Re: Segment fault when using wireless (adhoc mode)

[Dan Williams <dcbw redhat com>]

On Fri, 2009-03-06 at 11:51 +0800, 陈杰 wrote:
>         These cause multiple schedules of the rest of the activation
>         process,
>         which clearly shouldn't happen.  It looks like the root cause
>         is that
>         the supplicant is sending reconnect events too fast for NM to
>         handle; if
>         stage3 is already scheduled, NM shouldn't schedule it again,
>         it should
>         simply ignore the event.  Any chance you could get some timing
>         information to see how long it is between calls of
>         supplicant_iface_connection_state_cb_handler() ?
> Fine, I replaced nm_info with nm_debug in the file "nm-device-wifi.c,
> line 2353", and ran NM in daemon mode.
>  
> After upgrading wpa_supplicant from 0.6.3 to 0.6.4, it seems to be
> easier to reproduce this problem (see attachment "adhoc_crash_syslog"
> and "adhoc_crash_wpa_supplicant.log")

Interesting.  Three observations:

1) NM has that bug I talked about where it can schedule the next stage
twice.  That I'll have to fix.

2) "Could not read SSID from driver."
     Seems wrong; whenever the driver is IFF_UP it should be returning
valid values (which of course can be a blank SSID to indicate "no SSID
has been set") instead of an error; the supplicant will only print this
out when the driver returned an error when asking for the SSID.

3) "Associated with 00:00:00:00:00:00"
	*Seriously* wrong.  Looks like you're using a custom supplicant driver
which is pretty broken.  It's sending EVENT_ASSOC with an all-zero BSSID
when it should probably be sending EVENT_DISASSOC.  EVENT_ASSOC should
*never* have an all-zero BSSID.  That would imply you've associated with
nothing.  All-zero BSSIDs coming from drivers via the WEXT SIOGIWAP
event always mean "disassociated".

Dan