On Mon, Feb 19, 2018 at 06:49:30PM +0300, Dmitrii Lebed wrote:
> If CAC is aborted, hostapd will continue wait for
> CAC completion and will stuck in this state forever.
> Adding CAC restart action on CAC aborted event,
> considering "CAC aborted" state as recoverable.
> CAC aborted can be generated in some complex configs,
> e.g. in repeater config, when wpa_supplicant is
> doing something on another virtual interface that
> can lead to generation of "CAC aborted" event.
Please note that nl80211 drivers will indicate CAC aborted if a radar is
detected during the CAC. It does not sound correct to try to restart CAC
in such cases since the channel can clearly not be usable.
> diff --git a/src/ap/dfs.c b/src/ap/dfs.c
> @@ -636,6 +637,34 @@ static unsigned int dfs_get_cac_time(struct hostapd_iface *iface,
> return cac_time_ms;
> }
>
> +static int dfs_start_cac(struct hostapd_iface *iface)
> +{
> + int res;
This is whitespace damaged (tabs seemed to be converted to spaces) which
makes it quite inconvenient to try to apply this.
> @@ -800,9 +831,21 @@ int hostapd_dfs_complete_cac(struct hostapd_iface *iface, int success, int freq,
> + } else {
So this is the !success case which is also entered if a radar is
detected during CAC.
> + if (dfs_offload) {
> + if (iface->state == HAPD_IFACE_ENABLED)
> + iface->cac_started = 0;
> + } else if (iface->state == HAPD_IFACE_DFS) {
> + /* Schedule CAC restart in 1 second */
> + eloop_register_timeout(1, 0, hostapd_dfs_cac_restart,
> + iface, NULL);
That eloop timeout was not canceled anywhere. There needs to be some
protection against the interface being removed during the wait since
otherwise there would be a callback coming in with a pointer to freed
memory.
Where does this one second time come from? Why not immediately? Or
should the extra time be used to wait to see if NL80211_CMD_RADAR_DETECT
is received and if so, cancel this restarting attempt of CAC?
> -int hostapd_dfs_start_cac(struct hostapd_iface *iface, int freq,
> +int hostapd_dfs_cac_started(struct hostapd_iface *iface, int freq,
That renaming broke wpa_supplicant build. I'd leave this renaming to a
separate patch (also covering the needed wpa_supplicant/ap.c change) or
maybe just drop the renaming completely to avoid unnecessary changes.
--
Jouni Malinen PGP id EFC895FA
_______________________________________________
Hostap mailing list
Hostap@xxxxxxxxxxxxxxxxxxx
http://lists.infradead.org/mailman/listinfo/hostap
