Oleg Nesterov wrote:
> On 08/06, Pavel Emelyanov wrote:
>> Oleg Nesterov wrote:
>>> On 07/26, Pavel Emelyanov wrote:
>>>> The reason to release namespaces after reparenting is that when task
>>>> exits it may send a signal to its parent (SIGCHLD), but if the parent
>>>> has already exited its namespaces there will be no way to decide what
>>>> pid to dever to him - parent can be from different namespace.
>>> I almost forgot about this one...
>>>
>>> After reading the whole series, I can't understand the above explanation
>>> any longer. The parent can't be from different namespace, either we have
>>> another sub-thread, or we reparent the child to /sbin/init which should
>>> be from the same namespace.
>> If the child that is a new namespace's init is exiting its parent is from the
>> different namespace.
> 
> In that case it doesn't have childs. The were SIGKILL'ed before exit_notify().

It does not, but it's parent - does :)

>> Moreover, we will probably want to implement "entering"
>> the pid namespace, so  having tasks with parents from another namespace will
>> be OK.
> 
> Well. I saw this word "entering", but I don't know the meaning. Just curious,
> could you explain?

"Entering" means "moving task to arbitrary namespace"

> And, if an exiting task has a child which is already from another namespace,
> why can't we release our namespace before re-parenting? I guess I need to
> know what "entering" means to understand this...

One of the desired actions was the following:
1. task X clones the new namespace with the child Y as this namespace's init;
2. task X waits for SIGCHILD to come informing that the namespace is dead.
In this scenario we need to set the Y's pid as it is seen from X's 
namespace in siginfo.

> Oleg.
> 
>