[PATCH v2] wireless: Allow registering more than one beacon listener.

[PATCH v2] wireless: Allow registering more than one beacon listener.

[greearb]

From: Ben Greear <greearb@candelatech.com>

The commit:

commit 5e760230e42cf759bd923457ca2753aacf2e656e
Author: Johannes Berg <johannes.berg@intel.com>
Date:   Fri Nov 4 11:18:17 2011 +0100

    cfg80211: allow registering to beacons

allowed only a single process to register for beacon events
per wiphy.  This breaks cases where a user may want two or
more VIFs on a wiphy and run a seperate hostapd process on
each vif.

This patch allows multiple beacon listeners, fixing the
regression.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
v2:  Suggested changes from Johannes Berg.

 include/net/cfg80211.h |    3 +-
 net/mac80211/rx.c      |    2 +-
 net/wireless/core.c    |    7 ++++
 net/wireless/core.h    |    7 +++-
 net/wireless/nl80211.c |   88 ++++++++++++++++++++++++++++++++---------------
 5 files changed, 75 insertions(+), 32 deletions(-)

diff --git a/include/net/cfg80211.h b/include/net/cfg80211.h
index a7c0d15..8e0fa15 100644
--- a/include/net/cfg80211.h
+++ b/include/net/cfg80211.h
@@ -3544,7 +3544,6 @@ void cfg80211_probe_status(struct net_device *dev, const u8 *addr,
  * @len: length of the frame
  * @freq: frequency the frame was received on
  * @sig_dbm: signal strength in mBm, or 0 if unknown
- * @gfp: allocation flags
  *
  * Use this function to report to userspace when a beacon was
  * received. It is not useful to call this when there is no
@@ -3552,7 +3551,7 @@ void cfg80211_probe_status(struct net_device *dev, const u8 *addr,
  */
 void cfg80211_report_obss_beacon(struct wiphy *wiphy,
 				 const u8 *frame, size_t len,
-				 int freq, int sig_dbm, gfp_t gfp);
+				 int freq, int sig_dbm);
 
 /**
  * cfg80211_can_beacon_sec_chan - test if ht40 on extension channel can be used
diff --git a/net/mac80211/rx.c b/net/mac80211/rx.c
index d07216a..7b1c5f3 100644
--- a/net/mac80211/rx.c
+++ b/net/mac80211/rx.c
@@ -2187,7 +2187,7 @@ ieee80211_rx_h_mgmt_check(struct ieee80211_rx_data *rx)
 
 		cfg80211_report_obss_beacon(rx->local->hw.wiphy,
 					    rx->skb->data, rx->skb->len,
-					    status->freq, sig, GFP_ATOMIC);
+					    status->freq, sig);
 		rx->flags |= IEEE80211_RX_BEACON_REPORTED;
 	}
 
diff --git a/net/wireless/core.c b/net/wireless/core.c
index ce1ad77..9c3a657 100644
--- a/net/wireless/core.c
+++ b/net/wireless/core.c
@@ -326,6 +326,8 @@ struct wiphy *wiphy_new(const struct cfg80211_ops *ops, int sizeof_priv)
 	mutex_init(&rdev->devlist_mtx);
 	mutex_init(&rdev->sched_scan_mtx);
 	INIT_LIST_HEAD(&rdev->wdev_list);
+	INIT_LIST_HEAD(&rdev->beacon_registrations);
+	spin_lock_init(&rdev->beacon_registrations_lock);
 	spin_lock_init(&rdev->bss_lock);
 	INIT_LIST_HEAD(&rdev->bss_list);
 	INIT_WORK(&rdev->scan_done_wk, __cfg80211_scan_done);
@@ -699,10 +701,15 @@ EXPORT_SYMBOL(wiphy_unregister);
 void cfg80211_dev_free(struct cfg80211_registered_device *rdev)
 {
 	struct cfg80211_internal_bss *scan, *tmp;
+	struct cfg80211_beacon_registration *reg, *treg;
 	rfkill_destroy(rdev->rfkill);
 	mutex_destroy(&rdev->mtx);
 	mutex_destroy(&rdev->devlist_mtx);
 	mutex_destroy(&rdev->sched_scan_mtx);
+	list_for_each_entry_safe(reg, treg, &rdev->beacon_registrations, list) {
+		list_del(&reg->list);
+		kfree(reg);
+	}
 	list_for_each_entry_safe(scan, tmp, &rdev->bss_list, list)
 		cfg80211_put_bss(&scan->pub);
 	kfree(rdev);
diff --git a/net/wireless/core.h b/net/wireless/core.h
index b8eb743..e53831c 100644
--- a/net/wireless/core.h
+++ b/net/wireless/core.h
@@ -55,7 +55,8 @@ struct cfg80211_registered_device {
 	int opencount; /* also protected by devlist_mtx */
 	wait_queue_head_t dev_wait;
 
-	u32 ap_beacons_nlportid;
+	struct list_head beacon_registrations;
+	spinlock_t beacon_registrations_lock;
 
 	/* protected by RTNL only */
 	int num_running_ifaces;
@@ -260,6 +261,10 @@ enum cfg80211_chan_mode {
 	CHAN_MODE_EXCLUSIVE,
 };
 
+struct cfg80211_beacon_registration {
+	struct list_head list;
+	u32 nlportid;
+};
 
 /* free object */
 extern void cfg80211_dev_free(struct cfg80211_registered_device *rdev);
diff --git a/net/wireless/nl80211.c b/net/wireless/nl80211.c
index e408cbf..16dd2de 100644
--- a/net/wireless/nl80211.c
+++ b/net/wireless/nl80211.c
@@ -6914,16 +6914,35 @@ static int nl80211_probe_client(struct sk_buff *skb,
 static int nl80211_register_beacons(struct sk_buff *skb, struct genl_info *info)
 {
 	struct cfg80211_registered_device *rdev = info->user_ptr[0];
+	struct cfg80211_beacon_registration *reg, *nreg;
+	int rv;
 
 	if (!(rdev->wiphy.flags & WIPHY_FLAG_REPORTS_OBSS))
 		return -EOPNOTSUPP;
 
-	if (rdev->ap_beacons_nlportid)
-		return -EBUSY;
+	nreg = kzalloc(sizeof(*nreg), GFP_KERNEL);
+	if (!nreg)
+		return -ENOMEM;
+
+	/* First, check if already registered. */
+	spin_lock_bh(&rdev->beacon_registrations_lock);
+	list_for_each_entry(reg, &rdev->beacon_registrations, list) {
+		if (reg->nlportid == info->snd_portid) {
+			rv = -EALREADY;
+			goto out_err;
+		}
+	}
+	/* Add it to the list */
+	nreg->nlportid = info->snd_portid;
+	list_add(&nreg->list, &rdev->beacon_registrations);
 
-	rdev->ap_beacons_nlportid = info->snd_portid;
+	spin_unlock_bh(&rdev->beacon_registrations_lock);
 
 	return 0;
+out_err:
+	spin_unlock_bh(&rdev->beacon_registrations_lock);
+	kfree(nreg);
+	return rv;
 }
 
 static int nl80211_start_p2p_device(struct sk_buff *skb, struct genl_info *info)
@@ -8929,43 +8948,46 @@ EXPORT_SYMBOL(cfg80211_probe_status);
 
 void cfg80211_report_obss_beacon(struct wiphy *wiphy,
 				 const u8 *frame, size_t len,
-				 int freq, int sig_dbm, gfp_t gfp)
+				 int freq, int sig_dbm)
 {
 	struct cfg80211_registered_device *rdev = wiphy_to_dev(wiphy);
 	struct sk_buff *msg;
 	void *hdr;
-	u32 nlportid = ACCESS_ONCE(rdev->ap_beacons_nlportid);
+	struct cfg80211_beacon_registration *reg;
 
 	trace_cfg80211_report_obss_beacon(wiphy, frame, len, freq, sig_dbm);
 
-	if (!nlportid)
-		return;
-
-	msg = nlmsg_new(len + 100, gfp);
-	if (!msg)
-		return;
+	spin_lock_bh(&rdev->beacon_registrations_lock);
+	list_for_each_entry(reg, &rdev->beacon_registrations, list) {
+		msg = nlmsg_new(len + 100, GFP_ATOMIC);
+		if (!msg) {
+			spin_unlock_bh(&rdev->beacon_registrations_lock);
+			return;
+		}
 
-	hdr = nl80211hdr_put(msg, 0, 0, 0, NL80211_CMD_FRAME);
-	if (!hdr) {
-		nlmsg_free(msg);
-		return;
-	}
+		hdr = nl80211hdr_put(msg, 0, 0, 0, NL80211_CMD_FRAME);
+		if (!hdr)
+			goto nla_put_failure;
 
-	if (nla_put_u32(msg, NL80211_ATTR_WIPHY, rdev->wiphy_idx) ||
-	    (freq &&
-	     nla_put_u32(msg, NL80211_ATTR_WIPHY_FREQ, freq)) ||
-	    (sig_dbm &&
-	     nla_put_u32(msg, NL80211_ATTR_RX_SIGNAL_DBM, sig_dbm)) ||
-	    nla_put(msg, NL80211_ATTR_FRAME, len, frame))
-		goto nla_put_failure;
+		if (nla_put_u32(msg, NL80211_ATTR_WIPHY, rdev->wiphy_idx) ||
+		    (freq &&
+		     nla_put_u32(msg, NL80211_ATTR_WIPHY_FREQ, freq)) ||
+		    (sig_dbm &&
+		     nla_put_u32(msg, NL80211_ATTR_RX_SIGNAL_DBM, sig_dbm)) ||
+		    nla_put(msg, NL80211_ATTR_FRAME, len, frame))
+			goto nla_put_failure;
 
-	genlmsg_end(msg, hdr);
+		genlmsg_end(msg, hdr);
 
-	genlmsg_unicast(wiphy_net(&rdev->wiphy), msg, nlportid);
+		genlmsg_unicast(wiphy_net(&rdev->wiphy), msg, reg->nlportid);
+	}
+	spin_unlock_bh(&rdev->beacon_registrations_lock);
 	return;
 
  nla_put_failure:
-	genlmsg_cancel(msg, hdr);
+	spin_unlock_bh(&rdev->beacon_registrations_lock);
+	if (hdr)
+		genlmsg_cancel(msg, hdr);
 	nlmsg_free(msg);
 }
 EXPORT_SYMBOL(cfg80211_report_obss_beacon);
@@ -8977,6 +8999,7 @@ static int nl80211_netlink_notify(struct notifier_block * nb,
 	struct netlink_notify *notify = _notify;
 	struct cfg80211_registered_device *rdev;
 	struct wireless_dev *wdev;
+	struct cfg80211_beacon_registration *reg, *tmp;
 
 	if (state != NETLINK_URELEASE)
 		return NOTIFY_DONE;
@@ -8986,8 +9009,17 @@ static int nl80211_netlink_notify(struct notifier_block * nb,
 	list_for_each_entry_rcu(rdev, &cfg80211_rdev_list, list) {
 		list_for_each_entry_rcu(wdev, &rdev->wdev_list, list)
 			cfg80211_mlme_unregister_socket(wdev, notify->portid);
-		if (rdev->ap_beacons_nlportid == notify->portid)
-			rdev->ap_beacons_nlportid = 0;
+
+		spin_lock_bh(&rdev->beacon_registrations_lock);
+		list_for_each_entry_safe(reg, tmp, &rdev->beacon_registrations,
+					 list) {
+			if (reg->nlportid == notify->portid) {
+				list_del(&reg->list);
+				kfree(reg);
+				break;
+			}
+		}
+		spin_unlock_bh(&rdev->beacon_registrations_lock);
 	}
 
 	rcu_read_unlock();
-- 
1.7.3.4

Re: [PATCH v2] wireless: Allow registering more than one beacon listener.

[Johannes Berg]

Thanks Ben,

I was going to apply this but it looks like I missed something on the
previous review:

>  void cfg80211_report_obss_beacon(struct wiphy *wiphy,
>  				 const u8 *frame, size_t len,
> -				 int freq, int sig_dbm, gfp_t gfp)
> +				 int freq, int sig_dbm)
>  {

...

> +	spin_lock_bh(&rdev->beacon_registrations_lock);
...
> +	spin_unlock_bh(&rdev->beacon_registrations_lock);

I have a feeling this is incorrect, this is called with BHs disabled, so
if we disable/enable BHs here we end up with BHs enabled even if the
caller wanted them disabled, no?

So I suppose unless we need to be able to call this from hard interrupt
in some driver (unlikely) we should add a change like this:

diff --git a/include/net/cfg80211.h b/include/net/cfg80211.h
index 10a7f45..3b6a11a 100644
--- a/include/net/cfg80211.h
+++ b/include/net/cfg80211.h
@@ -3544,6 +3544,7 @@ void cfg80211_probe_status(struct net_device *dev, const u8 *addr,
  * Use this function to report to userspace when a beacon was
  * received. It is not useful to call this when there is no
  * netdev that is in AP/GO mode.
+ * The function must be called with BHs disabled.
  */
 void cfg80211_report_obss_beacon(struct wiphy *wiphy,
 				 const u8 *frame, size_t len,
diff --git a/net/wireless/nl80211.c b/net/wireless/nl80211.c
index 15ddbb8..5ff24ef 100644
--- a/net/wireless/nl80211.c
+++ b/net/wireless/nl80211.c
@@ -8942,11 +8942,11 @@ void cfg80211_report_obss_beacon(struct wiphy *wiphy,
 
 	trace_cfg80211_report_obss_beacon(wiphy, frame, len, freq, sig_dbm);
 
-	spin_lock_bh(&rdev->beacon_registrations_lock);
+	spin_lock(&rdev->beacon_registrations_lock);
 	list_for_each_entry(reg, &rdev->beacon_registrations, list) {
 		msg = nlmsg_new(len + 100, GFP_ATOMIC);
 		if (!msg) {
-			spin_unlock_bh(&rdev->beacon_registrations_lock);
+			spin_unlock(&rdev->beacon_registrations_lock);
 			return;
 		}
 
@@ -8966,11 +8966,11 @@ void cfg80211_report_obss_beacon(struct wiphy *wiphy,
 
 		genlmsg_unicast(wiphy_net(&rdev->wiphy), msg, reg->nlportid);
 	}
-	spin_unlock_bh(&rdev->beacon_registrations_lock);
+	spin_unlock(&rdev->beacon_registrations_lock);
 	return;
 
  nla_put_failure:
-	spin_unlock_bh(&rdev->beacon_registrations_lock);
+	spin_unlock(&rdev->beacon_registrations_lock);
 	if (hdr)
 		genlmsg_cancel(msg, hdr);
 	nlmsg_free(msg);


johannes

Re: [PATCH v2] wireless: Allow registering more than one beacon listener.

[Ben Greear]

On 10/29/2012 02:04 AM, Johannes Berg wrote:
> Thanks Ben,
>
> I was going to apply this but it looks like I missed something on the
> previous review:
>
>>   void cfg80211_report_obss_beacon(struct wiphy *wiphy,
>>   				 const u8 *frame, size_t len,
>> -				 int freq, int sig_dbm, gfp_t gfp)
>> +				 int freq, int sig_dbm)
>>   {
>
> ...
>
>> +	spin_lock_bh(&rdev->beacon_registrations_lock);
> ...
>> +	spin_unlock_bh(&rdev->beacon_registrations_lock);
>
> I have a feeling this is incorrect, this is called with BHs disabled, so
> if we disable/enable BHs here we end up with BHs enabled even if the
> caller wanted them disabled, no?
>
> So I suppose unless we need to be able to call this from hard interrupt
> in some driver (unlikely) we should add a change like this:

I get confused with the various types of spin-locks,
and I just copied this code from the mgt frames logic.

I did run some tests and my code seems stable, but I could
just be getting lucky.

Do you want me to roll in your changes below and re-submit, or
do you want to just make the updates?

Thanks,
Ben

>
> diff --git a/include/net/cfg80211.h b/include/net/cfg80211.h
> index 10a7f45..3b6a11a 100644
> --- a/include/net/cfg80211.h
> +++ b/include/net/cfg80211.h
> @@ -3544,6 +3544,7 @@ void cfg80211_probe_status(struct net_device *dev, const u8 *addr,
>    * Use this function to report to userspace when a beacon was
>    * received. It is not useful to call this when there is no
>    * netdev that is in AP/GO mode.
> + * The function must be called with BHs disabled.
>    */
>   void cfg80211_report_obss_beacon(struct wiphy *wiphy,
>   				 const u8 *frame, size_t len,
> diff --git a/net/wireless/nl80211.c b/net/wireless/nl80211.c
> index 15ddbb8..5ff24ef 100644
> --- a/net/wireless/nl80211.c
> +++ b/net/wireless/nl80211.c
> @@ -8942,11 +8942,11 @@ void cfg80211_report_obss_beacon(struct wiphy *wiphy,
>
>   	trace_cfg80211_report_obss_beacon(wiphy, frame, len, freq, sig_dbm);
>
> -	spin_lock_bh(&rdev->beacon_registrations_lock);
> +	spin_lock(&rdev->beacon_registrations_lock);
>   	list_for_each_entry(reg, &rdev->beacon_registrations, list) {
>   		msg = nlmsg_new(len + 100, GFP_ATOMIC);
>   		if (!msg) {
> -			spin_unlock_bh(&rdev->beacon_registrations_lock);
> +			spin_unlock(&rdev->beacon_registrations_lock);
>   			return;
>   		}
>
> @@ -8966,11 +8966,11 @@ void cfg80211_report_obss_beacon(struct wiphy *wiphy,
>
>   		genlmsg_unicast(wiphy_net(&rdev->wiphy), msg, reg->nlportid);
>   	}
> -	spin_unlock_bh(&rdev->beacon_registrations_lock);
> +	spin_unlock(&rdev->beacon_registrations_lock);
>   	return;
>
>    nla_put_failure:
> -	spin_unlock_bh(&rdev->beacon_registrations_lock);
> +	spin_unlock(&rdev->beacon_registrations_lock);
>   	if (hdr)
>   		genlmsg_cancel(msg, hdr);
>   	nlmsg_free(msg);
>
>
> johannes
>


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: [PATCH v2] wireless: Allow registering more than one beacon listener.

[Johannes Berg]

On Mon, 2012-10-29 at 09:17 -0700, Ben Greear wrote:

> >> +	spin_lock_bh(&rdev->beacon_registrations_lock);
> > ...
> >> +	spin_unlock_bh(&rdev->beacon_registrations_lock);
> >
> > I have a feeling this is incorrect, this is called with BHs disabled, so
> > if we disable/enable BHs here we end up with BHs enabled even if the
> > caller wanted them disabled, no?
> >
> > So I suppose unless we need to be able to call this from hard interrupt
> > in some driver (unlikely) we should add a change like this:
> 
> I get confused with the various types of spin-locks,
> and I just copied this code from the mgt frames logic.

Uh, that looks like a bug there as well then?

> I did run some tests and my code seems stable, but I could
> just be getting lucky.

Hmm I guess it depends on the driver, I suppose enabling BHs inside one
doesn't actually do anything, and it would only really matter for
drivers using ieee80211_rx_ni()?

> Do you want me to roll in your changes below and re-submit, or
> do you want to just make the updates?

I can do the update, I guess I'll sort out the mgmt frame part as well.

johannes

Re: [PATCH v2] wireless: Allow registering more than one beacon listener.

[Felix Fietkau]

On 2012-10-29 5:26 PM, Johannes Berg wrote:
> On Mon, 2012-10-29 at 09:17 -0700, Ben Greear wrote:
> 
>> >> +	spin_lock_bh(&rdev->beacon_registrations_lock);
>> > ...
>> >> +	spin_unlock_bh(&rdev->beacon_registrations_lock);
>> >
>> > I have a feeling this is incorrect, this is called with BHs disabled, so
>> > if we disable/enable BHs here we end up with BHs enabled even if the
>> > caller wanted them disabled, no?
>> >
>> > So I suppose unless we need to be able to call this from hard interrupt
>> > in some driver (unlikely) we should add a change like this:
>> 
>> I get confused with the various types of spin-locks,
>> and I just copied this code from the mgt frames logic.
> 
> Uh, that looks like a bug there as well then?
As far as I know, local_bh_enable/local_bh_disable (which are used by
spin_lock_bh) are re-entrant. Using them from a BH-disabled context is
safe. All I can find with a few simple google searches confirms this.

- Felix

Re: [PATCH v2] wireless: Allow registering more than one beacon listener.

[Johannes Berg]

On Mon, 2012-10-29 at 17:54 +0100, Felix Fietkau wrote:
> On 2012-10-29 5:26 PM, Johannes Berg wrote:
> > On Mon, 2012-10-29 at 09:17 -0700, Ben Greear wrote:
> > 
> >> >> +	spin_lock_bh(&rdev->beacon_registrations_lock);
> >> > ...
> >> >> +	spin_unlock_bh(&rdev->beacon_registrations_lock);
> >> >
> >> > I have a feeling this is incorrect, this is called with BHs disabled, so
> >> > if we disable/enable BHs here we end up with BHs enabled even if the
> >> > caller wanted them disabled, no?
> >> >
> >> > So I suppose unless we need to be able to call this from hard interrupt
> >> > in some driver (unlikely) we should add a change like this:
> >> 
> >> I get confused with the various types of spin-locks,
> >> and I just copied this code from the mgt frames logic.
> > 
> > Uh, that looks like a bug there as well then?
> As far as I know, local_bh_enable/local_bh_disable (which are used by
> spin_lock_bh) are re-entrant. Using them from a BH-disabled context is
> safe. All I can find with a few simple google searches confirms this.

Hm yeah I must've confused it with something, I'll apply the patch.

johannes