]> bbs.cooldavid.org Git - net-next-2.6.git/blame - include/net/sch_generic.h
net_sched: fix estimator lock selection for mq child qdiscs
[net-next-2.6.git] / include / net / sch_generic.h
CommitLineData
1da177e4
LT
1#ifndef __NET_SCHED_GENERIC_H
2#define __NET_SCHED_GENERIC_H
3
1da177e4
LT
4#include <linux/netdevice.h>
5#include <linux/types.h>
6#include <linux/rcupdate.h>
7#include <linux/module.h>
1da177e4
LT
8#include <linux/pkt_sched.h>
9#include <linux/pkt_cls.h>
10#include <net/gen_stats.h>
be577ddc 11#include <net/rtnetlink.h>
1da177e4
LT
12
13struct Qdisc_ops;
14struct qdisc_walker;
15struct tcf_walker;
16struct module;
17
18struct qdisc_rate_table
19{
20 struct tc_ratespec rate;
21 u32 data[256];
22 struct qdisc_rate_table *next;
23 int refcnt;
24};
25
e2627c8c
DM
26enum qdisc_state_t
27{
28 __QDISC_STATE_RUNNING,
37437bb2 29 __QDISC_STATE_SCHED,
a9312ae8 30 __QDISC_STATE_DEACTIVATED,
e2627c8c
DM
31};
32
175f9c1b
JK
33struct qdisc_size_table {
34 struct list_head list;
35 struct tc_sizespec szopts;
36 int refcnt;
37 u16 data[];
38};
39
1da177e4
LT
40struct Qdisc
41{
42 int (*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
43 struct sk_buff * (*dequeue)(struct Qdisc *dev);
44 unsigned flags;
b00355db
JP
45#define TCQ_F_BUILTIN 1
46#define TCQ_F_THROTTLED 2
47#define TCQ_F_INGRESS 4
bbd8a0d3 48#define TCQ_F_CAN_BYPASS 8
23bcf634 49#define TCQ_F_MQROOT 16
b00355db 50#define TCQ_F_WARN_NONWC (1 << 16)
1da177e4
LT
51 int padded;
52 struct Qdisc_ops *ops;
175f9c1b 53 struct qdisc_size_table *stab;
5e140dfc 54 struct list_head list;
1da177e4
LT
55 u32 handle;
56 u32 parent;
57 atomic_t refcnt;
1da177e4 58 struct gnet_stats_rate_est rate_est;
1da177e4
LT
59 int (*reshape_fail)(struct sk_buff *skb,
60 struct Qdisc *q);
61
72b25a91
DM
62 void *u32_node;
63
1da177e4
LT
64 /* This field is deprecated, but it is still used by CBQ
65 * and it will live until better solution will be invented.
66 */
67 struct Qdisc *__parent;
5e140dfc
ED
68 struct netdev_queue *dev_queue;
69 struct Qdisc *next_sched;
70
71 struct sk_buff *gso_skb;
72 /*
73 * For performance sake on SMP, we put highly modified fields at the end
74 */
75 unsigned long state;
76 struct sk_buff_head q;
c1a8f1f1 77 struct gnet_stats_basic_packed bstats;
5e140dfc 78 struct gnet_stats_queue qstats;
1da177e4
LT
79};
80
81struct Qdisc_class_ops
82{
83 /* Child qdisc manipulation */
6ec1c69a 84 unsigned int (*select_queue)(struct Qdisc *, struct tcmsg *);
1da177e4
LT
85 int (*graft)(struct Qdisc *, unsigned long cl,
86 struct Qdisc *, struct Qdisc **);
87 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
43effa1e 88 void (*qlen_notify)(struct Qdisc *, unsigned long);
1da177e4
LT
89
90 /* Class manipulation routines */
91 unsigned long (*get)(struct Qdisc *, u32 classid);
92 void (*put)(struct Qdisc *, unsigned long);
93 int (*change)(struct Qdisc *, u32, u32,
1e90474c 94 struct nlattr **, unsigned long *);
1da177e4
LT
95 int (*delete)(struct Qdisc *, unsigned long);
96 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
97
98 /* Filter manipulation */
99 struct tcf_proto ** (*tcf_chain)(struct Qdisc *, unsigned long);
100 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
101 u32 classid);
102 void (*unbind_tcf)(struct Qdisc *, unsigned long);
103
104 /* rtnetlink specific */
105 int (*dump)(struct Qdisc *, unsigned long,
106 struct sk_buff *skb, struct tcmsg*);
107 int (*dump_stats)(struct Qdisc *, unsigned long,
108 struct gnet_dump *);
109};
110
111struct Qdisc_ops
112{
113 struct Qdisc_ops *next;
20fea08b 114 const struct Qdisc_class_ops *cl_ops;
1da177e4
LT
115 char id[IFNAMSIZ];
116 int priv_size;
117
118 int (*enqueue)(struct sk_buff *, struct Qdisc *);
119 struct sk_buff * (*dequeue)(struct Qdisc *);
90d841fd 120 struct sk_buff * (*peek)(struct Qdisc *);
1da177e4
LT
121 unsigned int (*drop)(struct Qdisc *);
122
1e90474c 123 int (*init)(struct Qdisc *, struct nlattr *arg);
1da177e4
LT
124 void (*reset)(struct Qdisc *);
125 void (*destroy)(struct Qdisc *);
1e90474c 126 int (*change)(struct Qdisc *, struct nlattr *arg);
6ec1c69a 127 void (*attach)(struct Qdisc *);
1da177e4
LT
128
129 int (*dump)(struct Qdisc *, struct sk_buff *);
130 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
131
132 struct module *owner;
133};
134
135
136struct tcf_result
137{
138 unsigned long class;
139 u32 classid;
140};
141
142struct tcf_proto_ops
143{
144 struct tcf_proto_ops *next;
145 char kind[IFNAMSIZ];
146
147 int (*classify)(struct sk_buff*, struct tcf_proto*,
148 struct tcf_result *);
149 int (*init)(struct tcf_proto*);
150 void (*destroy)(struct tcf_proto*);
151
152 unsigned long (*get)(struct tcf_proto*, u32 handle);
153 void (*put)(struct tcf_proto*, unsigned long);
154 int (*change)(struct tcf_proto*, unsigned long,
add93b61 155 u32 handle, struct nlattr **,
1da177e4
LT
156 unsigned long *);
157 int (*delete)(struct tcf_proto*, unsigned long);
158 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
159
160 /* rtnetlink specific */
161 int (*dump)(struct tcf_proto*, unsigned long,
162 struct sk_buff *skb, struct tcmsg*);
163
164 struct module *owner;
165};
166
167struct tcf_proto
168{
169 /* Fast access part */
170 struct tcf_proto *next;
171 void *root;
172 int (*classify)(struct sk_buff*, struct tcf_proto*,
173 struct tcf_result *);
66c6f529 174 __be16 protocol;
1da177e4
LT
175
176 /* All the rest */
177 u32 prio;
178 u32 classid;
179 struct Qdisc *q;
180 void *data;
181 struct tcf_proto_ops *ops;
182};
183
175f9c1b
JK
184struct qdisc_skb_cb {
185 unsigned int pkt_len;
186 char data[];
187};
188
bbd8a0d3
KK
189static inline int qdisc_qlen(struct Qdisc *q)
190{
191 return q->q.qlen;
192}
193
175f9c1b
JK
194static inline struct qdisc_skb_cb *qdisc_skb_cb(struct sk_buff *skb)
195{
196 return (struct qdisc_skb_cb *)skb->cb;
197}
198
83874000
DM
199static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
200{
201 return &qdisc->q.lock;
202}
203
7698b4fc
DM
204static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
205{
206 return qdisc->dev_queue->qdisc;
207}
208
2540e051
JP
209static inline struct Qdisc *qdisc_root_sleeping(struct Qdisc *qdisc)
210{
211 return qdisc->dev_queue->qdisc_sleeping;
212}
213
7e43f112
DM
214/* The qdisc root lock is a mechanism by which to top level
215 * of a qdisc tree can be locked from any qdisc node in the
216 * forest. This allows changing the configuration of some
217 * aspect of the qdisc tree while blocking out asynchronous
218 * qdisc access in the packet processing paths.
219 *
220 * It is only legal to do this when the root will not change
221 * on us. Otherwise we'll potentially lock the wrong qdisc
222 * root. This is enforced by holding the RTNL semaphore, which
223 * all users of this lock accessor must do.
224 */
7698b4fc
DM
225static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
226{
227 struct Qdisc *root = qdisc_root(qdisc);
228
7e43f112 229 ASSERT_RTNL();
83874000 230 return qdisc_lock(root);
7698b4fc
DM
231}
232
f6f9b93f
JP
233static inline spinlock_t *qdisc_root_sleeping_lock(struct Qdisc *qdisc)
234{
235 struct Qdisc *root = qdisc_root_sleeping(qdisc);
236
237 ASSERT_RTNL();
238 return qdisc_lock(root);
239}
240
5ce2d488
DM
241static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
242{
243 return qdisc->dev_queue->dev;
244}
1da177e4 245
78a5b30b
DM
246static inline void sch_tree_lock(struct Qdisc *q)
247{
fe439dd0 248 spin_lock_bh(qdisc_root_sleeping_lock(q));
78a5b30b
DM
249}
250
251static inline void sch_tree_unlock(struct Qdisc *q)
252{
fe439dd0 253 spin_unlock_bh(qdisc_root_sleeping_lock(q));
78a5b30b
DM
254}
255
256#define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
257#define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
1da177e4 258
e41a33e6
TG
259extern struct Qdisc noop_qdisc;
260extern struct Qdisc_ops noop_qdisc_ops;
6ec1c69a
DM
261extern struct Qdisc_ops pfifo_fast_ops;
262extern struct Qdisc_ops mq_qdisc_ops;
e41a33e6 263
6fe1c7a5
PM
264struct Qdisc_class_common
265{
266 u32 classid;
267 struct hlist_node hnode;
268};
269
270struct Qdisc_class_hash
271{
272 struct hlist_head *hash;
273 unsigned int hashsize;
274 unsigned int hashmask;
275 unsigned int hashelems;
276};
277
278static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
279{
280 id ^= id >> 8;
281 id ^= id >> 4;
282 return id & mask;
283}
284
285static inline struct Qdisc_class_common *
286qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
287{
288 struct Qdisc_class_common *cl;
289 struct hlist_node *n;
290 unsigned int h;
291
292 h = qdisc_class_hash(id, hash->hashmask);
293 hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
294 if (cl->classid == id)
295 return cl;
296 }
297 return NULL;
298}
299
300extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
301extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
302extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
303extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
304extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
305
e41a33e6
TG
306extern void dev_init_scheduler(struct net_device *dev);
307extern void dev_shutdown(struct net_device *dev);
308extern void dev_activate(struct net_device *dev);
309extern void dev_deactivate(struct net_device *dev);
589983cd
PM
310extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
311 struct Qdisc *qdisc);
e41a33e6
TG
312extern void qdisc_reset(struct Qdisc *qdisc);
313extern void qdisc_destroy(struct Qdisc *qdisc);
43effa1e 314extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
5ce2d488 315extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
bb949fbd 316 struct Qdisc_ops *ops);
e41a33e6 317extern struct Qdisc *qdisc_create_dflt(struct net_device *dev,
bb949fbd 318 struct netdev_queue *dev_queue,
9f9afec4 319 struct Qdisc_ops *ops, u32 parentid);
175f9c1b
JK
320extern void qdisc_calculate_pkt_len(struct sk_buff *skb,
321 struct qdisc_size_table *stab);
a48b5a61 322extern void tcf_destroy(struct tcf_proto *tp);
ff31ab56 323extern void tcf_destroy_chain(struct tcf_proto **fl);
1da177e4 324
5aa70995
DM
325/* Reset all TX qdiscs of a device. */
326static inline void qdisc_reset_all_tx(struct net_device *dev)
327{
e8a0464c
DM
328 unsigned int i;
329 for (i = 0; i < dev->num_tx_queues; i++)
330 qdisc_reset(netdev_get_tx_queue(dev, i)->qdisc);
5aa70995
DM
331}
332
3e745dd6
DM
333/* Are all TX queues of the device empty? */
334static inline bool qdisc_all_tx_empty(const struct net_device *dev)
335{
e8a0464c
DM
336 unsigned int i;
337 for (i = 0; i < dev->num_tx_queues; i++) {
338 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
339 const struct Qdisc *q = txq->qdisc;
3e745dd6 340
e8a0464c
DM
341 if (q->q.qlen)
342 return false;
343 }
344 return true;
3e745dd6
DM
345}
346
6fa9864b
DM
347/* Are any of the TX qdiscs changing? */
348static inline bool qdisc_tx_changing(struct net_device *dev)
349{
e8a0464c
DM
350 unsigned int i;
351 for (i = 0; i < dev->num_tx_queues; i++) {
352 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
353 if (txq->qdisc != txq->qdisc_sleeping)
354 return true;
355 }
356 return false;
6fa9864b
DM
357}
358
e8a0464c 359/* Is the device using the noop qdisc on all queues? */
05297949
DM
360static inline bool qdisc_tx_is_noop(const struct net_device *dev)
361{
e8a0464c
DM
362 unsigned int i;
363 for (i = 0; i < dev->num_tx_queues; i++) {
364 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
365 if (txq->qdisc != &noop_qdisc)
366 return false;
367 }
368 return true;
05297949
DM
369}
370
0abf77e5
JK
371static inline unsigned int qdisc_pkt_len(struct sk_buff *skb)
372{
175f9c1b 373 return qdisc_skb_cb(skb)->pkt_len;
0abf77e5
JK
374}
375
c27f339a 376/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
378a2f09
JP
377enum net_xmit_qdisc_t {
378 __NET_XMIT_STOLEN = 0x00010000,
c27f339a 379 __NET_XMIT_BYPASS = 0x00020000,
378a2f09
JP
380};
381
c27f339a 382#ifdef CONFIG_NET_CLS_ACT
378a2f09 383#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
378a2f09
JP
384#else
385#define net_xmit_drop_count(e) (1)
386#endif
387
5f86173b
JK
388static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
389{
3a682fbd 390#ifdef CONFIG_NET_SCHED
175f9c1b
JK
391 if (sch->stab)
392 qdisc_calculate_pkt_len(skb, sch->stab);
3a682fbd 393#endif
5f86173b
JK
394 return sch->enqueue(skb, sch);
395}
396
397static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
398{
175f9c1b 399 qdisc_skb_cb(skb)->pkt_len = skb->len;
378a2f09 400 return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
5f86173b
JK
401}
402
bbd8a0d3
KK
403static inline void __qdisc_update_bstats(struct Qdisc *sch, unsigned int len)
404{
405 sch->bstats.bytes += len;
406 sch->bstats.packets++;
407}
408
9972b25d
TG
409static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
410 struct sk_buff_head *list)
411{
412 __skb_queue_tail(list, skb);
0abf77e5 413 sch->qstats.backlog += qdisc_pkt_len(skb);
bbd8a0d3 414 __qdisc_update_bstats(sch, qdisc_pkt_len(skb));
9972b25d
TG
415
416 return NET_XMIT_SUCCESS;
417}
418
419static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
420{
421 return __qdisc_enqueue_tail(skb, sch, &sch->q);
422}
423
424static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
425 struct sk_buff_head *list)
426{
427 struct sk_buff *skb = __skb_dequeue(list);
428
429 if (likely(skb != NULL))
0abf77e5 430 sch->qstats.backlog -= qdisc_pkt_len(skb);
9972b25d
TG
431
432 return skb;
433}
434
435static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
436{
437 return __qdisc_dequeue_head(sch, &sch->q);
438}
439
440static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
441 struct sk_buff_head *list)
442{
443 struct sk_buff *skb = __skb_dequeue_tail(list);
444
445 if (likely(skb != NULL))
0abf77e5 446 sch->qstats.backlog -= qdisc_pkt_len(skb);
9972b25d
TG
447
448 return skb;
449}
450
451static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
452{
453 return __qdisc_dequeue_tail(sch, &sch->q);
454}
455
48a8f519
PM
456static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
457{
458 return skb_peek(&sch->q);
459}
460
77be155c
JP
461/* generic pseudo peek method for non-work-conserving qdisc */
462static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
463{
464 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
61c9eaf9 465 if (!sch->gso_skb) {
77be155c 466 sch->gso_skb = sch->dequeue(sch);
61c9eaf9
JP
467 if (sch->gso_skb)
468 /* it's still part of the queue */
469 sch->q.qlen++;
470 }
77be155c
JP
471
472 return sch->gso_skb;
473}
474
475/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
476static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
477{
478 struct sk_buff *skb = sch->gso_skb;
479
61c9eaf9 480 if (skb) {
77be155c 481 sch->gso_skb = NULL;
61c9eaf9
JP
482 sch->q.qlen--;
483 } else {
77be155c 484 skb = sch->dequeue(sch);
61c9eaf9 485 }
77be155c
JP
486
487 return skb;
488}
489
9972b25d
TG
490static inline void __qdisc_reset_queue(struct Qdisc *sch,
491 struct sk_buff_head *list)
492{
493 /*
494 * We do not know the backlog in bytes of this list, it
495 * is up to the caller to correct it
496 */
93245dd6 497 __skb_queue_purge(list);
9972b25d
TG
498}
499
500static inline void qdisc_reset_queue(struct Qdisc *sch)
501{
502 __qdisc_reset_queue(sch, &sch->q);
503 sch->qstats.backlog = 0;
504}
505
506static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
507 struct sk_buff_head *list)
508{
509 struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
510
511 if (likely(skb != NULL)) {
0abf77e5 512 unsigned int len = qdisc_pkt_len(skb);
9972b25d
TG
513 kfree_skb(skb);
514 return len;
515 }
516
517 return 0;
518}
519
520static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
521{
522 return __qdisc_queue_drop(sch, &sch->q);
523}
524
525static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
526{
527 kfree_skb(skb);
528 sch->qstats.drops++;
529
530 return NET_XMIT_DROP;
531}
532
533static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
534{
535 sch->qstats.drops++;
536
c3bc7cff 537#ifdef CONFIG_NET_CLS_ACT
9972b25d
TG
538 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
539 goto drop;
540
541 return NET_XMIT_SUCCESS;
542
543drop:
544#endif
545 kfree_skb(skb);
546 return NET_XMIT_DROP;
547}
548
e9bef55d
JDB
549/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
550 long it will take to send a packet given its size.
551 */
552static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
553{
e08b0998
JDB
554 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
555 if (slot < 0)
556 slot = 0;
e9bef55d
JDB
557 slot >>= rtab->rate.cell_log;
558 if (slot > 255)
559 return (rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]);
560 return rtab->data[slot];
561}
562
12da81d1
JHS
563#ifdef CONFIG_NET_CLS_ACT
564static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask)
565{
566 struct sk_buff *n = skb_clone(skb, gfp_mask);
567
568 if (n) {
569 n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
570 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
571 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
12da81d1
JHS
572 }
573 return n;
574}
575#endif
576
1da177e4 577#endif