1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __NET_SCHED_GENERIC_H
3 #define __NET_SCHED_GENERIC_H
5 #include <linux/netdevice.h>
6 #include <linux/types.h>
7 #include <linux/rcupdate.h>
8 #include <linux/pkt_sched.h>
9 #include <linux/pkt_cls.h>
10 #include <linux/percpu.h>
11 #include <linux/dynamic_queue_limits.h>
12 #include <linux/list.h>
13 #include <linux/refcount.h>
14 #include <linux/workqueue.h>
15 #include <net/gen_stats.h>
16 #include <net/rtnetlink.h>
23 struct qdisc_rate_table {
24 struct tc_ratespec rate;
26 struct qdisc_rate_table *next;
32 __QDISC_STATE_DEACTIVATED,
35 struct qdisc_size_table {
37 struct list_head list;
38 struct tc_sizespec szopts;
43 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
44 struct qdisc_skb_head {
52 int (*enqueue)(struct sk_buff *skb,
54 struct sk_buff **to_free);
55 struct sk_buff * (*dequeue)(struct Qdisc *sch);
57 #define TCQ_F_BUILTIN 1
58 #define TCQ_F_INGRESS 2
59 #define TCQ_F_CAN_BYPASS 4
60 #define TCQ_F_MQROOT 8
61 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
62 * q->dev_queue : It can test
63 * netif_xmit_frozen_or_stopped() before
64 * dequeueing next packet.
65 * Its true for MQ/MQPRIO slaves, or non
68 #define TCQ_F_WARN_NONWC (1 << 16)
69 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
70 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
71 * qdisc_tree_decrease_qlen() should stop.
73 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
75 const struct Qdisc_ops *ops;
76 struct qdisc_size_table __rcu *stab;
77 struct hlist_node hash;
81 struct netdev_queue *dev_queue;
83 struct net_rate_estimator __rcu *rate_est;
84 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
85 struct gnet_stats_queue __percpu *cpu_qstats;
88 * For performance sake on SMP, we put highly modified fields at the end
90 struct sk_buff *gso_skb ____cacheline_aligned_in_smp;
91 struct qdisc_skb_head q;
92 struct gnet_stats_basic_packed bstats;
94 struct gnet_stats_queue qstats;
96 struct Qdisc *next_sched;
97 struct sk_buff *skb_bad_txq;
98 struct rcu_head rcu_head;
102 spinlock_t busylock ____cacheline_aligned_in_smp;
105 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
107 if (qdisc->flags & TCQ_F_BUILTIN)
109 refcount_inc(&qdisc->refcnt);
112 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
114 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
117 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
119 if (qdisc_is_running(qdisc))
121 /* Variant of write_seqcount_begin() telling lockdep a trylock
124 raw_write_seqcount_begin(&qdisc->running);
125 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
129 static inline void qdisc_run_end(struct Qdisc *qdisc)
131 write_seqcount_end(&qdisc->running);
134 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
136 return qdisc->flags & TCQ_F_ONETXQUEUE;
139 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
142 /* Non-BQL migrated drivers will return 0, too. */
143 return dql_avail(&txq->dql);
149 struct Qdisc_class_ops {
150 /* Child qdisc manipulation */
151 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
152 int (*graft)(struct Qdisc *, unsigned long cl,
153 struct Qdisc *, struct Qdisc **);
154 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
155 void (*qlen_notify)(struct Qdisc *, unsigned long);
157 /* Class manipulation routines */
158 unsigned long (*find)(struct Qdisc *, u32 classid);
159 int (*change)(struct Qdisc *, u32, u32,
160 struct nlattr **, unsigned long *);
161 int (*delete)(struct Qdisc *, unsigned long);
162 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
164 /* Filter manipulation */
165 struct tcf_block * (*tcf_block)(struct Qdisc *, unsigned long);
166 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
168 void (*unbind_tcf)(struct Qdisc *, unsigned long);
170 /* rtnetlink specific */
171 int (*dump)(struct Qdisc *, unsigned long,
172 struct sk_buff *skb, struct tcmsg*);
173 int (*dump_stats)(struct Qdisc *, unsigned long,
178 struct Qdisc_ops *next;
179 const struct Qdisc_class_ops *cl_ops;
183 int (*enqueue)(struct sk_buff *skb,
185 struct sk_buff **to_free);
186 struct sk_buff * (*dequeue)(struct Qdisc *);
187 struct sk_buff * (*peek)(struct Qdisc *);
189 int (*init)(struct Qdisc *, struct nlattr *arg);
190 void (*reset)(struct Qdisc *);
191 void (*destroy)(struct Qdisc *);
192 int (*change)(struct Qdisc *, struct nlattr *arg);
193 void (*attach)(struct Qdisc *);
195 int (*dump)(struct Qdisc *, struct sk_buff *);
196 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
198 struct module *owner;
208 const struct tcf_proto *goto_tp;
212 struct tcf_proto_ops {
213 struct list_head head;
216 int (*classify)(struct sk_buff *,
217 const struct tcf_proto *,
218 struct tcf_result *);
219 int (*init)(struct tcf_proto*);
220 void (*destroy)(struct tcf_proto*);
222 void* (*get)(struct tcf_proto*, u32 handle);
223 int (*change)(struct net *net, struct sk_buff *,
224 struct tcf_proto*, unsigned long,
225 u32 handle, struct nlattr **,
227 int (*delete)(struct tcf_proto*, void *, bool*);
228 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
229 void (*bind_class)(void *, u32, unsigned long);
231 /* rtnetlink specific */
232 int (*dump)(struct net*, struct tcf_proto*, void *,
233 struct sk_buff *skb, struct tcmsg*);
235 struct module *owner;
239 /* Fast access part */
240 struct tcf_proto __rcu *next;
242 int (*classify)(struct sk_buff *,
243 const struct tcf_proto *,
244 struct tcf_result *);
252 const struct tcf_proto_ops *ops;
253 struct tcf_chain *chain;
257 struct qdisc_skb_cb {
258 unsigned int pkt_len;
259 u16 slave_dev_queue_mapping;
261 #define QDISC_CB_PRIV_LEN 20
262 unsigned char data[QDISC_CB_PRIV_LEN];
266 struct tcf_proto __rcu *filter_chain;
267 struct tcf_proto __rcu **p_filter_chain;
268 struct list_head list;
269 struct tcf_block *block;
270 u32 index; /* chain index */
275 struct list_head chain_list;
276 struct work_struct work;
279 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
281 struct qdisc_skb_cb *qcb;
283 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
284 BUILD_BUG_ON(sizeof(qcb->data) < sz);
287 static inline int qdisc_qlen(const struct Qdisc *q)
292 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
294 return (struct qdisc_skb_cb *)skb->cb;
297 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
299 return &qdisc->q.lock;
302 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
304 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
309 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
311 return qdisc->dev_queue->qdisc_sleeping;
314 /* The qdisc root lock is a mechanism by which to top level
315 * of a qdisc tree can be locked from any qdisc node in the
316 * forest. This allows changing the configuration of some
317 * aspect of the qdisc tree while blocking out asynchronous
318 * qdisc access in the packet processing paths.
320 * It is only legal to do this when the root will not change
321 * on us. Otherwise we'll potentially lock the wrong qdisc
322 * root. This is enforced by holding the RTNL semaphore, which
323 * all users of this lock accessor must do.
325 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
327 struct Qdisc *root = qdisc_root(qdisc);
330 return qdisc_lock(root);
333 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
335 struct Qdisc *root = qdisc_root_sleeping(qdisc);
338 return qdisc_lock(root);
341 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
343 struct Qdisc *root = qdisc_root_sleeping(qdisc);
346 return &root->running;
349 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
351 return qdisc->dev_queue->dev;
354 static inline void sch_tree_lock(const struct Qdisc *q)
356 spin_lock_bh(qdisc_root_sleeping_lock(q));
359 static inline void sch_tree_unlock(const struct Qdisc *q)
361 spin_unlock_bh(qdisc_root_sleeping_lock(q));
364 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
365 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
367 extern struct Qdisc noop_qdisc;
368 extern struct Qdisc_ops noop_qdisc_ops;
369 extern struct Qdisc_ops pfifo_fast_ops;
370 extern struct Qdisc_ops mq_qdisc_ops;
371 extern struct Qdisc_ops noqueue_qdisc_ops;
372 extern const struct Qdisc_ops *default_qdisc_ops;
373 static inline const struct Qdisc_ops *
374 get_default_qdisc_ops(const struct net_device *dev, int ntx)
376 return ntx < dev->real_num_tx_queues ?
377 default_qdisc_ops : &pfifo_fast_ops;
380 struct Qdisc_class_common {
382 struct hlist_node hnode;
385 struct Qdisc_class_hash {
386 struct hlist_head *hash;
387 unsigned int hashsize;
388 unsigned int hashmask;
389 unsigned int hashelems;
392 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
399 static inline struct Qdisc_class_common *
400 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
402 struct Qdisc_class_common *cl;
408 h = qdisc_class_hash(id, hash->hashmask);
409 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
410 if (cl->classid == id)
416 int qdisc_class_hash_init(struct Qdisc_class_hash *);
417 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
418 struct Qdisc_class_common *);
419 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
420 struct Qdisc_class_common *);
421 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
422 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
424 void dev_init_scheduler(struct net_device *dev);
425 void dev_shutdown(struct net_device *dev);
426 void dev_activate(struct net_device *dev);
427 void dev_deactivate(struct net_device *dev);
428 void dev_deactivate_many(struct list_head *head);
429 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
430 struct Qdisc *qdisc);
431 void qdisc_reset(struct Qdisc *qdisc);
432 void qdisc_destroy(struct Qdisc *qdisc);
433 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
435 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
436 const struct Qdisc_ops *ops);
437 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
438 const struct Qdisc_ops *ops, u32 parentid);
439 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
440 const struct qdisc_size_table *stab);
441 int skb_do_redirect(struct sk_buff *);
443 static inline void skb_reset_tc(struct sk_buff *skb)
445 #ifdef CONFIG_NET_CLS_ACT
446 skb->tc_redirected = 0;
450 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
452 #ifdef CONFIG_NET_CLS_ACT
453 return skb->tc_at_ingress;
459 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
461 #ifdef CONFIG_NET_CLS_ACT
462 if (skb->tc_skip_classify) {
463 skb->tc_skip_classify = 0;
470 /* Reset all TX qdiscs greater then index of a device. */
471 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
475 for (; i < dev->num_tx_queues; i++) {
476 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
478 spin_lock_bh(qdisc_lock(qdisc));
480 spin_unlock_bh(qdisc_lock(qdisc));
485 static inline void qdisc_reset_all_tx(struct net_device *dev)
487 qdisc_reset_all_tx_gt(dev, 0);
490 /* Are all TX queues of the device empty? */
491 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
496 for (i = 0; i < dev->num_tx_queues; i++) {
497 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
498 const struct Qdisc *q = rcu_dereference(txq->qdisc);
509 /* Are any of the TX qdiscs changing? */
510 static inline bool qdisc_tx_changing(const struct net_device *dev)
514 for (i = 0; i < dev->num_tx_queues; i++) {
515 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
516 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
522 /* Is the device using the noop qdisc on all queues? */
523 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
527 for (i = 0; i < dev->num_tx_queues; i++) {
528 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
529 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
535 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
537 return qdisc_skb_cb(skb)->pkt_len;
540 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
541 enum net_xmit_qdisc_t {
542 __NET_XMIT_STOLEN = 0x00010000,
543 __NET_XMIT_BYPASS = 0x00020000,
546 #ifdef CONFIG_NET_CLS_ACT
547 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
549 #define net_xmit_drop_count(e) (1)
552 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
553 const struct Qdisc *sch)
555 #ifdef CONFIG_NET_SCHED
556 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
559 __qdisc_calculate_pkt_len(skb, stab);
563 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
564 struct sk_buff **to_free)
566 qdisc_calculate_pkt_len(skb, sch);
567 return sch->enqueue(skb, sch, to_free);
570 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
572 return q->flags & TCQ_F_CPUSTATS;
575 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
576 __u64 bytes, __u32 packets)
578 bstats->bytes += bytes;
579 bstats->packets += packets;
582 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
583 const struct sk_buff *skb)
585 _bstats_update(bstats,
587 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
590 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
591 __u64 bytes, __u32 packets)
593 u64_stats_update_begin(&bstats->syncp);
594 _bstats_update(&bstats->bstats, bytes, packets);
595 u64_stats_update_end(&bstats->syncp);
598 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
599 const struct sk_buff *skb)
601 u64_stats_update_begin(&bstats->syncp);
602 bstats_update(&bstats->bstats, skb);
603 u64_stats_update_end(&bstats->syncp);
606 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
607 const struct sk_buff *skb)
609 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
612 static inline void qdisc_bstats_update(struct Qdisc *sch,
613 const struct sk_buff *skb)
615 bstats_update(&sch->bstats, skb);
618 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
619 const struct sk_buff *skb)
621 sch->qstats.backlog -= qdisc_pkt_len(skb);
624 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
625 const struct sk_buff *skb)
627 sch->qstats.backlog += qdisc_pkt_len(skb);
630 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
632 sch->qstats.drops += count;
635 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
640 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
642 qstats->overlimits++;
645 static inline void qdisc_qstats_drop(struct Qdisc *sch)
647 qstats_drop_inc(&sch->qstats);
650 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
652 this_cpu_inc(sch->cpu_qstats->drops);
655 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
657 sch->qstats.overlimits++;
660 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
667 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
668 struct qdisc_skb_head *qh)
670 struct sk_buff *last = qh->tail;
681 qdisc_qstats_backlog_inc(sch, skb);
683 return NET_XMIT_SUCCESS;
686 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
688 return __qdisc_enqueue_tail(skb, sch, &sch->q);
691 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
693 struct sk_buff *skb = qh->head;
695 if (likely(skb != NULL)) {
696 qh->head = skb->next;
698 if (qh->head == NULL)
706 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
708 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
710 if (likely(skb != NULL)) {
711 qdisc_qstats_backlog_dec(sch, skb);
712 qdisc_bstats_update(sch, skb);
718 /* Instead of calling kfree_skb() while root qdisc lock is held,
719 * queue the skb for future freeing at end of __dev_xmit_skb()
721 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
723 skb->next = *to_free;
727 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
728 struct qdisc_skb_head *qh,
729 struct sk_buff **to_free)
731 struct sk_buff *skb = __qdisc_dequeue_head(qh);
733 if (likely(skb != NULL)) {
734 unsigned int len = qdisc_pkt_len(skb);
736 qdisc_qstats_backlog_dec(sch, skb);
737 __qdisc_drop(skb, to_free);
744 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
745 struct sk_buff **to_free)
747 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
750 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
752 const struct qdisc_skb_head *qh = &sch->q;
757 /* generic pseudo peek method for non-work-conserving qdisc */
758 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
760 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
762 sch->gso_skb = sch->dequeue(sch);
764 /* it's still part of the queue */
765 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
773 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
774 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
776 struct sk_buff *skb = sch->gso_skb;
780 qdisc_qstats_backlog_dec(sch, skb);
783 skb = sch->dequeue(sch);
789 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
792 * We do not know the backlog in bytes of this list, it
793 * is up to the caller to correct it
797 rtnl_kfree_skbs(qh->head, qh->tail);
805 static inline void qdisc_reset_queue(struct Qdisc *sch)
807 __qdisc_reset_queue(&sch->q);
808 sch->qstats.backlog = 0;
811 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
820 unsigned int qlen = old->q.qlen;
821 unsigned int backlog = old->qstats.backlog;
824 qdisc_tree_reduce_backlog(old, qlen, backlog);
826 sch_tree_unlock(sch);
831 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
833 rtnl_kfree_skbs(skb, skb);
834 qdisc_qstats_drop(sch);
838 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
839 struct sk_buff **to_free)
841 __qdisc_drop(skb, to_free);
842 qdisc_qstats_drop(sch);
844 return NET_XMIT_DROP;
847 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
848 long it will take to send a packet given its size.
850 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
852 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
855 slot >>= rtab->rate.cell_log;
857 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
858 return rtab->data[slot];
861 struct psched_ratecfg {
862 u64 rate_bytes_ps; /* bytes per second */
869 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
874 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
875 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
877 return ((u64)len * r->mult) >> r->shift;
880 void psched_ratecfg_precompute(struct psched_ratecfg *r,
881 const struct tc_ratespec *conf,
884 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
885 const struct psched_ratecfg *r)
887 memset(res, 0, sizeof(*res));
889 /* legacy struct tc_ratespec has a 32bit @rate field
890 * Qdisc using 64bit rate should add new attributes
891 * in order to maintain compatibility.
893 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
895 res->overhead = r->overhead;
896 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);