xref: /linux/include/net/sch_generic.h (revision 8341c989ac77d712c7d6e2bce29e8a4bcb2eeae4)
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 <linux/mutex.h>
16 #include <linux/rwsem.h>
17 #include <linux/atomic.h>
18 #include <linux/hashtable.h>
19 #include <net/gen_stats.h>
20 #include <net/rtnetlink.h>
21 #include <net/flow_offload.h>
22 #include <linux/xarray.h>
23 #include <net/dropreason-qdisc.h>
24 
25 struct Qdisc_ops;
26 struct qdisc_walker;
27 struct tcf_walker;
28 struct module;
29 struct bpf_flow_keys;
30 struct Qdisc;
31 struct netdev_queue;
32 
33 struct qdisc_rate_table {
34 	struct tc_ratespec rate;
35 	u32		data[256];
36 	struct qdisc_rate_table *next;
37 	int		refcnt;
38 };
39 
40 enum qdisc_state_t {
41 	__QDISC_STATE_SCHED,
42 	__QDISC_STATE_DEACTIVATED,
43 	__QDISC_STATE_MISSED,
44 	__QDISC_STATE_DRAINING,
45 };
46 
47 #define QDISC_STATE_MISSED	BIT(__QDISC_STATE_MISSED)
48 #define QDISC_STATE_DRAINING	BIT(__QDISC_STATE_DRAINING)
49 
50 #define QDISC_STATE_NON_EMPTY	(QDISC_STATE_MISSED | \
51 					QDISC_STATE_DRAINING)
52 
53 struct qdisc_size_table {
54 	struct rcu_head		rcu;
55 	struct list_head	list;
56 	struct tc_sizespec	szopts;
57 	int			refcnt;
58 	u16			data[];
59 };
60 
61 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
62 struct qdisc_skb_head {
63 	struct sk_buff	*head;
64 	struct sk_buff	*tail;
65 	__u32		qlen;
66 	spinlock_t	lock;
67 };
68 
69 struct Qdisc {
70 	int 			(*enqueue)(struct sk_buff *skb,
71 					   struct Qdisc *sch,
72 					   struct sk_buff **to_free);
73 	struct sk_buff *	(*dequeue)(struct Qdisc *sch);
74 	unsigned int		flags;
75 #define TCQ_F_BUILTIN		1
76 #define TCQ_F_INGRESS		2
77 #define TCQ_F_CAN_BYPASS	4
78 #define TCQ_F_MQROOT		8
79 #define TCQ_F_ONETXQUEUE	0x10 /* dequeue_skb() can assume all skbs are for
80 				      * q->dev_queue : It can test
81 				      * netif_xmit_frozen_or_stopped() before
82 				      * dequeueing next packet.
83 				      * Its true for MQ/MQPRIO slaves, or non
84 				      * multiqueue device.
85 				      */
86 #define TCQ_F_WARN_NONWC	(1 << 16)
87 #define TCQ_F_CPUSTATS		0x20 /* run using percpu statistics */
88 #define TCQ_F_NOPARENT		0x40 /* root of its hierarchy :
89 				      * qdisc_tree_decrease_qlen() should stop.
90 				      */
91 #define TCQ_F_INVISIBLE		0x80 /* invisible by default in dump */
92 #define TCQ_F_NOLOCK		0x100 /* qdisc does not require locking */
93 #define TCQ_F_OFFLOADED		0x200 /* qdisc is offloaded to HW */
94 #define TCQ_F_DEQUEUE_DROPS	0x400 /* ->dequeue() can drop packets in q->to_free */
95 
96 	u32			limit;
97 	const struct Qdisc_ops	*ops;
98 	struct qdisc_size_table	__rcu *stab;
99 	struct hlist_node       hash;
100 	u32			handle;
101 	u32			parent;
102 
103 	struct netdev_queue	*dev_queue;
104 
105 	struct net_rate_estimator __rcu *rate_est;
106 	struct gnet_stats_basic_sync __percpu *cpu_bstats;
107 	struct gnet_stats_queue	__percpu *cpu_qstats;
108 	int			pad;
109 	refcount_t		refcnt;
110 
111 	/* Cache line potentially dirtied in dequeue() or __netif_reschedule(). */
112 	__cacheline_group_begin(Qdisc_read_mostly) ____cacheline_aligned;
113 		struct sk_buff_head	gso_skb;
114 		struct Qdisc		*next_sched;
115 		struct sk_buff_head	skb_bad_txq;
116 	__cacheline_group_end(Qdisc_read_mostly);
117 
118 	/* Fields dirtied in dequeue() fast path. */
119 	__cacheline_group_begin(Qdisc_write) ____cacheline_aligned;
120 		struct qdisc_skb_head	q;
121 		unsigned long		state;
122 		struct gnet_stats_basic_sync bstats;
123 		bool			running; /* must be written under qdisc spinlock */
124 
125 		/* Note : we only change qstats.backlog in fast path. */
126 		struct gnet_stats_queue	qstats;
127 
128 		struct sk_buff		*to_free;
129 	__cacheline_group_end(Qdisc_write);
130 
131 
132 	atomic_long_t		defer_count ____cacheline_aligned_in_smp;
133 	struct llist_head	defer_list;
134 
135 	spinlock_t		seqlock;
136 
137 	struct rcu_head		rcu;
138 	netdevice_tracker	dev_tracker;
139 	struct lock_class_key	root_lock_key;
140 	/* private data */
141 	long privdata[] ____cacheline_aligned;
142 };
143 
144 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
145 {
146 	if (qdisc->flags & TCQ_F_BUILTIN)
147 		return;
148 	refcount_inc(&qdisc->refcnt);
149 }
150 
151 static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc)
152 {
153 	if (qdisc->flags & TCQ_F_BUILTIN)
154 		return true;
155 	return refcount_dec_if_one(&qdisc->refcnt);
156 }
157 
158 /* Intended to be used by unlocked users, when concurrent qdisc release is
159  * possible.
160  */
161 
162 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
163 {
164 	if (qdisc->flags & TCQ_F_BUILTIN)
165 		return qdisc;
166 	if (refcount_inc_not_zero(&qdisc->refcnt))
167 		return qdisc;
168 	return NULL;
169 }
170 
171 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
172  * root_lock section, or provide their own memory barriers -- ordering
173  * against qdisc_run_begin/end() atomic bit operations.
174  */
175 static inline bool qdisc_is_running(struct Qdisc *qdisc)
176 {
177 	if (qdisc->flags & TCQ_F_NOLOCK)
178 		return spin_is_locked(&qdisc->seqlock);
179 	return READ_ONCE(qdisc->running);
180 }
181 
182 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
183 {
184 	return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
185 }
186 
187 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
188 {
189 	return q->flags & TCQ_F_CPUSTATS;
190 }
191 
192 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
193 {
194 	if (qdisc_is_percpu_stats(qdisc))
195 		return nolock_qdisc_is_empty(qdisc);
196 	return !READ_ONCE(qdisc->q.qlen);
197 }
198 
199 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
200  * the qdisc root lock acquired.
201  */
202 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
203 {
204 	if (qdisc->flags & TCQ_F_NOLOCK) {
205 		if (spin_trylock(&qdisc->seqlock))
206 			return true;
207 
208 		/* No need to insist if the MISSED flag was already set.
209 		 * Note that test_and_set_bit() also gives us memory ordering
210 		 * guarantees wrt potential earlier enqueue() and below
211 		 * spin_trylock(), both of which are necessary to prevent races
212 		 */
213 		if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
214 			return false;
215 
216 		/* Try to take the lock again to make sure that we will either
217 		 * grab it or the CPU that still has it will see MISSED set
218 		 * when testing it in qdisc_run_end()
219 		 */
220 		return spin_trylock(&qdisc->seqlock);
221 	}
222 	if (READ_ONCE(qdisc->running))
223 		return false;
224 	WRITE_ONCE(qdisc->running, true);
225 	return true;
226 }
227 
228 static inline struct sk_buff *qdisc_run_end(struct Qdisc *qdisc)
229 {
230 	struct sk_buff *to_free = NULL;
231 
232 	if (qdisc->flags & TCQ_F_NOLOCK) {
233 		spin_unlock(&qdisc->seqlock);
234 
235 		/* spin_unlock() only has store-release semantic. The unlock
236 		 * and test_bit() ordering is a store-load ordering, so a full
237 		 * memory barrier is needed here.
238 		 */
239 		smp_mb();
240 
241 		if (unlikely(test_bit(__QDISC_STATE_MISSED,
242 				      &qdisc->state)))
243 			__netif_schedule(qdisc);
244 		return NULL;
245 	}
246 
247 	if (qdisc->flags & TCQ_F_DEQUEUE_DROPS) {
248 		to_free = qdisc->to_free;
249 		if (to_free)
250 			qdisc->to_free = NULL;
251 	}
252 	WRITE_ONCE(qdisc->running, false);
253 	return to_free;
254 }
255 
256 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
257 {
258 	return qdisc->flags & TCQ_F_ONETXQUEUE;
259 }
260 
261 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
262 {
263 	return netdev_queue_dql_avail(txq);
264 }
265 
266 struct Qdisc_class_ops {
267 	unsigned int		flags;
268 	/* Child qdisc manipulation */
269 	struct netdev_queue *	(*select_queue)(struct Qdisc *, struct tcmsg *);
270 	int			(*graft)(struct Qdisc *, unsigned long cl,
271 					struct Qdisc *, struct Qdisc **,
272 					struct netlink_ext_ack *extack);
273 	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
274 	void			(*qlen_notify)(struct Qdisc *, unsigned long);
275 
276 	/* Class manipulation routines */
277 	unsigned long		(*find)(struct Qdisc *, u32 classid);
278 	int			(*change)(struct Qdisc *, u32, u32,
279 					struct nlattr **, unsigned long *,
280 					struct netlink_ext_ack *);
281 	int			(*delete)(struct Qdisc *, unsigned long,
282 					  struct netlink_ext_ack *);
283 	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
284 
285 	/* Filter manipulation */
286 	struct tcf_block *	(*tcf_block)(struct Qdisc *sch,
287 					     unsigned long arg,
288 					     struct netlink_ext_ack *extack);
289 	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
290 					u32 classid);
291 	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
292 
293 	/* rtnetlink specific */
294 	int			(*dump)(struct Qdisc *, unsigned long,
295 					struct sk_buff *skb, struct tcmsg*);
296 	int			(*dump_stats)(struct Qdisc *, unsigned long,
297 					struct gnet_dump *);
298 };
299 
300 /* Qdisc_class_ops flag values */
301 
302 /* Implements API that doesn't require rtnl lock */
303 enum qdisc_class_ops_flags {
304 	QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
305 };
306 
307 struct Qdisc_ops {
308 	struct Qdisc_ops	*next;
309 	const struct Qdisc_class_ops	*cl_ops;
310 	char			id[IFNAMSIZ];
311 	int			priv_size;
312 	unsigned int		static_flags;
313 
314 	int 			(*enqueue)(struct sk_buff *skb,
315 					   struct Qdisc *sch,
316 					   struct sk_buff **to_free);
317 	struct sk_buff *	(*dequeue)(struct Qdisc *);
318 	struct sk_buff *	(*peek)(struct Qdisc *);
319 
320 	int			(*init)(struct Qdisc *sch, struct nlattr *arg,
321 					struct netlink_ext_ack *extack);
322 	void			(*reset)(struct Qdisc *);
323 	void			(*destroy)(struct Qdisc *);
324 	int			(*change)(struct Qdisc *sch,
325 					  struct nlattr *arg,
326 					  struct netlink_ext_ack *extack);
327 	void			(*attach)(struct Qdisc *sch);
328 	int			(*change_tx_queue_len)(struct Qdisc *, unsigned int);
329 	void			(*change_real_num_tx)(struct Qdisc *sch,
330 						      unsigned int new_real_tx);
331 
332 	int			(*dump)(struct Qdisc *, struct sk_buff *);
333 	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
334 
335 	void			(*ingress_block_set)(struct Qdisc *sch,
336 						     u32 block_index);
337 	void			(*egress_block_set)(struct Qdisc *sch,
338 						    u32 block_index);
339 	u32			(*ingress_block_get)(struct Qdisc *sch);
340 	u32			(*egress_block_get)(struct Qdisc *sch);
341 
342 	struct module		*owner;
343 };
344 
345 struct tcf_result {
346 	union {
347 		struct {
348 			unsigned long	class;
349 			u32		classid;
350 		};
351 		const struct tcf_proto *goto_tp;
352 	};
353 };
354 
355 struct tcf_chain;
356 
357 struct tcf_proto_ops {
358 	struct list_head	head;
359 	char			kind[IFNAMSIZ];
360 
361 	int			(*classify)(struct sk_buff *,
362 					    const struct tcf_proto *,
363 					    struct tcf_result *);
364 	int			(*init)(struct tcf_proto*);
365 	void			(*destroy)(struct tcf_proto *tp, bool rtnl_held,
366 					   struct netlink_ext_ack *extack);
367 
368 	void*			(*get)(struct tcf_proto*, u32 handle);
369 	void			(*put)(struct tcf_proto *tp, void *f);
370 	int			(*change)(struct net *net, struct sk_buff *,
371 					struct tcf_proto*, unsigned long,
372 					u32 handle, struct nlattr **,
373 					void **, u32,
374 					struct netlink_ext_ack *);
375 	int			(*delete)(struct tcf_proto *tp, void *arg,
376 					  bool *last, bool rtnl_held,
377 					  struct netlink_ext_ack *);
378 	bool			(*delete_empty)(struct tcf_proto *tp);
379 	void			(*walk)(struct tcf_proto *tp,
380 					struct tcf_walker *arg, bool rtnl_held);
381 	int			(*reoffload)(struct tcf_proto *tp, bool add,
382 					     flow_setup_cb_t *cb, void *cb_priv,
383 					     struct netlink_ext_ack *extack);
384 	void			(*hw_add)(struct tcf_proto *tp,
385 					  void *type_data);
386 	void			(*hw_del)(struct tcf_proto *tp,
387 					  void *type_data);
388 	void			(*bind_class)(void *, u32, unsigned long,
389 					      void *, unsigned long);
390 	void *			(*tmplt_create)(struct net *net,
391 						struct tcf_chain *chain,
392 						struct nlattr **tca,
393 						struct netlink_ext_ack *extack);
394 	void			(*tmplt_destroy)(void *tmplt_priv);
395 	void			(*tmplt_reoffload)(struct tcf_chain *chain,
396 						   bool add,
397 						   flow_setup_cb_t *cb,
398 						   void *cb_priv);
399 	struct tcf_exts *	(*get_exts)(const struct tcf_proto *tp,
400 					    u32 handle);
401 
402 	/* rtnetlink specific */
403 	int			(*dump)(struct net*, struct tcf_proto*, void *,
404 					struct sk_buff *skb, struct tcmsg*,
405 					bool);
406 	int			(*terse_dump)(struct net *net,
407 					      struct tcf_proto *tp, void *fh,
408 					      struct sk_buff *skb,
409 					      struct tcmsg *t, bool rtnl_held);
410 	int			(*tmplt_dump)(struct sk_buff *skb,
411 					      struct net *net,
412 					      void *tmplt_priv);
413 
414 	struct module		*owner;
415 	int			flags;
416 };
417 
418 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
419  * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
420  * conditions can occur when filters are inserted/deleted simultaneously.
421  */
422 enum tcf_proto_ops_flags {
423 	TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
424 };
425 
426 struct tcf_proto {
427 	/* Fast access part */
428 	struct tcf_proto __rcu	*next;
429 	void __rcu		*root;
430 
431 	/* called under RCU BH lock*/
432 	int			(*classify)(struct sk_buff *,
433 					    const struct tcf_proto *,
434 					    struct tcf_result *);
435 	__be16			protocol;
436 
437 	/* All the rest */
438 	u32			prio;
439 	void			*data;
440 	const struct tcf_proto_ops	*ops;
441 	struct tcf_chain	*chain;
442 	/* Lock protects tcf_proto shared state and can be used by unlocked
443 	 * classifiers to protect their private data.
444 	 */
445 	spinlock_t		lock;
446 	bool			deleting;
447 	bool			counted;
448 	bool			usesw;
449 	refcount_t		refcnt;
450 	struct rcu_head		rcu;
451 	struct hlist_node	destroy_ht_node;
452 };
453 
454 struct qdisc_skb_cb {
455 	unsigned int		pkt_len;
456 	u16			pkt_segs;
457 	u16			tc_classid;
458 #define QDISC_CB_PRIV_LEN 20
459 	unsigned char		data[QDISC_CB_PRIV_LEN];
460 
461 	u16			slave_dev_queue_mapping;
462 	u8			post_ct:1;
463 	u8			post_ct_snat:1;
464 	u8			post_ct_dnat:1;
465 };
466 
467 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
468 
469 struct tcf_chain {
470 	/* Protects filter_chain. */
471 	struct mutex filter_chain_lock;
472 	struct tcf_proto __rcu *filter_chain;
473 	struct list_head list;
474 	struct tcf_block *block;
475 	u32 index; /* chain index */
476 	unsigned int refcnt;
477 	unsigned int action_refcnt;
478 	bool explicitly_created;
479 	bool flushing;
480 	const struct tcf_proto_ops *tmplt_ops;
481 	void *tmplt_priv;
482 	struct rcu_head rcu;
483 };
484 
485 struct tcf_block {
486 	struct xarray ports; /* datapath accessible */
487 	/* Lock protects tcf_block and lifetime-management data of chains
488 	 * attached to the block (refcnt, action_refcnt, explicitly_created).
489 	 */
490 	struct mutex lock;
491 	struct list_head chain_list;
492 	u32 index; /* block index for shared blocks */
493 	u32 classid; /* which class this block belongs to */
494 	refcount_t refcnt;
495 	struct net *net;
496 	struct Qdisc *q;
497 	struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
498 	struct flow_block flow_block;
499 	struct list_head owner_list;
500 	bool keep_dst;
501 	atomic_t useswcnt;
502 	atomic_t offloadcnt; /* Number of oddloaded filters */
503 	unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
504 	unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
505 	struct {
506 		struct tcf_chain *chain;
507 		struct list_head filter_chain_list;
508 	} chain0;
509 	struct rcu_head rcu;
510 	DECLARE_HASHTABLE(proto_destroy_ht, 7);
511 	struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
512 };
513 
514 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index);
515 
516 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
517 {
518 	return lockdep_is_held(&chain->filter_chain_lock);
519 }
520 
521 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
522 {
523 	return lockdep_is_held(&tp->lock);
524 }
525 
526 #define tcf_chain_dereference(p, chain)					\
527 	rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
528 
529 #define tcf_proto_dereference(p, tp)					\
530 	rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
531 
532 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
533 {
534 	struct qdisc_skb_cb *qcb;
535 
536 	BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
537 	BUILD_BUG_ON(sizeof(qcb->data) < sz);
538 }
539 
540 static inline int qdisc_qlen(const struct Qdisc *q)
541 {
542 	return q->q.qlen;
543 }
544 
545 static inline int qdisc_qlen_sum(const struct Qdisc *q)
546 {
547 	__u32 qlen = q->qstats.qlen;
548 	int i;
549 
550 	if (qdisc_is_percpu_stats(q)) {
551 		for_each_possible_cpu(i)
552 			qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
553 	} else {
554 		qlen += q->q.qlen;
555 	}
556 
557 	return qlen;
558 }
559 
560 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
561 {
562 	return (struct qdisc_skb_cb *)skb->cb;
563 }
564 
565 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
566 {
567 	return &qdisc->q.lock;
568 }
569 
570 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
571 {
572 	struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
573 
574 	return q;
575 }
576 
577 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
578 {
579 	return rcu_dereference_bh(qdisc->dev_queue->qdisc);
580 }
581 
582 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
583 {
584 	return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
585 }
586 
587 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
588 {
589 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
590 
591 	ASSERT_RTNL();
592 	return qdisc_lock(root);
593 }
594 
595 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
596 {
597 	return qdisc->dev_queue->dev;
598 }
599 
600 static inline void sch_tree_lock(struct Qdisc *q)
601 {
602 	if (q->flags & TCQ_F_MQROOT)
603 		spin_lock_bh(qdisc_lock(q));
604 	else
605 		spin_lock_bh(qdisc_root_sleeping_lock(q));
606 }
607 
608 static inline void sch_tree_unlock(struct Qdisc *q)
609 {
610 	if (q->flags & TCQ_F_MQROOT)
611 		spin_unlock_bh(qdisc_lock(q));
612 	else
613 		spin_unlock_bh(qdisc_root_sleeping_lock(q));
614 }
615 
616 extern struct Qdisc noop_qdisc;
617 extern struct Qdisc_ops noop_qdisc_ops;
618 extern struct Qdisc_ops pfifo_fast_ops;
619 extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
620 extern struct Qdisc_ops mq_qdisc_ops;
621 extern struct Qdisc_ops noqueue_qdisc_ops;
622 extern const struct Qdisc_ops *default_qdisc_ops;
623 static inline const struct Qdisc_ops *
624 get_default_qdisc_ops(const struct net_device *dev, int ntx)
625 {
626 	return ntx < dev->real_num_tx_queues ?
627 			default_qdisc_ops : &pfifo_fast_ops;
628 }
629 
630 struct Qdisc_class_common {
631 	u32			classid;
632 	unsigned int		filter_cnt;
633 	struct hlist_node	hnode;
634 };
635 
636 struct Qdisc_class_hash {
637 	struct hlist_head	*hash;
638 	unsigned int		hashsize;
639 	unsigned int		hashmask;
640 	unsigned int		hashelems;
641 };
642 
643 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
644 {
645 	id ^= id >> 8;
646 	id ^= id >> 4;
647 	return id & mask;
648 }
649 
650 static inline struct Qdisc_class_common *
651 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
652 {
653 	struct Qdisc_class_common *cl;
654 	unsigned int h;
655 
656 	if (!id)
657 		return NULL;
658 
659 	h = qdisc_class_hash(id, hash->hashmask);
660 	hlist_for_each_entry(cl, &hash->hash[h], hnode) {
661 		if (cl->classid == id)
662 			return cl;
663 	}
664 	return NULL;
665 }
666 
667 static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
668 {
669 	return cl->filter_cnt > 0;
670 }
671 
672 static inline void qdisc_class_get(struct Qdisc_class_common *cl)
673 {
674 	unsigned int res;
675 
676 	if (check_add_overflow(cl->filter_cnt, 1, &res))
677 		WARN(1, "Qdisc class overflow");
678 
679 	cl->filter_cnt = res;
680 }
681 
682 static inline void qdisc_class_put(struct Qdisc_class_common *cl)
683 {
684 	unsigned int res;
685 
686 	if (check_sub_overflow(cl->filter_cnt, 1, &res))
687 		WARN(1, "Qdisc class underflow");
688 
689 	cl->filter_cnt = res;
690 }
691 
692 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
693 {
694 	u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
695 
696 	return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
697 }
698 
699 int qdisc_class_hash_init(struct Qdisc_class_hash *);
700 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
701 			     struct Qdisc_class_common *);
702 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
703 			     struct Qdisc_class_common *);
704 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
705 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
706 
707 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
708 void dev_qdisc_change_real_num_tx(struct net_device *dev,
709 				  unsigned int new_real_tx);
710 void dev_init_scheduler(struct net_device *dev);
711 void dev_shutdown(struct net_device *dev);
712 void dev_activate(struct net_device *dev);
713 void dev_deactivate(struct net_device *dev);
714 void dev_deactivate_many(struct list_head *head);
715 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
716 			      struct Qdisc *qdisc);
717 void qdisc_reset(struct Qdisc *qdisc);
718 void qdisc_destroy(struct Qdisc *qdisc);
719 void qdisc_put(struct Qdisc *qdisc);
720 void qdisc_put_unlocked(struct Qdisc *qdisc);
721 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
722 #ifdef CONFIG_NET_SCHED
723 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
724 			      void *type_data);
725 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
726 				struct Qdisc *new, struct Qdisc *old,
727 				enum tc_setup_type type, void *type_data,
728 				struct netlink_ext_ack *extack);
729 #else
730 static inline int
731 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
732 			  void *type_data)
733 {
734 	q->flags &= ~TCQ_F_OFFLOADED;
735 	return 0;
736 }
737 
738 static inline void
739 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
740 			   struct Qdisc *new, struct Qdisc *old,
741 			   enum tc_setup_type type, void *type_data,
742 			   struct netlink_ext_ack *extack)
743 {
744 }
745 #endif
746 void qdisc_offload_query_caps(struct net_device *dev,
747 			      enum tc_setup_type type,
748 			      void *caps, size_t caps_len);
749 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
750 			  const struct Qdisc_ops *ops,
751 			  struct netlink_ext_ack *extack);
752 void qdisc_free(struct Qdisc *qdisc);
753 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
754 				const struct Qdisc_ops *ops, u32 parentid,
755 				struct netlink_ext_ack *extack);
756 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
757 			       const struct qdisc_size_table *stab);
758 int skb_do_redirect(struct sk_buff *);
759 
760 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
761 {
762 #ifdef CONFIG_NET_XGRESS
763 	return skb->tc_at_ingress;
764 #else
765 	return false;
766 #endif
767 }
768 
769 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
770 {
771 #ifdef CONFIG_NET_CLS_ACT
772 	if (skb->tc_skip_classify) {
773 		skb->tc_skip_classify = 0;
774 		return true;
775 	}
776 #endif
777 	return false;
778 }
779 
780 /* Reset all TX qdiscs greater than index of a device.  */
781 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
782 {
783 	struct Qdisc *qdisc;
784 
785 	for (; i < dev->num_tx_queues; i++) {
786 		qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
787 		if (qdisc) {
788 			spin_lock_bh(qdisc_lock(qdisc));
789 			qdisc_reset(qdisc);
790 			spin_unlock_bh(qdisc_lock(qdisc));
791 		}
792 	}
793 }
794 
795 /* Are all TX queues of the device empty?  */
796 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
797 {
798 	unsigned int i;
799 
800 	rcu_read_lock();
801 	for (i = 0; i < dev->num_tx_queues; i++) {
802 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
803 		const struct Qdisc *q = rcu_dereference(txq->qdisc);
804 
805 		if (!qdisc_is_empty(q)) {
806 			rcu_read_unlock();
807 			return false;
808 		}
809 	}
810 	rcu_read_unlock();
811 	return true;
812 }
813 
814 /* Are any of the TX qdiscs changing?  */
815 static inline bool qdisc_tx_changing(const struct net_device *dev)
816 {
817 	unsigned int i;
818 
819 	for (i = 0; i < dev->num_tx_queues; i++) {
820 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
821 
822 		if (rcu_access_pointer(txq->qdisc) !=
823 		    rcu_access_pointer(txq->qdisc_sleeping))
824 			return true;
825 	}
826 	return false;
827 }
828 
829 /* "noqueue" qdisc identified by not having any enqueue, see noqueue_init() */
830 static inline bool qdisc_txq_has_no_queue(const struct netdev_queue *txq)
831 {
832 	struct Qdisc *qdisc = rcu_access_pointer(txq->qdisc);
833 
834 	return qdisc->enqueue == NULL;
835 }
836 
837 /* Is the device using the noop qdisc on all queues?  */
838 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
839 {
840 	unsigned int i;
841 
842 	for (i = 0; i < dev->num_tx_queues; i++) {
843 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
844 		if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
845 			return false;
846 	}
847 	return true;
848 }
849 
850 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
851 {
852 	return qdisc_skb_cb(skb)->pkt_len;
853 }
854 
855 static inline unsigned int qdisc_pkt_segs(const struct sk_buff *skb)
856 {
857 	u32 pkt_segs = qdisc_skb_cb(skb)->pkt_segs;
858 
859 	DEBUG_NET_WARN_ON_ONCE(pkt_segs !=
860 			(skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1));
861 	return pkt_segs;
862 }
863 
864 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
865 enum net_xmit_qdisc_t {
866 	__NET_XMIT_STOLEN = 0x00010000,
867 	__NET_XMIT_BYPASS = 0x00020000,
868 };
869 
870 #ifdef CONFIG_NET_CLS_ACT
871 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
872 #else
873 #define net_xmit_drop_count(e)	(1)
874 #endif
875 
876 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
877 					   const struct Qdisc *sch)
878 {
879 #ifdef CONFIG_NET_SCHED
880 	struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
881 
882 	if (stab)
883 		__qdisc_calculate_pkt_len(skb, stab);
884 #endif
885 }
886 
887 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
888 				struct sk_buff **to_free)
889 {
890 	return sch->enqueue(skb, sch, to_free);
891 }
892 
893 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
894 				  __u64 bytes, __u64 packets)
895 {
896 	u64_stats_update_begin(&bstats->syncp);
897 	u64_stats_add(&bstats->bytes, bytes);
898 	u64_stats_add(&bstats->packets, packets);
899 	u64_stats_update_end(&bstats->syncp);
900 }
901 
902 static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
903 				 const struct sk_buff *skb)
904 {
905 	_bstats_update(bstats, qdisc_pkt_len(skb), qdisc_pkt_segs(skb));
906 }
907 
908 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
909 					   const struct sk_buff *skb)
910 {
911 	bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
912 }
913 
914 static inline void qdisc_bstats_update(struct Qdisc *sch,
915 				       const struct sk_buff *skb)
916 {
917 	bstats_update(&sch->bstats, skb);
918 }
919 
920 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
921 					    const struct sk_buff *skb)
922 {
923 	sch->qstats.backlog -= qdisc_pkt_len(skb);
924 }
925 
926 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
927 						const struct sk_buff *skb)
928 {
929 	this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
930 }
931 
932 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
933 					    const struct sk_buff *skb)
934 {
935 	sch->qstats.backlog += qdisc_pkt_len(skb);
936 }
937 
938 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
939 						const struct sk_buff *skb)
940 {
941 	this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
942 }
943 
944 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
945 {
946 	this_cpu_inc(sch->cpu_qstats->qlen);
947 }
948 
949 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
950 {
951 	this_cpu_dec(sch->cpu_qstats->qlen);
952 }
953 
954 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
955 {
956 	this_cpu_inc(sch->cpu_qstats->requeues);
957 }
958 
959 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
960 {
961 	sch->qstats.drops += count;
962 }
963 
964 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
965 {
966 	qstats->drops++;
967 }
968 
969 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
970 {
971 	qstats->overlimits++;
972 }
973 
974 static inline void qdisc_qstats_drop(struct Qdisc *sch)
975 {
976 	qstats_drop_inc(&sch->qstats);
977 }
978 
979 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
980 {
981 	this_cpu_inc(sch->cpu_qstats->drops);
982 }
983 
984 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
985 {
986 	sch->qstats.overlimits++;
987 }
988 
989 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
990 {
991 	__u32 qlen = qdisc_qlen_sum(sch);
992 
993 	return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
994 }
995 
996 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch,  __u32 *qlen,
997 					     __u32 *backlog)
998 {
999 	struct gnet_stats_queue qstats = { 0 };
1000 
1001 	gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
1002 	*qlen = qstats.qlen + qdisc_qlen(sch);
1003 	*backlog = qstats.backlog;
1004 }
1005 
1006 static inline void qdisc_purge_queue(struct Qdisc *sch)
1007 {
1008 	__u32 qlen, backlog;
1009 
1010 	qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
1011 	qdisc_reset(sch);
1012 	qdisc_tree_reduce_backlog(sch, qlen, backlog);
1013 }
1014 
1015 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
1016 					struct qdisc_skb_head *qh)
1017 {
1018 	struct sk_buff *last = qh->tail;
1019 
1020 	if (last) {
1021 		skb->next = NULL;
1022 		last->next = skb;
1023 		qh->tail = skb;
1024 	} else {
1025 		qh->tail = skb;
1026 		qh->head = skb;
1027 	}
1028 	qh->qlen++;
1029 }
1030 
1031 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1032 {
1033 	__qdisc_enqueue_tail(skb, &sch->q);
1034 	qdisc_qstats_backlog_inc(sch, skb);
1035 	return NET_XMIT_SUCCESS;
1036 }
1037 
1038 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1039 					struct qdisc_skb_head *qh)
1040 {
1041 	skb->next = qh->head;
1042 
1043 	if (!qh->head)
1044 		qh->tail = skb;
1045 	qh->head = skb;
1046 	qh->qlen++;
1047 }
1048 
1049 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1050 {
1051 	struct sk_buff *skb = qh->head;
1052 
1053 	if (likely(skb != NULL)) {
1054 		qh->head = skb->next;
1055 		qh->qlen--;
1056 		if (qh->head == NULL)
1057 			qh->tail = NULL;
1058 		skb->next = NULL;
1059 	}
1060 
1061 	return skb;
1062 }
1063 
1064 static inline struct sk_buff *qdisc_dequeue_internal(struct Qdisc *sch, bool direct)
1065 {
1066 	struct sk_buff *skb;
1067 
1068 	skb = __skb_dequeue(&sch->gso_skb);
1069 	if (skb) {
1070 		sch->q.qlen--;
1071 		qdisc_qstats_backlog_dec(sch, skb);
1072 		return skb;
1073 	}
1074 	if (direct) {
1075 		skb = __qdisc_dequeue_head(&sch->q);
1076 		if (skb)
1077 			qdisc_qstats_backlog_dec(sch, skb);
1078 		return skb;
1079 	} else {
1080 		return sch->dequeue(sch);
1081 	}
1082 }
1083 
1084 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1085 {
1086 	struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1087 
1088 	if (likely(skb != NULL)) {
1089 		qdisc_qstats_backlog_dec(sch, skb);
1090 		qdisc_bstats_update(sch, skb);
1091 	}
1092 
1093 	return skb;
1094 }
1095 
1096 struct tc_skb_cb {
1097 	struct qdisc_skb_cb qdisc_cb;
1098 	u32 drop_reason;
1099 
1100 	u16 zone; /* Only valid if qdisc_skb_cb(skb)->post_ct = true */
1101 	u16 mru;
1102 };
1103 
1104 static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
1105 {
1106 	struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
1107 
1108 	BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
1109 	return cb;
1110 }
1111 
1112 /* TC classifier accessors - use enum skb_drop_reason */
1113 static inline enum skb_drop_reason
1114 tcf_get_drop_reason(const struct sk_buff *skb)
1115 {
1116 	return (enum skb_drop_reason)tc_skb_cb(skb)->drop_reason;
1117 }
1118 
1119 static inline void tcf_set_drop_reason(const struct sk_buff *skb,
1120 				       enum skb_drop_reason reason)
1121 {
1122 	tc_skb_cb(skb)->drop_reason = (enum qdisc_drop_reason)reason;
1123 }
1124 
1125 /* Qdisc accessors - use enum qdisc_drop_reason */
1126 static inline enum qdisc_drop_reason
1127 tcf_get_qdisc_drop_reason(const struct sk_buff *skb)
1128 {
1129 	return tc_skb_cb(skb)->drop_reason;
1130 }
1131 
1132 static inline void tcf_set_qdisc_drop_reason(const struct sk_buff *skb,
1133 					     enum qdisc_drop_reason reason)
1134 {
1135 	tc_skb_cb(skb)->drop_reason = reason;
1136 }
1137 
1138 void __tcf_kfree_skb_list(struct sk_buff *skb, struct Qdisc *q,
1139 			  struct netdev_queue *txq, struct net_device *dev);
1140 
1141 static inline void tcf_kfree_skb_list(struct sk_buff *skb, struct Qdisc *q,
1142 				      struct netdev_queue *txq,
1143 				      struct net_device *dev)
1144 {
1145 	if (unlikely(skb))
1146 		__tcf_kfree_skb_list(skb, q, txq, dev);
1147 }
1148 
1149 static inline void qdisc_dequeue_drop(struct Qdisc *q, struct sk_buff *skb,
1150 				      enum qdisc_drop_reason reason)
1151 {
1152 	DEBUG_NET_WARN_ON_ONCE(!(q->flags & TCQ_F_DEQUEUE_DROPS));
1153 	DEBUG_NET_WARN_ON_ONCE(q->flags & TCQ_F_NOLOCK);
1154 
1155 	tcf_set_qdisc_drop_reason(skb, reason);
1156 	skb->next = q->to_free;
1157 	q->to_free = skb;
1158 }
1159 
1160 /* Instead of calling kfree_skb() while root qdisc lock is held,
1161  * queue the skb for future freeing at end of __dev_xmit_skb()
1162  */
1163 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1164 {
1165 	skb->next = *to_free;
1166 	*to_free = skb;
1167 }
1168 
1169 static inline void __qdisc_drop_all(struct sk_buff *skb,
1170 				    struct sk_buff **to_free)
1171 {
1172 	if (skb->prev)
1173 		skb->prev->next = *to_free;
1174 	else
1175 		skb->next = *to_free;
1176 	*to_free = skb;
1177 }
1178 
1179 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1180 						   struct qdisc_skb_head *qh,
1181 						   struct sk_buff **to_free)
1182 {
1183 	struct sk_buff *skb = __qdisc_dequeue_head(qh);
1184 
1185 	if (likely(skb != NULL)) {
1186 		unsigned int len = qdisc_pkt_len(skb);
1187 
1188 		qdisc_qstats_backlog_dec(sch, skb);
1189 		__qdisc_drop(skb, to_free);
1190 		return len;
1191 	}
1192 
1193 	return 0;
1194 }
1195 
1196 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1197 {
1198 	const struct qdisc_skb_head *qh = &sch->q;
1199 
1200 	return qh->head;
1201 }
1202 
1203 /* generic pseudo peek method for non-work-conserving qdisc */
1204 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1205 {
1206 	struct sk_buff *skb = skb_peek(&sch->gso_skb);
1207 
1208 	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1209 	if (!skb) {
1210 		skb = sch->dequeue(sch);
1211 
1212 		if (skb) {
1213 			__skb_queue_head(&sch->gso_skb, skb);
1214 			/* it's still part of the queue */
1215 			qdisc_qstats_backlog_inc(sch, skb);
1216 			sch->q.qlen++;
1217 		}
1218 	}
1219 
1220 	return skb;
1221 }
1222 
1223 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1224 						 struct sk_buff *skb)
1225 {
1226 	if (qdisc_is_percpu_stats(sch)) {
1227 		qdisc_qstats_cpu_backlog_dec(sch, skb);
1228 		qdisc_bstats_cpu_update(sch, skb);
1229 		qdisc_qstats_cpu_qlen_dec(sch);
1230 	} else {
1231 		qdisc_qstats_backlog_dec(sch, skb);
1232 		qdisc_bstats_update(sch, skb);
1233 		sch->q.qlen--;
1234 	}
1235 }
1236 
1237 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1238 						 unsigned int pkt_len)
1239 {
1240 	if (qdisc_is_percpu_stats(sch)) {
1241 		qdisc_qstats_cpu_qlen_inc(sch);
1242 		this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1243 	} else {
1244 		sch->qstats.backlog += pkt_len;
1245 		sch->q.qlen++;
1246 	}
1247 }
1248 
1249 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1250 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1251 {
1252 	struct sk_buff *skb = skb_peek(&sch->gso_skb);
1253 
1254 	if (skb) {
1255 		skb = __skb_dequeue(&sch->gso_skb);
1256 		if (qdisc_is_percpu_stats(sch)) {
1257 			qdisc_qstats_cpu_backlog_dec(sch, skb);
1258 			qdisc_qstats_cpu_qlen_dec(sch);
1259 		} else {
1260 			qdisc_qstats_backlog_dec(sch, skb);
1261 			sch->q.qlen--;
1262 		}
1263 	} else {
1264 		skb = sch->dequeue(sch);
1265 	}
1266 
1267 	return skb;
1268 }
1269 
1270 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1271 {
1272 	/*
1273 	 * We do not know the backlog in bytes of this list, it
1274 	 * is up to the caller to correct it
1275 	 */
1276 	ASSERT_RTNL();
1277 	if (qh->qlen) {
1278 		rtnl_kfree_skbs(qh->head, qh->tail);
1279 
1280 		qh->head = NULL;
1281 		qh->tail = NULL;
1282 		qh->qlen = 0;
1283 	}
1284 }
1285 
1286 static inline void qdisc_reset_queue(struct Qdisc *sch)
1287 {
1288 	__qdisc_reset_queue(&sch->q);
1289 }
1290 
1291 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1292 					  struct Qdisc **pold)
1293 {
1294 	struct Qdisc *old;
1295 
1296 	sch_tree_lock(sch);
1297 	old = *pold;
1298 	*pold = new;
1299 	if (old != NULL)
1300 		qdisc_purge_queue(old);
1301 	sch_tree_unlock(sch);
1302 
1303 	return old;
1304 }
1305 
1306 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1307 {
1308 	rtnl_kfree_skbs(skb, skb);
1309 	qdisc_qstats_drop(sch);
1310 }
1311 
1312 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1313 				 struct sk_buff **to_free)
1314 {
1315 	__qdisc_drop(skb, to_free);
1316 	qdisc_qstats_cpu_drop(sch);
1317 
1318 	return NET_XMIT_DROP;
1319 }
1320 
1321 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1322 			     struct sk_buff **to_free)
1323 {
1324 	__qdisc_drop(skb, to_free);
1325 	qdisc_qstats_drop(sch);
1326 
1327 	return NET_XMIT_DROP;
1328 }
1329 
1330 static inline int qdisc_drop_reason(struct sk_buff *skb, struct Qdisc *sch,
1331 				    struct sk_buff **to_free,
1332 				    enum qdisc_drop_reason reason)
1333 {
1334 	tcf_set_qdisc_drop_reason(skb, reason);
1335 	return qdisc_drop(skb, sch, to_free);
1336 }
1337 
1338 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1339 				 struct sk_buff **to_free)
1340 {
1341 	__qdisc_drop_all(skb, to_free);
1342 	qdisc_qstats_drop(sch);
1343 
1344 	return NET_XMIT_DROP;
1345 }
1346 
1347 struct psched_ratecfg {
1348 	u64	rate_bytes_ps; /* bytes per second */
1349 	u32	mult;
1350 	u16	overhead;
1351 	u16	mpu;
1352 	u8	linklayer;
1353 	u8	shift;
1354 };
1355 
1356 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1357 				unsigned int len)
1358 {
1359 	len += r->overhead;
1360 
1361 	if (len < r->mpu)
1362 		len = r->mpu;
1363 
1364 	if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1365 		return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1366 
1367 	return ((u64)len * r->mult) >> r->shift;
1368 }
1369 
1370 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1371 			       const struct tc_ratespec *conf,
1372 			       u64 rate64);
1373 
1374 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1375 					  const struct psched_ratecfg *r)
1376 {
1377 	memset(res, 0, sizeof(*res));
1378 
1379 	/* legacy struct tc_ratespec has a 32bit @rate field
1380 	 * Qdisc using 64bit rate should add new attributes
1381 	 * in order to maintain compatibility.
1382 	 */
1383 	res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1384 
1385 	res->overhead = r->overhead;
1386 	res->mpu = r->mpu;
1387 	res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1388 }
1389 
1390 struct psched_pktrate {
1391 	u64	rate_pkts_ps; /* packets per second */
1392 	u32	mult;
1393 	u8	shift;
1394 };
1395 
1396 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1397 				  unsigned int pkt_num)
1398 {
1399 	return ((u64)pkt_num * r->mult) >> r->shift;
1400 }
1401 
1402 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1403 
1404 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1405  * The fast path only needs to access filter list and to update stats
1406  */
1407 struct mini_Qdisc {
1408 	struct tcf_proto *filter_list;
1409 	struct tcf_block *block;
1410 	struct gnet_stats_basic_sync __percpu *cpu_bstats;
1411 	struct gnet_stats_queue	__percpu *cpu_qstats;
1412 	unsigned long rcu_state;
1413 };
1414 
1415 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1416 						const struct sk_buff *skb)
1417 {
1418 	bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1419 }
1420 
1421 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1422 {
1423 	this_cpu_inc(miniq->cpu_qstats->drops);
1424 }
1425 
1426 struct mini_Qdisc_pair {
1427 	struct mini_Qdisc miniq1;
1428 	struct mini_Qdisc miniq2;
1429 	struct mini_Qdisc __rcu **p_miniq;
1430 };
1431 
1432 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1433 			  struct tcf_proto *tp_head);
1434 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1435 			  struct mini_Qdisc __rcu **p_miniq);
1436 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1437 				struct tcf_block *block);
1438 
1439 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1440 
1441 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1442 
1443 /* Make sure qdisc is no longer in SCHED state. */
1444 static inline void qdisc_synchronize(const struct Qdisc *q)
1445 {
1446 	while (test_bit(__QDISC_STATE_SCHED, &q->state))
1447 		msleep(1);
1448 }
1449 
1450 #endif
1451