Acegi Security System for Spring

The Acegi Security System for Spring is widely used by members of the Spring Community. The APIs are considered stable and only minor changes are expected. Having said that, like many other projects we need to strike a balance between backward compatibility and improvement. Effective version 0.6.1, Acegi Security uses the Apache Portable Runtime Project versioning guidelines, available from http://apr.apache.org/versioning.html.

We are now at release 0.9.0, and a lot of effort has been made to implement all non-backward compatible changes either in or before this release. Some minor improvements are currently intended to the 1.0.0 release, although they will in no way modify the project's central interfaces or classes. Users of Acegi Security System for Spring should therefore be comfortable depending on the current version of the project in their applications. Please note that we will be changing the package name prefix in the 1.0.0 release, but this should be a simple "find and replace" type operation in your code.

Most enterprise applications have four basic security requirements. First, they need to be able to authenticate a principal. Second, they need to be able to secure web requests. Third, enterprise applications need to be able to secure services layer methods. Finally, quite often an enterprise application will need to secure domain object instances. Acegi Security provides a comprehensive framework for achieving all of these four common enterprise application security requirements.

The Acegi Security System for Spring essentially comprises eight key functional parts:

A "secure object" interceptor executes most of the Acegi Security key classes and in doing so delivers the framework's major features. Given its importance, Figure 1 shows the key relationships and concrete implementations of AbstractSecurityInterceptor.

Figure 1: The key "secure object" model

Each "secure object" interceptor (hereinafter called a "security interceptor") works with a particular type of "secure object". So, what is a secure object? Secure objects refer to any type of object that can have security applied to it. A secure object must provide some form of callback, so that the security interceptor can transparently do its work as required, and callback the object when it is time for it to proceed with the requested operation. If secure objects cannot provide a native callback approach, a wrapper needs to be written so this becomes possible.

Each secure object has its own package under org.acegisecurity.intercept. Every other package in the security system is secure object independent, in that it can support any type of secure object presented.

Only developers contemplating an entirely new way of intercepting and authorizing requests would need to use secure objects directly. For example, it would be possible to build a new secure object to secure calls to a messaging system that does not use MethodInvocations. Most Spring applications will simply use the three currently supported secure object types (AOP Alliance MethodInvocation, AspectJ JoinPoint and web request FilterInterceptor) with complete transparency.

Each of the eight key parts of Acegi Security are discussed in detail throughout this document.

As shown in the base of Figure 1, the Acegi Security System for Spring currently supports three secure objects.

The first handles an AOP Alliance MethodInvocation. This is the secure object type used to protect Spring beans. Developers will generally use this secure object type to secure their business objects. To make a standard Spring-hosted bean available as a MethodInvocation, the bean is simply published through a ProxyFactoryBean or BeanNameAutoProxyCreator or DefaultAdvisorAutoProxyCreator. Most Spring developers would already be familiar with these due to their use in transactions and other areas of Spring.

The second type is an AspectJ JoinPoint. AspectJ has a particular use in securing domain object instances, as these are most often managed outside the Spring bean container. By using AspectJ, standard constructs such as new Person(); can be used and full security will be applied to them by Acegi Security. The AspectJSecurityInterceptor is still managed by Spring, which creates the aspect singleton and wires it with the appropriate authentication managers, access decision managers and so on.

The third type is a FilterInvocation. This is an object included with the Acegi Security System for Spring. It is created by an included filter and simply wraps the HTTP ServletRequest, ServletResponse and FilterChain. The FilterInvocation enables HTTP resources to be secured. Developers do not usually need to understand the mechanics of how this works, because they just add the filters to their web.xml and let the security system do its work.

Every secure object can represent an infinite number of individual requests. For example, a MethodInvocation can represent the invocation of any method with any arguments, whilst a FilterInvocation can represent any HTTP URL.

The Acegi Security System for Spring needs to record the configuration that applies to each of these possible requests. The security configuration of a request to BankManager.getBalance(int accountNumber) needs to be very different from the security configuration of a request to BankManager.approveLoan(int applicationNumber). Similarly, the security configuration of a request to http://some.bank.com/index.htm needs to be very different from the security configuration of http://some.bank.com/manage/timesheet.jsp.

To store the various security configurations associated with different requests, a configuration attribute is used. At an implementation level a configuration attribute is represented by the ConfigAttribute interface. One concrete implementation of ConfigAttribute is provided, SecurityConfig, which simply stores a configuration attribute as a String.

The collection of ConfigAttributes associated with a particular request is held in a ConfigAttributeDefinition. This concrete class is simply a holder of ConfigAttributes and does nothing special.

When a request is received by the security interceptor, it needs to determine which configuration attributes apply. In other words, it needs to find the ConfigAttributeDefinition which applies to the request. This decision is handled by the ObjectDefinitionSource interface. The main method provided by this interface is public ConfigAttributeDefinition getAttributes(Object object), with the Object being the secure object. Recall the secure object contains details of the request, so the ObjectDefinitionSource implementation will be able to extract the details it requires to lookup the relevant ConfigAttributeDefinition.

Prior to release 0.9.0, Acegi Security used a ContextHolder to store a Context between sessions. A particular subclass of Context, SecureContext defined an interface used for storage of the Authentication object. The ContextHolder was a ThreadLocal. A fuller discussion of the ThreadLocal usage with Acegi Security follows in this document. ContextHolder and SecureContext was removed from 0.9.0 after discussion with other Spring developers for the sake of consistency. See for example http://article.gmane.org/gmane.comp.java.springframework.devel/8290 and JIRA task SEC-77. This history is mentioned as the long period ContextHolder was used will likely mean that certain documentation you encounter concerning Acegi Security might still refer to ContextHolder. Generally you can just substitute "SecurityContextHolder" for "ContextHolder", and "SecurityContext" for "SecureContext", and you'll have the primary meaning of such documentation.

The Acegi Security System for Spring uses a SecurityContextHolder to store the SecurityContext. The SecurityContext contains a single getter/setter for Authentication. All Acegi Security classes query the SecurityContextHolder for obtaining the current SecurityContext (and in turn the principal). SecurityContextHolder is a ThreadLocal, meaning it is associated with the current thread of execution.

Central to Acegi Security's design is that the contents of the SecurityContextHolder (which is simply a SecurityContext implementation) can be stored between web requests. This is so that a successfully authenticated principal can be identified on subsequent requests through the Authentication stored inside the SecurityContext obtained from the SecurityContextHolder. The HttpSessionContextIntegrationFilter exists to automatically copy the contents of a well-defined HttpSession attribute into the SecurityContextHolder, then at the end of each request, copy the SecurityContextHolder contents back into the HttpSession ready for next request.

It is essential - and an extremely common error of end users - that HttpSessionContextIntegrationFilter appears before any other Acegi Security filter. Acegi Security filters expect to be able to modify the SecurityContextHolder contents as they see fit, and something else (namely HttpSessionContextIntegrationFilter) will store those between requests if necessary. This is why HttpSessionContextIntegrationFilter must be the first filter used.

You can define a custom SecurityContext implementation be used in your application by setting the context property on the HttpSessionContextIntegrationFilter bean.

From 1.0.0, Acegi Security supports localization of exception messages that end users are likely to see. Such exceptions include authentication failures and access being denied (authorization failures). Exceptions and logging that is focused on developers or system deployers (including incorrect attributes, interface contract violations, using incorrect constructors, startup time validation, debug-level logging) etc are not localized and instead are hard-coded in English within Acegi Security's code.

Shipping in the acegi-security-xx.jar inside the org.acegisecurity package is a messages.properties file. This should be referred to by your ApplicationContext, as Acegi Security classes implement Spring's MessageSourceAware interface and expect the message resolver to be dependency injected at application context startup time. Usually all you need to do is register a bean inside your application context to refer to the messages. An example is shown below:

<bean id="messageSource" class="org.springframework.context.support.ReloadableResourceBundleMessageSource">
  <property name="basename"><value>org/acegisecurity/messages</value></property>
</bean>

The messages.properties is named in accordance with standard resource bundles and represents the default language supported by Acegi Securtiy messages. This default file is in English. If you do not register a message source, Acegi Security will still work correctly and fallback to hard-coded English versions of the messages.

If you wish to customize the messages.properties file, or support other languages, you should copy the file, rename it accordingly, and register it inside the above bean definition. There are not a large number of message keys inside this file, so localization should not be considered a major initiative. If you do perform localization of this file, please consider sharing your work with the community by logging a JIRA task and attaching your appropriately-named localized version of messages.properties.

Rounding out the discussion on localization is the Spring ThreadLocal known as org.springframework.context.i18n.LocaleContextHolder. You should set the LocaleContextHolder to represent the preferred Locale of each user. Acegi Security will attempt to locate a message from the message source using the Locale obtained from this ThreadLocal. Please refer to Spring documentation for further details on using LocaleContextHolder and the helper classes that can automatically set it for you (eg AcceptHeaderLocaleResolver, CookieLocaleResolver, FixedLocaleResolver, SessionLocaleResolver etc)

As described in the High Level Design section, each secure object has its own security interceptor which is responsible for handling each request. Handling involves a number of operations:

Whilst this may seem quite involved, don't worry. Developers interact with the security process by simply implementing basic interfaces (such as AccessDecisionManager), which are fully discussed below.

The AbstractSecurityInterceptor handles the majority of the flow listed above. As shown in Figure 1, each secure object has its own security interceptor which subclasses AbstractSecurityInterceptor. Each of these secure object-specific security interceptors are discussed below.

To secure MethodInvocations, developers simply add a properly configured MethodSecurityInterceptor into the application context. Next the beans requiring security are chained into the interceptor. This chaining is accomplished using Spring’s ProxyFactoryBean or BeanNameAutoProxyCreator, as commonly used by many other parts of Spring (refer to the sample application for examples). Alternatively, Acegi Security provides a MethodDefinitionSourceAdvisor which may be used with Spring's DefaultAdvisorAutoProxyCreator to automatically chain the security interceptor in front of any beans defined against the MethodSecurityInterceptor. The MethodSecurityInterceptor itself is configured as follows:

<bean id="bankManagerSecurity" class="org.acegisecurity.intercept.method.aopalliance.MethodSecurityInterceptor">
  <property name="validateConfigAttributes"><value>true</value></property>
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
  <property name="accessDecisionManager"><ref bean="accessDecisionManager"/></property>
  <property name="runAsManager"><ref bean="runAsManager"/></property>
  <property name="afterInvocationManager"><ref bean="afterInvocationManager"/></property>
  <property name="objectDefinitionSource">
    <value>
      org.acegisecurity.context.BankManager.delete*=ROLE_SUPERVISOR,RUN_AS_SERVER
      org.acegisecurity.context.BankManager.getBalance=ROLE_TELLER,ROLE_SUPERVISOR,BANKSECURITY_CUSTOMER,RUN_AS_SERVER
    </value>
  </property>
</bean>

As shown above, the MethodSecurityInterceptor is configured with a reference to an AuthenticationManager, AccessDecisionManager and RunAsManager, which are each discussed in separate sections below. In this case we've also defined an AfterInvocationManager, although this is entirely optional. The MethodSecurityInterceptor is also configured with configuration attributes that apply to different method signatures. A full discussion of configuration attributes is provided in the High Level Design section of this document.

The MethodSecurityInterceptor can be configured with configuration attributes in three ways. The first is via a property editor and the application context, which is shown above. The second is via defining the configuration attributes in your source code using Jakarta Commons Attributes or Java 5 Annotations. The third is via writing your own ObjectDefinitionSource, although this is beyond the scope of this document. Irrespective of the approach used, the ObjectDefinitionSource is responsible for returning a ConfigAttributeDefinition object that contains all of the configuration attributes associated with a single secure method.

It should be noted that the MethodSecurityInterceptor.setObjectDefinitionSource() method actually expects an instance of MethodDefinitionSource. This is a marker interface which subclasses ObjectDefinitionSource. It simply denotes the ObjectDefinitionSource understands MethodInvocations. In the interests of simplicity we'll continue to refer to the MethodDefinitionSource as an ObjectDefinitionSource, as the distinction is of little relevance to most users of the MethodSecurityInterceptor.

If using the application context property editor approach (as shown above), commas are used to delimit the different configuration attributes that apply to a given method pattern. Each configuration attribute is assigned into its own SecurityConfig object. The SecurityConfig object is discussed in the High Level Design section.

If you are using the Jakarta Commons Attributes approach, your bean context will be configured differently:

<bean id="attributes" class="org.springframework.metadata.commons.CommonsAttributes"/>
<bean id="objectDefinitionSource" class="org.acegisecurity.intercept.method.MethodDefinitionAttributes">
  <property name="attributes"><ref local="attributes"/></property>
</bean>

<bean id="bankManagerSecurity" class="org.acegisecurity.intercept.method.aopalliance.MethodSecurityInterceptor">
  <property name="validateConfigAttributes"><value>false</value></property>
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
  <property name="accessDecisionManager"><ref bean="accessDecisionManager"/></property>
  <property name="runAsManager"><ref bean="runAsManager"/></property>
  <property name="objectDefinitionSource"><ref bean="objectDefinitionSource"/></property>
</bean>

In addition, your source code will contain Jakarta Commons Attributes tags that refer to a concrete implementation of ConfigAttribute. The following example uses the SecurityConfig implementation to represent the configuration attributes, and results in the same security configuration as provided by the property editor approach above:

public interface BankManager {

    /**
     * @@SecurityConfig("ROLE_SUPERVISOR")
     * @@SecurityConfig("RUN_AS_SERVER")
     */
    public void deleteSomething(int id);

    /**
     * @@SecurityConfig("ROLE_SUPERVISOR")
     * @@SecurityConfig("RUN_AS_SERVER")
     */
    public void deleteAnother(int id);

    /**
     * @@SecurityConfig("ROLE_TELLER")
     * @@SecurityConfig("ROLE_SUPERVISOR")
     * @@SecurityConfig("BANKSECURITY_CUSTOMER")
     * @@SecurityConfig("RUN_AS_SERVER")
     */
    public float getBalance(int id);
}

If you are using the Spring Security Java 5 Annotations approach, your bean context will be configured as follows:

<bean id="attributes" class="org.acegisecurity.annotation.SecurityAnnotationAttributes"/>
<bean id="objectDefinitionSource" class="org.acegisecurity.intercept.method.MethodDefinitionAttributes">
  <property name="attributes"><ref local="attributes"/></property>
</bean>

<bean id="bankManagerSecurity" class="org.acegisecurity.intercept.method.aopalliance.MethodSecurityInterceptor">
  <property name="validateConfigAttributes"><value>false</value></property>
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
  <property name="accessDecisionManager"><ref bean="accessDecisionManager"/></property>
  <property name="runAsManager"><ref bean="runAsManager"/></property>
  <property name="objectDefinitionSource"><ref bean="objectDefinitionSource"/></property>
</bean>

In addition, your source code will contain the Acegi Java 5 Security Annotations that represent the ConfigAttribute. The following example uses the @Secured annotations to represent the configuration attributes, and results in the same security configuration as provided by the property editor approach:

import org.acegisecurity.annotation.Secured;

public interface BankManager {

    /**
     * Delete something
     */
    @Secured({"ROLE_SUPERVISOR","RUN_AS_SERVER" })
    public void deleteSomething(int id);

    /**
     * Delete another
     */
    @Secured({"ROLE_SUPERVISOR","RUN_AS_SERVER" })
    public void deleteAnother(int id);

    /**
     * Get balance
     */
    @Secured({"ROLE_TELLER","ROLE_SUPERVISOR","BANKSECURITY_CUSTOMER","RUN_AS_SERVER" })
    public float getBalance(int id);
}

You might have noticed the validateConfigAttributes property in the above MethodSecurityInterceptor examples. When set to true (the default), at startup time the MethodSecurityInterceptor will evaluate if the provided configuration attributes are valid. It does this by checking each configuration attribute can be processed by either the AccessDecisionManager or the RunAsManager. If neither of these can process a given configuration attribute, an exception is thrown. If using the Jakarta Commons Attributes method of configuration, you should set validateConfigAttributes to false.

The AspectJ security interceptor is very similar to the AOP Alliance security interceptor discussed in the previous section. Indeed we will only discuss the differences in this section.

The AspectJ interceptor is named AspectJSecurityInterceptor. Unlike the AOP Alliance security interceptor, which relies on the Spring application context to weave in the security interceptor via proxying, the AspectJSecurityInterceptor is weaved in via the AspectJ compiler. It would not be uncommon to use both types of security interceptors in the same application, with AspectJSecurityInterceptor being used for domain object instance security and the AOP Alliance MethodSecurityInterceptor being used for services layer security.

Let's first consider how the AspectJSecurityInterceptor is configured in the Spring application context:

<bean id="bankManagerSecurity" class="org.acegisecurity.intercept.method.aspectj.AspectJSecurityInterceptor">
  <property name="validateConfigAttributes"><value>true</value></property>
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
  <property name="accessDecisionManager"><ref bean="accessDecisionManager"/></property>
  <property name="runAsManager"><ref bean="runAsManager"/></property>
  <property name="afterInvocationManager"><ref bean="afterInvocationManager"/></property>
  <property name="objectDefinitionSource">
    <value>
      org.acegisecurity.context.BankManager.delete*=ROLE_SUPERVISOR,RUN_AS_SERVER
      org.acegisecurity.context.BankManager.getBalance=ROLE_TELLER,ROLE_SUPERVISOR,BANKSECURITY_CUSTOMER,RUN_AS_SERVER
    </value>
  </property>
</bean>

As you can see, aside from the class name, the AspectJSecurityInterceptor is exactly the same as the AOP Alliance security interceptor. Indeed the two interceptors can share the same objectDefinitionSource, as the ObjectDefinitionSource works with java.lang.reflect.Methods rather than an AOP library-specific class. Of course, your access decisions have access to the relevant AOP library-specific invocation (ie MethodInvocation or JoinPoint) and as such can consider a range of addition criteria when making access decisions (such as method arguments).

Next you'll need to define an AspectJ aspect. For example:

package org.acegisecurity.samples.aspectj;

import org.acegisecurity.intercept.method.aspectj.AspectJSecurityInterceptor;
import org.acegisecurity.intercept.method.aspectj.AspectJCallback;
import org.springframework.beans.factory.InitializingBean;

public aspect DomainObjectInstanceSecurityAspect implements InitializingBean {

  private AspectJSecurityInterceptor securityInterceptor;

  pointcut domainObjectInstanceExecution(): target(PersistableEntity) 
             && execution(public * *(..)) && !within(DomainObjectInstanceSecurityAspect);

  Object around(): domainObjectInstanceExecution() {
    if (this.securityInterceptor != null) {
      AspectJCallback callback = new AspectJCallback() {
        public Object proceedWithObject() {
        return proceed();
      }
    };
    return this.securityInterceptor.invoke(thisJoinPoint, callback);
    } else {
      return proceed();
    }
  }

  public AspectJSecurityInterceptor getSecurityInterceptor() {
    return securityInterceptor;
  }

  public void setSecurityInterceptor(AspectJSecurityInterceptor securityInterceptor) {
    this.securityInterceptor = securityInterceptor;
  }

  public void afterPropertiesSet() throws Exception {
    if (this.securityInterceptor == null)
      throw new IllegalArgumentException("securityInterceptor required");
  }
}

In the above example, the security interceptor will be applied to every instance of PersistableEntity, which is an abstract class not shown (you can use any other class or pointcut expression you like). For those curious, AspectJCallback is needed because the proceed(); statement has special meaning only within an around() body. The AspectJSecurityInterceptor calls this anonymous AspectJCallback class when it wants the target object to continue.

You will need to configure Spring to load the aspect and wire it with the AspectJSecurityInterceptor. A bean declaration which achieves this is shown below:

<bean id="domainObjectInstanceSecurityAspect" 
    class="org.acegisecurity.samples.aspectj.DomainObjectInstanceSecurityAspect"
    factory-method="aspectOf">
  <property name="securityInterceptor"><ref bean="aspectJSecurityInterceptor"/></property>
</bean>

That's it! Now you can create your beans from anywhere within your application, using whatever means you think fit (eg new Person();) and they will have the security interceptor applied.

To secure FilterInvocations, developers need to add a filter to their web.xml that delegates to the SecurityEnforcementFilter. A typical configuration example is provided below:

<filter>
  <filter-name>Acegi HTTP Request Security Filter</filter-name>
  <filter-class>org.acegisecurity.util.FilterToBeanProxy</filter-class>
  <init-param>
    <param-name>targetClass</param-name>
    <param-value>org.acegisecurity.intercept.web.SecurityEnforcementFilter</param-value>
  </init-param>
</filter>

<filter-mapping>
  <filter-name>Acegi HTTP Request Security Filter</filter-name>
  <url-pattern>/*</url-pattern>
</filter-mapping>

Notice that the filter is actually a FilterToBeanProxy. Most of the filters used by the Acegi Security System for Spring use this class. Refer to the Filters section to learn more about this bean.

In the application context you will need to configure three beans:

<bean id="securityEnforcementFilter" class="org.acegisecurity.intercept.web.SecurityEnforcementFilter">
  <property name="filterSecurityInterceptor"><ref bean="filterInvocationInterceptor"/></property>
  <property name="authenticationEntryPoint"><ref bean="authenticationEntryPoint"/></property>
</bean>

<bean id="authenticationEntryPoint" class="org.acegisecurity.ui.webapp.AuthenticationProcessingFilterEntryPoint">
  <property name="loginFormUrl"><value>/acegilogin.jsp</value></property>
  <property name="forceHttps"><value>false</value></property>
</bean>

<bean id="filterInvocationInterceptor" class="org.acegisecurity.intercept.web.FilterSecurityInterceptor">
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
  <property name="accessDecisionManager"><ref bean="accessDecisionManager"/></property>
  <property name="runAsManager"><ref bean="runAsManager"/></property>
  <property name="objectDefinitionSource">
    <value>
      CONVERT_URL_TO_LOWERCASE_BEFORE_COMPARISON
      \A/secure/super/.*\Z=ROLE_WE_DONT_HAVE
      \A/secure/.*\Z=ROLE_SUPERVISOR,ROLE_TELLER
    </value>
  </property>
</bean>

The AuthenticationEntryPoint will be called if the user requests a secure HTTP resource but they are not authenticated. The class handles presenting the appropriate response to the user so that authentication can begin. Three concrete implementations are provided with the Acegi Security System for Spring: AuthenticationProcessingFilterEntryPoint for commencing a form-based authentication, BasicProcessingFilterEntryPoint for commencing a HTTP Basic authentication process, and CasProcessingFilterEntryPoint for commencing a Yale Central Authentication Service (CAS) login. The AuthenticationProcessingFilterEntryPoint and CasProcessingFilterEntryPoint have optional properties related to forcing the use of HTTPS, so please refer to the JavaDocs if you require this.

The PortMapper provides information on which HTTPS ports correspond to which HTTP ports. This is used by the AuthenticationProcessingFilterEntryPoint and several other beans. The default implementation, PortMapperImpl, knows the common HTTP ports 80 and 8080 map to HTTPS ports 443 and 8443 respectively. You can customise this mapping if desired.

The SecurityEnforcementFilter primarily provides session management support and initiates authentication when required. It delegates actual FilterInvocation security decisions to the configured FilterSecurityInterceptor.

Like any other security interceptor, the FilterSecurityInterceptor requires a reference to an AuthenticationManager, AccessDecisionManager and RunAsManager, which are each discussed in separate sections below. The FilterSecurityInterceptor is also configured with configuration attributes that apply to different HTTP URL requests. A full discussion of configuration attributes is provided in the High Level Design section of this document.

The FilterSecurityInterceptor can be configured with configuration attributes in two ways. The first is via a property editor and the application context, which is shown above. The second is via writing your own ObjectDefinitionSource, although this is beyond the scope of this document. Irrespective of the approach used, the ObjectDefinitionSource is responsible for returning a ConfigAttributeDefinition object that contains all of the configuration attributes associated with a single secure HTTP URL.

It should be noted that the FilterSecurityInterceptor.setObjectDefinitionSource() method actually expects an instance of FilterInvocationDefinitionSource. This is a marker interface which subclasses ObjectDefinitionSource. It simply denotes the ObjectDefinitionSource understands FilterInvocations. In the interests of simplicity we'll continue to refer to the FilterInvocationDefinitionSource as an ObjectDefinitionSource, as the distinction is of little relevance to most users of the FilterSecurityInterceptor.

If using the application context property editor approach (as shown above), commas are used to delimit the different configuration attributes that apply to each HTTP URL. Each configuration attribute is assigned into its own SecurityConfig object. The SecurityConfig object is discussed in the High Level Design section. The ObjectDefinitionSource created by the property editor, FilterInvocationDefinitionSource, matches configuration attributes against FilterInvocations based on expression evaluation of the request URL. Two standard expression syntaxes are supported. The default is to treat all expressions as regular expressions. Alternatively, the presence of a PATTERN_TYPE_APACHE_ANT directive will cause all expressions to be treated as Apache Ant paths. It is not possible to mix expression syntaxes within the same definition. For example, the earlier configuration could be generated using Apache Ant paths as follows:

<bean id="filterInvocationInterceptor" class="org.acegisecurity.intercept.web.FilterSecurityInterceptor">
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
  <property name="accessDecisionManager"><ref bean="accessDecisionManager"/></property>
  <property name="runAsManager"><ref bean="runAsManager"/></property>
  <property name="objectDefinitionSource">
    <value>
      CONVERT_URL_TO_LOWERCASE_BEFORE_COMPARISON
      PATTERN_TYPE_APACHE_ANT
      /secure/super/**=ROLE_WE_DONT_HAVE
      /secure/**=ROLE_SUPERVISOR,ROLE_TELLER
    </value>
  </property>
</bean>

Irrespective of the type of expression syntax used, expressions are always evaluated in the order they are defined. Thus it is important that more specific expressions are defined higher in the list than less specific expressions. This is reflected in our example above, where the more specific /secure/super/ pattern appears higher than the less specific /secure/ pattern. If they were reversed, the /secure/ pattern would always match and the /secure/super/ pattern would never be evaluated.

The special keyword CONVERT_URL_TO_LOWERCASE_BEFORE_COMPARISON causes the FilterInvocationDefinitionSource to automatically convert a request URL to lowercase before comparison against the expressions. Whilst by default the case of the request URL is not converted, it is generally recommended to use CONVERT_URL_TO_LOWERCASE_BEFORE_COMPARISON and write each expression assuming lowercase.

As with other security interceptors, the validateConfigAttributes property is observed. When set to true (the default), at startup time the FilterSecurityInterceptor will evaluate if the provided configuration attributes are valid. It does this by checking each configuration attribute can be processed by either the AccessDecisionManager or the RunAsManager. If neither of these can process a given configuration attribute, an exception is thrown.

Authentication requires a way for client code to present its security identification to the Acegi Security System for Spring. This is the role of the Authentication interface. The Authentication interface holds three important objects: the principal (the identity of the caller), the credentials (the proof of the identity of the caller, such as a password), and the authorities that have been granted to the principal. The principal and its credentials are populated by the client code, whilst the granted authorities are populated by the AuthenticationManager.

Figure 3: Key Authentication Architecture

As shown in Figure 3, the Acegi Security System for Spring includes several concrete Authentication implementations:

The authorities granted to a principal are represented by the GrantedAuthority interface. The GrantedAuthority interface is discussed at length in the Authorization section.

As discussed in the Security Interception section, the AbstractSecurityInterceptor extracts the Authentication object from the SecurityContext in the SecurityContextHolder. This is then passed to an AuthenticationManager. The AuthenticationManager interface is very simple:

public Authentication authenticate(Authentication authentication) throws AuthenticationException;

Implementations of AuthenticationManager are required to throw an AuthenticationException should authentication fail, or return a fully populated Authentication object. In particular, the returned Authentication object should contain an array of GrantedAuthority objects. The SecurityInterceptor places the populated Authentication object back in the SecurityContext in the SecurityContextHolder, overwriting the original Authentication object.

The AuthenticationException has a number of subclasses. The most important are BadCredentialsException (an incorrect principal or credentials), DisabledException and LockedException. The latter two exceptions indicate the principal was found, but the credentials were not checked and authentication is denied. An AuthenticationServiceException is also provided, which indicates the authentication system could not process the request (eg a database was unavailable). AuthenticationException also has a CredentialsExpiredException and AccoungtExpiredException subclass, although these are less commonly used.

Whilst the basic Authentication and AuthenticationManager interfaces enable users to develop their own authentication systems, users should consider using the provider-based authentication packages provided by the Acegi Security System for Spring. The key class, ProviderManager, is configured via the bean context with a list of AuthenticationProviders:

<bean id="authenticationManager" class="org.acegisecurity.providers.ProviderManager">
  <property name="providers">
    <list>
      <ref bean="daoAuthenticationProvider"/>
      <ref bean="someOtherAuthenticationProvider"/>
    </list>
  </property>
</bean>

ProviderManager calls a series of registered AuthenticationProvider implementations, until one is found that indicates it is able to authenticate a given Authentication class. When the first compatible AuthenticationProvider is located, it is passed the authentication request. The AuthenticationProvider will then either throw an AuthenticationException or return a fully populated Authentication object.

Note the ProviderManager may throw a ProviderNotFoundException (a subclass of AuthenticationException) if it none of the registered AuthenticationProviders can validate the Authentication object.

The ProviderManager also has several other important functions. It integrates with concurrent session handling supoprt, and it also converts any exceptions thrown by an AuthenticationProvider and publishes a suitable event. The events that are published are located in the org.acegisecurity.event.authentication package and advanced users can map different exceptions to different events by configuring the ProviderManager.exceptionMappings property (generally this is not required and the default event propagation is appropriate - especially as events will simply be ignored if you don't have an ApplicationListener configured in the ApplicationContext).

Several AuthenticationProvider implementations are provided with the Acegi Security System for Spring:

Acegi Security is able to stop the same principal authenticating to the same web application multiple times concurrently. Put differently, you can stop user "Batman" from logging into a web application twice at the same time.

To use concurrent session support, you'll need to add the following to web.xml:

<listener>
  <listener-class>org.acegisecurity.ui.session.HttpSessionEventPublisher</listener-class>
</listener>

In addition, you will need to add the org.acegisecurity.concurrent.ConcurrentSessionFilter to your FilterChainProxy. The ConcurrentSessionFilter requires only one property, sessionRegistry, which generally points to an instance of SessionRegistryImpl.

The web.xml HttpSessionEventPublisher causes an ApplicationEvent to be published to the Spring ApplicationContext every time a HttpSession commences or terminates. This is critical, as it allows the SessionRegistryImpl to be notified when a session ends.

You will also need to wire up the ConcurrentSessionControllerImpl and refer to it from your ProviderManager bean:

<bean id="authenticationManager" class="org.acegisecurity.providers.ProviderManager">
  <property name="providers">
    <!-- your providers go here -->
  </property>
  <property name="sessionController"><ref bean="concurrentSessionController"/></property>
</bean>

<bean id="concurrentSessionController" class="org.acegisecurity.concurrent.ConcurrentSessionControllerImpl">
  <property name="maximumSessions"><value>1</value></property>
  <property name="sessionRegistry"><ref local="sessionRegistry"/></property>
</bean>

<bean id="sessionRegistry" class="org.acegisecurity.concurrent.SessionRegistryImpl"/>

The Acegi Security System for Spring includes a production-quality AuthenticationProvider implementation called DaoAuthenticationProvider. This authentication provider is able to authenticate a UsernamePasswordAuthenticationToken by obtaining authentication details from a data access object configured at bean creation time:

<bean id="daoAuthenticationProvider" class="org.acegisecurity.providers.dao.DaoAuthenticationProvider">
  <property name="userDetailsService"><ref bean="inMemoryDaoImpl"/></property>
  <property name="saltSource"><ref bean="saltSource"/></property>
  <property name="passwordEncoder"><ref bean="passwordEncoder"/></property>
</bean>

The PasswordEncoder and SaltSource are optional. A PasswordEncoder provides encoding and decoding of passwords obtained from the authentication repository. A SaltSource enables the passwords to be populated with a "salt", which enhances the security of the passwords in the authentication repository. PasswordEncoder implementations are provided with the Acegi Security System for Spring covering MD5, SHA and cleartext encodings. Two SaltSource implementations are also provided: SystemWideSaltSource which encodes all passwords with the same salt, and ReflectionSaltSource, which inspects a given property of the returned UserDetails object to obtain the salt. Please refer to the JavaDocs for further details on these optional features.

In addition to the properties above, the DaoAuthenticationProvider supports optional caching of UserDetails objects. The UserCache interface enables the DaoAuthenticationProvider to place a UserDetails object into the cache, and retrieve it from the cache upon subsequent authentication attempts for the same username. By default the DaoAuthenticationProvider uses the NullUserCache, which performs no caching. A usable caching implementation is also provided, EhCacheBasedUserCache, which is configured as follows:

<bean id="daoAuthenticationProvider" class="org.acegisecurity.providers.dao.DaoAuthenticationProvider">
  <property name="userDetailsService"><ref bean="userDetailsService"/></property>
  <property name="userCache"><ref bean="userCache"/></property>
</bean>

<bean id="cacheManager" class="org.springframework.cache.ehcache.EhCacheManagerFactoryBean">
  <property name="configLocation">
    <value>classpath:/ehcache-failsafe.xml</value>
  </property>
</bean>
    
<bean id="userCacheBackend" class="org.springframework.cache.ehcache.EhCacheFactoryBean">
  <property name="cacheManager">
    <ref local="cacheManager"/>
  </property>
  <property name="cacheName">
    <value>userCache</value>
  </property>
</bean>
   
<bean id="userCache" class="org.acegisecurity.providers.dao.cache.EhCacheBasedUserCache">
  <property name="cache"><ref local="userCacheBackend"/></property>
</bean>

All Acegi Security EH-CACHE implementations (including EhCacheBasedUserCache) require an EH-CACHE Cache object. The Cache object can be obtained from wherever you like, although we recommend you use Spring's factory classes as shown in the above configuration. If using Spring's factory classes, please refer to the Spring documentation for further details on how to optimise the cache storage location, memory usage, eviction policies, timeouts etc.

For a class to be able to provide the DaoAuthenticationProvider with access to an authentication repository, it must implement the UserDetailsService interface:

public UserDetails loadUserByUsername(String username) throws UsernameNotFoundException, DataAccessException;

The UserDetails is an interface that provides getters that guarantee non-null provision of basic authentication information such as the username, password, granted authorities and whether the user is enabled or disabled. A concrete implementation, User, is also provided. Acegi Security users will need to decide when writing their UserDetailsService what type of UserDetails to return. In most cases User will be used directly or subclassed, although special circumstances (such as object relational mappers) may require users to write their own UserDetails implementation from scratch. UserDetails is often used to store additional principal-related properties (such as their telephone number and email address), so they can be easily used by web views.

Given UserDetailsService is so simple to implement, it should be easy for users to retrieve authentication information using a persistence strategy of their choice.

A design decision was made not to support account locking in the DaoAuthenticationProvider, as doing so would have increased the complexity of the UserDetailsService interface. For instance, a method would be required to increase the count of unsuccessful authentication attempts. Such functionality could be easily provided by leveraging the application event publishing features discussed below.

DaoAuthenticationProvider returns an Authentication object which in turn has its principal property set. The principal will be either a String (which is essentially the username) or a UserDetails object (which was looked up from the UserDetailsService). By default the UserDetails is returned, as this enables applications to add extra properties potentially of use in applications, such as the user's full name, email address etc. If using container adapters, or if your applications were written to operate with Strings (as was the case for releases prior to Acegi Security 0.6), you should set the DaoAuthenticationProvider.forcePrincipalAsString property to true in your application context.

Whilst it is easy to use the DaoAuthenticationProvider and create a custom UserDetailsService implementation that extracts information from a persistence engine of choice, many applications do not require such complexity. One alternative is to configure an authentication repository in the application context itself using the InMemoryDaoImpl:

<bean id="inMemoryDaoImpl" class="org.acegisecurity.userdetails.memory.InMemoryDaoImpl">
  <property name="userMap">
    <value>
      marissa=koala,ROLE_TELLER,ROLE_SUPERVISOR
      dianne=emu,ROLE_TELLER
      scott=wombat,ROLE_TELLER
      peter=opal,disabled,ROLE_TELLER
    </value>
  </property>
</bean>

The userMap property contains each of the usernames, passwords, a list of granted authorities and an optional enabled/disabled keyword. Commas delimit each token. The username must appear to the left of the equals sign, and the password must be the first token to the right of the equals sign. The enabled and disabled keywords (case insensitive) may appear in the second or any subsequent token. Any remaining tokens are treated as granted authorities, which are created as GrantedAuthorityImpl objects (refer to the Authorization section for further discussion on granted authorities). Note that if a user has no password and/or no granted authorities, the user will not be created in the in-memory authentication repository.

InMemoryDaoImpl also offers a setUserProperties(Properties) method, which allows you to externalise the java.util.Properties in another Spring configured bean or an external properties file. This might prove useful for simple applications that have a larger number of users, or deployment-time configuration changes, but do not wish to use a full database for authentication details.

The Acegi Security System for Spring also includes an authentication provider that can obtain authentication information from a JDBC data source. The typical configuration for the JdbcDaoImpl is shown below:

<bean id="dataSource" class="org.springframework.jdbc.datasource.DriverManagerDataSource">
  <property name="driverClassName"><value>org.hsqldb.jdbcDriver</value></property>
  <property name="url"><value>jdbc:hsqldb:hsql://localhost:9001</value></property>
  <property name="username"><value>sa</value></property>
  <property name="password"><value></value></property>
</bean>

<bean id="jdbcDaoImpl" class="org.acegisecurity.userdetails.jdbc.JdbcDaoImpl">
  <property name="dataSource"><ref bean="dataSource"/></property>
</bean>

You can use different relational database management systems by modifying the DriverManagerDataSource shown above. Irrespective of the database used, a standard schema must be used as indicated in dbinit.txt.

If you default schema is unsuitable for your needs, JdbcDaoImpl provides two properties that allow customisation of the SQL statements. You may also subclass the JdbcDaoImpl if further customisation is necessary. Please refer to the JavaDocs for details.

Acegi Security provides a package able to delegate authentication requests to the Java Authentication and Authorization Service (JAAS). This package is discussed in detail below.

Central to JAAS operation are login configuration files. To learn more about JAAS login configuration files, consult the JAAS reference documentation available from Sun Microsystems. We expect you to have a basic understanding of JAAS and its login configuration file syntax in order to understand this section.

JAASTest {
  sample.SampleLoginModule required;
};

<bean id="jaasAuthenticationProvider" class="org.acegisecurity.providers.jaas.JaasAuthenticationProvider">
  <property name="loginConfig">
    <value>/WEB-INF/login.conf</value>
  </property>
  <property name="loginContextName">
    <value>JAASTest</value>
  </property>
  <property name="callbackHandlers">
    <list>
      <bean class="org.acegisecurity.providers.jaas.JaasNameCallbackHandler"/>
      <bean class="org.acegisecurity.providers.jaas.JaasPasswordCallbackHandler"/>
    </list>
  </property>
  <property name="authorityGranters">
    <list>
      <bean class="org.acegisecurity.providers.jaas.TestAuthorityGranter"/>
    </list>
  </property>
</bean>

Acegi Security provides a web filter (org.acegisecurity.ui.webapp.SiteminderAuthenticationProcessingFilter) that can be used to process requests that have been pre-authenticated by Computer Associates' Siteminder. This filter assumes that you're using Siteminder for authentication, and your application (or backing datasource) is used for authorization. The use of Siteminder for authorization is not yet directly supported by Acegi.

Recall that a Siteminder agent is set up on your web server to intercept a user's first call to your application. This agent redirects the initial request to a login page, and only after successful authentication does your application receive the request. Authenticated requests contain one or more HTTP headers populated by the Siteminder agent. Below we'll assume that the request header key containing the user's identity is "SM_USER", but of course your header values may be different based on Siteminder policy server configuration. Please refer to your company's "single sign-on" group for header details.

   <!-- ======================== SITEMINDER AUTHENTICATION PROCESSING FILTER ======================= --> 
   <bean id="authenticationProcessingFilter" class="org.acegisecurity.ui.webapp.SiteminderAuthenticationProcessingFilter">
      <property name="authenticationManager"><ref bean="authenticationManager"/></property>
      <property name="authenticationFailureUrl"><value>/login.jsp?login_error=1</value></property>
      <property name="defaultTargetUrl"><value>/security.do?method=getMainMenu</value></property>
      <property name="filterProcessesUrl"><value>/j_acegi_security_check</value></property>
      <property name="siteminderUsernameHeaderKey"><value>SM_USER</value></property>
      <property name="siteminderPasswordHeaderKey"><value>SM_USER</value></property>
   </bean>

    <!-- ======================== AUTHENTICATION ======================= -->
    <!--
      - The top-level Authentication Manager is responsible for all application AUTHENTICATION
      - operations.  Note that it must reference one or more provider(s) defined below.
      -->
    <bean id="authenticationManager" class="org.acegisecurity.providers.ProviderManager">
        <property name="providers">
            <list>
                <ref local="daoAuthenticationProvider"/>
            </list>
        </property>
    </bean>

    <!-- ======================== FILTER CHAIN ======================= -->
    <!--
      - The web.xml file has a single filter reference to this top-level bean, which
      - invokes the chain of sub-filters specified below.
      --> 
    <bean id="filterChainProxy" class="org.acegisecurity.util.FilterChainProxy">
      <property name="filterInvocationDefinitionSource">
         <value>
             CONVERT_URL_TO_LOWERCASE_BEFORE_COMPARISON
             PATTERN_TYPE_APACHE_ANT
            /**=httpSessionContextIntegrationFilter,authenticationProcessingFilter,securityEnforcementFilter
         </value>
      </property>
    </bean>

With the heavy use of interfaces throughout the authentication system (Authentication, AuthenticationManager, AuthenticationProvider and UserDetailsService) it might be confusing to a new user to know which part of the authentication system to customize. In general, the following is recommended:

As briefly mentioned in the Authentication section, all Authentication implementations are required to store an array of GrantedAuthority objects. These represent the authorities that have been granted to the principal. The GrantedAuthority objects are inserted into the Authentication object by the AuthenticationManager and are later read by AccessDecisionManagers when making authorization decisions.

GrantedAuthority is an interface with only one method:

public String getAuthority();

This method allows AccessDecisionManagers to obtain a precise String representation of the GrantedAuthority. By returning a representation as a String, a GrantedAuthority can be easily "read" by most AccessDecisionManagers. If a GrantedAuthority cannot be precisely represented as a String, the GrantedAuthority is considered "complex" and getAuthority() must return null.

An example of a "complex" GrantedAuthority would be an implementation that stores a list of operations and authority thresholds that apply to different customer account numbers. Representing this complex GrantedAuthority as a String would be quite complex, and as a result the getAuthority() method should return null. This will indicate to any AccessDecisionManager that it will need to specifically support the GrantedAuthority implementation in order to understand its contents.

The Acegi Security System for Spring includes one concrete GrantedAuthority implementation, GrantedAuthorityImpl. This allows any user-specified String to be converted into a GrantedAuthority. All AuthenticationProviders included with the security architecture use GrantedAuthorityImpl to populate the Authentication object.

The AccessDecisionManager is called by the AbstractSecurityInterceptor and is responsible for making final access control decisions. The AccessDecisionManager interface contains three methods:

public void decide(Authentication authentication, Object object, ConfigAttributeDefinition config) throws AccessDeniedException;
public boolean supports(ConfigAttribute attribute);
public boolean supports(Class clazz);

As can be seen from the first method, the AccessDecisionManager is passed via method parameters all information that is likely to be of value in assessing an authorization decision. In particular, passing the secure Object enables those arguments contained in the actual secure object invocation to be inspected. For example, let's assume the secure object was a MethodInvocation. It would be easy to query the MethodInvocation for any Customer argument, and then implement some sort of security logic in the AccessDecisionManager to ensure the principal is permitted to operate on that customer. Implementations are expected to throw an AccessDeniedException if access is denied.

The supports(ConfigAttribute) method is called by the AbstractSecurityInterceptor at startup time to determine if the AccessDecisionManager can process the passed ConfigAttribute. The supports(Class) method is called by a security interceptor implementation to ensure the configured AccessDecisionManager supports the type of secure object that the security interceptor will present.

Whilst users can implement their own AccessDecisionManager to control all aspects of authorization, the Acegi Security System for Spring includes several AccessDecisionManager implementations that are based on voting. Figure 4 illustrates the relevant classes.

Figure 4: Voting Decision Manager

Using this approach, a series of AccessDecisionVoter implementations are polled on an authorization decision. The AccessDecisionManager then decides whether or not to throw an AccessDeniedException based on its assessment of the votes.

The AccessDecisionVoter interface has three methods:

public int vote(Authentication authentication, Object object, ConfigAttributeDefinition config);
public boolean supports(ConfigAttribute attribute);
public boolean supports(Class clazz);

Concrete implementations return an int, with possible values being reflected in the AccessDecisionVoter static fields ACCESS_ABSTAIN, ACCESS_DENIED and ACCESS_GRANTED. A voting implementation will return ACCESS_ABSTAIN if it has no opinion on an authorization decision. If it does have an opinion, it must return either ACCESS_DENIED or ACCESS_GRANTED.

There are three concrete AccessDecisionManagers provided with the Acegi Security System for Spring that tally the votes. The ConsensusBased implementation will grant or deny access based on the consensus of non-abstain votes. Properties are provided to control behavior in the event of an equality of votes or if all votes are abstain. The AffirmativeBased implementation will grant access if one or more ACCESS_GRANTED votes were received (ie a deny vote will be ignored, provided there was at least one grant vote). Like the ConsensusBased implementation, there is a parameter that controls the behavior if all voters abstain. The UnanimousBased provider expects unanimous ACCESS_GRANTED votes in order to grant access, ignoring abstains. It will deny access if there is any ACCESS_DENIED vote. Like the other implementations, there is a parameter that controls the behaviour if all voters abstain.

It is possible to implement a custom AccessDecisionManager that tallies votes differently. For example, votes from a particular AccessDecisionVoter might receive additional weighting, whilst a deny vote from a particular voter may have a veto effect.

There are two concrete AccessDecisionVoter implementations provided with the Acegi Security System for Spring. The RoleVoter class will vote if any ConfigAttribute begins with ROLE_. It will vote to grant access if there is a GrantedAuthority which returns a String representation (via the getAuthority() method) exactly equal to one or more ConfigAttributes starting with ROLE_. If there is no exact match of any ConfigAttribute starting with ROLE_, the RoleVoter will vote to deny access. If no ConfigAttribute begins with ROLE_, the voter will abstain. RoleVoter is case sensitive on comparisons as well as the ROLE_ prefix.

BasicAclEntryVoter is the other concrete voter included with Acegi Security. It integrates with Acegi Security's AclManager (discussed later). This voter is designed to have multiple instances in the same application context, such as:

<bean id="aclContactReadVoter" class="org.acegisecurity.vote.BasicAclEntryVoter">
  <property name="processConfigAttribute"><value>ACL_CONTACT_READ</value></property>
  <property name="processDomainObjectClass"><value>sample.contact.Contact</value></property>
  <property name="aclManager"><ref local="aclManager"/></property>
  <property name="requirePermission">
    <list>
      <ref local="org.acegisecurity.acl.basic.SimpleAclEntry.ADMINISTRATION"/>
      <ref local="org.acegisecurity.acl.basic.SimpleAclEntry.READ"/>
    </list>
  </property>
</bean>

<bean id="aclContactDeleteVoter" class="org.acegisecurity.vote.BasicAclEntryVoter">
  <property name="processConfigAttribute"><value>ACL_CONTACT_DELETE</value></property>
  <property name="processDomainObjectClass"><value>sample.contact.Contact</value></property>
  <property name="aclManager"><ref local="aclManager"/></property>
  <property name="requirePermission">
    <list>
      <ref local="org.acegisecurity.acl.basic.SimpleAclEntry.ADMINISTRATION"/>
      <ref local="org.acegisecurity.acl.basic.SimpleAclEntry.DELETE"/>
    </list>
  </property>
</bean>

In the above example, you'd define ACL_CONTACT_READ or ACL_CONTACT_DELETE against some methods on a MethodSecurityInterceptor or AspectJSecurityInterceptor. When those methods are invoked, the above applicable voter defined above would vote to grant or deny access. The voter would look at the method invocation to locate the first argument of type sample.contact.Contact, and then pass that Contact to the AclManager. The AclManager will then return an access control list (ACL) that applies to the current Authentication. Assuming that ACL contains one of the listed requirePermissions, the voter will vote to grant access. If the ACL does not contain one of the permissions defined against the voter, the voter will vote to deny access. BasicAclEntryVoter is an important class as it allows you to build truly complex applications with domain object security entirely defined in the application context. If you're interested in learning more about Acegi Security's ACL capabilities and how best to apply them, please see the ACL and "After Invocation" sections of this reference guide, and the Contacts sample application.

It is also possible to implement a custom AccessDecisionVoter. Several examples are provided in the Acegi Security System for Spring unit tests, including ContactSecurityVoter and DenyVoter. The ContactSecurityVoter abstains from voting decisions where a CONTACT_OWNED_BY_CURRENT_USER ConfigAttribute is not found. If voting, it queries the MethodInvocation to extract the owner of the Contact object that is subject of the method call. It votes to grant access if the Contact owner matches the principal presented in the Authentication object. It could have just as easily compared the Contact owner with some GrantedAuthority the Authentication object presented. All of this is achieved with relatively few lines of code and demonstrates the flexibility of the authorization model.

The Acegi Security System for Spring comes bundled with several JSP tag libraries that eases JSP writing. The tag libraries are known as authz and provide a range of different services.

All taglib classes are included in the core acegi-security-xx.jar file, with the authz.tld located in the JAR's META-INF directory. This means for JSP 1.2+ web containers you can simply include the JAR in the WAR's WEB-INF/lib directory and it will be available. If you're using a JSP 1.1 container, you'll need to declare the JSP taglib in your web.xml file, and include authz.tld in the WEB-INF/lib directory. The following fragment is added to web.xml:

<taglib>
       <taglib-uri>http://acegisecurity.sf.net/authz</taglib-uri>
       <taglib-location>/WEB-INF/authz.tld</taglib-location>
</taglib>

<authz:authorize ifAllGranted="ROLE_SUPERVISOR">
  <td>
    <A HREF="del.htm?id=<c:out value="${contact.id}"/>">Del</A>
  </td>
</authz:authorize>

<authz:authentication operation="username"/>

<authz:acl domainObject="${contact}" hasPermission="16,1">
  <td><A HREF="<c:url value="del.htm"><c:param name="contactId" value="${contact.id}"/></c:url>">Del</A></td>
</authz:acl>

Given there are several ways to achieve similar authorization outcomes in the Acegi Security System for Spring, the following general recommendations are made:

Whilst the AccessDecisionManager is called by the AbstractSecurityInterceptor before proceeding with the secure object invocation, some applications need a way of modifying the object actually returned by the secure object invocation. Whilst you could easily implement your own AOP concern to achieve this, Acegi Security provides a convenient hook that has several concrete implementations that integrate with its ACL capabilities.

Figure 5 illustrates Acegi Security's AfterInvocationManager and its concrete implementations.

Figure 5: After Invocation Implementation

Like many other parts of Acegi Security, AfterInvocationManager has a single concrete implementation, AfterInvocationProvider, which polls a list of AfterInvocationProviders. Each AfterInvocationProvider is allowed to modify the return object or throw an AccessDeniedException. Indeed multiple providers can modify the object, as the result of the previous provider is passed to the next in the list. Let's now consider our ACL-aware implementations of AfterInvocationProvider.

Please be aware that if you're using AfterInvocationManager, you will still need configuration attributes that allow the MethodSecurityInterceptor's AccessDecisionManager to allow an operation. If you're using the typical Acegi Security included AccessDecisionManager implementations, having no configuration attributes defined for a particular secure method invocation will cause each AccessDecisionVoter to abstain from voting. In turn, if the AccessDecisionManager property "allowIfAllAbstainDecisions" is false, an AccessDeniedException will be thrown. You may avoid this potential issue by either (i) setting "allowIfAllAbstainDecisions" to true (although this is generally not recommended) or (ii) simply ensure that there is at least one configuration attribute that an AccessDecisionVoter will vote to grant access for. This latter (recommended) approach is usually achieved through a ROLE_USER or ROLE_AUTHENTICATED configuration attribute.

A common services layer method we've all written at one stage or another looks like this:

public Contact getById(Integer id);

Quite often, only principals with permission to read the Contact should be allowed to obtain it. In this situation the AccessDecisionManager approach provided by the AbstractSecurityInterceptor will not suffice. This is because the identity of the Contact is all that is available before the secure object is invoked. The BasicAclAfterInvocationProvider delivers a solution, and is configured as follows:

<bean id="afterAclRead" class="org.acegisecurity.afterinvocation.BasicAclEntryAfterInvocationProvider">
  <property name="aclManager"><ref local="aclManager"/></property>
  <property name="requirePermission">
    <list>
      <ref local="org.acegisecurity.acl.basic.SimpleAclEntry.ADMINISTRATION"/>
      <ref local="org.acegisecurity.acl.basic.SimpleAclEntry.READ"/>
    </list>
  </property>
</bean>

In the above example, the Contact will be retrieved and passed to the BasicAclEntryAfterInvocationProvider. The provider will thrown an AccessDeniedException if one of the listed requirePermissions is not held by the Authentication. The BasicAclEntryAfterInvocationProvider queries the AclManager to determine the ACL that applies for this domain object to this Authentication.

Similar to the BasicAclEntryAfterInvocationProvider is BasicAclEntryAfterInvocationCollectionFilteringProvider. It is designed to remove Collection or array elements for which a principal does not have access. It never thrown an AccessDeniedException - simply silently removes the offending elements. The provider is configured as follows:

<bean id="afterAclCollectionRead" class="org.acegisecurity.afterinvocation.BasicAclEntryAfterInvocationCollectionFilteringProvider">
  <property name="aclManager"><ref local="aclManager"/></property>
  <property name="requirePermission">
    <list>
      <ref local="org.acegisecurity.acl.basic.SimpleAclEntry.ADMINISTRATION"/>
      <ref local="org.acegisecurity.acl.basic.SimpleAclEntry.READ"/>
    </list>
  </property>
</bean>

As you can imagine, the returned Object must be a Collection or array for this provider to operate. It will remove any element if the AclManager indicates the Authentication does not hold one of the listed requirePermissions.

The Contacts sample application demonstrates these two AfterInvocationProviders.

The AbstractSecurityInterceptor is able to temporarily replace the Authentication object in the SecurityContext and SecurityContextHolder during the SecurityInterceptorCallback. This only occurs if the original Authentication object was successfully processed by the AuthenticationManager and AccessDecisionManager. The RunAsManager will indicate the replacement Authentication object, if any, that should be used during the SecurityInterceptorCallback.

By temporarily replacing the Authentication object during a SecurityInterceptorCallback, the secured invocation will be able to call other objects which require different authentication and authorization credentials. It will also be able to perform any internal security checks for specific GrantedAuthority objects. Because Acegi Security provides a number of helper classes that automatically configure remoting protocols based on the contents of the ContextHolder, these run-as replacements are particularly useful when calling remote web services.

A RunAsManager interface is provided by the Acegi Security System for Spring:

public Authentication buildRunAs(Authentication authentication, Object object, ConfigAttributeDefinition config);
public boolean supports(ConfigAttribute attribute);
public boolean supports(Class clazz);

The first method returns the Authentication object that should replace the existing Authentication object for the duration of the method invocation. If the method returns null, it indicates no replacement should be made. The second method is used by the AbstractSecurityInterceptor as part of its startup validation of configuration attributes. The supports(Class) method is called by a security interceptor implementation to ensure the configured RunAsManager supports the type of secure object that the security interceptor will present.

One concrete implementation of a RunAsManager is provided with the Acegi Security System for Spring. The RunAsManagerImpl class returns a replacement RunAsUserToken if any ConfigAttribute starts with RUN_AS_. If any such ConfigAttribute is found, the replacement RunAsUserToken will contain the same principal, credentials and granted authorities as the original Authentication object, along with a new GrantedAuthorityImpl for each RUN_AS_ ConfigAttribute. Each new GrantedAuthorityImpl will be prefixed with ROLE_, followed by the RUN_AS ConfigAttribute. For example, a RUN_AS_SERVER will result in the replacement RunAsUserToken containing a ROLE_RUN_AS_SERVER granted authority.

The replacement RunAsUserToken is just like any other Authentication object. It needs to be authenticated by the AuthenticationManager, probably via delegation to a suitable AuthenticationProvider. The RunAsImplAuthenticationProvider performs such authentication. It simply accepts as valid any RunAsUserToken presented.

To ensure malicious code does not create a RunAsUserToken and present it for guaranteed acceptance by the RunAsImplAuthenticationProvider, the hash of a key is stored in all generated tokens. The RunAsManagerImpl and RunAsImplAuthenticationProvider is created in the bean context with the same key:

<bean id="runAsManager" class="org.acegisecurity.runas.RunAsManagerImpl">
  <property name="key"><value>my_run_as_password</value></property>
</bean>

<bean id="runAsAuthenticationProvider" class="org.acegisecurity.runas.RunAsImplAuthenticationProvider">
  <property name="key"><value>my_run_as_password</value></property>
</bean>

By using the same key, each RunAsUserToken can be validated it was created by an approved RunAsManagerImpl. The RunAsUserToken is immutable after creation for security reasons.

Everything presented so far assumes one thing: the SecurityContextHolder is populated with a valid SecurityContext, which in turn contains a valid Authentication object. Developers are free to do this in whichever way they like, such as directly calling the relevant objects at runtime. However, several classes have been provided to make this process transparent in many situations. We call these classes "authentication mechanisms".

The org.acegisecurity.ui package provides what we call "authentication processing mechanisms". An authentication processing mechanism is solely concerned with received an authentication request from the principal, testing if it seems valid, and if so, placing the authentication request token onto the SecurityContextHolder. Of course, if the authentication request is invalid, the authentication processing mechanism is responsible for informing the principal in whatever way is appropriate to the protocol.

Recall the HttpSessionContextIntegrationFilter (discussed in the context section) is responsible for storing the SecurityContextHolder contents between invocations. This means no authentication processing mechanism need ever interact directly with HttpSession. Indeed HttpSessionContextIntegrationFilter has been designed to minimise the unnecessary creation of HttpSessions, as might occur when using Basic authentication for example.

There are several authentication processing mechanisms included with Acegi Security, which will be briefly discussed in this chapter. The most popular (and almost always recommended) approach is HTTP Form Authentication, which uses a login form to authenticate the user. Another approach (commonly use with web services) is HTTP Basic Authentication, which allows clients to use HTTP headers to present authentication information to the Acegi Security System for Spring. Alternatively, you can also use Yale Central Authentication Service (CAS) for enterprise-wide single sign on. The final (and generally unrecommended) approach is via Container Adapters, which allow supported web containers to perform the authentication themselves. HTTP Form Authentication and Basic Authentication is discussed below, whilst CAS and Container Adapters are discussed in separate sections of this document.

HTTP Form Authentication involves using the AuthenticationProcessingFilter to process a login form. The login form simply contains j_username and j_password input fields, and posts to a URL that is monitored by the filter (by default j_acegi_security_check). The filter is defined in web.xml behind a FilterToBeanProxy as follows:

<filter>
  <filter-name>Acegi Authentication Processing Filter</filter-name>
  <filter-class>org.acegisecurity.util.FilterToBeanProxy</filter-class>
  <init-param>
    <param-name>targetClass</param-name>
    <param-value>org.acegisecurity.ui.webapp.AuthenticationProcessingFilter</param-value>
  </init-param>
</filter>

<filter-mapping>
  <filter-name>Acegi Authentication Processing Filter</filter-name>
  <url-pattern>/*</url-pattern>
</filter-mapping>

For a discussion of FilterToBeanProxy, please refer to the Filters section. The application context will need to define the AuthenticationProcessingFilter:

<bean id="authenticationProcessingFilter" class="org.acegisecurity.ui.webapp.AuthenticationProcessingFilter">
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
  <property name="authenticationFailureUrl"><value>/acegilogin.jsp?login_error=1</value></property>
  <property name="defaultTargetUrl"><value>/</value></property>
  <property name="filterProcessesUrl"><value>/j_acegi_security_check</value></property>
</bean>

The configured AuthenticationManager processes each authentication request. If authentication fails, the browser will be redirected to the authenticationFailureUrl. The AuthenticationException will be placed into the HttpSession attribute indicated by AbstractProcessingFilter.ACEGI_SECURITY_LAST_EXCEPTION_KEY, enabling a reason to be provided to the user on the error page.

If authentication is successful, the resulting Authentication object will be placed into the SecurityContextHolder.

Once the SecurityContextHolder has been updated, the browser will need to be redirected to the target URL. The target URL is usually indicated by the HttpSession attribute specified by AbstractProcessingFilter.ACEGI_SECURITY_TARGET_URL_KEY. This attribute is automatically set by the SecurityEnforcementFilter when an AuthenticationException occurs, so that after login is completed the user can return to what they were trying to access. If for some reason the HttpSession does not indicate the target URL, the browser will be redirected to the defaultTargetUrl property.

Because this authentication approach is fully contained within a single web application, HTTP Form Authentication is recommended to be used instead of Container Adapters.

The Acegi Security System for Spring provides a BasicProcessingFilter which is capable of processing basic authentication credentials presented in HTTP headers. This can be used for authenticating calls made by Spring remoting protocols (such as Hessian and Burlap), as well as normal user agents (such as Internet Explorer and Navigator). The standard governing HTTP Basic Authentication is defined by RFC 1945, Section 11, and the BasicProcessingFilter conforms with this RFC. Basic Authentication is an attractive approach to authentication, because it is very widely deployed in user agents and implementation is extremely simple (it's just a Base64 encoding of the username:password, specified in a HTTP header).

To implement HTTP Basic Authentication, it is necessary to define BasicProcessingFilter in the fitler chain. The application context will need to define the BasicProcessingFilter and its required collaborator:

<bean id="basicProcessingFilter" class="org.acegisecurity.ui.basicauth.BasicProcessingFilter">
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
  <property name="authenticationEntryPoint"><ref bean="authenticationEntryPoint"/></property>
</bean>

<bean id="authenticationEntryPoint" class="org.acegisecurity.ui.basicauth.BasicProcessingFilterEntryPoint">
  <property name="realmName"><value>Name Of Your Realm</value></property>
</bean>

The configured AuthenticationManager processes each authentication request. If authentication fails, the configured AuthenticationEntryPoint will be used to retry the authentication process. Usually you will use the BasicProcessingFilterEntryPoint, which returns a 401 response with a suitable header to retry HTTP Basic authentication. If authentication is successful, the resulting Authentication object will be placed into the SecurityContextHolder.

If the authentication event was successful, or authentication was not attempted because the HTTP header did not contain a supported authentication request, the filter chain will continue as normal. The only time the filter chain will be interrupted is if authentication fails and the AuthenticationEntryPoint is called, as discussed in the previous paragraph.

The Acegi Security System for Spring provides a DigestProcessingFilter which is capable of processing digest authentication credentials presented in HTTP headers. Digest Authentication attempts to solve many of the weakenesses of Basic authentication, specifically by ensuring credentials are never sent in clear text across the wire. Many user agents support Digest Authentication, including FireFox and Internet Explorer. The standard governing HTTP Digest Authentication is defined by RFC 2617, which updates an earlier version of the Digest Authentication standard prescribed by RFC 2069. Most user agents implement RFC 2617. The Acegi Security DigestProcessingFilter is compatible with the "auth" quality of protection (qop) prescribed by RFC 2617, which also provides backward compatibility with RFC 2069. Digest Authentication is a highly attractive option if you need to use unencrypted HTTP (ie no TLS/HTTPS) and wish to maximise security of the authentication process. Indeed Digest Authentication is a mandatory requirement for the WebDAV protocol, as noted by RFC 2518 Section 17.1, so we should expect to see it increasingly deployed and replacing Basic Authentication.

Digest Authentication is definitely the most secure choice between Form Authentication, Basic Authentication and Digest Authentication, although extra security also means more complex user agent implementations. Central to Digest Authentication is a "nonce". This is a value the server generates. Acegi Security's nonce adopts the following format:

base64(expirationTime + ":" + md5Hex(expirationTime + ":" + key))

expirationTime:   The date and time when the nonce expires, expressed in milliseconds
key:              A private key to prevent modification of the nonce token

The DigestProcessingFilterEntryPoint has a property specifying the key used for generating the nonce tokens, along with a nonceValiditySeconds property for determining the expiration time (default 300, which equals five minutes). Whilstever the nonce is valid, the digest is computed by concatenating various strings including the username, password, nonce, URI being requested, a client-generated nonce (merely a random value which the user agent generates each request), the realm name etc, then performing an MD5 hash. Both the server and user agent perform this digest computation, resulting in different hash codes if they disagree on an included value (eg password). In the Acegi Security implementation, if the server-generated nonce has merely expired (but the digest was otherwise valid), the DigestProcessingFilterEntryPoint will send a "stale=true" header. This tells the user agent there is no need to disturb the user (as the password and username etc is correct), but simply to try again using a new nonce.

An appropriate value for DigestProcessingFilterEntryPoint's nonceValiditySeconds parameter will depend on your application. Extremely secure applications should note that an intercepted authentication header can be used to impersonate the principal until the expirationTime contained in the nonce is reached. This is the key principle when selecting an appropriate setting, but it would be unusual for immensly secure applications to not be running over TLS/HTTPS in the first instance.

Because of the more complex implementation of Digest Authentication, there are often user agent issues. For example, Internet Explorer fails to present an "opaque" token on subsequent requests in the same session. The Acegi Security filters therefore encapsulate all state information into the "nonce" token instead. In our testing, the Acegi Security implementation works reliably with FireFox and Internet Explorer, correctly handling nonce timeouts etc.

Now that we've reviewed the theory, let's see how to use it. To implement HTTP Digest Authentication, it is necessary to define DigestProcessingFilter in the fitler chain. The application context will need to define the DigestProcessingFilter and its required collaborators:

<bean id="digestProcessingFilter" class="org.acegisecurity.ui.digestauth.DigestProcessingFilter">
  <property name="userDetailsService"><ref local="jdbcDaoImpl"/></property>
  <property name="authenticationEntryPoint"><ref local="digestProcessingFilterEntryPoint"/></property>
  <property name="userCache"><ref local="userCache"/></property>
</bean>

<bean id="digestProcessingFilterEntryPoint" class="org.acegisecurity.ui.digestauth.DigestProcessingFilterEntryPoint">
  <property name="realmName"><value>Contacts Realm via Digest Authentication</value></property>
  <property name="key"><value>acegi</value></property>
  <property name="nonceValiditySeconds"><value>10</value></property>
</bean>

The configured UserDetailsService is needed because DigestProcessingFilter must have direct access to the clear text password of a user. Digest Authentication will NOT work if you are using encoded passwords in your DAO. The DAO collaborator, along with the UserCache, are typically shared directly with a DaoAuthenticationProvider. The authenticationEntryPoint property must be DigestProcessingFilterEntryPoint, so that DigestProcessingFilter can obtain the correct realmName and key for digest calculations.

Like BasicAuthenticationFilter, if authentication is successful an Authentication request token will be placed into the SecurityContextHolder. If the authentication event was successful, or authentication was not attempted because the HTTP header did not contain a Digest Authentication request, the filter chain will continue as normal. The only time the filter chain will be interrupted is if authentication fails and the AuthenticationEntryPoint is called, as discussed in the previous paragraph.

Digest Authentication's RFC offers a range of additional features to further increase security. For example, the nonce can be changed on every request. Despite this, the Acegi Security implementation was designed to minimise the complexity of the implementation (and the doubtless user agent incompatibilities that would emerge), and avoid needing to store server-side state. You are invited to review RFC 2617 if you wish to explore these features in more detail. As far as we are aware, the Acegi Security implementation does comply with the minimum standards of this RFC.

Particularly in the case of web request URI security, sometimes it is more convenient to assign configuration attributes against every possible secure object invocation. Put differently, sometimes it is nice to say ROLE_SOMETHING is required by default and only allow certain exceptions to this rule, such as for login, logout and home pages of an application. There are also other situations where anonymous authentication would be desired, such as when an auditing interceptor queries the SecurityContextHolder to identify which principal was responsible for a given operation. Such classes can be authored with more robustness if they know the SecurityContextHolder always contains an Authentication object, and never null.

Acegi Security provides three classes that together provide an anoymous authentication feature. AnonymousAuthenticationToken is an implementation of Authentication, and stores the GrantedAuthority[]s which apply to the anonymous principal. There is a corresponding AnonymousAuthenticationProvider, which is chained into the ProviderManager so that AnonymousAuthenticationTokens are accepted. Finally, there is an AnonymousProcessingFilter, which is chained after the normal authentication mechanisms and automatically add an AnonymousAuthenticationToken to the SecurityContextHolder if there is no existing Authentication held there. The definition of the filter and authentication provider appears as follows:

<bean id="anonymousProcessingFilter" class="org.acegisecurity.providers.anonymous.AnonymousProcessingFilter">
  <property name="key"><value>foobar</value></property>
  <property name="userAttribute"><value>anonymousUser,ROLE_ANONYMOUS</value></property>
</bean>

<bean id="anonymousAuthenticationProvider" class="org.acegisecurity.providers.anonymous.AnonymousAuthenticationProvider">
  <property name="key"><value>foobar</value></property>
</bean>

The key is shared between the filter and authentication provider, so that tokens created by the former are accepted by the latter. The userAttribute is expressed in the form of usernameInTheAuthenticationToken,grantedAuthority[,grantedAuthority]. This is the same syntax as used after the equals sign for InMemoryDaoImpl's userMap property.

As explained earlier, the benefit of anonymous authentication is that all URI patterns can have security applied to them. For example:

<bean id="filterInvocationInterceptor" class="org.acegisecurity.intercept.web.FilterSecurityInterceptor">
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
  <property name="accessDecisionManager"><ref local="httpRequestAccessDecisionManager"/></property>
  <property name="objectDefinitionSource">
    <value>
      CONVERT_URL_TO_LOWERCASE_BEFORE_COMPARISON
      PATTERN_TYPE_APACHE_ANT
      /index.jsp=ROLE_ANONYMOUS,ROLE_USER
      /hello.htm=ROLE_ANONYMOUS,ROLE_USER
      /logoff.jsp=ROLE_ANONYMOUS,ROLE_USER
      /acegilogin.jsp*=ROLE_ANONYMOUS,ROLE_USER
      /**=ROLE_USER
    </value>
  </property>
</bean>

Remember-me authentication refers to web sites being able to remember the identity of a principal between sessions. This is typically accomplished by sending a cookie to the browser, with the cookie being detected during future sessions and causing automated login to take place. Acegi Security provides the necessary hooks so that such operations can take place, along with providing a concrete implementation that uses hashing to preserve the security of cookie-based tokens.

Remember-me authentication is not used with digest or basic authentication, given they are often not used with HttpSessions. Remember-me is used with AuthenticationProcessingFilter, and is implemented via hooks in the AbstractProcessingFilter superclass. The hooks will invoke a concrete RememberMeServices at the appropriate times. The interface looks like this:

public Authentication autoLogin(HttpServletRequest request, HttpServletResponse response);
public void loginFail(HttpServletRequest request, HttpServletResponse response);
public void loginSuccess(HttpServletRequest request, HttpServletResponse response, Authentication successfulAuthentication);

Please refer to JavaDocs for a fuller discussion on what the methods do, although note at this stage AbstractProcessingFilter only calls the loginFail() and loginSuccess() methods. The autoLogin() method is called by RememberMeProcessingFilter whenever the SecurityContextHolder does not contain an Authentication. This interface therefore provides the underlaying remember-me implementation with sufficient notification of authentication-related events, and delegates to the implementation whenever a candidate web request might contain a cookie and wish to be remembered.

This design allows any number of remember-me implementation strategies. In the interests of simplicity and avoiding the need for DAO implementations that specify write and create methods, Acegi Security's only concrete implementation, TokenBasedRememberMeServices, uses hashing to achieve a useful remember-me strategy. In essence a cookie is sent to the browser upon successful interactive authentication, with that cookie being composed as follows:

base64(username + ":" + expirationTime + ":" + md5Hex(username + ":" + expirationTime + ":" password + ":" + key))

username:         As identifiable to TokenBasedRememberMeServices.getUserDetailsService()
password:         That matches the relevant UserDetails retrieved from TokenBasedRememberMeServices.getUserDetailsService()
expirationTime:   The date and time when the remember-me token expires, expressed in milliseconds
key:              A private key to prevent modification of the remember-me token

As such the remember-me token is valid only for the period specified, and provided that the username, password and key does not change. Notably, this has a potential security issue issue in that a captured remember-me token will be usable from any user agent until such time as the token expires. This is the same issue as with digest authentication. If a principal is aware a token has been captured, they can easily change their password and immediately invalidate all remember-me tokens on issue. However, if more significant security is needed a rolling token approach should be used (this would require a database) or remember-me services should simply not be used.

TokenBasedRememberMeServices generates a RememberMeAuthenticationToken, which is processed by RememberMeAuthenticationProvider. A key is shared between this authentication provider and the TokenBasedRememberMeServices. In addition, TokenBasedRememberMeServices requires A UserDetailsService from which it can retrieve the username and password for signature comparison purposes, and generate the RememberMeAuthenticationToken to contain the correct GrantedAuthority[]s. Some sort of logout command should be provided by the application (typically via a JSP) that invalidates the cookie upon user request. See the Contacts Sample application's logout.jsp for an example.

The beans required in an application context to enable remember-me services are as follows:

<bean id="rememberMeProcessingFilter" class="org.acegisecurity.ui.rememberme.RememberMeProcessingFilter">
  <property name="rememberMeServices"><ref local="rememberMeServices"/></property>
</bean>

<bean id="rememberMeServices" class="org.acegisecurity.ui.rememberme.TokenBasedRememberMeServices">
  <property name="userDetailsService"><ref local="jdbcDaoImpl"/></property>
  <property name="key"><value>springRocks</value></property>
</bean>
   
<bean id="rememberMeAuthenticationProvider" class="org.acegisecurity.providers.rememberme.RememberMeAuthenticationProvider">
  <property name="key"><value>springRocks</value></property>
</bean>

Prior to release 0.8.0, Acegi Security referred to "well-known locations" in discussions about storing the Authentication. This approach did not explicitly separate the function of HttpSession storage of SecurityContextHolder contents from the processing of authentication requests received through various protocols. In addition, the previous approach did not facilitate storage of non-Authentication objects between requests, which was limiting usefulness of the SecurityContextHolder system to member of the community. For these reasons, the notion of well-known locations was abandoned, the HttpSessionContextIntegrationFilter was established, and the purpose of authentication processing mechanisms was explicitly defined and limited to interaction with the SecurityContextHolder only. There is no need to refer to well-known locations any more and we hope this clearer separation of responsibilities enhances understanding of the design.

Very early versions of the Acegi Security System for Spring exclusively used Container Adapters for interfacing authentication with end users. Whilst this worked well, it required considerable time to support multiple container versions and the configuration itself was relatively time-consuming for developers. For this reason the HTTP Form Authentication and HTTP Basic Authentication approaches were developed, and are today recommended for almost all applications.

Container Adapters enable the Acegi Security System for Spring to integrate directly with the containers used to host end user applications. This integration means that applications can continue to leverage the authentication and authorization capabilities built into containers (such as isUserInRole() and form-based or basic authentication), whilst benefiting from the enhanced security interception capabilities provided by the Acegi Security System for Spring (it should be noted that Acegi Security also offers ContextHolderAwareRequestWrapper to deliver isUserInRole() and similar Servlet Specification compatibility methods).

The integration between a container and the Acegi Security System for Spring is achieved through an adapter. The adapter provides a container-compatible user authentication provider, and needs to return a container-compatible user object.

The adapter is instantiated by the container and is defined in a container-specific configuration file. The adapter then loads a Spring application context which defines the normal authentication manager settings, such as the authentication providers that can be used to authenticate the request. The application context is usually named acegisecurity.xml and is placed in a container-specific location.

The Acegi Security System for Spring currently supports Jetty, Catalina (Tomcat), JBoss and Resin. Additional container adapters can easily be written.

As is always the case, the container adapter generated Authentication object still needs to be authenticated by an AuthenticationManager when requested to do so by the AbstractSecurityInterceptor. The AuthenticationManager needs to be certain the adapter-provided Authentication object is valid and was actually authenticated by a trusted adapter.

Adapters create Authentication objects which are immutable and implement the AuthByAdapter interface. These objects store the hash of a key that is defined by the adapter. This allows the Authentication object to be validated by the AuthByAdapterProvider. This authentication provider is defined as follows:

<bean id="authByAdapterProvider" class="org.acegisecurity.adapters.AuthByAdapterProvider">
  <property name="key"><value>my_password</value></property>
</bean>

The key must match the key that is defined in the container-specific configuration file that starts the adapter. The AuthByAdapterProvider automatically accepts as valid any AuthByAdapter implementation that returns the expected hash of the key.

To reiterate, this means the adapter will perform the initial authentication using providers such as DaoAuthenticationProvider, returning an AuthByAdapter instance that contains a hash code of the key. Later, when an application calls a security interceptor managed resource, the AuthByAdapter instance in the SecurityContext in the SecurityContextHolder will be tested by the application's AuthByAdapterProvider. There is no requirement for additional authentication providers such as DaoAuthenticationProvider within the application-specific application context, as the only type of Authentication instance that will be presented by the application is from the container adapter.

Classloader issues are frequent with containers and the use of container adapters illustrates this further. Each container requires a very specific configuration. The installation instructions are provided below. Once installed, please take the time to try the sample application to ensure your container adapter is properly configured.

When using container adapters with the DaoAuthenticationProvider, ensure you set its forcePrincipalAsString property to true.

The following was tested with Jakarta Tomcat 4.1.30 and 5.0.19.

$CATALINA_HOME refers to the root of your Catalina (Tomcat) installation.

Edit your $CATALINA_HOME/conf/server.xml file so the <Engine> section contains only one active <Realm> entry. An example realm entry:

      <Realm className="org.acegisecurity.adapters.catalina.CatalinaAcegiUserRealm"
             appContextLocation="conf/acegisecurity.xml"
             key="my_password" />

Be sure to remove any other <Realm> entry from your <Engine> section.

Copy acegisecurity.xml into $CATALINA_HOME/conf.

Copy acegi-security-catalina-XX.jar into $CATALINA_HOME/server/lib.

Copy the following files into $CATALINA_HOME/common/lib:

None of the above JAR files (or acegi-security-XX.jar) should be in your application's WEB-INF/lib. The realm name indicated in your web.xml does not matter with Catalina.

We have received reports of problems using this Container Adapter with Mac OS X. A work-around is to use a script such as follows:

#!/bin/sh
export CATALINA_HOME="/Library/Tomcat"
export JAVA_HOME="/Library/Java/Home"
cd /
$CATALINA_HOME/bin/startup.sh

The following was tested with Jetty 4.2.18.

$JETTY_HOME refers to the root of your Jetty installation.

Edit your $JETTY_HOME/etc/jetty.xml file so the <Configure class> section has a new addRealm call:

  <Call name="addRealm">
    <Arg>
      <New class="org.acegisecurity.adapters.jetty.JettyAcegiUserRealm">
        <Arg>Spring Powered Realm</Arg>
        <Arg>my_password</Arg>
        <Arg>etc/acegisecurity.xml</Arg>
      </New>
    </Arg>
  </Call>

Copy acegisecurity.xml into $JETTY_HOME/etc.

Copy the following files into $JETTY_HOME/ext:

None of the above JAR files (or acegi-security-XX.jar) should be in your application's WEB-INF/lib. The realm name indicated in your web.xml does matter with Jetty. The web.xml must express the same <realm-name> as your jetty.xml (in the example above, "Spring Powered Realm").

The following was tested with JBoss 3.2.6.

$JBOSS_HOME refers to the root of your JBoss installation.

There are two different ways of making spring context available to the Jboss integration classes.

The first approach is by editing your $JBOSS_HOME/server/your_config/conf/login-config.xml file so that it contains a new entry under the <Policy> section:

    <application-policy name = "SpringPoweredRealm">
       <authentication>
          <login-module code = "org.acegisecurity.adapters.jboss.JbossSpringLoginModule"
            flag = "required">
            <module-option name = "appContextLocation">acegisecurity.xml</module-option>
            <module-option name = "key">my_password</module-option>
         </login-module>
       </authentication>
    </application-policy>

Copy acegisecurity.xml into $JBOSS_HOME/server/your_config/conf.

In this configuration acegisecurity.xml contains the spring context definition including all the authentication manager beans. You have to bear in mind though, that SecurityContext is created and destroyed on each login request, so the login operation might become costly. Alternatively, the second approach is to use Spring singleton capabilities through org.springframework.beans.factory.access.SingletonBeanFactoryLocator. The required configuration for this approach is:

    <application-policy name = "SpringPoweredRealm">
       <authentication>
          <login-module code = "org.acegisecurity.adapters.jboss.JbossSpringLoginModule"
            flag = "required">
            <module-option name = "singletonId">springRealm</module-option>
            <module-option name = "key">my_password</module-option>
            <module-option name = "authenticationManager">authenticationManager</module-option>
         </login-module>
       </authentication>
    </application-policy>

In the above code fragment, authenticationManager is a helper property that defines the expected name of the AuthenticationManager in case you have several defined in the IoC container. The singletonId property references a bean defined in a beanRefFactory.xml file. This file needs to be available from anywhere on the JBoss classpath, including $JBOSS_HOME/server/your_config/conf. The beanRefFactory.xml contains the following declaration:

<beans>
  <bean id="springRealm" singleton="true" lazy-init="true" class="org.springframework.context.support.ClassPathXmlApplicationContext">
    <constructor-arg>
      <list>
        <value>acegisecurity.xml</value>
      </list>
    </constructor-arg>
  </bean>
</beans>

Finally, irrespective of the configuration approach you need to copy the following files into $JBOSS_HOME/server/your_config/lib:

None of the above JAR files (or acegi-security-XX.jar) should be in your application's WEB-INF/lib. The realm name indicated in your web.xml does not matter with JBoss. However, your web application's WEB-INF/jboss-web.xml must express the same <security-domain> as your login-config.xml. For example, to match the above example, your jboss-web.xml would look like this:

<jboss-web>
  <security-domain>java:/jaas/SpringPoweredRealm</security-domain>
</jboss-web>

The following was tested with Resin 3.0.6.

$RESIN_HOME refers to the root of your Resin installation.

Resin provides several ways to support the container adapter. In the instructions below we have elected to maximise consistency with other container adapter configurations. This will allow Resin users to simply deploy the sample application and confirm correct configuration. Developers comfortable with Resin are naturally able to use its capabilities to package the JARs with the web application itself, and/or support single sign-on.

Copy the following files into $RESIN_HOME/lib:

Unlike the container-wide acegisecurity.xml files used by other container adapters, each Resin web application will contain its own WEB-INF/resin-acegisecurity.xml file. Each web application will also contain a resin-web.xml file which Resin uses to start the container adapter:

<web-app>
  <authenticator>
    <type>org.acegisecurity.adapters.resin.ResinAcegiAuthenticator</type>
    <init>
      <app-context-location>WEB-INF/resin-acegisecurity.xml</app-context-location>
      <key>my_password</key>
    </init>
  </authenticator>
</web-app>

With the basic configuration provided above, none of the JAR files listed (or acegi-security-XX.jar) should be in your application's WEB-INF/lib. The realm name indicated in your web.xml does not matter with Resin, as the relevant authentication class is indicated by the <authenticator> setting.

Yale University produces an enterprise-wide single sign on system known as CAS. Unlike other initiatives, Yale's Central Authentication Service is open source, widely used, simple to understand, platform independent, and supports proxy capabilities. The Acegi Security System for Spring fully supports CAS, and provides an easy migration path from single-application deployments of Acegi Security through to multiple-application deployments secured by an enterprise-wide CAS server.

You can learn more about CAS at http://www.yale.edu/tp/auth/. You will need to visit this URL to download the CAS Server files. Whilst the Acegi Security System for Spring includes two CAS libraries in the "-with-dependencies" ZIP file, you will still need the CAS Java Server Pages and web.xml to customise and deploy your CAS server.

Whilst the CAS web site above contains two documents that detail the architecture of CAS, we present the general overview again here within the context of the Acegi Security System for Spring. The following refers to CAS 2.0, being the version of CAS that Acegi Security System for Spring supports.

Somewhere in your enterprise you will need to setup a CAS server. The CAS server is simply a standard WAR file, so there isn't anything difficult about setting up your server. Inside the WAR file you will customise the login and other single sign on pages displayed to users. You will also need to specify in the web.xml a PasswordHandler. The PasswordHandler has a simple method that returns a boolean as to whether a given username and password is valid. Your PasswordHandler implementation will need to link into some type of backend authentication repository, such as an LDAP server or database.

If you are already running an existing CAS server instance, you will have already established a PasswordHandler. If you do not already have a PasswordHandler, you might prefer to use the Acegi Security System for Spring CasPasswordHandler class. This class delegates through to the standard Acegi Security AuthenticationManager, enabling you to use a security configuration you might already have in place. You do not need to use the CasPasswordHandler class on your CAS server if you do not wish. The Acegi Security System for Spring will function as a CAS client successfully irrespective of the PasswordHandler you've chosen for your CAS server.

Apart from the CAS server itself, the other key player is of course the secure web applications deployed throughout your enterprise. These web applications are known as "services". There are two types of services: standard services and proxy services. A proxy service is able to request resources from other services on behalf of the user. This will be explained more fully later.

Services can be developed in a large variety of languages, due to CAS 2.0's very light XML-based protocol. The Yale CAS home page contains a clients archive which demonstrates CAS clients in Java, Active Server Pages, Perl, Python and others. Naturally, Java support is very strong given the CAS server is written in Java. You do not need to use any of CAS' client classes in applications secured by the Acegi Security System for Spring. This is handled transparently for you.

The basic interaction between a web browser, CAS server and an Acegi Security for System Spring secured service is as follows:

It's good that you're still here! It might sound involved, but you can relax as the Acegi Security System for Spring classes hide much of the complexity. Let's now look at how this is configured.

As mentioned above, the Acegi Security System for Spring includes a PasswordHandler that bridges your existing AuthenticationManager into CAS. You do not need to use this PasswordHandler to use Acegi Security on the client side (any CAS PasswordHandler will do).

To install, you will need to download and extract the CAS server archive. We used version 2.0.12. There will be a /web directory in the root of the deployment. Copy an applicationContext.xml containing your AuthenticationManager as well as the CasPasswordHandler into the /web/WEB-INF directory. A sample applicationContext.xml is included below:

<bean id="inMemoryDaoImpl" class="org.acegisecurity.userdetails.memory.InMemoryDaoImpl">
  <property name="userMap">
    <value>
      marissa=koala,ROLES_IGNORED_BY_CAS
      dianne=emu,ROLES_IGNORED_BY_CAS
      scott=wombat,ROLES_IGNORED_BY_CAS
      peter=opal,disabled,ROLES_IGNORED_BY_CAS
    </value>
  </property>
</bean>

<bean id="daoAuthenticationProvider" class="org.acegisecurity.providers.dao.DaoAuthenticationProvider">
  <property name="userDetailsService"><ref bean="inMemoryDaoImpl"/></property>
</bean>

<bean id="authenticationManager" class="org.acegisecurity.providers.ProviderManager">
  <property name="providers">
    <list>
      <ref bean="daoAuthenticationProvider"/>
    </list>
  </property>
</bean>

<bean id="casPasswordHandler" class="org.acegisecurity.adapters.cas.CasPasswordHandler">
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
</bean>

Note the granted authorities are ignored by CAS because it has no way of communicating the granted authorities to calling applications. CAS is only concerned with username and passwords (and the enabled/disabled status).

Next you will need to edit the existing /web/WEB-INF/web.xml file. Add (or edit in the case of the authHandler property) the following lines:

<context-param>
  <param-name>edu.yale.its.tp.cas.authHandler</param-name>
  <param-value>org.acegisecurity.adapters.cas.CasPasswordHandlerProxy</param-value>
</context-param>

<context-param>
  <param-name>contextConfigLocation</param-name>
  <param-value>/WEB-INF/applicationContext.xml</param-value>
</context-param>

<listener>
  <listener-class>org.springframework.web.context.ContextLoaderListener</listener-class>
</listener>

Copy the spring.jar and acegi-security.jar files into /web/WEB-INF/lib. Now use the ant dist task in the build.xml in the root of the directory structure. This will create /lib/cas.war, which is ready for deployment to your servlet container.

Note CAS heavily relies on HTTPS. You can't even test the system without a HTTPS certificate. Whilst you should refer to your web container's documentation on setting up HTTPS, if you need some additional help or a test certificate you might like to check the samples/contacts/etc/ssl directory.

The web application side of CAS is made easy due to the Acegi Security System for Spring. It is assumed you already know the basics of using the Acegi Security System for Spring, so these are not covered again below. Only the CAS-specific beans are mentioned.

You will need to add a ServiceProperties bean to your application context. This represents your service:

<bean id="serviceProperties" class="org.acegisecurity.ui.cas.ServiceProperties">
  <property name="service"><value>https://localhost:8443/contacts-cas/j_acegi_cas_security_check</value></property>
  <property name="sendRenew"><value>false</value></property>
</bean>

The service must equal a URL that will be monitored by the CasProcessingFilter. The sendRenew defaults to false, but should be set to true if your application is particularly sensitive. What this parameter does is tell the CAS login service that a single sign on login is unacceptable. Instead, the user will need to re-enter their username and password in order to gain access to the service.

The following beans should be configured to commence the CAS authentication process:

<bean id="casProcessingFilter" class="org.acegisecurity.ui.cas.CasProcessingFilter">
  <property name="authenticationManager"><ref bean="authenticationManager"/></property>
  <property name="authenticationFailureUrl"><value>/casfailed.jsp</value></property>
  <property name="defaultTargetUrl"><value>/</value></property>
  <property name="filterProcessesUrl"><value>/j_acegi_cas_security_check</value></property>
</bean>

<bean id="securityEnforcementFilter" class="org.acegisecurity.intercept.web.SecurityEnforcementFilter">
  <property name="filterSecurityInterceptor"><ref bean="filterInvocationInterceptor"/></property>
  <property name="authenticationEntryPoint"><ref bean="casProcessingFilterEntryPoint"/></property>
</bean>

<bean id="casProcessingFilterEntryPoint" class="org.acegisecurity.ui.cas.CasProcessingFilterEntryPoint">
  <property name="loginUrl"><value>https://localhost:8443/cas/login</value></property>
  <property name="serviceProperties"><ref bean="serviceProperties"/></property>
</bean>

You will also need to add the CasProcessingFilter to web.xml:

<filter>
  <filter-name>Acegi CAS Processing Filter</filter-name>
  <filter-class>org.acegisecurity.util.FilterToBeanProxy</filter-class>
  <init-param>
    <param-name>targetClass</param-name>
    <param-value>org.acegisecurity.ui.cas.CasProcessingFilter</param-value>
  </init-param>
</filter>

<filter-mapping>
  <filter-name>Acegi CAS Processing Filter</filter-name>
  <url-pattern>/*</url-pattern>
</filter-mapping>

The CasProcessingFilter has very similar properties to the AuthenticationProcessingFilter (used for form-based logins). Each property is self-explanatory.

For CAS to operate, the SecurityEnforcementFilter must have its authenticationEntryPoint property set to the CasProcessingFilterEntryPoint bean.

The CasProcessingFilterEntryPoint must refer to the ServiceProperties bean (discussed above), which provides the URL to the enterprise's CAS login server. This is where the user's browser will be redirected.

Next you need to add an AuthenticationManager that uses CasAuthenticationProvider and its collaborators:

<bean id="authenticationManager" class="org.acegisecurity.providers.ProviderManager">
  <property name="providers">
    <list>
      <ref bean="casAuthenticationProvider"/>
    </list>
  </property>
</bean>

<bean id="casAuthenticationProvider" class="org.acegisecurity.providers.cas.CasAuthenticationProvider">
  <property name="casAuthoritiesPopulator"><ref bean="casAuthoritiesPopulator"/></property>
  <property name="casProxyDecider"><ref bean="casProxyDecider"/></property>
  <property name="ticketValidator"><ref bean="casProxyTicketValidator"/></property>
  <property name="statelessTicketCache"><ref bean="statelessTicketCache"/></property>
  <property name="key"><value>my_password_for_this_auth_provider_only</value></property>
</bean>

<bean id="casProxyTicketValidator" class="org.acegisecurity.providers.cas.ticketvalidator.CasProxyTicketValidator">
  <property name="casValidate"><value>https://localhost:8443/cas/proxyValidate</value></property>
  <property name="proxyCallbackUrl"><value>https://localhost:8443/contacts-cas/casProxy/receptor</value></property>
  <property name="serviceProperties"><ref bean="serviceProperties"/></property>
  <!-- <property name="trustStore"><value>/some/path/to/your/lib/security/cacerts</value></property> -->
</bean>

<bean id="cacheManager" class="org.springframework.cache.ehcache.EhCacheManagerFactoryBean">
  <property name="configLocation">
    <value>classpath:/ehcache-failsafe.xml</value>
  </property>
</bean>
    
<bean id="ticketCacheBackend" class="org.springframework.cache.ehcache.EhCacheFactoryBean">
  <property name="cacheManager">
    <ref local="cacheManager"/>
  </property>
  <property name="cacheName">
    <value>ticketCache</value>
  </property>
</bean>
   
<bean id="statelessTicketCache" class="org.acegisecurity.providers.cas.cache.EhCacheBasedTicketCache">
  <property name="cache"><ref local="ticketCacheBackend"/></property>
</bean>

<bean id="casAuthoritiesPopulator" class="org.acegisecurity.providers.cas.populator.DaoCasAuthoritiesPopulator">
  <property name="userDetailsService"><ref bean="inMemoryDaoImpl"/></property>
</bean>

<bean id="casProxyDecider" class="org.acegisecurity.providers.cas.proxy.RejectProxyTickets"/>

The beans are all reasonable self-explanatory if you refer back to the "How CAS Works" section. Careful readers might notice one surprise: the statelessTicketCache property of the CasAuthenticationProvider. This is discussed in detail in the "Advanced CAS Usage" section.

Note the CasProxyTicketValidator has a remarked out trustStore property. This property might be helpful if you experience HTTPS certificate issues. Also note the proxyCallbackUrl is set so the service can receive a proxy-granting ticket. As mentioned above, this is optional and unnecessary if you do not require proxy-granting tickets. If you do use this feature, you will need to configure a suitable servlet to receive the proxy-granting tickets. We suggest you use CAS' ProxyTicketReceptor by adding the following to your web application's web.xml:

<servlet>
  <servlet-name>casproxy</servlet-name>
  <servlet-class>edu.yale.its.tp.cas.proxy.ProxyTicketReceptor</servlet-class>
</servlet>

<servlet-mapping>
  <servlet-name>casproxy</servlet-name>
  <url-pattern>/casProxy/*</url-pattern>
</servlet-mapping>

This completes the configuration of CAS. If you haven't made any mistakes, your web application should happily work within the framework of CAS single sign on. No other parts of the Acegi Security System for Spring need to be concerned about the fact CAS handled authentication.

There is also a contacts-cas.war file in the sample applications directory. This sample application uses the above settings and can be deployed to see CAS in operation.

The CasAuthenticationProvider distinguishes between stateful and stateless clients. A stateful client is considered any that originates via the CasProcessingFilter. A stateless client is any that presents an authentication request via the UsernamePasswordAuthenticationToken with a principal equal to CasProcessingFilter.CAS_STATELESS_IDENTIFIER.

Stateless clients are likely to be via remoting protocols such as Hessian and Burlap. The BasicProcessingFilter is still used in this case, but the remoting protocol client is expected to present a username equal to the static string above, and a password equal to a CAS service ticket. Clients should acquire a CAS service ticket directly from the CAS server.

Because remoting protocols have no way of presenting themselves within the context of a HttpSession, it isn't possible to rely on the HttpSession's HttpSessionIntegrationFilter.ACEGI_SECURITY_AUTHENTICATION_KEY attribute to locate the CasAuthenticationToken. Furthermore, because the CAS server invalidates a service ticket after it has been validated by the TicketValidator, presenting the same service ticket on subsequent requests will not work. It is similarly very difficult to obtain a proxy-granting ticket for a remoting protocol client, as they are often deployed on client machines which rarely have HTTPS URLs that would be accessible to the CAS server.

One obvious option is to not use CAS at all for remoting protocol clients. However, this would eliminate many of the desirable features of CAS.

As a middle-ground, the CasAuthenticationProvider uses a StatelessTicketCache. This is used solely for requests with a principal equal to CasProcessingFilter.CAS_STATELESS_IDENTIFIER. What happens is the CasAuthenticationProvider will store the resulting CasAuthenticationToken in the StatelessTicketCache, keyed on the service ticket. Accordingly, remoting protocol clients can present the same service ticket and the CasAuthenticationProvider will not need to contact the CAS server for validation (aside from the first request).

The other aspect of advanced CAS usage involves creating proxy tickets from the proxy-granting ticket. As indicated above, we recommend you use CAS' ProxyTicketReceptor to receive these tickets. The ProxyTicketReceptor provides a static method that enables you to obtain a proxy ticket by presenting the proxy-granting IOU ticket. You can obtain the proxy-granting IOU ticket by calling CasAuthenticationToken.getProxyGrantingTicketIou().

It is hoped you find CAS integration easy and useful with the Acegi Security System for Spring classes. Welcome to enterprise-wide single sign on!

The most common use of X509 certificate authentication is in verifying the identity of a server when using SSL, most commonly when using HTTPS from a browser. The browser will automatically check that the certificate presented by a server has been issued (ie digitally signed) by one of a list of trusted certificate authorities which it maintains.

You can also use SSL with “mutual authentication”; the server will then request a valid certificate from the client as part of the SSL handshake. The server will authenticate the client by checking that it's certificate is signed by an acceptable authority. If a valid certificate has been provided, it can be obtained through the servlet API in an application. The Acegi Security X509 module extracts the certificate using a filter and passes it to the configured X509 authentication provider to allow any additional application-specific checks to be applied. It also maps the certificate to an application user and loads that user's set of granted authorities for use with the standard Acegi Security infrastructure.

You should be familiar with using certificates and setting up client authentication for your servlet container before attempting to use it with Acegi Security. Most of the work is in creating and installing suitable certificates and keys. For example, if you're using Tomcat then read the instructions here http://jakarta.apache.org/tomcat/tomcat-5.0-doc/ssl-howto.html. It's important that you get this working before trying it out with Acegi Security.

With X509 authentication, there is no explicit login procedure so the implementation is relatively simple; there is no need to redirect requests in order to interact with the user. As a result, some of the classes behave slightly differently from their equivalents in other packages. For example, the default “entry point” class, which is normally responsible for starting the authentication process, is only invoked if the certificate is rejected and it always returns an error to the user. With a suitable bean configuration, the normal sequence of events is as follows

There is a version of the Contacts Sample Application which uses X509. Copy the beans and filter setup from this as a starting point for configuring your own application. A set of example certificates is also included which you can use to configure your server. These are

<!-- SSL/TLS Connector configuration -->
<Connector port="8443" address="${jboss.bind.address}"
  maxThreads="100" minSpareThreads="5" maxSpareThreads="15"
  scheme="https" secure="true"
  sslProtocol = "TLS"
  clientAuth="true" keystoreFile="${jboss.server.home.dir}/conf/server.p12"
  keystoreType="PKCS12" keystorePass="password"
  truststoreFile="${jboss.server.home.dir}/conf/ca.jks"
  truststoreType="JKS" truststorePass="password"
/>

In addition to coordinating the authentication and authorization requirements of your application, the Acegi Security System for Spring is also able to ensure unauthenticated web requests have certain properties. These properties may include being of a particular transport type, having a particular HttpSession attribute set and so on. The most common requirement is for your web requests to be received using a particular transport protocol, such as HTTPS.

An important issue in considering transport security is that of session hijacking. Your web container manages a HttpSession by reference to a jsessionid that is sent to user agents either via a cookie or URL rewriting. If the jsessionid is ever sent over HTTP, there is a possibility that session identifier can be intercepted and used to impersonate the user after they complete the authentication process. This is because most web containers maintain the same session identifier for a given user, even after they switch from HTTP to HTTPS pages.

If session hijacking is considered too significant a risk for your particular application, the only option is to use HTTPS for every request. This means the jsessionid is never sent across an insecure channel. You will need to ensure your web.xml-defined <welcome-file> points to a HTTPS location, and the application never directs the user to a HTTP location. The Acegi Security System for Spring provides a solution to assist with the latter.

To utilise Acegi Security's channel security services, add the following lines to web.xml:

<filter>
  <filter-name>Acegi Channel Processing Filter</filter-name>
  <filter-class>org.acegisecurity.util.FilterToBeanProxy</filter-class>
  <init-param>
    <param-name>targetClass</param-name>
    <param-value>org.acegisecurity.securechannel.ChannelProcessingFilter</param-value>
  </init-param>
</filter>

<filter-mapping>
  <filter-name>Acegi Channel Processing Filter</filter-name>
  <url-pattern>/*</url-pattern>
</filter-mapping>

As usual when running FilterToBeanProxy, you will also need to configure the filter in your application context:

<bean id="channelProcessingFilter" class="org.acegisecurity.securechannel.ChannelProcessingFilter">
  <property name="channelDecisionManager"><ref bean="channelDecisionManager"/></property>
  <property name="filterInvocationDefinitionSource">
    <value>
      CONVERT_URL_TO_LOWERCASE_BEFORE_COMPARISON
      \A/secure/.*\Z=REQUIRES_SECURE_CHANNEL
      \A/acegilogin.jsp.*\Z=REQUIRES_SECURE_CHANNEL
      \A/j_acegi_security_check.*\Z=REQUIRES_SECURE_CHANNEL	
      \A.*\Z=REQUIRES_INSECURE_CHANNEL
    </value>
  </property>
</bean>

<bean id="channelDecisionManager" class="org.acegisecurity.securechannel.ChannelDecisionManagerImpl">
  <property name="channelProcessors">
    <list>
      <ref bean="secureChannelProcessor"/>
      <ref bean="insecureChannelProcessor"/>
    </list>
  </property>
</bean>

<bean id="secureChannelProcessor" class="org.acegisecurity.securechannel.SecureChannelProcessor"/>
<bean id="insecureChannelProcessor" class="org.acegisecurity.securechannel.InsecureChannelProcessor"/>

Like FilterSecurityInterceptor, Apache Ant style paths are also supported by the ChannelProcessingFilter.

The ChannelProcessingFilter operates by filtering all web requests and determining the configuration attributes that apply. It then delegates to the ChannelDecisionManager. The default implementation, ChannelDecisionManagerImpl, should suffice in most cases. It simply delegates through the list of configured ChannelProcessor instances. A ChannelProcessor will review the request, and if it is unhappy with the request (eg it was received across the incorrect transport protocol), it will perform a redirect, throw an exception or take whatever other action is appropriate.

Included with the Acegi Security System for Spring are two concrete ChannelProcessor implementations: SecureChannelProcessor ensures requests with a configuration attribute of REQUIRES_SECURE_CHANNEL are received over HTTPS, whilst InsecureChannelProcessor ensures requests with a configuration attribute of REQUIRES_INSECURE_CHANNEL are received over HTTP. Both implementations delegate to a ChannelEntryPoint if the required transport protocol is not used. The two ChannelEntryPoint implementations included with Acegi Security simply redirect the request to HTTP and HTTPS as appropriate. Appropriate defaults are assigned to the ChannelProcessor implementations for the configuration attribute keywords they respond to and the ChannelEntryPoint they delegate to, although you have the ability to override these using the application context.

Note that the redirections are absolute (eg http://www.company.com:8080/app/page), not relative (eg /app/page). During testing it was discovered that Internet Explorer 6 Service Pack 1 has a bug whereby it does not respond correctly to a redirection instruction which also changes the port to use. Accordingly, absolute URLs are used in conjunction with bug detection logic in the PortResolverImpl that is wired up by default to many Acegi Security beans. Please refer to the JavaDocs for PortResolverImpl for further details.

Once configured, using the channel security filter is very easy. Simply request pages without regard to the protocol (ie HTTP or HTTPS) or port (eg 80, 8080, 443, 8443 etc). Obviously you'll still need a way of making the initial request (probably via the web.xml <welcome-file> or a well-known home page URL), but once this is done the filter will perform redirects as defined by your application context.

You can also add your own ChannelProcessor implementations to the ChannelDecisionManagerImpl. For example, you might set a HttpSession attribute when a human user is detected via a "enter the contents of this graphic" procedure. Your ChannelProcessor would respond to say REQUIRES_HUMAN_USER configuration attributes and redirect to an appropriate entry point to start the human user validation process if the HttpSession attribute is not currently set.

To decide whether a security check belongs in a ChannelProcessor or an AccessDecisionVoter, remember that the former is designed to handle unauthenticated requests, whilst the latter is designed to handle authenticated requests. The latter therefore has access to the granted authorities of the authenticated principal. In addition, problems detected by a ChannelProcessor will generally cause a HTTP/HTTPS redirection so its requirements can be met, whilst problems detected by an AccessDecisionVoter will ultimately result in an AccessDeniedException (depending on the governing AccessDecisionManager).

Complex applications often will find the need to define access permissions not simply at a web request or method invocation level. Instead, security decisions need to comprise both who (Authentication), where (MethodInvocation) and what (SomeDomainObject). In other words, authorization decisions also need to consider the actual domain object instance subject of a method invocation.

Imagine you're designing an application for a pet clinic. There will be two main groups of users of your Spring-based application: staff of the pet clinic, as well as the pet clinic's customers. The staff will have access to all of the data, whilst your customers will only be able to see their own customer records. To make it a little more interesting, your customers can allow other users to see their customer records, such as their "puppy preschool "mentor or president of their local "Pony Club". Using Acegi Security System for Spring as the foundation, you have several approaches that can be used:

Each one of these approaches is perfectly legitimate. However, the first couples your authorization checking to your business code. The main problems with this include the enhanced difficulty of unit testing and the fact it would be more difficult to reuse the Customer authorization logic elsewhere. Obtaining the GrantedAuthority[]s from the Authentication object is also fine, but will not scale to large numbers of Customers. If a user might be able to access 5,000 Customers (unlikely in this case, but imagine if it were a popular vet for a large Pony Club!) the amount of memory consumed and time required to construct the Authentication object would be undesirable. The final method, opening the Customer directly from external code, is probably the best of the three. It achieves separation of concerns, and doesn't misuse memory or CPU cycles, but it is still inefficient in that both the AccessDecisionVoter and the eventual business method itself will perform a call to the DAO responsible for retrieving the Customer object. Two accesses per method invocation is clearly undesirable. In addition, with every approach listed you'll need to write your own access control list (ACL) persistence and business logic from scratch.

Fortunately, there is another alternative, which we'll talk about below.

The org.acegisecurity.acl package is very simple, comprising only a handful of interfaces and a single class, as shown in Figure 6. It provides the basic foundation for access control list (ACL) lookups.

Figure 6: Access Control List Manager

The central interface is AclManager, which is defined by two methods:

public AclEntry[] getAcls(java.lang.Object domainInstance);
public AclEntry[] getAcls(java.lang.Object domainInstance, Authentication authentication);

AclManager is intended to be used as a collaborator against your business objects, or, more desirably, AccessDecisionVoters. This means you use Spring's normal ApplicationContext features to wire up your AccessDecisionVoter (or business method) with an AclManager. Consideration was given to placing the ACL information in the ContextHolder, but it was felt this would be inefficient both in terms of memory usage as well as the time spent loading potentially unused ACL information. The trade-off of needing to wire up a collaborator for those objects requiring ACL information is rather minor, particularly in a Spring-managed application.

The first method of the AclManager will return all ACLs applying to the domain object instance passed to it. The second method does the same, but only returns those ACLs which apply to the passed Authentication object.

The AclEntry interface returned by AclManager is merely a marker interface. You will need to provide an implementation that reflects that ACL permissions for your application.

Rounding out the org.acegisecurity.acl package is an AclProviderManager class, with a corresponding AclProvider interface. AclProviderManager is a concrete implementation of AclManager, which iterates through registered AclProviders. The first AclProvider that indicates it can authoritatively provide ACL information for the presented domain object instance will be used. This is very similar to the AuthenticationProvider interface used for authentication.

With this background, let's now look at a usable ACL implementation.

Acegi Security System for Spring includes a production-quality ACL provider implementation, which is shown in Figure 7.

Figure 7: Basic ACL Manager

The implementation is based on integer masking, which is commonly used for ACL permissions given its flexibility and speed. Anyone who has used Unix's chmod command will know all about this type of permission masking (eg chmod 777). You'll find the classes and interfaces for the integer masking ACL package under org.acegisecurity.acl.basic.

Extending the AclEntry interface is a BasicAclEntry interface, with the main methods shown below:

public AclObjectIdentity getAclObjectIdentity();
public AclObjectIdentity getAclObjectParentIdentity();
public int getMask();
public java.lang.Object getRecipient();

As shown, each BasicAclEntry has four main properties. The mask is the integer that represents the permissions granted to the recipient. The aclObjectIdentity is able to identify the domain object instance for which the ACL applies, and the aclObjectParentIdentity optionally specifies the parent of the domain object instance. Multiple BasicAclEntrys usually exist against a single domain object instance, and as suggested by the parent identity property, permissions granted higher in the object hierarchy will trickle down and be inherited (unless blocked by integer zero).

BasicAclEntry implementations typically provide convenience methods, such as isReadAllowed(), to avoid application classes needing to perform bit masking themselves. The SimpleAclEntry and AbstractBasicAclEntry demonstrate and provide much of this bit masking logic.

The AclObjectIdentity itself is merely a marker interface, so you need to provide implementations for your domain objects. However, the package does include a NamedEntityObjectIdentity implementation which will suit many needs. The NamedEntityObjectIdentity identifies a given domain object instance by the classname of the instance and the identity of the instance. A NamedEntityObjectIdentity can be constructed manually (by calling the constructor and providing the classname and identity Strings), or by passing in any domain object that contains a getId() method.

The actual AclProvider implementation is named BasicAclProvider. It has adopted a similar design to that used by the authentication-related DaoAuthenticationProvder. Specifically, you define a BasicAclDao against the provider, so different ACL repository types can be accessed in a pluggable manner. The BasicAclProvider also supports pluggable cache providers (with Acegi Security System for Spring including an implementation that fronts EH-CACHE).

The BasicAclDao interface is very simple to implement:

public BasicAclEntry[] getAcls(AclObjectIdentity aclObjectIdentity);

A BasicAclDao implementation needs to understand the presented AclObjectIdentity and how it maps to a storage repository, find the relevant records, and create appropriate BasicAclEntry objects and return them.

Acegi Security includes a single BasicAclDao implementation called JdbcDaoImpl. As implied by the name, JdbcDaoImpl accesses ACL information from a JDBC database. There is also an extended version of this DAO, JdbcExtendedDaoImpl, which provides CRUD operations on the JDBC database, although we won't discuss these features here. The default database schema and some sample data will aid in understanding its function:

CREATE TABLE acl_object_identity (
     id IDENTITY NOT NULL,
     object_identity VARCHAR_IGNORECASE(250) NOT NULL,
     parent_object INTEGER,
     acl_class VARCHAR_IGNORECASE(250) NOT NULL,
     CONSTRAINT unique_object_identity UNIQUE(object_identity),
     FOREIGN KEY (parent_object) REFERENCES acl_object_identity(id)
);

CREATE TABLE acl_permission (
     id IDENTITY NOT NULL,
     acl_object_identity INTEGER NOT NULL,
     recipient VARCHAR_IGNORECASE(100) NOT NULL,
     mask INTEGER NOT NULL,
     CONSTRAINT unique_recipient UNIQUE(acl_object_identity, recipient),
     FOREIGN KEY (acl_object_identity) REFERENCES acl_object_identity(id)
);

INSERT INTO acl_object_identity VALUES (1, 'corp.DomainObject:1', null, 'org.acegisecurity.acl.basic.SimpleAclEntry');
INSERT INTO acl_object_identity VALUES (2, 'corp.DomainObject:2', 1, 'org.acegisecurity.acl.basic.SimpleAclEntry');
INSERT INTO acl_object_identity VALUES (3, 'corp.DomainObject:3', 1, 'org.acegisecurity.acl.basic.SimpleAclEntry');
INSERT INTO acl_object_identity VALUES (4, 'corp.DomainObject:4', 1, 'org.acegisecurity.acl.basic.SimpleAclEntry');
INSERT INTO acl_object_identity VALUES (5, 'corp.DomainObject:5', 3, 'org.acegisecurity.acl.basic.SimpleAclEntry');
INSERT INTO acl_object_identity VALUES (6, 'corp.DomainObject:6', 3, 'org.acegisecurity.acl.basic.SimpleAclEntry');

INSERT INTO acl_permission VALUES (null, 1, 'ROLE_SUPERVISOR', 1);
INSERT INTO acl_permission VALUES (null, 2, 'ROLE_SUPERVISOR', 0);
INSERT INTO acl_permission VALUES (null, 2, 'marissa', 2);
INSERT INTO acl_permission VALUES (null, 3, 'scott', 14);
INSERT INTO acl_permission VALUES (null, 6, 'scott', 1);

As can be seen, database-specific constraints are used extensively to ensure the integrity of the ACL information. If you need to use a different database (Hypersonic SQL statements are shown above), you should try to implement equivalent constraints.

The JdbcDaoImpl will only respond to requests for NamedEntityObjectIdentitys. It converts such identities into a single String, comprising the NamedEntityObjectIdentity.getClassname() + ":" + NamedEntityObjectIdentity.getId(). This yields the type of object_identity values shown above. As indicated by the sample data, each database row corresponds to a single BasicAclEntry. As stated earlier and demonstrated by corp.DomainObject:2 in the above sample data, each domain object instance will often have multiple BasicAclEntry[]s.

As JdbcDaoImpl is required to return concrete BasicAclEntry classes, it needs to know which BasicAclEntry implementation it is to create and populate. This is the role of the acl_class column. JdbcDaoImpl will create the indicated class and set its mask, recipient, aclObjectIdentity and aclObjectParentIdentity properties.

As you can probably tell from the sample data, the parent_object_identity value can either be null or in the same format as the object_identity. If non-null, JdbcDaoImpl will create a NamedEntityObjectIdentity to place inside the returned BasicAclEntry class.

Returning to the BasicAclProvider, before it can poll the BasicAclDao implementation it needs to convert the domain object instance it was passed into an AclObjectIdentity. BasicAclProvider has a protected AclObjectIdentity obtainIdentity(Object domainInstance) method that is responsible for this. As a protected method, it enables subclasses to easily override. The normal implementation checks whether the passed domain object instance implements the AclObjectIdentityAware interface, which is merely a getter for an AclObjectIdentity. If the domain object does implement this interface, that is the identity returned. If the domain object does not implement this interface, the method will attempt to create an AclObjectIdentity by passing the domain object instance to the constructor of a class defined by the BasicAclProvider.getDefaultAclObjectIdentity() method. By default the defined class is NamedEntityObjectIdentity, which was described in more detail above. Therefore, you will need to either (i) provide a getId() method on your domain objects, (ii) implement AclObjectIdentityAware on your domain objects, (iii) provide an alternative AclObjectIdentity implementation that will accept your domain object in its constructor, or (iv) override the obtainIdentity(Object) method.

Once the AclObjectIdentity of the domain object instance is determined, the BasicAclProvider will poll the DAO to obtain its BasicAclEntry[]s. If any of the entries returned by the DAO indicate there is a parent, that parent will be polled, and the process will repeat until there is no further parent. The permissions assigned to a recipient closest to the domain object instance will always take priority and override any inherited permissions. From the sample data above, the following inherited permissions would apply:

--- Mask integer 0  = no permissions
--- Mask integer 1  = administer
--- Mask integer 2  = read
--- Mask integer 6  = read and write permissions
--- Mask integer 14 = read and write and create permissions

---------------------------------------------------------------------
--- *** INHERITED RIGHTS FOR DIFFERENT INSTANCES AND RECIPIENTS ***
--- INSTANCE  RECIPIENT         PERMISSION(S) (COMMENT #INSTANCE)
---------------------------------------------------------------------
---    1      ROLE_SUPERVISOR   Administer
---    2      ROLE_SUPERVISOR   None (overrides parent #1)
---           marissa           Read
---    3      ROLE_SUPERVISOR   Administer (from parent #1)
---           scott             Read, Write, Create
---    4      ROLE_SUPERVISOR   Administer (from parent #1)
---    5      ROLE_SUPERVISOR   Administer (from parent #3)
---           scott             Read, Write, Create (from parent #3)
---    6      ROLE_SUPERVISOR   Administer (from parent #3)
---           scott             Administer (overrides parent #3)

So the above explains how a domain object instance has its AclObjectIdentity discovered, and the BasicAclDao will be polled successively until an array of inherited permissions is constructed for the domain object instance. The final step is to determine the BasicAclEntry[]s that are actually applicable to a given Authentication object.

As you would recall, the AclManager (and all delegates, up to and including BasicAclProvider) provides a method which returns only those BasicAclEntry[]s applying to a passed Authentication object. BasicAclProvider delivers this functionality by delegating the filtering operation to an EffectiveAclsResolver implementation. The default implementation, GrantedAuthorityEffectiveAclsResolver, will iterate through the BasicAclEntry[]s and include only those where the recipient is equal to either the Authentication's principal or any of the Authentication's GrantedAuthority[]s. Please refer to the JavaDocs for more information.

Figure 8: ACL Instantiation Approach

Acegi Security's instance-specific ACL packages shield you from much of the complexity of developing your own ACL approach. The interfaces and classes detailed above provide a scalable, customisable ACL solution that is decoupled from your application code. Whilst the reference documentation may suggest complexity, the basic implementation is able to support most typical applications out-of-the-box.

The Acegi Security System for Spring uses filters extensively. Each filter is covered in detail in a respective section of this document. This section includes information that applies to all filters.

Most filters are configured using the FilterToBeanProxy. An example configuration from web.xml follows:

<filter>
  <filter-name>Acegi HTTP Request Security Filter</filter-name>
  <filter-class>org.acegisecurity.util.FilterToBeanProxy</filter-class>
  <init-param>
    <param-name>targetClass</param-name>
    <param-value>org.acegisecurity.ClassThatImplementsFilter</param-value>
  </init-param>
</filter>

Notice that the filter in web.xml is actually a FilterToBeanProxy, and not the filter that will actually implements the logic of the filter. What FilterToBeanProxy does is delegate the Filter's methods through to a bean which is obtained from the Spring application context. This enables the bean to benefit from the Spring application context lifecycle support and configuration flexibility. The bean must implement javax.servlet.Filter.

The FilterToBeanProxy only requires a single initialization parameter, targetClass or targetBean. The targetClass parameter locates the first object in the application context of the specified class, whilst targetBean locates the object by bean name. Like standard Spring web applications, the FilterToBeanProxy accesses the application context via WebApplicationContextUtils.getWebApplicationContext(ServletContext), so you should configure a ContextLoaderListener in web.xml.

There is a lifecycle issue to consider when hosting Filters in an IoC container instead of a servlet container. Specifically, which container should be responsible for calling the Filter's "startup" and "shutdown" methods? It is noted that the order of initialization and destruction of a Filter can vary by servlet container, and this can cause problems if one Filter depends on configuration settings established by an earlier initialized Filter. The Spring IoC container on the other hand has more comprehensive lifecycle/IoC interfaces (such as InitializingBean, DisposableBean, BeanNameAware, ApplicationContextAware and many others) as well as a well-understood interface contract, predictable method invocation ordering, autowiring support, and even options to avoid implementing Spring interfaces (eg the destroy-method attribute in Spring XML). For this reason we recommend the use of Spring lifecycle services instead of servlet container lifecycle services wherever possible. By default FilterToBeanProxy will not delegate init(FilterConfig) and destroy() methods through to the proxied bean. If you do require such invocations to be delegated, set the lifecycle initialization parameter to servlet-container-managed.

We strongly recommend to use FilterChainProxy instead of adding multiple filters to web.xml.

Whilst FilterToBeanProxy is a very useful class, the problem is that the lines of code required for <filter> and <filter-mapping> entries in web.xml explodes when using more than a few filters. To overcome this issue, Acegi Security provides a FilterChainProxy class. It is wired using a FilterToBeanProxy (just like in the example above), but the target class is org.acegisecurity.util.FilterChainProxy. The filter chain is then declared in the application context, using code such as this:

<bean id="filterChainProxy" class="org.acegisecurity.util.FilterChainProxy">
  <property name="filterInvocationDefinitionSource">
    <value>
      CONVERT_URL_TO_LOWERCASE_BEFORE_COMPARISON
      PATTERN_TYPE_APACHE_ANT
      /webServices/**=httpSessionContextIntegrationFilterWithASCFalse,basicProcessingFilter,securityEnforcementFilter
      /**=httpSessionContextIntegrationFilterWithASCTrue,authenticationProcessingFilter,securityEnforcementFilter
    </value>
  </property>
</bean>

You may notice similarities with the way SecurityEnforcementFilter is declared. Both regular expressions and Ant Paths are supported, and the most specific URIs appear first. At runtime the FilterChainProxy will locate the first URI pattern that matches the current web request. Each of the corresponding configuration attributes represent the name of a bean defined in the application context. The filters will then be invoked in the order they are specified, with standard FilterChain behaviour being respected (a Filter can elect not to proceed with the chain if it wishes to end processing).

As you can see, FitlerChainProxy requires the duplication of filter names for different request patterns (in the above example, httpSessionContextIntegrationFilter and securityEnforcementFilter are duplicated). This design decision was made to enable FilterChainProxy to specify different Filter invocation orders for different URI patterns, and also to improve both the expressiveness (in terms of regular expressions, Ant Paths, and any custom FilterInvocationDefinitionSource implementations) and clarity of which Filters should be invoked.

You may have noticed we have declared two HttpSessionContextIntegrationFilters in the filter chain (ASC is short for allowSessionCreation, a property of HttpSessionContextIntegrationFilter). As web services will never present a jsessionid on future requests, creating HttpSessions for such user agents would be wasteful. If you had a high-volume application which required maximum scalability, we recommend you use the approach shown above. For smaller applications, using a single HttpSessionContextIntegrationFilter (with its default allowSessionCreation as true) would likely be sufficient.

In relation to lifecycle issues, the FilterChainProxy will always delegate init(FilterConfig) and destroy() methods through to the underlaying Filters if such methods are called against FilterChainProxy itself. In this case, FilterChainProxy guarantees to only initialize and destroy each Filter once, irrespective of how many times it is declared by the FilterInvocationDefinitionSource. You control the overall choice as to whether these methods are called or not via the lifecycle initialization parameter of the FilterToBeanProxy that proxies FilterChainProxy. As discussed above, by default any servlet container lifecycle invocations are not delegated through to FilterChainProxy.

The order that filters are defined in web.xml is important. NB: THE FILTER ORDER CHANGED FROM VERSION 0.8.0.

Irrespective of which filters you are actually using, the order of the <filter-mapping>s should be as follows:

All of the above filters use FilterToBeanProxy or FilterChainProxy, which is discussed in the previous sections. It is recommended that a single FilterToBeanProxy proxy through to a single FilterChainProxy for each application, with that FilterChainProxy defining all of the Acegi Security Filters.

If you're using SiteMesh, ensure the Acegi Security filters execute before the SiteMesh filters are called. This enables the SecurityContextHolder to be populated in time for use by SiteMesh decorators.