Re: [PATCH 7/8] job control: Notify the real parent of job control events regardless of ptrace

[Tejun Heo <tj@kernel.org>]

Hello,

On Wed, Mar 23, 2011 at 06:07:08PM +0100, Oleg Nesterov wrote:
> On 03/23, Tejun Heo wrote:
> > On Wed, Mar 23, 2011 at 05:46:57PM +0100, Oleg Nesterov wrote:
> > > > Ooh, on a related note, we probably want to change
> > > > do_notify_parent_cldstop() too.  tsk->group_leader->real_parent is
> > > > used as the delivery target when !@for_ptracer.  This is the same with
> > > > tsk->real_parent and the code has been like this for a long time but
> > > > is a bit confusing.
> > >
> > > Yes, although in this case we do
> > >
> > > 		tsk = tsk->group_leader;
> > > 		parent = tsk->real_parent;
> > >
> > > We need to change tsk to report the correct si_pid. But we could do
> > >
> > > 		parent = tsk->real_parent;
> > > 		tsk = tsk->group_leader;
> > >
> > > not sure this looks less confusing.
> >
> > They themselves are about the same but the inconsistency with the
> > is_real_parent() test is a bit confusing, I think.  If we're testing
> > for duplicates by testing ->real_parent against ->parent then the
> > actual delivery should be also be using ->real_parent and ->parent,
> > so...
> 
> Hmm. So I misunderstood you. And still can't understand...
> 
> Could you explain how should we change do_notify_parent_cldstop()?
> I am just curious.

As you suggested above. :-) The inconsistency I'm concernced about is
the following.

 /*
  * Test whether the target task of the usual cldstop notification - the
  * real_parent of @child - is in the same group as the ptracer.
  */
 static bool real_parent_is_ptracer(struct task_struct *child)
 {
	 return same_thread_group(child->parent, child->real_parent);
 }

So, here, we're saying that if child->parent and child->real_parent
are in the same thread group, they're gonna be duplicates, but in
do_notify_parent_cldstop(), we target tsk->group_leader->real_parent,
which should be the same as tsk->real_parent but confusing
nonetheless.  That's actually the reason why I used
child->group_leader->real_parent in the first place.

Thanks.

-- 
tejun