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>
19 struct qdisc_rate_table {
20 struct tc_ratespec rate;
22 struct qdisc_rate_table *next;
28 __QDISC_STATE_DEACTIVATED,
31 struct qdisc_size_table {
33 struct list_head list;
34 struct tc_sizespec szopts;
39 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
40 struct qdisc_skb_head {
48 int (*enqueue)(struct sk_buff *skb,
50 struct sk_buff **to_free);
51 struct sk_buff * (*dequeue)(struct Qdisc *sch);
53 #define TCQ_F_BUILTIN 1
54 #define TCQ_F_INGRESS 2
55 #define TCQ_F_CAN_BYPASS 4
56 #define TCQ_F_MQROOT 8
57 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
58 * q->dev_queue : It can test
59 * netif_xmit_frozen_or_stopped() before
60 * dequeueing next packet.
61 * Its true for MQ/MQPRIO slaves, or non
64 #define TCQ_F_WARN_NONWC (1 << 16)
65 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
66 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
67 * qdisc_tree_decrease_qlen() should stop.
69 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
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 net_rate_estimator __rcu *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;
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(const struct Qdisc *qdisc)
104 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
107 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
109 if (qdisc_is_running(qdisc))
111 /* Variant of write_seqcount_begin() telling lockdep a trylock
114 raw_write_seqcount_begin(&qdisc->running);
115 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
119 static inline void qdisc_run_end(struct Qdisc *qdisc)
121 write_seqcount_end(&qdisc->running);
124 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
126 return qdisc->flags & TCQ_F_ONETXQUEUE;
129 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
132 /* Non-BQL migrated drivers will return 0, too. */
133 return dql_avail(&txq->dql);
139 struct Qdisc_class_ops {
140 /* Child qdisc manipulation */
141 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
142 int (*graft)(struct Qdisc *, unsigned long cl,
143 struct Qdisc *, struct Qdisc **);
144 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
145 void (*qlen_notify)(struct Qdisc *, unsigned long);
147 /* Class manipulation routines */
148 unsigned long (*get)(struct Qdisc *, u32 classid);
149 void (*put)(struct Qdisc *, unsigned long);
150 int (*change)(struct Qdisc *, u32, u32,
151 struct nlattr **, unsigned long *);
152 int (*delete)(struct Qdisc *, unsigned long);
153 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
155 /* Filter manipulation */
156 struct tcf_proto __rcu ** (*tcf_chain)(struct Qdisc *, unsigned long);
157 bool (*tcf_cl_offload)(u32 classid);
158 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
160 void (*unbind_tcf)(struct Qdisc *, unsigned long);
162 /* rtnetlink specific */
163 int (*dump)(struct Qdisc *, unsigned long,
164 struct sk_buff *skb, struct tcmsg*);
165 int (*dump_stats)(struct Qdisc *, unsigned long,
170 struct Qdisc_ops *next;
171 const struct Qdisc_class_ops *cl_ops;
175 int (*enqueue)(struct sk_buff *skb,
177 struct sk_buff **to_free);
178 struct sk_buff * (*dequeue)(struct Qdisc *);
179 struct sk_buff * (*peek)(struct Qdisc *);
181 int (*init)(struct Qdisc *, struct nlattr *arg);
182 void (*reset)(struct Qdisc *);
183 void (*destroy)(struct Qdisc *);
184 int (*change)(struct Qdisc *, struct nlattr *arg);
185 void (*attach)(struct Qdisc *);
187 int (*dump)(struct Qdisc *, struct sk_buff *);
188 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
190 struct module *owner;
199 struct tcf_proto_ops {
200 struct list_head head;
203 int (*classify)(struct sk_buff *,
204 const struct tcf_proto *,
205 struct tcf_result *);
206 int (*init)(struct tcf_proto*);
207 void (*destroy)(struct tcf_proto*);
209 unsigned long (*get)(struct tcf_proto*, u32 handle);
210 int (*change)(struct net *net, struct sk_buff *,
211 struct tcf_proto*, unsigned long,
212 u32 handle, struct nlattr **,
213 unsigned long *, bool);
214 int (*delete)(struct tcf_proto*, unsigned long, bool*);
215 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
217 /* rtnetlink specific */
218 int (*dump)(struct net*, struct tcf_proto*, unsigned long,
219 struct sk_buff *skb, struct tcmsg*);
221 struct module *owner;
225 /* Fast access part */
226 struct tcf_proto __rcu *next;
228 int (*classify)(struct sk_buff *,
229 const struct tcf_proto *,
230 struct tcf_result *);
238 const struct tcf_proto_ops *ops;
242 struct qdisc_skb_cb {
243 unsigned int pkt_len;
244 u16 slave_dev_queue_mapping;
246 #define QDISC_CB_PRIV_LEN 20
247 unsigned char data[QDISC_CB_PRIV_LEN];
250 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
252 struct qdisc_skb_cb *qcb;
254 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
255 BUILD_BUG_ON(sizeof(qcb->data) < sz);
258 static inline int qdisc_qlen(const struct Qdisc *q)
263 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
265 return (struct qdisc_skb_cb *)skb->cb;
268 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
270 return &qdisc->q.lock;
273 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
275 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
280 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
282 return qdisc->dev_queue->qdisc_sleeping;
285 /* The qdisc root lock is a mechanism by which to top level
286 * of a qdisc tree can be locked from any qdisc node in the
287 * forest. This allows changing the configuration of some
288 * aspect of the qdisc tree while blocking out asynchronous
289 * qdisc access in the packet processing paths.
291 * It is only legal to do this when the root will not change
292 * on us. Otherwise we'll potentially lock the wrong qdisc
293 * root. This is enforced by holding the RTNL semaphore, which
294 * all users of this lock accessor must do.
296 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
298 struct Qdisc *root = qdisc_root(qdisc);
301 return qdisc_lock(root);
304 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
306 struct Qdisc *root = qdisc_root_sleeping(qdisc);
309 return qdisc_lock(root);
312 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
314 struct Qdisc *root = qdisc_root_sleeping(qdisc);
317 return &root->running;
320 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
322 return qdisc->dev_queue->dev;
325 static inline void sch_tree_lock(const struct Qdisc *q)
327 spin_lock_bh(qdisc_root_sleeping_lock(q));
330 static inline void sch_tree_unlock(const struct Qdisc *q)
332 spin_unlock_bh(qdisc_root_sleeping_lock(q));
335 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
336 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
338 extern struct Qdisc noop_qdisc;
339 extern struct Qdisc_ops noop_qdisc_ops;
340 extern struct Qdisc_ops pfifo_fast_ops;
341 extern struct Qdisc_ops mq_qdisc_ops;
342 extern struct Qdisc_ops noqueue_qdisc_ops;
343 extern const struct Qdisc_ops *default_qdisc_ops;
344 static inline const struct Qdisc_ops *
345 get_default_qdisc_ops(const struct net_device *dev, int ntx)
347 return ntx < dev->real_num_tx_queues ?
348 default_qdisc_ops : &pfifo_fast_ops;
351 struct Qdisc_class_common {
353 struct hlist_node hnode;
356 struct Qdisc_class_hash {
357 struct hlist_head *hash;
358 unsigned int hashsize;
359 unsigned int hashmask;
360 unsigned int hashelems;
363 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
370 static inline struct Qdisc_class_common *
371 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
373 struct Qdisc_class_common *cl;
376 h = qdisc_class_hash(id, hash->hashmask);
377 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
378 if (cl->classid == id)
384 int qdisc_class_hash_init(struct Qdisc_class_hash *);
385 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
386 struct Qdisc_class_common *);
387 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
388 struct Qdisc_class_common *);
389 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
390 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
392 void dev_init_scheduler(struct net_device *dev);
393 void dev_shutdown(struct net_device *dev);
394 void dev_activate(struct net_device *dev);
395 void dev_deactivate(struct net_device *dev);
396 void dev_deactivate_many(struct list_head *head);
397 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
398 struct Qdisc *qdisc);
399 void qdisc_reset(struct Qdisc *qdisc);
400 void qdisc_destroy(struct Qdisc *qdisc);
401 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
403 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
404 const struct Qdisc_ops *ops);
405 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
406 const struct Qdisc_ops *ops, u32 parentid);
407 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
408 const struct qdisc_size_table *stab);
409 int skb_do_redirect(struct sk_buff *);
411 static inline void skb_reset_tc(struct sk_buff *skb)
413 #ifdef CONFIG_NET_CLS_ACT
414 skb->tc_redirected = 0;
418 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
420 #ifdef CONFIG_NET_CLS_ACT
421 return skb->tc_at_ingress;
427 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
429 #ifdef CONFIG_NET_CLS_ACT
430 if (skb->tc_skip_classify) {
431 skb->tc_skip_classify = 0;
438 /* Reset all TX qdiscs greater then index of a device. */
439 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
443 for (; i < dev->num_tx_queues; i++) {
444 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
446 spin_lock_bh(qdisc_lock(qdisc));
448 spin_unlock_bh(qdisc_lock(qdisc));
453 static inline void qdisc_reset_all_tx(struct net_device *dev)
455 qdisc_reset_all_tx_gt(dev, 0);
458 /* Are all TX queues of the device empty? */
459 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
464 for (i = 0; i < dev->num_tx_queues; i++) {
465 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
466 const struct Qdisc *q = rcu_dereference(txq->qdisc);
477 /* Are any of the TX qdiscs changing? */
478 static inline bool qdisc_tx_changing(const struct net_device *dev)
482 for (i = 0; i < dev->num_tx_queues; i++) {
483 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
484 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
490 /* Is the device using the noop qdisc on all queues? */
491 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
495 for (i = 0; i < dev->num_tx_queues; i++) {
496 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
497 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
503 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
505 return qdisc_skb_cb(skb)->pkt_len;
508 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
509 enum net_xmit_qdisc_t {
510 __NET_XMIT_STOLEN = 0x00010000,
511 __NET_XMIT_BYPASS = 0x00020000,
514 #ifdef CONFIG_NET_CLS_ACT
515 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
517 #define net_xmit_drop_count(e) (1)
520 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
521 const struct Qdisc *sch)
523 #ifdef CONFIG_NET_SCHED
524 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
527 __qdisc_calculate_pkt_len(skb, stab);
531 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
532 struct sk_buff **to_free)
534 qdisc_calculate_pkt_len(skb, sch);
535 return sch->enqueue(skb, sch, to_free);
538 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
540 return q->flags & TCQ_F_CPUSTATS;
543 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
544 __u64 bytes, __u32 packets)
546 bstats->bytes += bytes;
547 bstats->packets += packets;
550 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
551 const struct sk_buff *skb)
553 _bstats_update(bstats,
555 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
558 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
559 __u64 bytes, __u32 packets)
561 u64_stats_update_begin(&bstats->syncp);
562 _bstats_update(&bstats->bstats, bytes, packets);
563 u64_stats_update_end(&bstats->syncp);
566 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
567 const struct sk_buff *skb)
569 u64_stats_update_begin(&bstats->syncp);
570 bstats_update(&bstats->bstats, skb);
571 u64_stats_update_end(&bstats->syncp);
574 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
575 const struct sk_buff *skb)
577 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
580 static inline void qdisc_bstats_update(struct Qdisc *sch,
581 const struct sk_buff *skb)
583 bstats_update(&sch->bstats, skb);
586 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
587 const struct sk_buff *skb)
589 sch->qstats.backlog -= qdisc_pkt_len(skb);
592 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
593 const struct sk_buff *skb)
595 sch->qstats.backlog += qdisc_pkt_len(skb);
598 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
600 sch->qstats.drops += count;
603 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
608 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
610 qstats->overlimits++;
613 static inline void qdisc_qstats_drop(struct Qdisc *sch)
615 qstats_drop_inc(&sch->qstats);
618 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
620 this_cpu_inc(sch->cpu_qstats->drops);
623 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
625 sch->qstats.overlimits++;
628 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
635 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
636 struct qdisc_skb_head *qh)
638 struct sk_buff *last = qh->tail;
649 qdisc_qstats_backlog_inc(sch, skb);
651 return NET_XMIT_SUCCESS;
654 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
656 return __qdisc_enqueue_tail(skb, sch, &sch->q);
659 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
661 struct sk_buff *skb = qh->head;
663 if (likely(skb != NULL)) {
664 qh->head = skb->next;
666 if (qh->head == NULL)
674 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
676 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
678 if (likely(skb != NULL)) {
679 qdisc_qstats_backlog_dec(sch, skb);
680 qdisc_bstats_update(sch, skb);
686 /* Instead of calling kfree_skb() while root qdisc lock is held,
687 * queue the skb for future freeing at end of __dev_xmit_skb()
689 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
691 skb->next = *to_free;
695 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
696 struct qdisc_skb_head *qh,
697 struct sk_buff **to_free)
699 struct sk_buff *skb = __qdisc_dequeue_head(qh);
701 if (likely(skb != NULL)) {
702 unsigned int len = qdisc_pkt_len(skb);
704 qdisc_qstats_backlog_dec(sch, skb);
705 __qdisc_drop(skb, to_free);
712 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
713 struct sk_buff **to_free)
715 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
718 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
720 const struct qdisc_skb_head *qh = &sch->q;
725 /* generic pseudo peek method for non-work-conserving qdisc */
726 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
728 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
730 sch->gso_skb = sch->dequeue(sch);
732 /* it's still part of the queue */
733 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
741 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
742 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
744 struct sk_buff *skb = sch->gso_skb;
748 qdisc_qstats_backlog_dec(sch, skb);
751 skb = sch->dequeue(sch);
757 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
760 * We do not know the backlog in bytes of this list, it
761 * is up to the caller to correct it
765 rtnl_kfree_skbs(qh->head, qh->tail);
773 static inline void qdisc_reset_queue(struct Qdisc *sch)
775 __qdisc_reset_queue(&sch->q);
776 sch->qstats.backlog = 0;
779 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
788 qdisc_tree_reduce_backlog(old, old->q.qlen, old->qstats.backlog);
791 sch_tree_unlock(sch);
796 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
798 rtnl_kfree_skbs(skb, skb);
799 qdisc_qstats_drop(sch);
803 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
804 struct sk_buff **to_free)
806 __qdisc_drop(skb, to_free);
807 qdisc_qstats_drop(sch);
809 return NET_XMIT_DROP;
812 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
813 long it will take to send a packet given its size.
815 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
817 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
820 slot >>= rtab->rate.cell_log;
822 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
823 return rtab->data[slot];
826 struct psched_ratecfg {
827 u64 rate_bytes_ps; /* bytes per second */
834 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
839 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
840 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
842 return ((u64)len * r->mult) >> r->shift;
845 void psched_ratecfg_precompute(struct psched_ratecfg *r,
846 const struct tc_ratespec *conf,
849 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
850 const struct psched_ratecfg *r)
852 memset(res, 0, sizeof(*res));
854 /* legacy struct tc_ratespec has a 32bit @rate field
855 * Qdisc using 64bit rate should add new attributes
856 * in order to maintain compatibility.
858 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
860 res->overhead = r->overhead;
861 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);