On 03/03/2011 01:23 AM, Dan Smith wrote:
> DL> The posix semaphore use the mapped deleted hardlinked files.
>
> DL> AFAIK, the current implementation do not support that, no ?
> DL> How is it possible to CR programs using sem_* operations in this case ?
>
> You're right. This addresses the problem of /dev/shm/sem.foo going away
> when the container's /dev/shm mount is destroyed. It does not address
> the fact that the mmap()'d unlinked file is restored as an anonymous
> region and therefore becomes disjoint with the one on-disk. I expect
> the evolution of the unlinked files patches will resolve that.
The mapped file is deleted but hardlinked.
> Our naive tests are not discriminating enough to make this distinction
> at the moment, which certainly needs to be resolved :)
Well a simple test program doing sem_* operations should have spotted
the problem.
Let me know for this one:
#include <stdio.h>
#include <semaphore.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#define SEM "/dummy_sem_open"
int main(int argc, char *argv[])
{
sem_t *sem1, *sem2;
sem_unlink(SEM);
sem1 = sem_open(SEM, O_CREAT | O_EXCL, 0600, 1);
if (sem1 == SEM_FAILED) {
perror("sem_open (1)");
return -1;
}
printf("checkpoint/restart and press enter");
getchar();
sem2 = sem_open(SEM, 0);
if (sem2 == SEM_FAILED) {
perror("sem_open (2)");
return -1;
}
{
int val;
if (sem_getvalue(sem2, &val)) {
perror("sem_getvalue");
return -1;
}
printf("value is %d\n", val);
}
if (sem1 != sem2) {
fprintf(stderr, "sem addresses differ %p <> %p\n",
sem1, sem2);
return -1;
}
return 0;
}
_______________________________________________
Containers mailing list
Containers@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/containers
