The C Standard, subclause 7.2.1.1 [ISO/IEC 9899:2011], defines assert() to have the following behavior:

The assert macro puts diagnostic tests into programs; it expands to a void expression. When it is executed, if expression (which shall have a scalar type) is false (that is, compares equal to 0), the assert macro writes information about the particular call that failed (including the text of the argument, the name of the source file, the source line number, and the name of the enclosing function—the latter are respectively the values of the pre-processing macros __FILE__ and __LINE__ and of the identifier __func__) on the standard error stream in an implementation-defined format. It then calls the abort function.

Because assert() calls abort(), cleanup functions registered with atexit() are not called. If the intention of the programmer is to properly clean up in the case of a failed assertion, then runtime assertions should be replaced with static assertions where possible. (See DCL03-C. Use a static assertion to test the value of a constant expression.) When the assertion is based on runtime data, the assert should be replaced with a runtime check that implements the adopted error strategy (see ERR00-C. Adopt and implement a consistent and comprehensive error-handling policy).

See ERR04-C. Choose an appropriate termination strategy for more information on program termination strategies and MSC11-C. Incorporate diagnostic tests using assertions for more information on using the assert() macro.

Noncompliant Code Example

This noncompliant code example defines a function that is called before the program exits to clean up:

void cleanup(void) {
  /* Delete temporary files, restore consistent state, etc. */
}

int main(void) {
  if (atexit(cleanup) != 0) {
    /* Handle error */
  }

  /* ... */

  assert(/* Something bad didn't happen */);

  /* ... */
}

However, the code also has an assert, and if the assertion fails, the cleanup() function is not called.

Compliant Solution

In this compliant solution, the call to assert() is replaced with an if statement that calls exit() to ensure that the proper termination routines are run:

void cleanup(void) {
  /* Delete temporary files, restore consistent state, etc. */
}

int main(void) {
  if (atexit(cleanup) != 0) {
    /* Handle error */
  }

  /* ... */

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

  /* ... */
}

Risk Assessment

Unsafe use of abort() may leave files written in an inconsistent state. It may also leave sensitive temporary files on the file system.

Recommendation

Severity

Likelihood

Remediation Cost

Priority

Level

ERR06-C

Medium

Unlikely

Medium

P4

L3

Automated Detection

Tool

Version

Checker

Description

Astrée
24.04
bad-function
bad-macro-use
Supported
Compass/ROSE



Can detect some violations of this rule. However, it can only detect violations involving abort() because assert() is implemented as a macro

LDRA tool suite
9.7.1
44 SEnhanced enforcement
Parasoft C/C++test

2023.1

CERT_C-ERR06-a

Do not use assertions

PC-lint Plus

1.4

586

Fully supported

PVS-Studio

7.33

V2021


RuleChecker
24.04
bad-function
bad-macro-use
Supported

Related Vulnerabilities

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

Related Guidelines

Bibliography

[ISO/IEC 9899:2011]Subclause 7.2.1.1, "The assert Macro"



4 Comments

  1. Still need some way to verify that the signal is raised by abort/assert and not something else... otherwise somebody could send us a SIGABRT asynchronously and we would still end up calling exit()

    1. AFAIK there is no standard way to know if a signal was sent by the program (say by an assertion failure) or externally. One can prevent external signals by using masks, and one can always provide some internal mechanism (eg set a global flag around assertions) to detect synchronous signals.

      Assuming your cleanup() function does things not safe to do in a signal handler, you are basically asking attackers to kindly not send SIGABRT to your program...not a safe proposition.

      1. I agree with David. The only safe way to deal with signals is to implement an event-driven architecture with a main loop that occasionally polls the status of volatile sig_atomic_t flags and takes action appropriately (see SIG30-C. Call only asynchronous-safe functions within signal handlers) for example.

        I think some other solutions you should suggest as part of this recommendation are:

        1) don't use assert() for detecting runtime errors in systems where terminating the system with a call to abort() is not an option.
        2) use static assertions DCL03-C. Use a static assertion to test the value of a constant expression

        The compliant solution, in fact, may say "replace runtime assertions with static assertions where possible, and when the assertion is based on runtime data, replace the assert with a runtime check that implements the adopted error strategy ERR00-C. Adopt and implement a consistent and comprehensive error-handling policy.

        1. I changed the name of the rule and incorporated your suggestions.