sem_wait and forked processes

[Frank Bell_1]

Do shared semaphores work in user programs under HPUX 11.00?

Scenario:
Daemon creates semaphore using sem_init(sem, 1, 1) and grabs it using sem_wait OK.
Forks a child.
Child waits on the semaphore (sem_wait) until parent says "ready to go".
Parent does some work and releases the semaphore (sem_post) to inform the child to continue (execv a program).
Under HPUX, the child never wakes up ... sits on the sem_wait indefinitely.

I suspect that the fork is creating a brand new private copy of the semaphore (whilst in the locked_by_parent state) and so is never being cleared (i.e., it is not really a shared semaphore after all). Any ideas please?

[Frank Bell_1]

In case anyone else hits this problem, it appears to be solved by using mmap to explicitly allocate the memory for the semaphore as "shared" ...

sem_t *sem;
void *shmemblk;
...
shmemblk = mmap(NULL, sizeof(sem_t)
, PROT_READ | PROT_WRITE
, MAP_SHARED | MAP_ANONYMOUS
, -1, 0)
...
sem = (sem_t *) shmemblk;
...
sem_init(sem, 1, 1)
...

Many thanks to Andy Bennett (HP) for this information.