[Devel] Re: [RFC][PATCH 1/5] Virtualization/containers: startup

Hubertus Franke frankeh at watson.ibm.com
Fri Feb 3 12:34:46 PST 2006


Greg KH wrote:
> On Fri, Feb 03, 2006 at 10:34:01AM -0800, Dave Hansen wrote:
> 
>>Lastly, is this a place for krefs?  I don't see a real need for a
>>destructor yet, but the idea is fresh in my mind.
> 
> 
> Well, what happens when you drop the last reference to this container?
> Right now, your patch doesn't cause anything to happen, and if that's
> acceptable, then fine, you don't need to use a struct kref.
> 
> But if not, then why have a reference count at all?  :)
> 
> thanks,
> 
> greg k-h
> 

Greg ...

In our pid virtualization patches we removed the object automatically.
I presume that Kirill does the same .... although the code for that is not released.

I am working on converting our pid stuff so to follow the recommendation of Linux/Alan
to just do <container,pid> isolation through pidspaces.

I could release an allocation/ before hand. Based on Kirill's 1/5 patch adoption
(which took Linus's & Dave's naming comments into account)

How do we want to create the container?
In our patch we did it through a /proc/container filesystem.
Which created the container object and then on fork/exec switched over.

How about an additional sys_exec_container( exec_args + "container_name").
This does all the work like exec, but creates new container
with name "..." and attaches task to new container.
If name exists, an error -EEXIST will be raised !

-- Hubertus




More information about the Devel mailing list