ERR04-C. Choose an appropriate termination strategy

Some errors, such as out-of-range values, might be the result of erroneous user input. Interactive programs typically handle such errors by rejecting the input and prompting the user for an acceptable value. Servers reject invalid user input by indicating an error to the client while at the same time continuing to service other clients' valid requests. All robust programs must be prepared to gracefully handle resource exhaustion, such as low memory or disk space conditions, at a minimum by preventing the loss of user data kept in volatile storage. Interactive programs may give the user the option to save data on an alternative medium, whereas network servers may respond by reducing throughput or otherwise degrading the quality of service. However, when certain kinds of errors are detected, such as irrecoverable logic errors, rather than risk data corruption by continuing to execute in an indeterminate state, the appropriate strategy may be for the system to quickly shut down, allowing the operator to start it afresh in a determinate state.

ISO/IEC TR 24772:2013, Section 6.39, "Termination Strategy [REU]," [ISO/IEC TR 24772:2013], says:

When a fault is detected, there are many ways in which a system can react. The quickest and most noticeable way is to fail hard, also known as fail fast or fail stop. The reaction to a detected fault is to immediately halt the system. Alternatively, the reaction to a detected fault could be to fail soft. The system would keep working with the faults present, but the performance of the system would be degraded. Systems used in a high availability environment such as telephone switching centers, e-commerce, or other "always available" applications would likely use a fail soft approach. What is actually done in a fail soft approach can vary depending on whether the system is used for safety-critical or security critical purposes. For fail-safe systems, such as flight controllers, traffic signals, or medical monitoring systems, there would be no effort to meet normal operational requirements, but rather to limit the damage or danger caused by the fault. A system that fails securely, such as cryptologic systems, would maintain maximum security when a fault is detected, possibly through a denial of service.

And

The reaction to a fault in a system can depend on the criticality of the part in which the fault originates. When a program consists of several tasks, each task may be critical, or not. If a task is critical, it may or may not be restartable by the rest of the program. Ideally, a task that detects a fault within itself should be able to halt leaving its resources available for use by the rest of the program, halt clearing away its resources, or halt the entire program. The latency of task termination and whether tasks can ignore termination signals should be clearly specified. Having inconsistent reactions to a fault can potentially be a vulnerability.

C provides several options for program termination, including exit(), returning from main(), _Exit(), and abort().

exit()

Calling exit() causes normal program termination to occur. Other than returning from main(), calling exit() is the typical way to end a program. The function takes one argument of type int, which should be either EXIT_SUCCESS or EXIT_FAILURE, indicating successful or unsuccessful termination respectively. The value of EXIT_SUCCESS is guaranteed to be 0. The C Standard, subclause 7.22.4.4 [ISO/IEC 9899:2011], says, "If the value of status is zero or EXIT_SUCCESS, an implementation-defined form of the status successful termination is returned." The exit() function never returns.

#include <stdlib.h>
/* ... */

if (/* Something really bad happened */) {
  exit(EXIT_FAILURE);
}

Calling exit()

• Flushes unwritten buffered data.
• Closes all open files.
• Removes temporary files.
• Returns an integer exit status to the operating system.

The C Standard atexit() function can be used to customize exit() to perform additional actions at program termination.

For example, calling

atexit(turn_gizmo_off);

registers the turn_gizmo_off() function so that a subsequent call to exit() will invoke turn_gizmo_off() as it terminates the program. C requires that atexit() can register at least 32 functions.

Functions registered by the atexit() function are called by exit() or upon normal completion of main().

Note that the behavior of a program that calls exit() from an atexit handler is undefined. (See undefined behavior 182 in Annex J of the C Standard. See also ENV32-C. All exit handlers must return normally.)

return from main()

Returning from main() causes normal program termination to occur, which is the preferred way to terminate a program. Evaluating the return statement has the same effect as calling exit() with the same argument.

#include <stdlib.h>
 
int main(int argc, char **argv) {
  /* ... */
  if (/* Something really bad happened */) {
    return EXIT_FAILURE;
  }
  /* ... */
  return EXIT_SUCCESS;
}

The C Standard, subclause 5.1.2.2.3 [ISO/IEC 9899:2011], has this to say about returning from main():

If the return type of the main function is a type compatible with int, a return from the initial call to the main function is equivalent to calling the exit function with the value returned by the main function as its argument; reaching the } that terminates the main function returns a value of 0. If the return type is not compatible with int, the termination status returned to the host environment is unspecified.

Consequently, returning from main() is equivalent to calling exit(). Many compilers implement this behavior with something analogous to

void _start(void) {
  /* ... */
  exit(main(argc, argv));
}

However, exiting from main is conditional on correctly handling all errors in a way that does not force premature termination. (See ERR00-C. Adopt and implement a consistent and comprehensive error-handling policy and ERR05-C. Application-independent code should provide error detection without dictating error handling.)

_Exit()

Calling _Exit() causes normal program termination to occur. Like the exit() function, _Exit() takes one argument of type int and never returns. However, unlike exit(), whether _Exit() closes open streams, flushes stream buffers,^[1] or deletes temporary files is implementation-defined. Functions registered by atexit() are not executed.

[1] Note that POSIX strengthens the specification for _Exit() by prohibiting the function from flushing stream buffers. See the documentation of the function in The Open Group Base Specifications Issue 7, IEEE Std 1003.1, 2013 Edition [IEEE Std 1003.1:2013].

#include <stdlib.h>
/* ... */

if (/* Something really bad happened */) {
  _Exit(EXIT_FAILURE);
}

The _exit() function is an alias for _Exit().

abort()

Calling abort() causes abnormal program termination to occur unless the SIGABRT signal is caught and the signal handler calls exit() or _Exit():

#include <stdlib.h>
/* ... */

if (/* Something really bad happened */) {
  abort();
}

As with _Exit(), whether open streams with unwritten buffered data are flushed,^[2] open streams are closed, or temporary files are removed is implementation-defined. Functions registered by atexit() are not executed. (See ERR06-C. Understand the termination behavior of assert() and abort().)

[2] Unlike in the case of _Exit(), POSIX explicitly permits but does not require implementations to flush stream buffers. See the documentation of the function in The Open Group Base Specifications Issue 7, IEEE Std 1003.1, 2013 Edition [IEEE Std 1003.1:2013].

Summary

The following table summarizes the exit behavior of the program termination functions.

Table legend:

• – Yes. The specified action is performed.
• – No. The specified action is not performed.
• – Implementation-defined. Whether the specified action is performed depends on the implementation.

Noncompliant Code Example

The abort() function should not be called if it is important to perform application-specific cleanup before exiting. In this noncompliant code example, abort() is called after data is sent to an open file descriptor. The data may or may not be written to the file.

#include <stdlib.h>
#include <stdio.h>

int write_data(void) {
  const char *filename = "hello.txt";
  FILE *f = fopen(filename, "w");
  if (f == NULL) {
    /* Handle error */
  }
  fprintf(f, "Hello, World\n");
  /* ... */
  abort(); /* Oops! Data might not be written! */
  /* ... */
  return 0;
}

int main(void) {
  write_data();
  return EXIT_SUCCESS;
}

Compliant Solution

In this compliant solution, the call to abort() is replaced with exit(), which guarantees that buffered I/O data is flushed to the file descriptor and the file descriptor is properly closed:

#include <stdlib.h>
#include <stdio.h>

int write_data(void) {
  const char *filename = "hello.txt";
  FILE *f = fopen(filename, "w");
  if (f == NULL) {
    /* Handle error */
  }
  fprintf(f, "Hello, World\n");
  /* ... */
  exit(EXIT_FAILURE); /* Writes data and closes f */
  /* ... */
  return 0;
}

int main(void) {
  write_data();
  return EXIT_SUCCESS;
}

Although this particular example benefits from calling exit() over abort(), in some situations, abort() is the better choice. Usually, abort() is preferable when a programmer does not need to close any file descriptors or call any handlers registered with atexit(), for instance, if the speed of terminating the program is critical.

For more details on proper usage of abort(), see ERR06-C. Understand the termination behavior of assert() and abort().

Risk Assessment

As an example, using abort() or _Exit() in place of exit() may leave written files in an inconsistent state and may also leave sensitive temporary files on the file system.

Automated Detection

Related Vulnerabilities

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

Related Guidelines

Bibliography