[PATCH net] net/sched: sch_red: annotate data-races in red_dump_stats()

[PATCH net] net/sched: sch_red: annotate data-races in red_dump_stats()

[Eric Dumazet]

red_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_red_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>
---
 net/sched/sch_red.c | 31 +++++++++++++++++++++----------
 1 file changed, 21 insertions(+), 10 deletions(-)

diff --git a/net/sched/sch_red.c b/net/sched/sch_red.c
index c8d3d09f15e3919d6468964561130bfc79fb215b..432b8a3000a57b4688b3ddb5501f604d5752c67c 100644
--- a/net/sched/sch_red.c
+++ b/net/sched/sch_red.c
@@ -90,17 +90,20 @@ static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 	case RED_PROB_MARK:
 		qdisc_qstats_overlimit(sch);
 		if (!red_use_ecn(q)) {
-			q->stats.prob_drop++;
+			WRITE_ONCE(q->stats.prob_drop,
+				   q->stats.prob_drop + 1);
 			goto congestion_drop;
 		}
 
 		if (INET_ECN_set_ce(skb)) {
-			q->stats.prob_mark++;
+			WRITE_ONCE(q->stats.prob_mark,
+				   q->stats.prob_mark + 1);
 			skb = tcf_qevent_handle(&q->qe_mark, sch, skb, to_free, &ret);
 			if (!skb)
 				return NET_XMIT_CN | ret;
 		} else if (!red_use_nodrop(q)) {
-			q->stats.prob_drop++;
+			WRITE_ONCE(q->stats.prob_drop,
+				   q->stats.prob_drop + 1);
 			goto congestion_drop;
 		}
 
@@ -111,17 +114,20 @@ static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 		reason = QDISC_DROP_OVERLIMIT;
 		qdisc_qstats_overlimit(sch);
 		if (red_use_harddrop(q) || !red_use_ecn(q)) {
-			q->stats.forced_drop++;
+			WRITE_ONCE(q->stats.forced_drop,
+				   q->stats.forced_drop + 1);
 			goto congestion_drop;
 		}
 
 		if (INET_ECN_set_ce(skb)) {
-			q->stats.forced_mark++;
+			WRITE_ONCE(q->stats.forced_mark,
+				   q->stats.forced_mark + 1);
 			skb = tcf_qevent_handle(&q->qe_mark, sch, skb, to_free, &ret);
 			if (!skb)
 				return NET_XMIT_CN | ret;
 		} else if (!red_use_nodrop(q)) {
-			q->stats.forced_drop++;
+			WRITE_ONCE(q->stats.forced_drop,
+				   q->stats.forced_drop + 1);
 			goto congestion_drop;
 		}
 
@@ -135,7 +141,8 @@ static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 		sch->qstats.backlog += len;
 		sch->q.qlen++;
 	} else if (net_xmit_drop_count(ret)) {
-		q->stats.pdrop++;
+		WRITE_ONCE(q->stats.pdrop,
+			   q->stats.pdrop + 1);
 		qdisc_qstats_drop(sch);
 	}
 	return ret;
@@ -463,9 +470,13 @@ static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 		dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED,
 					      &hw_stats_request);
 	}
-	st.early = q->stats.prob_drop + q->stats.forced_drop;
-	st.pdrop = q->stats.pdrop;
-	st.marked = q->stats.prob_mark + q->stats.forced_mark;
+	st.early = READ_ONCE(q->stats.prob_drop) +
+		   READ_ONCE(q->stats.forced_drop);
+
+	st.pdrop = READ_ONCE(q->stats.pdrop);
+
+	st.marked = READ_ONCE(q->stats.prob_mark) +
+		    READ_ONCE(q->stats.forced_mark);
 
 	return gnet_stats_copy_app(d, &st, sizeof(st));
 }
-- 
2.54.0.rc2.533.g4f5dca5207-goog