Re: [PATCH v2] ns: do not block exit_task_namespaces() for a long time

["Kirill A. Shutemov" <kirill-oKw7cIdHH8eLwutG50LtGA@public.gmane.org>]

On Mon, Jul 16, 2012 at 05:53:01PM +0100, Al Viro wrote:
> On Mon, Jul 16, 2012 at 06:09:24PM +0300, Kirill A. Shutemov wrote:
> > From: "Kirill A. Shutemov" <kirill.shutemov-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
> > 
> > On exiting of the last task in a namespace we need to trigger freeing of
> > the namespace. Currently, we call synchronize_rcu() and free_nsproxy()
> > directly on do_exit() path.
> > 
> > On my machine synchronize_rcu() blocks for about 0.01 seconds. For
> > comparing: normal exit_group() syscall takes less than 0.0003 seconds.
> > 
> > Let's offload synchronize_rcu() and free_nsproxy() to a workqueue.
> > 
> > I also move synchronize_rcu() inside free_nsproxy(). It fixes racy
> > put_nsproxy() which calls free_nsproxy() without synchronize_rcu().
> > I guess it was missed during switch to RCU (see cf7b708).
> 
> NAK.  Making final umounts of anything in that namespace asynchronous,
> even though nothing is holding the stuff on them busy is simply
> wrong.  Note that they can take a _long_ time, so we are talking about
> minutes worth of delay in the worst case.  It's user-visible and
> it's a serious potential for trouble.

Good point.

Now in worst case we have a process which hang for a few minutes in
exit_group() syscall in D state, right? Why is that any better?
Does it provide better user experience or better accounting or what?

-- 
 Kirill A. Shutemov