[PATCH 1/2] hostapd: dfs: allow switch to available channel

[Janusz Dziedzic]

pt., 9 paź 2020 o 15:20 Janusz Dziedzic <janusz.dziedzic at gmail.com> napisał(a):
>
> pt., 9 paź 2020 o 14:11 Jouni Malinen <j at w1.fi> napisał(a):
> >
> > On Fri, Oct 09, 2020 at 01:44:32PM +0200, Janusz Dziedzic wrote:
> > > pt., 9 paź 2020 o 10:37 Jouni Malinen <j at w1.fi> napisał(a):
> > > >
> > > > On Sat, Oct 03, 2020 at 02:00:12PM +0200, Janusz Dziedzic wrote:
> > > > > For EU, where preCAC is allowed, we should
> > > > > allow switch to DFS available channels, instead
> > > > > of restart BSS.
> > > >
> > > > Could you please clarify how this is supposed to work in regulatory
> > > > domains that do not allow previously done CAC results to be used? The
> > > > test case added in 2/2 seems to pass even if running the AP with country
> > > > US or CA..
> > >
> > > Test itself choose/force EU country and expect AP-CSA-FINISHED
> > > +        # Toggle regulatory - clean all preCAC
> > > +        hostapd.cmd_execute(apdev[0], ['iw', 'reg', 'set', "US"])
> > > +
> > > +        hapd = start_dfs_ap(apdev[0], country="PL")
> >
> > And that "PL" is indeed what I replaced with "CA" and "US" to confirm
> > that patch 1/2 did not enable something that it shouldn't enable..
> > However, the test was still passing when the country code was pointing
> > to non-ETSI country. Can you please clarify why this is the case and how
> > does this protect expected functionality (new CAC required) outside the
> > ETSI regulatory domain?
> >
> Seems I see what is wrong. This is mainly because we cache channels state
> in hostapd and my code do fast chan_switch:
>
> CHAN_SWITCH 5 5180 ht
> CHAN_SWITCH 5 5260 ht
>
> and we get late DFS-PRE-CAC-EXPIRED
>
> When add 2 second sleep between chan_switch - fail correctly:
> @@ -662,7 +662,7 @@ def test_dfs_chan_switch_precac(dev, apdev):
>          # Toggle regulatory - clean all preCAC
>          hostapd.cmd_execute(apdev[0], ['iw', 'reg', 'set', "US"])
>
> -        hapd = start_dfs_ap(apdev[0], country="PL")
> +        hapd = start_dfs_ap(apdev[0], country="US")
>
>          ev = wait_dfs_event(hapd, "DFS-CAC-COMPLETED", 70)
>          if "success=1" not in ev:
> @@ -692,6 +692,8 @@ def test_dfs_chan_switch_precac(dev, apdev):
>          if freq != "5180":
>              raise Exception("Unexpected frequency")
>
> +        time.sleep(2)
> +
>
> 192.168.0.11/wlp2s0: <3>DFS-CAC-START freq=5260 chan=52 sec_chan=0,
> width=0, seg0=0, seg1=0, cac_time=60s
> 192.168.0.11/wlp2s0: CTRL: STATUS
> 192.168.0.11/wlp2s0: <3>DFS-CAC-COMPLETED success=1 freq=5260
> ht_enabled=1 chan_offset=0 chan_width=1 cf1=5260 cf2=0
> 192.168.0.11/wlp2s0: <3>AP-ENABLED
> 192.168.0.11/wlp2s0: CTRL: STATUS
> 192.168.0.11/wlp2s0: CTRL: CHAN_SWITCH 5 5180 ht
> 192.168.0.11/wlp2s0: <3>CTRL-EVENT-STARTED-CHANNEL-SWITCH freq=5180
> ht_enabled=1 ch_offset=0 ch_width=20 MHz cf1=5180 cf2=0 dfs=0
> 192.168.0.11/wlp2s0: <3>CTRL-EVENT-CHANNEL-SWITCH freq=5180
> ht_enabled=1 ch_offset=0 ch_width=20 MHz cf1=5180 cf2=0 dfs=0
> 192.168.0.11/wlp2s0: <3>AP-CSA-FINISHED freq=5180 dfs=0
> 192.168.0.11/wlp2s0: CTRL: STATUS
> 192.168.0.11/wlp2s0: CTRL: CHAN_SWITCH 5 5260 ht
> 192.168.0.11/wlp2s0: <3>DFS-PRE-CAC-EXPIRED freq=5260 ht_enabled=0
> chan_offset=0 chan_width=0 cf1=5260 cf2=0
> 192.168.0.11/wlp2s0: <3>AP-DISABLED
> 192.168.0.11/wlp2s0: <3>DFS-CAC-START freq=5260 chan=52 sec_chan=0,
> width=0, seg0=0, seg1=0, cac_time=60s
> 192.168.0.11/wlp2s0: CTRL: DISABLE
>
> Should I send patch with 2 seconds delay between chan_switch and add
> negative test also?
>
> But for the future shouldn't we always read channels state from
> kernel/driver/lower_layer?
> Each time we get some EVENT/CTRL re-read channel states?
>
Did fast check in cfg80211 code and seems we:

queue_delayed_work(cfg80211_wq, &rdev->dfs_update_channels_wk, 0);
send(NL80211_CMD_CH_SWITCH_NOTIFY)

Unfortunately delayed work recalc also dfs_state and send PRE-CAC-EXPIRED
First we get CH_SWITCH_NOTIFY and next PRE-CAC-EXPIRED..

More, we can run another chan_switch before set correct dfs_state
(clean CAC state in our case).
Seems like a cfg80211 bug, while we should send CH_SWITCH_NOTIFY after
dfs_state recalc.

BR
Janusz