[PATCH net] net/sched: sch_pie: annotate data-races in pie_dump_stats()

[PATCH net] net/sched: sch_pie: annotate data-races in pie_dump_stats()

[Eric Dumazet]

pie_dump_stats() only runs with RTNL held,
reading fields that can be changed in qdisc fast path.

Add READ_ONCE()/WRITE_ONCE() annotations.

Alternative would be to acquire the qdisc spinlock, but our long-term
goal is to make qdisc dump operations lockless as much as we can.

tc_pie_xstats fields don't need to be latched atomically,
otherwise this bug would have been caught earlier.

Fixes: edb09eb17ed8 ("net: sched: do not acquire qdisc spinlock in qdisc/class stats dump")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jamal Hadi Salim <jhs@mojatatu.com>
---
 include/net/pie.h   |  2 +-
 net/sched/sch_pie.c | 38 +++++++++++++++++++-------------------
 2 files changed, 20 insertions(+), 20 deletions(-)

diff --git a/include/net/pie.h b/include/net/pie.h
index 01cbc66825a40bd21c0a044b1180cbbc346785df..1f3db0c355149b41823a891c9156cac625122031 100644
--- a/include/net/pie.h
+++ b/include/net/pie.h
@@ -104,7 +104,7 @@ static inline void pie_vars_init(struct pie_vars *vars)
 	vars->dq_tstamp = DTIME_INVALID;
 	vars->accu_prob = 0;
 	vars->dq_count = DQCOUNT_INVALID;
-	vars->avg_dq_rate = 0;
+	WRITE_ONCE(vars->avg_dq_rate, 0);
 }
 
 static inline struct pie_skb_cb *get_pie_cb(const struct sk_buff *skb)
diff --git a/net/sched/sch_pie.c b/net/sched/sch_pie.c
index 16f3f629cb8e4be71431f7e50a278e3c7fdba8d0..fb53fbf0e328571be72b66ba4e75a938e1963422 100644
--- a/net/sched/sch_pie.c
+++ b/net/sched/sch_pie.c
@@ -90,7 +90,7 @@ static int pie_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 	bool enqueue = false;
 
 	if (unlikely(qdisc_qlen(sch) >= sch->limit)) {
-		q->stats.overlimit++;
+		WRITE_ONCE(q->stats.overlimit, q->stats.overlimit + 1);
 		goto out;
 	}
 
@@ -104,7 +104,7 @@ static int pie_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 		/* If packet is ecn capable, mark it if drop probability
 		 * is lower than 10%, else drop it.
 		 */
-		q->stats.ecn_mark++;
+		WRITE_ONCE(q->stats.ecn_mark, q->stats.ecn_mark + 1);
 		enqueue = true;
 	}
 
@@ -114,15 +114,15 @@ static int pie_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 		if (!q->params.dq_rate_estimator)
 			pie_set_enqueue_time(skb);
 
-		q->stats.packets_in++;
+		WRITE_ONCE(q->stats.packets_in, q->stats.packets_in + 1);
 		if (qdisc_qlen(sch) > q->stats.maxq)
-			q->stats.maxq = qdisc_qlen(sch);
+			WRITE_ONCE(q->stats.maxq, qdisc_qlen(sch));
 
 		return qdisc_enqueue_tail(skb, sch);
 	}
 
 out:
-	q->stats.dropped++;
+	WRITE_ONCE(q->stats.dropped, q->stats.dropped + 1);
 	q->vars.accu_prob = 0;
 	return qdisc_drop_reason(skb, sch, to_free, reason);
 }
@@ -267,11 +267,11 @@ void pie_process_dequeue(struct sk_buff *skb, struct pie_params *params,
 			count = count / dtime;
 
 			if (vars->avg_dq_rate == 0)
-				vars->avg_dq_rate = count;
+				WRITE_ONCE(vars->avg_dq_rate, count);
 			else
-				vars->avg_dq_rate =
+				WRITE_ONCE(vars->avg_dq_rate,
 				    (vars->avg_dq_rate -
-				     (vars->avg_dq_rate >> 3)) + (count >> 3);
+				     (vars->avg_dq_rate >> 3)) + (count >> 3));
 
 			/* If the queue has receded below the threshold, we hold
 			 * on to the last drain rate calculated, else we reset
@@ -381,7 +381,7 @@ void pie_calculate_probability(struct pie_params *params, struct pie_vars *vars,
 	if (delta > 0) {
 		/* prevent overflow */
 		if (vars->prob < oldprob) {
-			vars->prob = MAX_PROB;
+			WRITE_ONCE(vars->prob, MAX_PROB);
 			/* Prevent normalization error. If probability is at
 			 * maximum value already, we normalize it here, and
 			 * skip the check to do a non-linear drop in the next
@@ -392,7 +392,7 @@ void pie_calculate_probability(struct pie_params *params, struct pie_vars *vars,
 	} else {
 		/* prevent underflow */
 		if (vars->prob > oldprob)
-			vars->prob = 0;
+			WRITE_ONCE(vars->prob, 0);
 	}
 
 	/* Non-linear drop in probability: Reduce drop probability quickly if
@@ -403,7 +403,7 @@ void pie_calculate_probability(struct pie_params *params, struct pie_vars *vars,
 		/* Reduce drop probability to 98.4% */
 		vars->prob -= vars->prob / 64;
 
-	vars->qdelay = qdelay;
+	WRITE_ONCE(vars->qdelay, qdelay);
 	vars->backlog_old = backlog;
 
 	/* We restart the measurement cycle if the following conditions are met
@@ -502,21 +502,21 @@ static int pie_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 	struct pie_sched_data *q = qdisc_priv(sch);
 	struct tc_pie_xstats st = {
 		.prob		= q->vars.prob << BITS_PER_BYTE,
-		.delay		= ((u32)PSCHED_TICKS2NS(q->vars.qdelay)) /
+		.delay		= ((u32)PSCHED_TICKS2NS(READ_ONCE(q->vars.qdelay))) /
 				   NSEC_PER_USEC,
-		.packets_in	= q->stats.packets_in,
-		.overlimit	= q->stats.overlimit,
-		.maxq		= q->stats.maxq,
-		.dropped	= q->stats.dropped,
-		.ecn_mark	= q->stats.ecn_mark,
+		.packets_in	= READ_ONCE(q->stats.packets_in),
+		.overlimit	= READ_ONCE(q->stats.overlimit),
+		.maxq		= READ_ONCE(q->stats.maxq),
+		.dropped	= READ_ONCE(q->stats.dropped),
+		.ecn_mark	= READ_ONCE(q->stats.ecn_mark),
 	};
 
 	/* avg_dq_rate is only valid if dq_rate_estimator is enabled */
 	st.dq_rate_estimating = q->params.dq_rate_estimator;
 
 	/* unscale and return dq_rate in bytes per sec */
-	if (q->params.dq_rate_estimator)
-		st.avg_dq_rate = q->vars.avg_dq_rate *
+	if (st.dq_rate_estimating)
+		st.avg_dq_rate = READ_ONCE(q->vars.avg_dq_rate) *
 				 (PSCHED_TICKS_PER_SEC) >> PIE_SCALE;
 
 	return gnet_stats_copy_app(d, &st, sizeof(st));
-- 
2.54.0.rc2.533.g4f5dca5207-goog