[PATCH] dbus: add BSS changes notifications

[Witold Sowa]

Jouni Malinen pisze:
> On Mon, Jan 04, 2010 at 08:41:57PM +0100, Witold Sowa wrote:
>> Send PropertiesChanged signal from BSS object with dict of properties
>> that has changed and theirs values if any changes occurred.
> 
>> +static int are_ies_equal(const struct wpa_bss *old,
>> +		       const struct wpa_scan_res *new,
>> +		       u32 ie)
> 
>> +	switch (ie) {
>> +	case WPA_IE_VENDOR_TYPE:
>> +	case WPS_IE_VENDOR_TYPE:
>> +		old_ie = wpa_bss_get_vendor_ie(old, WPA_IE_VENDOR_TYPE);
>> +		new_ie = wpa_scan_get_vendor_ie(new, WPA_IE_VENDOR_TYPE);
> 
> I would assume the second argument was supposed to be 'ie' here (and
> same for RSN), i.e., the switch is just selecting between vendor IE and
> non-vendor IE.
> 
Yes, of course. A copy-paste error. Shame on me. :-(

>>  static void wpa_bss_update(struct wpa_supplicant *wpa_s, struct wpa_bss *bss,
>>  			   struct wpa_scan_res *res)
>>  {
>> +	notify_bss_changes(wpa_s, bss, res);
> 
> This call happens before the BSS entry is updated. This sounds a bit
> racy, i.e., none of the notification handlers would be able to fetch the
> new data from the BSS entry when processing the notification. That may
> work with the current D-Bus code due to the aggregation of signals, but
> it would not work some other things that I would expect to be using the
> notify.c callbacks. In other words, this should most likely do the
> comparison first (and say, store a bitfield of changed values) and then
> call the notification functions only after having updated the BSS entry.
> 
Agreed. I didn't take under consideration the fact that DBus won't be
the only beneficent of notifications callbacks forever.

Do you want me to resend the patch corrected?

Witek.