[Devel] Re: [PATCH 1/2][RFC] containers: improve automatic container naming

[Andrew Morton]

On Fri, 1 Jun 2007 16:48:09 -0500
"Serge E. Hallyn" <serue at us.ibm.com> wrote:

> This compiles and boots, but is not intended for inclusion in -mm (yet),
> just as an RFC for the naming scheme to fix the bug Andrew pointed out.
> 
> Seem ok overall?
> 
> thanks,
> -serge
> 
> >From 8e9b972f7482415777e982d3bc9a0d55cbaf862b Mon Sep 17 00:00:00 2001
> From: Serge E. Hallyn <serue at us.ibm.com>
> Date: Fri, 1 Jun 2007 15:32:15 -0400
> Subject: [PATCH 1/2] containers: improve automatic container naming
> 
> The automatic naming of containers created using container_clone()
> is currently broken (not protected from wraparound) and inconvenient.
> 
> Add a per-container counter for use in naming children of the container.
> Before two unshares in a row by one process, and a third in another,
> would result in
> 
> 	/node1/node2
> 	/node3
> 
> The current scheme should result in
> 
> 	/node1/node1
> 	/node2
> 
> Also, keep a hash table populated with used names, to protect
> against counter wrap-around.
> 
> ...
>
>  include/linux/container.h |    8 +++
>  kernel/container.c        |  116 +++++++++++++++++++++++++++++++++++++++++++--

gad, what's all this stuff?

I think an IDR tree would get you what you're after in much less code. 
Although it means that container IDs would get recycled quickly across a
remove+add.

Be aware that there are IDR enhancements in Greg's driver tree (and hence
in -mm) which are relevant to this application.


> +	if (cont->auto_cnt_set) {

Can we please stop using "cnt" and "cont" to refer to containers?  Let's
use "container", OK?