[Devel] Re: pspace child_reaper

[Cedric Le Goater]

Eric W. Biederman wrote:
> Cedric Le Goater <clg at fr.ibm.com> writes:
> 
>> Hello All,
>>
>> Eric, in your initial proof of concept on the pid namespace, you were
>> defining a child_reaper per pid namespace.
>>
>> IMO, we can't use init_task as a child_reaper in a pid namespace because we
>> will have pid collision which might result in a breakage of the init_task.
> 
> The kernel doesn't use init_task (The idle thread) once it starts
> init.  Reaping children is the job of pid == 1.

agree.

>> Here are some questions on the model you intended to follow :
>>
>> Do you think we should have a child_reaper task per container ?
> We have an init per container so yes.

hmm, have we always ? what if i don't start an /sbin/init process in my
newly created pid namespace or container. where do I collect all the SIGCHLD ?

>> Could we use a kthread to do the job ?
> Definitely not.

why ?

>> Could that kthread be global to all pid namespace ?
>
> Makes no sense.

if you don't have an init per container, we need to find someone for the job.

>> Any completely different idea on the topic ?
> Init reaps the children, and I believe there are parts of user space
> that depend on this.  We shouldn't change that semantic.

IMHO, the only semantic i see is in the kernel, which needs someone to take
care of sigchld. /sbin/init is a very good candidate bc it collects sigchld
anyway and discards the ones it doesn't know about.

C.

_______________________________________________
Containers mailing list
Containers at lists.osdl.org
https://lists.osdl.org/mailman/listinfo/containers