Merge branch 'for-3.11' of git://linux-nfs.org/~bfields/linux
[linux-2.6.git] / include / net / sch_generic.h
blob6eab63363e59fe82dc3fc6119633e5583b663dd1
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/pkt_sched.h>
8 #include <linux/pkt_cls.h>
9 #include <net/gen_stats.h>
10 #include <net/rtnetlink.h>
12 struct Qdisc_ops;
13 struct qdisc_walker;
14 struct tcf_walker;
15 struct module;
17 struct qdisc_rate_table {
18 struct tc_ratespec rate;
19 u32 data[256];
20 struct qdisc_rate_table *next;
21 int refcnt;
24 enum qdisc_state_t {
25 __QDISC_STATE_SCHED,
26 __QDISC_STATE_DEACTIVATED,
27 __QDISC_STATE_THROTTLED,
31 * following bits are only changed while qdisc lock is held
33 enum qdisc___state_t {
34 __QDISC___STATE_RUNNING = 1,
37 struct qdisc_size_table {
38 struct rcu_head rcu;
39 struct list_head list;
40 struct tc_sizespec szopts;
41 int refcnt;
42 u16 data[];
45 struct Qdisc {
46 int (*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
47 struct sk_buff * (*dequeue)(struct Qdisc *dev);
48 unsigned int flags;
49 #define TCQ_F_BUILTIN 1
50 #define TCQ_F_INGRESS 2
51 #define TCQ_F_CAN_BYPASS 4
52 #define TCQ_F_MQROOT 8
53 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
54 * q->dev_queue : It can test
55 * netif_xmit_frozen_or_stopped() before
56 * dequeueing next packet.
57 * Its true for MQ/MQPRIO slaves, or non
58 * multiqueue device.
60 #define TCQ_F_WARN_NONWC (1 << 16)
61 u32 limit;
62 const struct Qdisc_ops *ops;
63 struct qdisc_size_table __rcu *stab;
64 struct list_head list;
65 u32 handle;
66 u32 parent;
67 int (*reshape_fail)(struct sk_buff *skb,
68 struct Qdisc *q);
70 void *u32_node;
72 /* This field is deprecated, but it is still used by CBQ
73 * and it will live until better solution will be invented.
75 struct Qdisc *__parent;
76 struct netdev_queue *dev_queue;
78 struct gnet_stats_rate_est64 rate_est;
79 struct Qdisc *next_sched;
80 struct sk_buff *gso_skb;
82 * For performance sake on SMP, we put highly modified fields at the end
84 unsigned long state;
85 struct sk_buff_head q;
86 struct gnet_stats_basic_packed bstats;
87 unsigned int __state;
88 struct gnet_stats_queue qstats;
89 struct rcu_head rcu_head;
90 int padded;
91 atomic_t refcnt;
93 spinlock_t busylock ____cacheline_aligned_in_smp;
96 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
98 return (qdisc->__state & __QDISC___STATE_RUNNING) ? true : false;
101 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
103 if (qdisc_is_running(qdisc))
104 return false;
105 qdisc->__state |= __QDISC___STATE_RUNNING;
106 return true;
109 static inline void qdisc_run_end(struct Qdisc *qdisc)
111 qdisc->__state &= ~__QDISC___STATE_RUNNING;
114 static inline bool qdisc_is_throttled(const struct Qdisc *qdisc)
116 return test_bit(__QDISC_STATE_THROTTLED, &qdisc->state) ? true : false;
119 static inline void qdisc_throttled(struct Qdisc *qdisc)
121 set_bit(__QDISC_STATE_THROTTLED, &qdisc->state);
124 static inline void qdisc_unthrottled(struct Qdisc *qdisc)
126 clear_bit(__QDISC_STATE_THROTTLED, &qdisc->state);
129 struct Qdisc_class_ops {
130 /* Child qdisc manipulation */
131 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
132 int (*graft)(struct Qdisc *, unsigned long cl,
133 struct Qdisc *, struct Qdisc **);
134 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
135 void (*qlen_notify)(struct Qdisc *, unsigned long);
137 /* Class manipulation routines */
138 unsigned long (*get)(struct Qdisc *, u32 classid);
139 void (*put)(struct Qdisc *, unsigned long);
140 int (*change)(struct Qdisc *, u32, u32,
141 struct nlattr **, unsigned long *);
142 int (*delete)(struct Qdisc *, unsigned long);
143 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
145 /* Filter manipulation */
146 struct tcf_proto ** (*tcf_chain)(struct Qdisc *, unsigned long);
147 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
148 u32 classid);
149 void (*unbind_tcf)(struct Qdisc *, unsigned long);
151 /* rtnetlink specific */
152 int (*dump)(struct Qdisc *, unsigned long,
153 struct sk_buff *skb, struct tcmsg*);
154 int (*dump_stats)(struct Qdisc *, unsigned long,
155 struct gnet_dump *);
158 struct Qdisc_ops {
159 struct Qdisc_ops *next;
160 const struct Qdisc_class_ops *cl_ops;
161 char id[IFNAMSIZ];
162 int priv_size;
164 int (*enqueue)(struct sk_buff *, struct Qdisc *);
165 struct sk_buff * (*dequeue)(struct Qdisc *);
166 struct sk_buff * (*peek)(struct Qdisc *);
167 unsigned int (*drop)(struct Qdisc *);
169 int (*init)(struct Qdisc *, struct nlattr *arg);
170 void (*reset)(struct Qdisc *);
171 void (*destroy)(struct Qdisc *);
172 int (*change)(struct Qdisc *, struct nlattr *arg);
173 void (*attach)(struct Qdisc *);
175 int (*dump)(struct Qdisc *, struct sk_buff *);
176 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
178 struct module *owner;
182 struct tcf_result {
183 unsigned long class;
184 u32 classid;
187 struct tcf_proto_ops {
188 struct tcf_proto_ops *next;
189 char kind[IFNAMSIZ];
191 int (*classify)(struct sk_buff *,
192 const struct tcf_proto *,
193 struct tcf_result *);
194 int (*init)(struct tcf_proto*);
195 void (*destroy)(struct tcf_proto*);
197 unsigned long (*get)(struct tcf_proto*, u32 handle);
198 void (*put)(struct tcf_proto*, unsigned long);
199 int (*change)(struct net *net, struct sk_buff *,
200 struct tcf_proto*, unsigned long,
201 u32 handle, struct nlattr **,
202 unsigned long *);
203 int (*delete)(struct tcf_proto*, unsigned long);
204 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
206 /* rtnetlink specific */
207 int (*dump)(struct tcf_proto*, unsigned long,
208 struct sk_buff *skb, struct tcmsg*);
210 struct module *owner;
213 struct tcf_proto {
214 /* Fast access part */
215 struct tcf_proto *next;
216 void *root;
217 int (*classify)(struct sk_buff *,
218 const struct tcf_proto *,
219 struct tcf_result *);
220 __be16 protocol;
222 /* All the rest */
223 u32 prio;
224 u32 classid;
225 struct Qdisc *q;
226 void *data;
227 const struct tcf_proto_ops *ops;
230 struct qdisc_skb_cb {
231 unsigned int pkt_len;
232 u16 slave_dev_queue_mapping;
233 u16 _pad;
234 unsigned char data[20];
237 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
239 struct qdisc_skb_cb *qcb;
241 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
242 BUILD_BUG_ON(sizeof(qcb->data) < sz);
245 static inline int qdisc_qlen(const struct Qdisc *q)
247 return q->q.qlen;
250 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
252 return (struct qdisc_skb_cb *)skb->cb;
255 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
257 return &qdisc->q.lock;
260 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
262 return qdisc->dev_queue->qdisc;
265 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
267 return qdisc->dev_queue->qdisc_sleeping;
270 /* The qdisc root lock is a mechanism by which to top level
271 * of a qdisc tree can be locked from any qdisc node in the
272 * forest. This allows changing the configuration of some
273 * aspect of the qdisc tree while blocking out asynchronous
274 * qdisc access in the packet processing paths.
276 * It is only legal to do this when the root will not change
277 * on us. Otherwise we'll potentially lock the wrong qdisc
278 * root. This is enforced by holding the RTNL semaphore, which
279 * all users of this lock accessor must do.
281 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
283 struct Qdisc *root = qdisc_root(qdisc);
285 ASSERT_RTNL();
286 return qdisc_lock(root);
289 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
291 struct Qdisc *root = qdisc_root_sleeping(qdisc);
293 ASSERT_RTNL();
294 return qdisc_lock(root);
297 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
299 return qdisc->dev_queue->dev;
302 static inline void sch_tree_lock(const struct Qdisc *q)
304 spin_lock_bh(qdisc_root_sleeping_lock(q));
307 static inline void sch_tree_unlock(const struct Qdisc *q)
309 spin_unlock_bh(qdisc_root_sleeping_lock(q));
312 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
313 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
315 extern struct Qdisc noop_qdisc;
316 extern struct Qdisc_ops noop_qdisc_ops;
317 extern struct Qdisc_ops pfifo_fast_ops;
318 extern struct Qdisc_ops mq_qdisc_ops;
320 struct Qdisc_class_common {
321 u32 classid;
322 struct hlist_node hnode;
325 struct Qdisc_class_hash {
326 struct hlist_head *hash;
327 unsigned int hashsize;
328 unsigned int hashmask;
329 unsigned int hashelems;
332 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
334 id ^= id >> 8;
335 id ^= id >> 4;
336 return id & mask;
339 static inline struct Qdisc_class_common *
340 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
342 struct Qdisc_class_common *cl;
343 unsigned int h;
345 h = qdisc_class_hash(id, hash->hashmask);
346 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
347 if (cl->classid == id)
348 return cl;
350 return NULL;
353 extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
354 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
355 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
356 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
357 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
359 extern void dev_init_scheduler(struct net_device *dev);
360 extern void dev_shutdown(struct net_device *dev);
361 extern void dev_activate(struct net_device *dev);
362 extern void dev_deactivate(struct net_device *dev);
363 extern void dev_deactivate_many(struct list_head *head);
364 extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
365 struct Qdisc *qdisc);
366 extern void qdisc_reset(struct Qdisc *qdisc);
367 extern void qdisc_destroy(struct Qdisc *qdisc);
368 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
369 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
370 struct Qdisc_ops *ops);
371 extern struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
372 struct Qdisc_ops *ops, u32 parentid);
373 extern void __qdisc_calculate_pkt_len(struct sk_buff *skb,
374 const struct qdisc_size_table *stab);
375 extern void tcf_destroy(struct tcf_proto *tp);
376 extern void tcf_destroy_chain(struct tcf_proto **fl);
378 /* Reset all TX qdiscs greater then index of a device. */
379 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
381 struct Qdisc *qdisc;
383 for (; i < dev->num_tx_queues; i++) {
384 qdisc = netdev_get_tx_queue(dev, i)->qdisc;
385 if (qdisc) {
386 spin_lock_bh(qdisc_lock(qdisc));
387 qdisc_reset(qdisc);
388 spin_unlock_bh(qdisc_lock(qdisc));
393 static inline void qdisc_reset_all_tx(struct net_device *dev)
395 qdisc_reset_all_tx_gt(dev, 0);
398 /* Are all TX queues of the device empty? */
399 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
401 unsigned int i;
402 for (i = 0; i < dev->num_tx_queues; i++) {
403 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
404 const struct Qdisc *q = txq->qdisc;
406 if (q->q.qlen)
407 return false;
409 return true;
412 /* Are any of the TX qdiscs changing? */
413 static inline bool qdisc_tx_changing(const struct net_device *dev)
415 unsigned int i;
416 for (i = 0; i < dev->num_tx_queues; i++) {
417 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
418 if (txq->qdisc != txq->qdisc_sleeping)
419 return true;
421 return false;
424 /* Is the device using the noop qdisc on all queues? */
425 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
427 unsigned int i;
428 for (i = 0; i < dev->num_tx_queues; i++) {
429 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
430 if (txq->qdisc != &noop_qdisc)
431 return false;
433 return true;
436 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
438 return qdisc_skb_cb(skb)->pkt_len;
441 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
442 enum net_xmit_qdisc_t {
443 __NET_XMIT_STOLEN = 0x00010000,
444 __NET_XMIT_BYPASS = 0x00020000,
447 #ifdef CONFIG_NET_CLS_ACT
448 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
449 #else
450 #define net_xmit_drop_count(e) (1)
451 #endif
453 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
454 const struct Qdisc *sch)
456 #ifdef CONFIG_NET_SCHED
457 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
459 if (stab)
460 __qdisc_calculate_pkt_len(skb, stab);
461 #endif
464 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
466 qdisc_calculate_pkt_len(skb, sch);
467 return sch->enqueue(skb, sch);
470 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
472 qdisc_skb_cb(skb)->pkt_len = skb->len;
473 return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
477 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
478 const struct sk_buff *skb)
480 bstats->bytes += qdisc_pkt_len(skb);
481 bstats->packets += skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1;
484 static inline void qdisc_bstats_update(struct Qdisc *sch,
485 const struct sk_buff *skb)
487 bstats_update(&sch->bstats, skb);
490 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
491 struct sk_buff_head *list)
493 __skb_queue_tail(list, skb);
494 sch->qstats.backlog += qdisc_pkt_len(skb);
496 return NET_XMIT_SUCCESS;
499 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
501 return __qdisc_enqueue_tail(skb, sch, &sch->q);
504 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
505 struct sk_buff_head *list)
507 struct sk_buff *skb = __skb_dequeue(list);
509 if (likely(skb != NULL)) {
510 sch->qstats.backlog -= qdisc_pkt_len(skb);
511 qdisc_bstats_update(sch, skb);
514 return skb;
517 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
519 return __qdisc_dequeue_head(sch, &sch->q);
522 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
523 struct sk_buff_head *list)
525 struct sk_buff *skb = __skb_dequeue(list);
527 if (likely(skb != NULL)) {
528 unsigned int len = qdisc_pkt_len(skb);
529 sch->qstats.backlog -= len;
530 kfree_skb(skb);
531 return len;
534 return 0;
537 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch)
539 return __qdisc_queue_drop_head(sch, &sch->q);
542 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
543 struct sk_buff_head *list)
545 struct sk_buff *skb = __skb_dequeue_tail(list);
547 if (likely(skb != NULL))
548 sch->qstats.backlog -= qdisc_pkt_len(skb);
550 return skb;
553 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
555 return __qdisc_dequeue_tail(sch, &sch->q);
558 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
560 return skb_peek(&sch->q);
563 /* generic pseudo peek method for non-work-conserving qdisc */
564 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
566 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
567 if (!sch->gso_skb) {
568 sch->gso_skb = sch->dequeue(sch);
569 if (sch->gso_skb)
570 /* it's still part of the queue */
571 sch->q.qlen++;
574 return sch->gso_skb;
577 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
578 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
580 struct sk_buff *skb = sch->gso_skb;
582 if (skb) {
583 sch->gso_skb = NULL;
584 sch->q.qlen--;
585 } else {
586 skb = sch->dequeue(sch);
589 return skb;
592 static inline void __qdisc_reset_queue(struct Qdisc *sch,
593 struct sk_buff_head *list)
596 * We do not know the backlog in bytes of this list, it
597 * is up to the caller to correct it
599 __skb_queue_purge(list);
602 static inline void qdisc_reset_queue(struct Qdisc *sch)
604 __qdisc_reset_queue(sch, &sch->q);
605 sch->qstats.backlog = 0;
608 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
609 struct sk_buff_head *list)
611 struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
613 if (likely(skb != NULL)) {
614 unsigned int len = qdisc_pkt_len(skb);
615 kfree_skb(skb);
616 return len;
619 return 0;
622 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
624 return __qdisc_queue_drop(sch, &sch->q);
627 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
629 kfree_skb(skb);
630 sch->qstats.drops++;
632 return NET_XMIT_DROP;
635 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
637 sch->qstats.drops++;
639 #ifdef CONFIG_NET_CLS_ACT
640 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
641 goto drop;
643 return NET_XMIT_SUCCESS;
645 drop:
646 #endif
647 kfree_skb(skb);
648 return NET_XMIT_DROP;
651 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
652 long it will take to send a packet given its size.
654 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
656 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
657 if (slot < 0)
658 slot = 0;
659 slot >>= rtab->rate.cell_log;
660 if (slot > 255)
661 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
662 return rtab->data[slot];
665 #ifdef CONFIG_NET_CLS_ACT
666 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask,
667 int action)
669 struct sk_buff *n;
671 n = skb_clone(skb, gfp_mask);
673 if (n) {
674 n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
675 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
676 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
678 return n;
680 #endif
682 struct psched_ratecfg {
683 u64 rate_bytes_ps; /* bytes per second */
684 u32 mult;
685 u16 overhead;
686 u8 shift;
689 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
690 unsigned int len)
692 return ((u64)(len + r->overhead) * r->mult) >> r->shift;
695 extern void psched_ratecfg_precompute(struct psched_ratecfg *r, const struct tc_ratespec *conf);
697 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
698 const struct psched_ratecfg *r)
700 memset(res, 0, sizeof(*res));
701 res->rate = r->rate_bytes_ps;
702 res->overhead = r->overhead;
705 #endif