dhcpcd-discuss

Iain wrote:
> Roy Marples wrote:
>
> > My take on this is that it's a driver bug. If querying an interface 
> > returns ENODEV then dhcpcd - and other software - will rightly think 
> > it no longer exists.
>
> Well yes, however it looks like a deliberate design decision. It went in 
> here:
> <http://git.us.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=e655b9f03f41c7a84fb74d6619abf844d7f2ab65> 
>
>
> I'm not entirely sure, but it looks like they're trying to avoid 
> firmware load and various other suspend related things that are 
> effectively useless when rfkill has been enabled.

OK, hooking into rfkill is the wrong thing todo.
However, if the switch is toggled, the kernel should emit netlink 
messages to indicate this.
Can you see if this is the case? if-linux.c, lines 1-310 are of interest.

> > I would be interested in seeing dhcpcd output running in a terminal
> > dhcpcd -dB
> >
> > and then the rkwill switch is toggled a few times, about 10 secs 
> > between each toggle. Also, test starting dhcpcd with the switch on and 
> > then off.
> > If you could email back with those results, it should be interesting :)
>
> I'll grab some details tomorrow. There's not really much to see though, 
> as far as I can tell dhcpcd just ignores the interface due to that 
> ENODEV when it first probes for available network cards.

Ah, but what I'd like to see is this .....

dhcpcd reacts to new interface link messages sent via the kernel.
So if ifconfig UP fails with ENODEV, then I would logically expect a 
kernel announcement that the interface is available when the rfkill 
switch is toggled to indicate the interface is now ready.

Thanks

Roy