net_sched: long word align struct qdisc_skb_cb data
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / include / net / sch_generic.h
blob04f8556313d5f52c37ff9722ab7c1f38a2f955af
1 #ifndef __NET_SCHED_GENERIC_H
2 #define __NET_SCHED_GENERIC_H
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>
13 struct Qdisc_ops;
14 struct qdisc_walker;
15 struct tcf_walker;
16 struct module;
18 struct qdisc_rate_table {
19 struct tc_ratespec rate;
20 u32 data[256];
21 struct qdisc_rate_table *next;
22 int refcnt;
25 enum qdisc_state_t {
26 __QDISC_STATE_SCHED,
27 __QDISC_STATE_DEACTIVATED,
31 * following bits are only changed while qdisc lock is held
33 enum qdisc___state_t {
34 __QDISC___STATE_RUNNING,
37 struct qdisc_size_table {
38 struct list_head list;
39 struct tc_sizespec szopts;
40 int refcnt;
41 u16 data[];
44 struct Qdisc {
45 int (*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
46 struct sk_buff * (*dequeue)(struct Qdisc *dev);
47 unsigned flags;
48 #define TCQ_F_BUILTIN 1
49 #define TCQ_F_THROTTLED 2
50 #define TCQ_F_INGRESS 4
51 #define TCQ_F_CAN_BYPASS 8
52 #define TCQ_F_MQROOT 16
53 #define TCQ_F_WARN_NONWC (1 << 16)
54 int padded;
55 struct Qdisc_ops *ops;
56 struct qdisc_size_table *stab;
57 struct list_head list;
58 u32 handle;
59 u32 parent;
60 atomic_t refcnt;
61 struct gnet_stats_rate_est rate_est;
62 int (*reshape_fail)(struct sk_buff *skb,
63 struct Qdisc *q);
65 void *u32_node;
67 /* This field is deprecated, but it is still used by CBQ
68 * and it will live until better solution will be invented.
70 struct Qdisc *__parent;
71 struct netdev_queue *dev_queue;
72 struct Qdisc *next_sched;
74 struct sk_buff *gso_skb;
76 * For performance sake on SMP, we put highly modified fields at the end
78 unsigned long state;
79 struct sk_buff_head q;
80 struct gnet_stats_basic_packed bstats;
81 unsigned long __state;
82 struct gnet_stats_queue qstats;
83 struct rcu_head rcu_head;
84 spinlock_t busylock;
87 static inline bool qdisc_is_running(struct Qdisc *qdisc)
89 return test_bit(__QDISC___STATE_RUNNING, &qdisc->__state);
92 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
94 return !__test_and_set_bit(__QDISC___STATE_RUNNING, &qdisc->__state);
97 static inline void qdisc_run_end(struct Qdisc *qdisc)
99 __clear_bit(__QDISC___STATE_RUNNING, &qdisc->__state);
102 struct Qdisc_class_ops {
103 /* Child qdisc manipulation */
104 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
105 int (*graft)(struct Qdisc *, unsigned long cl,
106 struct Qdisc *, struct Qdisc **);
107 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
108 void (*qlen_notify)(struct Qdisc *, unsigned long);
110 /* Class manipulation routines */
111 unsigned long (*get)(struct Qdisc *, u32 classid);
112 void (*put)(struct Qdisc *, unsigned long);
113 int (*change)(struct Qdisc *, u32, u32,
114 struct nlattr **, unsigned long *);
115 int (*delete)(struct Qdisc *, unsigned long);
116 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
118 /* Filter manipulation */
119 struct tcf_proto ** (*tcf_chain)(struct Qdisc *, unsigned long);
120 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
121 u32 classid);
122 void (*unbind_tcf)(struct Qdisc *, unsigned long);
124 /* rtnetlink specific */
125 int (*dump)(struct Qdisc *, unsigned long,
126 struct sk_buff *skb, struct tcmsg*);
127 int (*dump_stats)(struct Qdisc *, unsigned long,
128 struct gnet_dump *);
131 struct Qdisc_ops {
132 struct Qdisc_ops *next;
133 const struct Qdisc_class_ops *cl_ops;
134 char id[IFNAMSIZ];
135 int priv_size;
137 int (*enqueue)(struct sk_buff *, struct Qdisc *);
138 struct sk_buff * (*dequeue)(struct Qdisc *);
139 struct sk_buff * (*peek)(struct Qdisc *);
140 unsigned int (*drop)(struct Qdisc *);
142 int (*init)(struct Qdisc *, struct nlattr *arg);
143 void (*reset)(struct Qdisc *);
144 void (*destroy)(struct Qdisc *);
145 int (*change)(struct Qdisc *, struct nlattr *arg);
146 void (*attach)(struct Qdisc *);
148 int (*dump)(struct Qdisc *, struct sk_buff *);
149 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
151 struct module *owner;
155 struct tcf_result {
156 unsigned long class;
157 u32 classid;
160 struct tcf_proto_ops {
161 struct tcf_proto_ops *next;
162 char kind[IFNAMSIZ];
164 int (*classify)(struct sk_buff*, struct tcf_proto*,
165 struct tcf_result *);
166 int (*init)(struct tcf_proto*);
167 void (*destroy)(struct tcf_proto*);
169 unsigned long (*get)(struct tcf_proto*, u32 handle);
170 void (*put)(struct tcf_proto*, unsigned long);
171 int (*change)(struct tcf_proto*, unsigned long,
172 u32 handle, struct nlattr **,
173 unsigned long *);
174 int (*delete)(struct tcf_proto*, unsigned long);
175 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
177 /* rtnetlink specific */
178 int (*dump)(struct tcf_proto*, unsigned long,
179 struct sk_buff *skb, struct tcmsg*);
181 struct module *owner;
184 struct tcf_proto {
185 /* Fast access part */
186 struct tcf_proto *next;
187 void *root;
188 int (*classify)(struct sk_buff*, struct tcf_proto*,
189 struct tcf_result *);
190 __be16 protocol;
192 /* All the rest */
193 u32 prio;
194 u32 classid;
195 struct Qdisc *q;
196 void *data;
197 struct tcf_proto_ops *ops;
200 struct qdisc_skb_cb {
201 unsigned int pkt_len;
202 long data[];
205 static inline int qdisc_qlen(struct Qdisc *q)
207 return q->q.qlen;
210 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
212 return (struct qdisc_skb_cb *)skb->cb;
215 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
217 return &qdisc->q.lock;
220 static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
222 return qdisc->dev_queue->qdisc;
225 static inline struct Qdisc *qdisc_root_sleeping(struct Qdisc *qdisc)
227 return qdisc->dev_queue->qdisc_sleeping;
230 /* The qdisc root lock is a mechanism by which to top level
231 * of a qdisc tree can be locked from any qdisc node in the
232 * forest. This allows changing the configuration of some
233 * aspect of the qdisc tree while blocking out asynchronous
234 * qdisc access in the packet processing paths.
236 * It is only legal to do this when the root will not change
237 * on us. Otherwise we'll potentially lock the wrong qdisc
238 * root. This is enforced by holding the RTNL semaphore, which
239 * all users of this lock accessor must do.
241 static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
243 struct Qdisc *root = qdisc_root(qdisc);
245 ASSERT_RTNL();
246 return qdisc_lock(root);
249 static inline spinlock_t *qdisc_root_sleeping_lock(struct Qdisc *qdisc)
251 struct Qdisc *root = qdisc_root_sleeping(qdisc);
253 ASSERT_RTNL();
254 return qdisc_lock(root);
257 static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
259 return qdisc->dev_queue->dev;
262 static inline void sch_tree_lock(struct Qdisc *q)
264 spin_lock_bh(qdisc_root_sleeping_lock(q));
267 static inline void sch_tree_unlock(struct Qdisc *q)
269 spin_unlock_bh(qdisc_root_sleeping_lock(q));
272 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
273 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
275 extern struct Qdisc noop_qdisc;
276 extern struct Qdisc_ops noop_qdisc_ops;
277 extern struct Qdisc_ops pfifo_fast_ops;
278 extern struct Qdisc_ops mq_qdisc_ops;
280 struct Qdisc_class_common {
281 u32 classid;
282 struct hlist_node hnode;
285 struct Qdisc_class_hash {
286 struct hlist_head *hash;
287 unsigned int hashsize;
288 unsigned int hashmask;
289 unsigned int hashelems;
292 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
294 id ^= id >> 8;
295 id ^= id >> 4;
296 return id & mask;
299 static inline struct Qdisc_class_common *
300 qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
302 struct Qdisc_class_common *cl;
303 struct hlist_node *n;
304 unsigned int h;
306 h = qdisc_class_hash(id, hash->hashmask);
307 hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
308 if (cl->classid == id)
309 return cl;
311 return NULL;
314 extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
315 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
316 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
317 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
318 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
320 extern void dev_init_scheduler(struct net_device *dev);
321 extern void dev_shutdown(struct net_device *dev);
322 extern void dev_activate(struct net_device *dev);
323 extern void dev_deactivate(struct net_device *dev);
324 extern void dev_deactivate_many(struct list_head *head);
325 extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
326 struct Qdisc *qdisc);
327 extern void qdisc_reset(struct Qdisc *qdisc);
328 extern void qdisc_destroy(struct Qdisc *qdisc);
329 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
330 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
331 struct Qdisc_ops *ops);
332 extern struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
333 struct Qdisc_ops *ops, u32 parentid);
334 extern void qdisc_calculate_pkt_len(struct sk_buff *skb,
335 struct qdisc_size_table *stab);
336 extern void tcf_destroy(struct tcf_proto *tp);
337 extern void tcf_destroy_chain(struct tcf_proto **fl);
339 /* Reset all TX qdiscs greater then index of a device. */
340 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
342 struct Qdisc *qdisc;
344 for (; i < dev->num_tx_queues; i++) {
345 qdisc = netdev_get_tx_queue(dev, i)->qdisc;
346 if (qdisc) {
347 spin_lock_bh(qdisc_lock(qdisc));
348 qdisc_reset(qdisc);
349 spin_unlock_bh(qdisc_lock(qdisc));
354 static inline void qdisc_reset_all_tx(struct net_device *dev)
356 qdisc_reset_all_tx_gt(dev, 0);
359 /* Are all TX queues of the device empty? */
360 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
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 const struct Qdisc *q = txq->qdisc;
367 if (q->q.qlen)
368 return false;
370 return true;
373 /* Are any of the TX qdiscs changing? */
374 static inline bool qdisc_tx_changing(struct net_device *dev)
376 unsigned int i;
377 for (i = 0; i < dev->num_tx_queues; i++) {
378 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
379 if (txq->qdisc != txq->qdisc_sleeping)
380 return true;
382 return false;
385 /* Is the device using the noop qdisc on all queues? */
386 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
388 unsigned int i;
389 for (i = 0; i < dev->num_tx_queues; i++) {
390 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
391 if (txq->qdisc != &noop_qdisc)
392 return false;
394 return true;
397 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
399 return qdisc_skb_cb(skb)->pkt_len;
402 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
403 enum net_xmit_qdisc_t {
404 __NET_XMIT_STOLEN = 0x00010000,
405 __NET_XMIT_BYPASS = 0x00020000,
408 #ifdef CONFIG_NET_CLS_ACT
409 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
410 #else
411 #define net_xmit_drop_count(e) (1)
412 #endif
414 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
416 #ifdef CONFIG_NET_SCHED
417 if (sch->stab)
418 qdisc_calculate_pkt_len(skb, sch->stab);
419 #endif
420 return sch->enqueue(skb, sch);
423 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
425 qdisc_skb_cb(skb)->pkt_len = skb->len;
426 return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
430 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
431 const struct sk_buff *skb)
433 bstats->bytes += qdisc_pkt_len(skb);
434 bstats->packets += skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1;
437 static inline void qdisc_bstats_update(struct Qdisc *sch,
438 const struct sk_buff *skb)
440 bstats_update(&sch->bstats, skb);
443 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
444 struct sk_buff_head *list)
446 __skb_queue_tail(list, skb);
447 sch->qstats.backlog += qdisc_pkt_len(skb);
449 return NET_XMIT_SUCCESS;
452 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
454 return __qdisc_enqueue_tail(skb, sch, &sch->q);
457 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
458 struct sk_buff_head *list)
460 struct sk_buff *skb = __skb_dequeue(list);
462 if (likely(skb != NULL)) {
463 sch->qstats.backlog -= qdisc_pkt_len(skb);
464 qdisc_bstats_update(sch, skb);
467 return skb;
470 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
472 return __qdisc_dequeue_head(sch, &sch->q);
475 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
476 struct sk_buff_head *list)
478 struct sk_buff *skb = __skb_dequeue(list);
480 if (likely(skb != NULL)) {
481 unsigned int len = qdisc_pkt_len(skb);
482 sch->qstats.backlog -= len;
483 kfree_skb(skb);
484 return len;
487 return 0;
490 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch)
492 return __qdisc_queue_drop_head(sch, &sch->q);
495 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
496 struct sk_buff_head *list)
498 struct sk_buff *skb = __skb_dequeue_tail(list);
500 if (likely(skb != NULL))
501 sch->qstats.backlog -= qdisc_pkt_len(skb);
503 return skb;
506 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
508 return __qdisc_dequeue_tail(sch, &sch->q);
511 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
513 return skb_peek(&sch->q);
516 /* generic pseudo peek method for non-work-conserving qdisc */
517 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
519 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
520 if (!sch->gso_skb) {
521 sch->gso_skb = sch->dequeue(sch);
522 if (sch->gso_skb)
523 /* it's still part of the queue */
524 sch->q.qlen++;
527 return sch->gso_skb;
530 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
531 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
533 struct sk_buff *skb = sch->gso_skb;
535 if (skb) {
536 sch->gso_skb = NULL;
537 sch->q.qlen--;
538 } else {
539 skb = sch->dequeue(sch);
542 return skb;
545 static inline void __qdisc_reset_queue(struct Qdisc *sch,
546 struct sk_buff_head *list)
549 * We do not know the backlog in bytes of this list, it
550 * is up to the caller to correct it
552 __skb_queue_purge(list);
555 static inline void qdisc_reset_queue(struct Qdisc *sch)
557 __qdisc_reset_queue(sch, &sch->q);
558 sch->qstats.backlog = 0;
561 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
562 struct sk_buff_head *list)
564 struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
566 if (likely(skb != NULL)) {
567 unsigned int len = qdisc_pkt_len(skb);
568 kfree_skb(skb);
569 return len;
572 return 0;
575 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
577 return __qdisc_queue_drop(sch, &sch->q);
580 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
582 kfree_skb(skb);
583 sch->qstats.drops++;
585 return NET_XMIT_DROP;
588 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
590 sch->qstats.drops++;
592 #ifdef CONFIG_NET_CLS_ACT
593 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
594 goto drop;
596 return NET_XMIT_SUCCESS;
598 drop:
599 #endif
600 kfree_skb(skb);
601 return NET_XMIT_DROP;
604 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
605 long it will take to send a packet given its size.
607 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
609 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
610 if (slot < 0)
611 slot = 0;
612 slot >>= rtab->rate.cell_log;
613 if (slot > 255)
614 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
615 return rtab->data[slot];
618 #ifdef CONFIG_NET_CLS_ACT
619 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask,
620 int action)
622 struct sk_buff *n;
624 n = skb_clone(skb, gfp_mask);
626 if (n) {
627 n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
628 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
629 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
631 return n;
633 #endif
635 #endif