xref: /linux/include/net/sch_generic.h (revision 7265706c8fd57722f622f336ec110cb35f83e739)
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 };
31 
32 struct qdisc_size_table {
33 	struct list_head	list;
34 	struct tc_sizespec	szopts;
35 	int			refcnt;
36 	u16			data[];
37 };
38 
39 struct Qdisc
40 {
41 	int 			(*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
42 	struct sk_buff *	(*dequeue)(struct Qdisc *dev);
43 	unsigned		flags;
44 #define TCQ_F_BUILTIN	1
45 #define TCQ_F_THROTTLED	2
46 #define TCQ_F_INGRESS	4
47 	int			padded;
48 	struct Qdisc_ops	*ops;
49 	struct qdisc_size_table	*stab;
50 	u32			handle;
51 	u32			parent;
52 	atomic_t		refcnt;
53 	unsigned long		state;
54 	struct sk_buff		*gso_skb;
55 	struct sk_buff_head	q;
56 	struct netdev_queue	*dev_queue;
57 	struct Qdisc		*next_sched;
58 	struct list_head	list;
59 
60 	struct gnet_stats_basic	bstats;
61 	struct gnet_stats_queue	qstats;
62 	struct gnet_stats_rate_est	rate_est;
63 	struct rcu_head 	q_rcu;
64 	int			(*reshape_fail)(struct sk_buff *skb,
65 					struct Qdisc *q);
66 
67 	void			*u32_node;
68 
69 	/* This field is deprecated, but it is still used by CBQ
70 	 * and it will live until better solution will be invented.
71 	 */
72 	struct Qdisc		*__parent;
73 };
74 
75 struct Qdisc_class_ops
76 {
77 	/* Child qdisc manipulation */
78 	int			(*graft)(struct Qdisc *, unsigned long cl,
79 					struct Qdisc *, struct Qdisc **);
80 	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
81 	void			(*qlen_notify)(struct Qdisc *, unsigned long);
82 
83 	/* Class manipulation routines */
84 	unsigned long		(*get)(struct Qdisc *, u32 classid);
85 	void			(*put)(struct Qdisc *, unsigned long);
86 	int			(*change)(struct Qdisc *, u32, u32,
87 					struct nlattr **, unsigned long *);
88 	int			(*delete)(struct Qdisc *, unsigned long);
89 	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
90 
91 	/* Filter manipulation */
92 	struct tcf_proto **	(*tcf_chain)(struct Qdisc *, unsigned long);
93 	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
94 					u32 classid);
95 	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
96 
97 	/* rtnetlink specific */
98 	int			(*dump)(struct Qdisc *, unsigned long,
99 					struct sk_buff *skb, struct tcmsg*);
100 	int			(*dump_stats)(struct Qdisc *, unsigned long,
101 					struct gnet_dump *);
102 };
103 
104 struct Qdisc_ops
105 {
106 	struct Qdisc_ops	*next;
107 	const struct Qdisc_class_ops	*cl_ops;
108 	char			id[IFNAMSIZ];
109 	int			priv_size;
110 
111 	int 			(*enqueue)(struct sk_buff *, struct Qdisc *);
112 	struct sk_buff *	(*dequeue)(struct Qdisc *);
113 	int 			(*requeue)(struct sk_buff *, struct Qdisc *);
114 	unsigned int		(*drop)(struct Qdisc *);
115 
116 	int			(*init)(struct Qdisc *, struct nlattr *arg);
117 	void			(*reset)(struct Qdisc *);
118 	void			(*destroy)(struct Qdisc *);
119 	int			(*change)(struct Qdisc *, struct nlattr *arg);
120 
121 	int			(*dump)(struct Qdisc *, struct sk_buff *);
122 	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
123 
124 	struct module		*owner;
125 };
126 
127 
128 struct tcf_result
129 {
130 	unsigned long	class;
131 	u32		classid;
132 };
133 
134 struct tcf_proto_ops
135 {
136 	struct tcf_proto_ops	*next;
137 	char			kind[IFNAMSIZ];
138 
139 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
140 					struct tcf_result *);
141 	int			(*init)(struct tcf_proto*);
142 	void			(*destroy)(struct tcf_proto*);
143 
144 	unsigned long		(*get)(struct tcf_proto*, u32 handle);
145 	void			(*put)(struct tcf_proto*, unsigned long);
146 	int			(*change)(struct tcf_proto*, unsigned long,
147 					u32 handle, struct nlattr **,
148 					unsigned long *);
149 	int			(*delete)(struct tcf_proto*, unsigned long);
150 	void			(*walk)(struct tcf_proto*, struct tcf_walker *arg);
151 
152 	/* rtnetlink specific */
153 	int			(*dump)(struct tcf_proto*, unsigned long,
154 					struct sk_buff *skb, struct tcmsg*);
155 
156 	struct module		*owner;
157 };
158 
159 struct tcf_proto
160 {
161 	/* Fast access part */
162 	struct tcf_proto	*next;
163 	void			*root;
164 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
165 					struct tcf_result *);
166 	__be16			protocol;
167 
168 	/* All the rest */
169 	u32			prio;
170 	u32			classid;
171 	struct Qdisc		*q;
172 	void			*data;
173 	struct tcf_proto_ops	*ops;
174 };
175 
176 struct qdisc_skb_cb {
177 	unsigned int		pkt_len;
178 	char			data[];
179 };
180 
181 static inline struct qdisc_skb_cb *qdisc_skb_cb(struct sk_buff *skb)
182 {
183 	return (struct qdisc_skb_cb *)skb->cb;
184 }
185 
186 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
187 {
188 	return &qdisc->q.lock;
189 }
190 
191 static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
192 {
193 	return qdisc->dev_queue->qdisc;
194 }
195 
196 /* The qdisc root lock is a mechanism by which to top level
197  * of a qdisc tree can be locked from any qdisc node in the
198  * forest.  This allows changing the configuration of some
199  * aspect of the qdisc tree while blocking out asynchronous
200  * qdisc access in the packet processing paths.
201  *
202  * It is only legal to do this when the root will not change
203  * on us.  Otherwise we'll potentially lock the wrong qdisc
204  * root.  This is enforced by holding the RTNL semaphore, which
205  * all users of this lock accessor must do.
206  */
207 static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
208 {
209 	struct Qdisc *root = qdisc_root(qdisc);
210 
211 	ASSERT_RTNL();
212 	return qdisc_lock(root);
213 }
214 
215 static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
216 {
217 	return qdisc->dev_queue->dev;
218 }
219 
220 static inline void sch_tree_lock(struct Qdisc *q)
221 {
222 	spin_lock_bh(qdisc_root_lock(q));
223 }
224 
225 static inline void sch_tree_unlock(struct Qdisc *q)
226 {
227 	spin_unlock_bh(qdisc_root_lock(q));
228 }
229 
230 #define tcf_tree_lock(tp)	sch_tree_lock((tp)->q)
231 #define tcf_tree_unlock(tp)	sch_tree_unlock((tp)->q)
232 
233 extern struct Qdisc noop_qdisc;
234 extern struct Qdisc_ops noop_qdisc_ops;
235 
236 struct Qdisc_class_common
237 {
238 	u32			classid;
239 	struct hlist_node	hnode;
240 };
241 
242 struct Qdisc_class_hash
243 {
244 	struct hlist_head	*hash;
245 	unsigned int		hashsize;
246 	unsigned int		hashmask;
247 	unsigned int		hashelems;
248 };
249 
250 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
251 {
252 	id ^= id >> 8;
253 	id ^= id >> 4;
254 	return id & mask;
255 }
256 
257 static inline struct Qdisc_class_common *
258 qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
259 {
260 	struct Qdisc_class_common *cl;
261 	struct hlist_node *n;
262 	unsigned int h;
263 
264 	h = qdisc_class_hash(id, hash->hashmask);
265 	hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
266 		if (cl->classid == id)
267 			return cl;
268 	}
269 	return NULL;
270 }
271 
272 extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
273 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
274 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
275 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
276 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
277 
278 extern void dev_init_scheduler(struct net_device *dev);
279 extern void dev_shutdown(struct net_device *dev);
280 extern void dev_activate(struct net_device *dev);
281 extern void dev_deactivate(struct net_device *dev);
282 extern void qdisc_reset(struct Qdisc *qdisc);
283 extern void qdisc_destroy(struct Qdisc *qdisc);
284 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
285 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
286 				 struct Qdisc_ops *ops);
287 extern struct Qdisc *qdisc_create_dflt(struct net_device *dev,
288 				       struct netdev_queue *dev_queue,
289 				       struct Qdisc_ops *ops, u32 parentid);
290 extern void qdisc_calculate_pkt_len(struct sk_buff *skb,
291 				   struct qdisc_size_table *stab);
292 extern void tcf_destroy(struct tcf_proto *tp);
293 extern void tcf_destroy_chain(struct tcf_proto **fl);
294 
295 /* Reset all TX qdiscs of a device.  */
296 static inline void qdisc_reset_all_tx(struct net_device *dev)
297 {
298 	unsigned int i;
299 	for (i = 0; i < dev->num_tx_queues; i++)
300 		qdisc_reset(netdev_get_tx_queue(dev, i)->qdisc);
301 }
302 
303 /* Are all TX queues of the device empty?  */
304 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
305 {
306 	unsigned int i;
307 	for (i = 0; i < dev->num_tx_queues; i++) {
308 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
309 		const struct Qdisc *q = txq->qdisc;
310 
311 		if (q->q.qlen)
312 			return false;
313 	}
314 	return true;
315 }
316 
317 /* Are any of the TX qdiscs changing?  */
318 static inline bool qdisc_tx_changing(struct net_device *dev)
319 {
320 	unsigned int i;
321 	for (i = 0; i < dev->num_tx_queues; i++) {
322 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
323 		if (txq->qdisc != txq->qdisc_sleeping)
324 			return true;
325 	}
326 	return false;
327 }
328 
329 /* Is the device using the noop qdisc on all queues?  */
330 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
331 {
332 	unsigned int i;
333 	for (i = 0; i < dev->num_tx_queues; i++) {
334 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
335 		if (txq->qdisc != &noop_qdisc)
336 			return false;
337 	}
338 	return true;
339 }
340 
341 static inline unsigned int qdisc_pkt_len(struct sk_buff *skb)
342 {
343 	return qdisc_skb_cb(skb)->pkt_len;
344 }
345 
346 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
347 enum net_xmit_qdisc_t {
348 	__NET_XMIT_STOLEN = 0x00010000,
349 	__NET_XMIT_BYPASS = 0x00020000,
350 };
351 
352 #ifdef CONFIG_NET_CLS_ACT
353 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
354 #else
355 #define net_xmit_drop_count(e)	(1)
356 #endif
357 
358 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
359 {
360 #ifdef CONFIG_NET_SCHED
361 	if (sch->stab)
362 		qdisc_calculate_pkt_len(skb, sch->stab);
363 #endif
364 	return sch->enqueue(skb, sch);
365 }
366 
367 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
368 {
369 	qdisc_skb_cb(skb)->pkt_len = skb->len;
370 	return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
371 }
372 
373 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
374 				       struct sk_buff_head *list)
375 {
376 	__skb_queue_tail(list, skb);
377 	sch->qstats.backlog += qdisc_pkt_len(skb);
378 	sch->bstats.bytes += qdisc_pkt_len(skb);
379 	sch->bstats.packets++;
380 
381 	return NET_XMIT_SUCCESS;
382 }
383 
384 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
385 {
386 	return __qdisc_enqueue_tail(skb, sch, &sch->q);
387 }
388 
389 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
390 						   struct sk_buff_head *list)
391 {
392 	struct sk_buff *skb = __skb_dequeue(list);
393 
394 	if (likely(skb != NULL))
395 		sch->qstats.backlog -= qdisc_pkt_len(skb);
396 
397 	return skb;
398 }
399 
400 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
401 {
402 	return __qdisc_dequeue_head(sch, &sch->q);
403 }
404 
405 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
406 						   struct sk_buff_head *list)
407 {
408 	struct sk_buff *skb = __skb_dequeue_tail(list);
409 
410 	if (likely(skb != NULL))
411 		sch->qstats.backlog -= qdisc_pkt_len(skb);
412 
413 	return skb;
414 }
415 
416 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
417 {
418 	return __qdisc_dequeue_tail(sch, &sch->q);
419 }
420 
421 static inline int __qdisc_requeue(struct sk_buff *skb, struct Qdisc *sch,
422 				  struct sk_buff_head *list)
423 {
424 	__skb_queue_head(list, skb);
425 	sch->qstats.backlog += qdisc_pkt_len(skb);
426 	sch->qstats.requeues++;
427 
428 	return NET_XMIT_SUCCESS;
429 }
430 
431 static inline int qdisc_requeue(struct sk_buff *skb, struct Qdisc *sch)
432 {
433 	return __qdisc_requeue(skb, sch, &sch->q);
434 }
435 
436 static inline void __qdisc_reset_queue(struct Qdisc *sch,
437 				       struct sk_buff_head *list)
438 {
439 	/*
440 	 * We do not know the backlog in bytes of this list, it
441 	 * is up to the caller to correct it
442 	 */
443 	__skb_queue_purge(list);
444 }
445 
446 static inline void qdisc_reset_queue(struct Qdisc *sch)
447 {
448 	__qdisc_reset_queue(sch, &sch->q);
449 	sch->qstats.backlog = 0;
450 }
451 
452 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
453 					      struct sk_buff_head *list)
454 {
455 	struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
456 
457 	if (likely(skb != NULL)) {
458 		unsigned int len = qdisc_pkt_len(skb);
459 		kfree_skb(skb);
460 		return len;
461 	}
462 
463 	return 0;
464 }
465 
466 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
467 {
468 	return __qdisc_queue_drop(sch, &sch->q);
469 }
470 
471 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
472 {
473 	kfree_skb(skb);
474 	sch->qstats.drops++;
475 
476 	return NET_XMIT_DROP;
477 }
478 
479 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
480 {
481 	sch->qstats.drops++;
482 
483 #ifdef CONFIG_NET_CLS_ACT
484 	if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
485 		goto drop;
486 
487 	return NET_XMIT_SUCCESS;
488 
489 drop:
490 #endif
491 	kfree_skb(skb);
492 	return NET_XMIT_DROP;
493 }
494 
495 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
496    long it will take to send a packet given its size.
497  */
498 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
499 {
500 	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
501 	if (slot < 0)
502 		slot = 0;
503 	slot >>= rtab->rate.cell_log;
504 	if (slot > 255)
505 		return (rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]);
506 	return rtab->data[slot];
507 }
508 
509 #ifdef CONFIG_NET_CLS_ACT
510 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask)
511 {
512 	struct sk_buff *n = skb_clone(skb, gfp_mask);
513 
514 	if (n) {
515 		n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
516 		n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
517 		n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
518 	}
519 	return n;
520 }
521 #endif
522 
523 #endif
524