2010-02-23 09:57:56

by Bruno Randolf

[permalink] [raw]
Subject: [PATCH] mac80211: fix rates setup on IBSS merge

when an IBSS merge happened, the supported rates for the newly added station
were left empty, causing the rate control module to be initialized with only
the basic rates.

also the section of the ibss code which deals with updating supported rates for
an already existing station fails to inform the rate control module about the
new rates. as i don't know how to fix this (minstrel does not have an update
function), i have just added a comment for now.

Signed-off-by: Bruno Randolf <[email protected]>
---

net/mac80211/ibss.c | 3 +++
1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/net/mac80211/ibss.c b/net/mac80211/ibss.c
index f3e9424..b840d90 100644
--- a/net/mac80211/ibss.c
+++ b/net/mac80211/ibss.c
@@ -276,6 +276,8 @@ static void ieee80211_rx_bss_info(struct ieee80211_sub_if_data *sdata,
(unsigned long long) sta->sta.supp_rates[band]);
#endif
rcu_read_unlock();
+
+ /* FIXME: update rate control */
} else {
rcu_read_unlock();
ieee80211_ibss_add_sta(sdata, mgmt->bssid, mgmt->sa,
@@ -370,6 +372,7 @@ static void ieee80211_rx_bss_info(struct ieee80211_sub_if_data *sdata,
sdata->name, mgmt->bssid);
#endif
ieee80211_sta_join_ibss(sdata, bss);
+ supp_rates = ieee80211_sta_get_rates(local, elems, band);
ieee80211_ibss_add_sta(sdata, mgmt->bssid, mgmt->sa,
supp_rates, GFP_KERNEL);
}



2010-02-23 19:28:26

by Luis R. Rodriguez

[permalink] [raw]
Subject: Re: [PATCH] mac80211: fix rates setup on IBSS merge

On Tue, Feb 23, 2010 at 1:51 AM, Bruno Randolf <[email protected]> wrote:
> when an IBSS merge happened, the supported rates for the newly added station
> were left empty, causing the rate control module to be initialized with only
> the basic rates.
>
> also the section of the ibss code which deals with updating supported rates for
> an already existing station fails to inform the rate control module about the
> new rates. as i don't know how to fix this (minstrel does not have an update
> function), i have just added a comment for now.
>
> Signed-off-by: Bruno Randolf <[email protected]>

This seems like a stable fix, if it applies can you please resend with a

Cc: [email protected]

on the commit log entry right below your own SOB.

Luis

2010-02-24 23:43:54

by Adam Wozniak

[permalink] [raw]
Subject: Re: [PATCH] mac80211: fix rates setup on IBSS merge

I noticed this a while back and tried (unsuccessfully) to get some
patches pushed through for it. Your patch looks cleaner than mine (I
don't think I understood as much at the time as I do now).

You can look through the linux-wireless archives for some of my old
posts. Things to note:

For minstrel, calling rate_control_init() is effectively an update. A
workable solution may be to add a minstrel rate_control_update() that
simply calls init(). I don't *think* PID has an update either, and
don't forget about the cases where the rate control algorithm is handled
in the firmware for the device.

Flow control in ieee80211_rx_bss_info() could use some cleanup. It's
possible to reach the end and use uninitialized variables.

To solve the problem for my devices, I used a patch to minstrel which
actually ignores the rates in the beacon, under the assumption that if
the other STA can't use them, they'll get 0% probability in the rate
table. This seems to work well for me.

--Adam

Bruno Randolf wrote:
> when an IBSS merge happened, the supported rates for the newly added station
> were left empty, causing the rate control module to be initialized with only
> the basic rates.
>
> also the section of the ibss code which deals with updating supported rates for
> an already existing station fails to inform the rate control module about the
> new rates. as i don't know how to fix this (minstrel does not have an update
> function), i have just added a comment for now.
>
> Signed-off-by: Bruno Randolf <[email protected]>
> ---
>
> net/mac80211/ibss.c | 3 +++
> 1 files changed, 3 insertions(+), 0 deletions(-)
>
> diff --git a/net/mac80211/ibss.c b/net/mac80211/ibss.c
> index f3e9424..b840d90 100644
> --- a/net/mac80211/ibss.c
> +++ b/net/mac80211/ibss.c
> @@ -276,6 +276,8 @@ static void ieee80211_rx_bss_info(struct ieee80211_sub_if_data *sdata,
> (unsigned long long) sta->sta.supp_rates[band]);
> #endif
> rcu_read_unlock();
> +
> + /* FIXME: update rate control */
> } else {
> rcu_read_unlock();
> ieee80211_ibss_add_sta(sdata, mgmt->bssid, mgmt->sa,
> @@ -370,6 +372,7 @@ static void ieee80211_rx_bss_info(struct ieee80211_sub_if_data *sdata,
> sdata->name, mgmt->bssid);
> #endif
> ieee80211_sta_join_ibss(sdata, bss);
> + supp_rates = ieee80211_sta_get_rates(local, elems, band);
> ieee80211_ibss_add_sta(sdata, mgmt->bssid, mgmt->sa,
> supp_rates, GFP_KERNEL);
> }
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>


2010-02-24 23:56:44

by Benoit Papillault

[permalink] [raw]
Subject: Re: [PATCH] mac80211: fix rates setup on IBSS merge

Luis R. Rodriguez a écrit :
> On Tue, Feb 23, 2010 at 1:51 AM, Bruno Randolf <[email protected]> wrote:
>
>> when an IBSS merge happened, the supported rates for the newly added station
>> were left empty, causing the rate control module to be initialized with only
>> the basic rates.
>>
>> also the section of the ibss code which deals with updating supported rates for
>> an already existing station fails to inform the rate control module about the
>> new rates. as i don't know how to fix this (minstrel does not have an update
>> function), i have just added a comment for now.
>>
>> Signed-off-by: Bruno Randolf <[email protected]>
>>
>
> This seems like a stable fix, if it applies can you please resend with a
>
> Cc: [email protected]
>
> on the commit log entry right below your own SOB.
>
> Luis
>
Hi Bruno,

I think the root cause is that IBSS neighbor entries are added whenever
we received any packet from a neighbor. However, the supported rates are
only available in the beacon/probe responses. I think one fix is to only
add IBSS neighbors on beacon/probe response reception (and moreover,
beacon/probe responses contains the timestamp that is needed for IBSS
merge).

Basically, I removed the call to ieee80211_ibss_add_sta in
net/mac80211/rx.c.

Can you try that and tell me about the result?

In fact, I did this change to have HT rates in IBSS mode (but I'm still
missing some stuff).

Regards,
Benoit

2010-03-01 21:17:34

by Benoit Papillault

[permalink] [raw]
Subject: Re: [PATCH] mac80211: fix rates setup on IBSS merge

Bruno Randolf a écrit :
> On Thursday 25 February 2010 08:56:40 Benoit PAPILLAULT wrote:
>
>> Luis R. Rodriguez a écrit :
>>
>>> On Tue, Feb 23, 2010 at 1:51 AM, Bruno Randolf <[email protected]> wrote:
>>>
>>>> when an IBSS merge happened, the supported rates for the newly added
>>>> station were left empty, causing the rate control module to be
>>>> initialized with only the basic rates.
>>>>
>>>> also the section of the ibss code which deals with updating supported
>>>> rates for an already existing station fails to inform the rate control
>>>> module about the new rates. as i don't know how to fix this (minstrel
>>>> does not have an update function), i have just added a comment for now.
>>>>
>>>> Signed-off-by: Bruno Randolf <[email protected]>
>>>>
>>> This seems like a stable fix, if it applies can you please resend with a
>>>
>>> Cc: [email protected]
>>>
>>> on the commit log entry right below your own SOB.
>>>
>>> Luis
>>>
>> Hi Bruno,
>>
>> I think the root cause is that IBSS neighbor entries are added whenever
>> we received any packet from a neighbor. However, the supported rates are
>> only available in the beacon/probe responses. I think one fix is to only
>> add IBSS neighbors on beacon/probe response reception (and moreover,
>> beacon/probe responses contains the timestamp that is needed for IBSS
>> merge).
>>
>> Basically, I removed the call to ieee80211_ibss_add_sta in
>> net/mac80211/rx.c.
>>
>
> i think we need to keep that.
>
> it can happen in normal IBSS operation, that we get a data packet from a node
> which we did not yet receive a beacon from. for example: he might be part of
> the IBSS, correctly merged and all, but have deferred from sending a beacon
> the last couple of beacon intervals because some other node sent a beacon
> first. or we might have missed his beacons due to interference. nevertheless
> we should try to be able to communicate with him.
>
> also there are many broken implementation out there (prominent example:
> madwifi) which fail to send beacons at some times. in order to communicate
> with then we also need to add stations on receiving data packets.
>
> so i think the proper way is to add an update function to the rate control
> module. (i will try to do that and send in another patch.)
>
> bruno
>
>
For 802.11n for instance, only the beacons/probe responses say the
sender is 802.11n. If we received a data packets from an unknown node,
maybe we could simply do a probe request in order to get its probe
response (despite the fact the 802.11 requires that only one node
replies to probe request in IBSS, this is impossible to implement in
practice). This will overcome if the sender node has not sent beacons
for various reason (including shared beaconing).

Just my 2 cents.

Regards,
Benoit

2010-03-01 08:48:46

by Bruno Randolf

[permalink] [raw]
Subject: Re: [PATCH] mac80211: fix rates setup on IBSS merge

On Thursday 25 February 2010 08:56:40 Benoit PAPILLAULT wrote:
> Luis R. Rodriguez a écrit :
> > On Tue, Feb 23, 2010 at 1:51 AM, Bruno Randolf <[email protected]> wrote:
> >> when an IBSS merge happened, the supported rates for the newly added
> >> station were left empty, causing the rate control module to be
> >> initialized with only the basic rates.
> >>
> >> also the section of the ibss code which deals with updating supported
> >> rates for an already existing station fails to inform the rate control
> >> module about the new rates. as i don't know how to fix this (minstrel
> >> does not have an update function), i have just added a comment for now.
> >>
> >> Signed-off-by: Bruno Randolf <[email protected]>
> >
> > This seems like a stable fix, if it applies can you please resend with a
> >
> > Cc: [email protected]
> >
> > on the commit log entry right below your own SOB.
> >
> > Luis
>
> Hi Bruno,
>
> I think the root cause is that IBSS neighbor entries are added whenever
> we received any packet from a neighbor. However, the supported rates are
> only available in the beacon/probe responses. I think one fix is to only
> add IBSS neighbors on beacon/probe response reception (and moreover,
> beacon/probe responses contains the timestamp that is needed for IBSS
> merge).
>
> Basically, I removed the call to ieee80211_ibss_add_sta in
> net/mac80211/rx.c.

i think we need to keep that.

it can happen in normal IBSS operation, that we get a data packet from a node
which we did not yet receive a beacon from. for example: he might be part of
the IBSS, correctly merged and all, but have deferred from sending a beacon
the last couple of beacon intervals because some other node sent a beacon
first. or we might have missed his beacons due to interference. nevertheless
we should try to be able to communicate with him.

also there are many broken implementation out there (prominent example:
madwifi) which fail to send beacons at some times. in order to communicate
with then we also need to add stations on receiving data packets.

so i think the proper way is to add an update function to the rate control
module. (i will try to do that and send in another patch.)

bruno