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 void qdisc_refcount_inc(struct Qdisc *qdisc)
106 if (qdisc->flags & TCQ_F_BUILTIN)
108 refcount_inc(&qdisc->refcnt);
111 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
113 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
116 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
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_);
128 static inline void qdisc_run_end(struct Qdisc *qdisc)
130 write_seqcount_end(&qdisc->running);
133 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
135 return qdisc->flags & TCQ_F_ONETXQUEUE;
138 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
141 /* Non-BQL migrated drivers will return 0, too. */
142 return dql_avail(&txq->dql);
148 struct Qdisc_class_ops {
149 /* Child qdisc manipulation */
150 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
151 int (*graft)(struct Qdisc *, unsigned long cl,
152 struct Qdisc *, struct Qdisc **);
153 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
154 void (*qlen_notify)(struct Qdisc *, unsigned long);
156 /* Class manipulation routines */
157 unsigned long (*get)(struct Qdisc *, u32 classid);
158 void (*put)(struct Qdisc *, unsigned long);
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 bool (*tcf_cl_offload)(u32 classid);
167 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
169 void (*unbind_tcf)(struct Qdisc *, unsigned long);
171 /* rtnetlink specific */
172 int (*dump)(struct Qdisc *, unsigned long,
173 struct sk_buff *skb, struct tcmsg*);
174 int (*dump_stats)(struct Qdisc *, unsigned long,
179 struct Qdisc_ops *next;
180 const struct Qdisc_class_ops *cl_ops;
184 int (*enqueue)(struct sk_buff *skb,
186 struct sk_buff **to_free);
187 struct sk_buff * (*dequeue)(struct Qdisc *);
188 struct sk_buff * (*peek)(struct Qdisc *);
190 int (*init)(struct Qdisc *, struct nlattr *arg);
191 void (*reset)(struct Qdisc *);
192 void (*destroy)(struct Qdisc *);
193 int (*change)(struct Qdisc *, struct nlattr *arg);
194 void (*attach)(struct Qdisc *);
196 int (*dump)(struct Qdisc *, struct sk_buff *);
197 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
199 struct module *owner;
209 const struct tcf_proto *goto_tp;
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 void (*destroy)(struct tcf_proto*);
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, bool*);
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;
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;
278 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
280 struct qdisc_skb_cb *qcb;
282 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
283 BUILD_BUG_ON(sizeof(qcb->data) < sz);
286 static inline int qdisc_qlen(const struct Qdisc *q)
291 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
293 return (struct qdisc_skb_cb *)skb->cb;
296 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
298 return &qdisc->q.lock;
301 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
303 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
308 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
310 return qdisc->dev_queue->qdisc_sleeping;
313 /* The qdisc root lock is a mechanism by which to top level
314 * of a qdisc tree can be locked from any qdisc node in the
315 * forest. This allows changing the configuration of some
316 * aspect of the qdisc tree while blocking out asynchronous
317 * qdisc access in the packet processing paths.
319 * It is only legal to do this when the root will not change
320 * on us. Otherwise we'll potentially lock the wrong qdisc
321 * root. This is enforced by holding the RTNL semaphore, which
322 * all users of this lock accessor must do.
324 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
326 struct Qdisc *root = qdisc_root(qdisc);
329 return qdisc_lock(root);
332 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
334 struct Qdisc *root = qdisc_root_sleeping(qdisc);
337 return qdisc_lock(root);
340 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
342 struct Qdisc *root = qdisc_root_sleeping(qdisc);
345 return &root->running;
348 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
350 return qdisc->dev_queue->dev;
353 static inline void sch_tree_lock(const struct Qdisc *q)
355 spin_lock_bh(qdisc_root_sleeping_lock(q));
358 static inline void sch_tree_unlock(const struct Qdisc *q)
360 spin_unlock_bh(qdisc_root_sleeping_lock(q));
363 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
364 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
366 extern struct Qdisc noop_qdisc;
367 extern struct Qdisc_ops noop_qdisc_ops;
368 extern struct Qdisc_ops pfifo_fast_ops;
369 extern struct Qdisc_ops mq_qdisc_ops;
370 extern struct Qdisc_ops noqueue_qdisc_ops;
371 extern const struct Qdisc_ops *default_qdisc_ops;
372 static inline const struct Qdisc_ops *
373 get_default_qdisc_ops(const struct net_device *dev, int ntx)
375 return ntx < dev->real_num_tx_queues ?
376 default_qdisc_ops : &pfifo_fast_ops;
379 struct Qdisc_class_common {
381 struct hlist_node hnode;
384 struct Qdisc_class_hash {
385 struct hlist_head *hash;
386 unsigned int hashsize;
387 unsigned int hashmask;
388 unsigned int hashelems;
391 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
398 static inline struct Qdisc_class_common *
399 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
401 struct Qdisc_class_common *cl;
404 h = qdisc_class_hash(id, hash->hashmask);
405 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
406 if (cl->classid == id)
412 int qdisc_class_hash_init(struct Qdisc_class_hash *);
413 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
414 struct Qdisc_class_common *);
415 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
416 struct Qdisc_class_common *);
417 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
418 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
420 void dev_init_scheduler(struct net_device *dev);
421 void dev_shutdown(struct net_device *dev);
422 void dev_activate(struct net_device *dev);
423 void dev_deactivate(struct net_device *dev);
424 void dev_deactivate_many(struct list_head *head);
425 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
426 struct Qdisc *qdisc);
427 void qdisc_reset(struct Qdisc *qdisc);
428 void qdisc_destroy(struct Qdisc *qdisc);
429 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
431 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
432 const struct Qdisc_ops *ops);
433 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
434 const struct Qdisc_ops *ops, u32 parentid);
435 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
436 const struct qdisc_size_table *stab);
437 int skb_do_redirect(struct sk_buff *);
439 static inline void skb_reset_tc(struct sk_buff *skb)
441 #ifdef CONFIG_NET_CLS_ACT
442 skb->tc_redirected = 0;
446 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
448 #ifdef CONFIG_NET_CLS_ACT
449 return skb->tc_at_ingress;
455 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
457 #ifdef CONFIG_NET_CLS_ACT
458 if (skb->tc_skip_classify) {
459 skb->tc_skip_classify = 0;
466 /* Reset all TX qdiscs greater then index of a device. */
467 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
471 for (; i < dev->num_tx_queues; i++) {
472 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
474 spin_lock_bh(qdisc_lock(qdisc));
476 spin_unlock_bh(qdisc_lock(qdisc));
481 static inline void qdisc_reset_all_tx(struct net_device *dev)
483 qdisc_reset_all_tx_gt(dev, 0);
486 /* Are all TX queues of the device empty? */
487 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
492 for (i = 0; i < dev->num_tx_queues; i++) {
493 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
494 const struct Qdisc *q = rcu_dereference(txq->qdisc);
505 /* Are any of the TX qdiscs changing? */
506 static inline bool qdisc_tx_changing(const struct net_device *dev)
510 for (i = 0; i < dev->num_tx_queues; i++) {
511 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
512 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
518 /* Is the device using the noop qdisc on all queues? */
519 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
523 for (i = 0; i < dev->num_tx_queues; i++) {
524 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
525 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
531 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
533 return qdisc_skb_cb(skb)->pkt_len;
536 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
537 enum net_xmit_qdisc_t {
538 __NET_XMIT_STOLEN = 0x00010000,
539 __NET_XMIT_BYPASS = 0x00020000,
542 #ifdef CONFIG_NET_CLS_ACT
543 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
545 #define net_xmit_drop_count(e) (1)
548 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
549 const struct Qdisc *sch)
551 #ifdef CONFIG_NET_SCHED
552 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
555 __qdisc_calculate_pkt_len(skb, stab);
559 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
560 struct sk_buff **to_free)
562 qdisc_calculate_pkt_len(skb, sch);
563 return sch->enqueue(skb, sch, to_free);
566 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
568 return q->flags & TCQ_F_CPUSTATS;
571 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
572 __u64 bytes, __u32 packets)
574 bstats->bytes += bytes;
575 bstats->packets += packets;
578 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
579 const struct sk_buff *skb)
581 _bstats_update(bstats,
583 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
586 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
587 __u64 bytes, __u32 packets)
589 u64_stats_update_begin(&bstats->syncp);
590 _bstats_update(&bstats->bstats, bytes, packets);
591 u64_stats_update_end(&bstats->syncp);
594 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
595 const struct sk_buff *skb)
597 u64_stats_update_begin(&bstats->syncp);
598 bstats_update(&bstats->bstats, skb);
599 u64_stats_update_end(&bstats->syncp);
602 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
603 const struct sk_buff *skb)
605 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
608 static inline void qdisc_bstats_update(struct Qdisc *sch,
609 const struct sk_buff *skb)
611 bstats_update(&sch->bstats, skb);
614 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
615 const struct sk_buff *skb)
617 sch->qstats.backlog -= qdisc_pkt_len(skb);
620 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
621 const struct sk_buff *skb)
623 sch->qstats.backlog += qdisc_pkt_len(skb);
626 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
628 sch->qstats.drops += count;
631 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
636 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
638 qstats->overlimits++;
641 static inline void qdisc_qstats_drop(struct Qdisc *sch)
643 qstats_drop_inc(&sch->qstats);
646 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
648 this_cpu_inc(sch->cpu_qstats->drops);
651 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
653 sch->qstats.overlimits++;
656 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
663 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
664 struct qdisc_skb_head *qh)
666 struct sk_buff *last = qh->tail;
677 qdisc_qstats_backlog_inc(sch, skb);
679 return NET_XMIT_SUCCESS;
682 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
684 return __qdisc_enqueue_tail(skb, sch, &sch->q);
687 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
689 struct sk_buff *skb = qh->head;
691 if (likely(skb != NULL)) {
692 qh->head = skb->next;
694 if (qh->head == NULL)
702 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
704 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
706 if (likely(skb != NULL)) {
707 qdisc_qstats_backlog_dec(sch, skb);
708 qdisc_bstats_update(sch, skb);
714 /* Instead of calling kfree_skb() while root qdisc lock is held,
715 * queue the skb for future freeing at end of __dev_xmit_skb()
717 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
719 skb->next = *to_free;
723 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
724 struct qdisc_skb_head *qh,
725 struct sk_buff **to_free)
727 struct sk_buff *skb = __qdisc_dequeue_head(qh);
729 if (likely(skb != NULL)) {
730 unsigned int len = qdisc_pkt_len(skb);
732 qdisc_qstats_backlog_dec(sch, skb);
733 __qdisc_drop(skb, to_free);
740 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
741 struct sk_buff **to_free)
743 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
746 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
748 const struct qdisc_skb_head *qh = &sch->q;
753 /* generic pseudo peek method for non-work-conserving qdisc */
754 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
756 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
758 sch->gso_skb = sch->dequeue(sch);
760 /* it's still part of the queue */
761 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
769 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
770 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
772 struct sk_buff *skb = sch->gso_skb;
776 qdisc_qstats_backlog_dec(sch, skb);
779 skb = sch->dequeue(sch);
785 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
788 * We do not know the backlog in bytes of this list, it
789 * is up to the caller to correct it
793 rtnl_kfree_skbs(qh->head, qh->tail);
801 static inline void qdisc_reset_queue(struct Qdisc *sch)
803 __qdisc_reset_queue(&sch->q);
804 sch->qstats.backlog = 0;
807 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
816 unsigned int qlen = old->q.qlen;
817 unsigned int backlog = old->qstats.backlog;
820 qdisc_tree_reduce_backlog(old, qlen, backlog);
822 sch_tree_unlock(sch);
827 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
829 rtnl_kfree_skbs(skb, skb);
830 qdisc_qstats_drop(sch);
834 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
835 struct sk_buff **to_free)
837 __qdisc_drop(skb, to_free);
838 qdisc_qstats_drop(sch);
840 return NET_XMIT_DROP;
843 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
844 long it will take to send a packet given its size.
846 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
848 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
851 slot >>= rtab->rate.cell_log;
853 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
854 return rtab->data[slot];
857 struct psched_ratecfg {
858 u64 rate_bytes_ps; /* bytes per second */
865 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
870 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
871 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
873 return ((u64)len * r->mult) >> r->shift;
876 void psched_ratecfg_precompute(struct psched_ratecfg *r,
877 const struct tc_ratespec *conf,
880 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
881 const struct psched_ratecfg *r)
883 memset(res, 0, sizeof(*res));
885 /* legacy struct tc_ratespec has a 32bit @rate field
886 * Qdisc using 64bit rate should add new attributes
887 * in order to maintain compatibility.
889 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
891 res->overhead = r->overhead;
892 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);