how to exit a child process - _exit() vs. exit

Question

Consider this code snippet:

pid_t cpid = fork();

if (cpid == -1) {
    perror("fork");
    exit(EXIT_FAILURE);
}

if (cpid == 0) { // in child
    execvp(argv[1], argv + 1);
    perror("execvp);
    _exit(EXIT_FAILURE);
}

// in parent

How shall I exit the child process if execvp returns? Shall I use exit() or _exit()?

Answer 1

execvp will exit the child if successfull so you don't have to exit.

On execve failure I simply use exit(EXIT_FAILURE); in the child.

Edit : i found that after some research : http://www.unixguide.net/unix/programming/1.1.3.shtml

So it's looks like it's better to use _exit() in a fork child specially when you are in C++ :p Thanks for your question i learned something :D

Answer 2

It depends on the behavior you want: man -s 3 exit and man _exit for more details on your system. In general I believe the _exit doesn't run functions that are registered with atexit() whereas exit does (these functions better not call exit - otherwise you get recursion).

In general I would prefer exit over _exit except for in functions registered with atexit, in those I would call _exit, if needed.

Answer 3

You should definitely use _Exit(). exit() calls the functions you added with atexit() and deletes files created with tmpfile(). Since the parent process is really the one that wants these things done when it exists, you should call _Exit(), which does none of these.

Notice _Exit() with a capital E. _exit(2) is probably not what you want to call directly. exit(3) and _Exit(3) will call this for you. If you don't have _Exit(3), then yes, _exit() is what you wanted.

Answer 4

The child of fork() should always call _exit().

Calling exit() instead is a good way to cause pending stdio buffers to be flushed twice.