Introduction

In C programming, there are two ways to terminate a program from the main function: using return and using exit().

int main() {
    printf("Hello, World!");
    return 0;    // Method 1: Normal termination
}

int main() {
    printf("Hello, World!");
    exit(0);     // Method 2：Normal termination
}

Why can both methods terminate the program correctly, even though they appear completely different?
In this article, we'll unravel this mystery by understanding how C programs actually start and terminate.
Note that this article focuses on the implementation in GNU/Linux environments, specifically using glibc.

How exit() Works

First, let's examine how the exit function works to understand the program termination mechanism.
The exit function is a standard library function that properly terminates a program.
Internally, the _exit function, which is called by exit, is implemented in glibc as follows:

void
_exit (int status)
{
  while (1)
    {
      INLINE_SYSCALL (exit_group, 1, status);

#ifdef ABORT_INSTRUCTION
      ABORT_INSTRUCTION;
#endif
    }
}

Looking at this implementation, we can see that the _exit function receives an exit status as its argument and calls exit_group (system call number 231).

This system call performs the following operations:

1. Sends a program termination notification to the kernel
2. The kernel performs cleanup operations:
   • Releases resources used by the process
   • Updates the process table
   • Performs additional cleanup procedures

Through these operations, the program terminates properly.

So, why does returning from main() also properly terminate the program?

C Program's Hidden Entry Point

To understand this, we need to know an important fact: C programs don't actually start from main.

Let's check the default settings of the linker (ld) to see the actual entry point:

$ ld --verbose | grep "ENTRY"
ENTRY(_start)

As this output shows, the actual entry point of a C program is the _start function. main is called after _start.
The _start function is implemented in the standard library, and in glibc, it looks like this:

_start:
    # Initialize stack pointer
    xorl %ebp, %ebp
    popq %rsi        # Get argc
    movq %rsp, %rdx  # Get argv

    # Setup arguments for main
    pushq %rsi       # Push argc
    pushq %rdx       # Push argv

    # Call __libc_start_main
    call __libc_start_main

The _start function has two main roles:

1. Initializes the stack frame required for program execution
2. Sets up command-line arguments (argc, argv) for the main function

After these initializations are complete, __libc_start_main is called.
This function is responsible for calling the main function.

Now, let's examine how __libc_start_main works in detail.

How __libc_start_main Makes return Work

__libc_start_call_main, which is called by __libc_start_main, is implemented as follows:

_Noreturn static void
__libc_start_call_main (int (*main) (int, char **, char ** MAIN_AUXVEC_DECL),
                        int argc, char **argv
#ifdef LIBC_START_MAIN_AUXVEC_ARG
                            , ElfW(auxv_t) *auxvec
#endif
                        )
{
  int result;

  /* Memory for the cancellation buffer.  */
  struct pthread_unwind_buf unwind_buf;

  int not_first_call;
  DIAG_PUSH_NEEDS_COMMENT;
#if __GNUC_PREREQ (7, 0)
  /* This call results in a -Wstringop-overflow warning because struct
     pthread_unwind_buf is smaller than jmp_buf.  setjmp and longjmp
     do not use anything beyond the common prefix (they never access
     the saved signal mask), so that is a false positive.  */
  DIAG_IGNORE_NEEDS_COMMENT (11, "-Wstringop-overflow=");
#endif
  not_first_call = setjmp ((struct __jmp_buf_tag *) unwind_buf.cancel_jmp_buf);
  DIAG_POP_NEEDS_COMMENT;
  if (__glibc_likely (! not_first_call))
    {
      struct pthread *self = THREAD_SELF;

      /* Store old info.  */
      unwind_buf.priv.data.prev = THREAD_GETMEM (self, cleanup_jmp_buf);
      unwind_buf.priv.data.cleanup = THREAD_GETMEM (self, cleanup);

      /* Store the new cleanup handler info.  */
      THREAD_SETMEM (self, cleanup_jmp_buf, &unwind_buf);

      /* Run the program.  */
      result = main (argc, argv, __environ MAIN_AUXVEC_PARAM);
    }
  else
    {
      /* Remove the thread-local data.  */
      __nptl_deallocate_tsd ();

      /* One less thread.  Decrement the counter.  If it is zero we
         terminate the entire process.  */
      result = 0;
      if (atomic_fetch_add_relaxed (&__nptl_nthreads, -1) != 1)
        /* Not much left to do but to exit the thread, not the process.  */
    while (1)
      INTERNAL_SYSCALL_CALL (exit, 0);
    }

  exit (result);
}

In this implementation, the key parts to focus on are as follows:

result = main (argc, argv, __environ MAIN_AUXVEC_PARAM);
exit(result);

Here, the important point is how the main function is executed and its return value is handled:

1. Executes the main function and stores its return value in result
2. Uses the return value from main as an argument for exit

Through this mechanism:

• When using return in main → The return value is passed to __libc_start_main, which then passes it to exit
• When exit() is called directly in main → The program terminates immediately

In either case, exit is ultimately called, ensuring proper program termination.

Conclusion

C programs have the following mechanism in place:

1. The program starts from _start
2. _start prepares for main's execution
3. main is executed through __libc_start_main
4. Receives main's return value and uses it as an argument for exit

Through this mechanism:

• Even when using return in main, the return value is automatically passed to exit
• As a result, both return and exit() terminate the program properly

Note that this mechanism is not limited to GNU/Linux; similar implementations exist in other operating systems (like Windows and macOS) and different C standard libraries.