net_sched: drop packets after root qdisc lock is released

Qdisc performance suffers when packets are dropped at enqueue()
time because drops (kfree_skb()) are done while qdisc lock is held,
delaying a dequeue() draining the queue.

Nominal throughput can be reduced by 50 % when this happens,
at a time we would like the dequeue() to proceed as fast as possible.

Even FQ is vulnerable to this problem, while one of FQ goals was
to provide some flow isolation.

This patch adds a 'struct sk_buff **to_free' parameter to all
qdisc->enqueue(), and in qdisc_drop() helper.

I measured a performance increase of up to 12 %, but this patch
is a prereq so that future batches in enqueue() can fly.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 30 insertions, 11 deletions

diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
index 4f7cee8344c4..04e84c07c94f 100644
--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -37,8 +37,10 @@ struct qdisc_size_table {
 };
 
 struct Qdisc {
-	int 			(*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
-	struct sk_buff *	(*dequeue)(struct Qdisc *dev);
+	int 			(*enqueue)(struct sk_buff *skb,
+					   struct Qdisc *sch,
+					   struct sk_buff **to_free);
+	struct sk_buff *	(*dequeue)(struct Qdisc *sch);
 	unsigned int		flags;
 #define TCQ_F_BUILTIN		1
 #define TCQ_F_INGRESS		2
@@ -160,7 +162,9 @@ struct Qdisc_ops {
 	char			id[IFNAMSIZ];
 	int			priv_size;
 
-	int 			(*enqueue)(struct sk_buff *, struct Qdisc *);
+	int 			(*enqueue)(struct sk_buff *skb,
+					   struct Qdisc *sch,
+					   struct sk_buff **to_free);
 	struct sk_buff *	(*dequeue)(struct Qdisc *);
 	struct sk_buff *	(*peek)(struct Qdisc *);
 
@@ -498,10 +502,11 @@ static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
 #endif
 }
 
-static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
+static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
+				struct sk_buff **to_free)
 {
 	qdisc_calculate_pkt_len(skb, sch);
-	return sch->enqueue(skb, sch);
+	return sch->enqueue(skb, sch, to_free);
 }
 
 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
@@ -626,24 +631,36 @@ static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
 	return __qdisc_dequeue_head(sch, &sch->q);
 }
 
+/* Instead of calling kfree_skb() while root qdisc lock is held,
+ * queue the skb for future freeing at end of __dev_xmit_skb()
+ */
+static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
+{
+	skb->next = *to_free;
+	*to_free = skb;
+}
+
 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
-					      struct sk_buff_head *list)
+						   struct sk_buff_head *list,
+						   struct sk_buff **to_free)
 {
 	struct sk_buff *skb = __skb_dequeue(list);
 
 	if (likely(skb != NULL)) {
 		unsigned int len = qdisc_pkt_len(skb);
+
 		qdisc_qstats_backlog_dec(sch, skb);
-		kfree_skb(skb);
+		__qdisc_drop(skb, to_free);
 		return len;
 	}
 
 	return 0;
 }
 
-static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch)
+static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
+						 struct sk_buff **to_free)
 {
-	return __qdisc_queue_drop_head(sch, &sch->q);
+	return __qdisc_queue_drop_head(sch, &sch->q, to_free);
 }
 
 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
@@ -724,9 +741,11 @@ static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
 	qdisc_qstats_drop(sch);
 }
 
-static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
+
+static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
+			     struct sk_buff **to_free)
 {
-	kfree_skb(skb);
+	__qdisc_drop(skb, to_free);
 	qdisc_qstats_drop(sch);
 
 	return NET_XMIT_DROP;