Page History

The C++ Standard, [stmt.dcl], paragraph 4 [ISO/IEC 14882-2014], states the following:

The zero-initialization (8.5) of all block-scope variables with static storage duration (3.7.1) or thread storage duration (3.7.2) is performed before any other initialization takes place. Constant initialization (3.6.2) of a block-scope entity with static storage duration, if applicable, is performed before its block is first entered. An implementation is permitted to perform early initialization of other block-scope variables with static or thread storage duration under the same conditions that an implementation is permitted to statically initialize a variable with static or thread storage duration in namespace scope (3.6.2). Otherwise such a variable is initialized the first time control passes through its declaration; such a variable is considered initialized upon the completion of its initialization. If the initialization exits by throwing an exception, the initialization is not complete, so it will be tried again the next time control enters the declaration. If control enters the declaration concurrently while the variable is being initialized, the concurrent execution shall wait for completion of the initialization. If control re-enters the declaration recursively while the variable is being initialized, the behavior is undefined.

Do not reenter a function during the initialization of a static variable declaration. If a function is reentered during the constant initialization of a static object inside that function, the behavior of the program is undefined. Please note that this is a different problem Infinite recursion is not the same as infinite recursion. For this problem to occur, a function only needs to recurse once.

\[[ISO/IEC 14882-2003|AA. C++ References#ISO/IEC 14882-2003]\] Section 6.7, "Declaration Statement" describes the initialization of static and thread storage duration objects.  In the case of static objects, recursive reentry into the initialization of a static storage duration causes undefined behavior and various results can be obtained when using different compilers.

Noncompliant Code Example

This noncompliant code example declares the variable y as a static int. The value of test( x) is assigned to y within the test(int x) function. However, when test(int x) is called with an input which results in reaching the initialization of y more than once, such as the value 12, undefined behavior occurs. Note that this code does not present an infinite recursion and still causes the undefined behavior mentioned.

int test(int x){
  x--;
  if(x < 0 || x > 10)
  {
    return 0;
  }
  else
  {
    static int y = test(x);  //<--undefined behavior occurs here
    return y;
  }
}

The behavior observed from running this code under various compilers differs.

In gcc3, this code will recurse as if y were a non-static variable.

In gcc4, upon reaching the initialization of y for the second time, the program will terminate with the following message:

terminate called after throwing an instance of
'__gnu_cxx::recursive_init'
  what():  N9__gnu_cxx14recursive_initE
Aborted (core dumped)

Compliant Solution (p with Block Scope)

In this compliant solution, p is declared with the same scope as str, preventing p from taking on an indeterminate value outside of this_is_OK().

void this_is_OK(void) {
    const char str[] = "Everything OK";
    const char *p = str;
    /* ... */
}
/* p is inaccessible outside the scope of string str */

Compliant Solution (p with File Scope)

If it is necessary for p to be defined with file scope, it can be set to NULL before str is destroyed. This prevents p from taking on an indeterminate value, although any references to p must check for NULL.

const char *p;
void is_this_OK(void) {
    const char str[] = "Everything OK?";
    p = str;
    /* ... */
    p = NULL;
}

Noncompliant Code Example (Return Values)

In this example, the function init_array() incorrectly returns a pointer to a local stack variable.

char *init_array(void) {
   char array[10];
   /* Initialize array */
   return array;
}

Some compilers generate a warning when a pointer to an automatic variable is returned from a function, as in this example. Compile your code at high warning levels and resolve any warnings (see MSC00-CPP. Compile cleanly at high warning levels).

Compliant Solution (Return Values)

Correcting this example depends on the intent of the programmer. If the intent is to modify the value of array and have that modification persist outside of the scope of init_array(), the desired behavior can be achieved by declaring array elsewhere and passing it as an argument to init_array().

void init_array(char array[]) {
   /* Initialize array */
   return;
}

int main(int argc, char *argv[]) {
   char array[10];
   init_array(array);
   /* ... */
   return 0;
}

Risk Assessment

required to trigger undefined behavior, the function need only recur once as part of the initialization. Due to thread-safe initialization of variables, a single, recursive call will often result in a deadlock due to locking a non-recursive synchronization primitive.

Additionally, the C++ Standard, [basic.start.init], paragraph 2, in part, states the following:

Dynamic initialization of a non-local variable with static storage duration is either ordered or unordered. Definitions of explicitly specialized class template static data members have ordered initialization. Other class template static data members (i.e., implicitly or explicitly instantiated specializations) have unordered initialization. Other non-local variables with static storage duration have ordered initialization. Variables with ordered initialization defined within a single translation unit shall be initialized in the order of their definitions in the translation unit. If a program starts a thread, the subsequent initialization of a variable is unsequenced with respect to the initialization of a variable defined in a different translation unit. Otherwise, the initialization of a variable is indeterminately sequenced with respect to the initialization of a variable defined in a different translation unit. If a program starts a thread, the subsequent unordered initialization of a variable is unsequenced with respect to every other dynamic initialization. Otherwise, the unordered initialization of a variable is indeterminately sequenced with respect to every other dynamic initialization.

Do not create an initialization interdependency between static objects with dynamic initialization unless they are ordered with respect to one another. Unordered initialization, especially prevalent across translation unit boundaries, results in unspecified behavior.

Noncompliant Code Example

This noncompliant example attempts to implement an efficient factorial function using caching. Because the initialization of the static local array cache involves recursion, the behavior of the function is undefined, even though the recursion is not infinite.

#include <stdexcept>
 
int fact(int i) noexcept(false) {
  if (i < 0) {
    // Negative factorials are undefined.
    throw std::domain_error("i must be >= 0");
  }
 
  static const int cache[] = {
    fact(0), fact(1), fact(2), fact(3), fact(4), fact(5),
    fact(6), fact(7), fact(8), fact(9), fact(10), fact(11),
    fact(12), fact(13), fact(14), fact(15), fact(16)
  };
 
  if (i < (sizeof(cache) / sizeof(int))) {
    return cache[i];
  }
 
  return i > 0 ? i * fact(i - 1) : 1;
}

Implementation Details

In Microsoft Visual Studio 2015 and GCC 6.1.0, the recursive initialization of cache deadlocks while initializing the static variable in a thread-safe manner.

Compliant Solution

This compliant solution avoids initializing the static local array cache and instead relies on zero-initialization to determine whether each member of the array has been assigned a value yet and, if not, recursively computes its value. It then returns the cached value when possible or computes the value as needed.

#include <stdexcept>
 
int fact(int i) noexcept(false) {
   if (i < 0) {
    // Negative factorials are undefined.
    throw std::domain_error("i must be >= 0");
  }

  // Use the lazy-initialized cache.
  static int cache[17];
  if (i < (sizeof(cache) / sizeof(int))) {
    if (0 == cache[i]) {
      cache[i] = i > 0 ? i * fact(i - 1) : 1;
    }
    return cache[i];
  }
 
  return i > 0 ? i * fact(i - 1) : 1;
}

Noncompliant Code Example

In this noncompliant code example, the value of numWheels in file1.cpp relies on c being initialized. However, because c is defined in a different translation unit (file2.cpp) than numWheels, there is no guarantee that c will be initialized by calling get_default_car() before numWheels is initialized by calling c.get_num_wheels(). This is often referred to as the "static initialization order fiasco," and the resulting behavior is unspecified.

// file.h
#ifndef FILE_H
#define FILE_H
 
class Car {
  int numWheels;
 
public:
  Car() : numWheels(4) {}
  explicit Car(int numWheels) : numWheels(numWheels) {}
 
  int get_num_wheels() const { return numWheels; }
};
#endif // FILE_H
 
// file1.cpp
#include "file.h"
#include <iostream>
 
extern Car c;
int numWheels = c.get_num_wheels();
 
int main() {
  std::cout << numWheels << std::endl;
}
 
// file2.cpp
#include "file.h"
 
Car get_default_car() { return Car(6); }
Car c = get_default_car();

Implementation Details

The value printed to the standard output stream will often rely on the order in which the translation units are linked. For instance, with Clang 3.8.0 on x86 Linux, the command clang++ file1.cpp file2.cpp && ./a.out will write 0 while clang++ file2.cpp file1.cpp && ./a.out will write 6.

Compliant Solution

This compliant solution uses the "construct on first use" idiom to resolve the static initialization order issue. The code for file.h and file2.cpp are unchanged; only the static numWheels in file1.cpp is moved into the body of a function. Consequently, the initialization of numWheels is guaranteed to happen when control flows over the point of declaration, ensuring control over the order. The global object c is initialized before execution of main() begins, so by the time get_num_wheels() is called, c is guaranteed to have already been dynamically initialized.

// file.h
#ifndef FILE_H
#define FILE_H

class Car {
  int numWheels;

public:
  Car() : numWheels(4) {}
  explicit Car(int numWheels) : numWheels(numWheels) {}

  int get_num_wheels() const { return numWheels; }
};
#endif // FILE_H

// file1.cpp
#include "file.h"
#include <iostream>

int &get_num_wheels() {
  extern Car c;
  static int numWheels = c.get_num_wheels();
  return numWheels;
}

int main() {
  std::cout << get_num_wheels() << std::endl;
}

// file2.cpp
#include "file.h"

Car get_default_car() { return Car(6); }
Car c = get_default_car();

Risk Assessment

Recursively reentering a function during the initialization of one of its static objects can result in an attacker being able to cause a crash or denial of service. Indeterminately ordered dynamic initialization can lead to undefined behavior due to accessing an uninitialized objectReferencing an object outside of its lifetime can result in an attacker being able to run arbitrary code.

Automated Detection

The LDRA tool suite Version 7.6.0 can detect violations of this rule.

Fortify SCA Version 5.0 can detect violations when an array is declared in a function and then a pointer to that array is returned.

Splint Version 3.1.1 can detect violations of this rule.

Compass/ROSE can detect violations of this rule. It automatically detects returning pointers to local variables. Detecting more general cases, such as examples where static pointers are set to local variables which then go out of scope would be difficult.

The Coverity Prevent RETURN_LOCAL checker finds many instances where a function will return a pointer to a local stack variable. Coverity Prevent cannot discover all violations of this rule, so further verification is necessary.

...

Related Vulnerabilities

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

Other Languages

This rule appears in the C Secure Coding Standard as DCL30-C. Declare objects with appropriate storage durations.

References

\[[Coverity 07|AA. C++ References#Coverity 07]\]
\[[ISO/IEC 14882-2003|AA. C++ References#ISO/IEC 14882-2003]\] Sections 3.7, "Storage duration"; 3.8, "Object Lifetime"
\[[Henricson 97|AA. C++ References#Henricson 97]\] Rule 5.9, "A function must never return, or in any other way give access to, references or pointers to local variables outside the scope in which they are declared."
\[[Lockheed Martin 05|AA. C++ References#Lockheed Martin 05]\] AV Rule 111, "A function shall not return a pointer or reference to a non-static local object."
\[[ISO/IEC PDTR 24772|AA. C++ References#ISO/IEC PDTR 24772]\] "DCM Dangling references to stack frames"
\[[MISRA 04|AA. C++ References#MISRA 04]\] Rule 8.6

Related Guidelines

Bibliography

...

Image Added Image Added Image AddedDCL17-CPP. Declare function parameters that are large data structures and are not changed by the function as const references      02. Declarations and Initialization (DCL)      DCL31-CPP. Do not define variadic functions