Page History

Functions that can fail spuriously should be wrapped in a loop.  The atomic_compare_exchange_weak() and atomic_compare_exchange_weak_explicit() functions both attempt to set an atomic variable to a new value but only if it currently possesses a known old value. Unlike the related functions atomic_compare_exchange_strong() and atomic_compare_exchange_strong_explicit(), these functions are permitted to fail spuriously. This makes these functions faster on some platforms—for example, on architectures that implement compare-and-exchange using load-linked/store-conditional instructions, such as Alpha, ARM, MIPS, and PowerPC. The C Standard, 7.17.7.4, paragraph 5 [ISO/IEC 9899:2024], describes this behavior:

A weak compare-and-exchange operation may fail spuriously. That is, even when the contents of memory referred to by expected and object are equal, it may return zero and store back to expected the same memory contents that were originally there.

Noncompliant Code Example

In this noncompliant code example, reorganize_data_structure() is to be used as an argument to thrd_create().  After reorganizing, the function attempts to replace the head pointer so that it points to the new version.  If no other thread has changed the head pointer since it was originally loaded, reorganize_data_structure() is intended to exit the thread with a result of true, indicating success.  Otherwise, the new reorganization attempt is discarded and the thread is exited with a result of false.  However, atomic_compare_exchange_weak() may fail even when the head pointer has not changed. Therefore, reorganize_data_structure() may perform the work and then discard it unnecessarily.

#include <stdatomic.h>
#include <stdbool.h>

struct data {
  struct data *next;
  /* ... */
};

extern void cleanup_data_structure(struct data *head);

int reorganize_data_structure(void *thread_arg) {
  struct data *_Atomic *ptr_to_head = thread_arg;
  struct data *old_head = atomic_load(ptr_to_head);
  struct data *new_head;
  bool success;

  /* ... Reorganize the data structure ... */

  success = atomic_compare_exchange_weak(ptr_to_head,
                                         &old_head, new_head);
  if (!success) {
    cleanup_data_structure(new_head);
  }
  return success; /* Exit the thread */
}

Compliant Solution (atomic_compare_exchange_weak())

To recover from spurious failures, a loop must be used.  However, atomic_compare_exchange_weak() might fail because the head pointer changed, or the failure may be spurious. In either case, the thread must perform the work repeatedly until the compare-and-exchange succeeds, as shown in this compliant solution:

#include <stdatomic.h>
#include <stdbool.h>
#include <stddef.h>

struct data {
  struct data *next;
  /* ... */
};

extern void cleanup_data_structure(struct data *head);

int reorganize_data_structure(void *thread_arg) {
  struct data *_Atomic *ptr_to_head = thread_arg;
  struct data *old_head = atomic_load(ptr_to_head);
  struct data *new_head = NULL;
  struct data *saved_old_head;
  bool success;

  do {
    if (new_head != NULL) {
      cleanup_data_structure(new_head);
    }
    saved_old_head = old_head;

  /* ... Reorganize the data structure ... */

  } while (!(success = atomic_compare_exchange_weak(
               ptr_to_head, &old_head, new_head
             )) && old_head == saved_old_head);
  return success; /* Exit the thread */
}

This loop could also be part of a larger control flow; for example, the thread from the noncompliant code example could be retried if it returns false.

Compliant Solution (atomic_compare_exchange_strong())

When a weak compare-and-exchange would require a loop and a strong one would not, the strong one is preferable, as in this compliant solution:

#include <stdatomic.h>
#include <stdbool.h>

struct data {
  struct data *next;
  /* ... */
};

extern void cleanup_data_structure(struct data *head);

int reorganize_data_structure(void *thread_arg) {
  struct data *_Atomic *ptr_to_head = thread_arg;
  struct data *old_head = atomic_load(ptr_to_head);
  struct data *new_head;
  bool success;

  /* ... Reorganize the data structure ... */

  success = atomic_compare_exchange_strong(
    ptr_to_head, &old_head, new_head
  );
  if (!success) {
    cleanup_data_structure(new_head);
  }
  return success; /* Exit the thread */
}

Automated Detection

Related Vulnerabilities

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

Related Guidelines

Key here (explains table format and definitions)

Bibliography

Image Added Image Added Image Added

The cnd_wait() and cnd_timedwait() functions temporarily cede possession of a mutex so that other threads that may be requesting the mutex can proceed. These functions must always be called from code that is protected by some kind of lock. The waiting thread resumes execution only after it has been notified, generally as the result of the invocation of the cnd_signal() or cnd_broadcast() function by some other thread. The cnd_wait() method must be invoked from a loop that checks whether a condition predicate holds. A condition predicate is an expression constructed from the variables of a function that must be true for a thread to be allowed to continue execution. The thread pauses execution, via cnd_wait(), cnd_timedwait(), or some other mechanism, and is resumed later, presumably when the condition predicate is true and when the thread is notified. Note that a condition predicate is typically the negation of the condition expression in the loop. For example, the condition predicate for removing an element from a linked list is (list->next != NULL), whereas the condition expression for the while loop condition is (list->next == NULL). Following is the correct way to invoke the cnd_wait() function when the list is empty.

struct node_t {
  void* node;
  struct node_t* next;
};

struct node_t list;
static mtx_t lock;
static cnd_t condition;
 
void consume_list_element() {
  int result;
  if ((result = mtx_lock(&lock)) != thrd_success) {
    /* handle error */
  }

  while (list.next == NULL) {
    if ((result = cnd_wait(&condition, &lock)) != thrd_success) {
      /* handle error */
    }
  }
 
  /* Resume when condition holds */

  if ((result = mtx_unlock(&lock)) != thrd_success) {
    /* handle error */
  }
}

The notification mechanism notifies the waiting thread and allows it to check its condition predicate. The invocation of notify() or notifyAll() in another thread cannot precisely determine which waiting thread will be resumed. Condition predicate statements allow notified threads to determine whether they should resume upon receiving the notification. Condition predicates are also useful when a thread is required to block until a condition becomes true, for example, when waiting for data to arrive on an input stream before reading the data.

Both safety and liveness are concerns when using the wait/notify mechanism. The safety property requires that all objects maintain consistent states in a multithreaded environment [Lea 2000]. The liveness property requires that every operation or method invocation execute to completion without interruption.

To guarantee liveness, programs must test the while loop condition before invoking the wait() method. This early test checks whether another thread has already satisfied the condition predicate and sent a notification. Invoking the wait() method after the notification has been sent results in indefinite blocking.

To guarantee safety, programs must test the while loop condition after returning from the wait() method. Although wait() is intended to block indefinitely until a notification is received, it must still be encased within a loop to prevent the following vulnerabilities [Bloch 2001]:

• Thread in the middle — A third thread can acquire the lock on the shared object during the interval between a notification being sent and the receiving thread resuming execution. This third thread can change the state of the object, leaving it inconsistent. This is a TOCTOU race condition.
• Malicious notification — A random or malicious notification can be received when the condition predicate is false. Such a notification would cancel the wait().
• Misdelivered notification — The order in which threads execute after receipt of a notifyAll() signal is unspecified. Consequently, an unrelated thread could start executing and discover that its condition predicate is satisfied. Consequently, it could resume execution, although it was required to remain dormant.
• Spurious wakeups — Certain JVM implementations are vulnerable to spurious wakeups that result in waiting threads waking up even without a notification [API 2006].

For these reasons, programs must check the condition predicate after the wait() method returns. A while loop is the best choice for checking the condition predicate both before and after invoking wait().

Similarly, the await() method of the Condition interface must also be invoked inside a loop. According to the Java API [API 2006], Interface Condition

When waiting upon a Condition, a "spurious wakeup" is permitted to occur, in general, as a concession to the underlying platform semantics. This has little practical impact on most application programs as a Condition should always be waited upon in a loop, testing the state predicate that is being waited for. An implementation is free to remove the possibility of spurious wakeups but it is recommended that applications programmers always assume that they can occur and so always wait in a loop.

New code should use the java.util.concurrent.locks concurrency utilities in place of the wait/notify mechanism. However, legacy code that complies with the other requirements of this rule is permitted to depend on the wait/notify mechanism.

Noncompliant Code Example

This noncompliant code example invokes the wait() method inside a traditional if block and fails to check the postcondition after the notification is received. If the notification were accidental or malicious, the thread could wake up prematurely.

Compliant Solution

This compliant solution calls the wait() method from within a while loop to check the condition both before and after the call to wait().

Invocations of the java.util.concurrent.locks.Condition.await() method must also be enclosed in a similar loop.

Risk Assessment

Failure to encase the wait() or await() methods inside a while loop can lead to indefinite blocking and denial of service (DoS).

Bibliography

Image Removed      Image Removed      Image Removed