WEP and enc restricted

Hi,

I am pretty sure that the mac80211 code doesn't handle WEP with shared keys correctly. Using the ath9k driver on my MacBook I am unable to put the driver into restricted mode (shared key). It stays in open mode. That is the following command has no effect: iwconfig wlan0 enc restricted. I can also confirm this when looking at the authentication packets that go over the air: they are sent with the open flag, and not the shared key flag.

Surprisingly nowhere in the mac80211 code the ifsta->auth_alg is set to WLAN_AUTH_SHARED_KEY so far. The attached small patch does just that. I am not 100% sure if that's the way to do it, so please review carefully.

With this patch authentication works. The AP now denies association, but that is another problem.

Cheers

Ortwin

@@ -929,6 +929,14 @@
 	else
 		idx--;
 
+	if (sdata->vif.type == IEEE80211_IF_TYPE_STA) {
+		struct ieee80211_if_sta *ifsta = &sdata->u.sta;
+		if (erq->flags & IW_ENCODE_OPEN)
+			ifsta->auth_alg = WLAN_AUTH_OPEN;
+		else if (erq->flags & IW_ENCODE_RESTRICTED)
+			ifsta->auth_alg = WLAN_AUTH_SHARED_KEY;
+	}
+
 	if (erq->flags & IW_ENCODE_DISABLED)
 		remove = 1;
 	else if (erq->length == 0) {

d keys=20

e driver=20

following=20

irm this=20

are sent=20

It looks like ieee80211_rx_mgmt_auth() in mlme.c should be cycling through the various auth methods when it gets an "auth not supported" response from the AP. I wonder why that's not working in your case. Can you put some printks in there and see if that code gets triggered and if not, poke around a bit to see why?

to=20

I am not=20

It looks like it does get set from the SIWAUTH handler if you pass in the right auth algs bits; but you're correct in that SIWENCODE won't se= t
the auth alg. The patch seems to be correct.

However, the submission format needs a few things: the signed-off-by line, and the patch should be inlined rather than attached to make for easier review. Please see here:

http://linux.yyz.us/patch-format.html

Thanks!
Dan

but that is=20

Yes, I've seen that. I'll have to check the packet log again, but as far as I remember the AP doesn't respond at all to these authentication packets. So the cycling will never happen.

In SIWAUTH only the available algorithms are set, not the actual one. It's a different variable:
sdata->u.sta.auth_alg*s* = data->value;

I'll resend the patch in the correct format.

Cheers

Ortwin