Re: [PATCH 09/10] netns: Enable tagging for net_class directories in sysfs

["Serge E. Hallyn" <serue@us.ibm.com>]

Quoting Greg KH (gregkh@suse.de):
> On Mon, Jun 02, 2008 at 03:46:08PM +0200, Benjamin Thery wrote:
> > net: Enable tagging for net_class directories in sysfs
> > 
> > The problem.  Network devices show up in sysfs and with the network
> > namespace active multiple devices with the same name can show up in
> > the same directory, ouch!
> > 
> > To avoid that problem and allow existing applications in network namespaces
> > to see the same interface that is currently presented in sysfs, this
> > patch enables the tagging directory support in sysfs.
> > 
> > By using the network namespace pointers as tags to separate out the
> > the sysfs directory entries we ensure that we don't have conflicts
> > in the directories and applications only see a limited set of
> > the network devices.
> 
> I don't like it how the network subsystem is starting to leach into the
> sysfs core here.  What happens when the next subsystem wants to do the
> same thing?  And then the next one?  Will they all have to do this kind
> of intrusive changes to sysfs?

At least as far as the tagging goes, each namespace which needs to do
this (network, devices, and user, at least) will add a field to the
tag structure and call sysfs_enable_tagging() on the relevant
directories.  So no more core sysfs changes should be needed, as Eric
tried to make this generic enough to be generally useful.

> Can't this be done only in the network subsystem?

sysfs/kobject layer has to somehow decide what to show for
/sys/class/net contents based on the mountpoint, right, so I don't see
how the network subsystem could do it.

The only non-tagging alternative I'd see would be to keep entirely
separate kobject pools for each namespace.  To do that we'd probably
want to break /sys/class/net into a separate fs that can be
remounted, so at least we don't have to keep the rest of the kobject
pools (/sys/firmware, kernel, etc) in sync...

Eric had mentioned before breaking /sys into multiple mountpoints, so
I'll assume the fact that he implemented tagging means that there was
too much cross-linking and whatnot across the /sys tree to make that
feasible.

More importantly, that approach would require more core sysfs changes
for the next namespace, whereas the tagging approach does not!

thanks,
-serge