xref: /openbmc/linux/include/net/sch_generic.h (revision a09d2831)
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_RUNNING,
27 	__QDISC_STATE_SCHED,
28 	__QDISC_STATE_DEACTIVATED,
29 };
30 
31 struct qdisc_size_table {
32 	struct list_head	list;
33 	struct tc_sizespec	szopts;
34 	int			refcnt;
35 	u16			data[];
36 };
37 
38 struct Qdisc {
39 	int 			(*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
40 	struct sk_buff *	(*dequeue)(struct Qdisc *dev);
41 	unsigned		flags;
42 #define TCQ_F_BUILTIN		1
43 #define TCQ_F_THROTTLED		2
44 #define TCQ_F_INGRESS		4
45 #define TCQ_F_CAN_BYPASS	8
46 #define TCQ_F_MQROOT		16
47 #define TCQ_F_WARN_NONWC	(1 << 16)
48 	int			padded;
49 	struct Qdisc_ops	*ops;
50 	struct qdisc_size_table	*stab;
51 	struct list_head	list;
52 	u32			handle;
53 	u32			parent;
54 	atomic_t		refcnt;
55 	struct gnet_stats_rate_est	rate_est;
56 	int			(*reshape_fail)(struct sk_buff *skb,
57 					struct Qdisc *q);
58 
59 	void			*u32_node;
60 
61 	/* This field is deprecated, but it is still used by CBQ
62 	 * and it will live until better solution will be invented.
63 	 */
64 	struct Qdisc		*__parent;
65 	struct netdev_queue	*dev_queue;
66 	struct Qdisc		*next_sched;
67 
68 	struct sk_buff		*gso_skb;
69 	/*
70 	 * For performance sake on SMP, we put highly modified fields at the end
71 	 */
72 	unsigned long		state;
73 	struct sk_buff_head	q;
74 	struct gnet_stats_basic_packed bstats;
75 	struct gnet_stats_queue	qstats;
76 };
77 
78 struct Qdisc_class_ops {
79 	/* Child qdisc manipulation */
80 	struct netdev_queue *	(*select_queue)(struct Qdisc *, struct tcmsg *);
81 	int			(*graft)(struct Qdisc *, unsigned long cl,
82 					struct Qdisc *, struct Qdisc **);
83 	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
84 	void			(*qlen_notify)(struct Qdisc *, unsigned long);
85 
86 	/* Class manipulation routines */
87 	unsigned long		(*get)(struct Qdisc *, u32 classid);
88 	void			(*put)(struct Qdisc *, unsigned long);
89 	int			(*change)(struct Qdisc *, u32, u32,
90 					struct nlattr **, unsigned long *);
91 	int			(*delete)(struct Qdisc *, unsigned long);
92 	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
93 
94 	/* Filter manipulation */
95 	struct tcf_proto **	(*tcf_chain)(struct Qdisc *, unsigned long);
96 	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
97 					u32 classid);
98 	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
99 
100 	/* rtnetlink specific */
101 	int			(*dump)(struct Qdisc *, unsigned long,
102 					struct sk_buff *skb, struct tcmsg*);
103 	int			(*dump_stats)(struct Qdisc *, unsigned long,
104 					struct gnet_dump *);
105 };
106 
107 struct Qdisc_ops {
108 	struct Qdisc_ops	*next;
109 	const struct Qdisc_class_ops	*cl_ops;
110 	char			id[IFNAMSIZ];
111 	int			priv_size;
112 
113 	int 			(*enqueue)(struct sk_buff *, struct Qdisc *);
114 	struct sk_buff *	(*dequeue)(struct Qdisc *);
115 	struct sk_buff *	(*peek)(struct Qdisc *);
116 	unsigned int		(*drop)(struct Qdisc *);
117 
118 	int			(*init)(struct Qdisc *, struct nlattr *arg);
119 	void			(*reset)(struct Qdisc *);
120 	void			(*destroy)(struct Qdisc *);
121 	int			(*change)(struct Qdisc *, struct nlattr *arg);
122 	void			(*attach)(struct Qdisc *);
123 
124 	int			(*dump)(struct Qdisc *, struct sk_buff *);
125 	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
126 
127 	struct module		*owner;
128 };
129 
130 
131 struct tcf_result {
132 	unsigned long	class;
133 	u32		classid;
134 };
135 
136 struct tcf_proto_ops {
137 	struct tcf_proto_ops	*next;
138 	char			kind[IFNAMSIZ];
139 
140 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
141 					struct tcf_result *);
142 	int			(*init)(struct tcf_proto*);
143 	void			(*destroy)(struct tcf_proto*);
144 
145 	unsigned long		(*get)(struct tcf_proto*, u32 handle);
146 	void			(*put)(struct tcf_proto*, unsigned long);
147 	int			(*change)(struct tcf_proto*, unsigned long,
148 					u32 handle, struct nlattr **,
149 					unsigned long *);
150 	int			(*delete)(struct tcf_proto*, unsigned long);
151 	void			(*walk)(struct tcf_proto*, struct tcf_walker *arg);
152 
153 	/* rtnetlink specific */
154 	int			(*dump)(struct tcf_proto*, unsigned long,
155 					struct sk_buff *skb, struct tcmsg*);
156 
157 	struct module		*owner;
158 };
159 
160 struct tcf_proto {
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 int qdisc_qlen(struct Qdisc *q)
182 {
183 	return q->q.qlen;
184 }
185 
186 static inline struct qdisc_skb_cb *qdisc_skb_cb(struct sk_buff *skb)
187 {
188 	return (struct qdisc_skb_cb *)skb->cb;
189 }
190 
191 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
192 {
193 	return &qdisc->q.lock;
194 }
195 
196 static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
197 {
198 	return qdisc->dev_queue->qdisc;
199 }
200 
201 static inline struct Qdisc *qdisc_root_sleeping(struct Qdisc *qdisc)
202 {
203 	return qdisc->dev_queue->qdisc_sleeping;
204 }
205 
206 /* The qdisc root lock is a mechanism by which to top level
207  * of a qdisc tree can be locked from any qdisc node in the
208  * forest.  This allows changing the configuration of some
209  * aspect of the qdisc tree while blocking out asynchronous
210  * qdisc access in the packet processing paths.
211  *
212  * It is only legal to do this when the root will not change
213  * on us.  Otherwise we'll potentially lock the wrong qdisc
214  * root.  This is enforced by holding the RTNL semaphore, which
215  * all users of this lock accessor must do.
216  */
217 static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
218 {
219 	struct Qdisc *root = qdisc_root(qdisc);
220 
221 	ASSERT_RTNL();
222 	return qdisc_lock(root);
223 }
224 
225 static inline spinlock_t *qdisc_root_sleeping_lock(struct Qdisc *qdisc)
226 {
227 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
228 
229 	ASSERT_RTNL();
230 	return qdisc_lock(root);
231 }
232 
233 static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
234 {
235 	return qdisc->dev_queue->dev;
236 }
237 
238 static inline void sch_tree_lock(struct Qdisc *q)
239 {
240 	spin_lock_bh(qdisc_root_sleeping_lock(q));
241 }
242 
243 static inline void sch_tree_unlock(struct Qdisc *q)
244 {
245 	spin_unlock_bh(qdisc_root_sleeping_lock(q));
246 }
247 
248 #define tcf_tree_lock(tp)	sch_tree_lock((tp)->q)
249 #define tcf_tree_unlock(tp)	sch_tree_unlock((tp)->q)
250 
251 extern struct Qdisc noop_qdisc;
252 extern struct Qdisc_ops noop_qdisc_ops;
253 extern struct Qdisc_ops pfifo_fast_ops;
254 extern struct Qdisc_ops mq_qdisc_ops;
255 
256 struct Qdisc_class_common {
257 	u32			classid;
258 	struct hlist_node	hnode;
259 };
260 
261 struct Qdisc_class_hash {
262 	struct hlist_head	*hash;
263 	unsigned int		hashsize;
264 	unsigned int		hashmask;
265 	unsigned int		hashelems;
266 };
267 
268 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
269 {
270 	id ^= id >> 8;
271 	id ^= id >> 4;
272 	return id & mask;
273 }
274 
275 static inline struct Qdisc_class_common *
276 qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
277 {
278 	struct Qdisc_class_common *cl;
279 	struct hlist_node *n;
280 	unsigned int h;
281 
282 	h = qdisc_class_hash(id, hash->hashmask);
283 	hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
284 		if (cl->classid == id)
285 			return cl;
286 	}
287 	return NULL;
288 }
289 
290 extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
291 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
292 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
293 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
294 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
295 
296 extern void dev_init_scheduler(struct net_device *dev);
297 extern void dev_shutdown(struct net_device *dev);
298 extern void dev_activate(struct net_device *dev);
299 extern void dev_deactivate(struct net_device *dev);
300 extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
301 				     struct Qdisc *qdisc);
302 extern void qdisc_reset(struct Qdisc *qdisc);
303 extern void qdisc_destroy(struct Qdisc *qdisc);
304 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
305 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
306 				 struct Qdisc_ops *ops);
307 extern struct Qdisc *qdisc_create_dflt(struct net_device *dev,
308 				       struct netdev_queue *dev_queue,
309 				       struct Qdisc_ops *ops, u32 parentid);
310 extern void qdisc_calculate_pkt_len(struct sk_buff *skb,
311 				   struct qdisc_size_table *stab);
312 extern void tcf_destroy(struct tcf_proto *tp);
313 extern void tcf_destroy_chain(struct tcf_proto **fl);
314 
315 /* Reset all TX qdiscs of a device.  */
316 static inline void qdisc_reset_all_tx(struct net_device *dev)
317 {
318 	unsigned int i;
319 	for (i = 0; i < dev->num_tx_queues; i++)
320 		qdisc_reset(netdev_get_tx_queue(dev, i)->qdisc);
321 }
322 
323 /* Are all TX queues of the device empty?  */
324 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
325 {
326 	unsigned int i;
327 	for (i = 0; i < dev->num_tx_queues; i++) {
328 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
329 		const struct Qdisc *q = txq->qdisc;
330 
331 		if (q->q.qlen)
332 			return false;
333 	}
334 	return true;
335 }
336 
337 /* Are any of the TX qdiscs changing?  */
338 static inline bool qdisc_tx_changing(struct net_device *dev)
339 {
340 	unsigned int i;
341 	for (i = 0; i < dev->num_tx_queues; i++) {
342 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
343 		if (txq->qdisc != txq->qdisc_sleeping)
344 			return true;
345 	}
346 	return false;
347 }
348 
349 /* Is the device using the noop qdisc on all queues?  */
350 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
351 {
352 	unsigned int i;
353 	for (i = 0; i < dev->num_tx_queues; i++) {
354 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
355 		if (txq->qdisc != &noop_qdisc)
356 			return false;
357 	}
358 	return true;
359 }
360 
361 static inline unsigned int qdisc_pkt_len(struct sk_buff *skb)
362 {
363 	return qdisc_skb_cb(skb)->pkt_len;
364 }
365 
366 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
367 enum net_xmit_qdisc_t {
368 	__NET_XMIT_STOLEN = 0x00010000,
369 	__NET_XMIT_BYPASS = 0x00020000,
370 };
371 
372 #ifdef CONFIG_NET_CLS_ACT
373 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
374 #else
375 #define net_xmit_drop_count(e)	(1)
376 #endif
377 
378 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
379 {
380 #ifdef CONFIG_NET_SCHED
381 	if (sch->stab)
382 		qdisc_calculate_pkt_len(skb, sch->stab);
383 #endif
384 	return sch->enqueue(skb, sch);
385 }
386 
387 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
388 {
389 	qdisc_skb_cb(skb)->pkt_len = skb->len;
390 	return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
391 }
392 
393 static inline void __qdisc_update_bstats(struct Qdisc *sch, unsigned int len)
394 {
395 	sch->bstats.bytes += len;
396 	sch->bstats.packets++;
397 }
398 
399 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
400 				       struct sk_buff_head *list)
401 {
402 	__skb_queue_tail(list, skb);
403 	sch->qstats.backlog += qdisc_pkt_len(skb);
404 	__qdisc_update_bstats(sch, qdisc_pkt_len(skb));
405 
406 	return NET_XMIT_SUCCESS;
407 }
408 
409 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
410 {
411 	return __qdisc_enqueue_tail(skb, sch, &sch->q);
412 }
413 
414 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
415 						   struct sk_buff_head *list)
416 {
417 	struct sk_buff *skb = __skb_dequeue(list);
418 
419 	if (likely(skb != NULL))
420 		sch->qstats.backlog -= qdisc_pkt_len(skb);
421 
422 	return skb;
423 }
424 
425 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
426 {
427 	return __qdisc_dequeue_head(sch, &sch->q);
428 }
429 
430 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
431 						   struct sk_buff_head *list)
432 {
433 	struct sk_buff *skb = __skb_dequeue_tail(list);
434 
435 	if (likely(skb != NULL))
436 		sch->qstats.backlog -= qdisc_pkt_len(skb);
437 
438 	return skb;
439 }
440 
441 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
442 {
443 	return __qdisc_dequeue_tail(sch, &sch->q);
444 }
445 
446 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
447 {
448 	return skb_peek(&sch->q);
449 }
450 
451 /* generic pseudo peek method for non-work-conserving qdisc */
452 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
453 {
454 	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
455 	if (!sch->gso_skb) {
456 		sch->gso_skb = sch->dequeue(sch);
457 		if (sch->gso_skb)
458 			/* it's still part of the queue */
459 			sch->q.qlen++;
460 	}
461 
462 	return sch->gso_skb;
463 }
464 
465 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
466 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
467 {
468 	struct sk_buff *skb = sch->gso_skb;
469 
470 	if (skb) {
471 		sch->gso_skb = NULL;
472 		sch->q.qlen--;
473 	} else {
474 		skb = sch->dequeue(sch);
475 	}
476 
477 	return skb;
478 }
479 
480 static inline void __qdisc_reset_queue(struct Qdisc *sch,
481 				       struct sk_buff_head *list)
482 {
483 	/*
484 	 * We do not know the backlog in bytes of this list, it
485 	 * is up to the caller to correct it
486 	 */
487 	__skb_queue_purge(list);
488 }
489 
490 static inline void qdisc_reset_queue(struct Qdisc *sch)
491 {
492 	__qdisc_reset_queue(sch, &sch->q);
493 	sch->qstats.backlog = 0;
494 }
495 
496 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
497 					      struct sk_buff_head *list)
498 {
499 	struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
500 
501 	if (likely(skb != NULL)) {
502 		unsigned int len = qdisc_pkt_len(skb);
503 		kfree_skb(skb);
504 		return len;
505 	}
506 
507 	return 0;
508 }
509 
510 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
511 {
512 	return __qdisc_queue_drop(sch, &sch->q);
513 }
514 
515 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
516 {
517 	kfree_skb(skb);
518 	sch->qstats.drops++;
519 
520 	return NET_XMIT_DROP;
521 }
522 
523 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
524 {
525 	sch->qstats.drops++;
526 
527 #ifdef CONFIG_NET_CLS_ACT
528 	if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
529 		goto drop;
530 
531 	return NET_XMIT_SUCCESS;
532 
533 drop:
534 #endif
535 	kfree_skb(skb);
536 	return NET_XMIT_DROP;
537 }
538 
539 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
540    long it will take to send a packet given its size.
541  */
542 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
543 {
544 	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
545 	if (slot < 0)
546 		slot = 0;
547 	slot >>= rtab->rate.cell_log;
548 	if (slot > 255)
549 		return (rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]);
550 	return rtab->data[slot];
551 }
552 
553 #ifdef CONFIG_NET_CLS_ACT
554 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask)
555 {
556 	struct sk_buff *n = skb_clone(skb, gfp_mask);
557 
558 	if (n) {
559 		n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
560 		n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
561 		n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
562 	}
563 	return n;
564 }
565 #endif
566 
567 #endif
568