Thursday, July 13, 2006

C/C++: Unix IPC Shared Memory Example

Some time ago I wrote a few articles on socket programming under Cygwin and process forking. The intended goal was to tie it all together with a third article about Shared Memory and IPC and demonstrate a simple server application. Unfortunately, time interfered with those plans. However, I hate leaving things under the ‘ole Task List for too long, and this article is a bit overdue. So, here is a quick rundown of using IPC mechanisms under Cygwin. I won’t go into detail because there are much more robust tutorials on Unix based IPC mechanism out there (here, here and my personal favorite here).

This program is pretty simple. It will simple create two shared memory segments, one to store a random integer that will be populated by a parent processed, and one to track the number of child processes. When a child dies, it will subtract from the process tracker. The child processes will poll the shared memory every few seconds and die after 20 attempts. The parent process will reset the value of the shared memory integer after a few seconds. The time delay will overlap to demonstrate the child processes are indeed sharing the memory with the parent processes and each other. This builds on the fork example mentioned above. In this example, I wrote it for specific use under Cygwin, using Cygwins IPC service for Windows. The IPC mechanism is Cygwin can get a little confusing since there are two different ones to use. If you are installing a newer Cygwin release, be sure to include this Windows service. I don’t even know if the IPCDaemon executables are options with Cygwin anymore. Anyway here is the example:

* A very simple server using Shared Memory example to communicate between processes.
* The parent will assign the shared memory segment to a random integer, and the child process will
* display what it sees.

//needed for the Shared Memory Functionality
//#include <sys/ipc.h>
#include <sys/shm.h>

//console based IO library
#include <iostream>

//For generating our keys
#include <cstdlib> //needed for the rand function
#include <ctime> //needed to work with time for seeding

using namespace std;

//These are our function declarations
void *get_shared_memory_segment(const int &, const char *); //getting shared memory segments

int main(int argc, char *argv[])
     //constants for shared memory sizes
     const int SHM_INT_SIZE = sizeof(int);
     const int MAX_CHILDREN = 7;

     //Message Count
     int *shared_int, *child_count, pid;
     //We do not want any geeky stuff happening with the child processes later on, so set this up so that
     //the parent does not need to wait for the child to terminate
     signal(SIGCLD, SIG_IGN);
     //seed the random number generator
     //Be verbose, let the console know how large the shared memory segment is set to
     cout << "Size of Shared Memory Segment for ints: " << SHM_INT_SIZE << endl;
     //get our shared memory segments
     shared_int = (int *)get_shared_memory_segment(SHM_INT_SIZE, argv[0]);
     child_count = (int *)get_shared_memory_segment(SHM_INT_SIZE, argv[0]);
     if ((!shared_int) || (!child_count))
          cerr << "Error allocating shared memory!" << endl;
          shared_int = NULL;
          child_count = NULL;
     //initialize our counts to 0
     *shared_int = 0;
     *child_count = 0;

     //Repeat indefinitly
     while (1) {
      if (*child_count < MAX_CHILDREN)
          *child_count = *child_count + 1;
          cout << "--------------------------------" << endl;
          cout << "Number of Children after fork: " << *child_count << endl;
          cout << "--------------------------------" << endl;
          pid = fork();
           if (pid < 0)
               cerr << "Error on fork!" << endl;
                shared_int = NULL;
                child_count = NULL;
           cout << "--------------------------------" << endl;
           cout << "Max children reached. I am not spawning anymore." << endl;
           cout << "--------------------------------" << endl;
      if (pid == 0)  {
          //Inside of child process
          for (int x = 0; x < 20; x++)
          cout << "I am a child " << getpid() << " and I see: " << *shared_int << endl;
          *child_count = *child_count - 1;
          cout << "--------------------------------" << endl;
          cout << "I am dieing... Number of Childred: " << *child_count << endl;
          cout << "--------------------------------" << endl;
          //Exit child process
          //Inside of parent process
          *shared_int = rand();
          cout << "I am the parent " << getpid() << " and I just set the shared memory segment to: " << *shared_int << endl;
     return 0; /* we never get here due to CTRL-C the parent */

//This will allocate the Shared Memory Segment and return a pointer to that segment
void *get_shared_memory_segment(const int &size, const char *filename)
     //Variables for our Shared Memory Segment
     key_t key;
     int shmid;
     int err = 0;

     //This is our local pointer to the shared memory segment
     void *temp_shm_pointer;

     //Create our shared memory key for client connections
     if ((key = ftok(filename, 1 + (rand() % 255) )) == -1)
                //error("Failed to FTOK");
          err = 1;
     //Connect to the shared memory segment and get the shared memory ID
     if ((shmid = shmget(key, size, IPC_CREAT | 0600)) == -1)
     //error("shmget attempt failed");
     err = 1;
     // point the clients to the segment
     temp_shm_pointer = (void *)shmat(shmid, 0, 0);
     if (temp_shm_pointer == (void *)(-1))
        //error("shmat failure");
     err = 1;

     //Return the pointer
     if (!err)
          return temp_shm_pointer;
     return 0;

If you’ve never seen an IPC call, let me explain what is going on here. In this program, we are using an IPC type called Shared Memory. Shared Memory is just that, memory that different programs can share to exchange data. There are other types of IPC’s, which are explained in the IPC tutorials listed above. While this is not really explained in the tutorials, the ftok (file to key) function can use any file that the process can access. In this example, I use the programs actual filename (whatever.exe) rather than some arbitrary filename as indicated in Beejs’s tutorial. I am also using random to help reduce the chances of using the same key. The rest should become clear after reading the tutorials. I will build on this a little more when I have a chance.

No comments: