Re: GSS-APIv3 sketch

[Nicolas Williams <[e-mail]>]

On Thu, Nov 12, 2009 at 08:20:18AM +0100, Volker Lendecke wrote:
> On Wed, Nov 11, 2009 at 05:04:42PM -0600, Nicolas Williams wrote:
> > I'd like to hear more about this.  Can you please describe the problem
> > with libraries creating worker threads?  You can use private e-mail if
> > you wish.  References to bug reports would be great.
> 
> The most nasty implications come from initialization and
> from fork. If you have several libraries doing threads, it
> is easy for unsuspecting applications creating races. The
> worst one is fork: You have to shut down all threads before
> the fork and potentially re-start them after it, because you
> need to drop all locks/mutexes/etc during the fork. If
> needed, I can try to contact the author and get his slides

I'd love to see that, please :)

OpenSolaris' fork() == fork1(), and it has forkx(2) for libraries.
forkx(2) allows library-forked child procs to not interfere with the
application running in the parent, e.g., by not triggering SIGCHLD and
not showing up in plain wait(3C) calls.  And libraries in OpenSolaris
that create worker threads take care to use pthread_atfork() and
atexit() handlers to do fork/exit cleanup.

Of course, pthread_atfork() and atexit() are standard functions...
Libraries on Linux should use them too.

If there are issues in OpenSolaris (apologies to the list) about this, I
really want to hear about them, because we need to make sure they are
addressed if it's at all possible.  I also care to hear about issues on
Linux and *BSD though (I do some development on Linux, but mostly I want
to be informed).

Nico
--