xref: /openbmc/linux/include/net/sch_generic.h (revision e190bfe5)
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 	struct rcu_head     rcu_head;
77 };
78 
79 struct Qdisc_class_ops {
80 	/* Child qdisc manipulation */
81 	struct netdev_queue *	(*select_queue)(struct Qdisc *, struct tcmsg *);
82 	int			(*graft)(struct Qdisc *, unsigned long cl,
83 					struct Qdisc *, struct Qdisc **);
84 	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
85 	void			(*qlen_notify)(struct Qdisc *, unsigned long);
86 
87 	/* Class manipulation routines */
88 	unsigned long		(*get)(struct Qdisc *, u32 classid);
89 	void			(*put)(struct Qdisc *, unsigned long);
90 	int			(*change)(struct Qdisc *, u32, u32,
91 					struct nlattr **, unsigned long *);
92 	int			(*delete)(struct Qdisc *, unsigned long);
93 	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
94 
95 	/* Filter manipulation */
96 	struct tcf_proto **	(*tcf_chain)(struct Qdisc *, unsigned long);
97 	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
98 					u32 classid);
99 	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
100 
101 	/* rtnetlink specific */
102 	int			(*dump)(struct Qdisc *, unsigned long,
103 					struct sk_buff *skb, struct tcmsg*);
104 	int			(*dump_stats)(struct Qdisc *, unsigned long,
105 					struct gnet_dump *);
106 };
107 
108 struct Qdisc_ops {
109 	struct Qdisc_ops	*next;
110 	const struct Qdisc_class_ops	*cl_ops;
111 	char			id[IFNAMSIZ];
112 	int			priv_size;
113 
114 	int 			(*enqueue)(struct sk_buff *, struct Qdisc *);
115 	struct sk_buff *	(*dequeue)(struct Qdisc *);
116 	struct sk_buff *	(*peek)(struct Qdisc *);
117 	unsigned int		(*drop)(struct Qdisc *);
118 
119 	int			(*init)(struct Qdisc *, struct nlattr *arg);
120 	void			(*reset)(struct Qdisc *);
121 	void			(*destroy)(struct Qdisc *);
122 	int			(*change)(struct Qdisc *, struct nlattr *arg);
123 	void			(*attach)(struct Qdisc *);
124 
125 	int			(*dump)(struct Qdisc *, struct sk_buff *);
126 	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
127 
128 	struct module		*owner;
129 };
130 
131 
132 struct tcf_result {
133 	unsigned long	class;
134 	u32		classid;
135 };
136 
137 struct tcf_proto_ops {
138 	struct tcf_proto_ops	*next;
139 	char			kind[IFNAMSIZ];
140 
141 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
142 					struct tcf_result *);
143 	int			(*init)(struct tcf_proto*);
144 	void			(*destroy)(struct tcf_proto*);
145 
146 	unsigned long		(*get)(struct tcf_proto*, u32 handle);
147 	void			(*put)(struct tcf_proto*, unsigned long);
148 	int			(*change)(struct tcf_proto*, unsigned long,
149 					u32 handle, struct nlattr **,
150 					unsigned long *);
151 	int			(*delete)(struct tcf_proto*, unsigned long);
152 	void			(*walk)(struct tcf_proto*, struct tcf_walker *arg);
153 
154 	/* rtnetlink specific */
155 	int			(*dump)(struct tcf_proto*, unsigned long,
156 					struct sk_buff *skb, struct tcmsg*);
157 
158 	struct module		*owner;
159 };
160 
161 struct tcf_proto {
162 	/* Fast access part */
163 	struct tcf_proto	*next;
164 	void			*root;
165 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
166 					struct tcf_result *);
167 	__be16			protocol;
168 
169 	/* All the rest */
170 	u32			prio;
171 	u32			classid;
172 	struct Qdisc		*q;
173 	void			*data;
174 	struct tcf_proto_ops	*ops;
175 };
176 
177 struct qdisc_skb_cb {
178 	unsigned int		pkt_len;
179 	char			data[];
180 };
181 
182 static inline int qdisc_qlen(struct Qdisc *q)
183 {
184 	return q->q.qlen;
185 }
186 
187 static inline struct qdisc_skb_cb *qdisc_skb_cb(struct sk_buff *skb)
188 {
189 	return (struct qdisc_skb_cb *)skb->cb;
190 }
191 
192 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
193 {
194 	return &qdisc->q.lock;
195 }
196 
197 static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
198 {
199 	return qdisc->dev_queue->qdisc;
200 }
201 
202 static inline struct Qdisc *qdisc_root_sleeping(struct Qdisc *qdisc)
203 {
204 	return qdisc->dev_queue->qdisc_sleeping;
205 }
206 
207 /* The qdisc root lock is a mechanism by which to top level
208  * of a qdisc tree can be locked from any qdisc node in the
209  * forest.  This allows changing the configuration of some
210  * aspect of the qdisc tree while blocking out asynchronous
211  * qdisc access in the packet processing paths.
212  *
213  * It is only legal to do this when the root will not change
214  * on us.  Otherwise we'll potentially lock the wrong qdisc
215  * root.  This is enforced by holding the RTNL semaphore, which
216  * all users of this lock accessor must do.
217  */
218 static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
219 {
220 	struct Qdisc *root = qdisc_root(qdisc);
221 
222 	ASSERT_RTNL();
223 	return qdisc_lock(root);
224 }
225 
226 static inline spinlock_t *qdisc_root_sleeping_lock(struct Qdisc *qdisc)
227 {
228 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
229 
230 	ASSERT_RTNL();
231 	return qdisc_lock(root);
232 }
233 
234 static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
235 {
236 	return qdisc->dev_queue->dev;
237 }
238 
239 static inline void sch_tree_lock(struct Qdisc *q)
240 {
241 	spin_lock_bh(qdisc_root_sleeping_lock(q));
242 }
243 
244 static inline void sch_tree_unlock(struct Qdisc *q)
245 {
246 	spin_unlock_bh(qdisc_root_sleeping_lock(q));
247 }
248 
249 #define tcf_tree_lock(tp)	sch_tree_lock((tp)->q)
250 #define tcf_tree_unlock(tp)	sch_tree_unlock((tp)->q)
251 
252 extern struct Qdisc noop_qdisc;
253 extern struct Qdisc_ops noop_qdisc_ops;
254 extern struct Qdisc_ops pfifo_fast_ops;
255 extern struct Qdisc_ops mq_qdisc_ops;
256 
257 struct Qdisc_class_common {
258 	u32			classid;
259 	struct hlist_node	hnode;
260 };
261 
262 struct Qdisc_class_hash {
263 	struct hlist_head	*hash;
264 	unsigned int		hashsize;
265 	unsigned int		hashmask;
266 	unsigned int		hashelems;
267 };
268 
269 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
270 {
271 	id ^= id >> 8;
272 	id ^= id >> 4;
273 	return id & mask;
274 }
275 
276 static inline struct Qdisc_class_common *
277 qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
278 {
279 	struct Qdisc_class_common *cl;
280 	struct hlist_node *n;
281 	unsigned int h;
282 
283 	h = qdisc_class_hash(id, hash->hashmask);
284 	hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
285 		if (cl->classid == id)
286 			return cl;
287 	}
288 	return NULL;
289 }
290 
291 extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
292 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
293 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
294 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
295 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
296 
297 extern void dev_init_scheduler(struct net_device *dev);
298 extern void dev_shutdown(struct net_device *dev);
299 extern void dev_activate(struct net_device *dev);
300 extern void dev_deactivate(struct net_device *dev);
301 extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
302 				     struct Qdisc *qdisc);
303 extern void qdisc_reset(struct Qdisc *qdisc);
304 extern void qdisc_destroy(struct Qdisc *qdisc);
305 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
306 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
307 				 struct Qdisc_ops *ops);
308 extern struct Qdisc *qdisc_create_dflt(struct net_device *dev,
309 				       struct netdev_queue *dev_queue,
310 				       struct Qdisc_ops *ops, u32 parentid);
311 extern void qdisc_calculate_pkt_len(struct sk_buff *skb,
312 				   struct qdisc_size_table *stab);
313 extern void tcf_destroy(struct tcf_proto *tp);
314 extern void tcf_destroy_chain(struct tcf_proto **fl);
315 
316 /* Reset all TX qdiscs of a device.  */
317 static inline void qdisc_reset_all_tx(struct net_device *dev)
318 {
319 	unsigned int i;
320 	for (i = 0; i < dev->num_tx_queues; i++)
321 		qdisc_reset(netdev_get_tx_queue(dev, i)->qdisc);
322 }
323 
324 /* Are all TX queues of the device empty?  */
325 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
326 {
327 	unsigned int i;
328 	for (i = 0; i < dev->num_tx_queues; i++) {
329 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
330 		const struct Qdisc *q = txq->qdisc;
331 
332 		if (q->q.qlen)
333 			return false;
334 	}
335 	return true;
336 }
337 
338 /* Are any of the TX qdiscs changing?  */
339 static inline bool qdisc_tx_changing(struct net_device *dev)
340 {
341 	unsigned int i;
342 	for (i = 0; i < dev->num_tx_queues; i++) {
343 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
344 		if (txq->qdisc != txq->qdisc_sleeping)
345 			return true;
346 	}
347 	return false;
348 }
349 
350 /* Is the device using the noop qdisc on all queues?  */
351 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
352 {
353 	unsigned int i;
354 	for (i = 0; i < dev->num_tx_queues; i++) {
355 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
356 		if (txq->qdisc != &noop_qdisc)
357 			return false;
358 	}
359 	return true;
360 }
361 
362 static inline unsigned int qdisc_pkt_len(struct sk_buff *skb)
363 {
364 	return qdisc_skb_cb(skb)->pkt_len;
365 }
366 
367 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
368 enum net_xmit_qdisc_t {
369 	__NET_XMIT_STOLEN = 0x00010000,
370 	__NET_XMIT_BYPASS = 0x00020000,
371 };
372 
373 #ifdef CONFIG_NET_CLS_ACT
374 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
375 #else
376 #define net_xmit_drop_count(e)	(1)
377 #endif
378 
379 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
380 {
381 #ifdef CONFIG_NET_SCHED
382 	if (sch->stab)
383 		qdisc_calculate_pkt_len(skb, sch->stab);
384 #endif
385 	return sch->enqueue(skb, sch);
386 }
387 
388 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
389 {
390 	qdisc_skb_cb(skb)->pkt_len = skb->len;
391 	return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
392 }
393 
394 static inline void __qdisc_update_bstats(struct Qdisc *sch, unsigned int len)
395 {
396 	sch->bstats.bytes += len;
397 	sch->bstats.packets++;
398 }
399 
400 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
401 				       struct sk_buff_head *list)
402 {
403 	__skb_queue_tail(list, skb);
404 	sch->qstats.backlog += qdisc_pkt_len(skb);
405 	__qdisc_update_bstats(sch, qdisc_pkt_len(skb));
406 
407 	return NET_XMIT_SUCCESS;
408 }
409 
410 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
411 {
412 	return __qdisc_enqueue_tail(skb, sch, &sch->q);
413 }
414 
415 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
416 						   struct sk_buff_head *list)
417 {
418 	struct sk_buff *skb = __skb_dequeue(list);
419 
420 	if (likely(skb != NULL))
421 		sch->qstats.backlog -= qdisc_pkt_len(skb);
422 
423 	return skb;
424 }
425 
426 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
427 {
428 	return __qdisc_dequeue_head(sch, &sch->q);
429 }
430 
431 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
432 					      struct sk_buff_head *list)
433 {
434 	struct sk_buff *skb = __qdisc_dequeue_head(sch, list);
435 
436 	if (likely(skb != NULL)) {
437 		unsigned int len = qdisc_pkt_len(skb);
438 		kfree_skb(skb);
439 		return len;
440 	}
441 
442 	return 0;
443 }
444 
445 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch)
446 {
447 	return __qdisc_queue_drop_head(sch, &sch->q);
448 }
449 
450 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
451 						   struct sk_buff_head *list)
452 {
453 	struct sk_buff *skb = __skb_dequeue_tail(list);
454 
455 	if (likely(skb != NULL))
456 		sch->qstats.backlog -= qdisc_pkt_len(skb);
457 
458 	return skb;
459 }
460 
461 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
462 {
463 	return __qdisc_dequeue_tail(sch, &sch->q);
464 }
465 
466 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
467 {
468 	return skb_peek(&sch->q);
469 }
470 
471 /* generic pseudo peek method for non-work-conserving qdisc */
472 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
473 {
474 	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
475 	if (!sch->gso_skb) {
476 		sch->gso_skb = sch->dequeue(sch);
477 		if (sch->gso_skb)
478 			/* it's still part of the queue */
479 			sch->q.qlen++;
480 	}
481 
482 	return sch->gso_skb;
483 }
484 
485 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
486 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
487 {
488 	struct sk_buff *skb = sch->gso_skb;
489 
490 	if (skb) {
491 		sch->gso_skb = NULL;
492 		sch->q.qlen--;
493 	} else {
494 		skb = sch->dequeue(sch);
495 	}
496 
497 	return skb;
498 }
499 
500 static inline void __qdisc_reset_queue(struct Qdisc *sch,
501 				       struct sk_buff_head *list)
502 {
503 	/*
504 	 * We do not know the backlog in bytes of this list, it
505 	 * is up to the caller to correct it
506 	 */
507 	__skb_queue_purge(list);
508 }
509 
510 static inline void qdisc_reset_queue(struct Qdisc *sch)
511 {
512 	__qdisc_reset_queue(sch, &sch->q);
513 	sch->qstats.backlog = 0;
514 }
515 
516 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
517 					      struct sk_buff_head *list)
518 {
519 	struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
520 
521 	if (likely(skb != NULL)) {
522 		unsigned int len = qdisc_pkt_len(skb);
523 		kfree_skb(skb);
524 		return len;
525 	}
526 
527 	return 0;
528 }
529 
530 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
531 {
532 	return __qdisc_queue_drop(sch, &sch->q);
533 }
534 
535 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
536 {
537 	kfree_skb(skb);
538 	sch->qstats.drops++;
539 
540 	return NET_XMIT_DROP;
541 }
542 
543 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
544 {
545 	sch->qstats.drops++;
546 
547 #ifdef CONFIG_NET_CLS_ACT
548 	if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
549 		goto drop;
550 
551 	return NET_XMIT_SUCCESS;
552 
553 drop:
554 #endif
555 	kfree_skb(skb);
556 	return NET_XMIT_DROP;
557 }
558 
559 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
560    long it will take to send a packet given its size.
561  */
562 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
563 {
564 	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
565 	if (slot < 0)
566 		slot = 0;
567 	slot >>= rtab->rate.cell_log;
568 	if (slot > 255)
569 		return (rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]);
570 	return rtab->data[slot];
571 }
572 
573 #ifdef CONFIG_NET_CLS_ACT
574 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask)
575 {
576 	struct sk_buff *n = skb_clone(skb, gfp_mask);
577 
578 	if (n) {
579 		n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
580 		n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
581 		n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
582 	}
583 	return n;
584 }
585 #endif
586 
587 #endif
588