forking, waitpid problems in c

Question

For some reason this code executes the parental commands immediately, terminating my semaphores and screwing up my flow control of other programs. Can anyone tell me why the waitpid() isnt working?

    //Create child processes
pid = fork();
if(pid < 0){
    fprintf(stderr, "Fork Failed.\n");
    exit(1);
    return;
}else if(pid==0){
        if(execl("/home/tropix/hw11-2","/home/tropix/hw11-2",semarg,pipe_to_p3,pipe_to_p4,(char*)0)){
            fprintf(stderr, "File Exexecution of hw11-2 failed.\n");
            exit(1);
        }
} else {
    pid = fork();
    if(pid < 0){
        fprintf(stderr, "Fork Failed.\n");
        exit(1);
        return;
    } else if(pid==0){
        if(execl("/home/tropix/hw11-3","/home/tropix/hw11-3",shmarg,semarg,pipe_from_p2,pipe_to_p5_1, (char*)0)){
            fprintf(stderr, "File Execution of hw11-3 failed.\n");
            exit(1);
        }
    } else {
        pid = fork();
        if(pid < 0){
            fprintf(stderr, "Fork Failed.\n");
            exit(1);
            return;
        } else if (pid == 0){
            if(execl("/home/tropix/hw11-4","/home/tropix/hw11-4",shmarg,semarg,pipe_from_p2_2,pipe_to_p5_2, (char*)0)){
                fprintf(stderr, "File Execution of hw11-4 failed.\n");
                exit(1);
            }
        } else {
            pid = fork();
            if(pid < 0){
                fprintf(stderr, "Fork Failed.\n");
                exit(1);
                return;
            } else if (pid == 0){
                if(execl("/home/tropix/hw11-5","/home/tropix/hw11-5",semarg,pipe_from_p3,pipe_from_p4,(char*)0)){
                    fprintf(stderr, "File Execution of hw11-5 failed.\n");
                    exit(1);
                }
            } else if (pid > 0) {
            }
        }

    }

    //Closing Pipes
    close(pipe1[1]);
    close(pipe2[1]);
    close(pipe3[1]);
    close(pipe4[1]);
    close(pipe1[0]);
    close(pipe2[0]);
    close(pipe3[0]);
    close(pipe4[0]);

    //Wait for child process completetion
    waitpid(pid,NULL,0);
    printf("Child Processes Complete.\n");

    //Terminate Semaphores
    semctl(sem_id,0,IPC_RMID);

    //Terminate Shared Memory Segement
    shmctl(shmid, IPC_RMID, NULL);



}

}

Thanks!

EDIT: Ok, I replaced waitpid with:

while (pid = waitpid(-1, NULL, 0)) {
       if (errno == ECHILD) {
          break;
       }
    }

and that got me part of the way there. It isnt executing the parental controls immediately, but it seems to never execute now. As far as the pipe issue you talked about, program 1 (this one) is supposed to terminate all IPC elements, including the pipes. If there is a better way, I would love to hear it.

Thanks @Jonathan

Answer 1

You only wait for one process to complete - not for all processes to complete. That is probably one problem. Fix with a loop on waitpid() until it returns 'no more kids'.

The structure of the code leaves something to be desired - it is a rabbit's warren of nested if's; ick!

I worry that you are not closing enough pipes before the other commands are executed. You may be OK if the commands do not depend on detecting EOF on a pipe; otherwise, you are in for a long wait.

You need a function like:

#include <stdarg.h>

static void err_exit(const char *format, ...)
{
    va_list args;
    va_start(args, format);
    vfprintf(stderr, format, args);
    va_end(args);
    exit(EXIT_FAILURE);
}

This simplifies your error handling. You can also do things like automatically add the PID that is dying, or the error that triggered the exit, if you wish.

We can also create a function to run another command:

static pid_t run_command(const char *cmd, const char *shmarg, const char *semarg,
                         const char *fdarg1, const char *fdarg2)
{
    pid_t pid = fork();
    if (pid < 0)
        err_exit("Failed to fork\n");
    else if (pid == 0)
    {
        execl(cmd, cmd, shmarg, semarg, fdarg1, fdarg2, (char *)0);
        err_exit("Failed to exec %s\n", cmd);
    }
    return pid;
}

With those in place, we can look to reduce your code to this...

// Create child processes
pid_t pid1 = run_command("/home/tropix/hw11-2", semarg, pipe_to_p3,   pipe_to_p4);
pid_t pid2 = run_command("/home/tropix/hw11-3", shmarg, semarg, pipe_from_p2,   pipe_to_p5_1);
pid_t pid3 = run_command("/home/tropix/hw11-4", shmarg, semarg, pipe_from_p2_2, pipe_to_p5_2);
pid_t pid4 = run_command("/home/tropix/hw11-5", semarg, pipe_from_p3, pipe_from_p4);

Hmmm...some of these have the shmarg and some don't - is that inconsistency intentional or accidental? We'll assume intentional, so we need two versions of 'run_command()':

static pid_t run_cmd4(const char *cmd, const char *shmarg, const char *semarg,
                         const char *fdarg1, const char *fdarg2)
{
    pid_t pid = fork();
    if (pid < 0)
        err_exit("Failed to fork\n");
    else if (pid == 0)
    {
        execl(cmd, cmd, shmarg, semarg, fdarg1, fdarg2, (char *)0);
        err_exit("Failed to exec %s\n", cmd);
    }
    return pid;
}

static pid_t run_cmd3(const char *cmd, const char *semarg,
                         const char *fdarg1, const char *fdarg2)
{
    pid_t pid = fork();
    if (pid < 0)
        err_exit("Failed to fork\n");
    else if (pid == 0)
    {
        execl(cmd, cmd, semarg, fdarg1, fdarg2, (char *)0);
        err_exit("Failed to exec %s\n", cmd);
    }
    return pid;
}

And then:

// Create child processes
pid_t pid1 = run_cmd3("/home/tropix/hw11-2", semarg, pipe_to_p3,   pipe_to_p4);
pid_t pid2 = run_cmd4("/home/tropix/hw11-3", shmarg, semarg, pipe_from_p2,   pipe_to_p5_1);
pid_t pid3 = run_cmd4("/home/tropix/hw11-4", shmarg, semarg, pipe_from_p2_2, pipe_to_p5_2);
pid_t pid4 = run_cmd3("/home/tropix/hw11-5", semarg, pipe_from_p3, pipe_from_p4);

If it was my code, the names of the variables would be more uniform - and probably in arrays:

// Create child processes
pid_t pid1 = run_cmd3("/home/tropix/hw11-2",         semarg, pipearg[0], pipearg[1]);
pid_t pid2 = run_cmd4("/home/tropix/hw11-3", shmarg, semarg, pipearg[2], pipearg[3]);
pid_t pid3 = run_cmd4("/home/tropix/hw11-4", shmarg, semarg, pipearg[4], pipearg[5]);
pid_t pid4 = run_cmd3("/home/tropix/hw11-5",         semarg, pipearg[6], pipearg[7]);

Then, finally, you have the code:

// Closing Pipes
close(pipe1[1]);
close(pipe2[1]);
close(pipe3[1]);
close(pipe4[1]);
close(pipe1[0]);
close(pipe2[0]);
close(pipe3[0]);
close(pipe4[0]);

//Wait for child process completion
while (waitpid(0, NULL, 0) != 0)
    ;

printf("Child Processes Complete.\n");

// Remove Semaphores and Shared Memory
semctl(sem_id,0,IPC_RMID);
shmctl(shmid, IPC_RMID, NULL);

---

I am deeply suspicious that the run_cmdX() functions also need to close a large selection of the pipes - at least every descriptor of the pipes not intended for communication with their sub-process.

Organizing that cleanly is trickier - but can be done with care. I'd probably create the pipes in a single array:

if (pipe(&pipes[0]) != 0 || pipe(&pipes[2]) != 0 ||
    pipe(&pipes[4]) != 0 || pipe(&pipes[6]) != 0)
    err_exit("Failed to create a pipe\n");

Then I'd create a function:

void pipe_closer(int *pipes, int close_mask)
{
    for (i = 0; i < 8; i++)
    {
         if ((mask & (1 << i)) != 0)
             close(pipes[i]);
    }
}

Then it can be called to close the unneeded pipes:

pipe_closer(pipes, 0xFF);   // Close them all - main()
pipe_closer(pipes, 0xFC);   // All except 0, 1
pipe_closer(pipes, 0xF3);   // All except 2, 3
pipe_closer(pipes, 0xCF);   // All except 4, 5
pipe_closer(pipes, 0x3F);   // All except 6, 7

You just have to arrange for the right mask to be passed with each of the run_cmdN() functions, and the correct calls to be made. If the pipes array is not global, that will need to be passed too. I'd also look at how to encode the data tidily so that the calls to run_cmdN() are as regular and symmetric as possible.

---

Kernighan & Plauger's " The Elements of Programming Style " (2nd Edn, 1978; hard to find, I suspect) contains many magnificent quotes. The immediately apposite one is (bold emphasis added, italics in original):

• [T]he subroutine call permits us to summarize the irregularities in the argument list where we can quickly see what is going on.
• The subroutine itself summarizes the regularities of the code , so repeated patterns need not be used.

This can be viewed as part of the DRY (Don't Repeat Yourself) principle of programming. The err_exit() function call encapsulates three or four lines of code - a print and an exit plus the braces, depending on your preferred layout. The run_command() functions are a prime case of DRY. The proposed pipe_closer() is yet another.