Re: real POSIX.1b semaphores

Ulrich Drepper (
20 Nov 1996 19:13:22 +0100

"Theodore Y. Ts'o" <tytso@MIT.EDU> writes:

> I had read the earlier proposal to mean that named semaphores would also
> go in globally accessible memory, which would have been a real problem.

Yes, this is true. But there is no such thing as read permission,
write permission. Only a general permission the use the semaphore.
So the use of a mapped page for semaphores might still be usable.
Or has anybody a better proposal?

> Because of the security problems, I don't believe global unnamed
> semaphores are useful.

But they have to be available. You can place the semaphore
descriptor in a shared memory segment and make it available to
other processes.

> Process local semaphores are useful for places
> where you need speed, and named semaphores for global access. But if
> you have global unnamed semaphores with no security, they really are
> pointless....

POSIX.1 always says:

>>Either the implemenation shall support the XXX function as described
above or the XXX function shall fail.<<

This means we really have to implement it.

-- Uli
--------------. ,-. Rubensstrasse 5
Ulrich Drepper \ ,--------------------' \ 76149 Karlsruhe/Germany
Cygnus Support `--' `------------------------