RFC: Allow multiple STA connected to same AP.

PATCH Ben Greear 2 participants 7 messages Sep 9, 2010 – Sep 10, 2010 mac80211

Share & tools

Ben Greear

author

2010-09-09 22:03 UTC #0 lore raw

With the patch below against latest wireless-testing, I can create two STA on the same ath9k phy0 and have them send traffic to each other. I tested only un-encrypted STAs at this point.

It seems the tx logic hangs after a bit, so there are still issues, but I'm not sure if that is something introduced by my patch or some existing ath9k bug.

Please let me know if this is a viable approach.

Thanks,
Ben

net/mac80211/main.c +18 -0

index 4935b84..21401c9 100644
@@ -44,6 +44,8 @@ void ieee80211_configure_filter(struct ieee80211_local *local)
        u64 mc;
        unsigned int changed_flags;
        unsigned int new_flags = 0;
+       int avifs = 0;
+       struct ieee80211_sub_if_data *sdata;
 
        if (atomic_read(&local->iff_promiscs))
                new_flags |= FIF_PROMISC_IN_BSS;

@@ -78,6 +80,22 @@ void ieee80211_configure_filter(struct ieee80211_local *local)
        /* be a bit nasty */
        new_flags |= (1<<31);
 
+       /* If we have more than one virtual station, turn on PROMISC_IN_BSS
+        *  --Ben
+        */
+       list_for_each_entry_rcu(sdata, &local->interfaces, list) {
+               if (!sdata->dev || !netif_running(sdata->dev))
+                       continue;
+
+               if (sdata->vif.type == NL80211_IFTYPE_STATION) {
+                       avifs++;
+                       if (avifs > 1)
+                               break;
+               }
+       }
+       if (avifs > 1)
+               new_flags |= FIF_PROMISC_IN_BSS;
+
        drv_configure_filter(local, changed_flags, &new_flags, mc);
 
        WARN_ON(new_flags & (1<<31));

net/mac80211/sta_info.c +1 -1

index 687077e..db751f1 100644
@@ -440,7 +440,7 @@ int sta_info_insert_rcu(struct sta_info *sta) __acquires(RCU)
 
        spin_lock_irqsave(&local->sta_lock, flags);
        /* check if STA exists already */
-       if (sta_info_get_bss(sdata, sta->sta.addr)) {
+       if (sta_info_get(sdata, sta->sta.addr)) {
                spin_unlock_irqrestore(&local->sta_lock, flags);
                mutex_unlock(&local->sta_mtx);
                rcu_read_lock();

Signature · 2 lines

Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com

Johannes Berg

maintainer

2010-09-09 22:51 UTC #1 lore raw

Quoted 3 lines · Ben Greear

With the patch below against latest wireless-testing, I can create two STA on the same ath9k phy0 and have them send traffic to each other. I tested only un-encrypted STAs at this point.

After connecting them to the same AP, presumably?

Quoted 7 lines

It seems the tx logic hangs after a bit, so there are still issues, but I'm not sure if that is something introduced by my patch or some existing ath9k bug.
Please let me know if this is a viable approach.
+ /* If we have more than one virtual station, turn on

PROMISC_IN_BSS

Quoted 5 lines

net/mac80211/main.c +5 -0

+        *  --Ben
+        */
+       list_for_each_entry_rcu(sdata, &local->interfaces, list) {
+               if (!sdata->dev || !netif_running(sdata->dev))
+                       continue;

net/mac80211/main.c void ieee80211_configure_filter(struct ieee80211_local *local) · line 98

Quoted 10 lines

net/mac80211/main.c +10 -0

+
+               if (sdata->vif.type == NL80211_IFTYPE_STATION) {
+                       avifs++;
+                       if (avifs > 1)
+                               break;
+               }
+       }
+       if (avifs > 1)
+               new_flags |= FIF_PROMISC_IN_BSS;
+

This seems weird. Why do you need to be *promisc* within a given BSS if you are multiple stations? You already have two MAC addrs that you need to accept, so that should be fine, i.e. you don't need to be promisc in the BSS. If ath9k needs to, it should be in the driver. So NAK on this change for sure.

net/mac80211/sta_info.c int sta_info_insert_rcu(struct sta_info *sta) · line 444

Quoted 7 lines

net/mac80211/sta_info.c +1 -1

@@ -440,7 +440,7 @@ int sta_info_insert_rcu(struct sta_info *sta)
 __acquires(RCU)
 
         spin_lock_irqsave(&local->sta_lock, flags);
         /* check if STA exists already */
-       if (sta_info_get_bss(sdata, sta->sta.addr)) {
+       if (sta_info_get(sdata, sta->sta.addr)) {

I don't understand this change. These functions are the same if sdata is in managed mode.

johannes

Ben Greear

author

2010-09-09 23:01 UTC #2 lore raw

Quoted 8 lines · Johannes Berg

Quoted 3 lines · Ben Greear
With the patch below against latest wireless-testing, I can create two STA on the same ath9k phy0 and have them send traffic to each other. I tested only un-encrypted STAs at this point.
After connecting them to the same AP, presumably?

Yes, though hopefully it would also work connecting to different APs so long as the APs can communicate properly.

I've also been testing WPA..and that definitely doesn't work yet. Seems they fail to authenticate with the AP, though they do talk and send EAPOL pkts back and forth.

Quoted 22 lines

Quoted 8 lines

It seems the tx logic hangs after a bit, so there are still issues, but I'm not sure if that is something introduced by my patch or some existing ath9k bug.
Please let me know if this is a viable approach.
+ /* If we have more than one virtual station, turn on

PROMISC_IN_BSS

Quoted 13 lines

net/mac80211/main.c +13 -0

+        *  --Ben
+        */
+       list_for_each_entry_rcu(sdata,&local->interfaces, list) {
+               if (!sdata->dev || !netif_running(sdata->dev))
+                       continue;
+
+               if (sdata->vif.type == NL80211_IFTYPE_STATION) {
+                       avifs++;
+                       if (avifs>  1)
+                               break;
+               }
+       }
+       if (avifs>  1)

Quoted 8 lines

net/mac80211/main.c void ieee80211_configure_filter(struct ieee80211_local *local) · line 98
Quoted 2 lines
net/mac80211/main.c +2 -0
+               new_flags |= FIF_PROMISC_IN_BSS;
+
This seems weird. Why do you need to be *promisc* within a given BSS if you are multiple stations? You already have two MAC addrs that you need to accept, so that should be fine, i.e. you don't need to be promisc in the BSS. If ath9k needs to, it should be in the driver. So NAK on this change for sure.

This seems to let the driver receive pkts for the 'real' mac on wlan0 and also the arbitrary mac on the second STA interface.

If there is supposed to be some other way for the driver to figure out it's supposed to go promisc, please let me know and I'll try to debug it.

Quoted 10 lines

net/mac80211/sta_info.c int sta_info_insert_rcu(struct sta_info *sta) · line 444
Quoted 7 lines
net/mac80211/sta_info.c +1 -1
@@ -440,7 +440,7 @@ int sta_info_insert_rcu(struct sta_info *sta)
 __acquires(RCU)
 
          spin_lock_irqsave(&local->sta_lock, flags);
          /* check if STA exists already */
-       if (sta_info_get_bss(sdata, sta->sta.addr)) {
+       if (sta_info_get(sdata, sta->sta.addr)) {
I don't understand this change. These functions are the same if sdata is in managed mode.

The _bss method returns anything in same bss or with same sdata. The sta_info_get only returns if MAC and sdata match.

W/out this change, the second STA fails this check because it has same BSS as the first one.

Thanks,
Ben

Signature · 2 lines

Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com

Johannes Berg

maintainer

2010-09-09 23:08 UTC #3 lore raw

Quoted 24 lines · Ben Greear

Quoted 15 lines
Quoted 1 line
+ /* If we have more than one virtual station, turn on
PROMISC_IN_BSS
Quoted 13 lines
net/mac80211/main.c +13 -0
+        *  --Ben
+        */
+       list_for_each_entry_rcu(sdata,&local->interfaces, list) {
+               if (!sdata->dev || !netif_running(sdata->dev))
+                       continue;
+
+               if (sdata->vif.type == NL80211_IFTYPE_STATION) {
+                       avifs++;
+                       if (avifs>  1)
+                               break;
+               }
+       }
+       if (avifs>  1)
Quoted 8 lines
net/mac80211/main.c void ieee80211_configure_filter(struct ieee80211_local *local) · line 98
Quoted 2 lines
net/mac80211/main.c +2 -0
+               new_flags |= FIF_PROMISC_IN_BSS;
+
This seems weird. Why do you need to be *promisc* within a given BSS if you are multiple stations? You already have two MAC addrs that you need to accept, so that should be fine, i.e. you don't need to be promisc in the BSS. If ath9k needs to, it should be in the driver. So NAK on this change for sure.

Quoted 6 lines

This seems to let the driver receive pkts for the 'real' mac on wlan0 and also the arbitrary mac on the second STA interface.
If there is supposed to be some other way for the driver to figure out it's supposed to go promisc, please let me know and I'll try to debug it.

But you can do the exact same thing in the driver. I just think it doesn't belong into mac80211, since mac80211 doesn't need you to be promisc, it just needs you to receive frames for those *two* addresses that you've added. And you already know about those, since you were notified about the interfaces going up and down and associating etc.

Quoted 11 lines

Quoted 10 lines
net/mac80211/sta_info.c int sta_info_insert_rcu(struct sta_info *sta) · line 444
Quoted 7 lines
net/mac80211/sta_info.c +1 -1
@@ -440,7 +440,7 @@ int sta_info_insert_rcu(struct sta_info *sta)
 __acquires(RCU)
 
          spin_lock_irqsave(&local->sta_lock, flags);
          /* check if STA exists already */
-       if (sta_info_get_bss(sdata, sta->sta.addr)) {
+       if (sta_info_get(sdata, sta->sta.addr)) {
I don't understand this change. These functions are the same if sdata is in managed mode.

Quoted 5 lines

The _bss method returns anything in same bss or with same sdata. The sta_info_get only returns if MAC and sdata match.
W/out this change, the second STA fails this check because it has same BSS as the first one.

I'll have to look at this, I may misremember the way this works internally, but this change will certainly break some things since it'd allow adding the same STA to multiple VLANs that sit on top of the same BSS.

johannes

Ben Greear

author

2010-09-10 04:52 UTC #4 lore raw

Quoted 34 lines · Johannes Berg

Quoted 24 lines · Ben Greear
Quoted 15 lines
Quoted 1 line
+ /* If we have more than one virtual station, turn on
PROMISC_IN_BSS
Quoted 13 lines
net/mac80211/main.c +13 -0
+        *  --Ben
+        */
+       list_for_each_entry_rcu(sdata,&local->interfaces, list) {
+               if (!sdata->dev || !netif_running(sdata->dev))
+                       continue;
+
+               if (sdata->vif.type == NL80211_IFTYPE_STATION) {
+                       avifs++;
+                       if (avifs>   1)
+                               break;
+               }
+       }
+       if (avifs>   1)
Quoted 8 lines
net/mac80211/main.c void ieee80211_configure_filter(struct ieee80211_local *local) · line 98
Quoted 2 lines
net/mac80211/main.c +2 -0
+               new_flags |= FIF_PROMISC_IN_BSS;
+
This seems weird. Why do you need to be *promisc* within a given BSS if you are multiple stations? You already have two MAC addrs that you need to accept, so that should be fine, i.e. you don't need to be promisc in the BSS. If ath9k needs to, it should be in the driver. So NAK on this change for sure.
Quoted 6 lines
This seems to let the driver receive pkts for the 'real' mac on wlan0 and also the arbitrary mac on the second STA interface.
If there is supposed to be some other way for the driver to figure out it's supposed to go promisc, please let me know and I'll try to debug it.

Quoted 6 lines

But you can do the exact same thing in the driver. I just think it doesn't belong into mac80211, since mac80211 doesn't need you to be promisc, it just needs you to receive frames for those *two* addresses that you've added. And you already know about those, since you were notified about the interfaces going up and down and associating etc.

After some more debugging, it seems at least part of the problem is that ath9k uses ieee80211_iterate_active_interfaces_atomic to determine which MACs to add to the BSSID mask. But, before things are associated properly, sdata isn't 'running', so ath9k doesn't calc the mask properly.

Looks a bit like chicken-and-egg issue, and for some reason, it doesn't seem to happen with WPA (but then, WPA doesn't authenticate properly either).

Maybe a new iterator is needed to get interfaces that want to go active (maybe by determining if sdata->dev is 'UP') ?

Thanks,
Ben

Signature · 2 lines

Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com

Johannes Berg

maintainer

2010-09-10 16:37 UTC #5 lore raw

Quoted 3 lines · Ben Greear

After some more debugging, it seems at least part of the problem is that ath9k uses ieee80211_iterate_active_interfaces_atomic to determine which MACs to add to the BSSID mask. But, before things are associated

properly,

Quoted 8 lines

sdata isn't 'running', so ath9k doesn't calc the mask properly.
Looks a bit like chicken-and-egg issue, and for some reason, it doesn't seem to happen with WPA (but then, WPA doesn't authenticate properly either).
Maybe a new iterator is needed to get interfaces that want to go active (maybe by determining if sdata->dev is 'UP') ?

Err, that's exactly what the active interfaces iterator does. sdata is in fact 'running' as soon as the interface is UP.

johannes

Ben Greear

author

2010-09-10 16:53 UTC #6 lore raw

Quoted 14 lines · Johannes Berg

Quoted 3 lines · Ben Greear
After some more debugging, it seems at least part of the problem is that ath9k uses ieee80211_iterate_active_interfaces_atomic to determine which MACs to add to the BSSID mask. But, before things are associated
properly,
Quoted 6 lines
sdata isn't 'running', so ath9k doesn't calc the mask properly.
Looks a bit like chicken-and-egg issue, and for some reason, it doesn't seem to happen with WPA (but then, WPA doesn't authenticate properly either).

Quoted 5 lines

Quoted 2 lines
Maybe a new iterator is needed to get interfaces that want to go active (maybe by determining if sdata->dev is 'UP') ?
Err, that's exactly what the active interfaces iterator does. sdata is in fact 'running' as soon as the interface is UP.

Maybe that bssidmask logic runs before sdata is actually brought up then. I'll keep poking at things.

Thanks,
Ben

Signature · 2 lines

Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com