Virtual Interface for P2P operations

[Johannes Berg]

On Wed, 2011-07-13 at 10:41 -0400, Pavel Roskin wrote:
> On 07/13/2011 03:19 AM, Johannes Berg wrote:
> > On Tue, 2011-07-12 at 18:55 -0700, Jithu Jance wrote:
> >> Thanks Jouni for your feedback. Agree with your suggestion that using
> >> primary mac address for sending Tx NL_CMD_FRAME looks ugly and is not
> >> the correct way to do it. May be exposing a network interface for p2p
> >> device address interface would be a better approach and should work
> >> with current supplicant. I will try this way.  Thanks again for your
> >> valuable suggestions and sorry for the trouble.
> >
> > I think the correct way to handle this, short of exposing a separate
> > netdev, would be to expose new nl80211 "start p2p" and "stop p2p"
> > operations that wpa_s could use. These might internally create a (maybe
> > hidden) netdev to use with the right operations.
> >
> > Having the driver just expose an interface isn't really a good solution
> > imho, since for example its lifetime wouldn't be controlled the right
> > way.
> 
> Can we use socket options?  As long as the socket is open, the hidden 
> netdev exists.  When the socket closes for whatever reason, the hidden 
> netdev is removed.

Not really with netlink.

> This way, the socket is remains bound to a specific vif.  It just 
> becomes a socket with special properties.

That's still easy to do with netlink sockets, see code we already have
for instance for frame subscriptions.

johannes