在调用execl（）之后退出（0）是否有意义？

Question

Consider the following code:

close(channel_data->pty_master);

if (login_tty(channel_data->pty_slave) != 0) // new terminal session
{
    exit(1); // fail
}

execl("/bin/sh", "sh", mode, command, NULL); // replace process image
exit(0);

According to the documentation of execl() , the current process image is being replaced and the call returns only on errors.

But why call exit() after a call to execl() if the process image is replaced?

Answer 1

Exec calls may fail. A common epilogue would be:

perror("Some eror message");
_exit(127); /*or _exit(1); (127 is what shells return) */

You would usually run _exit rather than exit here, in order to skip atexit hooks and stdio buffer flushes.

Answer 2

It does make sense to call exit after some exec(3) function because they can fail (eg when execve(2) is failing). The execve(2) page lists a number of failure reasons.

It should better be exit(EXIT_FAILURE) or some other (non 0) exit code (conventionally a high exit code like 127 or 126 would be used for that usage, to separate failure of exec vs errors in the program it would run), and I recommend calling perror just before that exit . As explained by PSKocick there are good reasons to call _exit (but his arguments could be reversed, one would want to run atexit and standard fflushs by using exit instead).

In your case, failure is unlikely, but imagine however if some other process has removed /bin/sh (eg the sysadmin making the stupid mistake of running /bin/rm -rf . in the root directory, or in /bin/ , perhaps in some other terminal window).

Still that execve could also fail when system resources are (temporarily) exhausted, eg for

ENOMEM Insufficient kernel memory was available.

And (in rare cases) this could even happen for /bin/sh ;

BTW your exec usage would probably fail (with E2BIG ) if (by mistake) command was a string of a million of non-null bytes.

As a general coding rule, all important system calls should be checked against failure.

Answer 3

You'll want to call exit because you failed to exec the program in question and you typically don't want that process to hang around since it's not running what you wanted it to run. Since execl only returns on failure, there's no need to check the return status.

In many cases, it also makes sense to print an error message to see why it failed. You should also use an exit code other than 0. A non-zero exit code is used to indicate an abnormal exit, and the parent process can capture that when it calls wait .

execl("/bin/sh", "sh", mode, command, NULL);
perror("command failed");
exit(1);

So yes, it makes sense to cal exit , but not necessarily exit(0) .

Answer 4

But why call exit() after a call to execl() if the process image is replaced?

As you said, execl() returns only on errors:

execl("/bin/sh", "sh", mode, command, NULL); // replace process image
exit(0);

In the code above, exit() is called only if the execl() call failed.

---

As Jonathan Leffler suggested in his comment it may be a very good idea to return a value other than zero , since zero indicates success and the code did indeed fail if the program's control flow ever reached the exit() call in the code above.