Page History

Serialization can be used maliciously, for example, to violate the intended invariants of a class. Deserialization is equivalent to object construction; consequently, all invariants enforced during object construction must also be enforced during deserialization. The default serialized form lacks any enforcement of class invariants; consequently, programs must not use the default serialized form for any class with implementation-defined invariants.

The deserialization process creates a new instance of the class without invoking any of the class's constructors. Consequently, any input validation checks in constructors are bypassed. Moreover, transient and static fields may fail to reflect their true values because such fields are bypassed during the serialization procedure and consequently cannot be restored from the object stream. As a result, any class that has transient fields or that performs validation checks in its constructors must also perform similar validation checks when being deserialized.

Validating deserialized objects establishes that the object state is within defined limits and ensures that all transient and static fields have secure values. However, fields that are declared final with a constant value will always be restored to the same constant value after deserialization. For example, the value of the field private transient final n = 42 after deserialization will be 42 rather than 0. Deserialization produces default values for all other cases.

Noncompliant Code Example (Singleton)

In this noncompliant code example [Bloch 2005], a class with singleton semantics uses the default serialized form, which fails to enforce any implementation-defined invariants. Consequently, malicious code can create a second instance even though the class should have only a single instance. For purposes of this example, we assume that the class contains only nonsensitive data.

public class NumberData extends Number {
  // ... Implement abstract Number methods, like Number.doubleValue()...

  private static final NumberData INSTANCE = new NumberData ();
  public static NumberData getInstance() {
    return INSTANCE;
  }

  private NumberData() {
    // Perform security checks and parameter validation
  }

  protected int printData() {
    int data = 1000;
    // Print data
    return data;
  }
}

class Malicious

According to \[[API 06|AA. Java References#API 06]\] class {{Object}}, method {{hashcode()}} documentation:

Whenever it is invoked on the same object more than once during an execution of a Java application, the hashCode method must consistently return the same integer, provided no information used in equals comparisons on the object is modified. This integer need not remain consistent from one execution of an application to another execution of the same application.

Consequently, the hashcode of an object need not remain consistent across different executions of the application. Similarly, if an object is serialized, its hashcode may not stay consistent with the original value. This introduces several hurdles; for example, upon deserialization it is impossible to retrieve the object because its corresponding key could have changed.

Noncompliant Code Example

The {{Key}} class is being used as the key index for the hashtable in this noncompliant code example. According to \[[API 06|AA. Java References#API 06]\] class {{Hashtable}} documentation:

To successfully store and retrieve objects from a hashtable, the objects used as keys must implement the hashCode method and the equals method.

This example follows the above advice but serialization is still at the mercy of the implementation of the hashcode method. It is not required to produce a key value (hashcode) that is consistent across different executions of the program or during (de)serialization. Consequently, using the default serialized form of the hashtable may be inappropriate. In this example, it is not possible to retrieve the value of the object using the original key after deserialization.

class Key implements Serializable {
  //overrides hashcode and equals methods
}

class HashSer {
  public static void main(String[] args) throws IOException, ClassNotFoundException {
    Hashtable<Key,String>NumberData htsc = new Hashtable<Key, String>(NumberData) deepCopy(NumberData.getInstance());
    // KeyPrints keyfalse; =indicates new Key();instance
    htSystem.out.put(key, "Value");println(sc == NumberData.getInstance());  
    System.out.println("Entry:Balance = " + htsc.getprintData(key));
  }

  // retrieve using the key, works
	 
    // Serialize the Hashtable object
    FileOutputStream fos = new FileOutputStream("hashdata.ser");
    ObjectOutputStream oos = new ObjectOutputStream(fos);
    oos.writeObject(ht);
    oos.close();
	 
    // Deserialize the Hashtable object
    FileInputStream fis = new FileInputStream("hashdata.ser");
    ObjectInputStream ois = new ObjectInputStream(fis);
    Hashtable<Key, String> ht_in = (Hashtable)(ois.readObject());
    ois.close();
	 
    if(ht_in.contains("Value")) // check if the object actually exists in the Hashtable
      System.out.println("Value was found in deserialized object.");
	 
    if (ht_in.get(key) == null) // gets printed This method should not be used in production code
  public static Object deepCopy(Object obj) {
    try {
      ByteArrayOutputStream bos = new ByteArrayOutputStream();
      new ObjectOutputStream(bos).writeObject(obj);
      ByteArrayInputStream bin =
          new ByteArrayInputStream(bos.toByteArray());
      return new ObjectInputStream(bin).readObject();
    } catch (Exception e) { 
      throw new IllegalArgumentException(e);
    }
  }
}

Compliant Solution

This compliant solution adds a custom readResolve() method that replaces the deserialized instance with a reference to the appropriate singleton from the current execution. More complicated cases may also require custom writeObject() or readObject() methods in addition to (or instead of) the custom readResolve() method.

public class NumberData extends Number {
  // ...
  protected final Object readResolve() throws NotSerializableException {
    return INSTANCE;
  }
}

More information on singleton classes is available in MSC07-J. Prevent multiple instantiations of singleton objects.

Noncompliant Code Example

This noncompliant code example uses a custom-defined readObject() method but fails to perform input validation after deserialization. The design of the system requires the maximum ticket number of any lottery ticket to be 20,000 and the minimum ticket number be greater than 0. However, an attacker can manipulate the serialized array to generate a different number on deserialization. Such a number could be greater than 20,000 or could be 0 or negative.

public class Lottery implements Serializable {	
  private int ticket = 1;
  private SecureRandom draw = new SecureRandom();

  public Lottery(int ticket) {
    this.ticket = (int) (Math.abs(ticket % 20000) + 1);
  }

  public int getTicket() {
    return this.ticket;	
  }

  public int roll() {
    this.ticket = (int) ((Math.abs(draw.nextInt()) % 20000) + 1);
    return this.ticket;
  }

  public static void main(String[] args) {
    Lottery l = new Lottery(2);
    for (int i = 0; i < 10; i++) {
      l.roll();
      System.out.println("Object was not found when retrieved using the key.");	 l.getTicket());
    }
  }

  private void readObject(ObjectInputStream in) throws IOException, ClassNotFoundException {
    in.defaultReadObject();
  }
}

Compliant Solution

One solution is to change the type of the key value so that it remains consistent across different runs of the program and the multitude of JVMs. This can be achieved by using an Integer object, for example, to hold the key Any input validation performed in the constructors must also be implemented wherever an object can be deserialized. This compliant solution performs field-by-field validation by reading all fields of the object using the readFields() method and ObjectInputStream.GetField constructor. The value for each field must be fully validated before it is assigned to the object under construction. For more complicated invariants, validating before assignment may require reading multiple field values into local variables to enable checks that depend on combinations of field values.

class HashSerpublic final class Lottery implements Serializable { 
  public static// ...
  private synchronized void main(String[] args) readObject(java.io.ObjectInputStream s)
                       throws IOException, ClassNotFoundException {
    Hashtable<Integer,String> htObjectInputStream.GetField fields = new Hashtable<Integer, String>(s.readFields();
    int ticket = fields.get("ticket", 0);
    ht.put(new Integer(1), "Valueif (ticket > 20000 || ticket <= 0) {
      throw new InvalidObjectException("Not in range!");
     System.out.println("Entry: " + ht.get(1)); // retrieve using the key
	 
    // Serialize the Hashtable object
    FileOutputStream fos = new FileOutputStream("hashdata.ser");
    ObjectOutputStream oos = new ObjectOutputStream(fos);
    oos.writeObject(ht);
    oos.close();
	 
    // Deserialize the Hashtable object
    FileInputStream fis = new FileInputStream("hashdata.ser");
    ObjectInputStream ois = new ObjectInputStream(fis);
    Hashtable<Integer, String> ht_in = (Hashtable)(ois.readObject());
    ois.close();
	 
    if(ht_in.contains("Value")) // check if the object actually exists in the Hashtable
      System.out.println("Value was found in deserialized object.");
	 
    if (ht_in.get(1) == null)  // does not get printed
      System.out.println("Object was not found when retrieved using the key.");	 
  }
}

This problem can also be avoided by overriding the equals and the hashcode method in the Key class, though it is best to avoid employing hash tables that are known to use implementation defined parameters.

Risk Assessment

}
    // Validate draw
    this.ticket = ticket;
  }
}

Note that the class must be declared final to prevent a malicious subclass from carrying out a finalizer attack (see OBJ11-J. Be wary of letting constructors throw exceptions for more information about finalizer attacks). For extendable classes, an acceptable alternative is to use a flag that indicates whether the instance is safe for use. The flag can be set after validation and must be checked in every method before any operation is performed.

Additionally, any transient or static fields must be explicitly set to an appropriate value within readObject().

Note that this compliant solution is insufficient to protect sensitive data (see SER03-J. Do not serialize unencrypted sensitive data for additional information).

Compliant Solution (Transient)

This compliant solution marks the fields as transient, so they are not serialized. The readObject() method initializes them using the roll() method. This class need not be final because its fields are private and cannot be tampered with by subclasses, and its methods have been declared final to prevent subclasses from overriding and ignoring them.

public class Lottery implements Serializable {
  private transient int ticket = 1;
  private transient SecureRandom draw = new SecureRandom();

  public Lottery(int ticket) {
    this.ticket = (int) (Math.abs(ticket % 20000) + 1);
  }

  public final int getTicket() {
    return this.ticket;
  }

  public final int roll() {
    this.ticket = (int) ((Math.abs(draw.nextInt()) % 20000) + 1);
    return this.ticket;
  }

  public static void main(String[] args) {
    Lottery l = new Lottery(2);
    for (int i = 0; i < 10; i++) {
      l.roll();
      System.out.println(l.getTicket());
    }
  }

  private void readObject(ObjectInputStream in)
          throws IOException, ClassNotFoundException {
    in.defaultReadObject();
    this.draw = new SecureRandom();
    roll();
  }
}

Compliant Solution (Nonserializable)

This compliant solution simply does not mark the Lottery class serializable:

public final class Lottery {	
  // ...
}

Noncompliant Code Example (AtomicReferenceArray<>)

CVE-2012-0507 describes an exploit that managed to bypass Java's applet security sandbox and run malicious code on a remote user's machine. The exploit deserialized a malicious object that subverted Java's type system. The malicious object was an array of two objects. The second object, of type AtomicReferenceArray<>, contained an array, but this array was also the first object. This data structure could not be created without serialization because the array referenced by AtomicReferenceArray<> should be private. However, whereas the first object was an array of objects of type Help (which inherited from ClassLoader), the AtomicReferenceArray<>'s internal array type was Object, enabling the malicious code to use AtomicReferenceArray.set(ClassLoader) to create an object that was subsequently interpreted as being of type Help object with no cast necessary. A cast would have caught this type mismatch. This exploit allowed attackers to create their own ClassLoader object, which is forbidden by the applet security manager.

This exploit worked because, in Java versions prior to 1.7.0_02, the object of type AtomicReferenceArray<> performed no validation on its internal array.

public class AtomicReferenceArray<E> implements java.io.Serializable {
  private static final long serialVersionUID = -6209656149925076980L;

  // Rest of class...
  // No readObject() method, relies on default readObject
}

Compliant Solution (AtomicReferenceArray<>)

This exploit was mitigated in Java 1.7.0_03 by having the object of type AtomicReferenceArray<> validate its array upon deserialization. The readObject() method inspects the array contents, and if the array is of the wrong type, it makes a defensive copy of the array, foiling the exploit. This technique is recommended by OBJ06-J. Defensively copy mutable inputs and mutable internal components.

public class AtomicReferenceArray<E> implements java.io.Serializable {
  private static final long serialVersionUID = -6209656149925076980L;

  // Rest of class...

  /**
   * Reconstitutes the instance from a stream (that is, deserializes it).
   * @param s the stream
   */
  private void readObject(java.io.ObjectInputStream s)
    throws java.io.IOException, ClassNotFoundException {
    // Note: This must be changed if any additional fields are defined
    Object a = s.readFields().get("array", null);
    if (a == null || !a.getClass().isArray())
      throw new java.io.InvalidObjectException("Not array type");
    if (a.getClass() != Object[].class)
      a = Arrays.copyOf((Object[])a, Array.getLength(a), Object[].class);
    unsafe.putObjectVolatile(this, arrayFieldOffset, a);
  }
}

Risk Assessment

Using the default serialized form for any class with implementation-defined invariants may result in the malicious tampering of class invariantsSerializing objects with implementation defined characteristics can seriously corrupt the state of the object.

Automated Detection

TODO

Related Vulnerabilities

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

References

\[[API 06|AA. Java References#API 06]\] Class {{Object}}, Class {{Hashtable}}
\[[Bloch 08|AA. Java References#Bloch 08]\] Item 75: "Consider using a custom serialized form"

Related Guidelines

Bibliography

...

Image Added Image Added Image AddedSER35-J. Prevent overwriting of Externalizable Objects      12. Serialization (SER)      SER37-J. Do not deserialize from a privileged context