CON01-J. Do not assume that composite operations on primitive data are atomic

Programmers sometimes assume that composite operations on primitive data are atomic. However, this is not true in a multithreaded environment. Even code that does not involve a composite operation can be unsafe for use, for instance:

x = 3;

This code can be fixed by declaring the variable x as volatile. Declaring a shared mutable variable volatile ensures the visibility of the latest updates to it across other threads but does not guarantee the atomicity of composite operations. For example, the post-increment operation consisting of the sequence read-modify-write is not atomic even when the variable is declared volatile.

The statement shown below increments the value of x and is nonatomic because it encapsulates several operations - reading the current value of x, adding one to it, and setting x to the new, incremented value.

x++;

"Sequential consistency and/or freedom from data races still allows errors arising from groups of operations that need to be perceived atomically and are not." [[JLS 05]]. In such cases, the java.util.concurrent utilities can be used to atomically manipulate a shared variable. If these utilities do not provide the required functionality in the form of atomic methods, operations that use the variable should be correctly synchronized.

Note that, as with volatile, updated values are immediately visible to other threads when either of these two techniques is used. Synchronization provides a way to safely share object state across multiple threads without the need to reason about reorderings, compiler optimizations and hardware specific behavior.

This rule specifically deals with primitive operators such as ++. For atomicity of a grouping of calls to atomic methods of the existing Java API, see CON07-J. Do not assume that a grouping of calls to independently atomic methods is atomic.

The Java Language Specification also permits reads and writes of 64-bit values to be nonatomic though this is not an issue with most modern JVMs (see CON25-J. Ensure atomicity when reading and writing 64-bit values).

Noncompliant Code Example (post-decrement composite operation)

In this noncompliant code example, the field itemsInInventory can be accessed by multiple threads.

private int itemsInInventory = 100;

public final int removeItem() {
  if (itemsInInventory > 0) {
    return itemsInInventory--;  // Returns new count of items in inventory
  } 
  return -1; // Error code  
}

However, when a thread is updating the value of itemsInInventory, it is possible for other threads to read the original value (that is, the value before the update). Furthermore, it is possible for two threads to perform the -- operator simultaneously. If this happens, both will read the same value, decrement it, and write the decremented value. This causes itemsInInventory to be decremented by 1 even though it was decremented twice! This is because the post decrement operator is nonatomic.

Noncompliant Code Example (volatile)

This noncompliant code example attempts to fix the visibility issue by declaring itemsInInventory as volatile.

private volatile int itemsInInventory = 100;

public final int removeItem() {
  if (itemsInInventory > 0) {
    return itemsInInventory--;  // Returns new count of items in inventory
  } 
  return -1; // Error code  
}

This guarantees that once the update has taken place, it is immediately visible to all threads that read the field. However, when a thread is in the process of updating the value of itemsInInventory (after the read and modification, but before the write), it is still possible for other threads to read the original value (that is, the value before the update). Furthermore, two threads may still decrement itemsInInventory at the same time, resulting in an incorrect value. This is because the post decrement operator is nonatomic even when volatile is used.

Compliant Solution (java.util.concurrent.atomic classes)

Volatile variables are unsuitable when more than one load/store operation needs to be atomic. This compliant solution uses a java.util.concurrent.atomic.AtomicInteger variable which allows several composite operations to be performed atomically.

private final AtomicInteger itemsInInventory = new AtomicInteger(100);

private final int removeItem() {
  for (;;) {
    int old = itemsInInventory.get();
    if (old > 0) {
      int next = old - 1; // Decrement 
      if (itemsInInventory.compareAndSet(old, next)) {
        return next;  // Returns new count of items in inventory
      }
    } else {
      return -1; // Error code
    }
  }
}

According to the Java API [[API 06]], class AtomicInteger documentation:

[AtomicInteger is an] int value that may be updated atomically. An AtomicInteger is used in applications such as atomically incremented counters, and cannot be used as a replacement for an Integer. However, this class does extend Number to allow uniform access by tools and utilities that deal with numerically-based classes.

The compareAndSet() method takes two arguments, the expected value of a variable when the method is invoked and the updated value. This compliant solution uses this method to atomically set the value of itemsInInventory to the updated value if and only if the current value equals the expected value. [[API 06]]

Compliant Solution (method synchronization)

This compliant solution uses method synchronization to synchronize access to itemsInInventory. Consequently, access to itemsInInventory is mutually exclusive and its state consistent across all threads.

private final int itemsInInventory = 100;

public synchronized int removeItem() {
  if (itemsInInventory > 0) {
    return itemsInInventory--;  // Returns new count of items in inventory
  } 
  return -1; // Error Code 
}

Synchronization is more expensive than using the optimized java.util.concurrent utilities and should only be used when the utilities do not contain the facilities (methods) required to carry out the atomic operation. When synchronizing, care must be taken to avoid deadlocks (see CON12-J. Avoid deadlock by requesting and releasing locks in the same order).

Compliant Solution (block synchronization)

Constructors and methods can use an alternative representation called block synchronization which synchronizes a block of code rather than a method, as highlighted in this compliant solution.

private final int itemsInInventory = 100;

public int removeItem() {
  synchronized(this) {
    if (itemsInInventory > 0) {
      return itemsInInventory--;  // Returns new count of items in inventory    
    } 
    return -1; // Error code
  }
}

Block synchronization is preferable over method synchronization because it reduces the duration for which the lock is held and also protects against denial of service attacks. Block synchronization requires synchronizing on a an internal private lock object instead of the intrinsic lock of the class's object (see CON04-J. Use the private lock object idiom instead of the object's intrinsic locking mechanism).

When the number of items is 0 most of the time, the synchronized block may be moved inside the if condition to reduce the performance cost associated with synchronization. In that case, the variable itemsInInventory must be declared as volatile because the check to determine whether it is greater than 0 should rely on the latest value of itemsInInventory.

Compliant Solution (ReentrantLock)

This compliant solution uses a java.util.concurrent.locks.ReentrantLock to atomically perform the post-decrement operation.

private int itemsInInventory = 100;
private final Lock lock = new ReentrantLock();
    
public int removeItem() {
  Boolean myLock = false;
    
  try {
    myLock = lock.tryLock(); 
    	
    if (itemsInInventory > 0) {
      return itemsInInventory--;
    }  
  } finally {
    if (myLock) {
      lock.unlock();
    }
  }
  return -1; // Error code
}

Code that uses this lock behaves similar to synchronized code that uses the traditional monitor lock. ReentrantLock provides several other capabilities, for instance, the tryLock() method does not block waiting if another thread is already holding the lock. The class java.util.concurrent.locks.ReentrantReadWriteLock can be used when some thread requires a lock to write information while other threads require the lock to simultaneously read the information.

Noncompliant Code Example (addition)

In this noncompliant code example, the two fields a and b may be set by multiple threads, using the setValues() method.

private volatile int a;
private volatile int b;

public int getSum() {
  return a + b;
}

public int setValues(int a, int b) {
  this.a = a;
  this.b = b;
}

The getSum() method may return a different sum every time it is invoked from different threads. For instance, if a and b currently have the value 0, and one thread calls getSum() while another calls setValues(1, 1), then getSum() might return 0, 1, or 2. Of these, the value 1 is unacceptable; it is returned when the first thread reads a and b, after the second thread has set the value of a but before it has set the value of b.

Compliant Solution (addition)

This compliant solution synchronizes the setValues() method so that the entire operation is atomic.

private volatile int a;
private volatile int b;

public synchronized int getSum() {
  return a + b;
}

public synchronized int setValues(int a, int b) {
  this.a = a;
  this.b = b;
}

Unlike the noncompliant code example, if a and b currently have the value 0, and one thread calls getSum() while another calls setValues(1, 1), getSum() may return return 0, or 2, depending on which thread obtains the intrinsic lock first. The locking guarantees that getSum() will never return the unacceptable value 1.

Risk Assessment

If operations on shared variables are not atomic, unexpected results may be produced. For example, there can be inadvertent information disclosure as one user may be able to receive information about other users.

Automated Detection

TODO

Related Vulnerabilities

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

References

[[API 06]] Class AtomicInteger
[[JLS 05]] Chapter 17, Threads and Locks, section 17.4.5 Happens-before Order, section 17.4.3 Programs and Program Order, section 17.4.8 Executions and Causality Requirements
[[Tutorials 08]] Java Concurrency Tutorial
[[Lea 00]] Sections, 2.2.7 The Java Memory Model, 2.2.5 Deadlock, 2.1.1.1 Objects and locks
[[Bloch 08]] Item 66: Synchronize access to shared mutable data
[[Daconta 03]] Item 31: Instance Variables in Servlets
[[JavaThreads 04]] Section 5.2 Atomic Variables
[[Goetz 06]] 2.3. "Locking"
[[MITRE 09]] CWE ID 667 "Insufficient Locking", CWE ID 413 "Insufficient Resource Locking", CWE ID 366 "Race Condition within a Thread", CWE ID 567 "Unsynchronized Access to Shared Data"

11. Concurrency (CON)      11. Concurrency (CON)      CON02-J. Always synchronize on the appropriate object