-------------------
> On Mon 08 Jul 2002 at 10:11:37 +0900, Curt Sampson wrote:
> > On Mon, 8 Jul 2002, Olaf Seibert wrote:
> > 
> > > I had a very similar thought: why not put these shared memory
segements
> > > in some real directory in a real file system, and then just
implement
> > > access via mmap()? If one wants to remove an unused shared
segment, just
> > > rm the file?
> > 
> > I thought about that too, in fact. The one real advantage this
> > gives you is that you've greatly expanded the potential amount of
> > shared memory you can allocate.
> > 
> > As for the rest, well, we already have ipcs and ipcrm, and we have
> > to keep them for use with the semaphores and message queues anyway.
> > Why spend time re-implementing functionality we already have, when
we
> > can't even get rid of the old interface?
> 
> That is true. But it could be implemented entirely in userland, I
> suspect. So the kernel could be a bit smaller. shmat(2) and shmdt(2)
> could be done in libc witm mmap, shmget(2) would create a file, if
> necessary, and shmctl(2) would stat(2), unlink(2) or chmod(2) it.

There are performance and security issues with userland 
implementations.

> That's no solution for  semaphores and message queues but maybe we'll
> think of something for them too.

These can be thought of as special cases on top of shm.  No?

> 
> > cjs
> -Olaf.
> -- 
> ___ Olaf 'Rhialto' Seibert - rhialto@       -- Woe betide the one who
feels
> \X/ polderland.nl  -- remorse without sin - Tom Poes, "Het boze oog",
4444.
> 
Sincerely,

Simon Shapiro, CEO
Nomis Storage, Inc.