Re: Braindump on wpa_supplicant

[Dan Williams <dcbw redhat com>]

On Tue, 2 Aug 2005, Olivier Blin wrote:
> Dan Williams <dcbw redhat com> writes:
> 
> > Well, I'm talking about the whole thing, where we get to modify wpa_supplicant 
> > for slave mode and what not as well.  We don't want to simply let wpa_supplicant 
> > loose, since _NetworkManager_ is the program that's determining which network to 
> > connect to, what password/passphrase/key to use, what certificates
> > to use, etc.  
> 
> This should be possible with current wpa_supplicant, since you can
> restrain it on a specific network ID.

Well, I wasn't planning on launching a new wpa_supplicant _every_ time I want to 
connect to a network.  That's a bit bogus.  What I'd like to do is to launch a 
wpa_supplicant either when NM starts up, or if it can't be helped, each time an 
interface is plugged in or noticed.  I'd like to run as few instances of 
wpa_supplicant as few times as possible.

Even if wpa_supplicant can be restricted to a specific network, I do not want to 
write out a config file to tell it so.  I need to be able to tell wpa_supplicant 
what to do over its control socket.  The only command I want is "connect to this 
ESSID with this passphrase and/or this certificate".  I do not want 
wpa_supplicant to attempt to do any "smart" things by itself, since that's what 
NetworkManager is for.

> > And I'm not going to write out a config file for wpa_supplicant, that's just 
> > bogus.  This stuff needs to be tightly integrated, not writing config files for 
> > stuff all over the place.  The only reason we do that for 
> > named/caching-nameserver is because it wasn't dbus-enabled yet at the time we 
> > did the VPN stuff (it has since been made dbus-aware).
> 
> So, will you try to make wpa_supplicant dbus-aware?
> The current socket interface is quite fine, but only one program can
> listen to events.

Well, that would be nice, but I don't think that would be easy to get upstream.  
Upstream (ie, Jouni) seems to want wpa_supplicant to do everything under the 
sun, and to simply have NetworkManager be a dumb front-end that writes out a 
config file and points wpa_supplicant at the config file.  If he would accept a 
dbus patch, great, I'd love to write it.

Dan