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 <linux/refcount.h>
13 #include <net/gen_stats.h>
14 #include <net/rtnetlink.h>
21 struct qdisc_rate_table {
22 struct tc_ratespec rate;
24 struct qdisc_rate_table *next;
30 __QDISC_STATE_DEACTIVATED,
33 struct qdisc_size_table {
35 struct list_head list;
36 struct tc_sizespec szopts;
41 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
42 struct qdisc_skb_head {
50 int (*enqueue)(struct sk_buff *skb,
52 struct sk_buff **to_free);
53 struct sk_buff * (*dequeue)(struct Qdisc *sch);
55 #define TCQ_F_BUILTIN 1
56 #define TCQ_F_INGRESS 2
57 #define TCQ_F_CAN_BYPASS 4
58 #define TCQ_F_MQROOT 8
59 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
60 * q->dev_queue : It can test
61 * netif_xmit_frozen_or_stopped() before
62 * dequeueing next packet.
63 * Its true for MQ/MQPRIO slaves, or non
66 #define TCQ_F_WARN_NONWC (1 << 16)
67 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
68 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
69 * qdisc_tree_decrease_qlen() should stop.
71 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
73 const struct Qdisc_ops *ops;
74 struct qdisc_size_table __rcu *stab;
75 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;
99 spinlock_t busylock ____cacheline_aligned_in_smp;
102 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
104 if (qdisc->flags & TCQ_F_BUILTIN)
106 refcount_inc(&qdisc->refcnt);
109 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
111 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
114 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
116 if (qdisc_is_running(qdisc))
118 /* Variant of write_seqcount_begin() telling lockdep a trylock
121 raw_write_seqcount_begin(&qdisc->running);
122 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
126 static inline void qdisc_run_end(struct Qdisc *qdisc)
128 write_seqcount_end(&qdisc->running);
131 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
133 return qdisc->flags & TCQ_F_ONETXQUEUE;
136 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
139 /* Non-BQL migrated drivers will return 0, too. */
140 return dql_avail(&txq->dql);
146 struct Qdisc_class_ops {
147 /* Child qdisc manipulation */
148 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
149 int (*graft)(struct Qdisc *, unsigned long cl,
150 struct Qdisc *, struct Qdisc **);
151 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
152 void (*qlen_notify)(struct Qdisc *, unsigned long);
154 /* Class manipulation routines */
155 unsigned long (*find)(struct Qdisc *, u32 classid);
156 int (*change)(struct Qdisc *, u32, u32,
157 struct nlattr **, unsigned long *);
158 int (*delete)(struct Qdisc *, unsigned long);
159 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
161 /* Filter manipulation */
162 struct tcf_block * (*tcf_block)(struct Qdisc *, unsigned long);
163 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
165 void (*unbind_tcf)(struct Qdisc *, unsigned long);
167 /* rtnetlink specific */
168 int (*dump)(struct Qdisc *, unsigned long,
169 struct sk_buff *skb, struct tcmsg*);
170 int (*dump_stats)(struct Qdisc *, unsigned long,
175 struct Qdisc_ops *next;
176 const struct Qdisc_class_ops *cl_ops;
180 int (*enqueue)(struct sk_buff *skb,
182 struct sk_buff **to_free);
183 struct sk_buff * (*dequeue)(struct Qdisc *);
184 struct sk_buff * (*peek)(struct Qdisc *);
186 int (*init)(struct Qdisc *, struct nlattr *arg);
187 void (*reset)(struct Qdisc *);
188 void (*destroy)(struct Qdisc *);
189 int (*change)(struct Qdisc *, struct nlattr *arg);
190 void (*attach)(struct Qdisc *);
192 int (*dump)(struct Qdisc *, struct sk_buff *);
193 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
195 struct module *owner;
205 const struct tcf_proto *goto_tp;
209 struct tcf_proto_ops {
210 struct list_head head;
213 int (*classify)(struct sk_buff *,
214 const struct tcf_proto *,
215 struct tcf_result *);
216 int (*init)(struct tcf_proto*);
217 void (*destroy)(struct tcf_proto*);
219 void* (*get)(struct tcf_proto*, u32 handle);
220 int (*change)(struct net *net, struct sk_buff *,
221 struct tcf_proto*, unsigned long,
222 u32 handle, struct nlattr **,
224 int (*delete)(struct tcf_proto*, void *, bool*);
225 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
226 void (*bind_class)(void *, u32, unsigned long);
228 /* rtnetlink specific */
229 int (*dump)(struct net*, struct tcf_proto*, void *,
230 struct sk_buff *skb, struct tcmsg*);
232 struct module *owner;
236 /* Fast access part */
237 struct tcf_proto __rcu *next;
239 int (*classify)(struct sk_buff *,
240 const struct tcf_proto *,
241 struct tcf_result *);
249 const struct tcf_proto_ops *ops;
250 struct tcf_chain *chain;
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];
263 struct tcf_proto __rcu *filter_chain;
264 struct tcf_proto __rcu **p_filter_chain;
265 struct list_head list;
266 struct tcf_block *block;
267 u32 index; /* chain index */
272 struct list_head chain_list;
277 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
279 struct qdisc_skb_cb *qcb;
281 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
282 BUILD_BUG_ON(sizeof(qcb->data) < sz);
285 static inline int qdisc_qlen(const struct Qdisc *q)
290 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
292 return (struct qdisc_skb_cb *)skb->cb;
295 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
297 return &qdisc->q.lock;
300 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
302 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
307 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
309 return qdisc->dev_queue->qdisc_sleeping;
312 /* The qdisc root lock is a mechanism by which to top level
313 * of a qdisc tree can be locked from any qdisc node in the
314 * forest. This allows changing the configuration of some
315 * aspect of the qdisc tree while blocking out asynchronous
316 * qdisc access in the packet processing paths.
318 * It is only legal to do this when the root will not change
319 * on us. Otherwise we'll potentially lock the wrong qdisc
320 * root. This is enforced by holding the RTNL semaphore, which
321 * all users of this lock accessor must do.
323 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
325 struct Qdisc *root = qdisc_root(qdisc);
328 return qdisc_lock(root);
331 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
333 struct Qdisc *root = qdisc_root_sleeping(qdisc);
336 return qdisc_lock(root);
339 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
341 struct Qdisc *root = qdisc_root_sleeping(qdisc);
344 return &root->running;
347 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
349 return qdisc->dev_queue->dev;
352 static inline void sch_tree_lock(const struct Qdisc *q)
354 spin_lock_bh(qdisc_root_sleeping_lock(q));
357 static inline void sch_tree_unlock(const struct Qdisc *q)
359 spin_unlock_bh(qdisc_root_sleeping_lock(q));
362 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
363 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
365 extern struct Qdisc noop_qdisc;
366 extern struct Qdisc_ops noop_qdisc_ops;
367 extern struct Qdisc_ops pfifo_fast_ops;
368 extern struct Qdisc_ops mq_qdisc_ops;
369 extern struct Qdisc_ops noqueue_qdisc_ops;
370 extern const struct Qdisc_ops *default_qdisc_ops;
371 static inline const struct Qdisc_ops *
372 get_default_qdisc_ops(const struct net_device *dev, int ntx)
374 return ntx < dev->real_num_tx_queues ?
375 default_qdisc_ops : &pfifo_fast_ops;
378 struct Qdisc_class_common {
380 struct hlist_node hnode;
383 struct Qdisc_class_hash {
384 struct hlist_head *hash;
385 unsigned int hashsize;
386 unsigned int hashmask;
387 unsigned int hashelems;
390 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
397 static inline struct Qdisc_class_common *
398 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
400 struct Qdisc_class_common *cl;
406 h = qdisc_class_hash(id, hash->hashmask);
407 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
408 if (cl->classid == id)
414 int qdisc_class_hash_init(struct Qdisc_class_hash *);
415 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
416 struct Qdisc_class_common *);
417 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
418 struct Qdisc_class_common *);
419 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
420 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
422 void dev_init_scheduler(struct net_device *dev);
423 void dev_shutdown(struct net_device *dev);
424 void dev_activate(struct net_device *dev);
425 void dev_deactivate(struct net_device *dev);
426 void dev_deactivate_many(struct list_head *head);
427 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
428 struct Qdisc *qdisc);
429 void qdisc_reset(struct Qdisc *qdisc);
430 void qdisc_destroy(struct Qdisc *qdisc);
431 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
433 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
434 const struct Qdisc_ops *ops);
435 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
436 const struct Qdisc_ops *ops, u32 parentid);
437 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
438 const struct qdisc_size_table *stab);
439 int skb_do_redirect(struct sk_buff *);
441 static inline void skb_reset_tc(struct sk_buff *skb)
443 #ifdef CONFIG_NET_CLS_ACT
444 skb->tc_redirected = 0;
448 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
450 #ifdef CONFIG_NET_CLS_ACT
451 return skb->tc_at_ingress;
457 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
459 #ifdef CONFIG_NET_CLS_ACT
460 if (skb->tc_skip_classify) {
461 skb->tc_skip_classify = 0;
468 /* Reset all TX qdiscs greater then index of a device. */
469 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
473 for (; i < dev->num_tx_queues; i++) {
474 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
476 spin_lock_bh(qdisc_lock(qdisc));
478 spin_unlock_bh(qdisc_lock(qdisc));
483 static inline void qdisc_reset_all_tx(struct net_device *dev)
485 qdisc_reset_all_tx_gt(dev, 0);
488 /* Are all TX queues of the device empty? */
489 static inline bool qdisc_all_tx_empty(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 const struct Qdisc *q = rcu_dereference(txq->qdisc);
507 /* Are any of the TX qdiscs changing? */
508 static inline bool qdisc_tx_changing(const struct net_device *dev)
512 for (i = 0; i < dev->num_tx_queues; i++) {
513 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
514 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
520 /* Is the device using the noop qdisc on all queues? */
521 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
525 for (i = 0; i < dev->num_tx_queues; i++) {
526 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
527 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
533 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
535 return qdisc_skb_cb(skb)->pkt_len;
538 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
539 enum net_xmit_qdisc_t {
540 __NET_XMIT_STOLEN = 0x00010000,
541 __NET_XMIT_BYPASS = 0x00020000,
544 #ifdef CONFIG_NET_CLS_ACT
545 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
547 #define net_xmit_drop_count(e) (1)
550 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
551 const struct Qdisc *sch)
553 #ifdef CONFIG_NET_SCHED
554 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
557 __qdisc_calculate_pkt_len(skb, stab);
561 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
562 struct sk_buff **to_free)
564 qdisc_calculate_pkt_len(skb, sch);
565 return sch->enqueue(skb, sch, to_free);
568 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
570 return q->flags & TCQ_F_CPUSTATS;
573 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
574 __u64 bytes, __u32 packets)
576 bstats->bytes += bytes;
577 bstats->packets += packets;
580 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
581 const struct sk_buff *skb)
583 _bstats_update(bstats,
585 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
588 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
589 __u64 bytes, __u32 packets)
591 u64_stats_update_begin(&bstats->syncp);
592 _bstats_update(&bstats->bstats, bytes, packets);
593 u64_stats_update_end(&bstats->syncp);
596 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
597 const struct sk_buff *skb)
599 u64_stats_update_begin(&bstats->syncp);
600 bstats_update(&bstats->bstats, skb);
601 u64_stats_update_end(&bstats->syncp);
604 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
605 const struct sk_buff *skb)
607 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
610 static inline void qdisc_bstats_update(struct Qdisc *sch,
611 const struct sk_buff *skb)
613 bstats_update(&sch->bstats, skb);
616 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
617 const struct sk_buff *skb)
619 sch->qstats.backlog -= qdisc_pkt_len(skb);
622 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
623 const struct sk_buff *skb)
625 sch->qstats.backlog += qdisc_pkt_len(skb);
628 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
630 sch->qstats.drops += count;
633 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
638 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
640 qstats->overlimits++;
643 static inline void qdisc_qstats_drop(struct Qdisc *sch)
645 qstats_drop_inc(&sch->qstats);
648 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
650 this_cpu_inc(sch->cpu_qstats->drops);
653 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
655 sch->qstats.overlimits++;
658 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
665 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
666 struct qdisc_skb_head *qh)
668 struct sk_buff *last = qh->tail;
679 qdisc_qstats_backlog_inc(sch, skb);
681 return NET_XMIT_SUCCESS;
684 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
686 return __qdisc_enqueue_tail(skb, sch, &sch->q);
689 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
691 struct sk_buff *skb = qh->head;
693 if (likely(skb != NULL)) {
694 qh->head = skb->next;
696 if (qh->head == NULL)
704 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
706 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
708 if (likely(skb != NULL)) {
709 qdisc_qstats_backlog_dec(sch, skb);
710 qdisc_bstats_update(sch, skb);
716 /* Instead of calling kfree_skb() while root qdisc lock is held,
717 * queue the skb for future freeing at end of __dev_xmit_skb()
719 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
721 skb->next = *to_free;
725 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
726 struct qdisc_skb_head *qh,
727 struct sk_buff **to_free)
729 struct sk_buff *skb = __qdisc_dequeue_head(qh);
731 if (likely(skb != NULL)) {
732 unsigned int len = qdisc_pkt_len(skb);
734 qdisc_qstats_backlog_dec(sch, skb);
735 __qdisc_drop(skb, to_free);
742 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
743 struct sk_buff **to_free)
745 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
748 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
750 const struct qdisc_skb_head *qh = &sch->q;
755 /* generic pseudo peek method for non-work-conserving qdisc */
756 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
758 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
760 sch->gso_skb = sch->dequeue(sch);
762 /* it's still part of the queue */
763 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
771 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
772 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
774 struct sk_buff *skb = sch->gso_skb;
778 qdisc_qstats_backlog_dec(sch, skb);
781 skb = sch->dequeue(sch);
787 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
790 * We do not know the backlog in bytes of this list, it
791 * is up to the caller to correct it
795 rtnl_kfree_skbs(qh->head, qh->tail);
803 static inline void qdisc_reset_queue(struct Qdisc *sch)
805 __qdisc_reset_queue(&sch->q);
806 sch->qstats.backlog = 0;
809 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
818 unsigned int qlen = old->q.qlen;
819 unsigned int backlog = old->qstats.backlog;
822 qdisc_tree_reduce_backlog(old, qlen, backlog);
824 sch_tree_unlock(sch);
829 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
831 rtnl_kfree_skbs(skb, skb);
832 qdisc_qstats_drop(sch);
836 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
837 struct sk_buff **to_free)
839 __qdisc_drop(skb, to_free);
840 qdisc_qstats_drop(sch);
842 return NET_XMIT_DROP;
845 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
846 long it will take to send a packet given its size.
848 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
850 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
853 slot >>= rtab->rate.cell_log;
855 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
856 return rtab->data[slot];
859 struct psched_ratecfg {
860 u64 rate_bytes_ps; /* bytes per second */
867 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
872 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
873 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
875 return ((u64)len * r->mult) >> r->shift;
878 void psched_ratecfg_precompute(struct psched_ratecfg *r,
879 const struct tc_ratespec *conf,
882 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
883 const struct psched_ratecfg *r)
885 memset(res, 0, sizeof(*res));
887 /* legacy struct tc_ratespec has a 32bit @rate field
888 * Qdisc using 64bit rate should add new attributes
889 * in order to maintain compatibility.
891 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
893 res->overhead = r->overhead;
894 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);