?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 3.1-rc1
[linux-kbuild.git] / include / net / sch_generic.h
blob4fc88f3ccd5f870a1647024e33572233314f4681
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_SCHED,
27 __QDISC_STATE_DEACTIVATED,
28 __QDISC_STATE_THROTTLED,
29 };
30
31 /*
32 * following bits are only changed while qdisc lock is held
33 */
34 enum qdisc___state_t {
35 __QDISC___STATE_RUNNING = 1,
36 };
37
38 struct qdisc_size_table {
39 struct rcu_head rcu;
40 struct list_head list;
41 struct tc_sizespec szopts;
42 int refcnt;
43 u16 data[];
44 };
45
46 struct Qdisc {
47 int (*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
48 struct sk_buff * (*dequeue)(struct Qdisc *dev);
49 unsigned flags;
50 #define TCQ_F_BUILTIN 1
51 #define TCQ_F_INGRESS 2
52 #define TCQ_F_CAN_BYPASS 4
53 #define TCQ_F_MQROOT 8
54 #define TCQ_F_WARN_NONWC (1 << 16)
55 int padded;
56 struct Qdisc_ops *ops;
57 struct qdisc_size_table __rcu *stab;
58 struct list_head list;
59 u32 handle;
60 u32 parent;
61 atomic_t refcnt;
62 struct gnet_stats_rate_est rate_est;
63 int (*reshape_fail)(struct sk_buff *skb,
64 struct Qdisc *q);
65
66 void *u32_node;
67
68 /* This field is deprecated, but it is still used by CBQ
69 * and it will live until better solution will be invented.
70 */
71 struct Qdisc *__parent;
72 struct netdev_queue *dev_queue;
73 struct Qdisc *next_sched;
74
75 struct sk_buff *gso_skb;
76 /*
77 * For performance sake on SMP, we put highly modified fields at the end
78 */
79 unsigned long state;
80 struct sk_buff_head q;
81 struct gnet_stats_basic_packed bstats;
82 unsigned int __state;
83 struct gnet_stats_queue qstats;
84 struct rcu_head rcu_head;
85 spinlock_t busylock;
86 u32 limit;
87 };
88
89 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
90 {
91 return (qdisc->__state & __QDISC___STATE_RUNNING) ? true : false;
92 }
93
94 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
95 {
96 if (qdisc_is_running(qdisc))
97 return false;
98 qdisc->__state |= __QDISC___STATE_RUNNING;
99 return true;
100 }
101
102 static inline void qdisc_run_end(struct Qdisc *qdisc)
103 {
104 qdisc->__state &= ~__QDISC___STATE_RUNNING;
105 }
106
107 static inline bool qdisc_is_throttled(const struct Qdisc *qdisc)
108 {
109 return test_bit(__QDISC_STATE_THROTTLED, &qdisc->state) ? true : false;
110 }
111
112 static inline void qdisc_throttled(struct Qdisc *qdisc)
113 {
114 set_bit(__QDISC_STATE_THROTTLED, &qdisc->state);
115 }
116
117 static inline void qdisc_unthrottled(struct Qdisc *qdisc)
118 {
119 clear_bit(__QDISC_STATE_THROTTLED, &qdisc->state);
120 }
121
122 struct Qdisc_class_ops {
123 /* Child qdisc manipulation */
124 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
125 int (*graft)(struct Qdisc *, unsigned long cl,
126 struct Qdisc *, struct Qdisc **);
127 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
128 void (*qlen_notify)(struct Qdisc *, unsigned long);
129
130 /* Class manipulation routines */
131 unsigned long (*get)(struct Qdisc *, u32 classid);
132 void (*put)(struct Qdisc *, unsigned long);
133 int (*change)(struct Qdisc *, u32, u32,
134 struct nlattr **, unsigned long *);
135 int (*delete)(struct Qdisc *, unsigned long);
136 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
137
138 /* Filter manipulation */
139 struct tcf_proto ** (*tcf_chain)(struct Qdisc *, unsigned long);
140 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
141 u32 classid);
142 void (*unbind_tcf)(struct Qdisc *, unsigned long);
143
144 /* rtnetlink specific */
145 int (*dump)(struct Qdisc *, unsigned long,
146 struct sk_buff *skb, struct tcmsg*);
147 int (*dump_stats)(struct Qdisc *, unsigned long,
148 struct gnet_dump *);
149 };
150
151 struct Qdisc_ops {
152 struct Qdisc_ops *next;
153 const struct Qdisc_class_ops *cl_ops;
154 char id[IFNAMSIZ];
155 int priv_size;
156
157 int (*enqueue)(struct sk_buff *, struct Qdisc *);
158 struct sk_buff * (*dequeue)(struct Qdisc *);
159 struct sk_buff * (*peek)(struct Qdisc *);
160 unsigned int (*drop)(struct Qdisc *);
161
162 int (*init)(struct Qdisc *, struct nlattr *arg);
163 void (*reset)(struct Qdisc *);
164 void (*destroy)(struct Qdisc *);
165 int (*change)(struct Qdisc *, struct nlattr *arg);
166 void (*attach)(struct Qdisc *);
167
168 int (*dump)(struct Qdisc *, struct sk_buff *);
169 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
170
171 struct module *owner;
172 };
173
174
175 struct tcf_result {
176 unsigned long class;
177 u32 classid;
178 };
179
180 struct tcf_proto_ops {
181 struct tcf_proto_ops *next;
182 char kind[IFNAMSIZ];
183
184 int (*classify)(struct sk_buff *,
185 const struct tcf_proto *,
186 struct tcf_result *);
187 int (*init)(struct tcf_proto*);
188 void (*destroy)(struct tcf_proto*);
189
190 unsigned long (*get)(struct tcf_proto*, u32 handle);
191 void (*put)(struct tcf_proto*, unsigned long);
192 int (*change)(struct tcf_proto*, unsigned long,
193 u32 handle, struct nlattr **,
194 unsigned long *);
195 int (*delete)(struct tcf_proto*, unsigned long);
196 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
197
198 /* rtnetlink specific */
199 int (*dump)(struct tcf_proto*, unsigned long,
200 struct sk_buff *skb, struct tcmsg*);
201
202 struct module *owner;
203 };
204
205 struct tcf_proto {
206 /* Fast access part */
207 struct tcf_proto *next;
208 void *root;
209 int (*classify)(struct sk_buff *,
210 const struct tcf_proto *,
211 struct tcf_result *);
212 __be16 protocol;
213
214 /* All the rest */
215 u32 prio;
216 u32 classid;
217 struct Qdisc *q;
218 void *data;
219 const struct tcf_proto_ops *ops;
220 };
221
222 struct qdisc_skb_cb {
223 unsigned int pkt_len;
224 long data[];
225 };
226
227 static inline int qdisc_qlen(struct Qdisc *q)
228 {
229 return q->q.qlen;
230 }
231
232 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
233 {
234 return (struct qdisc_skb_cb *)skb->cb;
235 }
236
237 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
238 {
239 return &qdisc->q.lock;
240 }
241
242 static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
243 {
244 return qdisc->dev_queue->qdisc;
245 }
246
247 static inline struct Qdisc *qdisc_root_sleeping(struct Qdisc *qdisc)
248 {
249 return qdisc->dev_queue->qdisc_sleeping;
250 }
251
252 /* The qdisc root lock is a mechanism by which to top level
253 * of a qdisc tree can be locked from any qdisc node in the
254 * forest. This allows changing the configuration of some
255 * aspect of the qdisc tree while blocking out asynchronous
256 * qdisc access in the packet processing paths.
257 *
258 * It is only legal to do this when the root will not change
259 * on us. Otherwise we'll potentially lock the wrong qdisc
260 * root. This is enforced by holding the RTNL semaphore, which
261 * all users of this lock accessor must do.
262 */
263 static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
264 {
265 struct Qdisc *root = qdisc_root(qdisc);
266
267 ASSERT_RTNL();
268 return qdisc_lock(root);
269 }
270
271 static inline spinlock_t *qdisc_root_sleeping_lock(struct Qdisc *qdisc)
272 {
273 struct Qdisc *root = qdisc_root_sleeping(qdisc);
274
275 ASSERT_RTNL();
276 return qdisc_lock(root);
277 }
278
279 static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
280 {
281 return qdisc->dev_queue->dev;
282 }
283
284 static inline void sch_tree_lock(struct Qdisc *q)
285 {
286 spin_lock_bh(qdisc_root_sleeping_lock(q));
287 }
288
289 static inline void sch_tree_unlock(struct Qdisc *q)
290 {
291 spin_unlock_bh(qdisc_root_sleeping_lock(q));
292 }
293
294 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
295 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
296
297 extern struct Qdisc noop_qdisc;
298 extern struct Qdisc_ops noop_qdisc_ops;
299 extern struct Qdisc_ops pfifo_fast_ops;
300 extern struct Qdisc_ops mq_qdisc_ops;
301
302 struct Qdisc_class_common {
303 u32 classid;
304 struct hlist_node hnode;
305 };
306
307 struct Qdisc_class_hash {
308 struct hlist_head *hash;
309 unsigned int hashsize;
310 unsigned int hashmask;
311 unsigned int hashelems;
312 };
313
314 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
315 {
316 id ^= id >> 8;
317 id ^= id >> 4;
318 return id & mask;
319 }
320
321 static inline struct Qdisc_class_common *
322 qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
323 {
324 struct Qdisc_class_common *cl;
325 struct hlist_node *n;
326 unsigned int h;
327
328 h = qdisc_class_hash(id, hash->hashmask);
329 hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
330 if (cl->classid == id)
331 return cl;
332 }
333 return NULL;
334 }
335
336 extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
337 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
338 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
339 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
340 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
341
342 extern void dev_init_scheduler(struct net_device *dev);
343 extern void dev_shutdown(struct net_device *dev);
344 extern void dev_activate(struct net_device *dev);
345 extern void dev_deactivate(struct net_device *dev);
346 extern void dev_deactivate_many(struct list_head *head);
347 extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
348 struct Qdisc *qdisc);
349 extern void qdisc_reset(struct Qdisc *qdisc);
350 extern void qdisc_destroy(struct Qdisc *qdisc);
351 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
352 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
353 struct Qdisc_ops *ops);
354 extern struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
355 struct Qdisc_ops *ops, u32 parentid);
356 extern void __qdisc_calculate_pkt_len(struct sk_buff *skb,
357 const struct qdisc_size_table *stab);
358 extern void tcf_destroy(struct tcf_proto *tp);
359 extern void tcf_destroy_chain(struct tcf_proto **fl);
360
361 /* Reset all TX qdiscs greater then index of a device. */
362 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
363 {
364 struct Qdisc *qdisc;
365
366 for (; i < dev->num_tx_queues; i++) {
367 qdisc = netdev_get_tx_queue(dev, i)->qdisc;
368 if (qdisc) {
369 spin_lock_bh(qdisc_lock(qdisc));
370 qdisc_reset(qdisc);
371 spin_unlock_bh(qdisc_lock(qdisc));
372 }
373 }
374 }
375
376 static inline void qdisc_reset_all_tx(struct net_device *dev)
377 {
378 qdisc_reset_all_tx_gt(dev, 0);
379 }
380
381 /* Are all TX queues of the device empty? */
382 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
383 {
384 unsigned int i;
385 for (i = 0; i < dev->num_tx_queues; i++) {
386 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
387 const struct Qdisc *q = txq->qdisc;
388
389 if (q->q.qlen)
390 return false;
391 }
392 return true;
393 }
394
395 /* Are any of the TX qdiscs changing? */
396 static inline bool qdisc_tx_changing(struct net_device *dev)
397 {
398 unsigned int i;
399 for (i = 0; i < dev->num_tx_queues; i++) {
400 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
401 if (txq->qdisc != txq->qdisc_sleeping)
402 return true;
403 }
404 return false;
405 }
406
407 /* Is the device using the noop qdisc on all queues? */
408 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
409 {
410 unsigned int i;
411 for (i = 0; i < dev->num_tx_queues; i++) {
412 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
413 if (txq->qdisc != &noop_qdisc)
414 return false;
415 }
416 return true;
417 }
418
419 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
420 {
421 return qdisc_skb_cb(skb)->pkt_len;
422 }
423
424 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
425 enum net_xmit_qdisc_t {
426 __NET_XMIT_STOLEN = 0x00010000,
427 __NET_XMIT_BYPASS = 0x00020000,
428 };
429
430 #ifdef CONFIG_NET_CLS_ACT
431 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
432 #else
433 #define net_xmit_drop_count(e) (1)
434 #endif
435
436 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
437 const struct Qdisc *sch)
438 {
439 #ifdef CONFIG_NET_SCHED
440 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
441
442 if (stab)
443 __qdisc_calculate_pkt_len(skb, stab);
444 #endif
445 }
446
447 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
448 {
449 qdisc_calculate_pkt_len(skb, sch);
450 return sch->enqueue(skb, sch);
451 }
452
453 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
454 {
455 qdisc_skb_cb(skb)->pkt_len = skb->len;
456 return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
457 }
458
459
460 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
461 const struct sk_buff *skb)
462 {
463 bstats->bytes += qdisc_pkt_len(skb);
464 bstats->packets += skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1;
465 }
466
467 static inline void qdisc_bstats_update(struct Qdisc *sch,
468 const struct sk_buff *skb)
469 {
470 bstats_update(&sch->bstats, skb);
471 }
472
473 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
474 struct sk_buff_head *list)
475 {
476 __skb_queue_tail(list, skb);
477 sch->qstats.backlog += qdisc_pkt_len(skb);
478
479 return NET_XMIT_SUCCESS;
480 }
481
482 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
483 {
484 return __qdisc_enqueue_tail(skb, sch, &sch->q);
485 }
486
487 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
488 struct sk_buff_head *list)
489 {
490 struct sk_buff *skb = __skb_dequeue(list);
491
492 if (likely(skb != NULL)) {
493 sch->qstats.backlog -= qdisc_pkt_len(skb);
494 qdisc_bstats_update(sch, skb);
495 }
496
497 return skb;
498 }
499
500 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
501 {
502 return __qdisc_dequeue_head(sch, &sch->q);
503 }
504
505 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
506 struct sk_buff_head *list)
507 {
508 struct sk_buff *skb = __skb_dequeue(list);
509
510 if (likely(skb != NULL)) {
511 unsigned int len = qdisc_pkt_len(skb);
512 sch->qstats.backlog -= len;
513 kfree_skb(skb);
514 return len;
515 }
516
517 return 0;
518 }
519
520 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch)
521 {
522 return __qdisc_queue_drop_head(sch, &sch->q);
523 }
524
525 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
526 struct sk_buff_head *list)
527 {
528 struct sk_buff *skb = __skb_dequeue_tail(list);
529
530 if (likely(skb != NULL))
531 sch->qstats.backlog -= qdisc_pkt_len(skb);
532
533 return skb;
534 }
535
536 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
537 {
538 return __qdisc_dequeue_tail(sch, &sch->q);
539 }
540
541 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
542 {
543 return skb_peek(&sch->q);
544 }
545
546 /* generic pseudo peek method for non-work-conserving qdisc */
547 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
548 {
549 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
550 if (!sch->gso_skb) {
551 sch->gso_skb = sch->dequeue(sch);
552 if (sch->gso_skb)
553 /* it's still part of the queue */
554 sch->q.qlen++;
555 }
556
557 return sch->gso_skb;
558 }
559
560 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
561 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
562 {
563 struct sk_buff *skb = sch->gso_skb;
564
565 if (skb) {
566 sch->gso_skb = NULL;
567 sch->q.qlen--;
568 } else {
569 skb = sch->dequeue(sch);
570 }
571
572 return skb;
573 }
574
575 static inline void __qdisc_reset_queue(struct Qdisc *sch,
576 struct sk_buff_head *list)
577 {
578 /*
579 * We do not know the backlog in bytes of this list, it
580 * is up to the caller to correct it
581 */
582 __skb_queue_purge(list);
583 }
584
585 static inline void qdisc_reset_queue(struct Qdisc *sch)
586 {
587 __qdisc_reset_queue(sch, &sch->q);
588 sch->qstats.backlog = 0;
589 }
590
591 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
592 struct sk_buff_head *list)
593 {
594 struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
595
596 if (likely(skb != NULL)) {
597 unsigned int len = qdisc_pkt_len(skb);
598 kfree_skb(skb);
599 return len;
600 }
601
602 return 0;
603 }
604
605 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
606 {
607 return __qdisc_queue_drop(sch, &sch->q);
608 }
609
610 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
611 {
612 kfree_skb(skb);
613 sch->qstats.drops++;
614
615 return NET_XMIT_DROP;
616 }
617
618 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
619 {
620 sch->qstats.drops++;
621
622 #ifdef CONFIG_NET_CLS_ACT
623 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
624 goto drop;
625
626 return NET_XMIT_SUCCESS;
627
628 drop:
629 #endif
630 kfree_skb(skb);
631 return NET_XMIT_DROP;
632 }
633
634 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
635 long it will take to send a packet given its size.
636 */
637 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
638 {
639 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
640 if (slot < 0)
641 slot = 0;
642 slot >>= rtab->rate.cell_log;
643 if (slot > 255)
644 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
645 return rtab->data[slot];
646 }
647
648 #ifdef CONFIG_NET_CLS_ACT
649 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask,
650 int action)
651 {
652 struct sk_buff *n;
653
654 n = skb_clone(skb, gfp_mask);
655
656 if (n) {
657 n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
658 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
659 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
660 }
661 return n;
662 }
663 #endif
664
665 #endif
kbuild development tree