SEC04-J. Do not load trusted classes after allowing untrusted code to load arbitrary classes

The Java classes used by a program are not necessarily loaded upon program startup. Many JVMs load classes only when they need them.

If untrusted code is permitted to load classes, it may possess the ability to load sensitive classes required by trusted code. If the trusted code has not already loaded these classes, attempts to subsequently do so may result in untrusted classes being substituted for the sensitive classes. Therefore, if a program permits untrusted code to load classes, it must first 'preload' any sensitive classes it needs. Once properly loaded, these sensitive classes may not be 'overrideen' by untrusted code.

Noncompliant Code Example (Tomcat)

This noncompliant code example illustrates a vulnerability present in several versions of the Tomcat HTTP web server (fixed in v 6.0.20) that allows untrusted web applications to override the default XML parser used by the system to process web.xml, context.xml and tld files of other web applications deployed on the Tomcat instance. Consequently, untrusted web applications that install a parser could view and/or alter these files under limited circumstances.

The noncompliant code example shows the code associated with initialiation of a new Digester instance in the org.apache.catalina.startup.ContextConfig class. "A Digester processes an XML input stream by matching a series of element nesting patterns to execute Rules that have been added prior to the start of parsing" [[Tomcat 2009]]. The code to initialize the Digester follows:

protected static Digester webDigester = null;

if (webDigester == null) {
  webDigester = createWebDigester();
}

The createWebDigester() method is responsible for creating the Digester. This method internally calls createWebXMLDigester(), which invokes the method DigesterFactory.newDigester(). Thie method creates the new digester instance and sets a boolean flag useContextClassLoader to true.

// This method exists in the class DigesterFactory and is called by ContextConfig.createWebXmlDigester()
// which is in turn called by ContextConfig.createWebDigester()
// webDigester finally contains the value of digester defined in this method
public static Digester newDigester(boolean xmlValidation, boolean xmlNamespaceAware, RuleSet rule) {
  Digester digester = new Digester();
  // ...
  digester.setUseContextClassLoader(true);
  // ...
  return digester;
}

The useContextClassLoader flag is used by Digester to decide which ClassLoader to use when loading new classes. When true, it uses the WebappClassLoader, which is untrusted, because it loads whatever classes are requested by various web applications.

public ClassLoader getClassLoader() {
  // ...
  if (this.useContextClassLoader) {
    // Uses the context class loader which was previously set to the WebappClassLoader
    ClassLoader classLoader = Thread.currentThread().getContextClassLoader();
  }
  return classloader;
}

Later, the Digester.getParser() method is internally called by Tomcat to process web.xml and other files:

// Digester.getParser() calls this method. It is defined in class Digester
public SAXParserFactory getFactory() {
  if (factory == null) {
    factory = SAXParserFactory.newInstance(); // Uses WebappClassLoader
    // ...
  }
  return (factory);
}

The underlying problem is that the newInstance() method is being invoked on behalf of an untrusted web application's class loader, the WebappClassLoader. If a web application has previously loaded its own javax.xml.parsers.SAXParserFactory, then the web application's SAXParserFactory will be returned by getFactory(), instead of the standard SAXParserFactory used by Tomcat.

Compliant Solution (Tomcat)

This compliant solution initializes the SAXParserFactory when it creates the Digester. This guarantees that the SAXParserFactory is constructed using the container's class loader, instead of the WebappClassLoader.

The webDigester is also marked final. This prevents any subclasses from assigning a new object reference to webDigester. See [OBJ04-J. Do not use public static non-final variables] for more info. It also prevents a race condition where another thread could access webDigester before it is fully initialized; see [OBJ05-J. Prevent access to partially initialized objects] for more info.

protected static final Digester webDigester = init();

protected Digester init() {
  Digester digester = createWebDigester();
  digester.getParser(); // Does not use the context Classloader at initialization, so safe
  return digester;
  // ...
}

Later, even if the Tomcat server still uses the WebappClassLoader to create the parser instance when attempting to process the web.xml and other files, the explicit call to getParser() in init() ensures that the default parser is set during prior initialization and is impossible to replace. Because this is a one-time setting, future attempts to change the parser are futile.

Note that the Class.newInstance() method requires the class to contain a no-argument constructor. If this requirement is not satisfied, a runtime exception results, which indirectly prevents a security breach.

Risk Assessment

Allowing untrusted code to load classes enables untrusted code to replace benign classes with malicious classes.

Automated Detection

TODO

Related Vulnerabilities

Search for vulnerabilities resulting from the violation of this rule on the CERT website.

Bibliography

[!The CERT Oracle Secure Coding Standard for Java^button_arrow_left.png!]      [!The CERT Oracle Secure Coding Standard for Java^button_arrow_up.png!]      [!The CERT Oracle Secure Coding Standard for Java^button_arrow_right.png!]