[Lxc-users] Failed to move phys interface, invalid argument (Linux 3.1.0, lxc 0.7.5)

[Daniel Smith]

On Mon, Nov 21, 2011 at 4:50 PM, Ivan Vilata i Balaguer
<ivan at selidor.net> wrote:
> Indeed it does, thanks!  I didn't need to run iw under the container, it
> appeared there instantly after moving the interface from the host.  I don't
> know what you mean by adding vifs to the moved interface, but thanks for the
> advice. :)

So when I say wiphy I am referring to the phy0 for instance. When I
referred to a vif (virtual interface) I was meaning the network
interface attached to the phy, e.g. wlan0. With most mac80211 devices
has the capability to have multiple vifs.

As I recall the following steps would reliable lock up the system (and
really should be allowed to happen in the first place). Say we have a
host (host) and a container (cont1) with a ns pid 252.
 1. In host: iw phy0 set netns 252
 2. In cont1: iw phy0 interface add wlan0 type managed
 3. In cont1: ip l s wlan0 up
 4. In host: iw phy0 interface add wlan0 type monitor
 5. In host: ip l s wlan0 up
 6. In cont1: halt
 7. system locked

This was on a stock Ubuntu 11.04 system. I have been busy with other
problems so I have never gotten back to do a deeper analysis of the
issue.

> However, I think this should be automatically set up by the "lxc.network.type
> = phys" option, so if no one has a clear reason of why this isn't supported
> I'll file a new issue in the tracker.

Should this really be under phys type? Right now, if I am not
mistaken, phys is used to denote a physical network interface, e.g.
eth0, in this case the device is a PHY which could have one or more
network interfaces.

Daniel