by Johannes Berg

[permalink] [raw]

Subject: Re: [PATCH 3/4] cfg80211: Join and Leave handling for OCB mode

On Thu, 2014-09-11 at 16:30 +0200, Rostislav Lisovy wrote:
> Signed-off-by: Rostislav Lisovy <[email protected]>

A bit more commit log would be good :)

> +++ b/include/net/cfg80211.h
> @@ -375,6 +375,8 @@ static inline enum nl80211_channel_type
> cfg80211_get_chandef_type(const struct cfg80211_chan_def *chandef)
> {
> switch (chandef->width) {
> + case NL80211_CHAN_WIDTH_5:
> + case NL80211_CHAN_WIDTH_10:
> case NL80211_CHAN_WIDTH_20_NOHT:
> return NL80211_CHAN_NO_HT;

This is wrong, the old nl80211_channel_type simply cannot express a 5 or
10 MHz width - so this case should WARN() or something and really never
ever be called here.

Also, it doesn't really belong into this patch?

> +struct ocb_setup {
> + struct cfg80211_chan_def chandef;
> +};

kernel-doc would be appreciated.

> @@ -892,6 +893,13 @@ cfg80211_get_chan_state(struct wireless_dev *wdev,
> *radar_detect |= BIT(wdev->chandef.width);
> }
> return;
> + case NL80211_IFTYPE_OCB:
> + if (wdev->chandef.chan) {
> + *chan = wdev->chandef.chan;
> + *chanmode = CHAN_MODE_EXCLUSIVE;
> + return;
> + }

same as in the mac80211 patch - I don't see why this should be
exclusive.

> + WARN_ON(!setup->chandef.chan);

if (WARN_ON(...))
return -EINVAL;

would be better

johannes

2014-10-16 17:20:08

by Rostislav Lisovy

[permalink] [raw]

Subject: Re: [PATCH 1/4] mac80211: OCB mode + join and leave handling

On Thu, 2014-10-16 at 18:33 +0200, Rostislav Lisovy wrote:
> > > + mutex_lock(&sdata->local->mtx);
> > > + ieee80211_vif_release_channel(sdata);
> > > + mutex_unlock(&sdata->local->mtx);
> > > +
> > > + skb_queue_purge(&sdata->skb_queue);
> > > +
> > > + del_timer_sync(&sdata->u.ocb.housekeeping_timer);
> >
> > That might call the timer - is it safe if that happens here? Looks like
> > maybe the housekeeping would still get triggered or so.
>
> You are right. I hope the following is a reasonable solution (in form of
> a patch to my previous patch; comment stolen from some prehistoric
> version of mesh.c):
>
> @@ -127,6 +127,9 @@ void ieee80211_ocb_work(struct ieee80211_sub_if_data *sdata)
> struct ieee80211_if_ocb *ifocb = &sdata->u.ocb;
> struct sta_info *sta;
>
> + if (!netif_running(sdata->dev))
> + return;
> +
> sdata_lock(sdata);
>
> spin_lock_bh(&ifocb->incomplete_lock);
> @@ -229,6 +232,13 @@ int ieee80211_ocb_leave(struct ieee80211_sub_if_data *sdata)
> skb_queue_purge(&sdata->skb_queue);
>
> del_timer_sync(&sdata->u.ocb.housekeeping_timer);
> + /*
> + * If the timer fired while we waited for it, it will have
> + * requeued the work. Now the work will be running again
> + * but will not rearm the timer again because it checks
> + * whether the interface is running, which, at this point,
> + * it no longer is.
> + */
>
> return 0;
> }

Now I realized it is not that easy (and I confused "interface running"
and being "connected to the network"). There seems not to be a solid
indication that we are no longer "connected" to the network.
I think a field
enum {
IEEE80211_OCB_STOPPED,
IEEE80211_OCB_JOINED,
} state;
in struct ieee80211_if_ocb seems to be appropriate.

Best regards;
Rostislav