Filters

Java filters were introduced in the Java Servlet specification version 2.3. A filter intercepts requests and responses to transform or use the information. Filters typically do not themselves create responses but instead provide universal functions that can be "attached" to any type of servlet page.

Since the filter chain is responsible for request handling in Magnolia, the default chain illustrates how filters are used to process requests. This document provides minimal information on filters. For more, see Request processing and filters and the filters package.

Context

The first filter in the filter chain is 

. This filter initializes MgnlContext and configures MDC logging. MgnlContext is local to the request and available on every further filter. The context provides a lot of useful functions, see  for details.

Content type

The job of the ContentTypeFilter is to initialize the 

.

It's created and populated with:

• Character encoding
• Original URI
• Original URL
• Extension
• Query string

The AggregationState is accessible using MgnlContext.getAggregationState().

The Content-Type header is not set by ContentTypeFilter anymore. The MIME type was incorrectly set according to the request extension. It is now the responsibility of renderers/servlets to set the correct content type. For instance 

 sets the content type.  

Content type filter can be configured to match requests to content types. See Restricting responses to configured MIME types.

Cookies detector

 is a personalization filter that detects request cookies and adds them as traits to the aggregation state.

Date

 is a simple date  trait  filter for  personalization  that stores the current system time in the  TraitCollector.

Country

 is a GeoIP trait filter that detects the user's country using the IP address of the request for personalization. It adds the country to the aggregation state as a trait. If a GET parameter Country#REQUEST_PARAMETER with an IP address is supplied, this address is used to resolve the country, which is stored in the TraitCollector.

Visitor

 is a visitor trait filter that detects the type of visitor based on the current user and cookies for personalization. visitorCookies can be configured for returning and registered users.

Additional Properties

(Applicable to the cookie name nodes, i.e. to "returning", "registered" and "new" in the above examples.)

Device detection

 detects the type of device requesting content. Device detection is done using the Mobile Detect API.

Multipart request

 checks to see if there are any binary uploads such as form attachments in the request, extracts the binaries and persists them in the JCR.

Unicode Normalization

 normalizes the current URI to the NFC form that is used internally.

Registration

 checks the validity of the registration of a DX Core installation and delegates to the registration form so that the user can enter the license key. The license is also checked in other parts of the code such as in the STK.

Login

Handles incoming login requests and delegates to login handlers. The handlers are configured under this filter. 

 checks to verify that user credentials have been authenticated so that only authenticated users can be made active users. Magnolia uses JAAS for authentication. For more information see Security.

GZip

To increase the performance of the site 

 replaces text type responses with the gzipped response. The trick in this filter is that it passes a ResponseWrapper instead of the response object it got in the doFilter(..) call to the filter chain. After all the following filters have been executed, content is extracted from the ResponseWrapper, gzipped and written to the original response.

Channel

 resolves the channel to use by considering variations of the set channel. Resulting site gets set in .

Multisite

 is a DX Core filter. It detects which site definition should be used. The filter makes site related properties available in the .

Site merge

 merges channel variations with the site definition. Configurations under this filter override configuration done in the site definition. The filter sets the site definition in the aggregation state. In the Community Edition, this filter sets the site in the aggregation state. In DX Core, the multisite filter can also set the site.

Logout

 checks to see if the logout attribute mgnlLogout is set as a request parameter. If this flag is found, the user will be logged out and the filter chain will restart with the first filter.

Security callback

 handles 401 and 403 HTTP response codes and AccessDeniedExceptions. The filter renders an appropriate "login form" that can consist of a redirect or anything else.

Multiple HttpClientCallbacks with different configuration and behavior can be configured for this filter.

Here is the client callback configuration for the Travel demo members area redirect and login form.

 Both callback classes implement the 

 interface and support their own configuration properties. A custom callback should implement this interface.

 Classes:

• $webResourceManager.requireResource("info.magnolia.sys.confluence.artifact-info-plugin:javadoc-resource-macro-resources") RedirectClientCallback
   redirects to a configured path or URL. This is useful, for example in single sign-on (SSO) context where the login screen is handled by a different application, or to hide the login form from a public instance using a fronting server configuration.
• $webResourceManager.requireResource("info.magnolia.sys.confluence.artifact-info-plugin:javadoc-resource-macro-resources") FormClientCallback
   renders a login form using Freemarker and the template configured with loginForm.

 provides a number of filtering capabilities:

• url. Current request URL decoded and without the context path.
• urlPattern
• originalUrl. Original request URL decoded and without the context path, but not modified by any filter.
• originalUrlPattern
• hostPattern
• voters

For example, in a multisite installation for the request  http://demo.magnolia-cms.com/travel/about.html :

• url is /about.html
• originalUrl is /travel/about.html.

The Multisite filter removes the first-level node name from the URL.

CSRF security

 checks the HTTP referer header to ensure that the request is not a cross-site request forgery attack. When a possible CSRF attack is detected the system serves a 400 Bad Request status to the browser. Magnolia also logs a security warning to the audit log.

The CSRF security filter causes a request to fail if:

• The referer header is empty

  Host: mysite.com/.magnolia/pages/adminCentral.html
  Referer: 

• The host part of the referer header does not match the requested host.

  Host: mysite.com/.magnolia/pages/adminCentral.html
  Referer: hackersite.io

Bypassing the filter

You can bypass the CSRF security filter with a voter.

By default, the filter is bypassed if:

• The requested URL does not start with /.magnolia. Only AdminCentral URLs are vulnerable to CSRF attacks. Other URLs are not checked. 
• The user is not authenticated for AdminCentral access. Only authenticated requests are vulnerable to CSRF attacks.
• The request does not have query parameters.
• The requested resource is somewhere in AdminCentral. Vaadin has its own CSRF protection so Magnolia hands the responsibility over to Vaadin. 

You can create your own whitelist of referrer domains or URIs using a voter. The filter is bypassed for the whitelisted referrers. In this example, we bypass the filter for any requests referred by   http://www.trustedsite.com/  .

Voter properties:

CSRF and Internet Explorer 

Some builds of Internet Explorer don't send the HTTP request header referrer when submitting a form or when opening a popup. If the referrer is not in the HTTP request header, CsrfSecurityFilter#handlePossibleCsrf interprets the request as potential CSRF attack which forces the user to log in on the popup. (See  MAGNOLIA-6211 - Getting issue details... STATUS ). To overcome this issue, add voter class 

 to bypass CsrfSecurityFilter for Internet Explorer.

Below the node allowed you can add a list of regular expressions to match the HTTP header userAgent. In the example above we have bypassed Internet Explorer 6 and 11.

To ensure the filter is bypassed, make sure to have at least one property on the node allowed with a value which will match the userAgent of the browser for which you want to bypass the filter. (For Internet Explorer 11, the userAgent might be Mozilla/5.0 (Windows NT 6.3; Trident/7.0; rv:11.0) like Gecko.)

 Always test changes on test environment before applying it to a production system. When adding an erroneous regular expression - you won't access admin central anymore. (In such case you have to use Groovy Rescue App).

Cross-site security

 grants or denies permission to a site when the site is requested through a particular domain name. For example, if you only grant access to the travel site through www.travel.com, no other URL can be used to access the content. When a user tries to access one site's content through another site's domain name, the system displays a HTTP 404 error (page not found). See Cross-site security

URI security

 checks to see if the active user has permission to access the requested URI. In DX Core,  (included in the  Multisite module) extends UriSecurityFilter to provide site aware functionality.

The following constraints are considered in finding the permissions of the user:

• URI ACLs of the user's roles
• URI ACLs of the user's groups' roles
• Permissions on IP addresses

If the user does not have the permission to access the URI then JAAS will provide a login form. This default behavior of the URI security filter can be changed in JAAS configuration.

You can configure your own login form in the URI security filter to replace the default Magnolia login form. The form is configured in /server/filters/securityCallback/clientCallbacks. Here is an example of a custom form used to grant public users access to a restricted members area. Authentication is delegated to the custom form when a particular URI is accessed.

Range

 adds support for ranged requests. Ranged requests is used by iPhone and some other clients to stream videos. In contrast to Android phones, iPhone does not support any other method of streaming videos.

i18n content support

 detects the requested locale and sets the locale in the aggregation state. The filter rewrites the internal current URI, whether virtual or not. It does not rewrite the URI displayed to the user, however. See Language.

Cache

 manages the Magnolia cache.

The cache filter checks if a requested resource is already stored in the cache to avoid recreation of the resource. If the resource is in the cache, then it will be written to the response and the filter chain stops. If the resource is not found in the cache, then a ResponseWrapper, which not only writes to the "standard" response but also saves the response, is passed to the chain. After the filters that follow have been executed (and the requested resource created), the content is extracted from the response wrapper and stored in the cache.

The cache filter is part of Cache core and the respective configuration can be found in the module configuration.

CORS

SelfConfiguredCorsFilter and SiteAwareCorsFilter (Magnolia 6.2.4+) handle simple and pre-flight requests for Cross-origin resource sharing (CORS). For more details, see the CORS page.

Virtual URI

 checks if the requested URI matches a configured URI pattern and executes the URI mapping.

Servlets filter chain

The servlets configured in modules are installed in Magnolia's servlets filter chain using the 

 as the implementing filter class. If the servlet mapping matches the URI, then the service(..) method of the servlet is called. See Registering a servlet for more.

CMS filter chain

Finally, we arrive at the filter chain which does the page rendering and delivery. The filters are grouped in this filter chain so they share a co-bypass definition.

Repository mapping

 handles access to different workspaces. By default, Magnolia is connected with the website workspace. Therefore a request URI is interpreted as the path to a node in the website workspace. If you want to address nodes in other workspaces you need to specify a repository mapping in /server/URI2RepositoryMapping.

Content security

Whereas the URI security filter checks permissions on the URI, 

 checks if the current user has permission to access the requested content resource. The following constraints are considered in finding the permissions of the user:

• Workspace-specific ACLs of the user's roles
• Workspace-specific ACLs of the user's groups' roles

If the user does not have permission to the resource, then JAAS will provide a login form. This default behavior of the content security filter can be changed in JAAS configuration.

Aggregator

 analyses the request and stores the results in the . After this filter, every value the aggregation state can have is known.

Variant resolver

is a personalization filter that wraps variant nodes. The filter tries to resolve a variant from the current node (from AggregationState) using all available traits stored in the TraitCollector and wraps it accordingly, if required. It only uses PersonalizationNodeWrapper if a variant was resolved. Non-variants are not wrapped.

Property:

Model execution

 executes the component model before template rendering. The filter looks for a request parameter containing the UUID of the component to execute. The model can send output in which case page rendering is skipped, or return a URI prefixed by redirect, permanent or forward.

Rendering

Finally, 

 is responsible for delivering the requested resource. If the requested resource is data, such as a file, then the data is just copied to the response.

The rendering filter is terminal, meaning it ends the filter chain and filtering process. If no filter before it has been able to fulfill the request and the rendering filter cannot find the page either, then a 404 "Page not found" error is returned. This is the default behavior.

You can change the behavior by adding a terminateChain property under the rendering filter and setting it to false. When a request for a page such as /home/some/page is received and no such page exists in the JCR, your own servlets can have a go at fulfilling the request. The default value for the terminateChain property is true.

Restricting access to resources

Access to resources is defined in the /modules/resources/config/resourceFilter filter.  By default, the filter allows access to resources as follows:

• byType
  • css, map, js, htm(l), ico, woff(2), ttf, svg, gif, jp(e)g, tiff, bmp
• byLocation
  • when located in the 'webresources' directory

Adding HTTP headers

The AddHeadersFilter implementation class allows configuration of a filter for adding HTTP headers to enable, for example, Cross-origin resource sharing (CORS). The parameters configured in this filter are added to the HTTP header if the filter is triggered. You can restrict the filter's scope by adding and configuring a bypasses node to it. For details please refer to the Magnolia main filter page.

Example configuration for CORS

 Since Magnolia 6.2.4, you can use the cors filter to handle CORS requests.

The example allows CORS with header types X-Requested-With, Content-Type, Accept, with the GET method, and from any origin:

Properties used in the example:

Custom filters

If you need a custom filter, please read Custom filters.