xref: /openbmc/linux/include/net/sch_generic.h (revision f36fe1c4)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __NET_SCHED_GENERIC_H
3 #define __NET_SCHED_GENERIC_H
4 
5 #include <linux/netdevice.h>
6 #include <linux/types.h>
7 #include <linux/rcupdate.h>
8 #include <linux/pkt_sched.h>
9 #include <linux/pkt_cls.h>
10 #include <linux/percpu.h>
11 #include <linux/dynamic_queue_limits.h>
12 #include <linux/list.h>
13 #include <linux/refcount.h>
14 #include <linux/workqueue.h>
15 #include <net/gen_stats.h>
16 #include <net/rtnetlink.h>
17 
18 struct Qdisc_ops;
19 struct qdisc_walker;
20 struct tcf_walker;
21 struct module;
22 
23 struct qdisc_rate_table {
24 	struct tc_ratespec rate;
25 	u32		data[256];
26 	struct qdisc_rate_table *next;
27 	int		refcnt;
28 };
29 
30 enum qdisc_state_t {
31 	__QDISC_STATE_SCHED,
32 	__QDISC_STATE_DEACTIVATED,
33 };
34 
35 struct qdisc_size_table {
36 	struct rcu_head		rcu;
37 	struct list_head	list;
38 	struct tc_sizespec	szopts;
39 	int			refcnt;
40 	u16			data[];
41 };
42 
43 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
44 struct qdisc_skb_head {
45 	struct sk_buff	*head;
46 	struct sk_buff	*tail;
47 	__u32		qlen;
48 	spinlock_t	lock;
49 };
50 
51 struct Qdisc {
52 	int 			(*enqueue)(struct sk_buff *skb,
53 					   struct Qdisc *sch,
54 					   struct sk_buff **to_free);
55 	struct sk_buff *	(*dequeue)(struct Qdisc *sch);
56 	unsigned int		flags;
57 #define TCQ_F_BUILTIN		1
58 #define TCQ_F_INGRESS		2
59 #define TCQ_F_CAN_BYPASS	4
60 #define TCQ_F_MQROOT		8
61 #define TCQ_F_ONETXQUEUE	0x10 /* dequeue_skb() can assume all skbs are for
62 				      * q->dev_queue : It can test
63 				      * netif_xmit_frozen_or_stopped() before
64 				      * dequeueing next packet.
65 				      * Its true for MQ/MQPRIO slaves, or non
66 				      * multiqueue device.
67 				      */
68 #define TCQ_F_WARN_NONWC	(1 << 16)
69 #define TCQ_F_CPUSTATS		0x20 /* run using percpu statistics */
70 #define TCQ_F_NOPARENT		0x40 /* root of its hierarchy :
71 				      * qdisc_tree_decrease_qlen() should stop.
72 				      */
73 #define TCQ_F_INVISIBLE		0x80 /* invisible by default in dump */
74 #define TCQ_F_NOLOCK		0x100 /* qdisc does not require locking */
75 #define TCQ_F_OFFLOADED		0x200 /* qdisc is offloaded to HW */
76 	u32			limit;
77 	const struct Qdisc_ops	*ops;
78 	struct qdisc_size_table	__rcu *stab;
79 	struct hlist_node       hash;
80 	u32			handle;
81 	u32			parent;
82 
83 	struct netdev_queue	*dev_queue;
84 
85 	struct net_rate_estimator __rcu *rate_est;
86 	struct gnet_stats_basic_cpu __percpu *cpu_bstats;
87 	struct gnet_stats_queue	__percpu *cpu_qstats;
88 
89 	/*
90 	 * For performance sake on SMP, we put highly modified fields at the end
91 	 */
92 	struct sk_buff_head	gso_skb ____cacheline_aligned_in_smp;
93 	struct qdisc_skb_head	q;
94 	struct gnet_stats_basic_packed bstats;
95 	seqcount_t		running;
96 	struct gnet_stats_queue	qstats;
97 	unsigned long		state;
98 	struct Qdisc            *next_sched;
99 	struct sk_buff_head	skb_bad_txq;
100 	int			padded;
101 	refcount_t		refcnt;
102 
103 	spinlock_t		busylock ____cacheline_aligned_in_smp;
104 };
105 
106 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
107 {
108 	if (qdisc->flags & TCQ_F_BUILTIN)
109 		return;
110 	refcount_inc(&qdisc->refcnt);
111 }
112 
113 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
114 {
115 	return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
116 }
117 
118 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
119 {
120 	if (qdisc_is_running(qdisc))
121 		return false;
122 	/* Variant of write_seqcount_begin() telling lockdep a trylock
123 	 * was attempted.
124 	 */
125 	raw_write_seqcount_begin(&qdisc->running);
126 	seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
127 	return true;
128 }
129 
130 static inline void qdisc_run_end(struct Qdisc *qdisc)
131 {
132 	write_seqcount_end(&qdisc->running);
133 }
134 
135 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
136 {
137 	return qdisc->flags & TCQ_F_ONETXQUEUE;
138 }
139 
140 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
141 {
142 #ifdef CONFIG_BQL
143 	/* Non-BQL migrated drivers will return 0, too. */
144 	return dql_avail(&txq->dql);
145 #else
146 	return 0;
147 #endif
148 }
149 
150 struct Qdisc_class_ops {
151 	/* Child qdisc manipulation */
152 	struct netdev_queue *	(*select_queue)(struct Qdisc *, struct tcmsg *);
153 	int			(*graft)(struct Qdisc *, unsigned long cl,
154 					struct Qdisc *, struct Qdisc **,
155 					struct netlink_ext_ack *extack);
156 	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
157 	void			(*qlen_notify)(struct Qdisc *, unsigned long);
158 
159 	/* Class manipulation routines */
160 	unsigned long		(*find)(struct Qdisc *, u32 classid);
161 	int			(*change)(struct Qdisc *, u32, u32,
162 					struct nlattr **, unsigned long *,
163 					struct netlink_ext_ack *);
164 	int			(*delete)(struct Qdisc *, unsigned long);
165 	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
166 
167 	/* Filter manipulation */
168 	struct tcf_block *	(*tcf_block)(struct Qdisc *sch,
169 					     unsigned long arg,
170 					     struct netlink_ext_ack *extack);
171 	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
172 					u32 classid);
173 	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
174 
175 	/* rtnetlink specific */
176 	int			(*dump)(struct Qdisc *, unsigned long,
177 					struct sk_buff *skb, struct tcmsg*);
178 	int			(*dump_stats)(struct Qdisc *, unsigned long,
179 					struct gnet_dump *);
180 };
181 
182 struct Qdisc_ops {
183 	struct Qdisc_ops	*next;
184 	const struct Qdisc_class_ops	*cl_ops;
185 	char			id[IFNAMSIZ];
186 	int			priv_size;
187 	unsigned int		static_flags;
188 
189 	int 			(*enqueue)(struct sk_buff *skb,
190 					   struct Qdisc *sch,
191 					   struct sk_buff **to_free);
192 	struct sk_buff *	(*dequeue)(struct Qdisc *);
193 	struct sk_buff *	(*peek)(struct Qdisc *);
194 
195 	int			(*init)(struct Qdisc *sch, struct nlattr *arg,
196 					struct netlink_ext_ack *extack);
197 	void			(*reset)(struct Qdisc *);
198 	void			(*destroy)(struct Qdisc *);
199 	int			(*change)(struct Qdisc *sch,
200 					  struct nlattr *arg,
201 					  struct netlink_ext_ack *extack);
202 	void			(*attach)(struct Qdisc *sch);
203 
204 	int			(*dump)(struct Qdisc *, struct sk_buff *);
205 	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
206 
207 	struct module		*owner;
208 };
209 
210 
211 struct tcf_result {
212 	union {
213 		struct {
214 			unsigned long	class;
215 			u32		classid;
216 		};
217 		const struct tcf_proto *goto_tp;
218 	};
219 };
220 
221 struct tcf_proto_ops {
222 	struct list_head	head;
223 	char			kind[IFNAMSIZ];
224 
225 	int			(*classify)(struct sk_buff *,
226 					    const struct tcf_proto *,
227 					    struct tcf_result *);
228 	int			(*init)(struct tcf_proto*);
229 	void			(*destroy)(struct tcf_proto*);
230 
231 	void*			(*get)(struct tcf_proto*, u32 handle);
232 	int			(*change)(struct net *net, struct sk_buff *,
233 					struct tcf_proto*, unsigned long,
234 					u32 handle, struct nlattr **,
235 					void **, bool);
236 	int			(*delete)(struct tcf_proto*, void *, bool*);
237 	void			(*walk)(struct tcf_proto*, struct tcf_walker *arg);
238 	void			(*bind_class)(void *, u32, unsigned long);
239 
240 	/* rtnetlink specific */
241 	int			(*dump)(struct net*, struct tcf_proto*, void *,
242 					struct sk_buff *skb, struct tcmsg*);
243 
244 	struct module		*owner;
245 };
246 
247 struct tcf_proto {
248 	/* Fast access part */
249 	struct tcf_proto __rcu	*next;
250 	void __rcu		*root;
251 	int			(*classify)(struct sk_buff *,
252 					    const struct tcf_proto *,
253 					    struct tcf_result *);
254 	__be16			protocol;
255 
256 	/* All the rest */
257 	u32			prio;
258 	u32			classid;
259 	struct Qdisc		*q;
260 	void			*data;
261 	const struct tcf_proto_ops	*ops;
262 	struct tcf_chain	*chain;
263 	struct rcu_head		rcu;
264 };
265 
266 struct qdisc_skb_cb {
267 	unsigned int		pkt_len;
268 	u16			slave_dev_queue_mapping;
269 	u16			tc_classid;
270 #define QDISC_CB_PRIV_LEN 20
271 	unsigned char		data[QDISC_CB_PRIV_LEN];
272 };
273 
274 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
275 
276 struct tcf_chain {
277 	struct tcf_proto __rcu *filter_chain;
278 	struct list_head filter_chain_list;
279 	struct list_head list;
280 	struct tcf_block *block;
281 	u32 index; /* chain index */
282 	unsigned int refcnt;
283 };
284 
285 struct tcf_block {
286 	struct list_head chain_list;
287 	u32 index; /* block index for shared blocks */
288 	unsigned int refcnt;
289 	struct net *net;
290 	struct Qdisc *q;
291 	struct list_head cb_list;
292 	struct list_head owner_list;
293 	bool keep_dst;
294 };
295 
296 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
297 {
298 	struct qdisc_skb_cb *qcb;
299 
300 	BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
301 	BUILD_BUG_ON(sizeof(qcb->data) < sz);
302 }
303 
304 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
305 {
306 	return this_cpu_ptr(q->cpu_qstats)->qlen;
307 }
308 
309 static inline int qdisc_qlen(const struct Qdisc *q)
310 {
311 	return q->q.qlen;
312 }
313 
314 static inline int qdisc_qlen_sum(const struct Qdisc *q)
315 {
316 	__u32 qlen = 0;
317 	int i;
318 
319 	if (q->flags & TCQ_F_NOLOCK) {
320 		for_each_possible_cpu(i)
321 			qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
322 	} else {
323 		qlen = q->q.qlen;
324 	}
325 
326 	return qlen;
327 }
328 
329 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
330 {
331 	return (struct qdisc_skb_cb *)skb->cb;
332 }
333 
334 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
335 {
336 	return &qdisc->q.lock;
337 }
338 
339 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
340 {
341 	struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
342 
343 	return q;
344 }
345 
346 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
347 {
348 	return qdisc->dev_queue->qdisc_sleeping;
349 }
350 
351 /* The qdisc root lock is a mechanism by which to top level
352  * of a qdisc tree can be locked from any qdisc node in the
353  * forest.  This allows changing the configuration of some
354  * aspect of the qdisc tree while blocking out asynchronous
355  * qdisc access in the packet processing paths.
356  *
357  * It is only legal to do this when the root will not change
358  * on us.  Otherwise we'll potentially lock the wrong qdisc
359  * root.  This is enforced by holding the RTNL semaphore, which
360  * all users of this lock accessor must do.
361  */
362 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
363 {
364 	struct Qdisc *root = qdisc_root(qdisc);
365 
366 	ASSERT_RTNL();
367 	return qdisc_lock(root);
368 }
369 
370 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
371 {
372 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
373 
374 	ASSERT_RTNL();
375 	return qdisc_lock(root);
376 }
377 
378 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
379 {
380 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
381 
382 	ASSERT_RTNL();
383 	return &root->running;
384 }
385 
386 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
387 {
388 	return qdisc->dev_queue->dev;
389 }
390 
391 static inline void sch_tree_lock(const struct Qdisc *q)
392 {
393 	spin_lock_bh(qdisc_root_sleeping_lock(q));
394 }
395 
396 static inline void sch_tree_unlock(const struct Qdisc *q)
397 {
398 	spin_unlock_bh(qdisc_root_sleeping_lock(q));
399 }
400 
401 extern struct Qdisc noop_qdisc;
402 extern struct Qdisc_ops noop_qdisc_ops;
403 extern struct Qdisc_ops pfifo_fast_ops;
404 extern struct Qdisc_ops mq_qdisc_ops;
405 extern struct Qdisc_ops noqueue_qdisc_ops;
406 extern const struct Qdisc_ops *default_qdisc_ops;
407 static inline const struct Qdisc_ops *
408 get_default_qdisc_ops(const struct net_device *dev, int ntx)
409 {
410 	return ntx < dev->real_num_tx_queues ?
411 			default_qdisc_ops : &pfifo_fast_ops;
412 }
413 
414 struct Qdisc_class_common {
415 	u32			classid;
416 	struct hlist_node	hnode;
417 };
418 
419 struct Qdisc_class_hash {
420 	struct hlist_head	*hash;
421 	unsigned int		hashsize;
422 	unsigned int		hashmask;
423 	unsigned int		hashelems;
424 };
425 
426 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
427 {
428 	id ^= id >> 8;
429 	id ^= id >> 4;
430 	return id & mask;
431 }
432 
433 static inline struct Qdisc_class_common *
434 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
435 {
436 	struct Qdisc_class_common *cl;
437 	unsigned int h;
438 
439 	if (!id)
440 		return NULL;
441 
442 	h = qdisc_class_hash(id, hash->hashmask);
443 	hlist_for_each_entry(cl, &hash->hash[h], hnode) {
444 		if (cl->classid == id)
445 			return cl;
446 	}
447 	return NULL;
448 }
449 
450 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
451 {
452 	u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
453 
454 	return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
455 }
456 
457 int qdisc_class_hash_init(struct Qdisc_class_hash *);
458 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
459 			     struct Qdisc_class_common *);
460 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
461 			     struct Qdisc_class_common *);
462 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
463 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
464 
465 void dev_init_scheduler(struct net_device *dev);
466 void dev_shutdown(struct net_device *dev);
467 void dev_activate(struct net_device *dev);
468 void dev_deactivate(struct net_device *dev);
469 void dev_deactivate_many(struct list_head *head);
470 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
471 			      struct Qdisc *qdisc);
472 void qdisc_reset(struct Qdisc *qdisc);
473 void qdisc_destroy(struct Qdisc *qdisc);
474 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
475 			       unsigned int len);
476 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
477 			  const struct Qdisc_ops *ops,
478 			  struct netlink_ext_ack *extack);
479 void qdisc_free(struct Qdisc *qdisc);
480 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
481 				const struct Qdisc_ops *ops, u32 parentid,
482 				struct netlink_ext_ack *extack);
483 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
484 			       const struct qdisc_size_table *stab);
485 int skb_do_redirect(struct sk_buff *);
486 
487 static inline void skb_reset_tc(struct sk_buff *skb)
488 {
489 #ifdef CONFIG_NET_CLS_ACT
490 	skb->tc_redirected = 0;
491 #endif
492 }
493 
494 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
495 {
496 #ifdef CONFIG_NET_CLS_ACT
497 	return skb->tc_at_ingress;
498 #else
499 	return false;
500 #endif
501 }
502 
503 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
504 {
505 #ifdef CONFIG_NET_CLS_ACT
506 	if (skb->tc_skip_classify) {
507 		skb->tc_skip_classify = 0;
508 		return true;
509 	}
510 #endif
511 	return false;
512 }
513 
514 /* Reset all TX qdiscs greater then index of a device.  */
515 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
516 {
517 	struct Qdisc *qdisc;
518 
519 	for (; i < dev->num_tx_queues; i++) {
520 		qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
521 		if (qdisc) {
522 			spin_lock_bh(qdisc_lock(qdisc));
523 			qdisc_reset(qdisc);
524 			spin_unlock_bh(qdisc_lock(qdisc));
525 		}
526 	}
527 }
528 
529 static inline void qdisc_reset_all_tx(struct net_device *dev)
530 {
531 	qdisc_reset_all_tx_gt(dev, 0);
532 }
533 
534 /* Are all TX queues of the device empty?  */
535 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
536 {
537 	unsigned int i;
538 
539 	rcu_read_lock();
540 	for (i = 0; i < dev->num_tx_queues; i++) {
541 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
542 		const struct Qdisc *q = rcu_dereference(txq->qdisc);
543 
544 		if (q->q.qlen) {
545 			rcu_read_unlock();
546 			return false;
547 		}
548 	}
549 	rcu_read_unlock();
550 	return true;
551 }
552 
553 /* Are any of the TX qdiscs changing?  */
554 static inline bool qdisc_tx_changing(const struct net_device *dev)
555 {
556 	unsigned int i;
557 
558 	for (i = 0; i < dev->num_tx_queues; i++) {
559 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
560 		if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
561 			return true;
562 	}
563 	return false;
564 }
565 
566 /* Is the device using the noop qdisc on all queues?  */
567 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
568 {
569 	unsigned int i;
570 
571 	for (i = 0; i < dev->num_tx_queues; i++) {
572 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
573 		if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
574 			return false;
575 	}
576 	return true;
577 }
578 
579 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
580 {
581 	return qdisc_skb_cb(skb)->pkt_len;
582 }
583 
584 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
585 enum net_xmit_qdisc_t {
586 	__NET_XMIT_STOLEN = 0x00010000,
587 	__NET_XMIT_BYPASS = 0x00020000,
588 };
589 
590 #ifdef CONFIG_NET_CLS_ACT
591 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
592 #else
593 #define net_xmit_drop_count(e)	(1)
594 #endif
595 
596 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
597 					   const struct Qdisc *sch)
598 {
599 #ifdef CONFIG_NET_SCHED
600 	struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
601 
602 	if (stab)
603 		__qdisc_calculate_pkt_len(skb, stab);
604 #endif
605 }
606 
607 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
608 				struct sk_buff **to_free)
609 {
610 	qdisc_calculate_pkt_len(skb, sch);
611 	return sch->enqueue(skb, sch, to_free);
612 }
613 
614 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
615 {
616 	return q->flags & TCQ_F_CPUSTATS;
617 }
618 
619 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
620 				  __u64 bytes, __u32 packets)
621 {
622 	bstats->bytes += bytes;
623 	bstats->packets += packets;
624 }
625 
626 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
627 				 const struct sk_buff *skb)
628 {
629 	_bstats_update(bstats,
630 		       qdisc_pkt_len(skb),
631 		       skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
632 }
633 
634 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
635 				      __u64 bytes, __u32 packets)
636 {
637 	u64_stats_update_begin(&bstats->syncp);
638 	_bstats_update(&bstats->bstats, bytes, packets);
639 	u64_stats_update_end(&bstats->syncp);
640 }
641 
642 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
643 				     const struct sk_buff *skb)
644 {
645 	u64_stats_update_begin(&bstats->syncp);
646 	bstats_update(&bstats->bstats, skb);
647 	u64_stats_update_end(&bstats->syncp);
648 }
649 
650 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
651 					   const struct sk_buff *skb)
652 {
653 	bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
654 }
655 
656 static inline void qdisc_bstats_update(struct Qdisc *sch,
657 				       const struct sk_buff *skb)
658 {
659 	bstats_update(&sch->bstats, skb);
660 }
661 
662 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
663 					    const struct sk_buff *skb)
664 {
665 	sch->qstats.backlog -= qdisc_pkt_len(skb);
666 }
667 
668 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
669 						const struct sk_buff *skb)
670 {
671 	this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
672 }
673 
674 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
675 					    const struct sk_buff *skb)
676 {
677 	sch->qstats.backlog += qdisc_pkt_len(skb);
678 }
679 
680 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
681 						const struct sk_buff *skb)
682 {
683 	this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
684 }
685 
686 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
687 {
688 	this_cpu_inc(sch->cpu_qstats->qlen);
689 }
690 
691 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
692 {
693 	this_cpu_dec(sch->cpu_qstats->qlen);
694 }
695 
696 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
697 {
698 	this_cpu_inc(sch->cpu_qstats->requeues);
699 }
700 
701 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
702 {
703 	sch->qstats.drops += count;
704 }
705 
706 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
707 {
708 	qstats->drops++;
709 }
710 
711 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
712 {
713 	qstats->overlimits++;
714 }
715 
716 static inline void qdisc_qstats_drop(struct Qdisc *sch)
717 {
718 	qstats_drop_inc(&sch->qstats);
719 }
720 
721 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
722 {
723 	this_cpu_inc(sch->cpu_qstats->drops);
724 }
725 
726 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
727 {
728 	sch->qstats.overlimits++;
729 }
730 
731 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
732 {
733 	qh->head = NULL;
734 	qh->tail = NULL;
735 	qh->qlen = 0;
736 }
737 
738 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
739 				       struct qdisc_skb_head *qh)
740 {
741 	struct sk_buff *last = qh->tail;
742 
743 	if (last) {
744 		skb->next = NULL;
745 		last->next = skb;
746 		qh->tail = skb;
747 	} else {
748 		qh->tail = skb;
749 		qh->head = skb;
750 	}
751 	qh->qlen++;
752 	qdisc_qstats_backlog_inc(sch, skb);
753 
754 	return NET_XMIT_SUCCESS;
755 }
756 
757 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
758 {
759 	return __qdisc_enqueue_tail(skb, sch, &sch->q);
760 }
761 
762 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
763 {
764 	struct sk_buff *skb = qh->head;
765 
766 	if (likely(skb != NULL)) {
767 		qh->head = skb->next;
768 		qh->qlen--;
769 		if (qh->head == NULL)
770 			qh->tail = NULL;
771 		skb->next = NULL;
772 	}
773 
774 	return skb;
775 }
776 
777 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
778 {
779 	struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
780 
781 	if (likely(skb != NULL)) {
782 		qdisc_qstats_backlog_dec(sch, skb);
783 		qdisc_bstats_update(sch, skb);
784 	}
785 
786 	return skb;
787 }
788 
789 /* Instead of calling kfree_skb() while root qdisc lock is held,
790  * queue the skb for future freeing at end of __dev_xmit_skb()
791  */
792 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
793 {
794 	skb->next = *to_free;
795 	*to_free = skb;
796 }
797 
798 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
799 						   struct qdisc_skb_head *qh,
800 						   struct sk_buff **to_free)
801 {
802 	struct sk_buff *skb = __qdisc_dequeue_head(qh);
803 
804 	if (likely(skb != NULL)) {
805 		unsigned int len = qdisc_pkt_len(skb);
806 
807 		qdisc_qstats_backlog_dec(sch, skb);
808 		__qdisc_drop(skb, to_free);
809 		return len;
810 	}
811 
812 	return 0;
813 }
814 
815 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
816 						 struct sk_buff **to_free)
817 {
818 	return __qdisc_queue_drop_head(sch, &sch->q, to_free);
819 }
820 
821 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
822 {
823 	const struct qdisc_skb_head *qh = &sch->q;
824 
825 	return qh->head;
826 }
827 
828 /* generic pseudo peek method for non-work-conserving qdisc */
829 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
830 {
831 	struct sk_buff *skb = skb_peek(&sch->gso_skb);
832 
833 	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
834 	if (!skb) {
835 		skb = sch->dequeue(sch);
836 
837 		if (skb) {
838 			__skb_queue_head(&sch->gso_skb, skb);
839 			/* it's still part of the queue */
840 			qdisc_qstats_backlog_inc(sch, skb);
841 			sch->q.qlen++;
842 		}
843 	}
844 
845 	return skb;
846 }
847 
848 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
849 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
850 {
851 	struct sk_buff *skb = skb_peek(&sch->gso_skb);
852 
853 	if (skb) {
854 		skb = __skb_dequeue(&sch->gso_skb);
855 		qdisc_qstats_backlog_dec(sch, skb);
856 		sch->q.qlen--;
857 	} else {
858 		skb = sch->dequeue(sch);
859 	}
860 
861 	return skb;
862 }
863 
864 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
865 {
866 	/*
867 	 * We do not know the backlog in bytes of this list, it
868 	 * is up to the caller to correct it
869 	 */
870 	ASSERT_RTNL();
871 	if (qh->qlen) {
872 		rtnl_kfree_skbs(qh->head, qh->tail);
873 
874 		qh->head = NULL;
875 		qh->tail = NULL;
876 		qh->qlen = 0;
877 	}
878 }
879 
880 static inline void qdisc_reset_queue(struct Qdisc *sch)
881 {
882 	__qdisc_reset_queue(&sch->q);
883 	sch->qstats.backlog = 0;
884 }
885 
886 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
887 					  struct Qdisc **pold)
888 {
889 	struct Qdisc *old;
890 
891 	sch_tree_lock(sch);
892 	old = *pold;
893 	*pold = new;
894 	if (old != NULL) {
895 		unsigned int qlen = old->q.qlen;
896 		unsigned int backlog = old->qstats.backlog;
897 
898 		qdisc_reset(old);
899 		qdisc_tree_reduce_backlog(old, qlen, backlog);
900 	}
901 	sch_tree_unlock(sch);
902 
903 	return old;
904 }
905 
906 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
907 {
908 	rtnl_kfree_skbs(skb, skb);
909 	qdisc_qstats_drop(sch);
910 }
911 
912 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
913 				 struct sk_buff **to_free)
914 {
915 	__qdisc_drop(skb, to_free);
916 	qdisc_qstats_cpu_drop(sch);
917 
918 	return NET_XMIT_DROP;
919 }
920 
921 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
922 			     struct sk_buff **to_free)
923 {
924 	__qdisc_drop(skb, to_free);
925 	qdisc_qstats_drop(sch);
926 
927 	return NET_XMIT_DROP;
928 }
929 
930 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
931    long it will take to send a packet given its size.
932  */
933 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
934 {
935 	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
936 	if (slot < 0)
937 		slot = 0;
938 	slot >>= rtab->rate.cell_log;
939 	if (slot > 255)
940 		return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
941 	return rtab->data[slot];
942 }
943 
944 struct psched_ratecfg {
945 	u64	rate_bytes_ps; /* bytes per second */
946 	u32	mult;
947 	u16	overhead;
948 	u8	linklayer;
949 	u8	shift;
950 };
951 
952 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
953 				unsigned int len)
954 {
955 	len += r->overhead;
956 
957 	if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
958 		return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
959 
960 	return ((u64)len * r->mult) >> r->shift;
961 }
962 
963 void psched_ratecfg_precompute(struct psched_ratecfg *r,
964 			       const struct tc_ratespec *conf,
965 			       u64 rate64);
966 
967 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
968 					  const struct psched_ratecfg *r)
969 {
970 	memset(res, 0, sizeof(*res));
971 
972 	/* legacy struct tc_ratespec has a 32bit @rate field
973 	 * Qdisc using 64bit rate should add new attributes
974 	 * in order to maintain compatibility.
975 	 */
976 	res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
977 
978 	res->overhead = r->overhead;
979 	res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
980 }
981 
982 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
983  * The fast path only needs to access filter list and to update stats
984  */
985 struct mini_Qdisc {
986 	struct tcf_proto *filter_list;
987 	struct gnet_stats_basic_cpu __percpu *cpu_bstats;
988 	struct gnet_stats_queue	__percpu *cpu_qstats;
989 	struct rcu_head rcu;
990 };
991 
992 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
993 						const struct sk_buff *skb)
994 {
995 	bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
996 }
997 
998 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
999 {
1000 	this_cpu_inc(miniq->cpu_qstats->drops);
1001 }
1002 
1003 struct mini_Qdisc_pair {
1004 	struct mini_Qdisc miniq1;
1005 	struct mini_Qdisc miniq2;
1006 	struct mini_Qdisc __rcu **p_miniq;
1007 };
1008 
1009 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1010 			  struct tcf_proto *tp_head);
1011 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1012 			  struct mini_Qdisc __rcu **p_miniq);
1013 
1014 #endif
1015