Merge branch 'net-mini_Qdisc'

Jiri Pirko says:

====================
net: core: introduce mini_Qdisc and eliminate usage of tp->q for clsact fastpath

This patchset's main patch is patch number 2. It carries the
description. Patch 1 is just a dependency.

---
v3->v4:
- rebased to be applicable on top of the current net-next
v2->v3:
- Using head change callback to replace miniq pointer every time tp head
  changes. This eliminates one rcu dereference and makes the claim "without
  added overhead" valid.
v1->v2:
- Use dev instead of skb->dev in sch_handle_egress as pointed out by Daniel
- Fixed synchronize_rcu_bh() in mini_qdisc_disable and commented
====================

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

7 files changed, 166 insertions, 60 deletions

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 5e02f79b2110..7de7656550c2 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1559,6 +1559,8 @@ enum netdev_priv_flags {
  *
  *	@rx_handler:		handler for received packets
  *	@rx_handler_data: 	XXX: need comments on this one
+ *	@miniq_ingress:		ingress/clsact qdisc specific data for
+ *				ingress processing
  *	@ingress_queue:		XXX: need comments on this one
  *	@broadcast:		hw bcast address
  *
@@ -1576,7 +1578,8 @@ enum netdev_priv_flags {
  *	@tx_global_lock: 	XXX: need comments on this one
  *
  *	@xps_maps:	XXX: need comments on this one
- *
+ *	@miniq_egress:		clsact qdisc specific data for
+ *				egress processing
  *	@watchdog_timeo:	Represents the timeout that is used by
  *				the watchdog (see dev_watchdog())
  *	@watchdog_timer:	List of timers
@@ -1795,7 +1798,7 @@ struct net_device {
 	void __rcu		*rx_handler_data;
 
 #ifdef CONFIG_NET_CLS_ACT
-	struct tcf_proto __rcu  *ingress_cl_list;
+	struct mini_Qdisc __rcu	*miniq_ingress;
 #endif
 	struct netdev_queue __rcu *ingress_queue;
 #ifdef CONFIG_NETFILTER_INGRESS
@@ -1826,7 +1829,7 @@ struct net_device {
 	struct xps_dev_maps __rcu *xps_maps;
 #endif
 #ifdef CONFIG_NET_CLS_ACT
-	struct tcf_proto __rcu  *egress_cl_list;
+	struct mini_Qdisc __rcu	*miniq_egress;
 #endif
 
 	/* These may be needed for future network-power-down code. */
diff --git a/include/net/pkt_cls.h b/include/net/pkt_cls.h
index d15c40c7bde7..505d4b71975f 100644
--- a/include/net/pkt_cls.h
+++ b/include/net/pkt_cls.h
@@ -26,6 +26,8 @@ enum tcf_block_binder_type {
 
 struct tcf_block_ext_info {
 	enum tcf_block_binder_type binder_type;
+	tcf_chain_head_change_t *chain_head_change;
+	void *chain_head_change_priv;
 };
 
 struct tcf_block_cb;
@@ -37,12 +39,10 @@ struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 void tcf_chain_put(struct tcf_chain *chain);
 int tcf_block_get(struct tcf_block **p_block,
 		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q);
-int tcf_block_get_ext(struct tcf_block **p_block,
-		      struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
+int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 		      struct tcf_block_ext_info *ei);
 void tcf_block_put(struct tcf_block *block);
-void tcf_block_put_ext(struct tcf_block *block,
-		       struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
+void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
 		       struct tcf_block_ext_info *ei);
 
 static inline struct Qdisc *tcf_block_q(struct tcf_block *block)
@@ -82,8 +82,7 @@ int tcf_block_get(struct tcf_block **p_block,
 }
 
 static inline
-int tcf_block_get_ext(struct tcf_block **p_block,
-		      struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
+int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 		      struct tcf_block_ext_info *ei)
 {
 	return 0;
@@ -94,8 +93,7 @@ static inline void tcf_block_put(struct tcf_block *block)
 }
 
 static inline
-void tcf_block_put_ext(struct tcf_block *block,
-		       struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
+void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
 		       struct tcf_block_ext_info *ei)
 {
 }
diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
index c23e938f5b19..c64e62c9450a 100644
--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -260,9 +260,12 @@ struct qdisc_skb_cb {
 	unsigned char		data[QDISC_CB_PRIV_LEN];
 };
 
+typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
+
 struct tcf_chain {
 	struct tcf_proto __rcu *filter_chain;
-	struct tcf_proto __rcu **p_filter_chain;
+	tcf_chain_head_change_t *chain_head_change;
+	void *chain_head_change_priv;
 	struct list_head list;
 	struct tcf_block *block;
 	u32 index; /* chain index */
@@ -901,4 +904,36 @@ static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
 	res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
 }
 
+/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
+ * The fast path only needs to access filter list and to update stats
+ */
+struct mini_Qdisc {
+	struct tcf_proto *filter_list;
+	struct gnet_stats_basic_cpu __percpu *cpu_bstats;
+	struct gnet_stats_queue	__percpu *cpu_qstats;
+	struct rcu_head rcu;
+};
+
+static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
+						const struct sk_buff *skb)
+{
+	bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
+}
+
+static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
+{
+	this_cpu_inc(miniq->cpu_qstats->drops);
+}
+
+struct mini_Qdisc_pair {
+	struct mini_Qdisc miniq1;
+	struct mini_Qdisc miniq2;
+	struct mini_Qdisc __rcu **p_miniq;
+};
+
+void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
+			  struct tcf_proto *tp_head);
+void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
+			  struct mini_Qdisc __rcu **p_miniq);
+
 #endif
diff --git a/net/core/dev.c b/net/core/dev.c
index 24ac9083bc13..1423cf4d695c 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3274,22 +3274,22 @@ EXPORT_SYMBOL(dev_loopback_xmit);
 static struct sk_buff *
 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
 {
-	struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
+	struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
 	struct tcf_result cl_res;
 
-	if (!cl)
+	if (!miniq)
 		return skb;
 
 	/* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
-	qdisc_bstats_cpu_update(cl->q, skb);
+	mini_qdisc_bstats_cpu_update(miniq, skb);
 
-	switch (tcf_classify(skb, cl, &cl_res, false)) {
+	switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
 	case TC_ACT_OK:
 	case TC_ACT_RECLASSIFY:
 		skb->tc_index = TC_H_MIN(cl_res.classid);
 		break;
 	case TC_ACT_SHOT:
-		qdisc_qstats_cpu_drop(cl->q);
+		mini_qdisc_qstats_cpu_drop(miniq);
 		*ret = NET_XMIT_DROP;
 		kfree_skb(skb);
 		return NULL;
@@ -4189,7 +4189,7 @@ sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
 		   struct net_device *orig_dev)
 {
 #ifdef CONFIG_NET_CLS_ACT
-	struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
+	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
 	struct tcf_result cl_res;
 
 	/* If there's at least one ingress present somewhere (so
@@ -4197,8 +4197,9 @@ sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
 	 * that are not configured with an ingress qdisc will bail
 	 * out here.
 	 */
-	if (!cl)
+	if (!miniq)
 		return skb;
+
 	if (*pt_prev) {
 		*ret = deliver_skb(skb, *pt_prev, orig_dev);
 		*pt_prev = NULL;
@@ -4206,15 +4207,15 @@ sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
 
 	qdisc_skb_cb(skb)->pkt_len = skb->len;
 	skb->tc_at_ingress = 1;
-	qdisc_bstats_cpu_update(cl->q, skb);
+	mini_qdisc_bstats_cpu_update(miniq, skb);
 
-	switch (tcf_classify(skb, cl, &cl_res, false)) {
+	switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
 	case TC_ACT_OK:
 	case TC_ACT_RECLASSIFY:
 		skb->tc_index = TC_H_MIN(cl_res.classid);
 		break;
 	case TC_ACT_SHOT:
-		qdisc_qstats_cpu_drop(cl->q);
+		mini_qdisc_qstats_cpu_drop(miniq);
 		kfree_skb(skb);
 		return NULL;
 	case TC_ACT_STOLEN:
diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c
index 8d1885abee83..206e19f4fc01 100644
--- a/net/sched/cls_api.c
+++ b/net/sched/cls_api.c
@@ -195,12 +195,19 @@ static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 	return chain;
 }
 
+static void tcf_chain_head_change(struct tcf_chain *chain,
+				  struct tcf_proto *tp_head)
+{
+	if (chain->chain_head_change)
+		chain->chain_head_change(tp_head,
+					 chain->chain_head_change_priv);
+}
+
 static void tcf_chain_flush(struct tcf_chain *chain)
 {
 	struct tcf_proto *tp;
 
-	if (chain->p_filter_chain)
-		RCU_INIT_POINTER(*chain->p_filter_chain, NULL);
+	tcf_chain_head_change(chain, NULL);
 	while ((tp = rtnl_dereference(chain->filter_chain)) != NULL) {
 		RCU_INIT_POINTER(chain->filter_chain, tp->next);
 		tcf_chain_put(chain);
@@ -242,13 +249,6 @@ void tcf_chain_put(struct tcf_chain *chain)
 }
 EXPORT_SYMBOL(tcf_chain_put);
 
-static void
-tcf_chain_filter_chain_ptr_set(struct tcf_chain *chain,
-			       struct tcf_proto __rcu **p_filter_chain)
-{
-	chain->p_filter_chain = p_filter_chain;
-}
-
 static void tcf_block_offload_cmd(struct tcf_block *block, struct Qdisc *q,
 				  struct tcf_block_ext_info *ei,
 				  enum tc_block_command command)
@@ -276,8 +276,7 @@ static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 	tcf_block_offload_cmd(block, q, ei, TC_BLOCK_UNBIND);
 }
 
-int tcf_block_get_ext(struct tcf_block **p_block,
-		      struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
+int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 		      struct tcf_block_ext_info *ei)
 {
 	struct tcf_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
@@ -295,7 +294,9 @@ int tcf_block_get_ext(struct tcf_block **p_block,
 		err = -ENOMEM;
 		goto err_chain_create;
 	}
-	tcf_chain_filter_chain_ptr_set(chain, p_filter_chain);
+	WARN_ON(!ei->chain_head_change);
+	chain->chain_head_change = ei->chain_head_change;
+	chain->chain_head_change_priv = ei->chain_head_change_priv;
 	block->net = qdisc_net(q);
 	block->q = q;
 	tcf_block_offload_bind(block, q, ei);
@@ -308,12 +309,23 @@ err_chain_create:
 }
 EXPORT_SYMBOL(tcf_block_get_ext);
 
+static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
+{
+	struct tcf_proto __rcu **p_filter_chain = priv;
+
+	rcu_assign_pointer(*p_filter_chain, tp_head);
+}
+
 int tcf_block_get(struct tcf_block **p_block,
 		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q)
 {
-	struct tcf_block_ext_info ei = {0, };
+	struct tcf_block_ext_info ei = {
+		.chain_head_change = tcf_chain_head_change_dflt,
+		.chain_head_change_priv = p_filter_chain,
+	};
 
-	return tcf_block_get_ext(p_block, p_filter_chain, q, &ei);
+	WARN_ON(!p_filter_chain);
+	return tcf_block_get_ext(p_block, q, &ei);
 }
 EXPORT_SYMBOL(tcf_block_get);
 
@@ -334,8 +346,7 @@ static void tcf_block_put_final(struct work_struct *work)
  * actions should be all removed after flushing. However, filters are now
  * destroyed in tc filter workqueue with RTNL lock, they can not race here.
  */
-void tcf_block_put_ext(struct tcf_block *block,
-		       struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
+void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
 		       struct tcf_block_ext_info *ei)
 {
 	struct tcf_chain *chain, *tmp;
@@ -361,7 +372,7 @@ void tcf_block_put(struct tcf_block *block)
 
 	if (!block)
 		return;
-	tcf_block_put_ext(block, NULL, block->q, &ei);
+	tcf_block_put_ext(block, block->q, &ei);
 }
 
 EXPORT_SYMBOL(tcf_block_put);
@@ -537,9 +548,8 @@ static void tcf_chain_tp_insert(struct tcf_chain *chain,
 				struct tcf_chain_info *chain_info,
 				struct tcf_proto *tp)
 {
-	if (chain->p_filter_chain &&
-	    *chain_info->pprev == chain->filter_chain)
-		rcu_assign_pointer(*chain->p_filter_chain, tp);
+	if (*chain_info->pprev == chain->filter_chain)
+		tcf_chain_head_change(chain, tp);
 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
 	rcu_assign_pointer(*chain_info->pprev, tp);
 	tcf_chain_hold(chain);
@@ -551,8 +561,8 @@ static void tcf_chain_tp_remove(struct tcf_chain *chain,
 {
 	struct tcf_proto *next = rtnl_dereference(chain_info->next);
 
-	if (chain->p_filter_chain && tp == chain->filter_chain)
-		RCU_INIT_POINTER(*chain->p_filter_chain, next);
+	if (tp == chain->filter_chain)
+		tcf_chain_head_change(chain, next);
 	RCU_INIT_POINTER(*chain_info->pprev, next);
 	tcf_chain_put(chain);
 }
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index aa74aa42b5d7..3839cbbdc32b 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -1024,3 +1024,49 @@ void psched_ratecfg_precompute(struct psched_ratecfg *r,
 	}
 }
 EXPORT_SYMBOL(psched_ratecfg_precompute);
+
+static void mini_qdisc_rcu_func(struct rcu_head *head)
+{
+}
+
+void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
+			  struct tcf_proto *tp_head)
+{
+	struct mini_Qdisc *miniq_old = rtnl_dereference(*miniqp->p_miniq);
+	struct mini_Qdisc *miniq;
+
+	if (!tp_head) {
+		RCU_INIT_POINTER(*miniqp->p_miniq, NULL);
+		return;
+	}
+
+	miniq = !miniq_old || miniq_old == &miniqp->miniq2 ?
+		&miniqp->miniq1 : &miniqp->miniq2;
+
+	/* We need to make sure that readers won't see the miniq
+	 * we are about to modify. So wait until previous call_rcu_bh callback
+	 * is done.
+	 */
+	rcu_barrier_bh();
+	miniq->filter_list = tp_head;
+	rcu_assign_pointer(*miniqp->p_miniq, miniq);
+
+	if (miniq_old)
+		/* This is counterpart of the rcu barrier above. We need to
+		 * block potential new user of miniq_old until all readers
+		 * are not seeing it.
+		 */
+		call_rcu_bh(&miniq_old->rcu, mini_qdisc_rcu_func);
+}
+EXPORT_SYMBOL(mini_qdisc_pair_swap);
+
+void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
+			  struct mini_Qdisc __rcu **p_miniq)
+{
+	miniqp->miniq1.cpu_bstats = qdisc->cpu_bstats;
+	miniqp->miniq1.cpu_qstats = qdisc->cpu_qstats;
+	miniqp->miniq2.cpu_bstats = qdisc->cpu_bstats;
+	miniqp->miniq2.cpu_qstats = qdisc->cpu_qstats;
+	miniqp->p_miniq = p_miniq;
+}
+EXPORT_SYMBOL(mini_qdisc_pair_init);
diff --git a/net/sched/sch_ingress.c b/net/sched/sch_ingress.c
index b599db26d34b..5ecc38f35d47 100644
--- a/net/sched/sch_ingress.c
+++ b/net/sched/sch_ingress.c
@@ -21,6 +21,7 @@
 struct ingress_sched_data {
 	struct tcf_block *block;
 	struct tcf_block_ext_info block_info;
+	struct mini_Qdisc_pair miniqp;
 };
 
 static struct Qdisc *ingress_leaf(struct Qdisc *sch, unsigned long arg)
@@ -54,16 +55,26 @@ static struct tcf_block *ingress_tcf_block(struct Qdisc *sch, unsigned long cl)
 	return q->block;
 }
 
+static void clsact_chain_head_change(struct tcf_proto *tp_head, void *priv)
+{
+	struct mini_Qdisc_pair *miniqp = priv;
+
+	mini_qdisc_pair_swap(miniqp, tp_head);
+}
+
 static int ingress_init(struct Qdisc *sch, struct nlattr *opt)
 {
 	struct ingress_sched_data *q = qdisc_priv(sch);
 	struct net_device *dev = qdisc_dev(sch);
 	int err;
 
+	mini_qdisc_pair_init(&q->miniqp, sch, &dev->miniq_ingress);
+
 	q->block_info.binder_type = TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS;
+	q->block_info.chain_head_change = clsact_chain_head_change;
+	q->block_info.chain_head_change_priv = &q->miniqp;
 
-	err = tcf_block_get_ext(&q->block, &dev->ingress_cl_list,
-				sch, &q->block_info);
+	err = tcf_block_get_ext(&q->block, sch, &q->block_info);
 	if (err)
 		return err;
 
@@ -76,10 +87,8 @@ static int ingress_init(struct Qdisc *sch, struct nlattr *opt)
 static void ingress_destroy(struct Qdisc *sch)
 {
 	struct ingress_sched_data *q = qdisc_priv(sch);
-	struct net_device *dev = qdisc_dev(sch);
 
-	tcf_block_put_ext(q->block, &dev->ingress_cl_list,
-			  sch, &q->block_info);
+	tcf_block_put_ext(q->block, sch, &q->block_info);
 	net_dec_ingress_queue();
 }
 
@@ -122,6 +131,8 @@ struct clsact_sched_data {
 	struct tcf_block *egress_block;
 	struct tcf_block_ext_info ingress_block_info;
 	struct tcf_block_ext_info egress_block_info;
+	struct mini_Qdisc_pair miniqp_ingress;
+	struct mini_Qdisc_pair miniqp_egress;
 };
 
 static unsigned long clsact_find(struct Qdisc *sch, u32 classid)
@@ -161,17 +172,23 @@ static int clsact_init(struct Qdisc *sch, struct nlattr *opt)
 	struct net_device *dev = qdisc_dev(sch);
 	int err;
 
+	mini_qdisc_pair_init(&q->miniqp_ingress, sch, &dev->miniq_ingress);
+
 	q->ingress_block_info.binder_type = TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS;
+	q->ingress_block_info.chain_head_change = clsact_chain_head_change;
+	q->ingress_block_info.chain_head_change_priv = &q->miniqp_ingress;
 
-	err = tcf_block_get_ext(&q->ingress_block, &dev->ingress_cl_list,
-				sch, &q->ingress_block_info);
+	err = tcf_block_get_ext(&q->ingress_block, sch, &q->ingress_block_info);
 	if (err)
 		return err;
 
+	mini_qdisc_pair_init(&q->miniqp_egress, sch, &dev->miniq_egress);
+
 	q->egress_block_info.binder_type = TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS;
+	q->egress_block_info.chain_head_change = clsact_chain_head_change;
+	q->egress_block_info.chain_head_change_priv = &q->miniqp_egress;
 
-	err = tcf_block_get_ext(&q->egress_block, &dev->egress_cl_list,
-				sch, &q->egress_block_info);
+	err = tcf_block_get_ext(&q->egress_block, sch, &q->egress_block_info);
 	if (err)
 		goto err_egress_block_get;
 
@@ -183,20 +200,16 @@ static int clsact_init(struct Qdisc *sch, struct nlattr *opt)
 	return 0;
 
 err_egress_block_get:
-	tcf_block_put_ext(q->ingress_block, &dev->ingress_cl_list,
-			  sch, &q->ingress_block_info);
+	tcf_block_put_ext(q->ingress_block, sch, &q->ingress_block_info);
 	return err;
 }
 
 static void clsact_destroy(struct Qdisc *sch)
 {
 	struct clsact_sched_data *q = qdisc_priv(sch);
-	struct net_device *dev = qdisc_dev(sch);
 
-	tcf_block_put_ext(q->egress_block, &dev->egress_cl_list,
-			  sch, &q->egress_block_info);
-	tcf_block_put_ext(q->ingress_block, &dev->ingress_cl_list,
-			  sch, &q->ingress_block_info);
+	tcf_block_put_ext(q->egress_block, sch, &q->egress_block_info);
+	tcf_block_put_ext(q->ingress_block, sch, &q->ingress_block_info);
 
 	net_dec_ingress_queue();
 	net_dec_egress_queue();