亲子沟通导致系统死机

Question

I am trying to create some communication between two programs by forking in the child program within the parent program.

When I execute the child program separately, it works. The purpose of it is that if someone types 1, 2, or 3 followed by enter, that program prints that number as a word. But if one presses 0 and enter, the program exits.

Now I am trying to make the parent program execute the child program in a way where all it does is exit the program while showing the progress of action.

When I execute my program, I see:

Child to start
Parent running OK

Which suggest the child program is running, otherwise I would see:

Exec failed

So instead of me seeing any actual useful output, the system decides to gradually slow down to the point where at first the mouse cursor doesn't move smoothly when I move the mouse, then It got to the point where it wouldn't respond to the keyboard, so I literally had to hold the power button to reset my computer.

How do I fix this so that it can work with any program (that I use as a child) that can exit when I press 0 and enter from within it?

This is my code for the parent:

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/wait.h>

//Setup child read and write file handle named parr and parw respectively
//and parent read and write file handle named  to parr and parw respectively

#define kidr wrp[0]
#define kidw rdp[1]
#define parw wrp[1]
#define parr rdp[0]


int main(){
  int cmd=0;
  //setup and start pipes
  int wrp[2],rdp[2];
  if (pipe(wrp) == -1 || pipe(rdp) == -1){printf("ERROR: cant run pipes.\n");return -1;}

  //Start fork
  pid_t f=fork();

  if (f > 0){
  int wstat; //wait state data
  char buff[100]; //our data buffer
  close(kidr); //we are parent so close child handles
  close(kidw);
  struct timeval tv;
  fd_set readfds;
  tv.tv_sec = 1;
  tv.tv_usec = 0;
  printf("Parent running OK\n");
  while(1){
      //process other async events here

      pid_t wpid=waitpid(-1,&wstat,WNOHANG);
      if (wpid==-1){printf("Wait PID error\n");break;}
      if (wpid > 0){printf("Children closed OK\n");break;}
      //Process data only when child data is readable via pipe
      FD_ZERO(&readfds);FD_SET(parr, &readfds);
      select(parr+1, &readfds, NULL, NULL, &tv); 
      if(FD_ISSET(parr, &readfds)){
        memset(buff,0,99);
        int rd=read(parr, buff, 50);
        //doesnt seem to reach this point...
        if (rd > 0){
        printf("Got: %s\n", buff);
        }else{
          if (cmd==0){
        printf("sending data...\n");
        char*dat="0\n"; //parent sends 0 and the enter button.
        cmd++; //so this doesn't get called again
        write(parw,dat,strlen(dat));
          }
        }
      }
  }
  //close everything and exit
  close(parr);
  close(parw);
  return 0;
  }
  if (f==0){
  printf("Child to start\n");
  //Child mode.
  //Close parents
  close(parr);close(parw);
  //make stdio as child handles
  dup2(kidr,STDIN_FILENO);dup2(kidw,STDOUT_FILENO);
  //close old child handles
  close(kidw);close(kidr);
  execlp("/path/to/forkt","forkt",NULL);
  //We shouldn't get here unless 'ls' command isnt found
  printf("Exec failed\n");
  _exit(-1);
  }

  if (f==-1){
  //If fork() doesnt work...
  printf("Fork error\n");
  }

  return 0;
}

This is my code for the child and I compiled it so its named forkt.

#include <stdio.h>
#include <stdlib.h>
int main(){
  printf("The child has started\n\n");
  char c[100];
  while (1){
printf("Enter number or 0 to exit: \n>");
scanf("%s",c);
if (c[0]=='1'){printf("one\n");}
if (c[0]=='2'){printf("two\n");}
if (c[0]=='3'){printf("three\n");}
if (c[0]=='0'){return 0;}
  }
}

Update

I took a suggestion of running my parent code through the gdb debugger.

I compiled my code using the gcc -g switch then executed it with gdb a.out

Then in gdb, I set a break point to first line of code then used the "run" command then i kept using the "step" command until I found the crashing point which is here:

  pid_t f=fork();
  if (f > 0){ // <- right here

This suggests that somehow the child is creating the lockup(?) even though the child runs fine if it is run by itself without a parent?

Answer 1

AFAICS, the parent won't write anything to the child until the child sends something back, but the child won't send anything until it gets something from the parent. That's a deadlock. There's also a problem with buffering. The pipes are not 'interactive devices' so the output streams are not flushed until the buffer is full, the stream is closed, or you call fflush() .

Here's some alternative but very similar code to yours:

forkt.c

#include <stdio.h>

int main(void)
{
    printf("The child has started\n\n");
    fflush(stdout);
    char c[100];
    while (1)
    {
        printf("Enter number or 0 to exit:\n>");
        fflush(stdout);
        if (scanf("%s", c) != 1)
            return 0;
        fprintf(stderr, "Child received: [%s]\n", c);
        if (c[0] == '1')
        {
            printf("one\n");
        }
        if (c[0] == '2')
        {
            printf("two\n");
        }
        if (c[0] == '3')
        {
            printf("three\n");
        }
        if (c[0] == '0')
        {
            return 0;
        }
        fflush(stdout);
    }
}

This is mostly noticeable for a collection of calls fflush(stdout) .

parent.c

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/wait.h>
#include <unistd.h>

#define kidr wrp[0]
#define kidw rdp[1]
#define parr rdp[0]
#define parw wrp[1]

int main(void)
{
    int cmd = 0;
    int wrp[2], rdp[2];

    fprintf(stderr, "Parent process: PID %d\n", getpid());

    if (pipe(wrp) == -1 || pipe(rdp) == -1)
    {
        fprintf(stderr, "ERROR: cant run pipes.\n");
        exit(1);
    }

    pid_t f = fork();

    if (f == -1)
    {
        fprintf(stderr, "Fork error\n");
        exit(1);
    }

    if (f > 0)
    {
        int wstat;
        char buff[100];
        close(kidr);
        close(kidw);

        fprintf(stderr, "Parent running OK - child %d\n", f);
        while (1)
        {
            pid_t wpid = waitpid(-1, &wstat, WNOHANG);
            if (wpid == -1)
            {
                fprintf(stderr, "Wait PID error\n");
                break;
            }
            if (wpid > 0)
            {
                fprintf(stderr, "Child %d exited\n", wpid);
                break;
            }

            for (cmd = 3; cmd >= 0; cmd--)
            {
                char buffer[30];
                int nb = snprintf(buffer, sizeof(buffer), "%d\n", cmd);
                int wr = write(parw, buffer, strlen(buffer));
                if (wr != nb)
                {
                    fprintf(stderr, "Parent failed to write: %d\n", wr);
                    exit(1);
                }
                fprintf(stderr, "Parent sent: %s", buffer);
                memset(buff, 0, 99);
                int rd = read(parr, buff, 50);
                if (rd > 0)
                {
                    fprintf(stderr, "Got: [[%.*s]]\n", rd, buff);
                }
                else
                {
                    fprintf(stderr, "Parent read failed\n");
                    exit(1);
                }
            }
        }

        close(parr);
        close(parw);
        return 0;
    }

    if (f == 0)
    {
        fprintf(stderr, "Child %d to start\n", getpid());
        close(parr);
        close(parw);
        dup2(kidr, STDIN_FILENO);
        dup2(kidw, STDOUT_FILENO);
        close(kidw);
        close(kidr);
        execlp("forkt", "forkt", NULL);
        fprintf(stderr, "Exec failed\n");
        _exit(-1);
    }

    return 0;
}

The surgery here is more extensive.

When I ran the code, one time I got the output:

Parent process: PID 94693
Parent running OK - child 94694
Parent sent: 3
Child 94694 to start
Got: [[The child has started

]]
Parent sent: 2
Child received: [3]
Got: [[Enter number or 0 to exit:
>]]
Parent sent: 1
Child received: [2]
Got: [[three
Enter number or 0 to exit:
>]]
Parent sent: 0
Child received: [1]
Got: [[two
Enter number or 0 to exit:
>]]
Child received: [0]
Parent sent: 3
Got: [[one
Enter number or 0 to exit:
>]]
Parent sent: 2
Parent read failed

Note that the prompts from the child are mixed up with the output.