login information is available to the application. By getting the serialized information using a single JMX remote call back to the originating server, the following benefits are realized:
- You gain the function of retrieving login information from the original server.
- We do not need to perform any remote user registry calls because the application server can regenerate the Subject from the serialized information. Without this ability, the application server might make five to six separate remote calls.
Figure 1. Horizontal propagation
Performance implications for horizontal propagation
The performance implications of either the DRS or JMX remote call depends upon the environment. The DRS or JMX remote call is used for obtaining the original login attributes. Horizontal propagation reduces many of the remote user registry calls in cases where these calls cause the most performance problems for an application. However, the de-serialization of these objects also might cause performance degradation, but this degradation might be less than the remote user registry calls. IBM recommends that you test the environment with horizontal propagation enabled and disabled. In cases where we must use horizontal propagation for preserving original login attributes, test whether DRS or JMX provides better performance in the environment. Typically, IBM recommends that we configure DRS both for failover and performance reasons. However, because DRS propagates the information to all of the servers in the same replication domain (whether the servers are accessed or not), there might be a performance degradation if too many servers are in the same replication domain. In this case, either reduce the number of servers in the replication domain or do not configure the servers in a DRS replication domain. The later suggestion causes a JMX remote call to retrieve the attributes, when needed, which might be quicker overall.
Security cache (WSSecureMap)
In Figure 1, the security cache (WSSecureMap) is the dynamic cache used for security attribute propagation. The WSSecureMap cache stores security attributes used to recreate user credentials; it scales with the number of users who log in. The default lifetime of WSSecureMap is the same as LTPA token timeout. That is, WSSecureMap cache entries are released when the user is logged out. The usage pattern for WSSecureMap is regular.
You set the WSSecureMap cache size in the administrative console. (Security > Global Security > Custom properties > New) and define com.ibm.ws.security.WSSecureMapInitAtStartup and com.ibm.ws.security.WSSecureMapSize to govern how the cache is used.
Downstream propagation
In downstream propagation, a Subject is generated at the web front-end server, either by a propagation login or a user registry login. WAS propagates the security information downstream for enterprise bean invocations when both RMI outbound and inbound propagation are enabled.
Benefits of propagating security attributes
The security attribute propagation feature of WAS has the following benefits:
- Enables WAS to use the security attribute information for authentication and authorization
purposes. The propagation of security attributes can eliminate the need for user registry calls at each remote hop along an invocation.
Previous versions of WAS propagated
only the user name of the authenticated user, but ignored other security attribute information that needed to be regenerated downstream using remote user registry calls. To accentuate the benefits of this new
functionality, consider the following example:
In previous releases, we might use a reverse proxy server (RPSS), such as WebSEAL, to authenticate the user, gather group information, and gather other security attributes. As stated previously, WAS accepted the identity of the authenticated user, but disregarded the additional security attribute information. To create a JAAS Subject containing the needed WSCredential and WSPrincipal objects, WAS made 5 to 6 calls to the user registry. The WSCredential object contains various security information required to authorize a Java EE resource. The WSPrincipal object contains the realm name and the user that represents the principal for the Subject.
In the current release of the Application Server, information that is obtained from the reverse proxy server can be used by WAS and propagated downstream to other server resources without additional calls to the user registry. The retaining of the security attribute information enables us to protect server resources properly by making appropriate authorization and trust-based decisions User switches that occur because of Java EE RunAs configurations do not cause the application server to lose the original caller information. This information is stored in the PropagationToken located on the running thread.
- Enable third-party providers to plug in custom tokens. The token interface contains a getBytes method that enables the token implementation to define custom serialization, encryption methods, or both.
- Provides the ability to have multiple tokens of the same type within a Subject created by different providers. WAS can handle multiple tokens for the same purpose. For example, we might have multiple authorization tokens in the Subject and each token might have distinct authorization attributes generated by different providers.
- Provides the ability to have a unique ID for each token type used to formulate a more unique subject identifier than just the user name in cases where dynamic attributes might change the context of a user login. The token type has a getUniqueId() method used for returning a unique string for caching purposes. For example, we might need to propagate a location ID, which indicates the location from which the user logs into the system. This location ID can be generated during the original login using either an reverse proxy server or the WEB_INBOUND login configuration and added to the Subject prior to serialization. Other attributes might be added to the Subject as well and use a unique ID. All of the unique IDs must be considered for the uniqueness of the entire Subject. WAS has the ability to specify what is unique about the information in the Subject, which might affect how the user accesses the Subject later.
Subtopics
- Default authentication token
Do not use the default authentication token in service provider code. This default token is used by the WAS run-time code only and is authentication mechanism specific.