CHROMIUMOS: wpa_supplicant: add fast reconnect support

CHROMIUMOS: wpa_supplicant: add fast reconnect support

When dropped by an AP for inactivity or because the session has timed
out, try to re-join directly before falling back to a full scan and
the normal work.  This reduces the time to re-join a network from 5-10
seconds to ~300ms + time to renew DHCP.

Note we cannot directly re-authenticate because by the time we issue
a request mac80211 has discarded the BSS state which is required to
process an MLME request.  Instead we scan the channel where we last
saw the AP and then use the normal join mechanism. Since this is not
true for all drivers it might be worthwhile to try a direct Auth first
and only then fall back to scanning.

BUG=chromium-os:10733
TEST=Roaming + manual tests

Committed: http://chrome-svn/viewvc/chromeos?view=rev&revision=cea5703

[Paul Stewart, 2011-02-19 05:47:13]

LGTM.  Nice work.

[Paul Stewart, 2011-02-19 06:12:19]

Just as I was walking away, I thought about a few things...  Let me ~LGTM for a
second.

http://codereview.chromium.org/6538036/diff/1/wpa_supplicant/events.c
File wpa_supplicant/events.c (right):

http://codereview.chromium.org/6538036/diff/1/wpa_supplicant/events.c#newcode953
wpa_supplicant/events.c:953: } else if (wpa_s->fast_reconnect) {
If the scan succeeds (selected != 0), we connect, but if wpa_s->fast_reconnect
is non-zero, it stays that way until the next scan failure.  Is this a desired
side-effect?  Probably not.

http://codereview.chromium.org/6538036/diff/1/wpa_supplicant/events.c#newcode962
wpa_supplicant/events.c:962: bgscan_deinit(wpa_s);
You should not do this here.  bgscan should have been disabled before the scan.

http://codereview.chromium.org/6538036/diff/1/wpa_supplicant/events.c#newcode...
wpa_supplicant/events.c:1353: if (fast_reconnect) {
I think you should disable bgscan even in the fast_reconnect case.  Otherwise
you run the risk of interfering with yourself.  bgscan will come back once the
connection completes.

[Sam Leffler, 2011-02-22 21:35:26]

re-ran Roaming suite + doing manual tests now; won't commit before more manual
testing

http://codereview.chromium.org/6538036/diff/1/wpa_supplicant/events.c
File wpa_supplicant/events.c (right):

http://codereview.chromium.org/6538036/diff/1/wpa_supplicant/events.c#newcode953
wpa_supplicant/events.c:953: } else if (wpa_s->fast_reconnect) {
Good catch; fixed.

http://codereview.chromium.org/6538036/diff/1/wpa_supplicant/events.c#newcode962
wpa_supplicant/events.c:962: bgscan_deinit(wpa_s);
Done

[Paul Stewart, 2011-02-22 21:56:25]

LGTM modulo completion of tests.  One nit you can choose to ignore.

http://codereview.chromium.org/6538036/diff/6001/wpa_supplicant/events.c
File wpa_supplicant/events.c (right):

http://codereview.chromium.org/6538036/diff/6001/wpa_supplicant/events.c#newc...
wpa_supplicant/events.c:1289: bgscan_deinit(wpa_s);
Nit: Any reason why instead of adding these two lines here you don't just move
lines 1365,1366 above the "if (fast_reconnect)" conditional?  Seems a shame to
duplicate.

[Sam Leffler, 2011-02-22 22:11:31]

I worried the order was important so figured I'd wait for upstream  to optimize
(it's likely to come back different anyway).