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
