]> bbs.cooldavid.org Git - net-next-2.6.git/blame - include/net/sch_generic.h
pkt_sched: Make qdisc->gso_skb a list.
[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;
45#define TCQ_F_BUILTIN 1
46#define TCQ_F_THROTTLED 2
47#define TCQ_F_INGRESS 4
48 int padded;
49 struct Qdisc_ops *ops;
175f9c1b 50 struct qdisc_size_table *stab;
1da177e4
LT
51 u32 handle;
52 u32 parent;
53 atomic_t refcnt;
e2627c8c 54 unsigned long state;
242f8bfe 55 struct sk_buff_head requeue;
1da177e4 56 struct sk_buff_head q;
bb949fbd 57 struct netdev_queue *dev_queue;
37437bb2 58 struct Qdisc *next_sched;
1da177e4
LT
59 struct list_head list;
60
61 struct gnet_stats_basic bstats;
62 struct gnet_stats_queue qstats;
63 struct gnet_stats_rate_est rate_est;
1da177e4
LT
64 int (*reshape_fail)(struct sk_buff *skb,
65 struct Qdisc *q);
66
72b25a91
DM
67 void *u32_node;
68
1da177e4
LT
69 /* This field is deprecated, but it is still used by CBQ
70 * and it will live until better solution will be invented.
71 */
72 struct Qdisc *__parent;
73};
74
75struct Qdisc_class_ops
76{
77 /* Child qdisc manipulation */
78 int (*graft)(struct Qdisc *, unsigned long cl,
79 struct Qdisc *, struct Qdisc **);
80 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
43effa1e 81 void (*qlen_notify)(struct Qdisc *, unsigned long);
1da177e4
LT
82
83 /* Class manipulation routines */
84 unsigned long (*get)(struct Qdisc *, u32 classid);
85 void (*put)(struct Qdisc *, unsigned long);
86 int (*change)(struct Qdisc *, u32, u32,
1e90474c 87 struct nlattr **, unsigned long *);
1da177e4
LT
88 int (*delete)(struct Qdisc *, unsigned long);
89 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
90
91 /* Filter manipulation */
92 struct tcf_proto ** (*tcf_chain)(struct Qdisc *, unsigned long);
93 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
94 u32 classid);
95 void (*unbind_tcf)(struct Qdisc *, unsigned long);
96
97 /* rtnetlink specific */
98 int (*dump)(struct Qdisc *, unsigned long,
99 struct sk_buff *skb, struct tcmsg*);
100 int (*dump_stats)(struct Qdisc *, unsigned long,
101 struct gnet_dump *);
102};
103
104struct Qdisc_ops
105{
106 struct Qdisc_ops *next;
20fea08b 107 const struct Qdisc_class_ops *cl_ops;
1da177e4
LT
108 char id[IFNAMSIZ];
109 int priv_size;
110
111 int (*enqueue)(struct sk_buff *, struct Qdisc *);
112 struct sk_buff * (*dequeue)(struct Qdisc *);
113 int (*requeue)(struct sk_buff *, struct Qdisc *);
114 unsigned int (*drop)(struct Qdisc *);
115
1e90474c 116 int (*init)(struct Qdisc *, struct nlattr *arg);
1da177e4
LT
117 void (*reset)(struct Qdisc *);
118 void (*destroy)(struct Qdisc *);
1e90474c 119 int (*change)(struct Qdisc *, struct nlattr *arg);
1da177e4
LT
120
121 int (*dump)(struct Qdisc *, struct sk_buff *);
122 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
123
124 struct module *owner;
125};
126
127
128struct tcf_result
129{
130 unsigned long class;
131 u32 classid;
132};
133
134struct tcf_proto_ops
135{
136 struct tcf_proto_ops *next;
137 char kind[IFNAMSIZ];
138
139 int (*classify)(struct sk_buff*, struct tcf_proto*,
140 struct tcf_result *);
141 int (*init)(struct tcf_proto*);
142 void (*destroy)(struct tcf_proto*);
143
144 unsigned long (*get)(struct tcf_proto*, u32 handle);
145 void (*put)(struct tcf_proto*, unsigned long);
146 int (*change)(struct tcf_proto*, unsigned long,
add93b61 147 u32 handle, struct nlattr **,
1da177e4
LT
148 unsigned long *);
149 int (*delete)(struct tcf_proto*, unsigned long);
150 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
151
152 /* rtnetlink specific */
153 int (*dump)(struct tcf_proto*, unsigned long,
154 struct sk_buff *skb, struct tcmsg*);
155
156 struct module *owner;
157};
158
159struct tcf_proto
160{
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 *);
66c6f529 166 __be16 protocol;
1da177e4
LT
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
175f9c1b
JK
176struct qdisc_skb_cb {
177 unsigned int pkt_len;
178 char data[];
179};
180
181static inline struct qdisc_skb_cb *qdisc_skb_cb(struct sk_buff *skb)
182{
183 return (struct qdisc_skb_cb *)skb->cb;
184}
185
83874000
DM
186static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
187{
188 return &qdisc->q.lock;
189}
190
7698b4fc
DM
191static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
192{
193 return qdisc->dev_queue->qdisc;
194}
195
2540e051
JP
196static inline struct Qdisc *qdisc_root_sleeping(struct Qdisc *qdisc)
197{
198 return qdisc->dev_queue->qdisc_sleeping;
199}
200
7e43f112
DM
201/* The qdisc root lock is a mechanism by which to top level
202 * of a qdisc tree can be locked from any qdisc node in the
203 * forest. This allows changing the configuration of some
204 * aspect of the qdisc tree while blocking out asynchronous
205 * qdisc access in the packet processing paths.
206 *
207 * It is only legal to do this when the root will not change
208 * on us. Otherwise we'll potentially lock the wrong qdisc
209 * root. This is enforced by holding the RTNL semaphore, which
210 * all users of this lock accessor must do.
211 */
7698b4fc
DM
212static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
213{
214 struct Qdisc *root = qdisc_root(qdisc);
215
7e43f112 216 ASSERT_RTNL();
83874000 217 return qdisc_lock(root);
7698b4fc
DM
218}
219
f6f9b93f
JP
220static inline spinlock_t *qdisc_root_sleeping_lock(struct Qdisc *qdisc)
221{
222 struct Qdisc *root = qdisc_root_sleeping(qdisc);
223
224 ASSERT_RTNL();
225 return qdisc_lock(root);
226}
227
5ce2d488
DM
228static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
229{
230 return qdisc->dev_queue->dev;
231}
1da177e4 232
78a5b30b
DM
233static inline void sch_tree_lock(struct Qdisc *q)
234{
fe439dd0 235 spin_lock_bh(qdisc_root_sleeping_lock(q));
78a5b30b
DM
236}
237
238static inline void sch_tree_unlock(struct Qdisc *q)
239{
fe439dd0 240 spin_unlock_bh(qdisc_root_sleeping_lock(q));
78a5b30b
DM
241}
242
243#define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
244#define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
1da177e4 245
e41a33e6
TG
246extern struct Qdisc noop_qdisc;
247extern struct Qdisc_ops noop_qdisc_ops;
248
6fe1c7a5
PM
249struct Qdisc_class_common
250{
251 u32 classid;
252 struct hlist_node hnode;
253};
254
255struct Qdisc_class_hash
256{
257 struct hlist_head *hash;
258 unsigned int hashsize;
259 unsigned int hashmask;
260 unsigned int hashelems;
261};
262
263static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
264{
265 id ^= id >> 8;
266 id ^= id >> 4;
267 return id & mask;
268}
269
270static inline struct Qdisc_class_common *
271qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
272{
273 struct Qdisc_class_common *cl;
274 struct hlist_node *n;
275 unsigned int h;
276
277 h = qdisc_class_hash(id, hash->hashmask);
278 hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
279 if (cl->classid == id)
280 return cl;
281 }
282 return NULL;
283}
284
285extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
286extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
287extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
288extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
289extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
290
e41a33e6
TG
291extern void dev_init_scheduler(struct net_device *dev);
292extern void dev_shutdown(struct net_device *dev);
293extern void dev_activate(struct net_device *dev);
294extern void dev_deactivate(struct net_device *dev);
295extern void qdisc_reset(struct Qdisc *qdisc);
296extern void qdisc_destroy(struct Qdisc *qdisc);
43effa1e 297extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
5ce2d488 298extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
bb949fbd 299 struct Qdisc_ops *ops);
e41a33e6 300extern struct Qdisc *qdisc_create_dflt(struct net_device *dev,
bb949fbd 301 struct netdev_queue *dev_queue,
9f9afec4 302 struct Qdisc_ops *ops, u32 parentid);
175f9c1b
JK
303extern void qdisc_calculate_pkt_len(struct sk_buff *skb,
304 struct qdisc_size_table *stab);
a48b5a61 305extern void tcf_destroy(struct tcf_proto *tp);
ff31ab56 306extern void tcf_destroy_chain(struct tcf_proto **fl);
1da177e4 307
5aa70995
DM
308/* Reset all TX qdiscs of a device. */
309static inline void qdisc_reset_all_tx(struct net_device *dev)
310{
e8a0464c
DM
311 unsigned int i;
312 for (i = 0; i < dev->num_tx_queues; i++)
313 qdisc_reset(netdev_get_tx_queue(dev, i)->qdisc);
5aa70995
DM
314}
315
3e745dd6
DM
316/* Are all TX queues of the device empty? */
317static inline bool qdisc_all_tx_empty(const struct net_device *dev)
318{
e8a0464c
DM
319 unsigned int i;
320 for (i = 0; i < dev->num_tx_queues; i++) {
321 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
322 const struct Qdisc *q = txq->qdisc;
3e745dd6 323
e8a0464c
DM
324 if (q->q.qlen)
325 return false;
326 }
327 return true;
3e745dd6
DM
328}
329
6fa9864b
DM
330/* Are any of the TX qdiscs changing? */
331static inline bool qdisc_tx_changing(struct net_device *dev)
332{
e8a0464c
DM
333 unsigned int i;
334 for (i = 0; i < dev->num_tx_queues; i++) {
335 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
336 if (txq->qdisc != txq->qdisc_sleeping)
337 return true;
338 }
339 return false;
6fa9864b
DM
340}
341
e8a0464c 342/* Is the device using the noop qdisc on all queues? */
05297949
DM
343static inline bool qdisc_tx_is_noop(const struct net_device *dev)
344{
e8a0464c
DM
345 unsigned int i;
346 for (i = 0; i < dev->num_tx_queues; i++) {
347 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
348 if (txq->qdisc != &noop_qdisc)
349 return false;
350 }
351 return true;
05297949
DM
352}
353
0abf77e5
JK
354static inline unsigned int qdisc_pkt_len(struct sk_buff *skb)
355{
175f9c1b 356 return qdisc_skb_cb(skb)->pkt_len;
0abf77e5
JK
357}
358
c27f339a 359/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
378a2f09
JP
360enum net_xmit_qdisc_t {
361 __NET_XMIT_STOLEN = 0x00010000,
c27f339a 362 __NET_XMIT_BYPASS = 0x00020000,
378a2f09
JP
363};
364
c27f339a 365#ifdef CONFIG_NET_CLS_ACT
378a2f09 366#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
378a2f09
JP
367#else
368#define net_xmit_drop_count(e) (1)
369#endif
370
5f86173b
JK
371static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
372{
3a682fbd 373#ifdef CONFIG_NET_SCHED
175f9c1b
JK
374 if (sch->stab)
375 qdisc_calculate_pkt_len(skb, sch->stab);
3a682fbd 376#endif
5f86173b
JK
377 return sch->enqueue(skb, sch);
378}
379
380static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
381{
175f9c1b 382 qdisc_skb_cb(skb)->pkt_len = skb->len;
378a2f09 383 return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
5f86173b
JK
384}
385
9972b25d
TG
386static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
387 struct sk_buff_head *list)
388{
389 __skb_queue_tail(list, skb);
0abf77e5
JK
390 sch->qstats.backlog += qdisc_pkt_len(skb);
391 sch->bstats.bytes += qdisc_pkt_len(skb);
9972b25d
TG
392 sch->bstats.packets++;
393
394 return NET_XMIT_SUCCESS;
395}
396
397static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
398{
399 return __qdisc_enqueue_tail(skb, sch, &sch->q);
400}
401
402static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
403 struct sk_buff_head *list)
404{
405 struct sk_buff *skb = __skb_dequeue(list);
406
407 if (likely(skb != NULL))
0abf77e5 408 sch->qstats.backlog -= qdisc_pkt_len(skb);
9972b25d
TG
409
410 return skb;
411}
412
413static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
414{
415 return __qdisc_dequeue_head(sch, &sch->q);
416}
417
418static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
419 struct sk_buff_head *list)
420{
421 struct sk_buff *skb = __skb_dequeue_tail(list);
422
423 if (likely(skb != NULL))
0abf77e5 424 sch->qstats.backlog -= qdisc_pkt_len(skb);
9972b25d
TG
425
426 return skb;
427}
428
429static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
430{
431 return __qdisc_dequeue_tail(sch, &sch->q);
432}
433
434static inline int __qdisc_requeue(struct sk_buff *skb, struct Qdisc *sch,
435 struct sk_buff_head *list)
436{
437 __skb_queue_head(list, skb);
0abf77e5 438 sch->qstats.backlog += qdisc_pkt_len(skb);
9972b25d
TG
439 sch->qstats.requeues++;
440
441 return NET_XMIT_SUCCESS;
442}
443
444static inline int qdisc_requeue(struct sk_buff *skb, struct Qdisc *sch)
445{
446 return __qdisc_requeue(skb, sch, &sch->q);
447}
448
449static inline void __qdisc_reset_queue(struct Qdisc *sch,
450 struct sk_buff_head *list)
451{
452 /*
453 * We do not know the backlog in bytes of this list, it
454 * is up to the caller to correct it
455 */
93245dd6 456 __skb_queue_purge(list);
9972b25d
TG
457}
458
459static inline void qdisc_reset_queue(struct Qdisc *sch)
460{
461 __qdisc_reset_queue(sch, &sch->q);
462 sch->qstats.backlog = 0;
463}
464
465static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
466 struct sk_buff_head *list)
467{
468 struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
469
470 if (likely(skb != NULL)) {
0abf77e5 471 unsigned int len = qdisc_pkt_len(skb);
9972b25d
TG
472 kfree_skb(skb);
473 return len;
474 }
475
476 return 0;
477}
478
479static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
480{
481 return __qdisc_queue_drop(sch, &sch->q);
482}
483
484static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
485{
486 kfree_skb(skb);
487 sch->qstats.drops++;
488
489 return NET_XMIT_DROP;
490}
491
492static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
493{
494 sch->qstats.drops++;
495
c3bc7cff 496#ifdef CONFIG_NET_CLS_ACT
9972b25d
TG
497 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
498 goto drop;
499
500 return NET_XMIT_SUCCESS;
501
502drop:
503#endif
504 kfree_skb(skb);
505 return NET_XMIT_DROP;
506}
507
e9bef55d
JDB
508/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
509 long it will take to send a packet given its size.
510 */
511static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
512{
e08b0998
JDB
513 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
514 if (slot < 0)
515 slot = 0;
e9bef55d
JDB
516 slot >>= rtab->rate.cell_log;
517 if (slot > 255)
518 return (rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]);
519 return rtab->data[slot];
520}
521
12da81d1
JHS
522#ifdef CONFIG_NET_CLS_ACT
523static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask)
524{
525 struct sk_buff *n = skb_clone(skb, gfp_mask);
526
527 if (n) {
528 n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
529 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
530 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
12da81d1
JHS
531 }
532 return n;
533}
534#endif
535
1da177e4 536#endif