xref: /linux/include/net/sch_generic.h (revision b693b51e0829b96a5c43f45c3fba3d11f6f09d2f)
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_reduce_backlog() 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_lockless(const struct Qdisc *q)
546 {
547 	return READ_ONCE(q->q.qlen);
548 }
549 
550 static inline void qdisc_qlen_inc(struct Qdisc *q)
551 {
552 	WRITE_ONCE(q->q.qlen, q->q.qlen + 1);
553 }
554 
555 static inline void qdisc_qlen_dec(struct Qdisc *q)
556 {
557 	WRITE_ONCE(q->q.qlen, q->q.qlen - 1);
558 }
559 
560 static inline int qdisc_qlen_sum(const struct Qdisc *q)
561 {
562 	__u32 qlen = q->qstats.qlen;
563 	int i;
564 
565 	if (qdisc_is_percpu_stats(q)) {
566 		for_each_possible_cpu(i)
567 			qlen += READ_ONCE(per_cpu_ptr(q->cpu_qstats, i)->qlen);
568 	} else {
569 		qlen += qdisc_qlen_lockless(q);
570 	}
571 
572 	return qlen;
573 }
574 
575 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
576 {
577 	return (struct qdisc_skb_cb *)skb->cb;
578 }
579 
580 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
581 {
582 	return &qdisc->q.lock;
583 }
584 
585 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
586 {
587 	struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
588 
589 	return q;
590 }
591 
592 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
593 {
594 	return rcu_dereference_bh(qdisc->dev_queue->qdisc);
595 }
596 
597 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
598 {
599 	return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
600 }
601 
602 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
603 {
604 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
605 
606 	ASSERT_RTNL();
607 	return qdisc_lock(root);
608 }
609 
610 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
611 {
612 	return qdisc->dev_queue->dev;
613 }
614 
615 static inline void sch_tree_lock(struct Qdisc *q)
616 {
617 	if (q->flags & TCQ_F_MQROOT)
618 		spin_lock_bh(qdisc_lock(q));
619 	else
620 		spin_lock_bh(qdisc_root_sleeping_lock(q));
621 }
622 
623 static inline void sch_tree_unlock(struct Qdisc *q)
624 {
625 	if (q->flags & TCQ_F_MQROOT)
626 		spin_unlock_bh(qdisc_lock(q));
627 	else
628 		spin_unlock_bh(qdisc_root_sleeping_lock(q));
629 }
630 
631 extern struct Qdisc noop_qdisc;
632 extern struct Qdisc_ops noop_qdisc_ops;
633 extern struct Qdisc_ops pfifo_fast_ops;
634 extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
635 extern struct Qdisc_ops mq_qdisc_ops;
636 extern struct Qdisc_ops noqueue_qdisc_ops;
637 extern const struct Qdisc_ops *default_qdisc_ops;
638 static inline const struct Qdisc_ops *
639 get_default_qdisc_ops(const struct net_device *dev, int ntx)
640 {
641 	return ntx < dev->real_num_tx_queues ?
642 			default_qdisc_ops : &pfifo_fast_ops;
643 }
644 
645 struct Qdisc_class_common {
646 	u32			classid;
647 	unsigned int		filter_cnt;
648 	struct hlist_node	hnode;
649 };
650 
651 struct Qdisc_class_hash {
652 	struct hlist_head	*hash;
653 	unsigned int		hashsize;
654 	unsigned int		hashmask;
655 	unsigned int		hashelems;
656 };
657 
658 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
659 {
660 	id ^= id >> 8;
661 	id ^= id >> 4;
662 	return id & mask;
663 }
664 
665 static inline struct Qdisc_class_common *
666 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
667 {
668 	struct Qdisc_class_common *cl;
669 	unsigned int h;
670 
671 	if (!id)
672 		return NULL;
673 
674 	h = qdisc_class_hash(id, hash->hashmask);
675 	hlist_for_each_entry(cl, &hash->hash[h], hnode) {
676 		if (cl->classid == id)
677 			return cl;
678 	}
679 	return NULL;
680 }
681 
682 static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
683 {
684 	return cl->filter_cnt > 0;
685 }
686 
687 static inline void qdisc_class_get(struct Qdisc_class_common *cl)
688 {
689 	unsigned int res;
690 
691 	if (check_add_overflow(cl->filter_cnt, 1, &res))
692 		WARN(1, "Qdisc class overflow");
693 
694 	cl->filter_cnt = res;
695 }
696 
697 static inline void qdisc_class_put(struct Qdisc_class_common *cl)
698 {
699 	unsigned int res;
700 
701 	if (check_sub_overflow(cl->filter_cnt, 1, &res))
702 		WARN(1, "Qdisc class underflow");
703 
704 	cl->filter_cnt = res;
705 }
706 
707 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
708 {
709 	u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
710 
711 	return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
712 }
713 
714 int qdisc_class_hash_init(struct Qdisc_class_hash *);
715 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
716 			     struct Qdisc_class_common *);
717 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
718 			     struct Qdisc_class_common *);
719 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
720 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
721 
722 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
723 void dev_qdisc_change_real_num_tx(struct net_device *dev,
724 				  unsigned int new_real_tx);
725 void dev_init_scheduler(struct net_device *dev);
726 void dev_shutdown(struct net_device *dev);
727 void dev_activate(struct net_device *dev);
728 void dev_deactivate(struct net_device *dev, bool reset_needed);
729 void dev_deactivate_many(struct list_head *head, bool reset_needed);
730 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
731 			      struct Qdisc *qdisc);
732 void qdisc_reset(struct Qdisc *qdisc);
733 void qdisc_destroy(struct Qdisc *qdisc);
734 void qdisc_put(struct Qdisc *qdisc);
735 void qdisc_put_unlocked(struct Qdisc *qdisc);
736 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
737 
738 static inline void dev_reset_queue(struct net_device *dev,
739 				   struct netdev_queue *dev_queue,
740 				   void *_unused)
741 {
742 	struct Qdisc *qdisc;
743 	bool nolock;
744 
745 	qdisc = rtnl_dereference(dev_queue->qdisc_sleeping);
746 	if (!qdisc)
747 		return;
748 
749 	nolock = qdisc->flags & TCQ_F_NOLOCK;
750 
751 	if (nolock)
752 		spin_lock_bh(&qdisc->seqlock);
753 	spin_lock_bh(qdisc_lock(qdisc));
754 
755 	qdisc_reset(qdisc);
756 
757 	spin_unlock_bh(qdisc_lock(qdisc));
758 	if (nolock) {
759 		clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
760 		clear_bit(__QDISC_STATE_DRAINING, &qdisc->state);
761 		spin_unlock_bh(&qdisc->seqlock);
762 	}
763 }
764 
765 #ifdef CONFIG_NET_SCHED
766 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
767 			      void *type_data);
768 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
769 				struct Qdisc *new, struct Qdisc *old,
770 				enum tc_setup_type type, void *type_data,
771 				struct netlink_ext_ack *extack);
772 #else
773 static inline int
774 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
775 			  void *type_data)
776 {
777 	q->flags &= ~TCQ_F_OFFLOADED;
778 	return 0;
779 }
780 
781 static inline void
782 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
783 			   struct Qdisc *new, struct Qdisc *old,
784 			   enum tc_setup_type type, void *type_data,
785 			   struct netlink_ext_ack *extack)
786 {
787 }
788 #endif
789 void qdisc_offload_query_caps(struct net_device *dev,
790 			      enum tc_setup_type type,
791 			      void *caps, size_t caps_len);
792 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
793 			  const struct Qdisc_ops *ops,
794 			  struct netlink_ext_ack *extack);
795 void qdisc_free(struct Qdisc *qdisc);
796 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
797 				const struct Qdisc_ops *ops, u32 parentid,
798 				struct netlink_ext_ack *extack);
799 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
800 			       const struct qdisc_size_table *stab);
801 int skb_do_redirect(struct sk_buff *);
802 
803 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
804 {
805 #ifdef CONFIG_NET_XGRESS
806 	return skb->tc_at_ingress;
807 #else
808 	return false;
809 #endif
810 }
811 
812 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
813 {
814 #ifdef CONFIG_NET_CLS_ACT
815 	if (skb->tc_skip_classify) {
816 		skb->tc_skip_classify = 0;
817 		return true;
818 	}
819 #endif
820 	return false;
821 }
822 
823 /* Reset all TX qdiscs greater than index of a device.  */
824 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
825 {
826 	struct Qdisc *qdisc;
827 	bool nolock;
828 
829 	for (; i < dev->num_tx_queues; i++) {
830 		qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
831 		if (qdisc) {
832 			nolock = qdisc->flags & TCQ_F_NOLOCK;
833 
834 			if (nolock)
835 				spin_lock_bh(&qdisc->seqlock);
836 			spin_lock_bh(qdisc_lock(qdisc));
837 			qdisc_reset(qdisc);
838 			spin_unlock_bh(qdisc_lock(qdisc));
839 			if (nolock) {
840 				clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
841 				clear_bit(__QDISC_STATE_DRAINING, &qdisc->state);
842 				spin_unlock_bh(&qdisc->seqlock);
843 			}
844 		}
845 	}
846 }
847 
848 /* Are all TX queues of the device empty?  */
849 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
850 {
851 	unsigned int i;
852 
853 	rcu_read_lock();
854 	for (i = 0; i < dev->num_tx_queues; i++) {
855 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
856 		const struct Qdisc *q = rcu_dereference(txq->qdisc);
857 
858 		if (!qdisc_is_empty(q)) {
859 			rcu_read_unlock();
860 			return false;
861 		}
862 	}
863 	rcu_read_unlock();
864 	return true;
865 }
866 
867 /* Are any of the TX qdiscs changing?  */
868 static inline bool qdisc_tx_changing(const struct net_device *dev)
869 {
870 	unsigned int i;
871 
872 	for (i = 0; i < dev->num_tx_queues; i++) {
873 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
874 
875 		if (rcu_access_pointer(txq->qdisc) !=
876 		    rcu_access_pointer(txq->qdisc_sleeping))
877 			return true;
878 	}
879 	return false;
880 }
881 
882 /* "noqueue" qdisc identified by not having any enqueue, see noqueue_init() */
883 static inline bool qdisc_txq_has_no_queue(const struct netdev_queue *txq)
884 {
885 	struct Qdisc *qdisc = rcu_access_pointer(txq->qdisc);
886 
887 	return qdisc->enqueue == NULL;
888 }
889 
890 /* Is the device using the noop qdisc on all queues?  */
891 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
892 {
893 	unsigned int i;
894 
895 	for (i = 0; i < dev->num_tx_queues; i++) {
896 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
897 		if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
898 			return false;
899 	}
900 	return true;
901 }
902 
903 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
904 {
905 	return qdisc_skb_cb(skb)->pkt_len;
906 }
907 
908 static inline unsigned int qdisc_pkt_segs(const struct sk_buff *skb)
909 {
910 	u32 pkt_segs = qdisc_skb_cb(skb)->pkt_segs;
911 
912 	DEBUG_NET_WARN_ON_ONCE(pkt_segs !=
913 			(skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1));
914 	return pkt_segs;
915 }
916 
917 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
918 enum net_xmit_qdisc_t {
919 	__NET_XMIT_STOLEN = 0x00010000,
920 	__NET_XMIT_BYPASS = 0x00020000,
921 };
922 
923 #ifdef CONFIG_NET_CLS_ACT
924 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
925 #else
926 #define net_xmit_drop_count(e)	(1)
927 #endif
928 
929 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
930 					   const struct Qdisc *sch)
931 {
932 #ifdef CONFIG_NET_SCHED
933 	struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
934 
935 	if (stab)
936 		__qdisc_calculate_pkt_len(skb, stab);
937 #endif
938 }
939 
940 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
941 				struct sk_buff **to_free)
942 {
943 	return sch->enqueue(skb, sch, to_free);
944 }
945 
946 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
947 				  __u64 bytes, __u64 packets)
948 {
949 	u64_stats_update_begin(&bstats->syncp);
950 	u64_stats_add(&bstats->bytes, bytes);
951 	u64_stats_add(&bstats->packets, packets);
952 	u64_stats_update_end(&bstats->syncp);
953 }
954 
955 static inline void _bstats_set(struct gnet_stats_basic_sync *bstats,
956 			       u64 bytes, u64 packets)
957 {
958 	u64_stats_update_begin(&bstats->syncp);
959 	u64_stats_set(&bstats->bytes, bytes);
960 	u64_stats_set(&bstats->packets, packets);
961 	u64_stats_update_end(&bstats->syncp);
962 }
963 
964 static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
965 				 const struct sk_buff *skb)
966 {
967 	_bstats_update(bstats, qdisc_pkt_len(skb), qdisc_pkt_segs(skb));
968 }
969 
970 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
971 					   const struct sk_buff *skb)
972 {
973 	bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
974 }
975 
976 static inline void qdisc_bstats_update(struct Qdisc *sch,
977 				       const struct sk_buff *skb)
978 {
979 	bstats_update(&sch->bstats, skb);
980 }
981 
982 static inline void qstats_backlog_sub(struct Qdisc *sch, u32 val)
983 {
984 	WRITE_ONCE(sch->qstats.backlog, sch->qstats.backlog - val);
985 }
986 
987 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
988 					    const struct sk_buff *skb)
989 {
990 	qstats_backlog_sub(sch, qdisc_pkt_len(skb));
991 }
992 
993 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
994 						const struct sk_buff *skb)
995 {
996 	this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
997 }
998 
999 static inline void qstats_backlog_add(struct Qdisc *sch, u32 val)
1000 {
1001 	WRITE_ONCE(sch->qstats.backlog, sch->qstats.backlog + val);
1002 }
1003 
1004 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
1005 					    const struct sk_buff *skb)
1006 {
1007 	qstats_backlog_add(sch, qdisc_pkt_len(skb));
1008 }
1009 
1010 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
1011 						const struct sk_buff *skb)
1012 {
1013 	this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
1014 }
1015 
1016 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
1017 {
1018 	this_cpu_inc(sch->cpu_qstats->qlen);
1019 }
1020 
1021 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
1022 {
1023 	this_cpu_dec(sch->cpu_qstats->qlen);
1024 }
1025 
1026 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
1027 {
1028 	this_cpu_inc(sch->cpu_qstats->requeues);
1029 }
1030 
1031 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
1032 {
1033 	WRITE_ONCE(sch->qstats.drops, sch->qstats.drops + count);
1034 }
1035 
1036 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
1037 {
1038 	WRITE_ONCE(qstats->drops, qstats->drops + 1);
1039 }
1040 
1041 static inline void qstats_cpu_drop_inc(struct gnet_stats_queue __percpu *qstats)
1042 {
1043 	this_cpu_inc(qstats->drops);
1044 }
1045 
1046 static inline void qstats_cpu_overlimit_inc(struct gnet_stats_queue __percpu *qstats)
1047 {
1048 	this_cpu_inc(qstats->overlimits);
1049 }
1050 
1051 static inline void qdisc_qstats_drop(struct Qdisc *sch)
1052 {
1053 	qstats_drop_inc(&sch->qstats);
1054 }
1055 
1056 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
1057 {
1058 	this_cpu_inc(sch->cpu_qstats->drops);
1059 }
1060 
1061 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
1062 {
1063 	WRITE_ONCE(sch->qstats.overlimits, sch->qstats.overlimits + 1);
1064 }
1065 
1066 static inline int qdisc_qstats_copy(struct gnet_dump *d, const struct Qdisc *sch)
1067 {
1068 	__u32 qlen = qdisc_qlen_sum(sch);
1069 
1070 	return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
1071 }
1072 
1073 static inline void qdisc_qstats_qlen_backlog(const struct Qdisc *sch,
1074 					     u32 *qlen, u32 *backlog)
1075 {
1076 	struct gnet_stats_queue qstats = { 0 };
1077 
1078 	gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
1079 	*qlen = qstats.qlen + qdisc_qlen_lockless(sch);
1080 	*backlog = qstats.backlog;
1081 }
1082 
1083 static inline void qdisc_purge_queue(struct Qdisc *sch)
1084 {
1085 	__u32 qlen, backlog;
1086 
1087 	qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
1088 	qdisc_reset(sch);
1089 	qdisc_tree_reduce_backlog(sch, qlen, backlog);
1090 }
1091 
1092 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
1093 					struct qdisc_skb_head *qh)
1094 {
1095 	struct sk_buff *last = qh->tail;
1096 
1097 	if (last) {
1098 		skb->next = NULL;
1099 		last->next = skb;
1100 		qh->tail = skb;
1101 	} else {
1102 		qh->tail = skb;
1103 		qh->head = skb;
1104 	}
1105 	WRITE_ONCE(qh->qlen, qh->qlen + 1);
1106 }
1107 
1108 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1109 {
1110 	__qdisc_enqueue_tail(skb, &sch->q);
1111 	qdisc_qstats_backlog_inc(sch, skb);
1112 	return NET_XMIT_SUCCESS;
1113 }
1114 
1115 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1116 					struct qdisc_skb_head *qh)
1117 {
1118 	skb->next = qh->head;
1119 
1120 	if (!qh->head)
1121 		qh->tail = skb;
1122 	qh->head = skb;
1123 	WRITE_ONCE(qh->qlen, qh->qlen + 1);
1124 }
1125 
1126 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1127 {
1128 	struct sk_buff *skb = qh->head;
1129 
1130 	if (likely(skb != NULL)) {
1131 		qh->head = skb->next;
1132 		WRITE_ONCE(qh->qlen, qh->qlen - 1);
1133 		if (qh->head == NULL)
1134 			qh->tail = NULL;
1135 		skb->next = NULL;
1136 	}
1137 
1138 	return skb;
1139 }
1140 
1141 static inline struct sk_buff *qdisc_dequeue_internal(struct Qdisc *sch, bool direct)
1142 {
1143 	struct sk_buff *skb;
1144 
1145 	skb = __skb_dequeue(&sch->gso_skb);
1146 	if (skb) {
1147 		qdisc_qlen_dec(sch);
1148 		qdisc_qstats_backlog_dec(sch, skb);
1149 		return skb;
1150 	}
1151 	if (direct) {
1152 		skb = __qdisc_dequeue_head(&sch->q);
1153 		if (skb)
1154 			qdisc_qstats_backlog_dec(sch, skb);
1155 		return skb;
1156 	} else {
1157 		return sch->dequeue(sch);
1158 	}
1159 }
1160 
1161 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1162 {
1163 	struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1164 
1165 	if (likely(skb != NULL)) {
1166 		qdisc_qstats_backlog_dec(sch, skb);
1167 		qdisc_bstats_update(sch, skb);
1168 	}
1169 
1170 	return skb;
1171 }
1172 
1173 struct tc_skb_cb {
1174 	struct qdisc_skb_cb qdisc_cb;
1175 	u32 drop_reason;
1176 
1177 	u16 zone; /* Only valid if qdisc_skb_cb(skb)->post_ct = true */
1178 	u16 mru;
1179 };
1180 
1181 static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
1182 {
1183 	struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
1184 
1185 	BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
1186 	return cb;
1187 }
1188 
1189 /* TC classifier accessors - use enum skb_drop_reason */
1190 static inline enum skb_drop_reason
1191 tcf_get_drop_reason(const struct sk_buff *skb)
1192 {
1193 	return (enum skb_drop_reason)tc_skb_cb(skb)->drop_reason;
1194 }
1195 
1196 static inline void tcf_set_drop_reason(const struct sk_buff *skb,
1197 				       enum skb_drop_reason reason)
1198 {
1199 	tc_skb_cb(skb)->drop_reason = (enum qdisc_drop_reason)reason;
1200 }
1201 
1202 /* Qdisc accessors - use enum qdisc_drop_reason */
1203 static inline enum qdisc_drop_reason
1204 tcf_get_qdisc_drop_reason(const struct sk_buff *skb)
1205 {
1206 	return tc_skb_cb(skb)->drop_reason;
1207 }
1208 
1209 static inline void tcf_set_qdisc_drop_reason(const struct sk_buff *skb,
1210 					     enum qdisc_drop_reason reason)
1211 {
1212 	tc_skb_cb(skb)->drop_reason = reason;
1213 }
1214 
1215 void __tcf_kfree_skb_list(struct sk_buff *skb, struct Qdisc *q,
1216 			  struct netdev_queue *txq, struct net_device *dev);
1217 
1218 static inline void tcf_kfree_skb_list(struct sk_buff *skb, struct Qdisc *q,
1219 				      struct netdev_queue *txq,
1220 				      struct net_device *dev)
1221 {
1222 	if (unlikely(skb))
1223 		__tcf_kfree_skb_list(skb, q, txq, dev);
1224 }
1225 
1226 static inline void qdisc_dequeue_drop(struct Qdisc *q, struct sk_buff *skb,
1227 				      enum qdisc_drop_reason reason)
1228 {
1229 	struct Qdisc *root;
1230 
1231 	DEBUG_NET_WARN_ON_ONCE(!(q->flags & TCQ_F_DEQUEUE_DROPS));
1232 	DEBUG_NET_WARN_ON_ONCE(q->flags & TCQ_F_NOLOCK);
1233 
1234 	rcu_read_lock();
1235 	root = qdisc_root_sleeping(q);
1236 
1237 	if (root->flags & TCQ_F_DEQUEUE_DROPS) {
1238 		tcf_set_qdisc_drop_reason(skb, reason);
1239 		skb->next = root->to_free;
1240 		root->to_free = skb;
1241 	} else {
1242 		kfree_skb_reason(skb, (enum skb_drop_reason)reason);
1243 	}
1244 	rcu_read_unlock();
1245 }
1246 
1247 /* Instead of calling kfree_skb() while root qdisc lock is held,
1248  * queue the skb for future freeing at end of __dev_xmit_skb()
1249  */
1250 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1251 {
1252 	skb->next = *to_free;
1253 	*to_free = skb;
1254 }
1255 
1256 static inline void __qdisc_drop_all(struct sk_buff *skb,
1257 				    struct sk_buff **to_free)
1258 {
1259 	if (skb->prev)
1260 		skb->prev->next = *to_free;
1261 	else
1262 		skb->next = *to_free;
1263 	*to_free = skb;
1264 }
1265 
1266 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1267 						   struct qdisc_skb_head *qh,
1268 						   struct sk_buff **to_free)
1269 {
1270 	struct sk_buff *skb = __qdisc_dequeue_head(qh);
1271 
1272 	if (likely(skb != NULL)) {
1273 		unsigned int len = qdisc_pkt_len(skb);
1274 
1275 		qdisc_qstats_backlog_dec(sch, skb);
1276 		__qdisc_drop(skb, to_free);
1277 		return len;
1278 	}
1279 
1280 	return 0;
1281 }
1282 
1283 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1284 {
1285 	const struct qdisc_skb_head *qh = &sch->q;
1286 
1287 	return qh->head;
1288 }
1289 
1290 /* generic pseudo peek method for non-work-conserving qdisc */
1291 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1292 {
1293 	struct sk_buff *skb = skb_peek(&sch->gso_skb);
1294 
1295 	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1296 	if (!skb) {
1297 		skb = sch->dequeue(sch);
1298 
1299 		if (skb) {
1300 			__skb_queue_head(&sch->gso_skb, skb);
1301 			/* it's still part of the queue */
1302 			qdisc_qstats_backlog_inc(sch, skb);
1303 			qdisc_qlen_inc(sch);
1304 		}
1305 	}
1306 
1307 	return skb;
1308 }
1309 
1310 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1311 						 struct sk_buff *skb)
1312 {
1313 	if (qdisc_is_percpu_stats(sch)) {
1314 		qdisc_qstats_cpu_backlog_dec(sch, skb);
1315 		qdisc_bstats_cpu_update(sch, skb);
1316 		qdisc_qstats_cpu_qlen_dec(sch);
1317 	} else {
1318 		qdisc_qstats_backlog_dec(sch, skb);
1319 		qdisc_bstats_update(sch, skb);
1320 		qdisc_qlen_dec(sch);
1321 	}
1322 }
1323 
1324 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1325 						 unsigned int pkt_len)
1326 {
1327 	if (qdisc_is_percpu_stats(sch)) {
1328 		qdisc_qstats_cpu_qlen_inc(sch);
1329 		this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1330 	} else {
1331 		qstats_backlog_add(sch, pkt_len);
1332 		qdisc_qlen_inc(sch);
1333 	}
1334 }
1335 
1336 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1337 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1338 {
1339 	struct sk_buff *skb = skb_peek(&sch->gso_skb);
1340 
1341 	if (skb) {
1342 		skb = __skb_dequeue(&sch->gso_skb);
1343 		if (qdisc_is_percpu_stats(sch)) {
1344 			qdisc_qstats_cpu_backlog_dec(sch, skb);
1345 			qdisc_qstats_cpu_qlen_dec(sch);
1346 		} else {
1347 			qdisc_qstats_backlog_dec(sch, skb);
1348 			qdisc_qlen_dec(sch);
1349 		}
1350 	} else {
1351 		skb = sch->dequeue(sch);
1352 	}
1353 
1354 	return skb;
1355 }
1356 
1357 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1358 {
1359 	/*
1360 	 * We do not know the backlog in bytes of this list, it
1361 	 * is up to the caller to correct it
1362 	 */
1363 	ASSERT_RTNL();
1364 	if (qh->qlen) {
1365 		rtnl_kfree_skbs(qh->head, qh->tail);
1366 
1367 		qh->head = NULL;
1368 		qh->tail = NULL;
1369 		WRITE_ONCE(qh->qlen, 0);
1370 	}
1371 }
1372 
1373 static inline void qdisc_reset_queue(struct Qdisc *sch)
1374 {
1375 	__qdisc_reset_queue(&sch->q);
1376 }
1377 
1378 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1379 					  struct Qdisc **pold)
1380 {
1381 	struct Qdisc *old;
1382 
1383 	sch_tree_lock(sch);
1384 	old = *pold;
1385 	*pold = new;
1386 	if (old != NULL)
1387 		qdisc_purge_queue(old);
1388 	sch_tree_unlock(sch);
1389 
1390 	return old;
1391 }
1392 
1393 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1394 {
1395 	rtnl_kfree_skbs(skb, skb);
1396 	qdisc_qstats_drop(sch);
1397 }
1398 
1399 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1400 				 struct sk_buff **to_free)
1401 {
1402 	__qdisc_drop(skb, to_free);
1403 	qdisc_qstats_cpu_drop(sch);
1404 
1405 	return NET_XMIT_DROP;
1406 }
1407 
1408 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1409 			     struct sk_buff **to_free)
1410 {
1411 	__qdisc_drop(skb, to_free);
1412 	qdisc_qstats_drop(sch);
1413 
1414 	return NET_XMIT_DROP;
1415 }
1416 
1417 static inline int qdisc_drop_reason(struct sk_buff *skb, struct Qdisc *sch,
1418 				    struct sk_buff **to_free,
1419 				    enum qdisc_drop_reason reason)
1420 {
1421 	tcf_set_qdisc_drop_reason(skb, reason);
1422 	return qdisc_drop(skb, sch, to_free);
1423 }
1424 
1425 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1426 				 struct sk_buff **to_free)
1427 {
1428 	__qdisc_drop_all(skb, to_free);
1429 	qdisc_qstats_drop(sch);
1430 
1431 	return NET_XMIT_DROP;
1432 }
1433 
1434 struct psched_ratecfg {
1435 	u64	rate_bytes_ps; /* bytes per second */
1436 	u32	mult;
1437 	u16	overhead;
1438 	u16	mpu;
1439 	u8	linklayer;
1440 	u8	shift;
1441 };
1442 
1443 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1444 				unsigned int len)
1445 {
1446 	len += r->overhead;
1447 
1448 	if (len < r->mpu)
1449 		len = r->mpu;
1450 
1451 	if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1452 		return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1453 
1454 	return ((u64)len * r->mult) >> r->shift;
1455 }
1456 
1457 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1458 			       const struct tc_ratespec *conf,
1459 			       u64 rate64);
1460 
1461 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1462 					  const struct psched_ratecfg *r)
1463 {
1464 	memset(res, 0, sizeof(*res));
1465 
1466 	/* legacy struct tc_ratespec has a 32bit @rate field
1467 	 * Qdisc using 64bit rate should add new attributes
1468 	 * in order to maintain compatibility.
1469 	 */
1470 	res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1471 
1472 	res->overhead = r->overhead;
1473 	res->mpu = r->mpu;
1474 	res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1475 }
1476 
1477 struct psched_pktrate {
1478 	u64	rate_pkts_ps; /* packets per second */
1479 	u32	mult;
1480 	u8	shift;
1481 };
1482 
1483 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1484 				  unsigned int pkt_num)
1485 {
1486 	return ((u64)pkt_num * r->mult) >> r->shift;
1487 }
1488 
1489 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1490 
1491 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1492  * The fast path only needs to access filter list and to update stats
1493  */
1494 struct mini_Qdisc {
1495 	struct tcf_proto *filter_list;
1496 	struct tcf_block *block;
1497 	struct gnet_stats_basic_sync __percpu *cpu_bstats;
1498 	struct gnet_stats_queue	__percpu *cpu_qstats;
1499 	unsigned long rcu_state;
1500 };
1501 
1502 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1503 						const struct sk_buff *skb)
1504 {
1505 	bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1506 }
1507 
1508 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1509 {
1510 	this_cpu_inc(miniq->cpu_qstats->drops);
1511 }
1512 
1513 struct mini_Qdisc_pair {
1514 	struct mini_Qdisc miniq1;
1515 	struct mini_Qdisc miniq2;
1516 	struct mini_Qdisc __rcu **p_miniq;
1517 };
1518 
1519 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1520 			  struct tcf_proto *tp_head);
1521 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1522 			  struct mini_Qdisc __rcu **p_miniq);
1523 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1524 				struct tcf_block *block);
1525 
1526 static inline bool mini_qdisc_pair_inited(struct mini_Qdisc_pair *miniqp)
1527 {
1528 	return !!miniqp->p_miniq;
1529 }
1530 
1531 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1532 
1533 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1534 
1535 /* Make sure qdisc is no longer in SCHED state. */
1536 static inline void qdisc_synchronize(const struct Qdisc *q)
1537 {
1538 	while (test_bit(__QDISC_STATE_SCHED, &q->state))
1539 		msleep(1);
1540 }
1541 
1542 #endif
1543