[pkg-wpa-devel] Bug#428304: Bug#428304: should reassociate when ifup fails

[Kel Modderman]

On Wed, 13 Jun 2007 09:38:40 am Elmar Hoffmann wrote:
> Hi Kel,
>
> on Mon, Jun 11, 2007 at 12:27:43 +1000, you wrote:
> > Could the patch use the return value of ifup rather than grep'ing the
> > ifstate file?
>
> I'm quite sure my first attempt at this did do the former, but that
> did not work, IIRC (its been at 1.5 months since I did this). I'm
> going check whether this is true and were the problem lies, if so.
>
> > Also, this patch has the potential to bing back the "re-association
> > storm" phenomenon that the wpa_hysteresis_{event,check} functions were
> > designed to prevent. This could lead to an uncontrollable loop of failed
> > re-association and failed ifup events.
>
> Hmm yes, this might be an issue with cases other than DHCP (which
> takes a while before it gives up) that fail immediately.
> OTOH in my experience wpasupplicant also takes a while to react to an
> reassociate request (and any normal disconnect and connect events
> inbetween would be covered by the hysteresis), so that loop should not
> be too fast and may not be a problem. I should be able to simulate and
> test that with a broken interface stanza or sth. :)

Broken drivers are the acid test. broadcom after suspend and madwifi 
association brokenness are two things that have burnt either myself or others 
in the past.

>
> Still it is (unlike the hysteresis case) quite logical to not stay
> associated to an AP that is not usable as your interface did not
> configure. And while trying over and over again arguably does not help
> in a "one AP with broken DHCP" situation, it sure helps when there are
> more APs wpasupplicant can connect to or when the wireless connection
> is bad enough for DHCP to fail sometimes due to loss of DHCP packets
> but another attempt could work. The latter is the case that made me do
> these changes.

I agree with the logic. However, I'm not going to risk locking up ifupdown or 
causing system instability because the wifi interface gets stuck in an 
infinite loop of re-association. It has happened to me before. Sure, I was 
using a rather experimental driver (madwifi) but please name one linux wifi 
kernel driver that _always_ does the right thing these days.

This is definitely an area that can be improved in this roaming scheme, I just 
don't know how far these sh scripts should extend themselves when other more 
comprehensive network management suites are in development . . . 

Thanks, Kel.