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;
80 struct netdev_queue *dev_queue;
82 struct net_rate_estimator __rcu *rate_est;
83 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
84 struct gnet_stats_queue __percpu *cpu_qstats;
87 * For performance sake on SMP, we put highly modified fields at the end
89 struct sk_buff *gso_skb ____cacheline_aligned_in_smp;
90 struct qdisc_skb_head q;
91 struct gnet_stats_basic_packed bstats;
93 struct gnet_stats_queue qstats;
95 struct Qdisc *next_sched;
96 struct sk_buff *skb_bad_txq;
97 struct rcu_head rcu_head;
101 spinlock_t busylock ____cacheline_aligned_in_smp;
104 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
106 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
109 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
111 if (qdisc_is_running(qdisc))
113 /* Variant of write_seqcount_begin() telling lockdep a trylock
116 raw_write_seqcount_begin(&qdisc->running);
117 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
121 static inline void qdisc_run_end(struct Qdisc *qdisc)
123 write_seqcount_end(&qdisc->running);
126 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
128 return qdisc->flags & TCQ_F_ONETXQUEUE;
131 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
134 /* Non-BQL migrated drivers will return 0, too. */
135 return dql_avail(&txq->dql);
141 struct Qdisc_class_ops {
142 /* Child qdisc manipulation */
143 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
144 int (*graft)(struct Qdisc *, unsigned long cl,
145 struct Qdisc *, struct Qdisc **);
146 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
147 void (*qlen_notify)(struct Qdisc *, unsigned long);
149 /* Class manipulation routines */
150 unsigned long (*get)(struct Qdisc *, u32 classid);
151 void (*put)(struct Qdisc *, unsigned long);
152 int (*change)(struct Qdisc *, u32, u32,
153 struct nlattr **, unsigned long *);
154 int (*delete)(struct Qdisc *, unsigned long);
155 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
157 /* Filter manipulation */
158 struct tcf_block * (*tcf_block)(struct Qdisc *, unsigned long);
159 bool (*tcf_cl_offload)(u32 classid);
160 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
162 void (*unbind_tcf)(struct Qdisc *, unsigned long);
164 /* rtnetlink specific */
165 int (*dump)(struct Qdisc *, unsigned long,
166 struct sk_buff *skb, struct tcmsg*);
167 int (*dump_stats)(struct Qdisc *, unsigned long,
172 struct Qdisc_ops *next;
173 const struct Qdisc_class_ops *cl_ops;
177 int (*enqueue)(struct sk_buff *skb,
179 struct sk_buff **to_free);
180 struct sk_buff * (*dequeue)(struct Qdisc *);
181 struct sk_buff * (*peek)(struct Qdisc *);
183 int (*init)(struct Qdisc *, struct nlattr *arg);
184 void (*reset)(struct Qdisc *);
185 void (*destroy)(struct Qdisc *);
186 int (*change)(struct Qdisc *, struct nlattr *arg);
187 void (*attach)(struct Qdisc *);
189 int (*dump)(struct Qdisc *, struct sk_buff *);
190 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
192 struct module *owner;
202 const struct tcf_proto *goto_tp;
206 struct tcf_proto_ops {
207 struct list_head head;
210 int (*classify)(struct sk_buff *,
211 const struct tcf_proto *,
212 struct tcf_result *);
213 int (*init)(struct tcf_proto*);
214 void (*destroy)(struct tcf_proto*);
216 unsigned long (*get)(struct tcf_proto*, u32 handle);
217 int (*change)(struct net *net, struct sk_buff *,
218 struct tcf_proto*, unsigned long,
219 u32 handle, struct nlattr **,
220 unsigned long *, bool);
221 int (*delete)(struct tcf_proto*, unsigned long, bool*);
222 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
224 /* rtnetlink specific */
225 int (*dump)(struct net*, struct tcf_proto*, unsigned long,
226 struct sk_buff *skb, struct tcmsg*);
228 struct module *owner;
232 /* Fast access part */
233 struct tcf_proto __rcu *next;
235 int (*classify)(struct sk_buff *,
236 const struct tcf_proto *,
237 struct tcf_result *);
245 const struct tcf_proto_ops *ops;
246 struct tcf_chain *chain;
250 struct qdisc_skb_cb {
251 unsigned int pkt_len;
252 u16 slave_dev_queue_mapping;
254 #define QDISC_CB_PRIV_LEN 20
255 unsigned char data[QDISC_CB_PRIV_LEN];
259 struct tcf_proto __rcu *filter_chain;
260 struct tcf_proto __rcu **p_filter_chain;
261 struct list_head list;
262 struct tcf_block *block;
263 u32 index; /* chain index */
268 struct list_head chain_list;
271 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
273 struct qdisc_skb_cb *qcb;
275 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
276 BUILD_BUG_ON(sizeof(qcb->data) < sz);
279 static inline int qdisc_qlen(const struct Qdisc *q)
284 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
286 return (struct qdisc_skb_cb *)skb->cb;
289 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
291 return &qdisc->q.lock;
294 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
296 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
301 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
303 return qdisc->dev_queue->qdisc_sleeping;
306 /* The qdisc root lock is a mechanism by which to top level
307 * of a qdisc tree can be locked from any qdisc node in the
308 * forest. This allows changing the configuration of some
309 * aspect of the qdisc tree while blocking out asynchronous
310 * qdisc access in the packet processing paths.
312 * It is only legal to do this when the root will not change
313 * on us. Otherwise we'll potentially lock the wrong qdisc
314 * root. This is enforced by holding the RTNL semaphore, which
315 * all users of this lock accessor must do.
317 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
319 struct Qdisc *root = qdisc_root(qdisc);
322 return qdisc_lock(root);
325 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
327 struct Qdisc *root = qdisc_root_sleeping(qdisc);
330 return qdisc_lock(root);
333 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
335 struct Qdisc *root = qdisc_root_sleeping(qdisc);
338 return &root->running;
341 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
343 return qdisc->dev_queue->dev;
346 static inline void sch_tree_lock(const struct Qdisc *q)
348 spin_lock_bh(qdisc_root_sleeping_lock(q));
351 static inline void sch_tree_unlock(const struct Qdisc *q)
353 spin_unlock_bh(qdisc_root_sleeping_lock(q));
356 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
357 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
359 extern struct Qdisc noop_qdisc;
360 extern struct Qdisc_ops noop_qdisc_ops;
361 extern struct Qdisc_ops pfifo_fast_ops;
362 extern struct Qdisc_ops mq_qdisc_ops;
363 extern struct Qdisc_ops noqueue_qdisc_ops;
364 extern const struct Qdisc_ops *default_qdisc_ops;
365 static inline const struct Qdisc_ops *
366 get_default_qdisc_ops(const struct net_device *dev, int ntx)
368 return ntx < dev->real_num_tx_queues ?
369 default_qdisc_ops : &pfifo_fast_ops;
372 struct Qdisc_class_common {
374 struct hlist_node hnode;
377 struct Qdisc_class_hash {
378 struct hlist_head *hash;
379 unsigned int hashsize;
380 unsigned int hashmask;
381 unsigned int hashelems;
384 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
391 static inline struct Qdisc_class_common *
392 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
394 struct Qdisc_class_common *cl;
397 h = qdisc_class_hash(id, hash->hashmask);
398 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
399 if (cl->classid == id)
405 int qdisc_class_hash_init(struct Qdisc_class_hash *);
406 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
407 struct Qdisc_class_common *);
408 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
409 struct Qdisc_class_common *);
410 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
411 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
413 void dev_init_scheduler(struct net_device *dev);
414 void dev_shutdown(struct net_device *dev);
415 void dev_activate(struct net_device *dev);
416 void dev_deactivate(struct net_device *dev);
417 void dev_deactivate_many(struct list_head *head);
418 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
419 struct Qdisc *qdisc);
420 void qdisc_reset(struct Qdisc *qdisc);
421 void qdisc_destroy(struct Qdisc *qdisc);
422 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
424 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
425 const struct Qdisc_ops *ops);
426 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
427 const struct Qdisc_ops *ops, u32 parentid);
428 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
429 const struct qdisc_size_table *stab);
430 int skb_do_redirect(struct sk_buff *);
432 static inline void skb_reset_tc(struct sk_buff *skb)
434 #ifdef CONFIG_NET_CLS_ACT
435 skb->tc_redirected = 0;
439 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
441 #ifdef CONFIG_NET_CLS_ACT
442 return skb->tc_at_ingress;
448 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
450 #ifdef CONFIG_NET_CLS_ACT
451 if (skb->tc_skip_classify) {
452 skb->tc_skip_classify = 0;
459 /* Reset all TX qdiscs greater then index of a device. */
460 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
464 for (; i < dev->num_tx_queues; i++) {
465 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
467 spin_lock_bh(qdisc_lock(qdisc));
469 spin_unlock_bh(qdisc_lock(qdisc));
474 static inline void qdisc_reset_all_tx(struct net_device *dev)
476 qdisc_reset_all_tx_gt(dev, 0);
479 /* Are all TX queues of the device empty? */
480 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
485 for (i = 0; i < dev->num_tx_queues; i++) {
486 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
487 const struct Qdisc *q = rcu_dereference(txq->qdisc);
498 /* Are any of the TX qdiscs changing? */
499 static inline bool qdisc_tx_changing(const struct net_device *dev)
503 for (i = 0; i < dev->num_tx_queues; i++) {
504 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
505 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
511 /* Is the device using the noop qdisc on all queues? */
512 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
516 for (i = 0; i < dev->num_tx_queues; i++) {
517 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
518 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
524 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
526 return qdisc_skb_cb(skb)->pkt_len;
529 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
530 enum net_xmit_qdisc_t {
531 __NET_XMIT_STOLEN = 0x00010000,
532 __NET_XMIT_BYPASS = 0x00020000,
535 #ifdef CONFIG_NET_CLS_ACT
536 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
538 #define net_xmit_drop_count(e) (1)
541 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
542 const struct Qdisc *sch)
544 #ifdef CONFIG_NET_SCHED
545 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
548 __qdisc_calculate_pkt_len(skb, stab);
552 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
553 struct sk_buff **to_free)
555 qdisc_calculate_pkt_len(skb, sch);
556 return sch->enqueue(skb, sch, to_free);
559 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
561 return q->flags & TCQ_F_CPUSTATS;
564 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
565 __u64 bytes, __u32 packets)
567 bstats->bytes += bytes;
568 bstats->packets += packets;
571 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
572 const struct sk_buff *skb)
574 _bstats_update(bstats,
576 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
579 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
580 __u64 bytes, __u32 packets)
582 u64_stats_update_begin(&bstats->syncp);
583 _bstats_update(&bstats->bstats, bytes, packets);
584 u64_stats_update_end(&bstats->syncp);
587 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
588 const struct sk_buff *skb)
590 u64_stats_update_begin(&bstats->syncp);
591 bstats_update(&bstats->bstats, skb);
592 u64_stats_update_end(&bstats->syncp);
595 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
596 const struct sk_buff *skb)
598 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
601 static inline void qdisc_bstats_update(struct Qdisc *sch,
602 const struct sk_buff *skb)
604 bstats_update(&sch->bstats, skb);
607 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
608 const struct sk_buff *skb)
610 sch->qstats.backlog -= qdisc_pkt_len(skb);
613 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
614 const struct sk_buff *skb)
616 sch->qstats.backlog += qdisc_pkt_len(skb);
619 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
621 sch->qstats.drops += count;
624 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
629 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
631 qstats->overlimits++;
634 static inline void qdisc_qstats_drop(struct Qdisc *sch)
636 qstats_drop_inc(&sch->qstats);
639 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
641 this_cpu_inc(sch->cpu_qstats->drops);
644 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
646 sch->qstats.overlimits++;
649 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
656 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
657 struct qdisc_skb_head *qh)
659 struct sk_buff *last = qh->tail;
670 qdisc_qstats_backlog_inc(sch, skb);
672 return NET_XMIT_SUCCESS;
675 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
677 return __qdisc_enqueue_tail(skb, sch, &sch->q);
680 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
682 struct sk_buff *skb = qh->head;
684 if (likely(skb != NULL)) {
685 qh->head = skb->next;
687 if (qh->head == NULL)
695 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
697 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
699 if (likely(skb != NULL)) {
700 qdisc_qstats_backlog_dec(sch, skb);
701 qdisc_bstats_update(sch, skb);
707 /* Instead of calling kfree_skb() while root qdisc lock is held,
708 * queue the skb for future freeing at end of __dev_xmit_skb()
710 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
712 skb->next = *to_free;
716 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
717 struct qdisc_skb_head *qh,
718 struct sk_buff **to_free)
720 struct sk_buff *skb = __qdisc_dequeue_head(qh);
722 if (likely(skb != NULL)) {
723 unsigned int len = qdisc_pkt_len(skb);
725 qdisc_qstats_backlog_dec(sch, skb);
726 __qdisc_drop(skb, to_free);
733 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
734 struct sk_buff **to_free)
736 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
739 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
741 const struct qdisc_skb_head *qh = &sch->q;
746 /* generic pseudo peek method for non-work-conserving qdisc */
747 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
749 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
751 sch->gso_skb = sch->dequeue(sch);
753 /* it's still part of the queue */
754 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
762 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
763 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
765 struct sk_buff *skb = sch->gso_skb;
769 qdisc_qstats_backlog_dec(sch, skb);
772 skb = sch->dequeue(sch);
778 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
781 * We do not know the backlog in bytes of this list, it
782 * is up to the caller to correct it
786 rtnl_kfree_skbs(qh->head, qh->tail);
794 static inline void qdisc_reset_queue(struct Qdisc *sch)
796 __qdisc_reset_queue(&sch->q);
797 sch->qstats.backlog = 0;
800 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
809 qdisc_tree_reduce_backlog(old, old->q.qlen, old->qstats.backlog);
812 sch_tree_unlock(sch);
817 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
819 rtnl_kfree_skbs(skb, skb);
820 qdisc_qstats_drop(sch);
824 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
825 struct sk_buff **to_free)
827 __qdisc_drop(skb, to_free);
828 qdisc_qstats_drop(sch);
830 return NET_XMIT_DROP;
833 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
834 long it will take to send a packet given its size.
836 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
838 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
841 slot >>= rtab->rate.cell_log;
843 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
844 return rtab->data[slot];
847 struct psched_ratecfg {
848 u64 rate_bytes_ps; /* bytes per second */
855 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
860 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
861 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
863 return ((u64)len * r->mult) >> r->shift;
866 void psched_ratecfg_precompute(struct psched_ratecfg *r,
867 const struct tc_ratespec *conf,
870 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
871 const struct psched_ratecfg *r)
873 memset(res, 0, sizeof(*res));
875 /* legacy struct tc_ratespec has a 32bit @rate field
876 * Qdisc using 64bit rate should add new attributes
877 * in order to maintain compatibility.
879 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
881 res->overhead = r->overhead;
882 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);