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 <linux/percpu.h>
10 #include <linux/dynamic_queue_limits.h>
11 #include <net/gen_stats.h>
12 #include <net/rtnetlink.h>
19 struct qdisc_rate_table {
20 struct tc_ratespec rate;
22 struct qdisc_rate_table *next;
28 __QDISC_STATE_DEACTIVATED,
29 __QDISC_STATE_THROTTLED,
33 * following bits are only changed while qdisc lock is held
35 enum qdisc___state_t {
36 __QDISC___STATE_RUNNING = 1,
39 struct qdisc_size_table {
41 struct list_head list;
42 struct tc_sizespec szopts;
48 int (*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
49 struct sk_buff * (*dequeue)(struct Qdisc *dev);
51 #define TCQ_F_BUILTIN 1
52 #define TCQ_F_INGRESS 2
53 #define TCQ_F_CAN_BYPASS 4
54 #define TCQ_F_MQROOT 8
55 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
56 * q->dev_queue : It can test
57 * netif_xmit_frozen_or_stopped() before
58 * dequeueing next packet.
59 * Its true for MQ/MQPRIO slaves, or non
62 #define TCQ_F_WARN_NONWC (1 << 16)
63 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
64 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
65 * qdisc_tree_decrease_qlen() should stop.
68 const struct Qdisc_ops *ops;
69 struct qdisc_size_table __rcu *stab;
70 struct list_head list;
73 int (*reshape_fail)(struct sk_buff *skb,
78 /* This field is deprecated, but it is still used by CBQ
79 * and it will live until better solution will be invented.
81 struct Qdisc *__parent;
82 struct netdev_queue *dev_queue;
84 struct gnet_stats_rate_est64 rate_est;
85 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
86 struct gnet_stats_queue __percpu *cpu_qstats;
88 struct Qdisc *next_sched;
89 struct sk_buff *gso_skb;
91 * For performance sake on SMP, we put highly modified fields at the end
94 struct sk_buff_head q;
95 struct gnet_stats_basic_packed bstats;
97 struct gnet_stats_queue qstats;
98 struct rcu_head rcu_head;
102 spinlock_t busylock ____cacheline_aligned_in_smp;
105 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
107 return (qdisc->__state & __QDISC___STATE_RUNNING) ? true : false;
110 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
112 if (qdisc_is_running(qdisc))
114 qdisc->__state |= __QDISC___STATE_RUNNING;
118 static inline void qdisc_run_end(struct Qdisc *qdisc)
120 qdisc->__state &= ~__QDISC___STATE_RUNNING;
123 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
125 return qdisc->flags & TCQ_F_ONETXQUEUE;
128 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
131 /* Non-BQL migrated drivers will return 0, too. */
132 return dql_avail(&txq->dql);
138 static inline bool qdisc_is_throttled(const struct Qdisc *qdisc)
140 return test_bit(__QDISC_STATE_THROTTLED, &qdisc->state) ? true : false;
143 static inline void qdisc_throttled(struct Qdisc *qdisc)
145 set_bit(__QDISC_STATE_THROTTLED, &qdisc->state);
148 static inline void qdisc_unthrottled(struct Qdisc *qdisc)
150 clear_bit(__QDISC_STATE_THROTTLED, &qdisc->state);
153 struct Qdisc_class_ops {
154 /* Child qdisc manipulation */
155 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
156 int (*graft)(struct Qdisc *, unsigned long cl,
157 struct Qdisc *, struct Qdisc **);
158 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
159 void (*qlen_notify)(struct Qdisc *, unsigned long);
161 /* Class manipulation routines */
162 unsigned long (*get)(struct Qdisc *, u32 classid);
163 void (*put)(struct Qdisc *, unsigned long);
164 int (*change)(struct Qdisc *, u32, u32,
165 struct nlattr **, unsigned long *);
166 int (*delete)(struct Qdisc *, unsigned long);
167 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
169 /* Filter manipulation */
170 struct tcf_proto __rcu ** (*tcf_chain)(struct Qdisc *, unsigned long);
171 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
173 void (*unbind_tcf)(struct Qdisc *, unsigned long);
175 /* rtnetlink specific */
176 int (*dump)(struct Qdisc *, unsigned long,
177 struct sk_buff *skb, struct tcmsg*);
178 int (*dump_stats)(struct Qdisc *, unsigned long,
183 struct Qdisc_ops *next;
184 const struct Qdisc_class_ops *cl_ops;
188 int (*enqueue)(struct sk_buff *, struct Qdisc *);
189 struct sk_buff * (*dequeue)(struct Qdisc *);
190 struct sk_buff * (*peek)(struct Qdisc *);
191 unsigned int (*drop)(struct Qdisc *);
193 int (*init)(struct Qdisc *, struct nlattr *arg);
194 void (*reset)(struct Qdisc *);
195 void (*destroy)(struct Qdisc *);
196 int (*change)(struct Qdisc *, struct nlattr *arg);
197 void (*attach)(struct Qdisc *);
199 int (*dump)(struct Qdisc *, struct sk_buff *);
200 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
202 struct module *owner;
211 struct tcf_proto_ops {
212 struct list_head head;
215 int (*classify)(struct sk_buff *,
216 const struct tcf_proto *,
217 struct tcf_result *);
218 int (*init)(struct tcf_proto*);
219 bool (*destroy)(struct tcf_proto*, bool);
221 unsigned long (*get)(struct tcf_proto*, u32 handle);
222 int (*change)(struct net *net, struct sk_buff *,
223 struct tcf_proto*, unsigned long,
224 u32 handle, struct nlattr **,
225 unsigned long *, bool);
226 int (*delete)(struct tcf_proto*, unsigned long);
227 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
229 /* rtnetlink specific */
230 int (*dump)(struct net*, struct tcf_proto*, unsigned long,
231 struct sk_buff *skb, struct tcmsg*);
233 struct module *owner;
237 /* Fast access part */
238 struct tcf_proto __rcu *next;
240 int (*classify)(struct sk_buff *,
241 const struct tcf_proto *,
242 struct tcf_result *);
250 const struct tcf_proto_ops *ops;
254 struct qdisc_skb_cb {
255 unsigned int pkt_len;
256 u16 slave_dev_queue_mapping;
258 #define QDISC_CB_PRIV_LEN 20
259 unsigned char data[QDISC_CB_PRIV_LEN];
262 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
264 struct qdisc_skb_cb *qcb;
266 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
267 BUILD_BUG_ON(sizeof(qcb->data) < sz);
270 static inline int qdisc_qlen(const struct Qdisc *q)
275 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
277 return (struct qdisc_skb_cb *)skb->cb;
280 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
282 return &qdisc->q.lock;
285 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
287 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
292 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
294 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
297 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
299 return qdisc->dev_queue->qdisc_sleeping;
302 /* The qdisc root lock is a mechanism by which to top level
303 * of a qdisc tree can be locked from any qdisc node in the
304 * forest. This allows changing the configuration of some
305 * aspect of the qdisc tree while blocking out asynchronous
306 * qdisc access in the packet processing paths.
308 * It is only legal to do this when the root will not change
309 * on us. Otherwise we'll potentially lock the wrong qdisc
310 * root. This is enforced by holding the RTNL semaphore, which
311 * all users of this lock accessor must do.
313 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
315 struct Qdisc *root = qdisc_root(qdisc);
318 return qdisc_lock(root);
321 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
323 struct Qdisc *root = qdisc_root_sleeping(qdisc);
326 return qdisc_lock(root);
329 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
331 return qdisc->dev_queue->dev;
334 static inline void sch_tree_lock(const struct Qdisc *q)
336 spin_lock_bh(qdisc_root_sleeping_lock(q));
339 static inline void sch_tree_unlock(const struct Qdisc *q)
341 spin_unlock_bh(qdisc_root_sleeping_lock(q));
344 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
345 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
347 extern struct Qdisc noop_qdisc;
348 extern struct Qdisc_ops noop_qdisc_ops;
349 extern struct Qdisc_ops pfifo_fast_ops;
350 extern struct Qdisc_ops mq_qdisc_ops;
351 extern struct Qdisc_ops noqueue_qdisc_ops;
352 extern const struct Qdisc_ops *default_qdisc_ops;
354 struct Qdisc_class_common {
356 struct hlist_node hnode;
359 struct Qdisc_class_hash {
360 struct hlist_head *hash;
361 unsigned int hashsize;
362 unsigned int hashmask;
363 unsigned int hashelems;
366 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
373 static inline struct Qdisc_class_common *
374 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
376 struct Qdisc_class_common *cl;
379 h = qdisc_class_hash(id, hash->hashmask);
380 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
381 if (cl->classid == id)
387 int qdisc_class_hash_init(struct Qdisc_class_hash *);
388 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
389 struct Qdisc_class_common *);
390 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
391 struct Qdisc_class_common *);
392 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
393 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
395 void dev_init_scheduler(struct net_device *dev);
396 void dev_shutdown(struct net_device *dev);
397 void dev_activate(struct net_device *dev);
398 void dev_deactivate(struct net_device *dev);
399 void dev_deactivate_many(struct list_head *head);
400 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
401 struct Qdisc *qdisc);
402 void qdisc_reset(struct Qdisc *qdisc);
403 void qdisc_destroy(struct Qdisc *qdisc);
404 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
406 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
407 const struct Qdisc_ops *ops);
408 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
409 const struct Qdisc_ops *ops, u32 parentid);
410 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
411 const struct qdisc_size_table *stab);
412 bool tcf_destroy(struct tcf_proto *tp, bool force);
413 void tcf_destroy_chain(struct tcf_proto __rcu **fl);
414 int skb_do_redirect(struct sk_buff *);
416 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
418 #ifdef CONFIG_NET_CLS_ACT
419 return G_TC_AT(skb->tc_verd) & AT_INGRESS;
425 /* Reset all TX qdiscs greater then index of a device. */
426 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
430 for (; i < dev->num_tx_queues; i++) {
431 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
433 spin_lock_bh(qdisc_lock(qdisc));
435 spin_unlock_bh(qdisc_lock(qdisc));
440 static inline void qdisc_reset_all_tx(struct net_device *dev)
442 qdisc_reset_all_tx_gt(dev, 0);
445 /* Are all TX queues of the device empty? */
446 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
451 for (i = 0; i < dev->num_tx_queues; i++) {
452 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
453 const struct Qdisc *q = rcu_dereference(txq->qdisc);
464 /* Are any of the TX qdiscs changing? */
465 static inline bool qdisc_tx_changing(const struct net_device *dev)
469 for (i = 0; i < dev->num_tx_queues; i++) {
470 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
471 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
477 /* Is the device using the noop qdisc on all queues? */
478 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
482 for (i = 0; i < dev->num_tx_queues; i++) {
483 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
484 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
490 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
492 return qdisc_skb_cb(skb)->pkt_len;
495 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
496 enum net_xmit_qdisc_t {
497 __NET_XMIT_STOLEN = 0x00010000,
498 __NET_XMIT_BYPASS = 0x00020000,
501 #ifdef CONFIG_NET_CLS_ACT
502 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
504 #define net_xmit_drop_count(e) (1)
507 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
508 const struct Qdisc *sch)
510 #ifdef CONFIG_NET_SCHED
511 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
514 __qdisc_calculate_pkt_len(skb, stab);
518 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
520 qdisc_calculate_pkt_len(skb, sch);
521 return sch->enqueue(skb, sch);
524 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
526 return q->flags & TCQ_F_CPUSTATS;
529 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
530 const struct sk_buff *skb)
532 bstats->bytes += qdisc_pkt_len(skb);
533 bstats->packets += skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1;
536 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
537 const struct sk_buff *skb)
539 u64_stats_update_begin(&bstats->syncp);
540 bstats_update(&bstats->bstats, skb);
541 u64_stats_update_end(&bstats->syncp);
544 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
545 const struct sk_buff *skb)
547 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
550 static inline void qdisc_bstats_update(struct Qdisc *sch,
551 const struct sk_buff *skb)
553 bstats_update(&sch->bstats, skb);
556 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
557 const struct sk_buff *skb)
559 sch->qstats.backlog -= qdisc_pkt_len(skb);
562 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
563 const struct sk_buff *skb)
565 sch->qstats.backlog += qdisc_pkt_len(skb);
568 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
570 sch->qstats.drops += count;
573 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
578 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
580 qstats->overlimits++;
583 static inline void qdisc_qstats_drop(struct Qdisc *sch)
585 qstats_drop_inc(&sch->qstats);
588 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
590 qstats_drop_inc(this_cpu_ptr(sch->cpu_qstats));
593 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
595 sch->qstats.overlimits++;
598 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
599 struct sk_buff_head *list)
601 __skb_queue_tail(list, skb);
602 qdisc_qstats_backlog_inc(sch, skb);
604 return NET_XMIT_SUCCESS;
607 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
609 return __qdisc_enqueue_tail(skb, sch, &sch->q);
612 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
613 struct sk_buff_head *list)
615 struct sk_buff *skb = __skb_dequeue(list);
617 if (likely(skb != NULL)) {
618 qdisc_qstats_backlog_dec(sch, skb);
619 qdisc_bstats_update(sch, skb);
625 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
627 return __qdisc_dequeue_head(sch, &sch->q);
630 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
631 struct sk_buff_head *list)
633 struct sk_buff *skb = __skb_dequeue(list);
635 if (likely(skb != NULL)) {
636 unsigned int len = qdisc_pkt_len(skb);
637 qdisc_qstats_backlog_dec(sch, skb);
645 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch)
647 return __qdisc_queue_drop_head(sch, &sch->q);
650 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
651 struct sk_buff_head *list)
653 struct sk_buff *skb = __skb_dequeue_tail(list);
655 if (likely(skb != NULL))
656 qdisc_qstats_backlog_dec(sch, skb);
661 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
663 return __qdisc_dequeue_tail(sch, &sch->q);
666 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
668 return skb_peek(&sch->q);
671 /* generic pseudo peek method for non-work-conserving qdisc */
672 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
674 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
676 sch->gso_skb = sch->dequeue(sch);
678 /* it's still part of the queue */
679 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
687 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
688 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
690 struct sk_buff *skb = sch->gso_skb;
694 qdisc_qstats_backlog_dec(sch, skb);
697 skb = sch->dequeue(sch);
703 static inline void __qdisc_reset_queue(struct Qdisc *sch,
704 struct sk_buff_head *list)
707 * We do not know the backlog in bytes of this list, it
708 * is up to the caller to correct it
710 __skb_queue_purge(list);
713 static inline void qdisc_reset_queue(struct Qdisc *sch)
715 __qdisc_reset_queue(sch, &sch->q);
716 sch->qstats.backlog = 0;
719 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
728 unsigned int qlen = old->q.qlen;
729 unsigned int backlog = old->qstats.backlog;
732 qdisc_tree_reduce_backlog(old, qlen, backlog);
734 sch_tree_unlock(sch);
739 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
740 struct sk_buff_head *list)
742 struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
744 if (likely(skb != NULL)) {
745 unsigned int len = qdisc_pkt_len(skb);
753 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
755 return __qdisc_queue_drop(sch, &sch->q);
758 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
761 qdisc_qstats_drop(sch);
763 return NET_XMIT_DROP;
766 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
768 qdisc_qstats_drop(sch);
770 #ifdef CONFIG_NET_CLS_ACT
771 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
774 return NET_XMIT_SUCCESS;
779 return NET_XMIT_DROP;
782 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
783 long it will take to send a packet given its size.
785 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
787 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
790 slot >>= rtab->rate.cell_log;
792 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
793 return rtab->data[slot];
796 struct psched_ratecfg {
797 u64 rate_bytes_ps; /* bytes per second */
804 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
809 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
810 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
812 return ((u64)len * r->mult) >> r->shift;
815 void psched_ratecfg_precompute(struct psched_ratecfg *r,
816 const struct tc_ratespec *conf,
819 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
820 const struct psched_ratecfg *r)
822 memset(res, 0, sizeof(*res));
824 /* legacy struct tc_ratespec has a 32bit @rate field
825 * Qdisc using 64bit rate should add new attributes
826 * in order to maintain compatibility.
828 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
830 res->overhead = r->overhead;
831 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);