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>
13 #include <net/net_seq_lock.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.
71 const struct Qdisc_ops *ops;
72 struct qdisc_size_table __rcu *stab;
73 struct hlist_node hash;
78 struct netdev_queue *dev_queue;
80 struct gnet_stats_rate_est64 rate_est;
81 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
82 struct gnet_stats_queue __percpu *cpu_qstats;
85 * For performance sake on SMP, we put highly modified fields at the end
87 struct sk_buff *gso_skb ____cacheline_aligned_in_smp;
88 struct qdisc_skb_head q;
89 struct gnet_stats_basic_packed bstats;
90 net_seqlock_t running;
91 struct gnet_stats_queue qstats;
93 struct Qdisc *next_sched;
94 struct sk_buff *skb_bad_txq;
95 struct rcu_head rcu_head;
99 spinlock_t busylock ____cacheline_aligned_in_smp;
102 static inline bool qdisc_is_running(struct Qdisc *qdisc)
104 #ifdef CONFIG_PREEMPT_RT_BASE
105 return spin_is_locked(&qdisc->running.lock) ? true : false;
107 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
111 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
113 #ifdef CONFIG_PREEMPT_RT_BASE
114 if (try_write_seqlock(&qdisc->running))
118 if (qdisc_is_running(qdisc))
120 /* Variant of write_seqcount_begin() telling lockdep a trylock
123 raw_write_seqcount_begin(&qdisc->running);
124 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
129 static inline void qdisc_run_end(struct Qdisc *qdisc)
131 #ifdef CONFIG_PREEMPT_RT_BASE
132 write_sequnlock(&qdisc->running);
134 write_seqcount_end(&qdisc->running);
138 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
140 return qdisc->flags & TCQ_F_ONETXQUEUE;
143 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
146 /* Non-BQL migrated drivers will return 0, too. */
147 return dql_avail(&txq->dql);
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 bool (*tcf_cl_offload)(u32 classid);
172 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
174 void (*unbind_tcf)(struct Qdisc *, unsigned long);
176 /* rtnetlink specific */
177 int (*dump)(struct Qdisc *, unsigned long,
178 struct sk_buff *skb, struct tcmsg*);
179 int (*dump_stats)(struct Qdisc *, unsigned long,
184 struct Qdisc_ops *next;
185 const struct Qdisc_class_ops *cl_ops;
189 int (*enqueue)(struct sk_buff *skb,
191 struct sk_buff **to_free);
192 struct sk_buff * (*dequeue)(struct Qdisc *);
193 struct sk_buff * (*peek)(struct Qdisc *);
195 int (*init)(struct Qdisc *, struct nlattr *arg);
196 void (*reset)(struct Qdisc *);
197 void (*destroy)(struct Qdisc *);
198 int (*change)(struct Qdisc *, struct nlattr *arg);
199 void (*attach)(struct Qdisc *);
201 int (*dump)(struct Qdisc *, struct sk_buff *);
202 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
204 struct module *owner;
213 struct tcf_proto_ops {
214 struct list_head head;
217 int (*classify)(struct sk_buff *,
218 const struct tcf_proto *,
219 struct tcf_result *);
220 int (*init)(struct tcf_proto*);
221 bool (*destroy)(struct tcf_proto*, bool);
223 unsigned long (*get)(struct tcf_proto*, u32 handle);
224 int (*change)(struct net *net, struct sk_buff *,
225 struct tcf_proto*, unsigned long,
226 u32 handle, struct nlattr **,
227 unsigned long *, bool);
228 int (*delete)(struct tcf_proto*, unsigned long);
229 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
231 /* rtnetlink specific */
232 int (*dump)(struct net*, struct tcf_proto*, unsigned long,
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;
256 struct qdisc_skb_cb {
257 unsigned int pkt_len;
258 u16 slave_dev_queue_mapping;
260 #define QDISC_CB_PRIV_LEN 20
261 unsigned char data[QDISC_CB_PRIV_LEN];
264 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
266 struct qdisc_skb_cb *qcb;
268 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
269 BUILD_BUG_ON(sizeof(qcb->data) < sz);
272 static inline int qdisc_qlen(const struct Qdisc *q)
277 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
279 return (struct qdisc_skb_cb *)skb->cb;
282 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
284 return &qdisc->q.lock;
287 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
289 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
294 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
296 return qdisc->dev_queue->qdisc_sleeping;
299 /* The qdisc root lock is a mechanism by which to top level
300 * of a qdisc tree can be locked from any qdisc node in the
301 * forest. This allows changing the configuration of some
302 * aspect of the qdisc tree while blocking out asynchronous
303 * qdisc access in the packet processing paths.
305 * It is only legal to do this when the root will not change
306 * on us. Otherwise we'll potentially lock the wrong qdisc
307 * root. This is enforced by holding the RTNL semaphore, which
308 * all users of this lock accessor must do.
310 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
312 struct Qdisc *root = qdisc_root(qdisc);
315 return qdisc_lock(root);
318 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
320 struct Qdisc *root = qdisc_root_sleeping(qdisc);
323 return qdisc_lock(root);
326 static inline net_seqlock_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
328 struct Qdisc *root = qdisc_root_sleeping(qdisc);
331 return &root->running;
334 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
336 return qdisc->dev_queue->dev;
339 static inline void sch_tree_lock(const struct Qdisc *q)
341 spin_lock_bh(qdisc_root_sleeping_lock(q));
344 static inline void sch_tree_unlock(const struct Qdisc *q)
346 spin_unlock_bh(qdisc_root_sleeping_lock(q));
349 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
350 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
352 extern struct Qdisc noop_qdisc;
353 extern struct Qdisc_ops noop_qdisc_ops;
354 extern struct Qdisc_ops pfifo_fast_ops;
355 extern struct Qdisc_ops mq_qdisc_ops;
356 extern struct Qdisc_ops noqueue_qdisc_ops;
357 extern const struct Qdisc_ops *default_qdisc_ops;
358 static inline const struct Qdisc_ops *
359 get_default_qdisc_ops(const struct net_device *dev, int ntx)
361 return ntx < dev->real_num_tx_queues ?
362 default_qdisc_ops : &pfifo_fast_ops;
365 struct Qdisc_class_common {
367 struct hlist_node hnode;
370 struct Qdisc_class_hash {
371 struct hlist_head *hash;
372 unsigned int hashsize;
373 unsigned int hashmask;
374 unsigned int hashelems;
377 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
384 static inline struct Qdisc_class_common *
385 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
387 struct Qdisc_class_common *cl;
390 h = qdisc_class_hash(id, hash->hashmask);
391 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
392 if (cl->classid == id)
398 int qdisc_class_hash_init(struct Qdisc_class_hash *);
399 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
400 struct Qdisc_class_common *);
401 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
402 struct Qdisc_class_common *);
403 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
404 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
406 void dev_init_scheduler(struct net_device *dev);
407 void dev_shutdown(struct net_device *dev);
408 void dev_activate(struct net_device *dev);
409 void dev_deactivate(struct net_device *dev);
410 void dev_deactivate_many(struct list_head *head);
411 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
412 struct Qdisc *qdisc);
413 void qdisc_reset(struct Qdisc *qdisc);
414 void qdisc_destroy(struct Qdisc *qdisc);
415 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
417 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
418 const struct Qdisc_ops *ops);
419 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
420 const struct Qdisc_ops *ops, u32 parentid);
421 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
422 const struct qdisc_size_table *stab);
423 bool tcf_destroy(struct tcf_proto *tp, bool force);
424 void tcf_destroy_chain(struct tcf_proto __rcu **fl);
425 int skb_do_redirect(struct sk_buff *);
427 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
429 #ifdef CONFIG_NET_CLS_ACT
430 return G_TC_AT(skb->tc_verd) & AT_INGRESS;
436 /* Reset all TX qdiscs greater then index of a device. */
437 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
441 for (; i < dev->num_tx_queues; i++) {
442 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
444 spin_lock_bh(qdisc_lock(qdisc));
446 spin_unlock_bh(qdisc_lock(qdisc));
451 static inline void qdisc_reset_all_tx(struct net_device *dev)
453 qdisc_reset_all_tx_gt(dev, 0);
456 /* Are all TX queues of the device empty? */
457 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
462 for (i = 0; i < dev->num_tx_queues; i++) {
463 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
464 const struct Qdisc *q = rcu_dereference(txq->qdisc);
475 /* Are any of the TX qdiscs changing? */
476 static inline bool qdisc_tx_changing(const struct net_device *dev)
480 for (i = 0; i < dev->num_tx_queues; i++) {
481 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
482 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
488 /* Is the device using the noop qdisc on all queues? */
489 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
493 for (i = 0; i < dev->num_tx_queues; i++) {
494 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
495 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
501 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
503 return qdisc_skb_cb(skb)->pkt_len;
506 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
507 enum net_xmit_qdisc_t {
508 __NET_XMIT_STOLEN = 0x00010000,
509 __NET_XMIT_BYPASS = 0x00020000,
512 #ifdef CONFIG_NET_CLS_ACT
513 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
515 #define net_xmit_drop_count(e) (1)
518 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
519 const struct Qdisc *sch)
521 #ifdef CONFIG_NET_SCHED
522 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
525 __qdisc_calculate_pkt_len(skb, stab);
529 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
530 struct sk_buff **to_free)
532 qdisc_calculate_pkt_len(skb, sch);
533 return sch->enqueue(skb, sch, to_free);
536 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
538 return q->flags & TCQ_F_CPUSTATS;
541 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
542 __u64 bytes, __u32 packets)
544 bstats->bytes += bytes;
545 bstats->packets += packets;
548 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
549 const struct sk_buff *skb)
551 _bstats_update(bstats,
553 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
556 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
557 __u64 bytes, __u32 packets)
559 u64_stats_update_begin(&bstats->syncp);
560 _bstats_update(&bstats->bstats, bytes, packets);
561 u64_stats_update_end(&bstats->syncp);
564 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
565 const struct sk_buff *skb)
567 u64_stats_update_begin(&bstats->syncp);
568 bstats_update(&bstats->bstats, skb);
569 u64_stats_update_end(&bstats->syncp);
572 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
573 const struct sk_buff *skb)
575 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
578 static inline void qdisc_bstats_update(struct Qdisc *sch,
579 const struct sk_buff *skb)
581 bstats_update(&sch->bstats, skb);
584 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
585 const struct sk_buff *skb)
587 sch->qstats.backlog -= qdisc_pkt_len(skb);
590 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
591 const struct sk_buff *skb)
593 sch->qstats.backlog += qdisc_pkt_len(skb);
596 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
598 sch->qstats.drops += count;
601 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
606 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
608 qstats->overlimits++;
611 static inline void qdisc_qstats_drop(struct Qdisc *sch)
613 qstats_drop_inc(&sch->qstats);
616 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
618 this_cpu_inc(sch->cpu_qstats->drops);
621 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
623 sch->qstats.overlimits++;
626 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
633 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
634 struct qdisc_skb_head *qh)
636 struct sk_buff *last = qh->tail;
647 qdisc_qstats_backlog_inc(sch, skb);
649 return NET_XMIT_SUCCESS;
652 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
654 return __qdisc_enqueue_tail(skb, sch, &sch->q);
657 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
659 struct sk_buff *skb = qh->head;
661 if (likely(skb != NULL)) {
662 qh->head = skb->next;
664 if (qh->head == NULL)
672 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
674 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
676 if (likely(skb != NULL)) {
677 qdisc_qstats_backlog_dec(sch, skb);
678 qdisc_bstats_update(sch, skb);
684 /* Instead of calling kfree_skb() while root qdisc lock is held,
685 * queue the skb for future freeing at end of __dev_xmit_skb()
687 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
689 skb->next = *to_free;
693 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
694 struct qdisc_skb_head *qh,
695 struct sk_buff **to_free)
697 struct sk_buff *skb = __qdisc_dequeue_head(qh);
699 if (likely(skb != NULL)) {
700 unsigned int len = qdisc_pkt_len(skb);
702 qdisc_qstats_backlog_dec(sch, skb);
703 __qdisc_drop(skb, to_free);
710 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
711 struct sk_buff **to_free)
713 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
716 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
718 const struct qdisc_skb_head *qh = &sch->q;
723 /* generic pseudo peek method for non-work-conserving qdisc */
724 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
726 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
728 sch->gso_skb = sch->dequeue(sch);
730 /* it's still part of the queue */
731 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
739 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
740 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
742 struct sk_buff *skb = sch->gso_skb;
746 qdisc_qstats_backlog_dec(sch, skb);
749 skb = sch->dequeue(sch);
755 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
758 * We do not know the backlog in bytes of this list, it
759 * is up to the caller to correct it
763 rtnl_kfree_skbs(qh->head, qh->tail);
771 static inline void qdisc_reset_queue(struct Qdisc *sch)
773 __qdisc_reset_queue(&sch->q);
774 sch->qstats.backlog = 0;
777 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
786 qdisc_tree_reduce_backlog(old, old->q.qlen, old->qstats.backlog);
789 sch_tree_unlock(sch);
794 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
796 rtnl_kfree_skbs(skb, skb);
797 qdisc_qstats_drop(sch);
801 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
802 struct sk_buff **to_free)
804 __qdisc_drop(skb, to_free);
805 qdisc_qstats_drop(sch);
807 return NET_XMIT_DROP;
810 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
811 long it will take to send a packet given its size.
813 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
815 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
818 slot >>= rtab->rate.cell_log;
820 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
821 return rtab->data[slot];
824 struct psched_ratecfg {
825 u64 rate_bytes_ps; /* bytes per second */
832 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
837 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
838 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
840 return ((u64)len * r->mult) >> r->shift;
843 void psched_ratecfg_precompute(struct psched_ratecfg *r,
844 const struct tc_ratespec *conf,
847 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
848 const struct psched_ratecfg *r)
850 memset(res, 0, sizeof(*res));
852 /* legacy struct tc_ratespec has a 32bit @rate field
853 * Qdisc using 64bit rate should add new attributes
854 * in order to maintain compatibility.
856 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
858 res->overhead = r->overhead;
859 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);