On Tue, 2009-11-03 at 01:17 +0200, Maxim Levitsky wrote:

> > There should be no need for doing both disassoc and deauth; just send
> > deauth only if that is needed. This will save one extra frame
> > transmission and speeds up roaming a bit.
> > 
> > Though, is this enough to handle the roaming cases where wpa_supplicant
> > may not try to send either disassociation or deauthentication?

> Currently wpa_supplicant sends only disassociation.
> deauthentication is send only in rare cases, usually due to [suspected]
> authentication error.
> Other that that I don't know, what I need is a clear statement about how
> things should work that is:
> 
> 
> * Should kernel allow authentication while in authenticated? I guess yes
> To same AP? To different APs?

It does allow it to different APs (up to 3/4 at a time), but not to the
same AP -- I plan on fixing the latter case.

> * What should kernel do if it done authentication to several APs?, but
> not association.
> should it timeout, or let wpa_suplicant do it?
> Currently it allows 4 (or 3) such APs, and then then bugs out with
> -ENOSPC

IMHO it should not make the decision on how long authentications are
valid. Any such value might be wrong if wpa_supplicant is really slow,
say it's swapped out.

> * Should kernel allow scanning while in authenticated but not associated
> case? 
> I have send patch to do so, I hope it will be accepted.

I don't think it should, and as such don't think I want to accept the
patch. You need to make a much much better case for it if you want me to
accept it. I see no reason to ever be in a situation where there are
authentications (that aren't either killed off or used to associate) for
a long enough period of time for it to make sense to scan in that time.

johannes