xref: /openbmc/linux/include/net/sch_generic.h (revision a0ae2562c6c4b2721d9fddba63b7286c13517d9f)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __NET_SCHED_GENERIC_H
3 #define __NET_SCHED_GENERIC_H
4 
5 #include <linux/netdevice.h>
6 #include <linux/types.h>
7 #include <linux/rcupdate.h>
8 #include <linux/pkt_sched.h>
9 #include <linux/pkt_cls.h>
10 #include <linux/percpu.h>
11 #include <linux/dynamic_queue_limits.h>
12 #include <linux/list.h>
13 #include <linux/refcount.h>
14 #include <linux/workqueue.h>
15 #include <net/gen_stats.h>
16 #include <net/rtnetlink.h>
17 
18 struct Qdisc_ops;
19 struct qdisc_walker;
20 struct tcf_walker;
21 struct module;
22 
23 typedef int tc_setup_cb_t(enum tc_setup_type type,
24 			  void *type_data, void *cb_priv);
25 
26 struct qdisc_rate_table {
27 	struct tc_ratespec rate;
28 	u32		data[256];
29 	struct qdisc_rate_table *next;
30 	int		refcnt;
31 };
32 
33 enum qdisc_state_t {
34 	__QDISC_STATE_SCHED,
35 	__QDISC_STATE_DEACTIVATED,
36 };
37 
38 struct qdisc_size_table {
39 	struct rcu_head		rcu;
40 	struct list_head	list;
41 	struct tc_sizespec	szopts;
42 	int			refcnt;
43 	u16			data[];
44 };
45 
46 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
47 struct qdisc_skb_head {
48 	struct sk_buff	*head;
49 	struct sk_buff	*tail;
50 	__u32		qlen;
51 	spinlock_t	lock;
52 };
53 
54 struct Qdisc {
55 	int 			(*enqueue)(struct sk_buff *skb,
56 					   struct Qdisc *sch,
57 					   struct sk_buff **to_free);
58 	struct sk_buff *	(*dequeue)(struct Qdisc *sch);
59 	unsigned int		flags;
60 #define TCQ_F_BUILTIN		1
61 #define TCQ_F_INGRESS		2
62 #define TCQ_F_CAN_BYPASS	4
63 #define TCQ_F_MQROOT		8
64 #define TCQ_F_ONETXQUEUE	0x10 /* dequeue_skb() can assume all skbs are for
65 				      * q->dev_queue : It can test
66 				      * netif_xmit_frozen_or_stopped() before
67 				      * dequeueing next packet.
68 				      * Its true for MQ/MQPRIO slaves, or non
69 				      * multiqueue device.
70 				      */
71 #define TCQ_F_WARN_NONWC	(1 << 16)
72 #define TCQ_F_CPUSTATS		0x20 /* run using percpu statistics */
73 #define TCQ_F_NOPARENT		0x40 /* root of its hierarchy :
74 				      * qdisc_tree_decrease_qlen() should stop.
75 				      */
76 #define TCQ_F_INVISIBLE		0x80 /* invisible by default in dump */
77 #define TCQ_F_NOLOCK		0x100 /* qdisc does not require locking */
78 #define TCQ_F_OFFLOADED		0x200 /* qdisc is offloaded to HW */
79 	u32			limit;
80 	const struct Qdisc_ops	*ops;
81 	struct qdisc_size_table	__rcu *stab;
82 	struct hlist_node       hash;
83 	u32			handle;
84 	u32			parent;
85 
86 	struct netdev_queue	*dev_queue;
87 
88 	struct net_rate_estimator __rcu *rate_est;
89 	struct gnet_stats_basic_cpu __percpu *cpu_bstats;
90 	struct gnet_stats_queue	__percpu *cpu_qstats;
91 	int			padded;
92 	refcount_t		refcnt;
93 
94 	/*
95 	 * For performance sake on SMP, we put highly modified fields at the end
96 	 */
97 	struct sk_buff_head	gso_skb ____cacheline_aligned_in_smp;
98 	struct qdisc_skb_head	q;
99 	struct gnet_stats_basic_packed bstats;
100 	seqcount_t		running;
101 	struct gnet_stats_queue	qstats;
102 	unsigned long		state;
103 	struct Qdisc            *next_sched;
104 	struct sk_buff_head	skb_bad_txq;
105 
106 	spinlock_t		busylock ____cacheline_aligned_in_smp;
107 	spinlock_t		seqlock;
108 };
109 
110 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
111 {
112 	if (qdisc->flags & TCQ_F_BUILTIN)
113 		return;
114 	refcount_inc(&qdisc->refcnt);
115 }
116 
117 static inline bool qdisc_is_running(struct Qdisc *qdisc)
118 {
119 	if (qdisc->flags & TCQ_F_NOLOCK)
120 		return spin_is_locked(&qdisc->seqlock);
121 	return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
122 }
123 
124 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
125 {
126 	if (qdisc->flags & TCQ_F_NOLOCK) {
127 		if (!spin_trylock(&qdisc->seqlock))
128 			return false;
129 	} else if (qdisc_is_running(qdisc)) {
130 		return false;
131 	}
132 	/* Variant of write_seqcount_begin() telling lockdep a trylock
133 	 * was attempted.
134 	 */
135 	raw_write_seqcount_begin(&qdisc->running);
136 	seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
137 	return true;
138 }
139 
140 static inline void qdisc_run_end(struct Qdisc *qdisc)
141 {
142 	write_seqcount_end(&qdisc->running);
143 	if (qdisc->flags & TCQ_F_NOLOCK)
144 		spin_unlock(&qdisc->seqlock);
145 }
146 
147 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
148 {
149 	return qdisc->flags & TCQ_F_ONETXQUEUE;
150 }
151 
152 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
153 {
154 #ifdef CONFIG_BQL
155 	/* Non-BQL migrated drivers will return 0, too. */
156 	return dql_avail(&txq->dql);
157 #else
158 	return 0;
159 #endif
160 }
161 
162 struct Qdisc_class_ops {
163 	/* Child qdisc manipulation */
164 	struct netdev_queue *	(*select_queue)(struct Qdisc *, struct tcmsg *);
165 	int			(*graft)(struct Qdisc *, unsigned long cl,
166 					struct Qdisc *, struct Qdisc **,
167 					struct netlink_ext_ack *extack);
168 	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
169 	void			(*qlen_notify)(struct Qdisc *, unsigned long);
170 
171 	/* Class manipulation routines */
172 	unsigned long		(*find)(struct Qdisc *, u32 classid);
173 	int			(*change)(struct Qdisc *, u32, u32,
174 					struct nlattr **, unsigned long *,
175 					struct netlink_ext_ack *);
176 	int			(*delete)(struct Qdisc *, unsigned long);
177 	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
178 
179 	/* Filter manipulation */
180 	struct tcf_block *	(*tcf_block)(struct Qdisc *sch,
181 					     unsigned long arg,
182 					     struct netlink_ext_ack *extack);
183 	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
184 					u32 classid);
185 	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
186 
187 	/* rtnetlink specific */
188 	int			(*dump)(struct Qdisc *, unsigned long,
189 					struct sk_buff *skb, struct tcmsg*);
190 	int			(*dump_stats)(struct Qdisc *, unsigned long,
191 					struct gnet_dump *);
192 };
193 
194 struct Qdisc_ops {
195 	struct Qdisc_ops	*next;
196 	const struct Qdisc_class_ops	*cl_ops;
197 	char			id[IFNAMSIZ];
198 	int			priv_size;
199 	unsigned int		static_flags;
200 
201 	int 			(*enqueue)(struct sk_buff *skb,
202 					   struct Qdisc *sch,
203 					   struct sk_buff **to_free);
204 	struct sk_buff *	(*dequeue)(struct Qdisc *);
205 	struct sk_buff *	(*peek)(struct Qdisc *);
206 
207 	int			(*init)(struct Qdisc *sch, struct nlattr *arg,
208 					struct netlink_ext_ack *extack);
209 	void			(*reset)(struct Qdisc *);
210 	void			(*destroy)(struct Qdisc *);
211 	int			(*change)(struct Qdisc *sch,
212 					  struct nlattr *arg,
213 					  struct netlink_ext_ack *extack);
214 	void			(*attach)(struct Qdisc *sch);
215 	int			(*change_tx_queue_len)(struct Qdisc *, unsigned int);
216 
217 	int			(*dump)(struct Qdisc *, struct sk_buff *);
218 	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
219 
220 	void			(*ingress_block_set)(struct Qdisc *sch,
221 						     u32 block_index);
222 	void			(*egress_block_set)(struct Qdisc *sch,
223 						    u32 block_index);
224 	u32			(*ingress_block_get)(struct Qdisc *sch);
225 	u32			(*egress_block_get)(struct Qdisc *sch);
226 
227 	struct module		*owner;
228 };
229 
230 
231 struct tcf_result {
232 	union {
233 		struct {
234 			unsigned long	class;
235 			u32		classid;
236 		};
237 		const struct tcf_proto *goto_tp;
238 	};
239 };
240 
241 struct tcf_proto_ops {
242 	struct list_head	head;
243 	char			kind[IFNAMSIZ];
244 
245 	int			(*classify)(struct sk_buff *,
246 					    const struct tcf_proto *,
247 					    struct tcf_result *);
248 	int			(*init)(struct tcf_proto*);
249 	void			(*destroy)(struct tcf_proto *tp,
250 					   struct netlink_ext_ack *extack);
251 
252 	void*			(*get)(struct tcf_proto*, u32 handle);
253 	int			(*change)(struct net *net, struct sk_buff *,
254 					struct tcf_proto*, unsigned long,
255 					u32 handle, struct nlattr **,
256 					void **, bool,
257 					struct netlink_ext_ack *);
258 	int			(*delete)(struct tcf_proto *tp, void *arg,
259 					  bool *last,
260 					  struct netlink_ext_ack *);
261 	void			(*walk)(struct tcf_proto*, struct tcf_walker *arg);
262 	int			(*reoffload)(struct tcf_proto *tp, bool add,
263 					     tc_setup_cb_t *cb, void *cb_priv,
264 					     struct netlink_ext_ack *extack);
265 	void			(*bind_class)(void *, u32, unsigned long);
266 
267 	/* rtnetlink specific */
268 	int			(*dump)(struct net*, struct tcf_proto*, void *,
269 					struct sk_buff *skb, struct tcmsg*);
270 
271 	struct module		*owner;
272 };
273 
274 struct tcf_proto {
275 	/* Fast access part */
276 	struct tcf_proto __rcu	*next;
277 	void __rcu		*root;
278 	int			(*classify)(struct sk_buff *,
279 					    const struct tcf_proto *,
280 					    struct tcf_result *);
281 	__be16			protocol;
282 
283 	/* All the rest */
284 	u32			prio;
285 	void			*data;
286 	const struct tcf_proto_ops	*ops;
287 	struct tcf_chain	*chain;
288 	struct rcu_head		rcu;
289 };
290 
291 struct qdisc_skb_cb {
292 	unsigned int		pkt_len;
293 	u16			slave_dev_queue_mapping;
294 	u16			tc_classid;
295 #define QDISC_CB_PRIV_LEN 20
296 	unsigned char		data[QDISC_CB_PRIV_LEN];
297 };
298 
299 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
300 
301 struct tcf_chain {
302 	struct tcf_proto __rcu *filter_chain;
303 	struct list_head filter_chain_list;
304 	struct list_head list;
305 	struct tcf_block *block;
306 	u32 index; /* chain index */
307 	unsigned int refcnt;
308 };
309 
310 struct tcf_block {
311 	struct list_head chain_list;
312 	u32 index; /* block index for shared blocks */
313 	unsigned int refcnt;
314 	struct net *net;
315 	struct Qdisc *q;
316 	struct list_head cb_list;
317 	struct list_head owner_list;
318 	bool keep_dst;
319 	unsigned int offloadcnt; /* Number of oddloaded filters */
320 	unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
321 };
322 
323 static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
324 {
325 	if (*flags & TCA_CLS_FLAGS_IN_HW)
326 		return;
327 	*flags |= TCA_CLS_FLAGS_IN_HW;
328 	block->offloadcnt++;
329 }
330 
331 static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
332 {
333 	if (!(*flags & TCA_CLS_FLAGS_IN_HW))
334 		return;
335 	*flags &= ~TCA_CLS_FLAGS_IN_HW;
336 	block->offloadcnt--;
337 }
338 
339 static inline void
340 tc_cls_offload_cnt_update(struct tcf_block *block, unsigned int *cnt,
341 			  u32 *flags, bool add)
342 {
343 	if (add) {
344 		if (!*cnt)
345 			tcf_block_offload_inc(block, flags);
346 		(*cnt)++;
347 	} else {
348 		(*cnt)--;
349 		if (!*cnt)
350 			tcf_block_offload_dec(block, flags);
351 	}
352 }
353 
354 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
355 {
356 	struct qdisc_skb_cb *qcb;
357 
358 	BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
359 	BUILD_BUG_ON(sizeof(qcb->data) < sz);
360 }
361 
362 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
363 {
364 	return this_cpu_ptr(q->cpu_qstats)->qlen;
365 }
366 
367 static inline int qdisc_qlen(const struct Qdisc *q)
368 {
369 	return q->q.qlen;
370 }
371 
372 static inline int qdisc_qlen_sum(const struct Qdisc *q)
373 {
374 	__u32 qlen = q->qstats.qlen;
375 	int i;
376 
377 	if (q->flags & TCQ_F_NOLOCK) {
378 		for_each_possible_cpu(i)
379 			qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
380 	} else {
381 		qlen += q->q.qlen;
382 	}
383 
384 	return qlen;
385 }
386 
387 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
388 {
389 	return (struct qdisc_skb_cb *)skb->cb;
390 }
391 
392 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
393 {
394 	return &qdisc->q.lock;
395 }
396 
397 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
398 {
399 	struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
400 
401 	return q;
402 }
403 
404 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
405 {
406 	return qdisc->dev_queue->qdisc_sleeping;
407 }
408 
409 /* The qdisc root lock is a mechanism by which to top level
410  * of a qdisc tree can be locked from any qdisc node in the
411  * forest.  This allows changing the configuration of some
412  * aspect of the qdisc tree while blocking out asynchronous
413  * qdisc access in the packet processing paths.
414  *
415  * It is only legal to do this when the root will not change
416  * on us.  Otherwise we'll potentially lock the wrong qdisc
417  * root.  This is enforced by holding the RTNL semaphore, which
418  * all users of this lock accessor must do.
419  */
420 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
421 {
422 	struct Qdisc *root = qdisc_root(qdisc);
423 
424 	ASSERT_RTNL();
425 	return qdisc_lock(root);
426 }
427 
428 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
429 {
430 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
431 
432 	ASSERT_RTNL();
433 	return qdisc_lock(root);
434 }
435 
436 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
437 {
438 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
439 
440 	ASSERT_RTNL();
441 	return &root->running;
442 }
443 
444 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
445 {
446 	return qdisc->dev_queue->dev;
447 }
448 
449 static inline void sch_tree_lock(const struct Qdisc *q)
450 {
451 	spin_lock_bh(qdisc_root_sleeping_lock(q));
452 }
453 
454 static inline void sch_tree_unlock(const struct Qdisc *q)
455 {
456 	spin_unlock_bh(qdisc_root_sleeping_lock(q));
457 }
458 
459 extern struct Qdisc noop_qdisc;
460 extern struct Qdisc_ops noop_qdisc_ops;
461 extern struct Qdisc_ops pfifo_fast_ops;
462 extern struct Qdisc_ops mq_qdisc_ops;
463 extern struct Qdisc_ops noqueue_qdisc_ops;
464 extern const struct Qdisc_ops *default_qdisc_ops;
465 static inline const struct Qdisc_ops *
466 get_default_qdisc_ops(const struct net_device *dev, int ntx)
467 {
468 	return ntx < dev->real_num_tx_queues ?
469 			default_qdisc_ops : &pfifo_fast_ops;
470 }
471 
472 struct Qdisc_class_common {
473 	u32			classid;
474 	struct hlist_node	hnode;
475 };
476 
477 struct Qdisc_class_hash {
478 	struct hlist_head	*hash;
479 	unsigned int		hashsize;
480 	unsigned int		hashmask;
481 	unsigned int		hashelems;
482 };
483 
484 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
485 {
486 	id ^= id >> 8;
487 	id ^= id >> 4;
488 	return id & mask;
489 }
490 
491 static inline struct Qdisc_class_common *
492 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
493 {
494 	struct Qdisc_class_common *cl;
495 	unsigned int h;
496 
497 	if (!id)
498 		return NULL;
499 
500 	h = qdisc_class_hash(id, hash->hashmask);
501 	hlist_for_each_entry(cl, &hash->hash[h], hnode) {
502 		if (cl->classid == id)
503 			return cl;
504 	}
505 	return NULL;
506 }
507 
508 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
509 {
510 	u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
511 
512 	return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
513 }
514 
515 int qdisc_class_hash_init(struct Qdisc_class_hash *);
516 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
517 			     struct Qdisc_class_common *);
518 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
519 			     struct Qdisc_class_common *);
520 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
521 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
522 
523 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
524 void dev_init_scheduler(struct net_device *dev);
525 void dev_shutdown(struct net_device *dev);
526 void dev_activate(struct net_device *dev);
527 void dev_deactivate(struct net_device *dev);
528 void dev_deactivate_many(struct list_head *head);
529 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
530 			      struct Qdisc *qdisc);
531 void qdisc_reset(struct Qdisc *qdisc);
532 void qdisc_destroy(struct Qdisc *qdisc);
533 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
534 			       unsigned int len);
535 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
536 			  const struct Qdisc_ops *ops,
537 			  struct netlink_ext_ack *extack);
538 void qdisc_free(struct Qdisc *qdisc);
539 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
540 				const struct Qdisc_ops *ops, u32 parentid,
541 				struct netlink_ext_ack *extack);
542 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
543 			       const struct qdisc_size_table *stab);
544 int skb_do_redirect(struct sk_buff *);
545 
546 static inline void skb_reset_tc(struct sk_buff *skb)
547 {
548 #ifdef CONFIG_NET_CLS_ACT
549 	skb->tc_redirected = 0;
550 #endif
551 }
552 
553 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
554 {
555 #ifdef CONFIG_NET_CLS_ACT
556 	return skb->tc_at_ingress;
557 #else
558 	return false;
559 #endif
560 }
561 
562 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
563 {
564 #ifdef CONFIG_NET_CLS_ACT
565 	if (skb->tc_skip_classify) {
566 		skb->tc_skip_classify = 0;
567 		return true;
568 	}
569 #endif
570 	return false;
571 }
572 
573 /* Reset all TX qdiscs greater than index of a device.  */
574 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
575 {
576 	struct Qdisc *qdisc;
577 
578 	for (; i < dev->num_tx_queues; i++) {
579 		qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
580 		if (qdisc) {
581 			spin_lock_bh(qdisc_lock(qdisc));
582 			qdisc_reset(qdisc);
583 			spin_unlock_bh(qdisc_lock(qdisc));
584 		}
585 	}
586 }
587 
588 static inline void qdisc_reset_all_tx(struct net_device *dev)
589 {
590 	qdisc_reset_all_tx_gt(dev, 0);
591 }
592 
593 /* Are all TX queues of the device empty?  */
594 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
595 {
596 	unsigned int i;
597 
598 	rcu_read_lock();
599 	for (i = 0; i < dev->num_tx_queues; i++) {
600 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
601 		const struct Qdisc *q = rcu_dereference(txq->qdisc);
602 
603 		if (q->q.qlen) {
604 			rcu_read_unlock();
605 			return false;
606 		}
607 	}
608 	rcu_read_unlock();
609 	return true;
610 }
611 
612 /* Are any of the TX qdiscs changing?  */
613 static inline bool qdisc_tx_changing(const struct net_device *dev)
614 {
615 	unsigned int i;
616 
617 	for (i = 0; i < dev->num_tx_queues; i++) {
618 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
619 		if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
620 			return true;
621 	}
622 	return false;
623 }
624 
625 /* Is the device using the noop qdisc on all queues?  */
626 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
627 {
628 	unsigned int i;
629 
630 	for (i = 0; i < dev->num_tx_queues; i++) {
631 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
632 		if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
633 			return false;
634 	}
635 	return true;
636 }
637 
638 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
639 {
640 	return qdisc_skb_cb(skb)->pkt_len;
641 }
642 
643 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
644 enum net_xmit_qdisc_t {
645 	__NET_XMIT_STOLEN = 0x00010000,
646 	__NET_XMIT_BYPASS = 0x00020000,
647 };
648 
649 #ifdef CONFIG_NET_CLS_ACT
650 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
651 #else
652 #define net_xmit_drop_count(e)	(1)
653 #endif
654 
655 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
656 					   const struct Qdisc *sch)
657 {
658 #ifdef CONFIG_NET_SCHED
659 	struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
660 
661 	if (stab)
662 		__qdisc_calculate_pkt_len(skb, stab);
663 #endif
664 }
665 
666 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
667 				struct sk_buff **to_free)
668 {
669 	qdisc_calculate_pkt_len(skb, sch);
670 	return sch->enqueue(skb, sch, to_free);
671 }
672 
673 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
674 {
675 	return q->flags & TCQ_F_CPUSTATS;
676 }
677 
678 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
679 				  __u64 bytes, __u32 packets)
680 {
681 	bstats->bytes += bytes;
682 	bstats->packets += packets;
683 }
684 
685 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
686 				 const struct sk_buff *skb)
687 {
688 	_bstats_update(bstats,
689 		       qdisc_pkt_len(skb),
690 		       skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
691 }
692 
693 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
694 				      __u64 bytes, __u32 packets)
695 {
696 	u64_stats_update_begin(&bstats->syncp);
697 	_bstats_update(&bstats->bstats, bytes, packets);
698 	u64_stats_update_end(&bstats->syncp);
699 }
700 
701 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
702 				     const struct sk_buff *skb)
703 {
704 	u64_stats_update_begin(&bstats->syncp);
705 	bstats_update(&bstats->bstats, skb);
706 	u64_stats_update_end(&bstats->syncp);
707 }
708 
709 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
710 					   const struct sk_buff *skb)
711 {
712 	bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
713 }
714 
715 static inline void qdisc_bstats_update(struct Qdisc *sch,
716 				       const struct sk_buff *skb)
717 {
718 	bstats_update(&sch->bstats, skb);
719 }
720 
721 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
722 					    const struct sk_buff *skb)
723 {
724 	sch->qstats.backlog -= qdisc_pkt_len(skb);
725 }
726 
727 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
728 						const struct sk_buff *skb)
729 {
730 	this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
731 }
732 
733 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
734 					    const struct sk_buff *skb)
735 {
736 	sch->qstats.backlog += qdisc_pkt_len(skb);
737 }
738 
739 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
740 						const struct sk_buff *skb)
741 {
742 	this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
743 }
744 
745 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
746 {
747 	this_cpu_inc(sch->cpu_qstats->qlen);
748 }
749 
750 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
751 {
752 	this_cpu_dec(sch->cpu_qstats->qlen);
753 }
754 
755 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
756 {
757 	this_cpu_inc(sch->cpu_qstats->requeues);
758 }
759 
760 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
761 {
762 	sch->qstats.drops += count;
763 }
764 
765 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
766 {
767 	qstats->drops++;
768 }
769 
770 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
771 {
772 	qstats->overlimits++;
773 }
774 
775 static inline void qdisc_qstats_drop(struct Qdisc *sch)
776 {
777 	qstats_drop_inc(&sch->qstats);
778 }
779 
780 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
781 {
782 	this_cpu_inc(sch->cpu_qstats->drops);
783 }
784 
785 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
786 {
787 	sch->qstats.overlimits++;
788 }
789 
790 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
791 {
792 	qh->head = NULL;
793 	qh->tail = NULL;
794 	qh->qlen = 0;
795 }
796 
797 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
798 				       struct qdisc_skb_head *qh)
799 {
800 	struct sk_buff *last = qh->tail;
801 
802 	if (last) {
803 		skb->next = NULL;
804 		last->next = skb;
805 		qh->tail = skb;
806 	} else {
807 		qh->tail = skb;
808 		qh->head = skb;
809 	}
810 	qh->qlen++;
811 	qdisc_qstats_backlog_inc(sch, skb);
812 
813 	return NET_XMIT_SUCCESS;
814 }
815 
816 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
817 {
818 	return __qdisc_enqueue_tail(skb, sch, &sch->q);
819 }
820 
821 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
822 {
823 	struct sk_buff *skb = qh->head;
824 
825 	if (likely(skb != NULL)) {
826 		qh->head = skb->next;
827 		qh->qlen--;
828 		if (qh->head == NULL)
829 			qh->tail = NULL;
830 		skb->next = NULL;
831 	}
832 
833 	return skb;
834 }
835 
836 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
837 {
838 	struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
839 
840 	if (likely(skb != NULL)) {
841 		qdisc_qstats_backlog_dec(sch, skb);
842 		qdisc_bstats_update(sch, skb);
843 	}
844 
845 	return skb;
846 }
847 
848 /* Instead of calling kfree_skb() while root qdisc lock is held,
849  * queue the skb for future freeing at end of __dev_xmit_skb()
850  */
851 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
852 {
853 	skb->next = *to_free;
854 	*to_free = skb;
855 }
856 
857 static inline void __qdisc_drop_all(struct sk_buff *skb,
858 				    struct sk_buff **to_free)
859 {
860 	if (skb->prev)
861 		skb->prev->next = *to_free;
862 	else
863 		skb->next = *to_free;
864 	*to_free = skb;
865 }
866 
867 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
868 						   struct qdisc_skb_head *qh,
869 						   struct sk_buff **to_free)
870 {
871 	struct sk_buff *skb = __qdisc_dequeue_head(qh);
872 
873 	if (likely(skb != NULL)) {
874 		unsigned int len = qdisc_pkt_len(skb);
875 
876 		qdisc_qstats_backlog_dec(sch, skb);
877 		__qdisc_drop(skb, to_free);
878 		return len;
879 	}
880 
881 	return 0;
882 }
883 
884 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
885 						 struct sk_buff **to_free)
886 {
887 	return __qdisc_queue_drop_head(sch, &sch->q, to_free);
888 }
889 
890 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
891 {
892 	const struct qdisc_skb_head *qh = &sch->q;
893 
894 	return qh->head;
895 }
896 
897 /* generic pseudo peek method for non-work-conserving qdisc */
898 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
899 {
900 	struct sk_buff *skb = skb_peek(&sch->gso_skb);
901 
902 	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
903 	if (!skb) {
904 		skb = sch->dequeue(sch);
905 
906 		if (skb) {
907 			__skb_queue_head(&sch->gso_skb, skb);
908 			/* it's still part of the queue */
909 			qdisc_qstats_backlog_inc(sch, skb);
910 			sch->q.qlen++;
911 		}
912 	}
913 
914 	return skb;
915 }
916 
917 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
918 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
919 {
920 	struct sk_buff *skb = skb_peek(&sch->gso_skb);
921 
922 	if (skb) {
923 		skb = __skb_dequeue(&sch->gso_skb);
924 		qdisc_qstats_backlog_dec(sch, skb);
925 		sch->q.qlen--;
926 	} else {
927 		skb = sch->dequeue(sch);
928 	}
929 
930 	return skb;
931 }
932 
933 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
934 {
935 	/*
936 	 * We do not know the backlog in bytes of this list, it
937 	 * is up to the caller to correct it
938 	 */
939 	ASSERT_RTNL();
940 	if (qh->qlen) {
941 		rtnl_kfree_skbs(qh->head, qh->tail);
942 
943 		qh->head = NULL;
944 		qh->tail = NULL;
945 		qh->qlen = 0;
946 	}
947 }
948 
949 static inline void qdisc_reset_queue(struct Qdisc *sch)
950 {
951 	__qdisc_reset_queue(&sch->q);
952 	sch->qstats.backlog = 0;
953 }
954 
955 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
956 					  struct Qdisc **pold)
957 {
958 	struct Qdisc *old;
959 
960 	sch_tree_lock(sch);
961 	old = *pold;
962 	*pold = new;
963 	if (old != NULL) {
964 		unsigned int qlen = old->q.qlen;
965 		unsigned int backlog = old->qstats.backlog;
966 
967 		qdisc_reset(old);
968 		qdisc_tree_reduce_backlog(old, qlen, backlog);
969 	}
970 	sch_tree_unlock(sch);
971 
972 	return old;
973 }
974 
975 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
976 {
977 	rtnl_kfree_skbs(skb, skb);
978 	qdisc_qstats_drop(sch);
979 }
980 
981 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
982 				 struct sk_buff **to_free)
983 {
984 	__qdisc_drop(skb, to_free);
985 	qdisc_qstats_cpu_drop(sch);
986 
987 	return NET_XMIT_DROP;
988 }
989 
990 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
991 			     struct sk_buff **to_free)
992 {
993 	__qdisc_drop(skb, to_free);
994 	qdisc_qstats_drop(sch);
995 
996 	return NET_XMIT_DROP;
997 }
998 
999 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1000 				 struct sk_buff **to_free)
1001 {
1002 	__qdisc_drop_all(skb, to_free);
1003 	qdisc_qstats_drop(sch);
1004 
1005 	return NET_XMIT_DROP;
1006 }
1007 
1008 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1009    long it will take to send a packet given its size.
1010  */
1011 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1012 {
1013 	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1014 	if (slot < 0)
1015 		slot = 0;
1016 	slot >>= rtab->rate.cell_log;
1017 	if (slot > 255)
1018 		return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1019 	return rtab->data[slot];
1020 }
1021 
1022 struct psched_ratecfg {
1023 	u64	rate_bytes_ps; /* bytes per second */
1024 	u32	mult;
1025 	u16	overhead;
1026 	u8	linklayer;
1027 	u8	shift;
1028 };
1029 
1030 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1031 				unsigned int len)
1032 {
1033 	len += r->overhead;
1034 
1035 	if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1036 		return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1037 
1038 	return ((u64)len * r->mult) >> r->shift;
1039 }
1040 
1041 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1042 			       const struct tc_ratespec *conf,
1043 			       u64 rate64);
1044 
1045 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1046 					  const struct psched_ratecfg *r)
1047 {
1048 	memset(res, 0, sizeof(*res));
1049 
1050 	/* legacy struct tc_ratespec has a 32bit @rate field
1051 	 * Qdisc using 64bit rate should add new attributes
1052 	 * in order to maintain compatibility.
1053 	 */
1054 	res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1055 
1056 	res->overhead = r->overhead;
1057 	res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1058 }
1059 
1060 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1061  * The fast path only needs to access filter list and to update stats
1062  */
1063 struct mini_Qdisc {
1064 	struct tcf_proto *filter_list;
1065 	struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1066 	struct gnet_stats_queue	__percpu *cpu_qstats;
1067 	struct rcu_head rcu;
1068 };
1069 
1070 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1071 						const struct sk_buff *skb)
1072 {
1073 	bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1074 }
1075 
1076 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1077 {
1078 	this_cpu_inc(miniq->cpu_qstats->drops);
1079 }
1080 
1081 struct mini_Qdisc_pair {
1082 	struct mini_Qdisc miniq1;
1083 	struct mini_Qdisc miniq2;
1084 	struct mini_Qdisc __rcu **p_miniq;
1085 };
1086 
1087 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1088 			  struct tcf_proto *tp_head);
1089 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1090 			  struct mini_Qdisc __rcu **p_miniq);
1091 
1092 #endif
1093