Re: [PATCH 2/4] fs: define a firmware security filesystem named fwsecurityfs

From: Greg Kroah-Hartman
Date: Mon Nov 21 2022 - 10:12:33 EST


On Mon, Nov 21, 2022 at 09:03:18AM -0500, James Bottomley wrote:
> On Mon, 2022-11-21 at 12:05 +0100, Greg Kroah-Hartman wrote:
> > On Sun, Nov 20, 2022 at 10:14:26PM -0500, James Bottomley wrote:
> > > On Sun, 2022-11-20 at 17:13 +0100, Greg Kroah-Hartman wrote:
> > > > On Sat, Nov 19, 2022 at 01:20:09AM -0500, Nayna wrote:
> > > > >
> > > > > On 11/17/22 16:27, Greg Kroah-Hartman wrote:
> > > > > > On Mon, Nov 14, 2022 at 06:03:43PM -0500, Nayna wrote:
> > > > > > > On 11/10/22 04:58, Greg Kroah-Hartman wrote:
> > > [...]
> > > > > > > > I do not understand, sorry.  What does namespaces have to
> > > > > > > > do
> > > > > > > > with this?
> > > > > > > > sysfs can already handle namespaces just fine, why not
> > > > > > > > use
> > > > > > > > that?
> > > > > > > Firmware objects are not namespaced. I mentioned it here as
> > > > > > > an
> > > > > > > example of the difference between firmware and kernel
> > > > > > > objects.
> > > > > > > It is also in response to the feedback from James Bottomley
> > > > > > > in
> > > > > > > RFC v2 [
> > > > > > > https://lore.kernel.org/linuxppc-dev/41ca51e8db9907d9060cc38ad
> > > > > > > b59a66dcae4c59b.camel@xxxxxxxxxxxxxxxxxxxxx/].
> > > > > > I do not understand, sorry.  Do you want to use a namespace
> > > > > > for
> > > > > > these or not?  The code does not seem to be using
> > > > > > namespaces. 
> > > > > > You can use sysfs with, or without, a namespace so I don't
> > > > > > understand the issue here.
> > > > > >
> > > > > > With your code, there is no namespace.
> > > > >
> > > > > You are correct. There's no namespace for these.
> > > >
> > > > So again, I do not understand.  Do you want to use filesystem
> > > > namespaces, or do you not?
> > >
> > > Since this seems to go back to my email quoted again, let me
> > > repeat: the question isn't if this patch is namespaced; I think
> > > you've agreed several times it isn't.  The question is if the
> > > exposed properties would ever need to be namespaced.  This is a
> > > subtle and complex question which isn't at all explored by the
> > > above interchange.
> > >
> > > > How again can you not use sysfs or securityfs due to namespaces? 
> > > > What is missing?
> > >
> > > I already explained in the email that sysfs contains APIs like
> > > simple_pin_... which are completely inimical to namespacing.
> >
> > Then how does the networking code handle the namespace stuff in
> > sysfs?
> > That seems to work today, or am I missing something?
>
> have you actually tried?
>
> jejb@lingrow:~> sudo unshare --net bash
> lingrow:/home/jejb # ls /sys/class/net/
> lo tun0 tun10 wlan0
> lingrow:/home/jejb # ip link show
> 1: lo: <LOOPBACK> mtu 65536 qdisc noop state DOWN mode DEFAULT group
> default qlen 1000
> link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
>
> So, as you see, I've entered a network namespace and ip link shows me
> the only interface I can see in that namespace (a down loopback) but
> sysfs shows me every interface on the system outside the namespace.

Then all of the code in include/kobject_ns.h is not being used? We have
a whole kobject namespace set up for networking, I just assumed they
were using it. If not, I'm all for ripping it out.

thanks,

greg k-h