wpa_supplicant on bonding slave

[Jouni Malinen]

On Sat, Jan 31, 2009 at 01:50:21AM +0000, Brian J. Murrell wrote:

> I have configured wpa_supplicant successfully when the interface is not a 
> bonding slave and it works great.  But as soon as I try to move that 
> configuration to the slave wireless interface for whatever reason the 
> wireless fails to maintain a connection to the router.

Do you tell wpa_supplicant about the bonding device? I think that
bonding behaves similarly to bridging in the sense that the EAPOL frame
reception needs special handling and you will likely need to add
-b<bridge/bond ifname> to the wpa_supplicant command line.

Another potential problem is in which MAC address is actually used on
the bonded device. IEEE 802.11 does not allow client mode to transmit
using any other MAC address than the address of the card itself. In
other words you will need to make sure that the wireless interface MAC
address gets selected one way or another. I've never tested Linux
bonding, so I do not know what kind of tricks this may need.


> When I connect to the supplicant with wpa_cli this is what I see it doing 
> (with debug level set to 2 -- anything higher seems to disable debug at 
> all):

You should use wpa_supplicant debug output from stdout (start it
manually with -dd on command line) to get all debug messages. Lots of
the messages are not available through wpa_cli. Anyway, my guess would
be that wpa_supplicant was not seeing the EAPOL frames when bond0 is
used without information wpa_supplicant (-bbond0) about it.

> pairwise_cipher=CCMP
> group_cipher=TKIP
> key_mgmt=WPA2-PSK
> wpa_state=ASSOCIATED

> Which to my untrained eye looks like WPA negotiation succeeds but then
> for whatever reason it gets disconnected/disassociated.  No idea why
> though.

Nope.. ASSOCIATED is the state before WPA negotiation was started. This
changes when wpa_supplicant receives the EAPOL frame for the WPA
negotiation.

-- 
Jouni Malinen                                            PGP id EFC895FA