I'm using sem_init() to create an un-named semaphore.

The semaphore data structure (sem_t *) is stored in memory that is mmaped from a disk file.

This leads me to ask three questions:

- how can a process that opens the file and does an mmap() determine if sem_init() has already been called for the semaphore and if the semaphore is valid?

- if all the processes using the semaphore terminate, and then a new process starts and mmap()s the same file, is the semaphore still valid or must sem_init() be called again?

- if the machine is rebooted, and the process is started again after the reboot, does it need to call sem_init() again, or can it just mmap() the file and start using the semaphore in calls to sem_wait(), etc?

I realise that named semaphores offer a more obvious solution (because sem_open() with O_CREAT will only create a semaphore if one doesn't already exist), but I would prefer to use un-named semaphores if at all possible.

[original comment removed]

I just read the sem_init man page - apparently you can use unnamed semaphores across processes (via mmap).

Now the interesting bit - you can use another flag in the mmap to see if the semaphore has been initialized - except of course you need a lock of some sort to do that (which brings you back to a named semaphore).

How to determine if a semaphore has been initialized? Well, you can start out with an mmap() that's all zeroed out; that makes it a bit easier to check the semaphore. You can also look at the code to see how the semaphore is implemented and you should get an idea of how to test.

[yet another edit]

Blah ... back to IPC101 - according to the man pages:

So if the mmap area used by the semaphores is initialized with, say, -1, then you can easily test if the sempahore had been init'ed.