13. The Security Filter Chain

Spring Security’s web infrastructure is based entirely on standard servlet filters. It doesn’t use servlets or any other servlet-based frameworks (such as Spring MVC) internally, so it has no strong links to any particular web technology. It deals in HttpServletRequest s and HttpServletResponse s and doesn’t care whether the requests come from a browser, a web service client, an HttpInvoker or an AJAX application.

Spring Security maintains a filter chain internally where each of the filters has a particular responsibility and filters are added or removed from the configuration depending on which services are required. The ordering of the filters is important as there are dependencies between them. If you have been using namespace configuration, then the filters are automatically configured for you and you don’t have to define any Spring beans explicitly but here may be times when you want full control over the security filter chain, either because you are using features which aren’t supported in the namespace, or you are using your own customized versions of classes.

13.1 DelegatingFilterProxy

When using servlet filters, you obviously need to declare them in your web.xml, or they will be ignored by the servlet container. In Spring Security, the filter classes are also Spring beans defined in the application context and thus able to take advantage of Spring’s rich dependency-injection facilities and lifecycle interfaces. Spring’s DelegatingFilterProxy provides the link between web.xml and the application context.

When using DelegatingFilterProxy, you will see something like this in the web.xml file:

<filter>
<filter-name>myFilter</filter-name>
<filter-class>org.springframework.web.filter.DelegatingFilterProxy</filter-class>
</filter>

<filter-mapping>
<filter-name>myFilter</filter-name>
<url-pattern>/*</url-pattern>
</filter-mapping>

Notice that the filter is actually a DelegatingFilterProxy, and not the class that will actually implement the logic of the filter. What DelegatingFilterProxy 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 web application context lifecycle support and configuration flexibility. The bean must implement javax.servlet.Filter and it must have the same name as that in the filter-name element. Read the Javadoc for DelegatingFilterProxy for more information

13.2 FilterChainProxy

Spring Security’s web infrastructure should only be used by delegating to an instance of FilterChainProxy. The security filters should not be used by themselves. In theory you could declare each Spring Security filter bean that you require in your application context file and add a corresponding DelegatingFilterProxy entry to web.xml for each filter, making sure that they are ordered correctly, but this would be cumbersome and would clutter up the web.xml file quickly if you have a lot of filters. FilterChainProxy lets us add a single entry to web.xml and deal entirely with the application context file for managing our web security beans. It is wired using a DelegatingFilterProxy, just like in the example above, but with the filter-name set to the bean name "filterChainProxy". The filter chain is then declared in the application context with the same bean name. Here’s an example:

<bean id="filterChainProxy" class="org.springframework.security.web.FilterChainProxy">
<constructor-arg>
	<list>
	<sec:filter-chain pattern="/restful/**" filters="
		securityContextPersistenceFilterWithASCFalse,
		basicAuthenticationFilter,
		exceptionTranslationFilter,
		filterSecurityInterceptor" />
	<sec:filter-chain pattern="/**" filters="
		securityContextPersistenceFilterWithASCTrue,
		formLoginFilter,
		exceptionTranslationFilter,
		filterSecurityInterceptor" />
	</list>
</constructor-arg>
</bean>

The namespace element filter-chain is used for convenience to set up the security filter chain(s) which are required within the application. [6]. It maps a particular URL pattern to a list of filters built up from the bean names specified in the filters element, and combines them in a bean of type SecurityFilterChain. The pattern attribute takes an Ant Paths and the most specific URIs should appear first [7]. At runtime the FilterChainProxy will locate the first URI pattern that matches the current web request and the list of filter beans specified by the filters attribute will be applied to that request. The filters will be invoked in the order they are defined, so you have complete control over the filter chain which is applied to a particular URL.

You may have noticed we have declared two SecurityContextPersistenceFilter s in the filter chain (ASC is short for allowSessionCreation, a property of SecurityContextPersistenceFilter). As web services will never present a jsessionid on future requests, creating HttpSession s 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 SecurityContextPersistenceFilter (with its default allowSessionCreation as true) would likely be sufficient.

Note that FilterChainProxy does not invoke standard filter lifecycle methods on the filters it is configured with. We recommend you use Spring’s application context lifecycle interfaces as an alternative, just as you would for any other Spring bean.

When we looked at how to set up web security using namespace configuration, we used a DelegatingFilterProxy with the name "springSecurityFilterChain". You should now be able to see that this is the name of the FilterChainProxy which is created by the namespace.

13.2.1 Bypassing the Filter Chain

You can use the attribute filters = "none" as an alternative to supplying a filter bean list. This will omit the request pattern from the security filter chain entirely. Note that anything matching this path will then have no authentication or authorization services applied and will be freely accessible. If you want to make use of the contents of the SecurityContext contents during a request, then it must have passed through the security filter chain. Otherwise the SecurityContextHolder will not have been populated and the contents will be null.

13.3 Filter Ordering

The order that filters are defined in the chain is very important. Irrespective of which filters you are actually using, the order should be as follows:

  • ChannelProcessingFilter, because it might need to redirect to a different protocol
  • SecurityContextPersistenceFilter, so a SecurityContext can be set up in the SecurityContextHolder at the beginning of a web request, and any changes to the SecurityContext can be copied to the HttpSession when the web request ends (ready for use with the next web request)
  • ConcurrentSessionFilter, because it uses the SecurityContextHolder functionality and needs to update the SessionRegistry to reflect ongoing requests from the principal
  • Authentication processing mechanisms - UsernamePasswordAuthenticationFilter, CasAuthenticationFilter, BasicAuthenticationFilter etc - so that the SecurityContextHolder can be modified to contain a valid Authentication request token
  • The SecurityContextHolderAwareRequestFilter, if you are using it to install a Spring Security aware HttpServletRequestWrapper into your servlet container
  • The JaasApiIntegrationFilter, if a JaasAuthenticationToken is in the SecurityContextHolder this will process the FilterChain as the Subject in the JaasAuthenticationToken
  • RememberMeAuthenticationFilter, so that if no earlier authentication processing mechanism updated the SecurityContextHolder, and the request presents a cookie that enables remember-me services to take place, a suitable remembered Authentication object will be put there
  • AnonymousAuthenticationFilter, so that if no earlier authentication processing mechanism updated the SecurityContextHolder, an anonymous Authentication object will be put there
  • ExceptionTranslationFilter, to catch any Spring Security exceptions so that either an HTTP error response can be returned or an appropriate AuthenticationEntryPoint can be launched
  • FilterSecurityInterceptor, to protect web URIs and raise exceptions when access is denied

13.4 Request Matching and HttpFirewall

Spring Security has several areas where patterns you have defined are tested against incoming requests in order to decide how the request should be handled. This occurs when the FilterChainProxy decides which filter chain a request should be passed through and also when the FilterSecurityInterceptor decides which security constraints apply to a request. It’s important to understand what the mechanism is and what URL value is used when testing against the patterns that you define.

The Servlet Specification defines several properties for the HttpServletRequest which are accessible via getter methods, and which we might want to match against. These are the contextPath, servletPath, pathInfo and queryString. Spring Security is only interested in securing paths within the application, so the contextPath is ignored. Unfortunately, the servlet spec does not define exactly what the values of servletPath and pathInfo will contain for a particular request URI. For example, each path segment of a URL may contain parameters, as defined in RFC 2396 [8]. The Specification does not clearly state whether these should be included in the servletPath and pathInfo values and the behaviour varies between different servlet containers. There is a danger that when an application is deployed in a container which does not strip path parameters from these values, an attacker could add them to the requested URL in order to cause a pattern match to succeed or fail unexpectedly. [9]. Other variations in the incoming URL are also possible. For example, it could contain path-traversal sequences (like /../) or multiple forward slashes (//) which could also cause pattern-matches to fail. Some containers normalize these out before performing the servlet mapping, but others don’t. To protect against issues like these, FilterChainProxy uses an HttpFirewall strategy to check and wrap the request. Un-normalized requests are automatically rejected by default, and path parameters and duplicate slashes are removed for matching purposes. [10]. It is therefore essential that a FilterChainProxy is used to manage the security filter chain. Note that the servletPath and pathInfo values are decoded by the container, so your application should not have any valid paths which contain semi-colons, as these parts will be removed for matching purposes.

As mentioned above, the default strategy is to use Ant-style paths for matching and this is likely to be the best choice for most users. The strategy is implemented in the class AntPathRequestMatcher which uses Spring’s AntPathMatcher to perform a case-insensitive match of the pattern against the concatenated servletPath and pathInfo, ignoring the queryString.

If for some reason, you need a more powerful matching strategy, you can use regular expressions. The strategy implementation is then RegexRequestMatcher. See the Javadoc for this class for more information.

In practice we recommend that you use method security at your service layer, to control access to your application, and do not rely entirely on the use of security constraints defined at the web-application level. URLs change and it is difficult to take account of all the possible URLs that an application might support and how requests might be manipulated. You should try and restrict yourself to using a few simple ant paths which are simple to understand. Always try to use a "deny-by-default" approach where you have a catch-all wildcard ( / or ) defined last and denying access.

Security defined at the service layer is much more robust and harder to bypass, so you should always take advantage of Spring Security’s method security options.

The HttpFirewall also prevents HTTP Response Splitting by rejecting new line characters in the HTTP Response headers.

By default the StrictHttpFirewall is used. This implementation rejects requests that appear to be malicious. If it is too strict for your needs, then you can customize what types of requests are rejected. However, it is important that you do so knowing that this can open your application up to attacks. For example, if you wish to leverage Spring MVC’s Matrix Variables, the following configuration could be used in XML:

<b:bean id="httpFirewall"
      class="org.springframework.security.web.firewall.StrictHttpFirewall"
      p:allowSemicolon="true"/>

<http-firewall ref="httpFirewall"/>

The same thing can be achieved with Java Configuration by exposing a StrictHttpFirewall bean.

@Bean
public StrictHttpFirewall httpFirewall() {
    StrictHttpFirewall firewall = new StrictHttpFirewall();
    firewall.setAllowSemicolon(true);
    return firewall;
}

13.5 Use with other Filter-Based Frameworks

If you’re using some other framework that is also filter-based, then you need to make sure that the Spring Security filters come first. This enables the SecurityContextHolder to be populated in time for use by the other filters. Examples are the use of SiteMesh to decorate your web pages or a web framework like Wicket which uses a filter to handle its requests.

13.6 Advanced Namespace Configuration

As we saw earlier in the namespace chapter, it’s possible to use multiple http elements to define different security configurations for different URL patterns. Each element creates a filter chain within the internal FilterChainProxy and the URL pattern that should be mapped to it. The elements will be added in the order they are declared, so the most specific patterns must again be declared first. Here’s another example, for a similar situation to that above, where the application supports both a stateless RESTful API and also a normal web application which users log into using a form.

<!-- Stateless RESTful service using Basic authentication -->
<http pattern="/restful/**" create-session="stateless">
<intercept-url pattern='/**' access="hasRole('REMOTE')" />
<http-basic />
</http>

<!-- Empty filter chain for the login page -->
<http pattern="/login.htm*" security="none"/>

<!-- Additional filter chain for normal users, matching all other requests -->
<http>
<intercept-url pattern='/**' access="hasRole('USER')" />
<form-login login-page='/login.htm' default-target-url="/home.htm"/>
<logout />
</http>


[6] Note that you’ll need to include the security namespace in your application context XML file in order to use this syntax. The older syntax which used a filter-chain-map is still supported, but is deprecated in favour of the constructor argument injection.

[7] Instead of a path pattern, the request-matcher-ref attribute can be used to specify a RequestMatcher instance for more powerful matching

[8] You have probably seen this when a browser doesn’t support cookies and the jsessionid parameter is appended to the URL after a semi-colon. However the RFC allows the presence of these parameters in any path segment of the URL

[9] The original values will be returned once the request leaves the FilterChainProxy, so will still be available to the application.

[10] So, for example, an original request path /secure;hack=1/somefile.html;hack=2 will be returned as /secure/somefile.html.