xref: /openbmc/linux/include/net/sch_generic.h (revision b04b4f78)
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_WARN_NONWC	(1 << 16)
49 	int			padded;
50 	struct Qdisc_ops	*ops;
51 	struct qdisc_size_table	*stab;
52 	struct list_head	list;
53 	u32			handle;
54 	u32			parent;
55 	atomic_t		refcnt;
56 	struct gnet_stats_rate_est	rate_est;
57 	int			(*reshape_fail)(struct sk_buff *skb,
58 					struct Qdisc *q);
59 
60 	void			*u32_node;
61 
62 	/* This field is deprecated, but it is still used by CBQ
63 	 * and it will live until better solution will be invented.
64 	 */
65 	struct Qdisc		*__parent;
66 	struct netdev_queue	*dev_queue;
67 	struct Qdisc		*next_sched;
68 
69 	struct sk_buff		*gso_skb;
70 	/*
71 	 * For performance sake on SMP, we put highly modified fields at the end
72 	 */
73 	unsigned long		state;
74 	struct sk_buff_head	q;
75 	struct gnet_stats_basic bstats;
76 	struct gnet_stats_queue	qstats;
77 };
78 
79 struct Qdisc_class_ops
80 {
81 	/* Child qdisc manipulation */
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 {
110 	struct Qdisc_ops	*next;
111 	const struct Qdisc_class_ops	*cl_ops;
112 	char			id[IFNAMSIZ];
113 	int			priv_size;
114 
115 	int 			(*enqueue)(struct sk_buff *, struct Qdisc *);
116 	struct sk_buff *	(*dequeue)(struct Qdisc *);
117 	struct sk_buff *	(*peek)(struct Qdisc *);
118 	unsigned int		(*drop)(struct Qdisc *);
119 
120 	int			(*init)(struct Qdisc *, struct nlattr *arg);
121 	void			(*reset)(struct Qdisc *);
122 	void			(*destroy)(struct Qdisc *);
123 	int			(*change)(struct Qdisc *, struct nlattr *arg);
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 {
134 	unsigned long	class;
135 	u32		classid;
136 };
137 
138 struct tcf_proto_ops
139 {
140 	struct tcf_proto_ops	*next;
141 	char			kind[IFNAMSIZ];
142 
143 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
144 					struct tcf_result *);
145 	int			(*init)(struct tcf_proto*);
146 	void			(*destroy)(struct tcf_proto*);
147 
148 	unsigned long		(*get)(struct tcf_proto*, u32 handle);
149 	void			(*put)(struct tcf_proto*, unsigned long);
150 	int			(*change)(struct tcf_proto*, unsigned long,
151 					u32 handle, struct nlattr **,
152 					unsigned long *);
153 	int			(*delete)(struct tcf_proto*, unsigned long);
154 	void			(*walk)(struct tcf_proto*, struct tcf_walker *arg);
155 
156 	/* rtnetlink specific */
157 	int			(*dump)(struct tcf_proto*, unsigned long,
158 					struct sk_buff *skb, struct tcmsg*);
159 
160 	struct module		*owner;
161 };
162 
163 struct tcf_proto
164 {
165 	/* Fast access part */
166 	struct tcf_proto	*next;
167 	void			*root;
168 	int			(*classify)(struct sk_buff*, struct tcf_proto*,
169 					struct tcf_result *);
170 	__be16			protocol;
171 
172 	/* All the rest */
173 	u32			prio;
174 	u32			classid;
175 	struct Qdisc		*q;
176 	void			*data;
177 	struct tcf_proto_ops	*ops;
178 };
179 
180 struct qdisc_skb_cb {
181 	unsigned int		pkt_len;
182 	char			data[];
183 };
184 
185 static inline struct qdisc_skb_cb *qdisc_skb_cb(struct sk_buff *skb)
186 {
187 	return (struct qdisc_skb_cb *)skb->cb;
188 }
189 
190 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
191 {
192 	return &qdisc->q.lock;
193 }
194 
195 static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
196 {
197 	return qdisc->dev_queue->qdisc;
198 }
199 
200 static inline struct Qdisc *qdisc_root_sleeping(struct Qdisc *qdisc)
201 {
202 	return qdisc->dev_queue->qdisc_sleeping;
203 }
204 
205 /* The qdisc root lock is a mechanism by which to top level
206  * of a qdisc tree can be locked from any qdisc node in the
207  * forest.  This allows changing the configuration of some
208  * aspect of the qdisc tree while blocking out asynchronous
209  * qdisc access in the packet processing paths.
210  *
211  * It is only legal to do this when the root will not change
212  * on us.  Otherwise we'll potentially lock the wrong qdisc
213  * root.  This is enforced by holding the RTNL semaphore, which
214  * all users of this lock accessor must do.
215  */
216 static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
217 {
218 	struct Qdisc *root = qdisc_root(qdisc);
219 
220 	ASSERT_RTNL();
221 	return qdisc_lock(root);
222 }
223 
224 static inline spinlock_t *qdisc_root_sleeping_lock(struct Qdisc *qdisc)
225 {
226 	struct Qdisc *root = qdisc_root_sleeping(qdisc);
227 
228 	ASSERT_RTNL();
229 	return qdisc_lock(root);
230 }
231 
232 static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
233 {
234 	return qdisc->dev_queue->dev;
235 }
236 
237 static inline void sch_tree_lock(struct Qdisc *q)
238 {
239 	spin_lock_bh(qdisc_root_sleeping_lock(q));
240 }
241 
242 static inline void sch_tree_unlock(struct Qdisc *q)
243 {
244 	spin_unlock_bh(qdisc_root_sleeping_lock(q));
245 }
246 
247 #define tcf_tree_lock(tp)	sch_tree_lock((tp)->q)
248 #define tcf_tree_unlock(tp)	sch_tree_unlock((tp)->q)
249 
250 extern struct Qdisc noop_qdisc;
251 extern struct Qdisc_ops noop_qdisc_ops;
252 
253 struct Qdisc_class_common
254 {
255 	u32			classid;
256 	struct hlist_node	hnode;
257 };
258 
259 struct Qdisc_class_hash
260 {
261 	struct hlist_head	*hash;
262 	unsigned int		hashsize;
263 	unsigned int		hashmask;
264 	unsigned int		hashelems;
265 };
266 
267 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
268 {
269 	id ^= id >> 8;
270 	id ^= id >> 4;
271 	return id & mask;
272 }
273 
274 static inline struct Qdisc_class_common *
275 qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
276 {
277 	struct Qdisc_class_common *cl;
278 	struct hlist_node *n;
279 	unsigned int h;
280 
281 	h = qdisc_class_hash(id, hash->hashmask);
282 	hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
283 		if (cl->classid == id)
284 			return cl;
285 	}
286 	return NULL;
287 }
288 
289 extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
290 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
291 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
292 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
293 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
294 
295 extern void dev_init_scheduler(struct net_device *dev);
296 extern void dev_shutdown(struct net_device *dev);
297 extern void dev_activate(struct net_device *dev);
298 extern void dev_deactivate(struct net_device *dev);
299 extern void qdisc_reset(struct Qdisc *qdisc);
300 extern void qdisc_destroy(struct Qdisc *qdisc);
301 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
302 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
303 				 struct Qdisc_ops *ops);
304 extern struct Qdisc *qdisc_create_dflt(struct net_device *dev,
305 				       struct netdev_queue *dev_queue,
306 				       struct Qdisc_ops *ops, u32 parentid);
307 extern void qdisc_calculate_pkt_len(struct sk_buff *skb,
308 				   struct qdisc_size_table *stab);
309 extern void tcf_destroy(struct tcf_proto *tp);
310 extern void tcf_destroy_chain(struct tcf_proto **fl);
311 
312 /* Reset all TX qdiscs of a device.  */
313 static inline void qdisc_reset_all_tx(struct net_device *dev)
314 {
315 	unsigned int i;
316 	for (i = 0; i < dev->num_tx_queues; i++)
317 		qdisc_reset(netdev_get_tx_queue(dev, i)->qdisc);
318 }
319 
320 /* Are all TX queues of the device empty?  */
321 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
322 {
323 	unsigned int i;
324 	for (i = 0; i < dev->num_tx_queues; i++) {
325 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
326 		const struct Qdisc *q = txq->qdisc;
327 
328 		if (q->q.qlen)
329 			return false;
330 	}
331 	return true;
332 }
333 
334 /* Are any of the TX qdiscs changing?  */
335 static inline bool qdisc_tx_changing(struct net_device *dev)
336 {
337 	unsigned int i;
338 	for (i = 0; i < dev->num_tx_queues; i++) {
339 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
340 		if (txq->qdisc != txq->qdisc_sleeping)
341 			return true;
342 	}
343 	return false;
344 }
345 
346 /* Is the device using the noop qdisc on all queues?  */
347 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
348 {
349 	unsigned int i;
350 	for (i = 0; i < dev->num_tx_queues; i++) {
351 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
352 		if (txq->qdisc != &noop_qdisc)
353 			return false;
354 	}
355 	return true;
356 }
357 
358 static inline unsigned int qdisc_pkt_len(struct sk_buff *skb)
359 {
360 	return qdisc_skb_cb(skb)->pkt_len;
361 }
362 
363 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
364 enum net_xmit_qdisc_t {
365 	__NET_XMIT_STOLEN = 0x00010000,
366 	__NET_XMIT_BYPASS = 0x00020000,
367 };
368 
369 #ifdef CONFIG_NET_CLS_ACT
370 #define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
371 #else
372 #define net_xmit_drop_count(e)	(1)
373 #endif
374 
375 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
376 {
377 #ifdef CONFIG_NET_SCHED
378 	if (sch->stab)
379 		qdisc_calculate_pkt_len(skb, sch->stab);
380 #endif
381 	return sch->enqueue(skb, sch);
382 }
383 
384 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
385 {
386 	qdisc_skb_cb(skb)->pkt_len = skb->len;
387 	return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
388 }
389 
390 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
391 				       struct sk_buff_head *list)
392 {
393 	__skb_queue_tail(list, skb);
394 	sch->qstats.backlog += qdisc_pkt_len(skb);
395 	sch->bstats.bytes += qdisc_pkt_len(skb);
396 	sch->bstats.packets++;
397 
398 	return NET_XMIT_SUCCESS;
399 }
400 
401 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
402 {
403 	return __qdisc_enqueue_tail(skb, sch, &sch->q);
404 }
405 
406 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
407 						   struct sk_buff_head *list)
408 {
409 	struct sk_buff *skb = __skb_dequeue(list);
410 
411 	if (likely(skb != NULL))
412 		sch->qstats.backlog -= qdisc_pkt_len(skb);
413 
414 	return skb;
415 }
416 
417 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
418 {
419 	return __qdisc_dequeue_head(sch, &sch->q);
420 }
421 
422 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
423 						   struct sk_buff_head *list)
424 {
425 	struct sk_buff *skb = __skb_dequeue_tail(list);
426 
427 	if (likely(skb != NULL))
428 		sch->qstats.backlog -= qdisc_pkt_len(skb);
429 
430 	return skb;
431 }
432 
433 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
434 {
435 	return __qdisc_dequeue_tail(sch, &sch->q);
436 }
437 
438 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
439 {
440 	return skb_peek(&sch->q);
441 }
442 
443 /* generic pseudo peek method for non-work-conserving qdisc */
444 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
445 {
446 	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
447 	if (!sch->gso_skb) {
448 		sch->gso_skb = sch->dequeue(sch);
449 		if (sch->gso_skb)
450 			/* it's still part of the queue */
451 			sch->q.qlen++;
452 	}
453 
454 	return sch->gso_skb;
455 }
456 
457 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
458 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
459 {
460 	struct sk_buff *skb = sch->gso_skb;
461 
462 	if (skb) {
463 		sch->gso_skb = NULL;
464 		sch->q.qlen--;
465 	} else {
466 		skb = sch->dequeue(sch);
467 	}
468 
469 	return skb;
470 }
471 
472 static inline void __qdisc_reset_queue(struct Qdisc *sch,
473 				       struct sk_buff_head *list)
474 {
475 	/*
476 	 * We do not know the backlog in bytes of this list, it
477 	 * is up to the caller to correct it
478 	 */
479 	__skb_queue_purge(list);
480 }
481 
482 static inline void qdisc_reset_queue(struct Qdisc *sch)
483 {
484 	__qdisc_reset_queue(sch, &sch->q);
485 	sch->qstats.backlog = 0;
486 }
487 
488 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
489 					      struct sk_buff_head *list)
490 {
491 	struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
492 
493 	if (likely(skb != NULL)) {
494 		unsigned int len = qdisc_pkt_len(skb);
495 		kfree_skb(skb);
496 		return len;
497 	}
498 
499 	return 0;
500 }
501 
502 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
503 {
504 	return __qdisc_queue_drop(sch, &sch->q);
505 }
506 
507 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
508 {
509 	kfree_skb(skb);
510 	sch->qstats.drops++;
511 
512 	return NET_XMIT_DROP;
513 }
514 
515 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
516 {
517 	sch->qstats.drops++;
518 
519 #ifdef CONFIG_NET_CLS_ACT
520 	if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
521 		goto drop;
522 
523 	return NET_XMIT_SUCCESS;
524 
525 drop:
526 #endif
527 	kfree_skb(skb);
528 	return NET_XMIT_DROP;
529 }
530 
531 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
532    long it will take to send a packet given its size.
533  */
534 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
535 {
536 	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
537 	if (slot < 0)
538 		slot = 0;
539 	slot >>= rtab->rate.cell_log;
540 	if (slot > 255)
541 		return (rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]);
542 	return rtab->data[slot];
543 }
544 
545 #ifdef CONFIG_NET_CLS_ACT
546 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask)
547 {
548 	struct sk_buff *n = skb_clone(skb, gfp_mask);
549 
550 	if (n) {
551 		n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
552 		n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
553 		n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
554 	}
555 	return n;
556 }
557 #endif
558 
559 #endif
560