CON01-J. Ensure that compound operations on shared variables are atomic

Compound operations are operations that consist of more than one discrete operation. Expressions that include postfix and prefix increment (++), postfix or prefix decrement (--), or compound assignment operators always result in compound operations. Compound assignment expressions include operators *=, /=, %=, +=, -=, <<=, >>=, >>>=, ^= and |= [[JLS 05]]. Compound operations on shared variables must be performed atomically to prevent [data races] and [race conditions].

For atomicity of a grouping of calls to independently atomic methods belonging to the existing thread-safe classes of the Java SE 6 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 non-atomic 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 (bitwise compound operation)

This noncompliant code example declares a shared boolean variable flag and a toggle() method that negates the current value of flag.

final class Flag {
  private boolean flag = true;
 
  public void toggle() {  // Unsafe
     flag = !flag; 
  }

  public boolean getFlag() { // Unsafe 
    return flag;
  }
}

It is prone to a data race because the value of flag is read, negated, and written back.

Alternatively, the toggle() method can use the compound assignment operator ^= to negate the current value of flag.

final class Flag {
  private boolean flag = true;
 
  public void toggle() {  // Unsafe
    flag ^= true;  // Same as flag = !flag; 
  }

  public boolean getFlag() { // Unsafe 
    return flag;
  }
}

This attempt is also not thread-safe. A data race exists because ^= is a non-atomic compound operation.

Consider, for example, two threads that call toggle(). The effect of toggling flag twice is expected to restore it to its original value. However, the following scenario leaves flag in the wrong state:

As a result, the effect of the call by t₂ is not reflected in flag; the program behaves as if the call was never made.

Noncompliant Code Example (volatile variable)

This noncompliant code example derives from the preceding one but declares flag as volatile.

final class Flag {
  private volatile boolean flag = true;
 
  public void toggle() {  // Unsafe
    flag ^= true; 
  }

  public boolean getFlag() { // Safe
    return flag;
  }
}

It is still insecure for multithreaded use because volatile does not guarantee the visibility of updates to the shared variable flag when a compound operation is performed. In other words, the value of the write depends on the current value of the variable.

Compliant Solution (synchronization)

This compliant solution declares both the toggle() and getFlag() methods as synchronized.

final class Flag {
  private boolean flag = true;
 
  public synchronized void toggle() { 
    flag ^= true; // Same as flag = !flag; 
  }

  public synchronized boolean getFlag() { 
    return flag;
  }
}

This guards reads and writes to the flag field with a lock on the instance, that is, this. This compliant solution ensures that changes are visible to all the threads. Now, only two execution orders are possible, one of which is shown below.

The second execution order involves the same operations, just that t₂ starts and finishes before t₁.

Compliant Solution (cheap read-write lock trick)

It is also permissible to declare flag as volatile to ensure its visibility and while doing so, forgoing to synchronize the getFlag() method. The toggle() method still requires synchronization because it performs a non-atomic operation.

final class Flag {
  private volatile boolean flag = true;
 
  public synchronized void toggle() { 
    flag ^= true; // Same as flag = !flag; 
  }

  public boolean getFlag() { 
    return flag;
  }
}

This advanced technique is fragile in most other scenarios, such as, when a getter method performs operations other than just returning the value of the volatile field. The cheap read-write lock trick offers performance advantages because the method to read a value getFlag() is not synchronized. Unless read performance is critical, this method is not recommended. [[Goetz 06]]

The cheap read-write lock trick is also addressed in CON11-J. Do not assume that declaring an object reference volatile guarantees visibility of its members.

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

This compliant solution uses the java.util.concurrent.atomic.AtomicBoolean type to declare the flag.

final class Flag {
  private AtomicBoolean flag = new AtomicBoolean(true);
 
  public void toggle() { 
    boolean temp;
    do {
      temp = flag.get();
    } while(!flag.compareAndSet(temp, !temp));
  }

  public AtomicBoolean getFlag() { 
    return flag;
  }
}

It ensures that updates to the variable are carried out by using the compareAndSet() method of the class AtomicBoolean. All updates are made visible to other threads.

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.

final class Adder {
  private int a;
  private int b;

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

  public void 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.

Note that declaring the variables as volatile does not resolve the issue because these compound operations involve reads and writes of multiple variables. Also, this code does not prevent integer overflow. See INT00-J. Perform explicit range checking to ensure integer operations do not overflow for more information.

Noncompliant Code Example (overflow check, atomic integer fields)

The issues described in the previous noncompliant code example can also arise when the fields a and b of type int are replaced with atomic integers.

final class Adder {
  private final AtomicInteger a = new AtomicInteger();
  private final AtomicInteger b = new AtomicInteger();

  public int getSum() throws ArithmeticException {
    // Check for integer overflow
    if (b.get() > 0 ? a.get() > Integer.MAX_VALUE - b.get() : a.get() < Integer.MIN_VALUE - b.get()) {
      throw new ArithmeticException("Not in range");
    }
    return a.get() + b.get(); // or, return a.getAndAdd(b.get());
  }

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

For example, when a thread is executing setValues(), another thread may invoke getSum() and retrieve an incorrect result. Furthermore, in the absence of synchronization, there are data races in the check for integer overflow. For instance, a thread can call setValues() after a second thread that is attempting to add the numbers has read a, but before it has read b. In this case, the second thread will get an improper sum.

Even worse, a thread can call setValues() after a second thread has verified that overflow will not occur, but before the second thread reads the values to be added. This would cause the second thread to add two values without checking for overflow, yielding an incorrect sum. Even though a check for integer overflow is installed, it is ineffective because of the time-of-check-time-of-use (TOCTOU) condition between the overflow check and the addition operation.

Compliant Solution (addition, synchronized)

This compliant solution synchronizes the setValues() and getSum() methods so that the entire operation is atomic.

final class Adder {
  private int a;
  private int b;

  public synchronized int getSum() throws ArithmeticException {
    // Check for integer overflow
    if (b > 0 ? a > Integer.MAX_VALUE - b : a < Integer.MIN_VALUE - b) {
      throw new ArithmeticException("Not in range");
    }

    return a + b;
  }

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

Unlike the noncompliant code examples, 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() never returns the unacceptable value 1.

This compliant solution also ensures that there is no TOCTOU condition between checking for overflow and adding the fields.

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

Currently, SureLogic Flashlight does not detect all violations of this guideline. It detects:

It does not detect:

Details

Dynamic analysis tools with a Java concurrency focus, such as SureLogic Flashlight and Coverity Dynamic Analysis will uncover the race conditions shown in the noncompliant code examples above. To accomplish this, however, these tools would have to observe the noncompliant code being called by two or more threads. Such as in an integration or stress test environment. These tools use a dynamic lockset analysis to observe race conditions that occur as the program runs. This analysis intersects the set of locks that are observed to be held when each piece of shared state in the program is accessed. If the lockset for a piece of shared state is empty then a race condition may have been observed and the tool reports this to the user.

Heurisitics-based static analysis tools, such as FindBugs and PMD, do not detect problems with the noncompliant code examples shown above without some "hint" that the program code is intended to be thread-safe. For example, consider the compliant code below where the use of a synchronized method is a hint to the analysis tool that the class is intended to be used concurrently.

public class Foo {
  private boolean flag = true;

  public synchronized boolean toggleAndGet() {
    flag ^= true; // Same as flag = !flag;
    return flag;
  }
}

FindBugs and PMD will not report a warning about this implementation as they do not note any problems.

SureLogic JSure, an analysis-based verification tool, will complain that the lock is unknown to the tool and ask the user to annotate what state the lock protects, i.e., the tool wants to know the locking policy that the programmer intends for this class. To express this intent, the programmer adds two annotations:

@RegionLock("FlagLock is this protects flag")
@Promise("@Unique(return) for new()")
public class Foo {
  private boolean flag = true;

  public synchronized boolean toggleAndGet() {
    flag ^= true; // Same as flag = !flag;
    return flag;
  }
}

The @RegionLock annotation creates a locking policy, named FlagLock, that specifies that reads and writes to the field flag are to be guarded by a lock on the receiver, i.e., this. The second annotation, @Promise is used to place an annotation on the default constructor generated by the compiler. The @Unique("return") annotation promises that the receiver is not aliased during object construction, i.e., that a race condition cannot occur during construction. (CON14-J. Do not let the "this" reference escape during object construction provides further details.) If the constructor was explicit in the code then the annotations would be:

@RegionLock("FlagLock is this protects flag")
public class Foo {
  private boolean flag;

  @Unique("return")
  public Foo() {
    flag = true;
  }

  public synchronized boolean toggleAndGet() {
    flag ^= true; // Same as flag = !flag;
    return flag;
  }
}

The JSure verification tool provides a strong assurance that the annotated model holds for all possible executions of the program. If the below noncompliant code is later added to the class,

  public boolean getValue() {
    return flag;
  }

then JSure will report the violation of the locking policy to the user.

If the noncompliant getValue() method shown above is defined in the code for Foo, then FindBugs can also report a problem, again if the locking model is annotated. However, it uses a different annotation than JSure.

public class Foo {
  @GuardedBy("this")
  private boolean flag = true;

  public synchronized boolean toggleAndGet() {
    flag ^= true; // Same as flag = !flag;
    return flag;
  }

  public boolean getValue() {
    return flag;
  }
}

With the @GuardedBy annotation in place, and only with this annotation in place, FindBugs reports that the field is not guarded against concurrent access in the getValue() method.

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, 2.1.1.1 Objects and locks
[[Bloch 08]] Item 66: Synchronize access to shared mutable data
[[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