VLAN creation may lead to deadlock.

Hi,

I stumbled upon the following annoying situation: supposing eth0 is some
interface that is currently DOWN, the following command

  ip link add link eth0 name eth0.1 up type vlan id 1

never returns and the kernel printks the following lines every few seconds

  unregister_netdevice: waiting for eth0.1 to become free. Usage count = 2

I understand that the kernel should prevent me from creating a VLAN interface
that's initially up when loose-binding is off and the master interface is down
(and indeed there is absolutely no problem when loose-binding is on), but the
unregistering of the currently-being-registered device should not block like
that.  It appears there are too few dev_puts compared to dev_holds in there.

Regards,

Ignacy

-- 
Ignacy Gawędzki
R&D Engineer
Green Communications

On Fri, Feb 13, 2015 at 4:02 AM, Ignacy Gawedzki
[off-list ref] wrote:

Hi,

I stumbled upon the following annoying situation: supposing eth0 is some
interface that is currently DOWN, the following command

  ip link add link eth0 name eth0.1 up type vlan id 1

never returns and the kernel printks the following lines every few seconds

  unregister_netdevice: waiting for eth0.1 to become free. Usage count = 2

I understand that the kernel should prevent me from creating a VLAN interface
that's initially up when loose-binding is off and the master interface is down
(and indeed there is absolutely no problem when loose-binding is on), but the
unregistering of the currently-being-registered device should not block like
that.  It appears there are too few dev_puts compared to dev_holds in there.

Good catch! It looks like we forgot to call unregister_vlan_dev(),
I am cooking a patch.