Automatically Unlink POSIX Shared Mem?

[Michael B Allen]

I only want a shared memory segment to exist if 1 or more processes
are attached to it. I suppose I could store a link count in the shared
memory itself and have the last process to detach call shm_unlink but
I was hoping there was an elegant solution builtin.

Actually, what I'd _really_ like to do is have the shared memory
segment automatically unlink itself if no processes are attached after
a configurable amount of time like 5 minutes.

Is this possible?

Thanks,
Mike

[Gianni Mariani]

Michael B Allen wrote:

> I only want a shared memory segment to exist if 1 or more processes
> are attached to it. I suppose I could store a link count in the shared
> memory itself and have the last process to detach call shm_unlink but
> I was hoping there was an elegant solution builtin.
>
> Actually, what I'd _really_ like to do is have the shared memory
> segment automatically unlink itself if no processes are attached after
> a configurable amount of time like 5 minutes.
>
> Is this possible?

No. The kernel provides no such function.

Somthing comes close. If you create a file, map it to N processes and
then unlink the file, the file will remain available to all the
processes that have it mapped. The file is deleted when it is no longer
mapped. However, once the file is unlinked from the file system, it no
longer can be mapped by an independant process.

[James Antill]

On Sun, 06 Jul 2003 00:34:53 -0400, Michael B Allen wrote:

> I only want a shared memory segment to exist if 1 or more processes
> are attached to it. I suppose I could store a link count in the shared
> memory itself and have the last process to detach call shm_unlink but
> I was hoping there was an elegant solution builtin.

Why don't you just used mmap() and MAP_SHARED? This is a much cleaner
interface, however even then you can't say delete this in X seconds if
noone is using it.
There is a trick used in gimp/xscreensaver (and porbably elsewhere) that
lets you portably say "destroy this SHM segment if all attached processes
die".

--
James Antill -- [email]james@and.org[/email]
Need an efficent and powerful string library for C?
[url]http://www.and.org/vstr/[/url]

[Rich Gray]

Gianni Mariani wrote:

>
> Michael B Allen wrote:

> > I only want a shared memory segment to exist if 1 or more processes
> > are attached to it. I suppose I could store a link count in the shared
> > memory itself and have the last process to detach call shm_unlink but
> > I was hoping there was an elegant solution builtin.
> >
> > Actually, what I'd _really_ like to do is have the shared memory
> > segment automatically unlink itself if no processes are attached after
> > a configurable amount of time like 5 minutes.
> >
> > Is this possible?

>
> No. The kernel provides no such function.
>
> Somthing comes close. If you create a file, map it to N processes and
> then unlink the file, the file will remain available to all the
> processes that have it mapped. The file is deleted when it is no longer
> mapped. However, once the file is unlinked from the file system, it no
> longer can be mapped by an independant process.

But, I suppose an independant process could obtain a file handle to the
unlinked file by using some form of IPC to request a holder of the
file's
fd pass a descriptor using sendmsg() over a Unix domain socket (UNPv1
section 14.7) or via a door (UNPv2 section 15.8). I've never tried
either technique.

Rich
--
mailto:richNOgSPAM@plustechnologies.com (remove NO SPAM to reply)
[url]http://www.plustechnologies.com[/url]

[Martijn Sipkema]

[...]

> But, I suppose an independant process could obtain a file handle to the
> unlinked file by using some form of IPC to request a holder of the
> file's
> fd pass a descriptor using sendmsg() over a Unix domain socket (UNPv1
> section 14.7) or via a door (UNPv2 section 15.8). I've never tried
> either technique.

message queue descriptors are not file descriptors though, so that won't
work.

--ms

[Martijn Sipkema]

[...]

> True, but this thread has been about shared memory, which can be created
> with the mmap function which takes a file descriptor. The file
> descriptor
> could be passed to another process (even unrelated) and mapped there,
> even
> if the file has been unlinked.
>

> > so that won't work.
> >
> > --ms

>
> It _should_ work, but I've never tried it.

Oh, I'm sorry, I somehow thought message queues were ment. Yes, for POSIX
shared memory this will work.

--ms