xref: /openbmc/linux/include/net/sch_generic.h (revision c67e8ec0)
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 struct bpf_flow_keys;
23 
24 typedef int tc_setup_cb_t(enum tc_setup_type type,
25 			  void *type_data, void *cb_priv);
26 
27 typedef int tc_indr_block_bind_cb_t(struct net_device *dev, void *cb_priv,
28 				    enum tc_setup_type type, void *type_data);
29 
30 struct qdisc_rate_table {
31 	struct tc_ratespec rate;
32 	u32		data[256];
33 	struct qdisc_rate_table *next;
34 	int		refcnt;
35 };
36 
37 enum qdisc_state_t {
38 	__QDISC_STATE_SCHED,
39 	__QDISC_STATE_DEACTIVATED,
40 };
41 
42 struct qdisc_size_table {
43 	struct rcu_head		rcu;
44 	struct list_head	list;
45 	struct tc_sizespec	szopts;
46 	int			refcnt;
47 	u16			data[];
48 };
49 
50 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
51 struct qdisc_skb_head {
52 	struct sk_buff	*head;
53 	struct sk_buff	*tail;
54 	__u32		qlen;
55 	spinlock_t	lock;
56 };
57 
58 struct Qdisc {
59 	int 			(*enqueue)(struct sk_buff *skb,
60 					   struct Qdisc *sch,
61 					   struct sk_buff **to_free);
62 	struct sk_buff *	(*dequeue)(struct Qdisc *sch);
63 	unsigned int		flags;
64 #define TCQ_F_BUILTIN		1
65 #define TCQ_F_INGRESS		2
66 #define TCQ_F_CAN_BYPASS	4
67 #define TCQ_F_MQROOT		8
68 #define TCQ_F_ONETXQUEUE	0x10 /* dequeue_skb() can assume all skbs are for
69 				      * q->dev_queue : It can test
70 				      * netif_xmit_frozen_or_stopped() before
71 				      * dequeueing next packet.
72 				      * Its true for MQ/MQPRIO slaves, or non
73 				      * multiqueue device.
74 				      */
75 #define TCQ_F_WARN_NONWC	(1 << 16)
76 #define TCQ_F_CPUSTATS		0x20 /* run using percpu statistics */
77 #define TCQ_F_NOPARENT		0x40 /* root of its hierarchy :
78 				      * qdisc_tree_decrease_qlen() should stop.
79 				      */
80 #define TCQ_F_INVISIBLE		0x80 /* invisible by default in dump */
81 #define TCQ_F_NOLOCK		0x100 /* qdisc does not require locking */
82 #define TCQ_F_OFFLOADED		0x200 /* qdisc is offloaded to HW */
83 	u32			limit;
84 	const struct Qdisc_ops	*ops;
85 	struct qdisc_size_table	__rcu *stab;
86 	struct hlist_node       hash;
87 	u32			handle;
88 	u32			parent;
89 
90 	struct netdev_queue	*dev_queue;
91 
92 	struct net_rate_estimator __rcu *rate_est;
93 	struct gnet_stats_basic_cpu __percpu *cpu_bstats;
94 	struct gnet_stats_queue	__percpu *cpu_qstats;
95 	int			padded;
96 	refcount_t		refcnt;
97 
98 	/*
99 	 * For performance sake on SMP, we put highly modified fields at the end
100 	 */
101 	struct sk_buff_head	gso_skb ____cacheline_aligned_in_smp;
102 	struct qdisc_skb_head	q;
103 	struct gnet_stats_basic_packed bstats;
104 	seqcount_t		running;
105 	struct gnet_stats_queue	qstats;
106 	unsigned long		state;
107 	struct Qdisc            *next_sched;
108 	struct sk_buff_head	skb_bad_txq;
109 
110 	spinlock_t		busylock ____cacheline_aligned_in_smp;
111 	spinlock_t		seqlock;
112 	struct rcu_head		rcu;
113 };
114 
115 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
116 {
117 	if (qdisc->flags & TCQ_F_BUILTIN)
118 		return;
119 	refcount_inc(&qdisc->refcnt);
120 }
121 
122 /* Intended to be used by unlocked users, when concurrent qdisc release is
123  * possible.
124  */
125 
126 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
127 {
128 	if (qdisc->flags & TCQ_F_BUILTIN)
129 		return qdisc;
130 	if (refcount_inc_not_zero(&qdisc->refcnt))
131 		return qdisc;
132 	return NULL;
133 }
134 
135 static inline bool qdisc_is_running(struct Qdisc *qdisc)
136 {
137 	if (qdisc->flags & TCQ_F_NOLOCK)
138 		return spin_is_locked(&qdisc->seqlock);
139 	return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
140 }
141 
142 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
143 {
144 	if (qdisc->flags & TCQ_F_NOLOCK) {
145 		if (!spin_trylock(&qdisc->seqlock))
146 			return false;
147 	} else if (qdisc_is_running(qdisc)) {
148 		return false;
149 	}
150 	/* Variant of write_seqcount_begin() telling lockdep a trylock
151 	 * was attempted.
152 	 */
153 	raw_write_seqcount_begin(&qdisc->running);
154 	seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
155 	return true;
156 }
157 
158 static inline void qdisc_run_end(struct Qdisc *qdisc)
159 {
160 	write_seqcount_end(&qdisc->running);
161 	if (qdisc->flags & TCQ_F_NOLOCK)
162 		spin_unlock(&qdisc->seqlock);
163 }
164 
165 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
166 {
167 	return qdisc->flags & TCQ_F_ONETXQUEUE;
168 }
169 
170 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
171 {
172 #ifdef CONFIG_BQL
173 	/* Non-BQL migrated drivers will return 0, too. */
174 	return dql_avail(&txq->dql);
175 #else
176 	return 0;
177 #endif
178 }
179 
180 struct Qdisc_class_ops {
181 	/* Child qdisc manipulation */
182 	struct netdev_queue *	(*select_queue)(struct Qdisc *, struct tcmsg *);
183 	int			(*graft)(struct Qdisc *, unsigned long cl,
184 					struct Qdisc *, struct Qdisc **,
185 					struct netlink_ext_ack *extack);
186 	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
187 	void			(*qlen_notify)(struct Qdisc *, unsigned long);
188 
189 	/* Class manipulation routines */
190 	unsigned long		(*find)(struct Qdisc *, u32 classid);
191 	int			(*change)(struct Qdisc *, u32, u32,
192 					struct nlattr **, unsigned long *,
193 					struct netlink_ext_ack *);
194 	int			(*delete)(struct Qdisc *, unsigned long);
195 	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
196 
197 	/* Filter manipulation */
198 	struct tcf_block *	(*tcf_block)(struct Qdisc *sch,
199 					     unsigned long arg,
200 					     struct netlink_ext_ack *extack);
201 	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
202 					u32 classid);
203 	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
204 
205 	/* rtnetlink specific */
206 	int			(*dump)(struct Qdisc *, unsigned long,
207 					struct sk_buff *skb, struct tcmsg*);
208 	int			(*dump_stats)(struct Qdisc *, unsigned long,
209 					struct gnet_dump *);
210 };
211 
212 struct Qdisc_ops {
213 	struct Qdisc_ops	*next;
214 	const struct Qdisc_class_ops	*cl_ops;
215 	char			id[IFNAMSIZ];
216 	int			priv_size;
217 	unsigned int		static_flags;
218 
219 	int 			(*enqueue)(struct sk_buff *skb,
220 					   struct Qdisc *sch,
221 					   struct sk_buff **to_free);
222 	struct sk_buff *	(*dequeue)(struct Qdisc *);
223 	struct sk_buff *	(*peek)(struct Qdisc *);
224 
225 	int			(*init)(struct Qdisc *sch, struct nlattr *arg,
226 					struct netlink_ext_ack *extack);
227 	void			(*reset)(struct Qdisc *);
228 	void			(*destroy)(struct Qdisc *);
229 	int			(*change)(struct Qdisc *sch,
230 					  struct nlattr *arg,
231 					  struct netlink_ext_ack *extack);
232 	void			(*attach)(struct Qdisc *sch);
233 	int			(*change_tx_queue_len)(struct Qdisc *, unsigned int);
234 
235 	int			(*dump)(struct Qdisc *, struct sk_buff *);
236 	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
237 
238 	void			(*ingress_block_set)(struct Qdisc *sch,
239 						     u32 block_index);
240 	void			(*egress_block_set)(struct Qdisc *sch,
241 						    u32 block_index);
242 	u32			(*ingress_block_get)(struct Qdisc *sch);
243 	u32			(*egress_block_get)(struct Qdisc *sch);
244 
245 	struct module		*owner;
246 };
247 
248 
249 struct tcf_result {
250 	union {
251 		struct {
252 			unsigned long	class;
253 			u32		classid;
254 		};
255 		const struct tcf_proto *goto_tp;
256 
257 		/* used by the TC_ACT_REINSERT action */
258 		struct {
259 			bool		ingress;
260 			struct gnet_stats_queue *qstats;
261 		};
262 	};
263 };
264 
265 struct tcf_chain;
266 
267 struct tcf_proto_ops {
268 	struct list_head	head;
269 	char			kind[IFNAMSIZ];
270 
271 	int			(*classify)(struct sk_buff *,
272 					    const struct tcf_proto *,
273 					    struct tcf_result *);
274 	int			(*init)(struct tcf_proto*);
275 	void			(*destroy)(struct tcf_proto *tp,
276 					   struct netlink_ext_ack *extack);
277 
278 	void*			(*get)(struct tcf_proto*, u32 handle);
279 	int			(*change)(struct net *net, struct sk_buff *,
280 					struct tcf_proto*, unsigned long,
281 					u32 handle, struct nlattr **,
282 					void **, bool,
283 					struct netlink_ext_ack *);
284 	int			(*delete)(struct tcf_proto *tp, void *arg,
285 					  bool *last,
286 					  struct netlink_ext_ack *);
287 	void			(*walk)(struct tcf_proto*, struct tcf_walker *arg);
288 	int			(*reoffload)(struct tcf_proto *tp, bool add,
289 					     tc_setup_cb_t *cb, void *cb_priv,
290 					     struct netlink_ext_ack *extack);
291 	void			(*bind_class)(void *, u32, unsigned long);
292 	void *			(*tmplt_create)(struct net *net,
293 						struct tcf_chain *chain,
294 						struct nlattr **tca,
295 						struct netlink_ext_ack *extack);
296 	void			(*tmplt_destroy)(void *tmplt_priv);
297 
298 	/* rtnetlink specific */
299 	int			(*dump)(struct net*, struct tcf_proto*, void *,
300 					struct sk_buff *skb, struct tcmsg*);
301 	int			(*tmplt_dump)(struct sk_buff *skb,
302 					      struct net *net,
303 					      void *tmplt_priv);
304 
305 	struct module		*owner;
306 };
307 
308 struct tcf_proto {
309 	/* Fast access part */
310 	struct tcf_proto __rcu	*next;
311 	void __rcu		*root;
312 
313 	/* called under RCU BH lock*/
314 	int			(*classify)(struct sk_buff *,
315 					    const struct tcf_proto *,
316 					    struct tcf_result *);
317 	__be16			protocol;
318 
319 	/* All the rest */
320 	u32			prio;
321 	void			*data;
322 	const struct tcf_proto_ops	*ops;
323 	struct tcf_chain	*chain;
324 	struct rcu_head		rcu;
325 };
326 
327 struct qdisc_skb_cb {
328 	union {
329 		struct {
330 			unsigned int		pkt_len;
331 			u16			slave_dev_queue_mapping;
332 			u16			tc_classid;
333 		};
334 		struct bpf_flow_keys *flow_keys;
335 	};
336 #define QDISC_CB_PRIV_LEN 20
337 	unsigned char		data[QDISC_CB_PRIV_LEN];
338 };
339 
340 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
341 
342 struct tcf_chain {
343 	struct tcf_proto __rcu *filter_chain;
344 	struct list_head list;
345 	struct tcf_block *block;
346 	u32 index; /* chain index */
347 	unsigned int refcnt;
348 	unsigned int action_refcnt;
349 	bool explicitly_created;
350 	const struct tcf_proto_ops *tmplt_ops;
351 	void *tmplt_priv;
352 };
353 
354 struct tcf_block {
355 	struct list_head chain_list;
356 	u32 index; /* block index for shared blocks */
357 	refcount_t refcnt;
358 	struct net *net;
359 	struct Qdisc *q;
360 	struct list_head cb_list;
361 	struct list_head owner_list;
362 	bool keep_dst;
363 	unsigned int offloadcnt; /* Number of oddloaded filters */
364 	unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
365 	struct {
366 		struct tcf_chain *chain;
367 		struct list_head filter_chain_list;
368 	} chain0;
369 	struct rcu_head rcu;
370 };
371 
372 static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
373 {
374 	if (*flags & TCA_CLS_FLAGS_IN_HW)
375 		return;
376 	*flags |= TCA_CLS_FLAGS_IN_HW;
377 	block->offloadcnt++;
378 }
379 
380 static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
381 {
382 	if (!(*flags & TCA_CLS_FLAGS_IN_HW))
383 		return;
384 	*flags &= ~TCA_CLS_FLAGS_IN_HW;
385 	block->offloadcnt--;
386 }
387 
388 static inline void
389 tc_cls_offload_cnt_update(struct tcf_block *block, u32 *cnt,
390 			  u32 *flags, bool add)
391 {
392 	if (add) {
393 		if (!*cnt)
394 			tcf_block_offload_inc(block, flags);
395 		(*cnt)++;
396 	} else {
397 		(*cnt)--;
398 		if (!*cnt)
399 			tcf_block_offload_dec(block, flags);
400 	}
401 }
402 
403 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
404 {
405 	struct qdisc_skb_cb *qcb;
406 
407 	BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
408 	BUILD_BUG_ON(sizeof(qcb->data) < sz);
409 }
410 
411 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
412 {
413 	return this_cpu_ptr(q->cpu_qstats)->qlen;
414 }
415 
416 static inline int qdisc_qlen(const struct Qdisc *q)
417 {
418 	return q->q.qlen;
419 }
420 
421 static inline int qdisc_qlen_sum(const struct Qdisc *q)
422 {
423 	__u32 qlen = q->qstats.qlen;
424 	int i;
425 
426 	if (q->flags & TCQ_F_NOLOCK) {
427 		for_each_possible_cpu(i)
428 			qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
429 	} else {
430 		qlen += q->q.qlen;
431 	}
432 
433 	return qlen;
434 }
435 
436 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
437 {
438 	return (struct qdisc_skb_cb *)skb->cb;
439 }
440 
441 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
442 {
443 	return &qdisc->q.lock;
444 }
445 
446 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
447 {
448 	struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
449 
450 	return q;
451 }
452 
453 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
454 {
455 	return qdisc->dev_queue->qdisc_sleeping;
456 }
457 
458 /* The qdisc root lock is a mechanism by which to top level
459  * of a qdisc tree can be locked from any qdisc node in the
460  * forest.  This allows changing the configuration of some
461  * aspect of the qdisc tree while blocking out asynchronous
462  * qdisc access in the packet processing paths.
463  *
464  * It is only legal to do this when the root will not change
465  * on us.  Otherwise we'll potentially lock the wrong qdisc
466  * root.  This is enforced by holding the RTNL semaphore, which
467  * all users of this lock accessor must do.
468  */
469 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
470 {
471 	struct Qdisc *root = qdisc_root(qdisc);
472 
473 	ASSERT_RTNL();
474 	return qdisc_lock(root);
475 }
476 
477 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
478 {
479 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
480 
481 	ASSERT_RTNL();
482 	return qdisc_lock(root);
483 }
484 
485 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
486 {
487 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
488 
489 	ASSERT_RTNL();
490 	return &root->running;
491 }
492 
493 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
494 {
495 	return qdisc->dev_queue->dev;
496 }
497 
498 static inline void sch_tree_lock(const struct Qdisc *q)
499 {
500 	spin_lock_bh(qdisc_root_sleeping_lock(q));
501 }
502 
503 static inline void sch_tree_unlock(const struct Qdisc *q)
504 {
505 	spin_unlock_bh(qdisc_root_sleeping_lock(q));
506 }
507 
508 extern struct Qdisc noop_qdisc;
509 extern struct Qdisc_ops noop_qdisc_ops;
510 extern struct Qdisc_ops pfifo_fast_ops;
511 extern struct Qdisc_ops mq_qdisc_ops;
512 extern struct Qdisc_ops noqueue_qdisc_ops;
513 extern const struct Qdisc_ops *default_qdisc_ops;
514 static inline const struct Qdisc_ops *
515 get_default_qdisc_ops(const struct net_device *dev, int ntx)
516 {
517 	return ntx < dev->real_num_tx_queues ?
518 			default_qdisc_ops : &pfifo_fast_ops;
519 }
520 
521 struct Qdisc_class_common {
522 	u32			classid;
523 	struct hlist_node	hnode;
524 };
525 
526 struct Qdisc_class_hash {
527 	struct hlist_head	*hash;
528 	unsigned int		hashsize;
529 	unsigned int		hashmask;
530 	unsigned int		hashelems;
531 };
532 
533 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
534 {
535 	id ^= id >> 8;
536 	id ^= id >> 4;
537 	return id & mask;
538 }
539 
540 static inline struct Qdisc_class_common *
541 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
542 {
543 	struct Qdisc_class_common *cl;
544 	unsigned int h;
545 
546 	if (!id)
547 		return NULL;
548 
549 	h = qdisc_class_hash(id, hash->hashmask);
550 	hlist_for_each_entry(cl, &hash->hash[h], hnode) {
551 		if (cl->classid == id)
552 			return cl;
553 	}
554 	return NULL;
555 }
556 
557 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
558 {
559 	u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
560 
561 	return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
562 }
563 
564 int qdisc_class_hash_init(struct Qdisc_class_hash *);
565 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
566 			     struct Qdisc_class_common *);
567 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
568 			     struct Qdisc_class_common *);
569 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
570 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
571 
572 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
573 void dev_init_scheduler(struct net_device *dev);
574 void dev_shutdown(struct net_device *dev);
575 void dev_activate(struct net_device *dev);
576 void dev_deactivate(struct net_device *dev);
577 void dev_deactivate_many(struct list_head *head);
578 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
579 			      struct Qdisc *qdisc);
580 void qdisc_reset(struct Qdisc *qdisc);
581 void qdisc_put(struct Qdisc *qdisc);
582 void qdisc_put_unlocked(struct Qdisc *qdisc);
583 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
584 			       unsigned int len);
585 #ifdef CONFIG_NET_SCHED
586 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
587 			      void *type_data);
588 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
589 				struct Qdisc *new, struct Qdisc *old,
590 				enum tc_setup_type type, void *type_data,
591 				struct netlink_ext_ack *extack);
592 #else
593 static inline int
594 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
595 			  void *type_data)
596 {
597 	q->flags &= ~TCQ_F_OFFLOADED;
598 	return 0;
599 }
600 
601 static inline void
602 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
603 			   struct Qdisc *new, struct Qdisc *old,
604 			   enum tc_setup_type type, void *type_data,
605 			   struct netlink_ext_ack *extack)
606 {
607 }
608 #endif
609 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
610 			  const struct Qdisc_ops *ops,
611 			  struct netlink_ext_ack *extack);
612 void qdisc_free(struct Qdisc *qdisc);
613 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
614 				const struct Qdisc_ops *ops, u32 parentid,
615 				struct netlink_ext_ack *extack);
616 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
617 			       const struct qdisc_size_table *stab);
618 int skb_do_redirect(struct sk_buff *);
619 
620 static inline void skb_reset_tc(struct sk_buff *skb)
621 {
622 #ifdef CONFIG_NET_CLS_ACT
623 	skb->tc_redirected = 0;
624 #endif
625 }
626 
627 static inline bool skb_is_tc_redirected(const struct sk_buff *skb)
628 {
629 #ifdef CONFIG_NET_CLS_ACT
630 	return skb->tc_redirected;
631 #else
632 	return false;
633 #endif
634 }
635 
636 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
637 {
638 #ifdef CONFIG_NET_CLS_ACT
639 	return skb->tc_at_ingress;
640 #else
641 	return false;
642 #endif
643 }
644 
645 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
646 {
647 #ifdef CONFIG_NET_CLS_ACT
648 	if (skb->tc_skip_classify) {
649 		skb->tc_skip_classify = 0;
650 		return true;
651 	}
652 #endif
653 	return false;
654 }
655 
656 /* Reset all TX qdiscs greater than index of a device.  */
657 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
658 {
659 	struct Qdisc *qdisc;
660 
661 	for (; i < dev->num_tx_queues; i++) {
662 		qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
663 		if (qdisc) {
664 			spin_lock_bh(qdisc_lock(qdisc));
665 			qdisc_reset(qdisc);
666 			spin_unlock_bh(qdisc_lock(qdisc));
667 		}
668 	}
669 }
670 
671 static inline void qdisc_reset_all_tx(struct net_device *dev)
672 {
673 	qdisc_reset_all_tx_gt(dev, 0);
674 }
675 
676 /* Are all TX queues of the device empty?  */
677 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
678 {
679 	unsigned int i;
680 
681 	rcu_read_lock();
682 	for (i = 0; i < dev->num_tx_queues; i++) {
683 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
684 		const struct Qdisc *q = rcu_dereference(txq->qdisc);
685 
686 		if (q->q.qlen) {
687 			rcu_read_unlock();
688 			return false;
689 		}
690 	}
691 	rcu_read_unlock();
692 	return true;
693 }
694 
695 /* Are any of the TX qdiscs changing?  */
696 static inline bool qdisc_tx_changing(const struct net_device *dev)
697 {
698 	unsigned int i;
699 
700 	for (i = 0; i < dev->num_tx_queues; i++) {
701 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
702 		if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
703 			return true;
704 	}
705 	return false;
706 }
707 
708 /* Is the device using the noop qdisc on all queues?  */
709 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
710 {
711 	unsigned int i;
712 
713 	for (i = 0; i < dev->num_tx_queues; i++) {
714 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
715 		if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
716 			return false;
717 	}
718 	return true;
719 }
720 
721 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
722 {
723 	return qdisc_skb_cb(skb)->pkt_len;
724 }
725 
726 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
727 enum net_xmit_qdisc_t {
728 	__NET_XMIT_STOLEN = 0x00010000,
729 	__NET_XMIT_BYPASS = 0x00020000,
730 };
731 
732 #ifdef CONFIG_NET_CLS_ACT
733 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
734 #else
735 #define net_xmit_drop_count(e)	(1)
736 #endif
737 
738 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
739 					   const struct Qdisc *sch)
740 {
741 #ifdef CONFIG_NET_SCHED
742 	struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
743 
744 	if (stab)
745 		__qdisc_calculate_pkt_len(skb, stab);
746 #endif
747 }
748 
749 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
750 				struct sk_buff **to_free)
751 {
752 	qdisc_calculate_pkt_len(skb, sch);
753 	return sch->enqueue(skb, sch, to_free);
754 }
755 
756 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
757 {
758 	return q->flags & TCQ_F_CPUSTATS;
759 }
760 
761 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
762 				  __u64 bytes, __u32 packets)
763 {
764 	bstats->bytes += bytes;
765 	bstats->packets += packets;
766 }
767 
768 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
769 				 const struct sk_buff *skb)
770 {
771 	_bstats_update(bstats,
772 		       qdisc_pkt_len(skb),
773 		       skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
774 }
775 
776 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
777 				      __u64 bytes, __u32 packets)
778 {
779 	u64_stats_update_begin(&bstats->syncp);
780 	_bstats_update(&bstats->bstats, bytes, packets);
781 	u64_stats_update_end(&bstats->syncp);
782 }
783 
784 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
785 				     const struct sk_buff *skb)
786 {
787 	u64_stats_update_begin(&bstats->syncp);
788 	bstats_update(&bstats->bstats, skb);
789 	u64_stats_update_end(&bstats->syncp);
790 }
791 
792 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
793 					   const struct sk_buff *skb)
794 {
795 	bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
796 }
797 
798 static inline void qdisc_bstats_update(struct Qdisc *sch,
799 				       const struct sk_buff *skb)
800 {
801 	bstats_update(&sch->bstats, skb);
802 }
803 
804 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
805 					    const struct sk_buff *skb)
806 {
807 	sch->qstats.backlog -= qdisc_pkt_len(skb);
808 }
809 
810 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
811 						const struct sk_buff *skb)
812 {
813 	this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
814 }
815 
816 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
817 					    const struct sk_buff *skb)
818 {
819 	sch->qstats.backlog += qdisc_pkt_len(skb);
820 }
821 
822 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
823 						const struct sk_buff *skb)
824 {
825 	this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
826 }
827 
828 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
829 {
830 	this_cpu_inc(sch->cpu_qstats->qlen);
831 }
832 
833 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
834 {
835 	this_cpu_dec(sch->cpu_qstats->qlen);
836 }
837 
838 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
839 {
840 	this_cpu_inc(sch->cpu_qstats->requeues);
841 }
842 
843 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
844 {
845 	sch->qstats.drops += count;
846 }
847 
848 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
849 {
850 	qstats->drops++;
851 }
852 
853 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
854 {
855 	qstats->overlimits++;
856 }
857 
858 static inline void qdisc_qstats_drop(struct Qdisc *sch)
859 {
860 	qstats_drop_inc(&sch->qstats);
861 }
862 
863 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
864 {
865 	this_cpu_inc(sch->cpu_qstats->drops);
866 }
867 
868 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
869 {
870 	sch->qstats.overlimits++;
871 }
872 
873 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
874 {
875 	qh->head = NULL;
876 	qh->tail = NULL;
877 	qh->qlen = 0;
878 }
879 
880 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
881 					struct qdisc_skb_head *qh)
882 {
883 	struct sk_buff *last = qh->tail;
884 
885 	if (last) {
886 		skb->next = NULL;
887 		last->next = skb;
888 		qh->tail = skb;
889 	} else {
890 		qh->tail = skb;
891 		qh->head = skb;
892 	}
893 	qh->qlen++;
894 }
895 
896 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
897 {
898 	__qdisc_enqueue_tail(skb, &sch->q);
899 	qdisc_qstats_backlog_inc(sch, skb);
900 	return NET_XMIT_SUCCESS;
901 }
902 
903 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
904 					struct qdisc_skb_head *qh)
905 {
906 	skb->next = qh->head;
907 
908 	if (!qh->head)
909 		qh->tail = skb;
910 	qh->head = skb;
911 	qh->qlen++;
912 }
913 
914 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
915 {
916 	struct sk_buff *skb = qh->head;
917 
918 	if (likely(skb != NULL)) {
919 		qh->head = skb->next;
920 		qh->qlen--;
921 		if (qh->head == NULL)
922 			qh->tail = NULL;
923 		skb->next = NULL;
924 	}
925 
926 	return skb;
927 }
928 
929 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
930 {
931 	struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
932 
933 	if (likely(skb != NULL)) {
934 		qdisc_qstats_backlog_dec(sch, skb);
935 		qdisc_bstats_update(sch, skb);
936 	}
937 
938 	return skb;
939 }
940 
941 /* Instead of calling kfree_skb() while root qdisc lock is held,
942  * queue the skb for future freeing at end of __dev_xmit_skb()
943  */
944 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
945 {
946 	skb->next = *to_free;
947 	*to_free = skb;
948 }
949 
950 static inline void __qdisc_drop_all(struct sk_buff *skb,
951 				    struct sk_buff **to_free)
952 {
953 	if (skb->prev)
954 		skb->prev->next = *to_free;
955 	else
956 		skb->next = *to_free;
957 	*to_free = skb;
958 }
959 
960 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
961 						   struct qdisc_skb_head *qh,
962 						   struct sk_buff **to_free)
963 {
964 	struct sk_buff *skb = __qdisc_dequeue_head(qh);
965 
966 	if (likely(skb != NULL)) {
967 		unsigned int len = qdisc_pkt_len(skb);
968 
969 		qdisc_qstats_backlog_dec(sch, skb);
970 		__qdisc_drop(skb, to_free);
971 		return len;
972 	}
973 
974 	return 0;
975 }
976 
977 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
978 						 struct sk_buff **to_free)
979 {
980 	return __qdisc_queue_drop_head(sch, &sch->q, to_free);
981 }
982 
983 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
984 {
985 	const struct qdisc_skb_head *qh = &sch->q;
986 
987 	return qh->head;
988 }
989 
990 /* generic pseudo peek method for non-work-conserving qdisc */
991 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
992 {
993 	struct sk_buff *skb = skb_peek(&sch->gso_skb);
994 
995 	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
996 	if (!skb) {
997 		skb = sch->dequeue(sch);
998 
999 		if (skb) {
1000 			__skb_queue_head(&sch->gso_skb, skb);
1001 			/* it's still part of the queue */
1002 			qdisc_qstats_backlog_inc(sch, skb);
1003 			sch->q.qlen++;
1004 		}
1005 	}
1006 
1007 	return skb;
1008 }
1009 
1010 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1011 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1012 {
1013 	struct sk_buff *skb = skb_peek(&sch->gso_skb);
1014 
1015 	if (skb) {
1016 		skb = __skb_dequeue(&sch->gso_skb);
1017 		qdisc_qstats_backlog_dec(sch, skb);
1018 		sch->q.qlen--;
1019 	} else {
1020 		skb = sch->dequeue(sch);
1021 	}
1022 
1023 	return skb;
1024 }
1025 
1026 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1027 {
1028 	/*
1029 	 * We do not know the backlog in bytes of this list, it
1030 	 * is up to the caller to correct it
1031 	 */
1032 	ASSERT_RTNL();
1033 	if (qh->qlen) {
1034 		rtnl_kfree_skbs(qh->head, qh->tail);
1035 
1036 		qh->head = NULL;
1037 		qh->tail = NULL;
1038 		qh->qlen = 0;
1039 	}
1040 }
1041 
1042 static inline void qdisc_reset_queue(struct Qdisc *sch)
1043 {
1044 	__qdisc_reset_queue(&sch->q);
1045 	sch->qstats.backlog = 0;
1046 }
1047 
1048 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1049 					  struct Qdisc **pold)
1050 {
1051 	struct Qdisc *old;
1052 
1053 	sch_tree_lock(sch);
1054 	old = *pold;
1055 	*pold = new;
1056 	if (old != NULL) {
1057 		unsigned int qlen = old->q.qlen;
1058 		unsigned int backlog = old->qstats.backlog;
1059 
1060 		qdisc_reset(old);
1061 		qdisc_tree_reduce_backlog(old, qlen, backlog);
1062 	}
1063 	sch_tree_unlock(sch);
1064 
1065 	return old;
1066 }
1067 
1068 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1069 {
1070 	rtnl_kfree_skbs(skb, skb);
1071 	qdisc_qstats_drop(sch);
1072 }
1073 
1074 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1075 				 struct sk_buff **to_free)
1076 {
1077 	__qdisc_drop(skb, to_free);
1078 	qdisc_qstats_cpu_drop(sch);
1079 
1080 	return NET_XMIT_DROP;
1081 }
1082 
1083 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1084 			     struct sk_buff **to_free)
1085 {
1086 	__qdisc_drop(skb, to_free);
1087 	qdisc_qstats_drop(sch);
1088 
1089 	return NET_XMIT_DROP;
1090 }
1091 
1092 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1093 				 struct sk_buff **to_free)
1094 {
1095 	__qdisc_drop_all(skb, to_free);
1096 	qdisc_qstats_drop(sch);
1097 
1098 	return NET_XMIT_DROP;
1099 }
1100 
1101 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1102    long it will take to send a packet given its size.
1103  */
1104 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1105 {
1106 	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1107 	if (slot < 0)
1108 		slot = 0;
1109 	slot >>= rtab->rate.cell_log;
1110 	if (slot > 255)
1111 		return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1112 	return rtab->data[slot];
1113 }
1114 
1115 struct psched_ratecfg {
1116 	u64	rate_bytes_ps; /* bytes per second */
1117 	u32	mult;
1118 	u16	overhead;
1119 	u8	linklayer;
1120 	u8	shift;
1121 };
1122 
1123 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1124 				unsigned int len)
1125 {
1126 	len += r->overhead;
1127 
1128 	if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1129 		return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1130 
1131 	return ((u64)len * r->mult) >> r->shift;
1132 }
1133 
1134 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1135 			       const struct tc_ratespec *conf,
1136 			       u64 rate64);
1137 
1138 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1139 					  const struct psched_ratecfg *r)
1140 {
1141 	memset(res, 0, sizeof(*res));
1142 
1143 	/* legacy struct tc_ratespec has a 32bit @rate field
1144 	 * Qdisc using 64bit rate should add new attributes
1145 	 * in order to maintain compatibility.
1146 	 */
1147 	res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1148 
1149 	res->overhead = r->overhead;
1150 	res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1151 }
1152 
1153 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1154  * The fast path only needs to access filter list and to update stats
1155  */
1156 struct mini_Qdisc {
1157 	struct tcf_proto *filter_list;
1158 	struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1159 	struct gnet_stats_queue	__percpu *cpu_qstats;
1160 	struct rcu_head rcu;
1161 };
1162 
1163 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1164 						const struct sk_buff *skb)
1165 {
1166 	bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1167 }
1168 
1169 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1170 {
1171 	this_cpu_inc(miniq->cpu_qstats->drops);
1172 }
1173 
1174 struct mini_Qdisc_pair {
1175 	struct mini_Qdisc miniq1;
1176 	struct mini_Qdisc miniq2;
1177 	struct mini_Qdisc __rcu **p_miniq;
1178 };
1179 
1180 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1181 			  struct tcf_proto *tp_head);
1182 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1183 			  struct mini_Qdisc __rcu **p_miniq);
1184 
1185 static inline void skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res)
1186 {
1187 	struct gnet_stats_queue *stats = res->qstats;
1188 	int ret;
1189 
1190 	if (res->ingress)
1191 		ret = netif_receive_skb(skb);
1192 	else
1193 		ret = dev_queue_xmit(skb);
1194 	if (ret && stats)
1195 		qstats_overlimit_inc(res->qstats);
1196 }
1197 
1198 #endif
1199