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;
96 struct rcu_head rcu_head;
100 spinlock_t busylock ____cacheline_aligned_in_smp;
103 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
105 if (qdisc->flags & TCQ_F_BUILTIN)
107 refcount_inc(&qdisc->refcnt);
110 static inline bool qdisc_is_running(const struct Qdisc *qdisc)
112 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
115 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
117 if (qdisc_is_running(qdisc))
119 /* Variant of write_seqcount_begin() telling lockdep a trylock
122 raw_write_seqcount_begin(&qdisc->running);
123 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
127 static inline void qdisc_run_end(struct Qdisc *qdisc)
129 write_seqcount_end(&qdisc->running);
132 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
134 return qdisc->flags & TCQ_F_ONETXQUEUE;
137 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
140 /* Non-BQL migrated drivers will return 0, too. */
141 return dql_avail(&txq->dql);
147 struct Qdisc_class_ops {
148 /* Child qdisc manipulation */
149 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
150 int (*graft)(struct Qdisc *, unsigned long cl,
151 struct Qdisc *, struct Qdisc **);
152 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
153 void (*qlen_notify)(struct Qdisc *, unsigned long);
155 /* Class manipulation routines */
156 unsigned long (*find)(struct Qdisc *, u32 classid);
157 int (*change)(struct Qdisc *, u32, u32,
158 struct nlattr **, unsigned long *);
159 int (*delete)(struct Qdisc *, unsigned long);
160 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
162 /* Filter manipulation */
163 struct tcf_block * (*tcf_block)(struct Qdisc *, unsigned long);
164 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
166 void (*unbind_tcf)(struct Qdisc *, unsigned long);
168 /* rtnetlink specific */
169 int (*dump)(struct Qdisc *, unsigned long,
170 struct sk_buff *skb, struct tcmsg*);
171 int (*dump_stats)(struct Qdisc *, unsigned long,
176 struct Qdisc_ops *next;
177 const struct Qdisc_class_ops *cl_ops;
181 int (*enqueue)(struct sk_buff *skb,
183 struct sk_buff **to_free);
184 struct sk_buff * (*dequeue)(struct Qdisc *);
185 struct sk_buff * (*peek)(struct Qdisc *);
187 int (*init)(struct Qdisc *, struct nlattr *arg);
188 void (*reset)(struct Qdisc *);
189 void (*destroy)(struct Qdisc *);
190 int (*change)(struct Qdisc *, struct nlattr *arg);
191 void (*attach)(struct Qdisc *);
193 int (*dump)(struct Qdisc *, struct sk_buff *);
194 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
196 struct module *owner;
206 const struct tcf_proto *goto_tp;
210 struct tcf_proto_ops {
211 struct list_head head;
214 int (*classify)(struct sk_buff *,
215 const struct tcf_proto *,
216 struct tcf_result *);
217 int (*init)(struct tcf_proto*);
218 void (*destroy)(struct tcf_proto*);
220 void* (*get)(struct tcf_proto*, u32 handle);
221 int (*change)(struct net *net, struct sk_buff *,
222 struct tcf_proto*, unsigned long,
223 u32 handle, struct nlattr **,
225 int (*delete)(struct tcf_proto*, void *, bool*);
226 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
227 void (*bind_class)(void *, u32, unsigned long);
229 /* rtnetlink specific */
230 int (*dump)(struct net*, struct tcf_proto*, void *,
231 struct sk_buff *skb, struct tcmsg*);
233 struct module *owner;
237 /* Fast access part */
238 struct tcf_proto __rcu *next;
240 int (*classify)(struct sk_buff *,
241 const struct tcf_proto *,
242 struct tcf_result *);
250 const struct tcf_proto_ops *ops;
251 struct tcf_chain *chain;
255 struct qdisc_skb_cb {
256 unsigned int pkt_len;
257 u16 slave_dev_queue_mapping;
259 #define QDISC_CB_PRIV_LEN 20
260 unsigned char data[QDISC_CB_PRIV_LEN];
264 struct tcf_proto __rcu *filter_chain;
265 struct tcf_proto __rcu **p_filter_chain;
266 struct list_head list;
267 struct tcf_block *block;
268 u32 index; /* chain index */
273 struct list_head chain_list;
276 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
278 struct qdisc_skb_cb *qcb;
280 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
281 BUILD_BUG_ON(sizeof(qcb->data) < sz);
284 static inline int qdisc_qlen(const struct Qdisc *q)
289 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
291 return (struct qdisc_skb_cb *)skb->cb;
294 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
296 return &qdisc->q.lock;
299 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
301 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
306 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
308 return qdisc->dev_queue->qdisc_sleeping;
311 /* The qdisc root lock is a mechanism by which to top level
312 * of a qdisc tree can be locked from any qdisc node in the
313 * forest. This allows changing the configuration of some
314 * aspect of the qdisc tree while blocking out asynchronous
315 * qdisc access in the packet processing paths.
317 * It is only legal to do this when the root will not change
318 * on us. Otherwise we'll potentially lock the wrong qdisc
319 * root. This is enforced by holding the RTNL semaphore, which
320 * all users of this lock accessor must do.
322 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
324 struct Qdisc *root = qdisc_root(qdisc);
327 return qdisc_lock(root);
330 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
332 struct Qdisc *root = qdisc_root_sleeping(qdisc);
335 return qdisc_lock(root);
338 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
340 struct Qdisc *root = qdisc_root_sleeping(qdisc);
343 return &root->running;
346 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
348 return qdisc->dev_queue->dev;
351 static inline void sch_tree_lock(const struct Qdisc *q)
353 spin_lock_bh(qdisc_root_sleeping_lock(q));
356 static inline void sch_tree_unlock(const struct Qdisc *q)
358 spin_unlock_bh(qdisc_root_sleeping_lock(q));
361 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
362 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
364 extern struct Qdisc noop_qdisc;
365 extern struct Qdisc_ops noop_qdisc_ops;
366 extern struct Qdisc_ops pfifo_fast_ops;
367 extern struct Qdisc_ops mq_qdisc_ops;
368 extern struct Qdisc_ops noqueue_qdisc_ops;
369 extern const struct Qdisc_ops *default_qdisc_ops;
370 static inline const struct Qdisc_ops *
371 get_default_qdisc_ops(const struct net_device *dev, int ntx)
373 return ntx < dev->real_num_tx_queues ?
374 default_qdisc_ops : &pfifo_fast_ops;
377 struct Qdisc_class_common {
379 struct hlist_node hnode;
382 struct Qdisc_class_hash {
383 struct hlist_head *hash;
384 unsigned int hashsize;
385 unsigned int hashmask;
386 unsigned int hashelems;
389 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
396 static inline struct Qdisc_class_common *
397 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
399 struct Qdisc_class_common *cl;
405 h = qdisc_class_hash(id, hash->hashmask);
406 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
407 if (cl->classid == id)
413 int qdisc_class_hash_init(struct Qdisc_class_hash *);
414 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
415 struct Qdisc_class_common *);
416 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
417 struct Qdisc_class_common *);
418 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
419 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
421 void dev_init_scheduler(struct net_device *dev);
422 void dev_shutdown(struct net_device *dev);
423 void dev_activate(struct net_device *dev);
424 void dev_deactivate(struct net_device *dev);
425 void dev_deactivate_many(struct list_head *head);
426 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
427 struct Qdisc *qdisc);
428 void qdisc_reset(struct Qdisc *qdisc);
429 void qdisc_destroy(struct Qdisc *qdisc);
430 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
432 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
433 const struct Qdisc_ops *ops);
434 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
435 const struct Qdisc_ops *ops, u32 parentid);
436 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
437 const struct qdisc_size_table *stab);
438 int skb_do_redirect(struct sk_buff *);
440 static inline void skb_reset_tc(struct sk_buff *skb)
442 #ifdef CONFIG_NET_CLS_ACT
443 skb->tc_redirected = 0;
447 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
449 #ifdef CONFIG_NET_CLS_ACT
450 return skb->tc_at_ingress;
456 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
458 #ifdef CONFIG_NET_CLS_ACT
459 if (skb->tc_skip_classify) {
460 skb->tc_skip_classify = 0;
467 /* Reset all TX qdiscs greater then index of a device. */
468 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
472 for (; i < dev->num_tx_queues; i++) {
473 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
475 spin_lock_bh(qdisc_lock(qdisc));
477 spin_unlock_bh(qdisc_lock(qdisc));
482 static inline void qdisc_reset_all_tx(struct net_device *dev)
484 qdisc_reset_all_tx_gt(dev, 0);
487 /* Are all TX queues of the device empty? */
488 static inline bool qdisc_all_tx_empty(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 const struct Qdisc *q = rcu_dereference(txq->qdisc);
506 /* Are any of the TX qdiscs changing? */
507 static inline bool qdisc_tx_changing(const struct net_device *dev)
511 for (i = 0; i < dev->num_tx_queues; i++) {
512 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
513 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
519 /* Is the device using the noop qdisc on all queues? */
520 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
524 for (i = 0; i < dev->num_tx_queues; i++) {
525 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
526 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
532 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
534 return qdisc_skb_cb(skb)->pkt_len;
537 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
538 enum net_xmit_qdisc_t {
539 __NET_XMIT_STOLEN = 0x00010000,
540 __NET_XMIT_BYPASS = 0x00020000,
543 #ifdef CONFIG_NET_CLS_ACT
544 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
546 #define net_xmit_drop_count(e) (1)
549 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
550 const struct Qdisc *sch)
552 #ifdef CONFIG_NET_SCHED
553 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
556 __qdisc_calculate_pkt_len(skb, stab);
560 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
561 struct sk_buff **to_free)
563 qdisc_calculate_pkt_len(skb, sch);
564 return sch->enqueue(skb, sch, to_free);
567 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
569 return q->flags & TCQ_F_CPUSTATS;
572 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
573 __u64 bytes, __u32 packets)
575 bstats->bytes += bytes;
576 bstats->packets += packets;
579 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
580 const struct sk_buff *skb)
582 _bstats_update(bstats,
584 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
587 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
588 __u64 bytes, __u32 packets)
590 u64_stats_update_begin(&bstats->syncp);
591 _bstats_update(&bstats->bstats, bytes, packets);
592 u64_stats_update_end(&bstats->syncp);
595 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
596 const struct sk_buff *skb)
598 u64_stats_update_begin(&bstats->syncp);
599 bstats_update(&bstats->bstats, skb);
600 u64_stats_update_end(&bstats->syncp);
603 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
604 const struct sk_buff *skb)
606 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
609 static inline void qdisc_bstats_update(struct Qdisc *sch,
610 const struct sk_buff *skb)
612 bstats_update(&sch->bstats, skb);
615 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
616 const struct sk_buff *skb)
618 sch->qstats.backlog -= qdisc_pkt_len(skb);
621 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
622 const struct sk_buff *skb)
624 sch->qstats.backlog += qdisc_pkt_len(skb);
627 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
629 sch->qstats.drops += count;
632 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
637 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
639 qstats->overlimits++;
642 static inline void qdisc_qstats_drop(struct Qdisc *sch)
644 qstats_drop_inc(&sch->qstats);
647 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
649 this_cpu_inc(sch->cpu_qstats->drops);
652 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
654 sch->qstats.overlimits++;
657 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
664 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
665 struct qdisc_skb_head *qh)
667 struct sk_buff *last = qh->tail;
678 qdisc_qstats_backlog_inc(sch, skb);
680 return NET_XMIT_SUCCESS;
683 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
685 return __qdisc_enqueue_tail(skb, sch, &sch->q);
688 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
690 struct sk_buff *skb = qh->head;
692 if (likely(skb != NULL)) {
693 qh->head = skb->next;
695 if (qh->head == NULL)
703 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
705 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
707 if (likely(skb != NULL)) {
708 qdisc_qstats_backlog_dec(sch, skb);
709 qdisc_bstats_update(sch, skb);
715 /* Instead of calling kfree_skb() while root qdisc lock is held,
716 * queue the skb for future freeing at end of __dev_xmit_skb()
718 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
720 skb->next = *to_free;
724 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
725 struct qdisc_skb_head *qh,
726 struct sk_buff **to_free)
728 struct sk_buff *skb = __qdisc_dequeue_head(qh);
730 if (likely(skb != NULL)) {
731 unsigned int len = qdisc_pkt_len(skb);
733 qdisc_qstats_backlog_dec(sch, skb);
734 __qdisc_drop(skb, to_free);
741 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
742 struct sk_buff **to_free)
744 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
747 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
749 const struct qdisc_skb_head *qh = &sch->q;
754 /* generic pseudo peek method for non-work-conserving qdisc */
755 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
757 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
759 sch->gso_skb = sch->dequeue(sch);
761 /* it's still part of the queue */
762 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
770 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
771 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
773 struct sk_buff *skb = sch->gso_skb;
777 qdisc_qstats_backlog_dec(sch, skb);
780 skb = sch->dequeue(sch);
786 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
789 * We do not know the backlog in bytes of this list, it
790 * is up to the caller to correct it
794 rtnl_kfree_skbs(qh->head, qh->tail);
802 static inline void qdisc_reset_queue(struct Qdisc *sch)
804 __qdisc_reset_queue(&sch->q);
805 sch->qstats.backlog = 0;
808 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
817 unsigned int qlen = old->q.qlen;
818 unsigned int backlog = old->qstats.backlog;
821 qdisc_tree_reduce_backlog(old, qlen, backlog);
823 sch_tree_unlock(sch);
828 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
830 rtnl_kfree_skbs(skb, skb);
831 qdisc_qstats_drop(sch);
835 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
836 struct sk_buff **to_free)
838 __qdisc_drop(skb, to_free);
839 qdisc_qstats_drop(sch);
841 return NET_XMIT_DROP;
844 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
845 long it will take to send a packet given its size.
847 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
849 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
852 slot >>= rtab->rate.cell_log;
854 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
855 return rtab->data[slot];
858 struct psched_ratecfg {
859 u64 rate_bytes_ps; /* bytes per second */
866 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
871 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
872 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
874 return ((u64)len * r->mult) >> r->shift;
877 void psched_ratecfg_precompute(struct psched_ratecfg *r,
878 const struct tc_ratespec *conf,
881 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
882 const struct psched_ratecfg *r)
884 memset(res, 0, sizeof(*res));
886 /* legacy struct tc_ratespec has a 32bit @rate field
887 * Qdisc using 64bit rate should add new attributes
888 * in order to maintain compatibility.
890 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
892 res->overhead = r->overhead;
893 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);