Re: [PATCH 1/3] mac80211: fix - drop frames for sta with no valid rate

[Johannes Berg <johannes@sipsolutions.net>]

[-- Attachment #1: Type: text/plain, Size: 3600 bytes --]

On Tue, 2009-06-09 at 04:30 -0400, Luis R. Rodriguez wrote:
> If we're associated and scanning mac80211 will allow through
> nullfunc and probe request frames. When we're scanning on a
> different band than the one we're associated on we should
> not send nullfunc frames to the sta on that band as it would
> be the incorrect band.
> 
> Lets catch the case where no valid rate is usable when associated
> and discard those frames and warn only for the case the frame is
> not a nullfunc or probe request as those are the only accounted
> for frames mac80211 should allow through while scanning.
> 
> This fixes an oops which occured due to an assert in ath9k:
> 
> http://marc.info/?l=linux-wireless&m=124277331319024
> 
> The assert was happening because the rate control algorithm
> figures it should find at least one valid dual stream or
> single stream rate. Since we allow mac80211 to send get_rate
> callback for drivers for a sta on invalid band no valid will
> actually have been found and hence the assert.
> 
> Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
> ---
>  include/net/mac80211.h |   11 +++++++++++
>  net/mac80211/tx.c      |   38 ++++++++++++++++++++++++++++++++++++++
>  2 files changed, 49 insertions(+), 0 deletions(-)
> 
> diff --git a/include/net/mac80211.h b/include/net/mac80211.h
> index 17d61d1..9559efa 100644
> --- a/include/net/mac80211.h
> +++ b/include/net/mac80211.h
> @@ -2103,6 +2103,17 @@ rate_lowest_index(struct ieee80211_supported_band *sband,
>  	return 0;
>  }
>  
> +static inline
> +bool rate_usable_index_exists(struct ieee80211_supported_band *sband,
> +			      struct ieee80211_sta *sta)
> +{
> +	unsigned int i;
> +
> +	for (i = 0; i < sband->n_bitrates; i++)
> +		if (rate_supported(sta, sband->band, i))
> +			return true;
> +	return false;
> +}
>  
>  int ieee80211_rate_control_register(struct rate_control_ops *ops);
>  void ieee80211_rate_control_unregister(struct rate_control_ops *ops);
> diff --git a/net/mac80211/tx.c b/net/mac80211/tx.c
> index 1436f74..03f9a4e 100644
> --- a/net/mac80211/tx.c
> +++ b/net/mac80211/tx.c
> @@ -511,6 +511,7 @@ ieee80211_tx_h_rate_ctrl(struct ieee80211_tx_data *tx)
>  	int i, len;
>  	bool inval = false, rts = false, short_preamble = false;
>  	struct ieee80211_tx_rate_control txrc;
> +	u32 sta_flags;
>  
>  	memset(&txrc, 0, sizeof(txrc));
>  
> @@ -543,7 +544,44 @@ ieee80211_tx_h_rate_ctrl(struct ieee80211_tx_data *tx)
>  	     (tx->sta && test_sta_flags(tx->sta, WLAN_STA_SHORT_PREAMBLE))))
>  		txrc.short_preamble = short_preamble = true;
>  
> +	sta_flags = tx->sta ? get_sta_flags(tx->sta) : 0;
> +
> +	/*
> +	 * Lets not bother rate control if we're associated and cannot
> +	 * talk to the sta. It makes little since for this to happen,
> +	 * it should mean we're scanning on another band somehow some frames
> +	 * got through the TX queue -- these should have been not been added
> +	 * to our TX queue. There is one excemption to this, but we handle
> +	 * these below.
> +	 */
> +
> +	if (unlikely((tx->local->sw_scanning) &&
> +	    (sta_flags & WLAN_STA_ASSOC) &&
> +	    !rate_usable_index_exists(sband, &tx->sta->sta))) {
> +		/*
> +		 *  The only accounted for frames of this type in
> +		 *  mac80211 are probe requests and null func frames,
> +		 *  so just warn for other drop of frames. Drop the
> +		 *  frames anyway as we have no usable bit rate.
> +		 */

Can we not avoid this situation to start with by flushing the master
device queues before we go scan?

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 801 bytes --]