xref: /linux/include/net/sch_generic.h (revision bcefe12eff5dca6fdfa94ed85e5bee66380d5cd9)
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 {
20 	struct tc_ratespec rate;
21 	u32		data[256];
22 	struct qdisc_rate_table *next;
23 	int		refcnt;
24 };
25 
26 enum qdisc_state_t
27 {
28 	__QDISC_STATE_RUNNING,
29 	__QDISC_STATE_SCHED,
30 	__QDISC_STATE_DEACTIVATED,
31 };
32 
33 struct qdisc_size_table {
34 	struct list_head	list;
35 	struct tc_sizespec	szopts;
36 	int			refcnt;
37 	u16			data[];
38 };
39 
40 struct Qdisc
41 {
42 	int 			(*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
43 	struct sk_buff *	(*dequeue)(struct Qdisc *dev);
44 	unsigned		flags;
45 #define TCQ_F_BUILTIN		1
46 #define TCQ_F_THROTTLED		2
47 #define TCQ_F_INGRESS		4
48 #define TCQ_F_CAN_BYPASS	8
49 #define TCQ_F_MQROOT		16
50 #define TCQ_F_WARN_NONWC	(1 << 16)
51 	int			padded;
52 	struct Qdisc_ops	*ops;
53 	struct qdisc_size_table	*stab;
54 	struct list_head	list;
55 	u32			handle;
56 	u32			parent;
57 	atomic_t		refcnt;
58 	struct gnet_stats_rate_est	rate_est;
59 	int			(*reshape_fail)(struct sk_buff *skb,
60 					struct Qdisc *q);
61 
62 	void			*u32_node;
63 
64 	/* This field is deprecated, but it is still used by CBQ
65 	 * and it will live until better solution will be invented.
66 	 */
67 	struct Qdisc		*__parent;
68 	struct netdev_queue	*dev_queue;
69 	struct Qdisc		*next_sched;
70 
71 	struct sk_buff		*gso_skb;
72 	/*
73 	 * For performance sake on SMP, we put highly modified fields at the end
74 	 */
75 	unsigned long		state;
76 	struct sk_buff_head	q;
77 	struct gnet_stats_basic_packed bstats;
78 	struct gnet_stats_queue	qstats;
79 };
80 
81 struct Qdisc_class_ops
82 {
83 	/* Child qdisc manipulation */
84 	struct netdev_queue *	(*select_queue)(struct Qdisc *, struct tcmsg *);
85 	int			(*graft)(struct Qdisc *, unsigned long cl,
86 					struct Qdisc *, struct Qdisc **);
87 	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
88 	void			(*qlen_notify)(struct Qdisc *, unsigned long);
89 
90 	/* Class manipulation routines */
91 	unsigned long		(*get)(struct Qdisc *, u32 classid);
92 	void			(*put)(struct Qdisc *, unsigned long);
93 	int			(*change)(struct Qdisc *, u32, u32,
94 					struct nlattr **, unsigned long *);
95 	int			(*delete)(struct Qdisc *, unsigned long);
96 	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
97 
98 	/* Filter manipulation */
99 	struct tcf_proto **	(*tcf_chain)(struct Qdisc *, unsigned long);
100 	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
101 					u32 classid);
102 	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
103 
104 	/* rtnetlink specific */
105 	int			(*dump)(struct Qdisc *, unsigned long,
106 					struct sk_buff *skb, struct tcmsg*);
107 	int			(*dump_stats)(struct Qdisc *, unsigned long,
108 					struct gnet_dump *);
109 };
110 
111 struct Qdisc_ops
112 {
113 	struct Qdisc_ops	*next;
114 	const struct Qdisc_class_ops	*cl_ops;
115 	char			id[IFNAMSIZ];
116 	int			priv_size;
117 
118 	int 			(*enqueue)(struct sk_buff *, struct Qdisc *);
119 	struct sk_buff *	(*dequeue)(struct Qdisc *);
120 	struct sk_buff *	(*peek)(struct Qdisc *);
121 	unsigned int		(*drop)(struct Qdisc *);
122 
123 	int			(*init)(struct Qdisc *, struct nlattr *arg);
124 	void			(*reset)(struct Qdisc *);
125 	void			(*destroy)(struct Qdisc *);
126 	int			(*change)(struct Qdisc *, struct nlattr *arg);
127 	void			(*attach)(struct Qdisc *);
128 
129 	int			(*dump)(struct Qdisc *, struct sk_buff *);
130 	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
131 
132 	struct module		*owner;
133 };
134 
135 
136 struct tcf_result
137 {
138 	unsigned long	class;
139 	u32		classid;
140 };
141 
142 struct tcf_proto_ops
143 {
144 	struct tcf_proto_ops	*next;
145 	char			kind[IFNAMSIZ];
146 
147 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
148 					struct tcf_result *);
149 	int			(*init)(struct tcf_proto*);
150 	void			(*destroy)(struct tcf_proto*);
151 
152 	unsigned long		(*get)(struct tcf_proto*, u32 handle);
153 	void			(*put)(struct tcf_proto*, unsigned long);
154 	int			(*change)(struct tcf_proto*, unsigned long,
155 					u32 handle, struct nlattr **,
156 					unsigned long *);
157 	int			(*delete)(struct tcf_proto*, unsigned long);
158 	void			(*walk)(struct tcf_proto*, struct tcf_walker *arg);
159 
160 	/* rtnetlink specific */
161 	int			(*dump)(struct tcf_proto*, unsigned long,
162 					struct sk_buff *skb, struct tcmsg*);
163 
164 	struct module		*owner;
165 };
166 
167 struct tcf_proto
168 {
169 	/* Fast access part */
170 	struct tcf_proto	*next;
171 	void			*root;
172 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
173 					struct tcf_result *);
174 	__be16			protocol;
175 
176 	/* All the rest */
177 	u32			prio;
178 	u32			classid;
179 	struct Qdisc		*q;
180 	void			*data;
181 	struct tcf_proto_ops	*ops;
182 };
183 
184 struct qdisc_skb_cb {
185 	unsigned int		pkt_len;
186 	char			data[];
187 };
188 
189 static inline int qdisc_qlen(struct Qdisc *q)
190 {
191 	return q->q.qlen;
192 }
193 
194 static inline struct qdisc_skb_cb *qdisc_skb_cb(struct sk_buff *skb)
195 {
196 	return (struct qdisc_skb_cb *)skb->cb;
197 }
198 
199 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
200 {
201 	return &qdisc->q.lock;
202 }
203 
204 static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
205 {
206 	return qdisc->dev_queue->qdisc;
207 }
208 
209 static inline struct Qdisc *qdisc_root_sleeping(struct Qdisc *qdisc)
210 {
211 	return qdisc->dev_queue->qdisc_sleeping;
212 }
213 
214 /* The qdisc root lock is a mechanism by which to top level
215  * of a qdisc tree can be locked from any qdisc node in the
216  * forest.  This allows changing the configuration of some
217  * aspect of the qdisc tree while blocking out asynchronous
218  * qdisc access in the packet processing paths.
219  *
220  * It is only legal to do this when the root will not change
221  * on us.  Otherwise we'll potentially lock the wrong qdisc
222  * root.  This is enforced by holding the RTNL semaphore, which
223  * all users of this lock accessor must do.
224  */
225 static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
226 {
227 	struct Qdisc *root = qdisc_root(qdisc);
228 
229 	ASSERT_RTNL();
230 	return qdisc_lock(root);
231 }
232 
233 static inline spinlock_t *qdisc_root_sleeping_lock(struct Qdisc *qdisc)
234 {
235 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
236 
237 	ASSERT_RTNL();
238 	return qdisc_lock(root);
239 }
240 
241 static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
242 {
243 	return qdisc->dev_queue->dev;
244 }
245 
246 static inline void sch_tree_lock(struct Qdisc *q)
247 {
248 	spin_lock_bh(qdisc_root_sleeping_lock(q));
249 }
250 
251 static inline void sch_tree_unlock(struct Qdisc *q)
252 {
253 	spin_unlock_bh(qdisc_root_sleeping_lock(q));
254 }
255 
256 #define tcf_tree_lock(tp)	sch_tree_lock((tp)->q)
257 #define tcf_tree_unlock(tp)	sch_tree_unlock((tp)->q)
258 
259 extern struct Qdisc noop_qdisc;
260 extern struct Qdisc_ops noop_qdisc_ops;
261 extern struct Qdisc_ops pfifo_fast_ops;
262 extern struct Qdisc_ops mq_qdisc_ops;
263 
264 struct Qdisc_class_common
265 {
266 	u32			classid;
267 	struct hlist_node	hnode;
268 };
269 
270 struct Qdisc_class_hash
271 {
272 	struct hlist_head	*hash;
273 	unsigned int		hashsize;
274 	unsigned int		hashmask;
275 	unsigned int		hashelems;
276 };
277 
278 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
279 {
280 	id ^= id >> 8;
281 	id ^= id >> 4;
282 	return id & mask;
283 }
284 
285 static inline struct Qdisc_class_common *
286 qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
287 {
288 	struct Qdisc_class_common *cl;
289 	struct hlist_node *n;
290 	unsigned int h;
291 
292 	h = qdisc_class_hash(id, hash->hashmask);
293 	hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
294 		if (cl->classid == id)
295 			return cl;
296 	}
297 	return NULL;
298 }
299 
300 extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
301 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
302 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
303 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
304 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
305 
306 extern void dev_init_scheduler(struct net_device *dev);
307 extern void dev_shutdown(struct net_device *dev);
308 extern void dev_activate(struct net_device *dev);
309 extern void dev_deactivate(struct net_device *dev);
310 extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
311 				     struct Qdisc *qdisc);
312 extern void qdisc_reset(struct Qdisc *qdisc);
313 extern void qdisc_destroy(struct Qdisc *qdisc);
314 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
315 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
316 				 struct Qdisc_ops *ops);
317 extern struct Qdisc *qdisc_create_dflt(struct net_device *dev,
318 				       struct netdev_queue *dev_queue,
319 				       struct Qdisc_ops *ops, u32 parentid);
320 extern void qdisc_calculate_pkt_len(struct sk_buff *skb,
321 				   struct qdisc_size_table *stab);
322 extern void tcf_destroy(struct tcf_proto *tp);
323 extern void tcf_destroy_chain(struct tcf_proto **fl);
324 
325 /* Reset all TX qdiscs of a device.  */
326 static inline void qdisc_reset_all_tx(struct net_device *dev)
327 {
328 	unsigned int i;
329 	for (i = 0; i < dev->num_tx_queues; i++)
330 		qdisc_reset(netdev_get_tx_queue(dev, i)->qdisc);
331 }
332 
333 /* Are all TX queues of the device empty?  */
334 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
335 {
336 	unsigned int i;
337 	for (i = 0; i < dev->num_tx_queues; i++) {
338 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
339 		const struct Qdisc *q = txq->qdisc;
340 
341 		if (q->q.qlen)
342 			return false;
343 	}
344 	return true;
345 }
346 
347 /* Are any of the TX qdiscs changing?  */
348 static inline bool qdisc_tx_changing(struct net_device *dev)
349 {
350 	unsigned int i;
351 	for (i = 0; i < dev->num_tx_queues; i++) {
352 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
353 		if (txq->qdisc != txq->qdisc_sleeping)
354 			return true;
355 	}
356 	return false;
357 }
358 
359 /* Is the device using the noop qdisc on all queues?  */
360 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
361 {
362 	unsigned int i;
363 	for (i = 0; i < dev->num_tx_queues; i++) {
364 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
365 		if (txq->qdisc != &noop_qdisc)
366 			return false;
367 	}
368 	return true;
369 }
370 
371 static inline unsigned int qdisc_pkt_len(struct sk_buff *skb)
372 {
373 	return qdisc_skb_cb(skb)->pkt_len;
374 }
375 
376 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
377 enum net_xmit_qdisc_t {
378 	__NET_XMIT_STOLEN = 0x00010000,
379 	__NET_XMIT_BYPASS = 0x00020000,
380 };
381 
382 #ifdef CONFIG_NET_CLS_ACT
383 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
384 #else
385 #define net_xmit_drop_count(e)	(1)
386 #endif
387 
388 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
389 {
390 #ifdef CONFIG_NET_SCHED
391 	if (sch->stab)
392 		qdisc_calculate_pkt_len(skb, sch->stab);
393 #endif
394 	return sch->enqueue(skb, sch);
395 }
396 
397 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
398 {
399 	qdisc_skb_cb(skb)->pkt_len = skb->len;
400 	return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
401 }
402 
403 static inline void __qdisc_update_bstats(struct Qdisc *sch, unsigned int len)
404 {
405 	sch->bstats.bytes += len;
406 	sch->bstats.packets++;
407 }
408 
409 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
410 				       struct sk_buff_head *list)
411 {
412 	__skb_queue_tail(list, skb);
413 	sch->qstats.backlog += qdisc_pkt_len(skb);
414 	__qdisc_update_bstats(sch, qdisc_pkt_len(skb));
415 
416 	return NET_XMIT_SUCCESS;
417 }
418 
419 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
420 {
421 	return __qdisc_enqueue_tail(skb, sch, &sch->q);
422 }
423 
424 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
425 						   struct sk_buff_head *list)
426 {
427 	struct sk_buff *skb = __skb_dequeue(list);
428 
429 	if (likely(skb != NULL))
430 		sch->qstats.backlog -= qdisc_pkt_len(skb);
431 
432 	return skb;
433 }
434 
435 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
436 {
437 	return __qdisc_dequeue_head(sch, &sch->q);
438 }
439 
440 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
441 						   struct sk_buff_head *list)
442 {
443 	struct sk_buff *skb = __skb_dequeue_tail(list);
444 
445 	if (likely(skb != NULL))
446 		sch->qstats.backlog -= qdisc_pkt_len(skb);
447 
448 	return skb;
449 }
450 
451 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
452 {
453 	return __qdisc_dequeue_tail(sch, &sch->q);
454 }
455 
456 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
457 {
458 	return skb_peek(&sch->q);
459 }
460 
461 /* generic pseudo peek method for non-work-conserving qdisc */
462 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
463 {
464 	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
465 	if (!sch->gso_skb) {
466 		sch->gso_skb = sch->dequeue(sch);
467 		if (sch->gso_skb)
468 			/* it's still part of the queue */
469 			sch->q.qlen++;
470 	}
471 
472 	return sch->gso_skb;
473 }
474 
475 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
476 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
477 {
478 	struct sk_buff *skb = sch->gso_skb;
479 
480 	if (skb) {
481 		sch->gso_skb = NULL;
482 		sch->q.qlen--;
483 	} else {
484 		skb = sch->dequeue(sch);
485 	}
486 
487 	return skb;
488 }
489 
490 static inline void __qdisc_reset_queue(struct Qdisc *sch,
491 				       struct sk_buff_head *list)
492 {
493 	/*
494 	 * We do not know the backlog in bytes of this list, it
495 	 * is up to the caller to correct it
496 	 */
497 	__skb_queue_purge(list);
498 }
499 
500 static inline void qdisc_reset_queue(struct Qdisc *sch)
501 {
502 	__qdisc_reset_queue(sch, &sch->q);
503 	sch->qstats.backlog = 0;
504 }
505 
506 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
507 					      struct sk_buff_head *list)
508 {
509 	struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
510 
511 	if (likely(skb != NULL)) {
512 		unsigned int len = qdisc_pkt_len(skb);
513 		kfree_skb(skb);
514 		return len;
515 	}
516 
517 	return 0;
518 }
519 
520 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
521 {
522 	return __qdisc_queue_drop(sch, &sch->q);
523 }
524 
525 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
526 {
527 	kfree_skb(skb);
528 	sch->qstats.drops++;
529 
530 	return NET_XMIT_DROP;
531 }
532 
533 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
534 {
535 	sch->qstats.drops++;
536 
537 #ifdef CONFIG_NET_CLS_ACT
538 	if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
539 		goto drop;
540 
541 	return NET_XMIT_SUCCESS;
542 
543 drop:
544 #endif
545 	kfree_skb(skb);
546 	return NET_XMIT_DROP;
547 }
548 
549 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
550    long it will take to send a packet given its size.
551  */
552 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
553 {
554 	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
555 	if (slot < 0)
556 		slot = 0;
557 	slot >>= rtab->rate.cell_log;
558 	if (slot > 255)
559 		return (rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]);
560 	return rtab->data[slot];
561 }
562 
563 #ifdef CONFIG_NET_CLS_ACT
564 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask)
565 {
566 	struct sk_buff *n = skb_clone(skb, gfp_mask);
567 
568 	if (n) {
569 		n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
570 		n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
571 		n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
572 	}
573 	return n;
574 }
575 #endif
576 
577 #endif
578