Re: [RFC, PATCH] handle the multi-threaded init's exit() properly

[Andrew Morton <akpm@linux-foundation.org>]

On Tue, 7 Aug 2007 00:33:26 +0400 Oleg Nesterov <oleg@tv-sign.ru> wrote:

> On 08/06, Andrew Morton wrote:
> >
> > On Fri, 3 Aug 2007 01:20:09 +0400 Oleg Nesterov <oleg@tv-sign.ru> wrote:
> > 
> > > 2. We are playing games with ->nsproxy->pid_ns. This code is bogus today, and
> > >    it has to be changed anyway when we really support pid namespaces, just
> > >    remove it.
> >
> > This patch broke
> > 
> > pid-namespaces-define-and-use-task_active_pid_ns-wrapper.patch.  This hunk:
> > 
> > ***************
> > *** 908,915 ****
> >   	if (unlikely(!tsk->pid))
> >   		panic("Attempted to kill the idle task!");
> >   	if (unlikely(tsk == child_reaper(tsk))) {
> > - 		if (tsk->nsproxy->pid_ns != &init_pid_ns)
> > - 			tsk->nsproxy->pid_ns->child_reaper = init_pid_ns.child_reaper;
> >   		else
> >   			panic("Attempted to kill init!");
> >   	}
> > --- 908,916 ----
> >   	if (unlikely(!tsk->pid))
> >   		panic("Attempted to kill the idle task!");
> >   	if (unlikely(tsk == child_reaper(tsk))) {
> > + 		if (task_active_pid_ns(tsk) != &init_pid_ns)
> > + 			task_active_pid_ns(tsk)->child_reaper =
> > + 					init_pid_ns.child_reaper;
> >   		else
> >   			panic("Attempted to kill init!");
> >   	}
> > 
> > has no place to live any more, so I just removed it.
> 
> Ah, thanks. I should have done this patch against -mm tree.
> 
> I hope it is OK to drop this chunk of
> 	pid-namespaces-define-and-use-task_active_pid_ns-wrapper.patch
> 
> Because it can't work right now anyway, and Sukadev+Pavel already have
> new patches on top this one which make namespace switch actually work.
> 

OK, well I had to make a bit of on-the-fly adjustment to
pid-namespaces-rename-child_reaper-function.patch as well.

The diff-of-the-diff is:

@@ -48,7 +48,7 @@
 diff -puN kernel/exit.c~pid-namespaces-rename-child_reaper-function kernel/exit.c
 --- a/kernel/exit.c~pid-namespaces-rename-child_reaper-function
 +++ a/kernel/exit.c
-@@ -694,7 +694,7 @@ forget_original_parent(struct task_struc
+@@ -683,7 +683,7 @@ forget_original_parent(struct task_struc
  	do {
  		reaper = next_thread(reaper);
  		if (reaper == father) {
@@ -57,19 +57,10 @@
  			break;
  		}
  	} while (reaper->exit_state);
-@@ -907,7 +907,7 @@ fastcall NORET_TYPE void do_exit(long co
- 		panic("Aiee, killing interrupt handler!");
- 	if (unlikely(!tsk->pid))
- 		panic("Attempted to kill the idle task!");
--	if (unlikely(tsk == child_reaper(tsk))) {
-+	if (unlikely(tsk == task_child_reaper(tsk))) {
- 		if (task_active_pid_ns(tsk) != &init_pid_ns)
- 			task_active_pid_ns(tsk)->child_reaper =
- 					init_pid_ns.child_reaper;
 diff -puN kernel/signal.c~pid-namespaces-rename-child_reaper-function kernel/signal.c

Hopefully people can re-review and retest what's there in next -mm.

Or if that's too much work or too risky, option b) is to drop
handle-the-multi-threaded-inits-exit-properly.patch, go back to the
2.6.23-rc1-mm2 versions of
pid-namespaces-define-and-use-task_active_pid_ns-wrapper.patch and
pid-namespaces-rename-child_reaper-function.patch and to ask Oleg to cook a 2.6.23-rc1-mm2 version of handle-the-multi-threaded-inits-exit-properly.patch.

The downside of this approach is that
handle-the-multi-threaded-inits-exit-properly.patch looks more 2.6.24-ready
than all the container stuff (based just on overall impact and
speculativeness)