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 <linux/rwsem.h>
17 #include <linux/atomic.h>
18 #include <linux/hashtable.h>
19 #include <net/gen_stats.h>
20 #include <net/rtnetlink.h>
21 #include <net/flow_offload.h>
29 struct qdisc_rate_table {
30 struct tc_ratespec rate;
32 struct qdisc_rate_table *next;
38 __QDISC_STATE_DEACTIVATED,
42 struct qdisc_size_table {
44 struct list_head list;
45 struct tc_sizespec szopts;
50 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
51 struct qdisc_skb_head {
59 int (*enqueue)(struct sk_buff *skb,
61 struct sk_buff **to_free);
62 struct sk_buff * (*dequeue)(struct Qdisc *sch);
64 #define TCQ_F_BUILTIN 1
65 #define TCQ_F_INGRESS 2
66 #define TCQ_F_CAN_BYPASS 4
67 #define TCQ_F_MQROOT 8
68 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
69 * q->dev_queue : It can test
70 * netif_xmit_frozen_or_stopped() before
71 * dequeueing next packet.
72 * Its true for MQ/MQPRIO slaves, or non
75 #define TCQ_F_WARN_NONWC (1 << 16)
76 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
77 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
78 * qdisc_tree_decrease_qlen() should stop.
80 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
81 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
82 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
84 const struct Qdisc_ops *ops;
85 struct qdisc_size_table __rcu *stab;
86 struct hlist_node hash;
90 struct netdev_queue *dev_queue;
92 struct net_rate_estimator __rcu *rate_est;
93 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
94 struct gnet_stats_queue __percpu *cpu_qstats;
99 * For performance sake on SMP, we put highly modified fields at the end
101 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
102 struct qdisc_skb_head q;
103 struct gnet_stats_basic_packed bstats;
105 struct gnet_stats_queue qstats;
107 struct Qdisc *next_sched;
108 struct sk_buff_head skb_bad_txq;
110 spinlock_t busylock ____cacheline_aligned_in_smp;
113 /* for NOLOCK qdisc, true if there are no enqueued skbs */
118 long privdata[] ____cacheline_aligned;
121 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
123 if (qdisc->flags & TCQ_F_BUILTIN)
125 refcount_inc(&qdisc->refcnt);
128 /* Intended to be used by unlocked users, when concurrent qdisc release is
132 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
134 if (qdisc->flags & TCQ_F_BUILTIN)
136 if (refcount_inc_not_zero(&qdisc->refcnt))
141 static inline bool qdisc_is_running(struct Qdisc *qdisc)
143 if (qdisc->flags & TCQ_F_NOLOCK)
144 return spin_is_locked(&qdisc->seqlock);
145 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
148 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
150 return q->flags & TCQ_F_CPUSTATS;
153 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
155 if (qdisc_is_percpu_stats(qdisc))
156 return READ_ONCE(qdisc->empty);
157 return !READ_ONCE(qdisc->q.qlen);
160 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
162 if (qdisc->flags & TCQ_F_NOLOCK) {
163 if (spin_trylock(&qdisc->seqlock))
166 /* No need to insist if the MISSED flag was already set.
167 * Note that test_and_set_bit() also gives us memory ordering
168 * guarantees wrt potential earlier enqueue() and below
169 * spin_trylock(), both of which are necessary to prevent races
171 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
174 /* Try to take the lock again to make sure that we will either
175 * grab it or the CPU that still has it will see MISSED set
176 * when testing it in qdisc_run_end()
178 if (!spin_trylock(&qdisc->seqlock))
182 WRITE_ONCE(qdisc->empty, false);
183 } else if (qdisc_is_running(qdisc)) {
186 /* Variant of write_seqcount_begin() telling lockdep a trylock
189 raw_write_seqcount_begin(&qdisc->running);
190 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
194 static inline void qdisc_run_end(struct Qdisc *qdisc)
196 write_seqcount_end(&qdisc->running);
197 if (qdisc->flags & TCQ_F_NOLOCK) {
198 spin_unlock(&qdisc->seqlock);
200 /* spin_unlock() only has store-release semantic. The unlock
201 * and test_bit() ordering is a store-load ordering, so a full
202 * memory barrier is needed here.
206 if (unlikely(test_bit(__QDISC_STATE_MISSED,
208 clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
209 __netif_schedule(qdisc);
214 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
216 return qdisc->flags & TCQ_F_ONETXQUEUE;
219 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
222 /* Non-BQL migrated drivers will return 0, too. */
223 return dql_avail(&txq->dql);
229 struct Qdisc_class_ops {
231 /* Child qdisc manipulation */
232 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
233 int (*graft)(struct Qdisc *, unsigned long cl,
234 struct Qdisc *, struct Qdisc **,
235 struct netlink_ext_ack *extack);
236 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
237 void (*qlen_notify)(struct Qdisc *, unsigned long);
239 /* Class manipulation routines */
240 unsigned long (*find)(struct Qdisc *, u32 classid);
241 int (*change)(struct Qdisc *, u32, u32,
242 struct nlattr **, unsigned long *,
243 struct netlink_ext_ack *);
244 int (*delete)(struct Qdisc *, unsigned long);
245 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
247 /* Filter manipulation */
248 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
250 struct netlink_ext_ack *extack);
251 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
253 void (*unbind_tcf)(struct Qdisc *, unsigned long);
255 /* rtnetlink specific */
256 int (*dump)(struct Qdisc *, unsigned long,
257 struct sk_buff *skb, struct tcmsg*);
258 int (*dump_stats)(struct Qdisc *, unsigned long,
262 /* Qdisc_class_ops flag values */
264 /* Implements API that doesn't require rtnl lock */
265 enum qdisc_class_ops_flags {
266 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
270 struct Qdisc_ops *next;
271 const struct Qdisc_class_ops *cl_ops;
274 unsigned int static_flags;
276 int (*enqueue)(struct sk_buff *skb,
278 struct sk_buff **to_free);
279 struct sk_buff * (*dequeue)(struct Qdisc *);
280 struct sk_buff * (*peek)(struct Qdisc *);
282 int (*init)(struct Qdisc *sch, struct nlattr *arg,
283 struct netlink_ext_ack *extack);
284 void (*reset)(struct Qdisc *);
285 void (*destroy)(struct Qdisc *);
286 int (*change)(struct Qdisc *sch,
288 struct netlink_ext_ack *extack);
289 void (*attach)(struct Qdisc *sch);
290 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
291 void (*change_real_num_tx)(struct Qdisc *sch,
292 unsigned int new_real_tx);
294 int (*dump)(struct Qdisc *, struct sk_buff *);
295 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
297 void (*ingress_block_set)(struct Qdisc *sch,
299 void (*egress_block_set)(struct Qdisc *sch,
301 u32 (*ingress_block_get)(struct Qdisc *sch);
302 u32 (*egress_block_get)(struct Qdisc *sch);
304 struct module *owner;
314 const struct tcf_proto *goto_tp;
316 /* used in the skb_tc_reinsert function */
319 struct gnet_stats_queue *qstats;
326 struct tcf_proto_ops {
327 struct list_head head;
330 int (*classify)(struct sk_buff *,
331 const struct tcf_proto *,
332 struct tcf_result *);
333 int (*init)(struct tcf_proto*);
334 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
335 struct netlink_ext_ack *extack);
337 void* (*get)(struct tcf_proto*, u32 handle);
338 void (*put)(struct tcf_proto *tp, void *f);
339 int (*change)(struct net *net, struct sk_buff *,
340 struct tcf_proto*, unsigned long,
341 u32 handle, struct nlattr **,
343 struct netlink_ext_ack *);
344 int (*delete)(struct tcf_proto *tp, void *arg,
345 bool *last, bool rtnl_held,
346 struct netlink_ext_ack *);
347 bool (*delete_empty)(struct tcf_proto *tp);
348 void (*walk)(struct tcf_proto *tp,
349 struct tcf_walker *arg, bool rtnl_held);
350 int (*reoffload)(struct tcf_proto *tp, bool add,
351 flow_setup_cb_t *cb, void *cb_priv,
352 struct netlink_ext_ack *extack);
353 void (*hw_add)(struct tcf_proto *tp,
355 void (*hw_del)(struct tcf_proto *tp,
357 void (*bind_class)(void *, u32, unsigned long,
358 void *, unsigned long);
359 void * (*tmplt_create)(struct net *net,
360 struct tcf_chain *chain,
362 struct netlink_ext_ack *extack);
363 void (*tmplt_destroy)(void *tmplt_priv);
365 /* rtnetlink specific */
366 int (*dump)(struct net*, struct tcf_proto*, void *,
367 struct sk_buff *skb, struct tcmsg*,
369 int (*terse_dump)(struct net *net,
370 struct tcf_proto *tp, void *fh,
372 struct tcmsg *t, bool rtnl_held);
373 int (*tmplt_dump)(struct sk_buff *skb,
377 struct module *owner;
381 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
382 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
383 * conditions can occur when filters are inserted/deleted simultaneously.
385 enum tcf_proto_ops_flags {
386 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
390 /* Fast access part */
391 struct tcf_proto __rcu *next;
394 /* called under RCU BH lock*/
395 int (*classify)(struct sk_buff *,
396 const struct tcf_proto *,
397 struct tcf_result *);
403 const struct tcf_proto_ops *ops;
404 struct tcf_chain *chain;
405 /* Lock protects tcf_proto shared state and can be used by unlocked
406 * classifiers to protect their private data.
412 struct hlist_node destroy_ht_node;
415 struct qdisc_skb_cb {
417 unsigned int pkt_len;
418 u16 slave_dev_queue_mapping;
421 #define QDISC_CB_PRIV_LEN 20
422 unsigned char data[QDISC_CB_PRIV_LEN];
426 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
429 /* Protects filter_chain. */
430 struct mutex filter_chain_lock;
431 struct tcf_proto __rcu *filter_chain;
432 struct list_head list;
433 struct tcf_block *block;
434 u32 index; /* chain index */
436 unsigned int action_refcnt;
437 bool explicitly_created;
439 const struct tcf_proto_ops *tmplt_ops;
445 /* Lock protects tcf_block and lifetime-management data of chains
446 * attached to the block (refcnt, action_refcnt, explicitly_created).
449 struct list_head chain_list;
450 u32 index; /* block index for shared blocks */
451 u32 classid; /* which class this block belongs to */
455 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
456 struct flow_block flow_block;
457 struct list_head owner_list;
459 atomic_t offloadcnt; /* Number of oddloaded filters */
460 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
461 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
463 struct tcf_chain *chain;
464 struct list_head filter_chain_list;
467 DECLARE_HASHTABLE(proto_destroy_ht, 7);
468 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
471 #ifdef CONFIG_PROVE_LOCKING
472 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
474 return lockdep_is_held(&chain->filter_chain_lock);
477 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
479 return lockdep_is_held(&tp->lock);
482 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain)
487 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
491 #endif /* #ifdef CONFIG_PROVE_LOCKING */
493 #define tcf_chain_dereference(p, chain) \
494 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
496 #define tcf_proto_dereference(p, tp) \
497 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
499 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
501 struct qdisc_skb_cb *qcb;
503 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
504 BUILD_BUG_ON(sizeof(qcb->data) < sz);
507 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
509 return this_cpu_ptr(q->cpu_qstats)->qlen;
512 static inline int qdisc_qlen(const struct Qdisc *q)
517 static inline int qdisc_qlen_sum(const struct Qdisc *q)
519 __u32 qlen = q->qstats.qlen;
522 if (qdisc_is_percpu_stats(q)) {
523 for_each_possible_cpu(i)
524 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
532 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
534 return (struct qdisc_skb_cb *)skb->cb;
537 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
539 return &qdisc->q.lock;
542 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
544 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
549 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
551 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
554 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
556 return qdisc->dev_queue->qdisc_sleeping;
559 /* The qdisc root lock is a mechanism by which to top level
560 * of a qdisc tree can be locked from any qdisc node in the
561 * forest. This allows changing the configuration of some
562 * aspect of the qdisc tree while blocking out asynchronous
563 * qdisc access in the packet processing paths.
565 * It is only legal to do this when the root will not change
566 * on us. Otherwise we'll potentially lock the wrong qdisc
567 * root. This is enforced by holding the RTNL semaphore, which
568 * all users of this lock accessor must do.
570 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
572 struct Qdisc *root = qdisc_root(qdisc);
575 return qdisc_lock(root);
578 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
580 struct Qdisc *root = qdisc_root_sleeping(qdisc);
583 return qdisc_lock(root);
586 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
588 struct Qdisc *root = qdisc_root_sleeping(qdisc);
591 return &root->running;
594 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
596 return qdisc->dev_queue->dev;
599 static inline void sch_tree_lock(const struct Qdisc *q)
601 spin_lock_bh(qdisc_root_sleeping_lock(q));
604 static inline void sch_tree_unlock(const struct Qdisc *q)
606 spin_unlock_bh(qdisc_root_sleeping_lock(q));
609 extern struct Qdisc noop_qdisc;
610 extern struct Qdisc_ops noop_qdisc_ops;
611 extern struct Qdisc_ops pfifo_fast_ops;
612 extern struct Qdisc_ops mq_qdisc_ops;
613 extern struct Qdisc_ops noqueue_qdisc_ops;
614 extern const struct Qdisc_ops *default_qdisc_ops;
615 static inline const struct Qdisc_ops *
616 get_default_qdisc_ops(const struct net_device *dev, int ntx)
618 return ntx < dev->real_num_tx_queues ?
619 default_qdisc_ops : &pfifo_fast_ops;
622 struct Qdisc_class_common {
624 struct hlist_node hnode;
627 struct Qdisc_class_hash {
628 struct hlist_head *hash;
629 unsigned int hashsize;
630 unsigned int hashmask;
631 unsigned int hashelems;
634 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
641 static inline struct Qdisc_class_common *
642 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
644 struct Qdisc_class_common *cl;
650 h = qdisc_class_hash(id, hash->hashmask);
651 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
652 if (cl->classid == id)
658 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
660 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
662 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
665 int qdisc_class_hash_init(struct Qdisc_class_hash *);
666 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
667 struct Qdisc_class_common *);
668 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
669 struct Qdisc_class_common *);
670 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
671 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
673 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
674 void dev_qdisc_change_real_num_tx(struct net_device *dev,
675 unsigned int new_real_tx);
676 void dev_init_scheduler(struct net_device *dev);
677 void dev_shutdown(struct net_device *dev);
678 void dev_activate(struct net_device *dev);
679 void dev_deactivate(struct net_device *dev);
680 void dev_deactivate_many(struct list_head *head);
681 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
682 struct Qdisc *qdisc);
683 void qdisc_reset(struct Qdisc *qdisc);
684 void qdisc_put(struct Qdisc *qdisc);
685 void qdisc_put_unlocked(struct Qdisc *qdisc);
686 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
687 #ifdef CONFIG_NET_SCHED
688 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
690 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
691 struct Qdisc *new, struct Qdisc *old,
692 enum tc_setup_type type, void *type_data,
693 struct netlink_ext_ack *extack);
696 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
699 q->flags &= ~TCQ_F_OFFLOADED;
704 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
705 struct Qdisc *new, struct Qdisc *old,
706 enum tc_setup_type type, void *type_data,
707 struct netlink_ext_ack *extack)
711 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
712 const struct Qdisc_ops *ops,
713 struct netlink_ext_ack *extack);
714 void qdisc_free(struct Qdisc *qdisc);
715 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
716 const struct Qdisc_ops *ops, u32 parentid,
717 struct netlink_ext_ack *extack);
718 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
719 const struct qdisc_size_table *stab);
720 int skb_do_redirect(struct sk_buff *);
722 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
724 #ifdef CONFIG_NET_CLS_ACT
725 return skb->tc_at_ingress;
731 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
733 #ifdef CONFIG_NET_CLS_ACT
734 if (skb->tc_skip_classify) {
735 skb->tc_skip_classify = 0;
742 /* Reset all TX qdiscs greater than index of a device. */
743 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
747 for (; i < dev->num_tx_queues; i++) {
748 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
750 spin_lock_bh(qdisc_lock(qdisc));
752 spin_unlock_bh(qdisc_lock(qdisc));
757 /* Are all TX queues of the device empty? */
758 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
763 for (i = 0; i < dev->num_tx_queues; i++) {
764 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
765 const struct Qdisc *q = rcu_dereference(txq->qdisc);
767 if (!qdisc_is_empty(q)) {
776 /* Are any of the TX qdiscs changing? */
777 static inline bool qdisc_tx_changing(const struct net_device *dev)
781 for (i = 0; i < dev->num_tx_queues; i++) {
782 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
783 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
789 /* Is the device using the noop qdisc on all queues? */
790 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
794 for (i = 0; i < dev->num_tx_queues; i++) {
795 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
796 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
802 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
804 return qdisc_skb_cb(skb)->pkt_len;
807 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
808 enum net_xmit_qdisc_t {
809 __NET_XMIT_STOLEN = 0x00010000,
810 __NET_XMIT_BYPASS = 0x00020000,
813 #ifdef CONFIG_NET_CLS_ACT
814 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
816 #define net_xmit_drop_count(e) (1)
819 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
820 const struct Qdisc *sch)
822 #ifdef CONFIG_NET_SCHED
823 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
826 __qdisc_calculate_pkt_len(skb, stab);
830 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
831 struct sk_buff **to_free)
833 qdisc_calculate_pkt_len(skb, sch);
834 return sch->enqueue(skb, sch, to_free);
837 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
838 __u64 bytes, __u32 packets)
840 bstats->bytes += bytes;
841 bstats->packets += packets;
844 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
845 const struct sk_buff *skb)
847 _bstats_update(bstats,
849 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
852 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
853 __u64 bytes, __u32 packets)
855 u64_stats_update_begin(&bstats->syncp);
856 _bstats_update(&bstats->bstats, bytes, packets);
857 u64_stats_update_end(&bstats->syncp);
860 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
861 const struct sk_buff *skb)
863 u64_stats_update_begin(&bstats->syncp);
864 bstats_update(&bstats->bstats, skb);
865 u64_stats_update_end(&bstats->syncp);
868 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
869 const struct sk_buff *skb)
871 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
874 static inline void qdisc_bstats_update(struct Qdisc *sch,
875 const struct sk_buff *skb)
877 bstats_update(&sch->bstats, skb);
880 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
881 const struct sk_buff *skb)
883 sch->qstats.backlog -= qdisc_pkt_len(skb);
886 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
887 const struct sk_buff *skb)
889 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
892 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
893 const struct sk_buff *skb)
895 sch->qstats.backlog += qdisc_pkt_len(skb);
898 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
899 const struct sk_buff *skb)
901 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
904 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
906 this_cpu_inc(sch->cpu_qstats->qlen);
909 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
911 this_cpu_dec(sch->cpu_qstats->qlen);
914 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
916 this_cpu_inc(sch->cpu_qstats->requeues);
919 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
921 sch->qstats.drops += count;
924 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
929 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
931 qstats->overlimits++;
934 static inline void qdisc_qstats_drop(struct Qdisc *sch)
936 qstats_drop_inc(&sch->qstats);
939 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
941 this_cpu_inc(sch->cpu_qstats->drops);
944 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
946 sch->qstats.overlimits++;
949 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
951 __u32 qlen = qdisc_qlen_sum(sch);
953 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
956 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
959 struct gnet_stats_queue qstats = { 0 };
960 __u32 len = qdisc_qlen_sum(sch);
962 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
964 *backlog = qstats.backlog;
967 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
971 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
972 qdisc_tree_reduce_backlog(sch, qlen, backlog);
975 static inline void qdisc_purge_queue(struct Qdisc *sch)
979 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
981 qdisc_tree_reduce_backlog(sch, qlen, backlog);
984 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
991 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
992 struct qdisc_skb_head *qh)
994 struct sk_buff *last = qh->tail;
1007 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1009 __qdisc_enqueue_tail(skb, &sch->q);
1010 qdisc_qstats_backlog_inc(sch, skb);
1011 return NET_XMIT_SUCCESS;
1014 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1015 struct qdisc_skb_head *qh)
1017 skb->next = qh->head;
1025 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1027 struct sk_buff *skb = qh->head;
1029 if (likely(skb != NULL)) {
1030 qh->head = skb->next;
1032 if (qh->head == NULL)
1040 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1042 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1044 if (likely(skb != NULL)) {
1045 qdisc_qstats_backlog_dec(sch, skb);
1046 qdisc_bstats_update(sch, skb);
1052 /* Instead of calling kfree_skb() while root qdisc lock is held,
1053 * queue the skb for future freeing at end of __dev_xmit_skb()
1055 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1057 skb->next = *to_free;
1061 static inline void __qdisc_drop_all(struct sk_buff *skb,
1062 struct sk_buff **to_free)
1065 skb->prev->next = *to_free;
1067 skb->next = *to_free;
1071 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1072 struct qdisc_skb_head *qh,
1073 struct sk_buff **to_free)
1075 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1077 if (likely(skb != NULL)) {
1078 unsigned int len = qdisc_pkt_len(skb);
1080 qdisc_qstats_backlog_dec(sch, skb);
1081 __qdisc_drop(skb, to_free);
1088 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1090 const struct qdisc_skb_head *qh = &sch->q;
1095 /* generic pseudo peek method for non-work-conserving qdisc */
1096 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1098 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1100 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1102 skb = sch->dequeue(sch);
1105 __skb_queue_head(&sch->gso_skb, skb);
1106 /* it's still part of the queue */
1107 qdisc_qstats_backlog_inc(sch, skb);
1115 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1116 struct sk_buff *skb)
1118 if (qdisc_is_percpu_stats(sch)) {
1119 qdisc_qstats_cpu_backlog_dec(sch, skb);
1120 qdisc_bstats_cpu_update(sch, skb);
1121 qdisc_qstats_cpu_qlen_dec(sch);
1123 qdisc_qstats_backlog_dec(sch, skb);
1124 qdisc_bstats_update(sch, skb);
1129 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1130 unsigned int pkt_len)
1132 if (qdisc_is_percpu_stats(sch)) {
1133 qdisc_qstats_cpu_qlen_inc(sch);
1134 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1136 sch->qstats.backlog += pkt_len;
1141 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1142 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1144 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1147 skb = __skb_dequeue(&sch->gso_skb);
1148 if (qdisc_is_percpu_stats(sch)) {
1149 qdisc_qstats_cpu_backlog_dec(sch, skb);
1150 qdisc_qstats_cpu_qlen_dec(sch);
1152 qdisc_qstats_backlog_dec(sch, skb);
1156 skb = sch->dequeue(sch);
1162 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1165 * We do not know the backlog in bytes of this list, it
1166 * is up to the caller to correct it
1170 rtnl_kfree_skbs(qh->head, qh->tail);
1178 static inline void qdisc_reset_queue(struct Qdisc *sch)
1180 __qdisc_reset_queue(&sch->q);
1183 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1184 struct Qdisc **pold)
1192 qdisc_purge_queue(old);
1193 sch_tree_unlock(sch);
1198 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1200 rtnl_kfree_skbs(skb, skb);
1201 qdisc_qstats_drop(sch);
1204 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1205 struct sk_buff **to_free)
1207 __qdisc_drop(skb, to_free);
1208 qdisc_qstats_cpu_drop(sch);
1210 return NET_XMIT_DROP;
1213 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1214 struct sk_buff **to_free)
1216 __qdisc_drop(skb, to_free);
1217 qdisc_qstats_drop(sch);
1219 return NET_XMIT_DROP;
1222 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1223 struct sk_buff **to_free)
1225 __qdisc_drop_all(skb, to_free);
1226 qdisc_qstats_drop(sch);
1228 return NET_XMIT_DROP;
1231 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1232 long it will take to send a packet given its size.
1234 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1236 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1239 slot >>= rtab->rate.cell_log;
1241 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1242 return rtab->data[slot];
1245 struct psched_ratecfg {
1246 u64 rate_bytes_ps; /* bytes per second */
1254 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1262 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1263 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1265 return ((u64)len * r->mult) >> r->shift;
1268 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1269 const struct tc_ratespec *conf,
1272 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1273 const struct psched_ratecfg *r)
1275 memset(res, 0, sizeof(*res));
1277 /* legacy struct tc_ratespec has a 32bit @rate field
1278 * Qdisc using 64bit rate should add new attributes
1279 * in order to maintain compatibility.
1281 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1283 res->overhead = r->overhead;
1285 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1288 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1289 * The fast path only needs to access filter list and to update stats
1292 struct tcf_proto *filter_list;
1293 struct tcf_block *block;
1294 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1295 struct gnet_stats_queue __percpu *cpu_qstats;
1296 struct rcu_head rcu;
1299 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1300 const struct sk_buff *skb)
1302 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1305 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1307 this_cpu_inc(miniq->cpu_qstats->drops);
1310 struct mini_Qdisc_pair {
1311 struct mini_Qdisc miniq1;
1312 struct mini_Qdisc miniq2;
1313 struct mini_Qdisc __rcu **p_miniq;
1316 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1317 struct tcf_proto *tp_head);
1318 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1319 struct mini_Qdisc __rcu **p_miniq);
1320 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1321 struct tcf_block *block);
1323 /* Make sure qdisc is no longer in SCHED state. */
1324 static inline void qdisc_synchronize(const struct Qdisc *q)
1326 while (test_bit(__QDISC_STATE_SCHED, &q->state))