subprocess.popen (python) in C

Question

Like the question says, I am searching for the subprocess.open module in C

p = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE, preexec_fn=os.setsid)
print p.pid

I am trying to spawn a process and then get the process ID.

`cmd` is a c program that is spawned

In C, I believe that the getpid() does the job.

But after I spawn the process, I don't know how to tell getpid() to get that spawned process, pid.

void run_apps(pid) {
        printf("process ID is %d\n", (int) getpid());
}

That is obviously giving the current process ID.

Answer 1

In C, you get the maximum number of options by first calling fork() to create a new process (the eventual subprocess), then one of the exec*() family of functions to execute the subprocess. Both the original process and the new process will run concurrently, so you can exchange (read and/or write data) via pipes or socket pairs. Finally, use eg waitpid() in a loop to wait for the new process to exit, and "reap" its exit status. For example:

#include <unistd.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <errno.h>

int main(void)
{
    pid_t child, p;
    int   status;

    /*
     * Prepare pipes et cetera first.
    */

    /* Fork to create the subprocess. */
    child = fork();
    if (child == (pid_t)-1) {
        /* Cannot fork(); usually out of resources (user limits).
         * see errno for details. With <string.h>, you can use
         * strerror(errno) to obtain the error string itself. */
        return 1;

    } else
    if (!child) {
        /* This is the child process itself.
         * Do whatever cleanup is necessary, then
         * execute the subprocess command. */
        execlp("/bin/ls",  "ls", "-lA", NULL);

        /* This is only reached if the exec failed;
         * again, see errno for reason.
         * Always have the child process exit! */
        return 127;
    }

    /* This is only run by the parent process
     * (because the child always exits within the
     *  else if body above).
     *
     * The subprocess PID is 'child'.
    */

    /* Wait for the child process to exit. */
    do {
        status = 0;
        p = waitpid(child, &status, 0);
        if (p == (pid_t)-1 && errno != EINTR)
            break; /* Error */
    } while (p != child);

    if (p != child) {
        /* Child process was lost.
         * If (p == (pid_t)-1), errno describes the error.
        */

    } else
    if (WIFEXITED(status)) {
        /* Child process exited with WEXITSTATUS(status) status.
         * A status of 0 (or EXIT_SUCCESS) means success,
         * no errors occurred. Nonzero usually means an error,
         * but codes vary from binary to binary.
        */

    } else
    if (WIFSIGNALED(status)) {
        /* Child process died from WTERMSIG(status) signal.
         * If you include <string.h>, you can use
         *     strsignal(WTERMSIG(status))
         * to obtain the name (string) of the terminating signal.
        */

    } else {
        /* Child process died from unknown causes.
        */

    }

    /* All done. */
    return 0;
}

Personally, I prefer to use socketpair() to create Unix domain stream or datagram sockets between processes I control, and pipes created via pipe() if the subprocess is just some random binary to be run. In all cases you can replace the standard input ( STDIN_FILENO descriptor), standard output ( STDOUT_FILENO descriptor), and standard error ( STDERR_FILENO descriptor) with sockets or pipes, using the dup2() function. You can even access the pseudo-files under /proc/[child]/ from the parent to observe the state of the child process if you want.

Depending on how you need to communicate with the subprocess -- input/output from/to files? strings in memory? dynamically allocated buffer for output -- there are many variants. Usually code similar to above is used when precise control and/or full-duplex (both read and write) and/or asynchronous communications are needed.

You can do a search for "linux" "fork" "exec" in your favourite search engine for examples of varying quality.

---

If you want an easier solution, and you only need to capture the output of the command (supplying no input to the command, or perhaps supplying the input from a file), you could use some variant of

#include <unistd.h>
#include <stdio.h>
#include <errno.h>

int main(void)
{
    FILE *sub;
    pid_t subpid;
    int   status;

    sub = popen("setsid /bin/sh -c 'echo $$ ; exec command args' </dev/null", "rb");
    if (!sub) {
        /* popen() failed. */
        return 1;
    }

    /* Read the first line from sub. It contains the PID for the command. */
    {
        char buffer[32], *line, dummy;
        int  value;

        line = fgets(buffer, sizeof buffer, sub);
        if (!line) {
            /* setsid failed, or non-POSIXy system (Windows?) */
            pclose(sub);
            return 1;
        }
        if (sscanf(line, "%d%c", &value, &dummy) != 1 || value < 2) {
            /* Internal bug, or extra output? */
            pclose(sub);
            return 1;
        }

        /* subpid is also the session ID and the process group ID,
         * because it is the session leader. */
        subpid = value;
    }

    /* Read from sub using standard I/O, to capture command output. */

    /* After no more output to read from sub, reap the subprocess. */
    errno = 0;
    do {
        status = pclose(sub);
    } while (status == -1 && errno == EINTR);

    if (status) {
        /* Problem: sub exited with nonzero exit status 'status',
         * or if status == -1, some other error occurred. */

    } else {
        /* Sub exited with success (zero exit status). */
    }

    /* Done. */
    return 0;
}

In Linux, popen() uses the /bin/sh shell (as per POSIX.1 specification), and we can use the setsid command-line utility to create the new session. Within the command, the echo $$ is a sh command which outputs the shell PID, and exec CMD... replaces the shell with the command; thus we get the PID of the command even before the command gets executed.