Merge branch 'sched_skb_free_defer'

Eric Dumazet says:

====================
net_sched: defer skb freeing while changing qdiscs

qdiscs/classes are changed under RTNL protection and often
while blocking BH and root qdisc spinlock.

When lots of skbs need to be dropped, we free
them under these locks causing TX/RX freezes,
and more generally latency spikes.

I saw spikes of 50+ ms on quite fast hardware...

This patch series adds a simple queue protected by RTNL
where skbs can be placed until RTNL is released.

Note that this might also serve in the future for optional
reinjection of packets when a qdisc is replaced.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

13 files changed, 72 insertions, 35 deletions

diff --git a/include/linux/rtnetlink.h b/include/linux/rtnetlink.h
index c006cc9..2daece8 100644
--- a/include/linux/rtnetlink.h
+++ b/include/linux/rtnetlink.h
@@ -89,8 +89,9 @@ void net_inc_egress_queue(void);
 void net_dec_egress_queue(void);
 #endif
 
-extern void rtnetlink_init(void);
-extern void __rtnl_unlock(void);
+void rtnetlink_init(void);
+void __rtnl_unlock(void);
+void rtnl_kfree_skbs(struct sk_buff *head, struct sk_buff *tail);
 
 #define ASSERT_RTNL() do { \
 	if (unlikely(!rtnl_is_locked())) { \
diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
index 9a0d177..4f7cee8 100644
--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -683,19 +683,21 @@ static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
 	return skb;
 }
 
-static inline void __qdisc_reset_queue(struct Qdisc *sch,
-				       struct sk_buff_head *list)
+static inline void __qdisc_reset_queue(struct sk_buff_head *list)
 {
 	/*
 	 * We do not know the backlog in bytes of this list, it
 	 * is up to the caller to correct it
 	 */
-	__skb_queue_purge(list);
+	if (!skb_queue_empty(list)) {
+		rtnl_kfree_skbs(list->next, list->prev);
+		__skb_queue_head_init(list);
+	}
 }
 
 static inline void qdisc_reset_queue(struct Qdisc *sch)
 {
-	__qdisc_reset_queue(sch, &sch->q);
+	__qdisc_reset_queue(&sch->q);
 	sch->qstats.backlog = 0;
 }
 
@@ -716,6 +718,12 @@ static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
 	return old;
 }
 
+static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
+{
+	rtnl_kfree_skbs(skb, skb);
+	qdisc_qstats_drop(sch);
+}
+
 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
 {
 	kfree_skb(skb);
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index d69c464..eb49ca2 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -71,9 +71,31 @@ void rtnl_lock(void)
 }
 EXPORT_SYMBOL(rtnl_lock);
 
+static struct sk_buff *defer_kfree_skb_list;
+void rtnl_kfree_skbs(struct sk_buff *head, struct sk_buff *tail)
+{
+	if (head && tail) {
+		tail->next = defer_kfree_skb_list;
+		defer_kfree_skb_list = head;
+	}
+}
+EXPORT_SYMBOL(rtnl_kfree_skbs);
+
 void __rtnl_unlock(void)
 {
+	struct sk_buff *head = defer_kfree_skb_list;
+
+	defer_kfree_skb_list = NULL;
+
 	mutex_unlock(&rtnl_mutex);
+
+	while (head) {
+		struct sk_buff *next = head->next;
+
+		kfree_skb(head);
+		cond_resched();
+		head = next;
+	}
 }
 
 void rtnl_unlock(void)
diff --git a/net/sched/sch_choke.c b/net/sched/sch_choke.c
index 04e0b05..789b69e 100644
--- a/net/sched/sch_choke.c
+++ b/net/sched/sch_choke.c
@@ -375,11 +375,11 @@ static void choke_reset(struct Qdisc *sch)
 		q->head = (q->head + 1) & q->tab_mask;
 		if (!skb)
 			continue;
-		qdisc_qstats_backlog_dec(sch, skb);
-		--sch->q.qlen;
-		qdisc_drop(skb, sch);
+		rtnl_qdisc_drop(skb, sch);
 	}
 
+	sch->q.qlen = 0;
+	sch->qstats.backlog = 0;
 	memset(q->tab, 0, (q->tab_mask + 1) * sizeof(struct sk_buff *));
 	q->head = q->tail = 0;
 	red_restart(&q->vars);
@@ -455,7 +455,7 @@ static int choke_change(struct Qdisc *sch, struct nlattr *opt)
 				dropped += qdisc_pkt_len(skb);
 				qdisc_qstats_backlog_dec(sch, skb);
 				--sch->q.qlen;
-				qdisc_drop(skb, sch);
+				rtnl_qdisc_drop(skb, sch);
 			}
 			qdisc_tree_reduce_backlog(sch, oqlen - sch->q.qlen, dropped);
 			q->head = 0;
diff --git a/net/sched/sch_codel.c b/net/sched/sch_codel.c
index dddf3bb..c5bc424 100644
--- a/net/sched/sch_codel.c
+++ b/net/sched/sch_codel.c
@@ -174,7 +174,7 @@ static int codel_change(struct Qdisc *sch, struct nlattr *opt)
 
 		dropped += qdisc_pkt_len(skb);
 		qdisc_qstats_backlog_dec(sch, skb);
-		qdisc_drop(skb, sch);
+		rtnl_qdisc_drop(skb, sch);
 	}
 	qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped);
 
diff --git a/net/sched/sch_fq.c b/net/sched/sch_fq.c
index f49c81e..6eb0667 100644
--- a/net/sched/sch_fq.c
+++ b/net/sched/sch_fq.c
@@ -514,17 +514,25 @@ out:
 	return skb;
 }
 
+static void fq_flow_purge(struct fq_flow *flow)
+{
+	rtnl_kfree_skbs(flow->head, flow->tail);
+	flow->head = NULL;
+	flow->qlen = 0;
+}
+
 static void fq_reset(struct Qdisc *sch)
 {
 	struct fq_sched_data *q = qdisc_priv(sch);
 	struct rb_root *root;
-	struct sk_buff *skb;
 	struct rb_node *p;
 	struct fq_flow *f;
 	unsigned int idx;
 
-	while ((skb = fq_dequeue_head(sch, &q->internal)) != NULL)
-		kfree_skb(skb);
+	sch->q.qlen = 0;
+	sch->qstats.backlog = 0;
+
+	fq_flow_purge(&q->internal);
 
 	if (!q->fq_root)
 		return;
@@ -535,8 +543,7 @@ static void fq_reset(struct Qdisc *sch)
 			f = container_of(p, struct fq_flow, fq_node);
 			rb_erase(p, root);
 
-			while ((skb = fq_dequeue_head(sch, f)) != NULL)
-				kfree_skb(skb);
+			fq_flow_purge(f);
 
 			kmem_cache_free(fq_flow_cachep, f);
 		}
@@ -737,7 +744,7 @@ static int fq_change(struct Qdisc *sch, struct nlattr *opt)
 		if (!skb)
 			break;
 		drop_len += qdisc_pkt_len(skb);
-		kfree_skb(skb);
+		rtnl_kfree_skbs(skb, skb);
 		drop_count++;
 	}
 	qdisc_tree_reduce_backlog(sch, drop_count, drop_len);
diff --git a/net/sched/sch_fq_codel.c b/net/sched/sch_fq_codel.c
index a302e8e..2dc0a84 100644
--- a/net/sched/sch_fq_codel.c
+++ b/net/sched/sch_fq_codel.c
@@ -336,6 +336,12 @@ begin:
 	return skb;
 }
 
+static void fq_codel_flow_purge(struct fq_codel_flow *flow)
+{
+	rtnl_kfree_skbs(flow->head, flow->tail);
+	flow->head = NULL;
+}
+
 static void fq_codel_reset(struct Qdisc *sch)
 {
 	struct fq_codel_sched_data *q = qdisc_priv(sch);
@@ -346,18 +352,13 @@ static void fq_codel_reset(struct Qdisc *sch)
 	for (i = 0; i < q->flows_cnt; i++) {
 		struct fq_codel_flow *flow = q->flows + i;
 
-		while (flow->head) {
-			struct sk_buff *skb = dequeue_head(flow);
-
-			qdisc_qstats_backlog_dec(sch, skb);
-			kfree_skb(skb);
-		}
-
+		fq_codel_flow_purge(flow);
 		INIT_LIST_HEAD(&flow->flowchain);
 		codel_vars_init(&flow->cvars);
 	}
 	memset(q->backlogs, 0, q->flows_cnt * sizeof(u32));
 	sch->q.qlen = 0;
+	sch->qstats.backlog = 0;
 	q->memory_usage = 0;
 }
 
@@ -433,7 +434,7 @@ static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt)
 		struct sk_buff *skb = fq_codel_dequeue(sch);
 
 		q->cstats.drop_len += qdisc_pkt_len(skb);
-		kfree_skb(skb);
+		rtnl_kfree_skbs(skb, skb);
 		q->cstats.drop_count++;
 	}
 	qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, q->cstats.drop_len);
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index 0c9cb51..773b632 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -493,7 +493,7 @@ static void pfifo_fast_reset(struct Qdisc *qdisc)
 	struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
 
 	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
-		__qdisc_reset_queue(qdisc, band2list(priv, prio));
+		__qdisc_reset_queue(band2list(priv, prio));
 
 	priv->bitmap = 0;
 	qdisc->qstats.backlog = 0;
diff --git a/net/sched/sch_hhf.c b/net/sched/sch_hhf.c
index c517918..c44593b 100644
--- a/net/sched/sch_hhf.c
+++ b/net/sched/sch_hhf.c
@@ -464,7 +464,7 @@ static void hhf_reset(struct Qdisc *sch)
 	struct sk_buff *skb;
 
 	while ((skb = hhf_dequeue(sch)) != NULL)
-		kfree_skb(skb);
+		rtnl_kfree_skbs(skb, skb);
 }
 
 static void *hhf_zalloc(size_t sz)
@@ -574,7 +574,7 @@ static int hhf_change(struct Qdisc *sch, struct nlattr *opt)
 	while (sch->q.qlen > sch->limit) {
 		struct sk_buff *skb = hhf_dequeue(sch);
 
-		kfree_skb(skb);
+		rtnl_kfree_skbs(skb, skb);
 	}
 	qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen,
 				  prev_backlog - sch->qstats.backlog);
diff --git a/net/sched/sch_htb.c b/net/sched/sch_htb.c
index 07dcd29..a454605 100644
--- a/net/sched/sch_htb.c
+++ b/net/sched/sch_htb.c
@@ -957,7 +957,7 @@ static void htb_reset(struct Qdisc *sch)
 		}
 	}
 	qdisc_watchdog_cancel(&q->watchdog);
-	__skb_queue_purge(&q->direct_queue);
+	__qdisc_reset_queue(&q->direct_queue);
 	sch->q.qlen = 0;
 	sch->qstats.backlog = 0;
 	memset(q->hlevel, 0, sizeof(q->hlevel));
@@ -1231,7 +1231,7 @@ static void htb_destroy(struct Qdisc *sch)
 			htb_destroy_class(sch, cl);
 	}
 	qdisc_class_hash_destroy(&q->clhash);
-	__skb_queue_purge(&q->direct_queue);
+	__qdisc_reset_queue(&q->direct_queue);
 }
 
 static int htb_delete(struct Qdisc *sch, unsigned long arg)
diff --git a/net/sched/sch_netem.c b/net/sched/sch_netem.c
index 876df13..e271967 100644
--- a/net/sched/sch_netem.c
+++ b/net/sched/sch_netem.c
@@ -368,9 +368,7 @@ static void tfifo_reset(struct Qdisc *sch)
 		struct sk_buff *skb = netem_rb_to_skb(p);
 
 		rb_erase(p, &q->t_root);
-		skb->next = NULL;
-		skb->prev = NULL;
-		kfree_skb(skb);
+		rtnl_kfree_skbs(skb, skb);
 	}
 }
 
diff --git a/net/sched/sch_pie.c b/net/sched/sch_pie.c
index 71ae3b9..912a46a 100644
--- a/net/sched/sch_pie.c
+++ b/net/sched/sch_pie.c
@@ -234,7 +234,7 @@ static int pie_change(struct Qdisc *sch, struct nlattr *opt)
 
 		dropped += qdisc_pkt_len(skb);
 		qdisc_qstats_backlog_dec(sch, skb);
-		qdisc_drop(skb, sch);
+		rtnl_qdisc_drop(skb, sch);
 	}
 	qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped);
 
diff --git a/net/sched/sch_sfq.c b/net/sched/sch_sfq.c
index a2e0b85..57d118b 100644
--- a/net/sched/sch_sfq.c
+++ b/net/sched/sch_sfq.c
@@ -520,7 +520,7 @@ sfq_reset(struct Qdisc *sch)
 	struct sk_buff *skb;
 
 	while ((skb = sfq_dequeue(sch)) != NULL)
-		kfree_skb(skb);
+		rtnl_kfree_skbs(skb, skb);
 }
 
 /*