xref: /linux/include/net/sch_generic.h (revision cb299ba8b5ef2239429484072fea394cd7581bd7)
1 #ifndef __NET_SCHED_GENERIC_H
2 #define __NET_SCHED_GENERIC_H
3 
4 #include <linux/netdevice.h>
5 #include <linux/types.h>
6 #include <linux/rcupdate.h>
7 #include <linux/module.h>
8 #include <linux/pkt_sched.h>
9 #include <linux/pkt_cls.h>
10 #include <net/gen_stats.h>
11 #include <net/rtnetlink.h>
12 
13 struct Qdisc_ops;
14 struct qdisc_walker;
15 struct tcf_walker;
16 struct module;
17 
18 struct qdisc_rate_table {
19 	struct tc_ratespec rate;
20 	u32		data[256];
21 	struct qdisc_rate_table *next;
22 	int		refcnt;
23 };
24 
25 enum qdisc_state_t {
26 	__QDISC_STATE_SCHED,
27 	__QDISC_STATE_DEACTIVATED,
28 };
29 
30 /*
31  * following bits are only changed while qdisc lock is held
32  */
33 enum qdisc___state_t {
34 	__QDISC___STATE_RUNNING,
35 };
36 
37 struct qdisc_size_table {
38 	struct list_head	list;
39 	struct tc_sizespec	szopts;
40 	int			refcnt;
41 	u16			data[];
42 };
43 
44 struct Qdisc {
45 	int 			(*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
46 	struct sk_buff *	(*dequeue)(struct Qdisc *dev);
47 	unsigned		flags;
48 #define TCQ_F_BUILTIN		1
49 #define TCQ_F_THROTTLED		2
50 #define TCQ_F_INGRESS		4
51 #define TCQ_F_CAN_BYPASS	8
52 #define TCQ_F_MQROOT		16
53 #define TCQ_F_WARN_NONWC	(1 << 16)
54 	int			padded;
55 	struct Qdisc_ops	*ops;
56 	struct qdisc_size_table	*stab;
57 	struct list_head	list;
58 	u32			handle;
59 	u32			parent;
60 	atomic_t		refcnt;
61 	struct gnet_stats_rate_est	rate_est;
62 	int			(*reshape_fail)(struct sk_buff *skb,
63 					struct Qdisc *q);
64 
65 	void			*u32_node;
66 
67 	/* This field is deprecated, but it is still used by CBQ
68 	 * and it will live until better solution will be invented.
69 	 */
70 	struct Qdisc		*__parent;
71 	struct netdev_queue	*dev_queue;
72 	struct Qdisc		*next_sched;
73 
74 	struct sk_buff		*gso_skb;
75 	/*
76 	 * For performance sake on SMP, we put highly modified fields at the end
77 	 */
78 	unsigned long		state;
79 	struct sk_buff_head	q;
80 	struct gnet_stats_basic_packed bstats;
81 	unsigned long		__state;
82 	struct gnet_stats_queue	qstats;
83 	struct rcu_head		rcu_head;
84 	spinlock_t		busylock;
85 };
86 
87 static inline bool qdisc_is_running(struct Qdisc *qdisc)
88 {
89 	return test_bit(__QDISC___STATE_RUNNING, &qdisc->__state);
90 }
91 
92 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
93 {
94 	return !__test_and_set_bit(__QDISC___STATE_RUNNING, &qdisc->__state);
95 }
96 
97 static inline void qdisc_run_end(struct Qdisc *qdisc)
98 {
99 	__clear_bit(__QDISC___STATE_RUNNING, &qdisc->__state);
100 }
101 
102 struct Qdisc_class_ops {
103 	/* Child qdisc manipulation */
104 	struct netdev_queue *	(*select_queue)(struct Qdisc *, struct tcmsg *);
105 	int			(*graft)(struct Qdisc *, unsigned long cl,
106 					struct Qdisc *, struct Qdisc **);
107 	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
108 	void			(*qlen_notify)(struct Qdisc *, unsigned long);
109 
110 	/* Class manipulation routines */
111 	unsigned long		(*get)(struct Qdisc *, u32 classid);
112 	void			(*put)(struct Qdisc *, unsigned long);
113 	int			(*change)(struct Qdisc *, u32, u32,
114 					struct nlattr **, unsigned long *);
115 	int			(*delete)(struct Qdisc *, unsigned long);
116 	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
117 
118 	/* Filter manipulation */
119 	struct tcf_proto **	(*tcf_chain)(struct Qdisc *, unsigned long);
120 	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
121 					u32 classid);
122 	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
123 
124 	/* rtnetlink specific */
125 	int			(*dump)(struct Qdisc *, unsigned long,
126 					struct sk_buff *skb, struct tcmsg*);
127 	int			(*dump_stats)(struct Qdisc *, unsigned long,
128 					struct gnet_dump *);
129 };
130 
131 struct Qdisc_ops {
132 	struct Qdisc_ops	*next;
133 	const struct Qdisc_class_ops	*cl_ops;
134 	char			id[IFNAMSIZ];
135 	int			priv_size;
136 
137 	int 			(*enqueue)(struct sk_buff *, struct Qdisc *);
138 	struct sk_buff *	(*dequeue)(struct Qdisc *);
139 	struct sk_buff *	(*peek)(struct Qdisc *);
140 	unsigned int		(*drop)(struct Qdisc *);
141 
142 	int			(*init)(struct Qdisc *, struct nlattr *arg);
143 	void			(*reset)(struct Qdisc *);
144 	void			(*destroy)(struct Qdisc *);
145 	int			(*change)(struct Qdisc *, struct nlattr *arg);
146 	void			(*attach)(struct Qdisc *);
147 
148 	int			(*dump)(struct Qdisc *, struct sk_buff *);
149 	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
150 
151 	struct module		*owner;
152 };
153 
154 
155 struct tcf_result {
156 	unsigned long	class;
157 	u32		classid;
158 };
159 
160 struct tcf_proto_ops {
161 	struct tcf_proto_ops	*next;
162 	char			kind[IFNAMSIZ];
163 
164 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
165 					struct tcf_result *);
166 	int			(*init)(struct tcf_proto*);
167 	void			(*destroy)(struct tcf_proto*);
168 
169 	unsigned long		(*get)(struct tcf_proto*, u32 handle);
170 	void			(*put)(struct tcf_proto*, unsigned long);
171 	int			(*change)(struct tcf_proto*, unsigned long,
172 					u32 handle, struct nlattr **,
173 					unsigned long *);
174 	int			(*delete)(struct tcf_proto*, unsigned long);
175 	void			(*walk)(struct tcf_proto*, struct tcf_walker *arg);
176 
177 	/* rtnetlink specific */
178 	int			(*dump)(struct tcf_proto*, unsigned long,
179 					struct sk_buff *skb, struct tcmsg*);
180 
181 	struct module		*owner;
182 };
183 
184 struct tcf_proto {
185 	/* Fast access part */
186 	struct tcf_proto	*next;
187 	void			*root;
188 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
189 					struct tcf_result *);
190 	__be16			protocol;
191 
192 	/* All the rest */
193 	u32			prio;
194 	u32			classid;
195 	struct Qdisc		*q;
196 	void			*data;
197 	struct tcf_proto_ops	*ops;
198 };
199 
200 struct qdisc_skb_cb {
201 	unsigned int		pkt_len;
202 	char			data[];
203 };
204 
205 static inline int qdisc_qlen(struct Qdisc *q)
206 {
207 	return q->q.qlen;
208 }
209 
210 static inline struct qdisc_skb_cb *qdisc_skb_cb(struct sk_buff *skb)
211 {
212 	return (struct qdisc_skb_cb *)skb->cb;
213 }
214 
215 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
216 {
217 	return &qdisc->q.lock;
218 }
219 
220 static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
221 {
222 	return qdisc->dev_queue->qdisc;
223 }
224 
225 static inline struct Qdisc *qdisc_root_sleeping(struct Qdisc *qdisc)
226 {
227 	return qdisc->dev_queue->qdisc_sleeping;
228 }
229 
230 /* The qdisc root lock is a mechanism by which to top level
231  * of a qdisc tree can be locked from any qdisc node in the
232  * forest.  This allows changing the configuration of some
233  * aspect of the qdisc tree while blocking out asynchronous
234  * qdisc access in the packet processing paths.
235  *
236  * It is only legal to do this when the root will not change
237  * on us.  Otherwise we'll potentially lock the wrong qdisc
238  * root.  This is enforced by holding the RTNL semaphore, which
239  * all users of this lock accessor must do.
240  */
241 static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
242 {
243 	struct Qdisc *root = qdisc_root(qdisc);
244 
245 	ASSERT_RTNL();
246 	return qdisc_lock(root);
247 }
248 
249 static inline spinlock_t *qdisc_root_sleeping_lock(struct Qdisc *qdisc)
250 {
251 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
252 
253 	ASSERT_RTNL();
254 	return qdisc_lock(root);
255 }
256 
257 static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
258 {
259 	return qdisc->dev_queue->dev;
260 }
261 
262 static inline void sch_tree_lock(struct Qdisc *q)
263 {
264 	spin_lock_bh(qdisc_root_sleeping_lock(q));
265 }
266 
267 static inline void sch_tree_unlock(struct Qdisc *q)
268 {
269 	spin_unlock_bh(qdisc_root_sleeping_lock(q));
270 }
271 
272 #define tcf_tree_lock(tp)	sch_tree_lock((tp)->q)
273 #define tcf_tree_unlock(tp)	sch_tree_unlock((tp)->q)
274 
275 extern struct Qdisc noop_qdisc;
276 extern struct Qdisc_ops noop_qdisc_ops;
277 extern struct Qdisc_ops pfifo_fast_ops;
278 extern struct Qdisc_ops mq_qdisc_ops;
279 
280 struct Qdisc_class_common {
281 	u32			classid;
282 	struct hlist_node	hnode;
283 };
284 
285 struct Qdisc_class_hash {
286 	struct hlist_head	*hash;
287 	unsigned int		hashsize;
288 	unsigned int		hashmask;
289 	unsigned int		hashelems;
290 };
291 
292 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
293 {
294 	id ^= id >> 8;
295 	id ^= id >> 4;
296 	return id & mask;
297 }
298 
299 static inline struct Qdisc_class_common *
300 qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
301 {
302 	struct Qdisc_class_common *cl;
303 	struct hlist_node *n;
304 	unsigned int h;
305 
306 	h = qdisc_class_hash(id, hash->hashmask);
307 	hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
308 		if (cl->classid == id)
309 			return cl;
310 	}
311 	return NULL;
312 }
313 
314 extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
315 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
316 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
317 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
318 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
319 
320 extern void dev_init_scheduler(struct net_device *dev);
321 extern void dev_shutdown(struct net_device *dev);
322 extern void dev_activate(struct net_device *dev);
323 extern void dev_deactivate(struct net_device *dev);
324 extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
325 				     struct Qdisc *qdisc);
326 extern void qdisc_reset(struct Qdisc *qdisc);
327 extern void qdisc_destroy(struct Qdisc *qdisc);
328 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
329 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
330 				 struct Qdisc_ops *ops);
331 extern struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
332 				       struct Qdisc_ops *ops, u32 parentid);
333 extern void qdisc_calculate_pkt_len(struct sk_buff *skb,
334 				   struct qdisc_size_table *stab);
335 extern void tcf_destroy(struct tcf_proto *tp);
336 extern void tcf_destroy_chain(struct tcf_proto **fl);
337 
338 /* Reset all TX qdiscs greater then index of a device.  */
339 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
340 {
341 	struct Qdisc *qdisc;
342 
343 	for (; i < dev->num_tx_queues; i++) {
344 		qdisc = netdev_get_tx_queue(dev, i)->qdisc;
345 		if (qdisc) {
346 			spin_lock_bh(qdisc_lock(qdisc));
347 			qdisc_reset(qdisc);
348 			spin_unlock_bh(qdisc_lock(qdisc));
349 		}
350 	}
351 }
352 
353 static inline void qdisc_reset_all_tx(struct net_device *dev)
354 {
355 	qdisc_reset_all_tx_gt(dev, 0);
356 }
357 
358 /* Are all TX queues of the device empty?  */
359 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
360 {
361 	unsigned int i;
362 	for (i = 0; i < dev->num_tx_queues; i++) {
363 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
364 		const struct Qdisc *q = txq->qdisc;
365 
366 		if (q->q.qlen)
367 			return false;
368 	}
369 	return true;
370 }
371 
372 /* Are any of the TX qdiscs changing?  */
373 static inline bool qdisc_tx_changing(struct net_device *dev)
374 {
375 	unsigned int i;
376 	for (i = 0; i < dev->num_tx_queues; i++) {
377 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
378 		if (txq->qdisc != txq->qdisc_sleeping)
379 			return true;
380 	}
381 	return false;
382 }
383 
384 /* Is the device using the noop qdisc on all queues?  */
385 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
386 {
387 	unsigned int i;
388 	for (i = 0; i < dev->num_tx_queues; i++) {
389 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
390 		if (txq->qdisc != &noop_qdisc)
391 			return false;
392 	}
393 	return true;
394 }
395 
396 static inline unsigned int qdisc_pkt_len(struct sk_buff *skb)
397 {
398 	return qdisc_skb_cb(skb)->pkt_len;
399 }
400 
401 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
402 enum net_xmit_qdisc_t {
403 	__NET_XMIT_STOLEN = 0x00010000,
404 	__NET_XMIT_BYPASS = 0x00020000,
405 };
406 
407 #ifdef CONFIG_NET_CLS_ACT
408 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
409 #else
410 #define net_xmit_drop_count(e)	(1)
411 #endif
412 
413 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
414 {
415 #ifdef CONFIG_NET_SCHED
416 	if (sch->stab)
417 		qdisc_calculate_pkt_len(skb, sch->stab);
418 #endif
419 	return sch->enqueue(skb, sch);
420 }
421 
422 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
423 {
424 	qdisc_skb_cb(skb)->pkt_len = skb->len;
425 	return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
426 }
427 
428 static inline void __qdisc_update_bstats(struct Qdisc *sch, unsigned int len)
429 {
430 	sch->bstats.bytes += len;
431 	sch->bstats.packets++;
432 }
433 
434 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
435 				       struct sk_buff_head *list)
436 {
437 	__skb_queue_tail(list, skb);
438 	sch->qstats.backlog += qdisc_pkt_len(skb);
439 	__qdisc_update_bstats(sch, qdisc_pkt_len(skb));
440 
441 	return NET_XMIT_SUCCESS;
442 }
443 
444 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
445 {
446 	return __qdisc_enqueue_tail(skb, sch, &sch->q);
447 }
448 
449 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
450 						   struct sk_buff_head *list)
451 {
452 	struct sk_buff *skb = __skb_dequeue(list);
453 
454 	if (likely(skb != NULL))
455 		sch->qstats.backlog -= qdisc_pkt_len(skb);
456 
457 	return skb;
458 }
459 
460 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
461 {
462 	return __qdisc_dequeue_head(sch, &sch->q);
463 }
464 
465 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
466 					      struct sk_buff_head *list)
467 {
468 	struct sk_buff *skb = __qdisc_dequeue_head(sch, list);
469 
470 	if (likely(skb != NULL)) {
471 		unsigned int len = qdisc_pkt_len(skb);
472 		kfree_skb(skb);
473 		return len;
474 	}
475 
476 	return 0;
477 }
478 
479 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch)
480 {
481 	return __qdisc_queue_drop_head(sch, &sch->q);
482 }
483 
484 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
485 						   struct sk_buff_head *list)
486 {
487 	struct sk_buff *skb = __skb_dequeue_tail(list);
488 
489 	if (likely(skb != NULL))
490 		sch->qstats.backlog -= qdisc_pkt_len(skb);
491 
492 	return skb;
493 }
494 
495 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
496 {
497 	return __qdisc_dequeue_tail(sch, &sch->q);
498 }
499 
500 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
501 {
502 	return skb_peek(&sch->q);
503 }
504 
505 /* generic pseudo peek method for non-work-conserving qdisc */
506 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
507 {
508 	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
509 	if (!sch->gso_skb) {
510 		sch->gso_skb = sch->dequeue(sch);
511 		if (sch->gso_skb)
512 			/* it's still part of the queue */
513 			sch->q.qlen++;
514 	}
515 
516 	return sch->gso_skb;
517 }
518 
519 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
520 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
521 {
522 	struct sk_buff *skb = sch->gso_skb;
523 
524 	if (skb) {
525 		sch->gso_skb = NULL;
526 		sch->q.qlen--;
527 	} else {
528 		skb = sch->dequeue(sch);
529 	}
530 
531 	return skb;
532 }
533 
534 static inline void __qdisc_reset_queue(struct Qdisc *sch,
535 				       struct sk_buff_head *list)
536 {
537 	/*
538 	 * We do not know the backlog in bytes of this list, it
539 	 * is up to the caller to correct it
540 	 */
541 	__skb_queue_purge(list);
542 }
543 
544 static inline void qdisc_reset_queue(struct Qdisc *sch)
545 {
546 	__qdisc_reset_queue(sch, &sch->q);
547 	sch->qstats.backlog = 0;
548 }
549 
550 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
551 					      struct sk_buff_head *list)
552 {
553 	struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
554 
555 	if (likely(skb != NULL)) {
556 		unsigned int len = qdisc_pkt_len(skb);
557 		kfree_skb(skb);
558 		return len;
559 	}
560 
561 	return 0;
562 }
563 
564 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
565 {
566 	return __qdisc_queue_drop(sch, &sch->q);
567 }
568 
569 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
570 {
571 	kfree_skb(skb);
572 	sch->qstats.drops++;
573 
574 	return NET_XMIT_DROP;
575 }
576 
577 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
578 {
579 	sch->qstats.drops++;
580 
581 #ifdef CONFIG_NET_CLS_ACT
582 	if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
583 		goto drop;
584 
585 	return NET_XMIT_SUCCESS;
586 
587 drop:
588 #endif
589 	kfree_skb(skb);
590 	return NET_XMIT_DROP;
591 }
592 
593 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
594    long it will take to send a packet given its size.
595  */
596 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
597 {
598 	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
599 	if (slot < 0)
600 		slot = 0;
601 	slot >>= rtab->rate.cell_log;
602 	if (slot > 255)
603 		return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
604 	return rtab->data[slot];
605 }
606 
607 #ifdef CONFIG_NET_CLS_ACT
608 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask,
609 					    int action)
610 {
611 	struct sk_buff *n;
612 
613 	if ((action == TC_ACT_STOLEN || action == TC_ACT_QUEUED) &&
614 	    !skb_shared(skb))
615 		n = skb_get(skb);
616 	else
617 		n = skb_clone(skb, gfp_mask);
618 
619 	if (n) {
620 		n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
621 		n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
622 		n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
623 	}
624 	return n;
625 }
626 #endif
627 
628 #endif
629