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 <linux/list.h>
12 #include <net/gen_stats.h>
13 #include <net/rtnetlink.h>
20 struct qdisc_rate_table {
21 struct tc_ratespec rate;
23 struct qdisc_rate_table *next;
29 __QDISC_STATE_DEACTIVATED,
32 struct qdisc_size_table {
34 struct list_head list;
35 struct tc_sizespec szopts;
40 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
41 struct qdisc_skb_head {
49 int (*enqueue)(struct sk_buff *skb,
51 struct sk_buff **to_free);
52 struct sk_buff * (*dequeue)(struct Qdisc *sch);
54 #define TCQ_F_BUILTIN 1
55 #define TCQ_F_INGRESS 2
56 #define TCQ_F_CAN_BYPASS 4
57 #define TCQ_F_MQROOT 8
58 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
59 * q->dev_queue : It can test
60 * netif_xmit_frozen_or_stopped() before
61 * dequeueing next packet.
62 * Its true for MQ/MQPRIO slaves, or non
65 #define TCQ_F_WARN_NONWC (1 << 16)
66 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
67 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
68 * qdisc_tree_decrease_qlen() should stop.
70 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
72 const struct Qdisc_ops *ops;
73 struct qdisc_size_table __rcu *stab;
74 struct hlist_node hash;
79 struct netdev_queue *dev_queue;
81 struct net_rate_estimator __rcu *rate_est;
82 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
83 struct gnet_stats_queue __percpu *cpu_qstats;
86 * For performance sake on SMP, we put highly modified fields at the end
88 struct sk_buff *gso_skb ____cacheline_aligned_in_smp;
89 struct qdisc_skb_head q;
90 struct gnet_stats_basic_packed bstats;
92 struct gnet_stats_queue qstats;
94 struct Qdisc *next_sched;
95 struct sk_buff *skb_bad_txq;
96 struct rcu_head rcu_head;
100 spinlock_t busylock ____cacheline_aligned_in_smp;
103 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
105 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
108 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
110 if (qdisc_is_running(qdisc))
112 /* Variant of write_seqcount_begin() telling lockdep a trylock
115 raw_write_seqcount_begin(&qdisc->running);
116 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
120 static inline void qdisc_run_end(struct Qdisc *qdisc)
122 write_seqcount_end(&qdisc->running);
125 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
127 return qdisc->flags & TCQ_F_ONETXQUEUE;
130 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
133 /* Non-BQL migrated drivers will return 0, too. */
134 return dql_avail(&txq->dql);
140 struct Qdisc_class_ops {
141 /* Child qdisc manipulation */
142 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
143 int (*graft)(struct Qdisc *, unsigned long cl,
144 struct Qdisc *, struct Qdisc **);
145 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
146 void (*qlen_notify)(struct Qdisc *, unsigned long);
148 /* Class manipulation routines */
149 unsigned long (*get)(struct Qdisc *, u32 classid);
150 void (*put)(struct Qdisc *, unsigned long);
151 int (*change)(struct Qdisc *, u32, u32,
152 struct nlattr **, unsigned long *);
153 int (*delete)(struct Qdisc *, unsigned long);
154 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
156 /* Filter manipulation */
157 struct tcf_block * (*tcf_block)(struct Qdisc *, unsigned long);
158 bool (*tcf_cl_offload)(u32 classid);
159 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
161 void (*unbind_tcf)(struct Qdisc *, unsigned long);
163 /* rtnetlink specific */
164 int (*dump)(struct Qdisc *, unsigned long,
165 struct sk_buff *skb, struct tcmsg*);
166 int (*dump_stats)(struct Qdisc *, unsigned long,
171 struct Qdisc_ops *next;
172 const struct Qdisc_class_ops *cl_ops;
176 int (*enqueue)(struct sk_buff *skb,
178 struct sk_buff **to_free);
179 struct sk_buff * (*dequeue)(struct Qdisc *);
180 struct sk_buff * (*peek)(struct Qdisc *);
182 int (*init)(struct Qdisc *, struct nlattr *arg);
183 void (*reset)(struct Qdisc *);
184 void (*destroy)(struct Qdisc *);
185 int (*change)(struct Qdisc *, struct nlattr *arg);
186 void (*attach)(struct Qdisc *);
188 int (*dump)(struct Qdisc *, struct sk_buff *);
189 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
191 struct module *owner;
201 const struct tcf_proto *goto_tp;
205 struct tcf_proto_ops {
206 struct list_head head;
209 int (*classify)(struct sk_buff *,
210 const struct tcf_proto *,
211 struct tcf_result *);
212 int (*init)(struct tcf_proto*);
213 void (*destroy)(struct tcf_proto*);
215 unsigned long (*get)(struct tcf_proto*, u32 handle);
216 int (*change)(struct net *net, struct sk_buff *,
217 struct tcf_proto*, unsigned long,
218 u32 handle, struct nlattr **,
219 unsigned long *, bool);
220 int (*delete)(struct tcf_proto*, unsigned long, bool*);
221 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
223 /* rtnetlink specific */
224 int (*dump)(struct net*, struct tcf_proto*, unsigned long,
225 struct sk_buff *skb, struct tcmsg*);
227 struct module *owner;
231 /* Fast access part */
232 struct tcf_proto __rcu *next;
234 int (*classify)(struct sk_buff *,
235 const struct tcf_proto *,
236 struct tcf_result *);
244 const struct tcf_proto_ops *ops;
245 struct tcf_chain *chain;
249 struct qdisc_skb_cb {
250 unsigned int pkt_len;
251 u16 slave_dev_queue_mapping;
253 #define QDISC_CB_PRIV_LEN 20
254 unsigned char data[QDISC_CB_PRIV_LEN];
258 struct tcf_proto __rcu *filter_chain;
259 struct tcf_proto __rcu **p_filter_chain;
260 struct list_head list;
261 struct tcf_block *block;
262 u32 index; /* chain index */
267 struct list_head chain_list;
270 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
272 struct qdisc_skb_cb *qcb;
274 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
275 BUILD_BUG_ON(sizeof(qcb->data) < sz);
278 static inline int qdisc_qlen(const struct Qdisc *q)
283 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
285 return (struct qdisc_skb_cb *)skb->cb;
288 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
290 return &qdisc->q.lock;
293 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
295 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
300 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
302 return qdisc->dev_queue->qdisc_sleeping;
305 /* The qdisc root lock is a mechanism by which to top level
306 * of a qdisc tree can be locked from any qdisc node in the
307 * forest. This allows changing the configuration of some
308 * aspect of the qdisc tree while blocking out asynchronous
309 * qdisc access in the packet processing paths.
311 * It is only legal to do this when the root will not change
312 * on us. Otherwise we'll potentially lock the wrong qdisc
313 * root. This is enforced by holding the RTNL semaphore, which
314 * all users of this lock accessor must do.
316 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
318 struct Qdisc *root = qdisc_root(qdisc);
321 return qdisc_lock(root);
324 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
326 struct Qdisc *root = qdisc_root_sleeping(qdisc);
329 return qdisc_lock(root);
332 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
334 struct Qdisc *root = qdisc_root_sleeping(qdisc);
337 return &root->running;
340 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
342 return qdisc->dev_queue->dev;
345 static inline void sch_tree_lock(const struct Qdisc *q)
347 spin_lock_bh(qdisc_root_sleeping_lock(q));
350 static inline void sch_tree_unlock(const struct Qdisc *q)
352 spin_unlock_bh(qdisc_root_sleeping_lock(q));
355 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
356 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
358 extern struct Qdisc noop_qdisc;
359 extern struct Qdisc_ops noop_qdisc_ops;
360 extern struct Qdisc_ops pfifo_fast_ops;
361 extern struct Qdisc_ops mq_qdisc_ops;
362 extern struct Qdisc_ops noqueue_qdisc_ops;
363 extern const struct Qdisc_ops *default_qdisc_ops;
364 static inline const struct Qdisc_ops *
365 get_default_qdisc_ops(const struct net_device *dev, int ntx)
367 return ntx < dev->real_num_tx_queues ?
368 default_qdisc_ops : &pfifo_fast_ops;
371 struct Qdisc_class_common {
373 struct hlist_node hnode;
376 struct Qdisc_class_hash {
377 struct hlist_head *hash;
378 unsigned int hashsize;
379 unsigned int hashmask;
380 unsigned int hashelems;
383 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
390 static inline struct Qdisc_class_common *
391 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
393 struct Qdisc_class_common *cl;
396 h = qdisc_class_hash(id, hash->hashmask);
397 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
398 if (cl->classid == id)
404 int qdisc_class_hash_init(struct Qdisc_class_hash *);
405 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
406 struct Qdisc_class_common *);
407 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
408 struct Qdisc_class_common *);
409 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
410 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
412 void dev_init_scheduler(struct net_device *dev);
413 void dev_shutdown(struct net_device *dev);
414 void dev_activate(struct net_device *dev);
415 void dev_deactivate(struct net_device *dev);
416 void dev_deactivate_many(struct list_head *head);
417 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
418 struct Qdisc *qdisc);
419 void qdisc_reset(struct Qdisc *qdisc);
420 void qdisc_destroy(struct Qdisc *qdisc);
421 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
423 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
424 const struct Qdisc_ops *ops);
425 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
426 const struct Qdisc_ops *ops, u32 parentid);
427 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
428 const struct qdisc_size_table *stab);
429 int skb_do_redirect(struct sk_buff *);
431 static inline void skb_reset_tc(struct sk_buff *skb)
433 #ifdef CONFIG_NET_CLS_ACT
434 skb->tc_redirected = 0;
438 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
440 #ifdef CONFIG_NET_CLS_ACT
441 return skb->tc_at_ingress;
447 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
449 #ifdef CONFIG_NET_CLS_ACT
450 if (skb->tc_skip_classify) {
451 skb->tc_skip_classify = 0;
458 /* Reset all TX qdiscs greater then index of a device. */
459 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
463 for (; i < dev->num_tx_queues; i++) {
464 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
466 spin_lock_bh(qdisc_lock(qdisc));
468 spin_unlock_bh(qdisc_lock(qdisc));
473 static inline void qdisc_reset_all_tx(struct net_device *dev)
475 qdisc_reset_all_tx_gt(dev, 0);
478 /* Are all TX queues of the device empty? */
479 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
484 for (i = 0; i < dev->num_tx_queues; i++) {
485 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
486 const struct Qdisc *q = rcu_dereference(txq->qdisc);
497 /* Are any of the TX qdiscs changing? */
498 static inline bool qdisc_tx_changing(const struct net_device *dev)
502 for (i = 0; i < dev->num_tx_queues; i++) {
503 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
504 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
510 /* Is the device using the noop qdisc on all queues? */
511 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
515 for (i = 0; i < dev->num_tx_queues; i++) {
516 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
517 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
523 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
525 return qdisc_skb_cb(skb)->pkt_len;
528 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
529 enum net_xmit_qdisc_t {
530 __NET_XMIT_STOLEN = 0x00010000,
531 __NET_XMIT_BYPASS = 0x00020000,
534 #ifdef CONFIG_NET_CLS_ACT
535 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
537 #define net_xmit_drop_count(e) (1)
540 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
541 const struct Qdisc *sch)
543 #ifdef CONFIG_NET_SCHED
544 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
547 __qdisc_calculate_pkt_len(skb, stab);
551 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
552 struct sk_buff **to_free)
554 qdisc_calculate_pkt_len(skb, sch);
555 return sch->enqueue(skb, sch, to_free);
558 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
560 return q->flags & TCQ_F_CPUSTATS;
563 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
564 __u64 bytes, __u32 packets)
566 bstats->bytes += bytes;
567 bstats->packets += packets;
570 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
571 const struct sk_buff *skb)
573 _bstats_update(bstats,
575 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
578 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
579 __u64 bytes, __u32 packets)
581 u64_stats_update_begin(&bstats->syncp);
582 _bstats_update(&bstats->bstats, bytes, packets);
583 u64_stats_update_end(&bstats->syncp);
586 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
587 const struct sk_buff *skb)
589 u64_stats_update_begin(&bstats->syncp);
590 bstats_update(&bstats->bstats, skb);
591 u64_stats_update_end(&bstats->syncp);
594 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
595 const struct sk_buff *skb)
597 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
600 static inline void qdisc_bstats_update(struct Qdisc *sch,
601 const struct sk_buff *skb)
603 bstats_update(&sch->bstats, skb);
606 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
607 const struct sk_buff *skb)
609 sch->qstats.backlog -= qdisc_pkt_len(skb);
612 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
613 const struct sk_buff *skb)
615 sch->qstats.backlog += qdisc_pkt_len(skb);
618 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
620 sch->qstats.drops += count;
623 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
628 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
630 qstats->overlimits++;
633 static inline void qdisc_qstats_drop(struct Qdisc *sch)
635 qstats_drop_inc(&sch->qstats);
638 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
640 this_cpu_inc(sch->cpu_qstats->drops);
643 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
645 sch->qstats.overlimits++;
648 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
655 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
656 struct qdisc_skb_head *qh)
658 struct sk_buff *last = qh->tail;
669 qdisc_qstats_backlog_inc(sch, skb);
671 return NET_XMIT_SUCCESS;
674 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
676 return __qdisc_enqueue_tail(skb, sch, &sch->q);
679 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
681 struct sk_buff *skb = qh->head;
683 if (likely(skb != NULL)) {
684 qh->head = skb->next;
686 if (qh->head == NULL)
694 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
696 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
698 if (likely(skb != NULL)) {
699 qdisc_qstats_backlog_dec(sch, skb);
700 qdisc_bstats_update(sch, skb);
706 /* Instead of calling kfree_skb() while root qdisc lock is held,
707 * queue the skb for future freeing at end of __dev_xmit_skb()
709 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
711 skb->next = *to_free;
715 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
716 struct qdisc_skb_head *qh,
717 struct sk_buff **to_free)
719 struct sk_buff *skb = __qdisc_dequeue_head(qh);
721 if (likely(skb != NULL)) {
722 unsigned int len = qdisc_pkt_len(skb);
724 qdisc_qstats_backlog_dec(sch, skb);
725 __qdisc_drop(skb, to_free);
732 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
733 struct sk_buff **to_free)
735 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
738 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
740 const struct qdisc_skb_head *qh = &sch->q;
745 /* generic pseudo peek method for non-work-conserving qdisc */
746 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
748 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
750 sch->gso_skb = sch->dequeue(sch);
752 /* it's still part of the queue */
753 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
761 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
762 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
764 struct sk_buff *skb = sch->gso_skb;
768 qdisc_qstats_backlog_dec(sch, skb);
771 skb = sch->dequeue(sch);
777 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
780 * We do not know the backlog in bytes of this list, it
781 * is up to the caller to correct it
785 rtnl_kfree_skbs(qh->head, qh->tail);
793 static inline void qdisc_reset_queue(struct Qdisc *sch)
795 __qdisc_reset_queue(&sch->q);
796 sch->qstats.backlog = 0;
799 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
808 qdisc_tree_reduce_backlog(old, old->q.qlen, old->qstats.backlog);
811 sch_tree_unlock(sch);
816 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
818 rtnl_kfree_skbs(skb, skb);
819 qdisc_qstats_drop(sch);
823 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
824 struct sk_buff **to_free)
826 __qdisc_drop(skb, to_free);
827 qdisc_qstats_drop(sch);
829 return NET_XMIT_DROP;
832 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
833 long it will take to send a packet given its size.
835 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
837 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
840 slot >>= rtab->rate.cell_log;
842 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
843 return rtab->data[slot];
846 struct psched_ratecfg {
847 u64 rate_bytes_ps; /* bytes per second */
854 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
859 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
860 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
862 return ((u64)len * r->mult) >> r->shift;
865 void psched_ratecfg_precompute(struct psched_ratecfg *r,
866 const struct tc_ratespec *conf,
869 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
870 const struct psched_ratecfg *r)
872 memset(res, 0, sizeof(*res));
874 /* legacy struct tc_ratespec has a 32bit @rate field
875 * Qdisc using 64bit rate should add new attributes
876 * in order to maintain compatibility.
878 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
880 res->overhead = r->overhead;
881 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);