[Ocfs2-devel] [PATCH] [ocfs2] Use a kobject instead of a kset

[dann frazier]

On Fri, Feb 04, 2011 at 09:58:01AM -0800, Joel Becker wrote:
> On Fri, Feb 04, 2011 at 09:40:27AM -0700, dann.frazier at canonical.com wrote:
> > From: dann frazier <dann.frazier at canonical.com>
> > 
> > We currently allocate a kset and add attributes to its internal kobject.
> > However, Documentation/kobject.txt says only the kset code should manipulate
> > this internal object, and we can get the same results by just creating our
> > own kobject and avoiding a kset altogether.
> 
> 	Hmm, is this actually valid?  I thought you had to be a kset to
> have children.  Did this change at some point?

hey Joel,
 Well, I'm by no means an expert on kobjects - hopefully someone who
is will chime in. I came across this while debugging a sysfs teardown
issue and looking through this code to see if it might be related. I
read through kobject.txt & saw that note about a kset's kobj being for
internal-use only, and then saw that samples/kobject/kobject-example.c
seems to achieve the same goal (a sysfs dir w/ attributes) w/o using a
kset.

The sysfs entries look the same to me before & after this patch:

$ ls -l /sys/fs/ocfs2
total 0
-r--r--r-- 1 root root 4096 Feb  5 11:40 active_cluster_plugin
-rw-r--r-- 1 root root 4096 Feb  5 11:40 cluster_stack
-r--r--r-- 1 root root 4096 Feb  5 11:40 loaded_cluster_plugins
-r--r--r-- 1 root root 4096 Feb  5 11:40 max_locking_protocol

That said, I haven't been able to demonstate that the current code is
causing any actual problems - this was just a cleanup attempt.