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,
31 struct qdisc_size_table {
33 struct list_head list;
34 struct tc_sizespec szopts;
39 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
40 struct qdisc_skb_head {
48 int (*enqueue)(struct sk_buff *skb,
50 struct sk_buff **to_free);
51 struct sk_buff * (*dequeue)(struct Qdisc *sch);
53 #define TCQ_F_BUILTIN 1
54 #define TCQ_F_INGRESS 2
55 #define TCQ_F_CAN_BYPASS 4
56 #define TCQ_F_MQROOT 8
57 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
58 * q->dev_queue : It can test
59 * netif_xmit_frozen_or_stopped() before
60 * dequeueing next packet.
61 * Its true for MQ/MQPRIO slaves, or non
64 #define TCQ_F_WARN_NONWC (1 << 16)
65 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
66 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
67 * qdisc_tree_decrease_qlen() should stop.
70 const struct Qdisc_ops *ops;
71 struct qdisc_size_table __rcu *stab;
72 struct hlist_node hash;
77 struct netdev_queue *dev_queue;
79 struct net_rate_estimator __rcu *rate_est;
80 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
81 struct gnet_stats_queue __percpu *cpu_qstats;
84 * For performance sake on SMP, we put highly modified fields at the end
86 struct sk_buff *gso_skb ____cacheline_aligned_in_smp;
87 struct qdisc_skb_head q;
88 struct gnet_stats_basic_packed bstats;
90 struct gnet_stats_queue qstats;
92 struct Qdisc *next_sched;
93 struct sk_buff *skb_bad_txq;
94 struct rcu_head rcu_head;
98 spinlock_t busylock ____cacheline_aligned_in_smp;
101 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
103 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
106 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
108 if (qdisc_is_running(qdisc))
110 /* Variant of write_seqcount_begin() telling lockdep a trylock
113 raw_write_seqcount_begin(&qdisc->running);
114 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
118 static inline void qdisc_run_end(struct Qdisc *qdisc)
120 write_seqcount_end(&qdisc->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 struct Qdisc_class_ops {
139 /* Child qdisc manipulation */
140 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
141 int (*graft)(struct Qdisc *, unsigned long cl,
142 struct Qdisc *, struct Qdisc **);
143 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
144 void (*qlen_notify)(struct Qdisc *, unsigned long);
146 /* Class manipulation routines */
147 unsigned long (*get)(struct Qdisc *, u32 classid);
148 void (*put)(struct Qdisc *, unsigned long);
149 int (*change)(struct Qdisc *, u32, u32,
150 struct nlattr **, unsigned long *);
151 int (*delete)(struct Qdisc *, unsigned long);
152 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
154 /* Filter manipulation */
155 struct tcf_proto __rcu ** (*tcf_chain)(struct Qdisc *, unsigned long);
156 bool (*tcf_cl_offload)(u32 classid);
157 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
159 void (*unbind_tcf)(struct Qdisc *, unsigned long);
161 /* rtnetlink specific */
162 int (*dump)(struct Qdisc *, unsigned long,
163 struct sk_buff *skb, struct tcmsg*);
164 int (*dump_stats)(struct Qdisc *, unsigned long,
169 struct Qdisc_ops *next;
170 const struct Qdisc_class_ops *cl_ops;
174 int (*enqueue)(struct sk_buff *skb,
176 struct sk_buff **to_free);
177 struct sk_buff * (*dequeue)(struct Qdisc *);
178 struct sk_buff * (*peek)(struct Qdisc *);
180 int (*init)(struct Qdisc *, struct nlattr *arg);
181 void (*reset)(struct Qdisc *);
182 void (*destroy)(struct Qdisc *);
183 int (*change)(struct Qdisc *, struct nlattr *arg);
184 void (*attach)(struct Qdisc *);
186 int (*dump)(struct Qdisc *, struct sk_buff *);
187 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
189 struct module *owner;
198 struct tcf_proto_ops {
199 struct list_head head;
202 int (*classify)(struct sk_buff *,
203 const struct tcf_proto *,
204 struct tcf_result *);
205 int (*init)(struct tcf_proto*);
206 bool (*destroy)(struct tcf_proto*, bool);
208 unsigned long (*get)(struct tcf_proto*, u32 handle);
209 int (*change)(struct net *net, struct sk_buff *,
210 struct tcf_proto*, unsigned long,
211 u32 handle, struct nlattr **,
212 unsigned long *, bool);
213 int (*delete)(struct tcf_proto*, unsigned long);
214 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
216 /* rtnetlink specific */
217 int (*dump)(struct net*, struct tcf_proto*, unsigned long,
218 struct sk_buff *skb, struct tcmsg*);
220 struct module *owner;
224 /* Fast access part */
225 struct tcf_proto __rcu *next;
227 int (*classify)(struct sk_buff *,
228 const struct tcf_proto *,
229 struct tcf_result *);
237 const struct tcf_proto_ops *ops;
241 struct qdisc_skb_cb {
242 unsigned int pkt_len;
243 u16 slave_dev_queue_mapping;
245 #define QDISC_CB_PRIV_LEN 20
246 unsigned char data[QDISC_CB_PRIV_LEN];
249 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
251 struct qdisc_skb_cb *qcb;
253 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
254 BUILD_BUG_ON(sizeof(qcb->data) < sz);
257 static inline int qdisc_qlen(const struct Qdisc *q)
262 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
264 return (struct qdisc_skb_cb *)skb->cb;
267 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
269 return &qdisc->q.lock;
272 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
274 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
279 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
281 return qdisc->dev_queue->qdisc_sleeping;
284 /* The qdisc root lock is a mechanism by which to top level
285 * of a qdisc tree can be locked from any qdisc node in the
286 * forest. This allows changing the configuration of some
287 * aspect of the qdisc tree while blocking out asynchronous
288 * qdisc access in the packet processing paths.
290 * It is only legal to do this when the root will not change
291 * on us. Otherwise we'll potentially lock the wrong qdisc
292 * root. This is enforced by holding the RTNL semaphore, which
293 * all users of this lock accessor must do.
295 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
297 struct Qdisc *root = qdisc_root(qdisc);
300 return qdisc_lock(root);
303 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
305 struct Qdisc *root = qdisc_root_sleeping(qdisc);
308 return qdisc_lock(root);
311 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
313 struct Qdisc *root = qdisc_root_sleeping(qdisc);
316 return &root->running;
319 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
321 return qdisc->dev_queue->dev;
324 static inline void sch_tree_lock(const struct Qdisc *q)
326 spin_lock_bh(qdisc_root_sleeping_lock(q));
329 static inline void sch_tree_unlock(const struct Qdisc *q)
331 spin_unlock_bh(qdisc_root_sleeping_lock(q));
334 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
335 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
337 extern struct Qdisc noop_qdisc;
338 extern struct Qdisc_ops noop_qdisc_ops;
339 extern struct Qdisc_ops pfifo_fast_ops;
340 extern struct Qdisc_ops mq_qdisc_ops;
341 extern struct Qdisc_ops noqueue_qdisc_ops;
342 extern const struct Qdisc_ops *default_qdisc_ops;
343 static inline const struct Qdisc_ops *
344 get_default_qdisc_ops(const struct net_device *dev, int ntx)
346 return ntx < dev->real_num_tx_queues ?
347 default_qdisc_ops : &pfifo_fast_ops;
350 struct Qdisc_class_common {
352 struct hlist_node hnode;
355 struct Qdisc_class_hash {
356 struct hlist_head *hash;
357 unsigned int hashsize;
358 unsigned int hashmask;
359 unsigned int hashelems;
362 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
369 static inline struct Qdisc_class_common *
370 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
372 struct Qdisc_class_common *cl;
375 h = qdisc_class_hash(id, hash->hashmask);
376 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
377 if (cl->classid == id)
383 int qdisc_class_hash_init(struct Qdisc_class_hash *);
384 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
385 struct Qdisc_class_common *);
386 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
387 struct Qdisc_class_common *);
388 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
389 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
391 void dev_init_scheduler(struct net_device *dev);
392 void dev_shutdown(struct net_device *dev);
393 void dev_activate(struct net_device *dev);
394 void dev_deactivate(struct net_device *dev);
395 void dev_deactivate_many(struct list_head *head);
396 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
397 struct Qdisc *qdisc);
398 void qdisc_reset(struct Qdisc *qdisc);
399 void qdisc_destroy(struct Qdisc *qdisc);
400 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
402 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
403 const struct Qdisc_ops *ops);
404 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
405 const struct Qdisc_ops *ops, u32 parentid);
406 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
407 const struct qdisc_size_table *stab);
408 int skb_do_redirect(struct sk_buff *);
410 static inline void skb_reset_tc(struct sk_buff *skb)
412 #ifdef CONFIG_NET_CLS_ACT
413 skb->tc_redirected = 0;
417 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
419 #ifdef CONFIG_NET_CLS_ACT
420 return skb->tc_at_ingress;
426 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
428 #ifdef CONFIG_NET_CLS_ACT
429 if (skb->tc_skip_classify) {
430 skb->tc_skip_classify = 0;
437 /* Reset all TX qdiscs greater then index of a device. */
438 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
442 for (; i < dev->num_tx_queues; i++) {
443 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
445 spin_lock_bh(qdisc_lock(qdisc));
447 spin_unlock_bh(qdisc_lock(qdisc));
452 static inline void qdisc_reset_all_tx(struct net_device *dev)
454 qdisc_reset_all_tx_gt(dev, 0);
457 /* Are all TX queues of the device empty? */
458 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
463 for (i = 0; i < dev->num_tx_queues; i++) {
464 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
465 const struct Qdisc *q = rcu_dereference(txq->qdisc);
476 /* Are any of the TX qdiscs changing? */
477 static inline bool qdisc_tx_changing(const struct net_device *dev)
481 for (i = 0; i < dev->num_tx_queues; i++) {
482 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
483 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
489 /* Is the device using the noop qdisc on all queues? */
490 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
494 for (i = 0; i < dev->num_tx_queues; i++) {
495 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
496 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
502 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
504 return qdisc_skb_cb(skb)->pkt_len;
507 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
508 enum net_xmit_qdisc_t {
509 __NET_XMIT_STOLEN = 0x00010000,
510 __NET_XMIT_BYPASS = 0x00020000,
513 #ifdef CONFIG_NET_CLS_ACT
514 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
516 #define net_xmit_drop_count(e) (1)
519 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
520 const struct Qdisc *sch)
522 #ifdef CONFIG_NET_SCHED
523 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
526 __qdisc_calculate_pkt_len(skb, stab);
530 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
531 struct sk_buff **to_free)
533 qdisc_calculate_pkt_len(skb, sch);
534 return sch->enqueue(skb, sch, to_free);
537 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
539 return q->flags & TCQ_F_CPUSTATS;
542 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
543 __u64 bytes, __u32 packets)
545 bstats->bytes += bytes;
546 bstats->packets += packets;
549 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
550 const struct sk_buff *skb)
552 _bstats_update(bstats,
554 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
557 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
558 __u64 bytes, __u32 packets)
560 u64_stats_update_begin(&bstats->syncp);
561 _bstats_update(&bstats->bstats, bytes, packets);
562 u64_stats_update_end(&bstats->syncp);
565 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
566 const struct sk_buff *skb)
568 u64_stats_update_begin(&bstats->syncp);
569 bstats_update(&bstats->bstats, skb);
570 u64_stats_update_end(&bstats->syncp);
573 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
574 const struct sk_buff *skb)
576 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
579 static inline void qdisc_bstats_update(struct Qdisc *sch,
580 const struct sk_buff *skb)
582 bstats_update(&sch->bstats, skb);
585 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
586 const struct sk_buff *skb)
588 sch->qstats.backlog -= qdisc_pkt_len(skb);
591 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
592 const struct sk_buff *skb)
594 sch->qstats.backlog += qdisc_pkt_len(skb);
597 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
599 sch->qstats.drops += count;
602 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
607 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
609 qstats->overlimits++;
612 static inline void qdisc_qstats_drop(struct Qdisc *sch)
614 qstats_drop_inc(&sch->qstats);
617 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
619 this_cpu_inc(sch->cpu_qstats->drops);
622 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
624 sch->qstats.overlimits++;
627 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
634 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
635 struct qdisc_skb_head *qh)
637 struct sk_buff *last = qh->tail;
648 qdisc_qstats_backlog_inc(sch, skb);
650 return NET_XMIT_SUCCESS;
653 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
655 return __qdisc_enqueue_tail(skb, sch, &sch->q);
658 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
660 struct sk_buff *skb = qh->head;
662 if (likely(skb != NULL)) {
663 qh->head = skb->next;
665 if (qh->head == NULL)
673 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
675 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
677 if (likely(skb != NULL)) {
678 qdisc_qstats_backlog_dec(sch, skb);
679 qdisc_bstats_update(sch, skb);
685 /* Instead of calling kfree_skb() while root qdisc lock is held,
686 * queue the skb for future freeing at end of __dev_xmit_skb()
688 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
690 skb->next = *to_free;
694 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
695 struct qdisc_skb_head *qh,
696 struct sk_buff **to_free)
698 struct sk_buff *skb = __qdisc_dequeue_head(qh);
700 if (likely(skb != NULL)) {
701 unsigned int len = qdisc_pkt_len(skb);
703 qdisc_qstats_backlog_dec(sch, skb);
704 __qdisc_drop(skb, to_free);
711 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
712 struct sk_buff **to_free)
714 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
717 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
719 const struct qdisc_skb_head *qh = &sch->q;
724 /* generic pseudo peek method for non-work-conserving qdisc */
725 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
727 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
729 sch->gso_skb = sch->dequeue(sch);
731 /* it's still part of the queue */
732 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
740 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
741 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
743 struct sk_buff *skb = sch->gso_skb;
747 qdisc_qstats_backlog_dec(sch, skb);
750 skb = sch->dequeue(sch);
756 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
759 * We do not know the backlog in bytes of this list, it
760 * is up to the caller to correct it
764 rtnl_kfree_skbs(qh->head, qh->tail);
772 static inline void qdisc_reset_queue(struct Qdisc *sch)
774 __qdisc_reset_queue(&sch->q);
775 sch->qstats.backlog = 0;
778 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
787 qdisc_tree_reduce_backlog(old, old->q.qlen, old->qstats.backlog);
790 sch_tree_unlock(sch);
795 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
797 rtnl_kfree_skbs(skb, skb);
798 qdisc_qstats_drop(sch);
802 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
803 struct sk_buff **to_free)
805 __qdisc_drop(skb, to_free);
806 qdisc_qstats_drop(sch);
808 return NET_XMIT_DROP;
811 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
812 long it will take to send a packet given its size.
814 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
816 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
819 slot >>= rtab->rate.cell_log;
821 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
822 return rtab->data[slot];
825 struct psched_ratecfg {
826 u64 rate_bytes_ps; /* bytes per second */
833 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
838 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
839 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
841 return ((u64)len * r->mult) >> r->shift;
844 void psched_ratecfg_precompute(struct psched_ratecfg *r,
845 const struct tc_ratespec *conf,
848 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
849 const struct psched_ratecfg *r)
851 memset(res, 0, sizeof(*res));
853 /* legacy struct tc_ratespec has a 32bit @rate field
854 * Qdisc using 64bit rate should add new attributes
855 * in order to maintain compatibility.
857 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
859 res->overhead = r->overhead;
860 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);