Security Aspects

XStream is designed to be an easy to use library. It takes its main task seriously: converting Java objects to XML, and XML to Java objects. As a result, it is possible to create an instance of XStream with the default constructor, call a method to convert an object into XML, then call another method to turn the XML back into an equivalent Java object. By design, there are few limits to the type of objects XStream can handle.

This flexibility comes at a price. XStream applies various techniques under the hood to ensure it is able to handle all types of objects. This includes using undocumented Java features and reflection. The XML generated by XStream includes all information required to build objects of almost any type. This introduces a potential security problem.

The XML provided to XStream for conversion to a Java object can be manipulated to inject objects into the unmarshalled object graph, which were not present at marshalling time. An attacker could exploit this to execute arbitrary code or shell commands in the context of the server running the XStream process. This issue is identified by CVE-2013-7285.

Note that the XML data could be manipulated on different levels. For example, manipulating values on existing objects (such as a price value), or breaking the format and causing the XML parser to fail. The latter case will raise an exception, but the former case must be handled by validity checks in any application which processes user-supplied XML. The worst case scenario is injection of arbitrary code or shell commands, as noted above.

External Security

An active Java Security Manager can prevent actions required by XStream components or converters. The same applies to environments such as Google Application Engine. XStream tries to some extent to check the functionality of a converter before it claims to handle a type.

        

Therefore it is possible that XStream behaves differently in such an environment, because a converter suddenly no longer handles a special type or any type at all. It is essential that an application that will have to run in such an environment is tested at an early stage to prevent nasty surprises.

Implicit Security

As explained above, it is possible to inject other object instances if an attacker is able to define the data used to deserialize the Java objects (typically XML, but XStream supports other formats like JSON). A known exploit can be created with the help of the Java runtime library using the Java Bean EventHandler. As an instance for the InvocationHandler of a dynamic proxy, it can be used to install a redirect for an arbitrary call to the original object to the method of a completely different instance of an embedded object of the EventHandler itself.

        

This scenario can be used perfectly to replace/inject a dynamic proxy with such an EventHandler at any location in the XML where its parent expects an object of such an interface's type or a simple object instance (any list element will suffice). The usage of a ProcessBuilder as an embedded element, coupled with the redirection of any call to the ProcessBuilder's start() method allows an attacker to call shell commands. Knowing how to define such an attack is the only prerequisite.

        

Starting with XStream 1.4.7, an instance of the EventHandler is no longer handled by default. You have to explicitly register a ReflectionConverter for the EventHandler type, if your application has the requirement to persist such an object. However, you still have to take special care regarding the location of the persisted data, and how your application can ensure its integrity.

        

Note: this vulnerability is not even a special problem of XStream. XML being deserialized by XStream acts here like a script, and the scenario above can be created with any script that is executed within a Java runtime (e.g. using its JavaScript interpreter) if someone is able to manipulate it externally. The key message for application developers is that deserializing arbitrary user-supplied content is a dangerous proposition in all cases.

Explicit Security

While XStream implicitly avoids the vulnerability scenario with the EventHandler class, there might be other combinations with types from well-known and commonly-used Java libraries such as ASM, CGLIB, Groovy, or even in the Java runtime, that are currently simply unknown.

    

Starting with XStream 1.4.7, it is possible to define permissions for types, to check the type of an object that should be unmarshalled. Those permissions can be used to allow or deny types explicitly. With these permissions it is at least not possible to inject unexpected types into an object graph. Any application that deserializes data from an external source should at least use this feature to limit the danger of arbitrary command execution.

        

Apart from value manipulations, this implementation still allows the injection of allowed objects at wrong locations, e.g. inserting an integer into a list of strings.

        

Separate to the XStream security framework, it has always been possible to overwrite the setupConverter method of XStream to register only the required converters.

XML Validation

XML itself supports input validation using a schema and a validating parser. With XStream, you can use e.g. a StAX parser for validation, but it will take some effort to ensure that the XML read and written by XStream matches the schema in first place. Typically you will have to write some custom converters, but it can be worth the effort depending on the use case.

Security Framework

Noted above, it might be possible that other combinations are found with the Java runtime itself, or other commonly-used Java libraries that allow a similar vulnerability like the known case using the Java Beans EventHandler. To prevent such a possibility at all, XStream version 1.4.7 and above contains a security framework, allowing application developers to define which types are allowed to be unmarshalled with XStream.

        

The core interface is TypePermission. The SecurityMapper will evaluate a list of registered instances for every type that will be required while unmarshalling input data. The interface has one simple method:

boolean allow(Class<?>);
        

The XStream facade provides the following methods to register such type permissions within the SecurityMapper:

XStream.addPermission(TypePermission);
XStream.allowTypes(Class[]);
XStream.allowTypes(String[]);
XStream.allowTypesByRegExp(String[]);
XStream.allowTypesByRegExp(Pattern[]);
XStream.allowTypesByWildcard(String[]);
XStream.allowTypeHierary(Class);
XStream.denyPermission(TypePermission);
XStream.denyTypes(Class[]);
XStream.denyTypes(String[]);
XStream.denyTypesByRegExp(String[]);
XStream.denyTypesByRegExp(Pattern[]);
XStream.denyTypesByWildcard(String[]);
XStream.denyTypeHierary(Class);

The sequence of registration is essential. The most recently registered permission will be evaluated first.

        

Every TypePermission has three options to implement the allow method and make decisions on the provided type:

        

Predefined Permission Types

XStream provides some TypePermission implementations to allow any or no type at all, to allow primitive types and their counterpart, null, array types, implementations match the name of the type by regular or wildcard expression and one to invert a permission.

Permission Description Example Default
AnyTypePermission Allow any type. You may use the ANY instance directly. A registration of this permission will wipe any prior one.   yes
ArrayTypePermission Allow any array type. You may use the ARRAYS instance directly.   no
CGLIBProxyTypePermission Allow any CGLIB proxy type. You may use the PROXIES instance directly.   no
ExplicitTypePermission Allow types explicitly by name.  
InterfaceTypePermission Allow any interface type. You may use the INTERFACES instance directly.   no
NoPermission Invert any other permission. Instances of this type are used by XStream in the deny methods.   no
NoTypePermission Allow no type. You may use the NONE instance directly. A registration of this permission will wipe any prior one.  
NullPermission Allow null as type. You may use the NULL instance directly.   no
PrimitiveTypePermission Allow any primitive type and its boxed counterpart (incl void). You may use the PROXIES instance directly.   no
ProxyTypePermission Allow any Java proxy type. You may use the PROXIES instance directly.   no
RegExpTypePermission Allow any type that matches with its name a regular expression. .*\\.core\\..*
[^$]+
TypeHierarchyPermission Allow types of a hierarchy.  
WildcardTypePermission Allow any type that matches with its name a wildcard expression. java.lang.*
java.util.**

Example Code

XStream uses the AnyTypePermission by default, i.e. any type is accepted. You have to clear out this default and register your own permissions to activate the security framework (the Blog type is from the Alias Tutorial):

XStream xstream = new XStream();
// clear out existing permissions and set own ones
xstream.addPermission(NoPermissionType.NONE);
// allow some basics
xstream.addPermission(NullPermission.NULL);
xstream.addPermission(PrimitiveTypePermission.PRIMITIVES);
xstream.allowTypeHierarchy(Collection.class);
// allow any type from the same package
xstream.allowTypesByWildcard(new String[] {
    Blog.class.getPackage().getName()+".*"
});

You may have a further look at XStream's acceptance tests, the security framework is enabled there in general.