[Qemu-devel] [RFC PATCH 0/2] Decouple netdev from nic

[Vasilis Liaskovitis]

Virtualization management layers would like to decouple host network device
(netdev) creation/deletion/setup from guest network adapters.

This would allow to independently:
- provision VMs with NIC(s)
- create and manage network devices on the host

First discussed here:
http://lists.gnu.org/archive/html/qemu-devel/2011-10/msg03614.html

Explicit plug/unplug of a backend netdev device to a VM NIC at runtime would be
needed.  This patch introduces a command "netdev_set" to attach a netdev device
to a guest nic. This allows the following use cases:

1) qemu can start with detached nic and backend netdevs
E.g. at boot time nic0 and netdev are not connected:

-netdev type=tap,id=netdev0 -device virtio-net-pci,id=nic0
Warning: netdev guest0 has no peer
Warning: nic net0 has no peer

The following sequence would connect backend device netdev0 to nic nic0

(qemu) set_link nic0 off
(qemu) netdev_set nic0 netdev0
(qemu) set_link nic0 up

Setting the link up/down is done separately with the existing set_link command,
but could be added to the semantics of the new command if desired.

2) change of backend netdev at runtime. 
E.g. at boot time, nic0 is connected to netdev netdev0:

-netdev type=tap,id=netdev0 -device virtio-net-pci,netdev=netdev0,id=nic0

The following sequence would change the backend device for nic0 to netdev1:

(qemu) set_link nic0 off
(qemu) netdev_del netdevnic0
(qemu) netdev_add id=netdev1,type=tap
(qemu) netdev_set nic0 netdev1
(qemu) set_link nic0 up

(netdev_del will currently call the link_status_changed callback for the guest
 device it is attached to, so afaict the first step "set_link off" for the guest
 device is not strictly needed in this example)

- is this feature of interest upstream? If yes, what's the desired way to 
implement
it in qom? Is there a legitimate way to change a qdev property after a device 
has
been initialized? The netdev property of a guest nic is set at nic
initialization time - if we attach a new backend device, how can we update the
netdev property?

- A link property for a nic device (link=up/down) which denotes initial status 
of a
nic can also be implemented (not yet in this patch). This way a nic could be
initialized with disabled carrier at the command line using link=down.

thoughts, comments welcome.

series is based on 
http://repo.or.cz/w/qemu/qmp-unstable.git/qmp-wip/qapi-commands-conv/netdev-add/v2
and can be found here:
https://github.com/vliaskov/qemu-kvm/commits/netdev-decouple

Vasilis Liaskovitis (2):
  Implement netdev_set command
  virtio-net: re-initialize tap device for new netdev

 hmp-commands.hx  |   14 +++++++++++++
 hmp.c            |   10 +++++++++
 hmp.h            |    1 +
 hw/virtio-net.c  |   30 +++++++++++++++++++++++++++
 net.c            |   58 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 net.h            |    1 +
 qapi-schema.json |   16 ++++++++++++++
 qmp-commands.hx  |   25 +++++++++++++++++++++++
 8 files changed, 155 insertions(+), 0 deletions(-)

-- 
1.7.9