CON22-J. Do not use incorrect forms of the double-checked locking idiom

Sometimes it is required to limit the number of instances of a member object to just one (this is similar to a singleton, however, the member object may or may not be static). Instead of initializing using a constructor, a technique called lazy initialization can be used to defer the construction of the member object until an instance is actually required. Lazy initialization also helps in breaking harmful circularities in class and instance initialization, and performing other optimizations [[Bloch 05]].

A class or an instance method is used for lazy initialization, depending on whether the member object is static or not. The method checks whether the instance has already been created and if not, creates it. If the instance already exists, it simply returns it. This is shown below:

// Correct single threaded version using lazy initialization
final class Foo { 
  private Helper helper = null;
  
  public Helper getHelper() {
    if (helper == null) {
      helper = new Helper();
    }
    return helper;
  }
  // ...
}

In a multithreading scenario, the initialization must be synchronized so that two or more threads do not create multiple instances of the member object. The code shown below is safe for execution in a multithreaded environment, albeit slower than the previous, single threaded code example.

// Correct multithreaded version using synchronization
final class Foo { 
  private Helper helper = null;
  
  public synchronized Helper getHelper() {
    if (helper == null) {
      helper = new Helper();
    }
    return helper;
  }
  // ...
}

The double checked locking (DCL) idiom is used to provide lazy initialization in multithreaded code. In a multithreading scenario, traditional lazy initialization is supplemented by reducing the cost of synchronization for each method access by limiting the synchronization to the case where the instance is required to be created and forgoing it when retrieving an already created instance.

The double-checked locking pattern uses block synchronization instead of method synchronization. It strives to make the previous code example faster by installing a null check before attempting to synchronize. This makes expensive synchronization necessary only for initialization, and dispensable for the common case of retrieving the value of helper. The noncompliant code example shows the originally proposed DCL pattern.

Noncompliant Code Example

This noncompliant code example uses the incorrect form of the double checked locking idiom.

// "Double-Checked Locking" idiom
final class Foo { 
  private Helper helper = null;
  public Helper getHelper() {
    if (helper == null) { 
      synchronized(this) {
        if (helper == null) {
          helper = new Helper();
        }
      }    
    }
    return helper;
  }
  // Other methods and members...
}

According to the Java Memory Model (discussion reference) [[Pugh 04]]:

... writes that initialize the Helper object and the write to the helper field can be done or perceived out of order. As a result, a thread which invokes getHelper() could see a non-null reference to a helper object, but see the default values for fields of the helper object, rather than the values set in the constructor.

Even if the compiler does not reorder those writes, on a multiprocessor the processor or the memory system may reorder those writes, as perceived by a thread running on another processor.

This makes the originally proposed double-checked locking pattern insecure. The guideline CON26-J. Do not publish partially initialized objects further discusses the possible occurrence of a non-null reference to a Helper object that observes default values of fields in the Helper object.

Compliant Solution (volatile)

This compliant solution declares the Helper object as volatile and consequently, uses the correct form of the double-checked locking idiom.

// Works with acquire/release semantics for volatile
// Broken under JDK 1.4 and earlier
final class Foo {
  private volatile Helper helper = null;
  
  public Helper getHelper() { 
    if (helper == null) {
      synchronized(this) {
        if (helper == null) {
          helper = new Helper(); // If the helper is null, create a new instance
        }
      }
    }
    return helper; // If helper is non-null, return its instance
  }
}

If a thread initializes the Helper object, a [happens-before relationship] is established between this thread and another that retrieves and returns the instance. [[Pugh 04]] and [[Manson 04]]

"Today, the double-check idiom is the technique of choice for lazily initializing an instance field. While you can apply the double-check idiom to static fields as well, there is no reason to do so: the lazy initialization holder class idiom is a better choice." [[Bloch 08]].

Compliant Solution (static initialization)

This compliant solution initializes the helper field in the declaration of the static variable [[Manson 06]].

final class Foo {
  private static final Helper helper = new Helper();

  public static Helper getHelper() {
    return helper;
  }
}

Variables declared static are guaranteed to be initialized and instantly made visible to other threads. Static initializers also exhibit these properties. This approach should not be confused with eager initialization because in this case, the Java Language Specification [[JLS 05]] guarantees lazy initialization of the class when it is first used.

Compliant Solution (initialize-on-demand holder class idiom)

This compliant solution uses the initialize-on-demand holder class idiom that implicitly incorporates lazy initialization. It uses a static variable as suggested in the previous compliant solution. The variable is declared within a static inner class, Holder.

final class Foo {
  // Lazy initialization 
  private static class Holder {
    static Helper helper = new Helper();
  }

  public static Helper getInstance() {
    return Holder.helper;
  }
}

Initialization of the Holder class is deferred until the getInstance() method is called, following which the helper field is initialized. The only limitation of this method is that it works only for static fields and not instance fields [[Bloch 01]]. This idiom is a better choice than the double checked locking idiom for lazily initializing static fields [[Bloch 08]].

Compliant Solution (ThreadLocal storage)

This compliant solution (originally suggested by Alexander Terekhov [[Pugh 04]]) uses a ThreadLocal object to lazily create a Helper instance.

class Foo {
  // If perThreadInstance.get() returns a non-null value, this thread
  // has done synchronization needed to see initialization of helper
  private final ThreadLocal perThreadInstance = new ThreadLocal();
  private Helper helper = null;

  public Helper getHelper() {
    if (perThreadInstance.get() == null) {
      createHelper();
    }
    return helper;
  }
        
  private final void createHelper() {
    synchronized(this) {
      if (helper == null) {
        helper = new Helper();
      }
      // Any non-null value would do as the argument here
      perThreadInstance.set(perThreadInstance);
    }
  }
}

Compliant Solution (java.util.concurrent utilities)

This compliant solution uses an AtomicReference wrapper around the Helper object. It uses the standard compareAndSet (CAS) functionality to set a newly created Helper object if helperRef is null. Otherwise, it simply returns the already created instance. (Tom Hawtin, JMM Mailing List)

// Uses atomic utilities
final class Foo {
  private final AtomicReference<Helper> helperRef =
    new AtomicReference<Helper>();

  public Helper getHelper() {
    Helper helper = helperRef.get();
    if (helper != null) {
      return helper;
    }
    Helper newHelper = new Helper();
    return helperRef.compareAndSet(null, newHelper) ?
           newHelper :
           helperRef.get();
  }
}

While this code ensures that only one Helper object is preserved, it may potentially allow multiple Helper objects to be created, with all but one being garbage-collected. However, if constructing multiple Helper objects is infeasible or expensive, this solution may be inappropriate.

Compliant Solution (immutable)

In this compliant solution the Foo class is unchanged, but the Helper class is made immutable. Consequently, the Helper class is guaranteed to be fully constructed before becoming visible. The object must be truly immutable; it is not sufficient for the program to refrain from modifying the object.

public final class Helper {
  private final int n;

  public Helper(int n) {
    this.n = n;
  }

  // Other fields and methods, all fields are final
}

final class Foo {
  private Helper helper = null;
  
  public Helper getHelper() { 
    if (helper == null) {
      synchronized(this) {
        if (helper == null) {
          helper = new Helper(); // If the helper is null, create a new instance
        }
      }
    }
    return helper; // If helper is non-null, return its instance
  }
}

Note that if class Foo were mutable, the Helper field would need to be declared volatile as shown in CON09-J. Ensure visibility of shared references to immutable objects. Also, the method getHelper() is an instance method and the accessibility of the helper field is private. This allows safe publication of the Helper object, in that, a thread cannot observe a partially initialized Foo object (CON26-J. Do not publish partially initialized objects). The class Helper is also compliant with CON26-J. Do not publish partially initialized objects and consequently, cannot be observed to be in a partially initialized state.

Exceptions

EX1: Explicitly synchronized code (that uses method synchronization or proper block synchronization, that is, enclosing all initialization statements) does not require the use of double-checked locking.

EX2: "Although the [noncompliant form of the] double-checked locking idiom cannot be used for references to objects, it can work for 32-bit primitive values (e.g., int's or float's). Note that it does not work for long's or double's, since unsynchronized reads/writes of 64-bit primitives are not guaranteed to be atomic." [[Pugh 04]]. See [CON25-J. Ensure atomicity when reading and writing 64-bit values] for more information.

Risk Assessment

Using incorrect forms of the double checked locking idiom can lead to synchronization issues.

Automated Detection

TODO

Related Vulnerabilities

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

References

[[API 06]]
[[JLS 05]] 12.4 "Initialization of Classes and Interfaces"
[[Pugh 04]]
[[Bloch 01]] Item 48: "Synchronize access to shared mutable data"
[[Bloch 08]] Item 71: "Use lazy initialization judiciously"
[[MITRE 09]] CWE ID 609 "Double-Checked Locking"

CON21-J. Use thread pools to enable graceful degradation of service during traffic bursts      11. Concurrency (CON)      CON23-J. Address the shortcomings of the Singleton design pattern