1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __NET_SCHED_GENERIC_H
3 #define __NET_SCHED_GENERIC_H
5 #include <linux/netdevice.h>
6 #include <linux/types.h>
7 #include <linux/rcupdate.h>
8 #include <linux/pkt_sched.h>
9 #include <linux/pkt_cls.h>
10 #include <linux/percpu.h>
11 #include <linux/dynamic_queue_limits.h>
12 #include <linux/list.h>
13 #include <linux/refcount.h>
14 #include <linux/workqueue.h>
15 #include <linux/mutex.h>
16 #include <net/gen_stats.h>
17 #include <net/rtnetlink.h>
25 typedef int tc_setup_cb_t(enum tc_setup_type type,
26 void *type_data, void *cb_priv);
28 typedef int tc_indr_block_bind_cb_t(struct net_device *dev, void *cb_priv,
29 enum tc_setup_type type, void *type_data);
31 struct qdisc_rate_table {
32 struct tc_ratespec rate;
34 struct qdisc_rate_table *next;
40 __QDISC_STATE_DEACTIVATED,
43 struct qdisc_size_table {
45 struct list_head list;
46 struct tc_sizespec szopts;
51 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
52 struct qdisc_skb_head {
63 int (*enqueue)(struct sk_buff *skb,
65 struct sk_buff **to_free);
66 struct sk_buff * (*dequeue)(struct Qdisc *sch);
68 #define TCQ_F_BUILTIN 1
69 #define TCQ_F_INGRESS 2
70 #define TCQ_F_CAN_BYPASS 4
71 #define TCQ_F_MQROOT 8
72 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
73 * q->dev_queue : It can test
74 * netif_xmit_frozen_or_stopped() before
75 * dequeueing next packet.
76 * Its true for MQ/MQPRIO slaves, or non
79 #define TCQ_F_WARN_NONWC (1 << 16)
80 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
81 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
82 * qdisc_tree_decrease_qlen() should stop.
84 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
85 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
86 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
88 const struct Qdisc_ops *ops;
89 struct qdisc_size_table __rcu *stab;
90 struct hlist_node hash;
94 struct netdev_queue *dev_queue;
96 struct net_rate_estimator __rcu *rate_est;
97 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
98 struct gnet_stats_queue __percpu *cpu_qstats;
103 * For performance sake on SMP, we put highly modified fields at the end
105 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
106 struct qdisc_skb_head q;
107 struct gnet_stats_basic_packed bstats;
109 struct gnet_stats_queue qstats;
111 struct Qdisc *next_sched;
112 struct sk_buff_head skb_bad_txq;
114 spinlock_t busylock ____cacheline_aligned_in_smp;
119 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
121 if (qdisc->flags & TCQ_F_BUILTIN)
123 refcount_inc(&qdisc->refcnt);
126 /* Intended to be used by unlocked users, when concurrent qdisc release is
130 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
132 if (qdisc->flags & TCQ_F_BUILTIN)
134 if (refcount_inc_not_zero(&qdisc->refcnt))
139 static inline bool qdisc_is_running(struct Qdisc *qdisc)
141 if (qdisc->flags & TCQ_F_NOLOCK)
142 return spin_is_locked(&qdisc->seqlock);
143 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
146 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
148 if (qdisc->flags & TCQ_F_NOLOCK) {
149 if (!spin_trylock(&qdisc->seqlock))
151 } else if (qdisc_is_running(qdisc)) {
154 /* Variant of write_seqcount_begin() telling lockdep a trylock
157 raw_write_seqcount_begin(&qdisc->running);
158 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
162 static inline void qdisc_run_end(struct Qdisc *qdisc)
164 write_seqcount_end(&qdisc->running);
165 if (qdisc->flags & TCQ_F_NOLOCK)
166 spin_unlock(&qdisc->seqlock);
169 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
171 return qdisc->flags & TCQ_F_ONETXQUEUE;
174 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
177 /* Non-BQL migrated drivers will return 0, too. */
178 return dql_avail(&txq->dql);
184 struct Qdisc_class_ops {
186 /* Child qdisc manipulation */
187 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
188 int (*graft)(struct Qdisc *, unsigned long cl,
189 struct Qdisc *, struct Qdisc **,
190 struct netlink_ext_ack *extack);
191 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
192 void (*qlen_notify)(struct Qdisc *, unsigned long);
194 /* Class manipulation routines */
195 unsigned long (*find)(struct Qdisc *, u32 classid);
196 int (*change)(struct Qdisc *, u32, u32,
197 struct nlattr **, unsigned long *,
198 struct netlink_ext_ack *);
199 int (*delete)(struct Qdisc *, unsigned long);
200 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
202 /* Filter manipulation */
203 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
205 struct netlink_ext_ack *extack);
206 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
208 void (*unbind_tcf)(struct Qdisc *, unsigned long);
210 /* rtnetlink specific */
211 int (*dump)(struct Qdisc *, unsigned long,
212 struct sk_buff *skb, struct tcmsg*);
213 int (*dump_stats)(struct Qdisc *, unsigned long,
217 /* Qdisc_class_ops flag values */
219 /* Implements API that doesn't require rtnl lock */
220 enum qdisc_class_ops_flags {
221 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
225 struct Qdisc_ops *next;
226 const struct Qdisc_class_ops *cl_ops;
229 unsigned int static_flags;
231 int (*enqueue)(struct sk_buff *skb,
233 struct sk_buff **to_free);
234 struct sk_buff * (*dequeue)(struct Qdisc *);
235 struct sk_buff * (*peek)(struct Qdisc *);
237 int (*init)(struct Qdisc *sch, struct nlattr *arg,
238 struct netlink_ext_ack *extack);
239 void (*reset)(struct Qdisc *);
240 void (*destroy)(struct Qdisc *);
241 int (*change)(struct Qdisc *sch,
243 struct netlink_ext_ack *extack);
244 void (*attach)(struct Qdisc *sch);
245 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
247 int (*dump)(struct Qdisc *, struct sk_buff *);
248 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
250 void (*ingress_block_set)(struct Qdisc *sch,
252 void (*egress_block_set)(struct Qdisc *sch,
254 u32 (*ingress_block_get)(struct Qdisc *sch);
255 u32 (*egress_block_get)(struct Qdisc *sch);
257 struct module *owner;
267 const struct tcf_proto *goto_tp;
269 /* used by the TC_ACT_REINSERT action */
272 struct gnet_stats_queue *qstats;
279 struct tcf_proto_ops {
280 struct list_head head;
283 int (*classify)(struct sk_buff *,
284 const struct tcf_proto *,
285 struct tcf_result *);
286 int (*init)(struct tcf_proto*);
287 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
288 struct netlink_ext_ack *extack);
290 void* (*get)(struct tcf_proto*, u32 handle);
291 void (*put)(struct tcf_proto *tp, void *f);
292 int (*change)(struct net *net, struct sk_buff *,
293 struct tcf_proto*, unsigned long,
294 u32 handle, struct nlattr **,
296 struct netlink_ext_ack *);
297 int (*delete)(struct tcf_proto *tp, void *arg,
298 bool *last, bool rtnl_held,
299 struct netlink_ext_ack *);
300 void (*walk)(struct tcf_proto *tp,
301 struct tcf_walker *arg, bool rtnl_held);
302 int (*reoffload)(struct tcf_proto *tp, bool add,
303 tc_setup_cb_t *cb, void *cb_priv,
304 struct netlink_ext_ack *extack);
305 void (*bind_class)(void *, u32, unsigned long);
306 void * (*tmplt_create)(struct net *net,
307 struct tcf_chain *chain,
309 struct netlink_ext_ack *extack);
310 void (*tmplt_destroy)(void *tmplt_priv);
312 /* rtnetlink specific */
313 int (*dump)(struct net*, struct tcf_proto*, void *,
314 struct sk_buff *skb, struct tcmsg*,
316 int (*tmplt_dump)(struct sk_buff *skb,
320 struct module *owner;
324 enum tcf_proto_ops_flags {
325 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
329 /* Fast access part */
330 struct tcf_proto __rcu *next;
333 /* called under RCU BH lock*/
334 int (*classify)(struct sk_buff *,
335 const struct tcf_proto *,
336 struct tcf_result *);
342 const struct tcf_proto_ops *ops;
343 struct tcf_chain *chain;
344 /* Lock protects tcf_proto shared state and can be used by unlocked
345 * classifiers to protect their private data.
353 struct qdisc_skb_cb {
356 unsigned int pkt_len;
357 u16 slave_dev_queue_mapping;
360 struct bpf_flow_keys *flow_keys;
362 #define QDISC_CB_PRIV_LEN 20
363 unsigned char data[QDISC_CB_PRIV_LEN];
366 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
369 /* Protects filter_chain. */
370 struct mutex filter_chain_lock;
371 struct tcf_proto __rcu *filter_chain;
372 struct list_head list;
373 struct tcf_block *block;
374 u32 index; /* chain index */
376 unsigned int action_refcnt;
377 bool explicitly_created;
379 const struct tcf_proto_ops *tmplt_ops;
385 /* Lock protects tcf_block and lifetime-management data of chains
386 * attached to the block (refcnt, action_refcnt, explicitly_created).
389 struct list_head chain_list;
390 u32 index; /* block index for shared blocks */
394 struct list_head cb_list;
395 struct list_head owner_list;
397 unsigned int offloadcnt; /* Number of oddloaded filters */
398 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
400 struct tcf_chain *chain;
401 struct list_head filter_chain_list;
406 #ifdef CONFIG_PROVE_LOCKING
407 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
409 return lockdep_is_held(&chain->filter_chain_lock);
412 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
414 return lockdep_is_held(&tp->lock);
417 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain)
422 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
426 #endif /* #ifdef CONFIG_PROVE_LOCKING */
428 #define tcf_chain_dereference(p, chain) \
429 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
431 #define tcf_proto_dereference(p, tp) \
432 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
434 static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
436 if (*flags & TCA_CLS_FLAGS_IN_HW)
438 *flags |= TCA_CLS_FLAGS_IN_HW;
442 static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
444 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
446 *flags &= ~TCA_CLS_FLAGS_IN_HW;
451 tc_cls_offload_cnt_update(struct tcf_block *block, u32 *cnt,
452 u32 *flags, bool add)
456 tcf_block_offload_inc(block, flags);
461 tcf_block_offload_dec(block, flags);
465 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
467 struct qdisc_skb_cb *qcb;
469 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
470 BUILD_BUG_ON(sizeof(qcb->data) < sz);
473 static inline int qdisc_qlen(const struct Qdisc *q)
478 static inline u32 qdisc_qlen_sum(const struct Qdisc *q)
480 u32 qlen = q->qstats.qlen;
482 if (q->flags & TCQ_F_NOLOCK)
483 qlen += atomic_read(&q->q.atomic_qlen);
490 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
492 return (struct qdisc_skb_cb *)skb->cb;
495 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
497 return &qdisc->q.lock;
500 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
502 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
507 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
509 return qdisc->dev_queue->qdisc_sleeping;
512 /* The qdisc root lock is a mechanism by which to top level
513 * of a qdisc tree can be locked from any qdisc node in the
514 * forest. This allows changing the configuration of some
515 * aspect of the qdisc tree while blocking out asynchronous
516 * qdisc access in the packet processing paths.
518 * It is only legal to do this when the root will not change
519 * on us. Otherwise we'll potentially lock the wrong qdisc
520 * root. This is enforced by holding the RTNL semaphore, which
521 * all users of this lock accessor must do.
523 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
525 struct Qdisc *root = qdisc_root(qdisc);
528 return qdisc_lock(root);
531 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
533 struct Qdisc *root = qdisc_root_sleeping(qdisc);
536 return qdisc_lock(root);
539 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
541 struct Qdisc *root = qdisc_root_sleeping(qdisc);
544 return &root->running;
547 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
549 return qdisc->dev_queue->dev;
552 static inline void sch_tree_lock(const struct Qdisc *q)
554 spin_lock_bh(qdisc_root_sleeping_lock(q));
557 static inline void sch_tree_unlock(const struct Qdisc *q)
559 spin_unlock_bh(qdisc_root_sleeping_lock(q));
562 extern struct Qdisc noop_qdisc;
563 extern struct Qdisc_ops noop_qdisc_ops;
564 extern struct Qdisc_ops pfifo_fast_ops;
565 extern struct Qdisc_ops mq_qdisc_ops;
566 extern struct Qdisc_ops noqueue_qdisc_ops;
567 extern const struct Qdisc_ops *default_qdisc_ops;
568 static inline const struct Qdisc_ops *
569 get_default_qdisc_ops(const struct net_device *dev, int ntx)
571 return ntx < dev->real_num_tx_queues ?
572 default_qdisc_ops : &pfifo_fast_ops;
575 struct Qdisc_class_common {
577 struct hlist_node hnode;
580 struct Qdisc_class_hash {
581 struct hlist_head *hash;
582 unsigned int hashsize;
583 unsigned int hashmask;
584 unsigned int hashelems;
587 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
594 static inline struct Qdisc_class_common *
595 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
597 struct Qdisc_class_common *cl;
603 h = qdisc_class_hash(id, hash->hashmask);
604 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
605 if (cl->classid == id)
611 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
613 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
615 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
618 int qdisc_class_hash_init(struct Qdisc_class_hash *);
619 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
620 struct Qdisc_class_common *);
621 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
622 struct Qdisc_class_common *);
623 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
624 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
626 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
627 void dev_init_scheduler(struct net_device *dev);
628 void dev_shutdown(struct net_device *dev);
629 void dev_activate(struct net_device *dev);
630 void dev_deactivate(struct net_device *dev);
631 void dev_deactivate_many(struct list_head *head);
632 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
633 struct Qdisc *qdisc);
634 void qdisc_reset(struct Qdisc *qdisc);
635 void qdisc_put(struct Qdisc *qdisc);
636 void qdisc_put_unlocked(struct Qdisc *qdisc);
637 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
638 #ifdef CONFIG_NET_SCHED
639 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
641 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
642 struct Qdisc *new, struct Qdisc *old,
643 enum tc_setup_type type, void *type_data,
644 struct netlink_ext_ack *extack);
647 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
650 q->flags &= ~TCQ_F_OFFLOADED;
655 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
656 struct Qdisc *new, struct Qdisc *old,
657 enum tc_setup_type type, void *type_data,
658 struct netlink_ext_ack *extack)
662 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
663 const struct Qdisc_ops *ops,
664 struct netlink_ext_ack *extack);
665 void qdisc_free(struct Qdisc *qdisc);
666 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
667 const struct Qdisc_ops *ops, u32 parentid,
668 struct netlink_ext_ack *extack);
669 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
670 const struct qdisc_size_table *stab);
671 int skb_do_redirect(struct sk_buff *);
673 static inline void skb_reset_tc(struct sk_buff *skb)
675 #ifdef CONFIG_NET_CLS_ACT
676 skb->tc_redirected = 0;
680 static inline bool skb_is_tc_redirected(const struct sk_buff *skb)
682 #ifdef CONFIG_NET_CLS_ACT
683 return skb->tc_redirected;
689 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
691 #ifdef CONFIG_NET_CLS_ACT
692 return skb->tc_at_ingress;
698 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
700 #ifdef CONFIG_NET_CLS_ACT
701 if (skb->tc_skip_classify) {
702 skb->tc_skip_classify = 0;
709 /* Reset all TX qdiscs greater than index of a device. */
710 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
714 for (; i < dev->num_tx_queues; i++) {
715 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
717 spin_lock_bh(qdisc_lock(qdisc));
719 spin_unlock_bh(qdisc_lock(qdisc));
724 static inline void qdisc_reset_all_tx(struct net_device *dev)
726 qdisc_reset_all_tx_gt(dev, 0);
729 /* Are all TX queues of the device empty? */
730 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
735 for (i = 0; i < dev->num_tx_queues; i++) {
736 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
737 const struct Qdisc *q = rcu_dereference(txq->qdisc);
748 /* Are any of the TX qdiscs changing? */
749 static inline bool qdisc_tx_changing(const struct net_device *dev)
753 for (i = 0; i < dev->num_tx_queues; i++) {
754 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
755 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
761 /* Is the device using the noop qdisc on all queues? */
762 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
766 for (i = 0; i < dev->num_tx_queues; i++) {
767 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
768 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
774 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
776 return qdisc_skb_cb(skb)->pkt_len;
779 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
780 enum net_xmit_qdisc_t {
781 __NET_XMIT_STOLEN = 0x00010000,
782 __NET_XMIT_BYPASS = 0x00020000,
785 #ifdef CONFIG_NET_CLS_ACT
786 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
788 #define net_xmit_drop_count(e) (1)
791 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
792 const struct Qdisc *sch)
794 #ifdef CONFIG_NET_SCHED
795 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
798 __qdisc_calculate_pkt_len(skb, stab);
802 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
803 struct sk_buff **to_free)
805 qdisc_calculate_pkt_len(skb, sch);
806 return sch->enqueue(skb, sch, to_free);
809 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
811 return q->flags & TCQ_F_CPUSTATS;
814 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
815 __u64 bytes, __u32 packets)
817 bstats->bytes += bytes;
818 bstats->packets += packets;
821 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
822 const struct sk_buff *skb)
824 _bstats_update(bstats,
826 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
829 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
830 __u64 bytes, __u32 packets)
832 u64_stats_update_begin(&bstats->syncp);
833 _bstats_update(&bstats->bstats, bytes, packets);
834 u64_stats_update_end(&bstats->syncp);
837 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
838 const struct sk_buff *skb)
840 u64_stats_update_begin(&bstats->syncp);
841 bstats_update(&bstats->bstats, skb);
842 u64_stats_update_end(&bstats->syncp);
845 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
846 const struct sk_buff *skb)
848 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
851 static inline void qdisc_bstats_update(struct Qdisc *sch,
852 const struct sk_buff *skb)
854 bstats_update(&sch->bstats, skb);
857 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
858 const struct sk_buff *skb)
860 sch->qstats.backlog -= qdisc_pkt_len(skb);
863 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
864 const struct sk_buff *skb)
866 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
869 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
870 const struct sk_buff *skb)
872 sch->qstats.backlog += qdisc_pkt_len(skb);
875 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
876 const struct sk_buff *skb)
878 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
881 static inline void qdisc_qstats_atomic_qlen_inc(struct Qdisc *sch)
883 atomic_inc(&sch->q.atomic_qlen);
886 static inline void qdisc_qstats_atomic_qlen_dec(struct Qdisc *sch)
888 atomic_dec(&sch->q.atomic_qlen);
891 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
893 this_cpu_inc(sch->cpu_qstats->requeues);
896 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
898 sch->qstats.drops += count;
901 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
906 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
908 qstats->overlimits++;
911 static inline void qdisc_qstats_drop(struct Qdisc *sch)
913 qstats_drop_inc(&sch->qstats);
916 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
918 this_cpu_inc(sch->cpu_qstats->drops);
921 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
923 sch->qstats.overlimits++;
926 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
928 __u32 qlen = qdisc_qlen_sum(sch);
930 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
933 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
936 struct gnet_stats_queue qstats = { 0 };
937 __u32 len = qdisc_qlen_sum(sch);
939 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
941 *backlog = qstats.backlog;
944 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
948 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
949 qdisc_tree_reduce_backlog(sch, qlen, backlog);
952 static inline void qdisc_purge_queue(struct Qdisc *sch)
956 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
958 qdisc_tree_reduce_backlog(sch, qlen, backlog);
961 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
968 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
969 struct qdisc_skb_head *qh)
971 struct sk_buff *last = qh->tail;
984 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
986 __qdisc_enqueue_tail(skb, &sch->q);
987 qdisc_qstats_backlog_inc(sch, skb);
988 return NET_XMIT_SUCCESS;
991 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
992 struct qdisc_skb_head *qh)
994 skb->next = qh->head;
1002 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1004 struct sk_buff *skb = qh->head;
1006 if (likely(skb != NULL)) {
1007 qh->head = skb->next;
1009 if (qh->head == NULL)
1017 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1019 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1021 if (likely(skb != NULL)) {
1022 qdisc_qstats_backlog_dec(sch, skb);
1023 qdisc_bstats_update(sch, skb);
1029 /* Instead of calling kfree_skb() while root qdisc lock is held,
1030 * queue the skb for future freeing at end of __dev_xmit_skb()
1032 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1034 skb->next = *to_free;
1038 static inline void __qdisc_drop_all(struct sk_buff *skb,
1039 struct sk_buff **to_free)
1042 skb->prev->next = *to_free;
1044 skb->next = *to_free;
1048 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1049 struct qdisc_skb_head *qh,
1050 struct sk_buff **to_free)
1052 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1054 if (likely(skb != NULL)) {
1055 unsigned int len = qdisc_pkt_len(skb);
1057 qdisc_qstats_backlog_dec(sch, skb);
1058 __qdisc_drop(skb, to_free);
1065 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
1066 struct sk_buff **to_free)
1068 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
1071 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1073 const struct qdisc_skb_head *qh = &sch->q;
1078 /* generic pseudo peek method for non-work-conserving qdisc */
1079 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1081 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1083 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1085 skb = sch->dequeue(sch);
1088 __skb_queue_head(&sch->gso_skb, skb);
1089 /* it's still part of the queue */
1090 qdisc_qstats_backlog_inc(sch, skb);
1098 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1099 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1101 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1104 skb = __skb_dequeue(&sch->gso_skb);
1105 qdisc_qstats_backlog_dec(sch, skb);
1108 skb = sch->dequeue(sch);
1114 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1117 * We do not know the backlog in bytes of this list, it
1118 * is up to the caller to correct it
1122 rtnl_kfree_skbs(qh->head, qh->tail);
1130 static inline void qdisc_reset_queue(struct Qdisc *sch)
1132 __qdisc_reset_queue(&sch->q);
1133 sch->qstats.backlog = 0;
1136 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1137 struct Qdisc **pold)
1145 qdisc_tree_flush_backlog(old);
1146 sch_tree_unlock(sch);
1151 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1153 rtnl_kfree_skbs(skb, skb);
1154 qdisc_qstats_drop(sch);
1157 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1158 struct sk_buff **to_free)
1160 __qdisc_drop(skb, to_free);
1161 qdisc_qstats_cpu_drop(sch);
1163 return NET_XMIT_DROP;
1166 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1167 struct sk_buff **to_free)
1169 __qdisc_drop(skb, to_free);
1170 qdisc_qstats_drop(sch);
1172 return NET_XMIT_DROP;
1175 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1176 struct sk_buff **to_free)
1178 __qdisc_drop_all(skb, to_free);
1179 qdisc_qstats_drop(sch);
1181 return NET_XMIT_DROP;
1184 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1185 long it will take to send a packet given its size.
1187 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1189 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1192 slot >>= rtab->rate.cell_log;
1194 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1195 return rtab->data[slot];
1198 struct psched_ratecfg {
1199 u64 rate_bytes_ps; /* bytes per second */
1206 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1211 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1212 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1214 return ((u64)len * r->mult) >> r->shift;
1217 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1218 const struct tc_ratespec *conf,
1221 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1222 const struct psched_ratecfg *r)
1224 memset(res, 0, sizeof(*res));
1226 /* legacy struct tc_ratespec has a 32bit @rate field
1227 * Qdisc using 64bit rate should add new attributes
1228 * in order to maintain compatibility.
1230 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1232 res->overhead = r->overhead;
1233 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1236 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1237 * The fast path only needs to access filter list and to update stats
1240 struct tcf_proto *filter_list;
1241 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1242 struct gnet_stats_queue __percpu *cpu_qstats;
1243 struct rcu_head rcu;
1246 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1247 const struct sk_buff *skb)
1249 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1252 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1254 this_cpu_inc(miniq->cpu_qstats->drops);
1257 struct mini_Qdisc_pair {
1258 struct mini_Qdisc miniq1;
1259 struct mini_Qdisc miniq2;
1260 struct mini_Qdisc __rcu **p_miniq;
1263 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1264 struct tcf_proto *tp_head);
1265 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1266 struct mini_Qdisc __rcu **p_miniq);
1268 static inline void skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res)
1270 struct gnet_stats_queue *stats = res->qstats;
1274 ret = netif_receive_skb(skb);
1276 ret = dev_queue_xmit(skb);
1278 qstats_overlimit_inc(res->qstats);