Thread: Posix shared memory Q

If I have 2 apps and each one does the following:

Code:

void *shmaddr;
fd = shm_open();
ftruncate(fd, size);
shmaddr = mmap(..., size, ..., fd);

If either app does

memcpy(shmaddr, "hello", 5);

Will the other's shmaddr instantly be able to cast itself to a char * and then I can print that out? And until I do the memcpy, *shmaddr will be NULL?

Originally Posted by homer_3

If I have 2 apps and each one does the following:

Code:

void *shmaddr;
fd = shm_open();
ftruncate(fd, size);
shmaddr = mmap(..., size, ..., fd);

If either app does

memcpy(shmaddr, "hello", 5);

Will the other's shmaddr instantly be able to cast itself to a char * and then I can print that out?

That depends on the structure of the shm object, for ex. char shmbuf[1024]; writing data to an shm object that can be read by another process depends on the arguments given to the shm_open() and mmap() calls. Specifically whether the shm object was created as "read/write" and mapped as MAP_SHARED or MAP_PRIVATE.

Originally Posted by homer_3

And until I do the memcpy, *shmaddr will be NULL?

Depends on whether the OS gives you a clean or dirty page.

For both apps, I'm passing O_RDWR to shm_open and MAP_SHARED to mmap. I'm also memsetting shmaddr to 0.

One thing I wasn't sure about is if I should be checking if shmaddr == NULL or *shmaddr == NULL. I've been trying to test this myself, but even with passing O_CREAT to shm_open, I keep getting ENOENT.

Originally Posted by homer_3

For both apps, I'm passing O_RDWR to shm_open and MAP_SHARED to mmap. I'm also memsetting shmaddr to 0.

But are you creating the shm object by doing a bitwise OR of O_CREAT and O_RDWR?

Originally Posted by homer_3

One thing I wasn't sure about is if I should be checking if shmaddr == NULL or *shmaddr == NULL. I've been trying to test this myself, but even with passing O_CREAT to shm_open, I keep getting ENOENT.

Can't say how you are calling shm_open() and mmap() without looking at the code.

Yea, I guess it is easier when you can see the code. This is what I'm using, and it keeps giving me "shm_open failed ..." with errno being ENOENT.

Code:

#include <errno.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/ipc.h>
#include <sys/shm.h>
#include <sys/mman.h> 

size_t shmsize = 10;
int shmfd;
const char *shmname = "/tmp/shmtest";
unsigned *shmaddr;

int main()
{

  shmfd = shm_open(shmname, O_RDWR | O_CREAT, S_IRUSR | S_IWUSR | S_IWGRP | S_IRGRP | S_IWOTH | S_IROTH);

  if(shmfd > 0)
  {
    if(ftruncate(shmfd, shmsize) != -1)
    {
	shmaddr = (unsigned *)mmap(0, shmsize, PROT_READ | PROT_WRITE, MAP_SHARED, shmfd, 0);

	if(shmaddr != MAP_FAILED)
	{
  	  memset(shmaddr, 0, shmsize);
  	  memcpy(shmaddr, "hello", 5);
          printf("Coppied \"hello\" to shared memory.\n");
	}else
	{
          printf("mmap failed errno %d.\n", errno);
	}
    }else
    {
       printf("ftruncate failed errno %d.\n", errno);
    }
  }else
  {
    printf("shm_open failed errno %d.\n", errno);
  }

  return 0;
}

The code looks good and it runs w/o any issues on my machine, so I'm not sure where the problem is.
Check the permissions, correct owner/group etc., on the working directory and the executable.
Try running it as root if you're on a *nix system. And btw, what compiler and platform are you using?

I'm using CentOS5. g++ version is 4.1.2 and I'm logged in as root.

Originally Posted by homer_3

I'm using CentOS5. g++ version is 4.1.2 and I'm logged in as root.

Don't know anything about Linux distros, but I assume when you installed CentOS5 it created a "/tmp" directory; did it or not?

Yea, there's a /tmp directory. But even if there wasn't O_CREAT should make the path, right?

Originally Posted by homer_3

Yea, there's a /tmp directory. But even if there wasn't O_CREAT should make the path, right?

Nope! the O_CREAT flag applies only to the shm object, not the path to the object.