BUG: weird behaviour after killing hostapd in latest CVS.

[Jouni Malinen]

On Wed, Aug 06, 2003 at 12:53:31PM -0400, Pavel Roskin wrote:

> I could reproduce the panic on Linux 2.6.0-test2-bk5 with CVS HostAP.  The
> sequence:
> 
> cardctl insert
> hostapd hostapd.conf
> ifconfig wlan0 up
> killall hostapd
> 
> It's not needed to bring wlan0sta up.  wlan0ap was up already.

Odd, I cannot reproduce this by setting wlan0 up.. However, I can
certainly produce it by setting wlan0sta up and then killing hostapd.
wlan0ap is not an issue, since hostapd is supposed to close it before
asking the driver to remove that interface. Anyway, I can reproduce this
with WDS links too, so there is something odd with the new netdev
unregistering code in Host AP driver. It is just not immediately clear
what is wrong.

> The panic has been captured on the serial console.  It's attached.  I'm
> not sure if I'll be able to debug it soon.
> 
> It appears that the crash happens very early in prism2_rx_80211.

Hmm.. Actually, in my tests this was a deadlock (of rtnl_lock?), not
NULL pointer dereference..

After some more debugging, it is looking like delayed unregisteration
function is called too early or something similar. I tried to fix this
by unregistering the netdev immediately when and just delaying kfree..
Of course I managed to produce deadlock that prevented even hostapd
killing without setting any interfaces up ;-). I'm too tired to even
think about this now, so I'll take a closer look tomorrow.

-- 
Jouni Malinen                                            PGP id EFC895FA