Skip to contentSkip to conversation

mac80211: fix memory leak on filtered powersave frames

PATCH In review Felix Fietkau 3 participants 5 messages Nov 11, 2020 – Oct 11, 2021 mac80211

lore
Share & tools
lore
#0 lore raw

PATCH 1/3 mac80211: fix memory leak on filtered powersave frames

After the status rework, ieee80211_tx_status_ext is leaking un-acknowledged packets for stations in powersave mode.
To fix this, move the code handling those packets from __ieee80211_tx_status into ieee80211_tx_status_ext

Fixes 3318111cf63d ("mac80211: reduce duplication in tx status functions")
---
 net/mac80211/status.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
net/mac80211/status.c +8 -10
index 6feb45135020..3485610755ef 100644
@@ -49,7 +49,8 @@ static void ieee80211_handle_filtered_frame(struct ieee80211_local *local,
 	int ac;
 
 	if (info->flags & (IEEE80211_TX_CTL_NO_PS_BUFFER |
-			   IEEE80211_TX_CTL_AMPDU)) {
+			   IEEE80211_TX_CTL_AMPDU |
+			   IEEE80211_TX_CTL_HW_80211_ENCAP)) {
 		ieee80211_free_txskb(&local->hw, skb);
 		return;
 	}
@@ -915,15 +916,6 @@ static void __ieee80211_tx_status(struct ieee80211_hw *hw,
 			ieee80211_mpsp_trigger_process(
 				ieee80211_get_qos_ctl(hdr), sta, true, acked);
 
-		if (!acked && test_sta_flag(sta, WLAN_STA_PS_STA)) {
-			/*
-			 * The STA is in power save mode, so assume
-			 * that this TX packet failed because of that.
-			 */
-			ieee80211_handle_filtered_frame(local, sta, skb);
-			return;
-		}
-
 		if (ieee80211_hw_check(&local->hw, HAS_RATE_CONTROL) &&
 		    (ieee80211_is_data(hdr->frame_control)) &&
 		    (rates_idx != -1))
@@ -1150,6 +1142,12 @@ void ieee80211_tx_status_ext(struct ieee80211_hw *hw,
 							    -info->status.ack_signal);
 				}
 			} else if (test_sta_flag(sta, WLAN_STA_PS_STA)) {
+				/*
+				 * The STA is in power save mode, so assume
+				 * that this TX packet failed because of that.
+				 */
+				if (skb)
+					ieee80211_handle_filtered_frame(local, sta, skb);
 				return;
 			} else if (noack_success) {
 				/* nothing to do here, do not account as lost */
Signature · 1 line
2.28.0
#1 lore raw

PATCH 2/3 mac80211: minstrel: remove deferred sampling code

Deferring sampling attempts to the second stage has some bad interactions with drivers that process the rate table in hardware and use the probe flag to indicate probing packets (e.g. most mt76 drivers). On affected drivers it can lead to probing not working at all.

If the link conditions turn worse, it might not be such a good idea to do a lot of sampling for lower rates in this case.

Fix this by simply skipping the sample attempt instead of deferring it, but keep the checks that would allow it to be sampled if it was skipped too often, but only if it has less than 95% success probability.

Also ensure that IEEE80211_TX_CTL_RATE_CTRL_PROBE is set for all probing packets.

Fixes cccf129f820e ("mac80211: add the 'minstrel' rate control algorithm")
---
 net/mac80211/rc80211_minstrel.c | 25 ++++---------------------
 net/mac80211/rc80211_minstrel.h |  1 -
 2 files changed, 4 insertions(+), 22 deletions(-)
net/mac80211/rc80211_minstrel.c +4 -21
index 86bc469a28bc..a8a940e97a9a 100644
@@ -287,12 +287,6 @@ minstrel_tx_status(void *priv, struct ieee80211_supported_band *sband,
 			mi->r[ndx].stats.success += success;
 	}
 
-	if ((info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE) && (i >= 0))
-		mi->sample_packets++;
-
-	if (mi->sample_deferred > 0)
-		mi->sample_deferred--;
-
 	if (time_after(jiffies, mi->last_stats_update +
 				mp->update_interval / (mp->new_avg ? 2 : 1)))
 		minstrel_update_stats(mp, mi);
@@ -367,7 +361,7 @@ minstrel_get_rate(void *priv, struct ieee80211_sta *sta,
 		return;
 
 	delta = (mi->total_packets * sampling_ratio / 100) -
-			(mi->sample_packets + mi->sample_deferred / 2);
+			mi->sample_packets;
 
 	/* delta < 0: no sampling required */
 	prev_sample = mi->prev_sample;
@@ -376,7 +370,6 @@ minstrel_get_rate(void *priv, struct ieee80211_sta *sta,
 		return;
 
 	if (mi->total_packets >= 10000) {
-		mi->sample_deferred = 0;
 		mi->sample_packets = 0;
 		mi->total_packets = 0;
 	} else if (delta > mi->n_rates * 2) {
@@ -401,19 +394,8 @@ minstrel_get_rate(void *priv, struct ieee80211_sta *sta,
 	 * rate sampling method should be used.
 	 * Respect such rates that are not sampled for 20 interations.
 	 */
-	if (mrr_capable &&
-	    msr->perfect_tx_time > mr->perfect_tx_time &&
-	    msr->stats.sample_skipped < 20) {
-		/* Only use IEEE80211_TX_CTL_RATE_CTRL_PROBE to mark
-		 * packets that have the sampling rate deferred to the
-		 * second MRR stage. Increase the sample counter only
-		 * if the deferred sample rate was actually used.
-		 * Use the sample_deferred counter to make sure that
-		 * the sampling is not done in large bursts */
-		info->flags |= IEEE80211_TX_CTL_RATE_CTRL_PROBE;
-		rate++;
-		mi->sample_deferred++;
-	} else {
+	if (msr->perfect_tx_time < mr->perfect_tx_time ||
+	    msr->stats.sample_skipped >= 20) {
 		if (!msr->sample_limit)
 			return;
 
@@ -433,6 +415,7 @@ minstrel_get_rate(void *priv, struct ieee80211_sta *sta,
 
 	rate->idx = mi->r[ndx].rix;
 	rate->count = minstrel_get_retry_count(&mi->r[ndx], info);
+	info->flags |= IEEE80211_TX_CTL_RATE_CTRL_PROBE;
 }
 
 
net/mac80211/rc80211_minstrel.h +0 -1
index dbb43bcd3c45..86cd80b3ffde 100644
@@ -126,7 +126,6 @@ struct minstrel_sta_info {
 	u8 max_prob_rate;
 	unsigned int total_packets;
 	unsigned int sample_packets;
-	int sample_deferred;
 
 	unsigned int sample_row;
 	unsigned int sample_column;
Signature · 1 line
2.28.0
#2 lore raw

Re: [PATCH 5.10 2/3] mac80211: minstrel: remove deferred sampling code

Hi,

I tracked down an issue with using the Kernel 4.9.277 and 5.4.131 back to this patch.

In 802.11bg mode the effective transmit rate drops to 3MBit/s. With that patch and perfect SNR (> 45dB) I get and rc_stats:

*********************
best __________rate_________ ________statistics________ ________last_______ ______sum-of________
rate [name idx airtime max_tp] [avg(tp) avg(prob) sd(prob)] [prob.|retry|suc|att] [#success | #attempts]

        1     0    9738    0.9       0.9      93.7      0.5      75.0 
 1     3 4           562   572
        2     1    4922    1.6       1.6      97.0      1.5     100.0 
 1     2 2           559   576
        5.5   2    1858    4.7       4.7      99.9      0.0     100.0 
 2     2 2           542   565
       11     3     982    9.1       9.1      94.9      1.2     100.0 
 4     3 3           560   578
        6     4    1648    5.3       5.3      95.3      1.3     100.0 
 3     3 3           538   595
        9     5    1112    8.0       8.0      97.5      1.8     100.0 
 4     2 2           560   596
       12     6     844   10.5      10.5      92.4      2.0      75.0 
 5     3 4           572   615
       18     7     576   15.5      15.4      89.0      1.4      66.6 
 5     2 3           559   609
   D   24     8     440   20.4      20.4      99.4      1.1     100.0 
 6     2 2           560   604
  C    36     9     308   29.1      29.1      98.7      1.6     100.0 
 6     2 2           565   600
 B     48    10     240   37.3      36.3      87.3      1.4      66.6 
 6     2 3           608   671
A   P  54    11     216   41.6      41.6      97.3      1.7     100.0 6     3 3           565   620

Total packet count:: ideal 6111 lookaround 652
*********************

After reverting that patch the transmit rate is back to 20MBit/s:
*********************
best __________rate_________ ________statistics________ ________last_______ ______sum-of________
rate [name idx airtime max_tp] [avg(tp) avg(prob) sd(prob)] [prob.|retry|suc|att] [#success | #attempts]

        1     0    9738    0.9       0.9      99.8      0.6     100.0 
 1     0 0            41   44
        2     1    4922    1.6       1.6     100.0      0.0     100.0 
 1     0 0            45   45
        5.5   2    1858    4.7       4.7      99.9      0.0     100.0 
 2     0 0            43   44
       11     3     982    9.1       9.1      97.0      1.4     100.0 
 4     0 0            44   45
        6     4    1648    5.3       5.3      99.9      0.0     100.0 
 3     0 0            43   44
        9     5    1112    8.0       8.0      99.9      0.0     100.0 
 4     0 0            45   47
       12     6     844   10.5       8.8      74.9      0.7       0.0 
 5     0 0            43   47
       18     7     576   15.5      15.5     100.0      0.0     100.0 
 5     0 0            42   42
   D   24     8     440   20.4      20.4      99.9      0.0     100.0 
 6     0 0            60   62
  C P  36     9     308   29.1      29.1      99.9      0.0     100.0 
 6     0 0            44   46
A      48    10     240   37.3      37.3      93.8      1.2     100.0 6     1 1           587   618
 B     54    11     216   41.6      32.3      69.8      2.1      33.3 
 6     0 0         24506   25849

Total packet count:: ideal 4962 lookaround 552
*********************

I'm using the ath9k driver with a sparklan WPEA-121N card.

For my case reverting that patch is enough.
What I see is that IEEE80211_TX_CTL_RATE_CTRL_PROBE is set but not used anywhere in the code. Maybe there is an easy and better fix than just reverting.

Regards,

Thomas

Am 11.11.2020 um 19:33 schrieb Felix Fietkau:

Quoted 100 lines

Deferring sampling attempts to the second stage has some bad interactions with drivers that process the rate table in hardware and use the probe flag to indicate probing packets (e.g. most mt76 drivers). On affected drivers it can lead to probing not working at all.

If the link conditions turn worse, it might not be such a good idea to do a lot of sampling for lower rates in this case.

Fix this by simply skipping the sample attempt instead of deferring it, but keep the checks that would allow it to be sampled if it was skipped too often, but only if it has less than 95% success probability.

Also ensure that IEEE80211_TX_CTL_RATE_CTRL_PROBE is set for all probing packets.

Fixes cccf129f820e ("mac80211: add the 'minstrel' rate control algorithm")
---
  net/mac80211/rc80211_minstrel.c | 25 ++++---------------------
  net/mac80211/rc80211_minstrel.h |  1 -
  2 files changed, 4 insertions(+), 22 deletions(-)
net/mac80211/rc80211_minstrel.c +4 -21
index 86bc469a28bc..a8a940e97a9a 100644
@@ -287,12 +287,6 @@ minstrel_tx_status(void *priv, struct ieee80211_supported_band *sband,
  			mi->r[ndx].stats.success += success;
  	}
 
-	if ((info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE) && (i >= 0))
-		mi->sample_packets++;
-
-	if (mi->sample_deferred > 0)
-		mi->sample_deferred--;
-
  	if (time_after(jiffies, mi->last_stats_update +
  				mp->update_interval / (mp->new_avg ? 2 : 1)))
  		minstrel_update_stats(mp, mi);
@@ -367,7 +361,7 @@ minstrel_get_rate(void *priv, struct ieee80211_sta *sta,
  		return;
 
  	delta = (mi->total_packets * sampling_ratio / 100) -
-			(mi->sample_packets + mi->sample_deferred / 2);
+			mi->sample_packets;
 
  	/* delta < 0: no sampling required */
  	prev_sample = mi->prev_sample;
@@ -376,7 +370,6 @@ minstrel_get_rate(void *priv, struct ieee80211_sta *sta,
  		return;
 
  	if (mi->total_packets >= 10000) {
-		mi->sample_deferred = 0;
  		mi->sample_packets = 0;
  		mi->total_packets = 0;
  	} else if (delta > mi->n_rates * 2) {
@@ -401,19 +394,8 @@ minstrel_get_rate(void *priv, struct ieee80211_sta *sta,
  	 * rate sampling method should be used.
  	 * Respect such rates that are not sampled for 20 interations.
  	 */
-	if (mrr_capable &&
-	    msr->perfect_tx_time > mr->perfect_tx_time &&
-	    msr->stats.sample_skipped < 20) {
-		/* Only use IEEE80211_TX_CTL_RATE_CTRL_PROBE to mark
-		 * packets that have the sampling rate deferred to the
-		 * second MRR stage. Increase the sample counter only
-		 * if the deferred sample rate was actually used.
-		 * Use the sample_deferred counter to make sure that
-		 * the sampling is not done in large bursts */
-		info->flags |= IEEE80211_TX_CTL_RATE_CTRL_PROBE;
-		rate++;
-		mi->sample_deferred++;
-	} else {
+	if (msr->perfect_tx_time < mr->perfect_tx_time ||
+	    msr->stats.sample_skipped >= 20) {
  		if (!msr->sample_limit)
  			return;
 
@@ -433,6 +415,7 @@ minstrel_get_rate(void *priv, struct ieee80211_sta *sta,
 
  	rate->idx = mi->r[ndx].rix;
  	rate->count = minstrel_get_retry_count(&mi->r[ndx], info);
+	info->flags |= IEEE80211_TX_CTL_RATE_CTRL_PROBE;
  }
 
 
net/mac80211/rc80211_minstrel.h +0 -1
index dbb43bcd3c45..86cd80b3ffde 100644
@@ -126,7 +126,6 @@ struct minstrel_sta_info {
  	u8 max_prob_rate;
  	unsigned int total_packets;
  	unsigned int sample_packets;
-	int sample_deferred;
 
  	unsigned int sample_row;
  	unsigned int sample_column;
#3 lore raw

Re: [PATCH 5.10 2/3] mac80211: minstrel: remove deferred sampling code

Quoted 73 lines · Thomas Graf

I tracked down an issue with using the Kernel 4.9.277 and 5.4.131 back to this patch.

In 802.11bg mode the effective transmit rate drops to 3MBit/s. With that patch and perfect SNR (> 45dB) I get and rc_stats:

*********************
best __________rate_________ ________statistics________
________last_______ ______sum-of________

rate  [name idx airtime max_tp]  [avg(tp) avg(prob) sd(prob)]
[prob.|retry|suc|att]  [#success | #attempts]
        1     0    9738    0.9       0.9      93.7      0.5      75.0
1     3 4           562   572
        2     1    4922    1.6       1.6      97.0      1.5     100.0
1     2 2           559   576
        5.5   2    1858    4.7       4.7      99.9      0.0     100.0
2     2 2           542   565
       11     3     982    9.1       9.1      94.9      1.2     100.0
4     3 3           560   578
        6     4    1648    5.3       5.3      95.3      1.3     100.0
3     3 3           538   595
        9     5    1112    8.0       8.0      97.5      1.8     100.0
4     2 2           560   596
       12     6     844   10.5      10.5      92.4      2.0      75.0
5     3 4           572   615
       18     7     576   15.5      15.4      89.0      1.4      66.6
5     2 3           559   609
   D   24     8     440   20.4      20.4      99.4      1.1     100.0
6     2 2           560   604
  C    36     9     308   29.1      29.1      98.7      1.6     100.0
6     2 2           565   600
 B     48    10     240   37.3      36.3      87.3      1.4      66.6
6     2 3           608   671
A   P  54    11     216   41.6      41.6      97.3      1.7     100.0
6     3 3           565   620

Total packet count:: ideal 6111 lookaround 652
*********************

After reverting that patch the transmit rate is back to 20MBit/s:
*********************
best __________rate_________ ________statistics________
________last_______ ______sum-of________

rate  [name idx airtime max_tp]  [avg(tp) avg(prob) sd(prob)]
[prob.|retry|suc|att]  [#success | #attempts]
        1     0    9738    0.9       0.9      99.8      0.6     100.0
1     0 0            41   44
        2     1    4922    1.6       1.6     100.0      0.0     100.0
1     0 0            45   45
        5.5   2    1858    4.7       4.7      99.9      0.0     100.0
2     0 0            43   44
       11     3     982    9.1       9.1      97.0      1.4     100.0
4     0 0            44   45
        6     4    1648    5.3       5.3      99.9      0.0     100.0
3     0 0            43   44
        9     5    1112    8.0       8.0      99.9      0.0     100.0
4     0 0            45   47
       12     6     844   10.5       8.8      74.9      0.7       0.0
5     0 0            43   47
       18     7     576   15.5      15.5     100.0      0.0     100.0
5     0 0            42   42
   D   24     8     440   20.4      20.4      99.9      0.0     100.0
6     0 0            60   62
  C P  36     9     308   29.1      29.1      99.9      0.0     100.0
6     0 0            44   46
A      48    10     240   37.3      37.3      93.8      1.2     100.0
6     1 1           587   618
 B     54    11     216   41.6      32.3      69.8      2.1      33.3
6     0 0         24506   25849

Total packet count:: ideal 4962 lookaround 552
*********************

Quoted 6 lines

I'm using the ath9k driver with a sparklan WPEA-121N card.

For my case reverting that patch is enough.
What I see is that IEEE80211_TX_CTL_RATE_CTRL_PROBE is set but not used anywhere in the code. Maybe there is an easy and better fix than just reverting.

I lost track, is this regression resolved now or is it still open?

Signature · 3 lines
https://patchwork.kernel.org/project/linux-wireless/list/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
#4 lore raw

PATCH 3/3 mac80211: minstrel: fix tx status processing corner case

In reply to Felix Fietkau

Some drivers fill the status rate list without setting the rate index after the final rate to -1. minstrel_ht already deals with this, but minstrel doesn't, which causes it to get stuck at the lowest rate on these drivers.

Fix this by checking the count as well.

Fixes cccf129f820e ("mac80211: add the 'minstrel' rate control algorithm")
---
 net/mac80211/rc80211_minstrel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/mac80211/rc80211_minstrel.c +1 -1
index a8a940e97a9a..b13b1da19386 100644
@@ -274,7 +274,7 @@ minstrel_tx_status(void *priv, struct ieee80211_supported_band *sband,
 	success = !!(info->flags & IEEE80211_TX_STAT_ACK);
 
 	for (i = 0; i < IEEE80211_TX_MAX_RATES; i++) {
-		if (ar[i].idx < 0)
+		if (ar[i].idx < 0 || !ar[i].count)
 			break;
 
 		ndx = rix_to_ndx(mi, ar[i].idx);
Signature · 1 line
2.28.0