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,
40 __QDISC_STATE_DRAINING,
44 /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly.
45 * Use qdisc_run_begin/end() or qdisc_is_running() instead.
47 __QDISC_STATE2_RUNNING,
50 #define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED)
51 #define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING)
53 #define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \
56 struct qdisc_size_table {
58 struct list_head list;
59 struct tc_sizespec szopts;
64 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
65 struct qdisc_skb_head {
73 int (*enqueue)(struct sk_buff *skb,
75 struct sk_buff **to_free);
76 struct sk_buff * (*dequeue)(struct Qdisc *sch);
78 #define TCQ_F_BUILTIN 1
79 #define TCQ_F_INGRESS 2
80 #define TCQ_F_CAN_BYPASS 4
81 #define TCQ_F_MQROOT 8
82 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
83 * q->dev_queue : It can test
84 * netif_xmit_frozen_or_stopped() before
85 * dequeueing next packet.
86 * Its true for MQ/MQPRIO slaves, or non
89 #define TCQ_F_WARN_NONWC (1 << 16)
90 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
91 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
92 * qdisc_tree_decrease_qlen() should stop.
94 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
95 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
96 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
98 const struct Qdisc_ops *ops;
99 struct qdisc_size_table __rcu *stab;
100 struct hlist_node hash;
104 struct netdev_queue *dev_queue;
106 struct net_rate_estimator __rcu *rate_est;
107 struct gnet_stats_basic_sync __percpu *cpu_bstats;
108 struct gnet_stats_queue __percpu *cpu_qstats;
113 * For performance sake on SMP, we put highly modified fields at the end
115 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
116 struct qdisc_skb_head q;
117 struct gnet_stats_basic_sync bstats;
118 struct gnet_stats_queue qstats;
120 unsigned long state2; /* must be written under qdisc spinlock */
121 struct Qdisc *next_sched;
122 struct sk_buff_head skb_bad_txq;
124 spinlock_t busylock ____cacheline_aligned_in_smp;
128 netdevice_tracker dev_tracker;
130 long privdata[] ____cacheline_aligned;
133 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
135 if (qdisc->flags & TCQ_F_BUILTIN)
137 refcount_inc(&qdisc->refcnt);
140 /* Intended to be used by unlocked users, when concurrent qdisc release is
144 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
146 if (qdisc->flags & TCQ_F_BUILTIN)
148 if (refcount_inc_not_zero(&qdisc->refcnt))
153 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
154 * root_lock section, or provide their own memory barriers -- ordering
155 * against qdisc_run_begin/end() atomic bit operations.
157 static inline bool qdisc_is_running(struct Qdisc *qdisc)
159 if (qdisc->flags & TCQ_F_NOLOCK)
160 return spin_is_locked(&qdisc->seqlock);
161 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
164 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
166 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
169 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
171 return q->flags & TCQ_F_CPUSTATS;
174 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
176 if (qdisc_is_percpu_stats(qdisc))
177 return nolock_qdisc_is_empty(qdisc);
178 return !READ_ONCE(qdisc->q.qlen);
181 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
182 * the qdisc root lock acquired.
184 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
186 if (qdisc->flags & TCQ_F_NOLOCK) {
187 if (spin_trylock(&qdisc->seqlock))
190 /* No need to insist if the MISSED flag was already set.
191 * Note that test_and_set_bit() also gives us memory ordering
192 * guarantees wrt potential earlier enqueue() and below
193 * spin_trylock(), both of which are necessary to prevent races
195 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
198 /* Try to take the lock again to make sure that we will either
199 * grab it or the CPU that still has it will see MISSED set
200 * when testing it in qdisc_run_end()
202 return spin_trylock(&qdisc->seqlock);
204 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
207 static inline void qdisc_run_end(struct Qdisc *qdisc)
209 if (qdisc->flags & TCQ_F_NOLOCK) {
210 spin_unlock(&qdisc->seqlock);
212 /* spin_unlock() only has store-release semantic. The unlock
213 * and test_bit() ordering is a store-load ordering, so a full
214 * memory barrier is needed here.
218 if (unlikely(test_bit(__QDISC_STATE_MISSED,
220 __netif_schedule(qdisc);
222 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
226 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
228 return qdisc->flags & TCQ_F_ONETXQUEUE;
231 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
234 /* Non-BQL migrated drivers will return 0, too. */
235 return dql_avail(&txq->dql);
241 struct Qdisc_class_ops {
243 /* Child qdisc manipulation */
244 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
245 int (*graft)(struct Qdisc *, unsigned long cl,
246 struct Qdisc *, struct Qdisc **,
247 struct netlink_ext_ack *extack);
248 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
249 void (*qlen_notify)(struct Qdisc *, unsigned long);
251 /* Class manipulation routines */
252 unsigned long (*find)(struct Qdisc *, u32 classid);
253 int (*change)(struct Qdisc *, u32, u32,
254 struct nlattr **, unsigned long *,
255 struct netlink_ext_ack *);
256 int (*delete)(struct Qdisc *, unsigned long,
257 struct netlink_ext_ack *);
258 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
260 /* Filter manipulation */
261 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
263 struct netlink_ext_ack *extack);
264 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
266 void (*unbind_tcf)(struct Qdisc *, unsigned long);
268 /* rtnetlink specific */
269 int (*dump)(struct Qdisc *, unsigned long,
270 struct sk_buff *skb, struct tcmsg*);
271 int (*dump_stats)(struct Qdisc *, unsigned long,
275 /* Qdisc_class_ops flag values */
277 /* Implements API that doesn't require rtnl lock */
278 enum qdisc_class_ops_flags {
279 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
283 struct Qdisc_ops *next;
284 const struct Qdisc_class_ops *cl_ops;
287 unsigned int static_flags;
289 int (*enqueue)(struct sk_buff *skb,
291 struct sk_buff **to_free);
292 struct sk_buff * (*dequeue)(struct Qdisc *);
293 struct sk_buff * (*peek)(struct Qdisc *);
295 int (*init)(struct Qdisc *sch, struct nlattr *arg,
296 struct netlink_ext_ack *extack);
297 void (*reset)(struct Qdisc *);
298 void (*destroy)(struct Qdisc *);
299 int (*change)(struct Qdisc *sch,
301 struct netlink_ext_ack *extack);
302 void (*attach)(struct Qdisc *sch);
303 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
304 void (*change_real_num_tx)(struct Qdisc *sch,
305 unsigned int new_real_tx);
307 int (*dump)(struct Qdisc *, struct sk_buff *);
308 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
310 void (*ingress_block_set)(struct Qdisc *sch,
312 void (*egress_block_set)(struct Qdisc *sch,
314 u32 (*ingress_block_get)(struct Qdisc *sch);
315 u32 (*egress_block_get)(struct Qdisc *sch);
317 struct module *owner;
327 const struct tcf_proto *goto_tp;
334 struct tcf_proto_ops {
335 struct list_head head;
338 int (*classify)(struct sk_buff *,
339 const struct tcf_proto *,
340 struct tcf_result *);
341 int (*init)(struct tcf_proto*);
342 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
343 struct netlink_ext_ack *extack);
345 void* (*get)(struct tcf_proto*, u32 handle);
346 void (*put)(struct tcf_proto *tp, void *f);
347 int (*change)(struct net *net, struct sk_buff *,
348 struct tcf_proto*, unsigned long,
349 u32 handle, struct nlattr **,
351 struct netlink_ext_ack *);
352 int (*delete)(struct tcf_proto *tp, void *arg,
353 bool *last, bool rtnl_held,
354 struct netlink_ext_ack *);
355 bool (*delete_empty)(struct tcf_proto *tp);
356 void (*walk)(struct tcf_proto *tp,
357 struct tcf_walker *arg, bool rtnl_held);
358 int (*reoffload)(struct tcf_proto *tp, bool add,
359 flow_setup_cb_t *cb, void *cb_priv,
360 struct netlink_ext_ack *extack);
361 void (*hw_add)(struct tcf_proto *tp,
363 void (*hw_del)(struct tcf_proto *tp,
365 void (*bind_class)(void *, u32, unsigned long,
366 void *, unsigned long);
367 void * (*tmplt_create)(struct net *net,
368 struct tcf_chain *chain,
370 struct netlink_ext_ack *extack);
371 void (*tmplt_destroy)(void *tmplt_priv);
372 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp,
375 /* rtnetlink specific */
376 int (*dump)(struct net*, struct tcf_proto*, void *,
377 struct sk_buff *skb, struct tcmsg*,
379 int (*terse_dump)(struct net *net,
380 struct tcf_proto *tp, void *fh,
382 struct tcmsg *t, bool rtnl_held);
383 int (*tmplt_dump)(struct sk_buff *skb,
387 struct module *owner;
391 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
392 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
393 * conditions can occur when filters are inserted/deleted simultaneously.
395 enum tcf_proto_ops_flags {
396 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
400 /* Fast access part */
401 struct tcf_proto __rcu *next;
404 /* called under RCU BH lock*/
405 int (*classify)(struct sk_buff *,
406 const struct tcf_proto *,
407 struct tcf_result *);
413 const struct tcf_proto_ops *ops;
414 struct tcf_chain *chain;
415 /* Lock protects tcf_proto shared state and can be used by unlocked
416 * classifiers to protect their private data.
422 struct hlist_node destroy_ht_node;
425 struct qdisc_skb_cb {
427 unsigned int pkt_len;
428 u16 slave_dev_queue_mapping;
431 #define QDISC_CB_PRIV_LEN 20
432 unsigned char data[QDISC_CB_PRIV_LEN];
435 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
438 /* Protects filter_chain. */
439 struct mutex filter_chain_lock;
440 struct tcf_proto __rcu *filter_chain;
441 struct list_head list;
442 struct tcf_block *block;
443 u32 index; /* chain index */
445 unsigned int action_refcnt;
446 bool explicitly_created;
448 const struct tcf_proto_ops *tmplt_ops;
454 /* Lock protects tcf_block and lifetime-management data of chains
455 * attached to the block (refcnt, action_refcnt, explicitly_created).
458 struct list_head chain_list;
459 u32 index; /* block index for shared blocks */
460 u32 classid; /* which class this block belongs to */
464 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
465 struct flow_block flow_block;
466 struct list_head owner_list;
468 atomic_t offloadcnt; /* Number of oddloaded filters */
469 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
470 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
472 struct tcf_chain *chain;
473 struct list_head filter_chain_list;
476 DECLARE_HASHTABLE(proto_destroy_ht, 7);
477 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
480 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
482 return lockdep_is_held(&chain->filter_chain_lock);
485 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
487 return lockdep_is_held(&tp->lock);
490 #define tcf_chain_dereference(p, chain) \
491 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
493 #define tcf_proto_dereference(p, tp) \
494 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
496 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
498 struct qdisc_skb_cb *qcb;
500 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
501 BUILD_BUG_ON(sizeof(qcb->data) < sz);
504 static inline int qdisc_qlen(const struct Qdisc *q)
509 static inline int qdisc_qlen_sum(const struct Qdisc *q)
511 __u32 qlen = q->qstats.qlen;
514 if (qdisc_is_percpu_stats(q)) {
515 for_each_possible_cpu(i)
516 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
524 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
526 return (struct qdisc_skb_cb *)skb->cb;
529 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
531 return &qdisc->q.lock;
534 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
536 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
541 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
543 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
546 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
548 return qdisc->dev_queue->qdisc_sleeping;
551 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
553 struct Qdisc *root = qdisc_root_sleeping(qdisc);
556 return qdisc_lock(root);
559 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
561 return qdisc->dev_queue->dev;
564 static inline void sch_tree_lock(struct Qdisc *q)
566 if (q->flags & TCQ_F_MQROOT)
567 spin_lock_bh(qdisc_lock(q));
569 spin_lock_bh(qdisc_root_sleeping_lock(q));
572 static inline void sch_tree_unlock(struct Qdisc *q)
574 if (q->flags & TCQ_F_MQROOT)
575 spin_unlock_bh(qdisc_lock(q));
577 spin_unlock_bh(qdisc_root_sleeping_lock(q));
580 extern struct Qdisc noop_qdisc;
581 extern struct Qdisc_ops noop_qdisc_ops;
582 extern struct Qdisc_ops pfifo_fast_ops;
583 extern struct Qdisc_ops mq_qdisc_ops;
584 extern struct Qdisc_ops noqueue_qdisc_ops;
585 extern const struct Qdisc_ops *default_qdisc_ops;
586 static inline const struct Qdisc_ops *
587 get_default_qdisc_ops(const struct net_device *dev, int ntx)
589 return ntx < dev->real_num_tx_queues ?
590 default_qdisc_ops : &pfifo_fast_ops;
593 struct Qdisc_class_common {
595 struct hlist_node hnode;
598 struct Qdisc_class_hash {
599 struct hlist_head *hash;
600 unsigned int hashsize;
601 unsigned int hashmask;
602 unsigned int hashelems;
605 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
612 static inline struct Qdisc_class_common *
613 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
615 struct Qdisc_class_common *cl;
621 h = qdisc_class_hash(id, hash->hashmask);
622 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
623 if (cl->classid == id)
629 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
631 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
633 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
636 int qdisc_class_hash_init(struct Qdisc_class_hash *);
637 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
638 struct Qdisc_class_common *);
639 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
640 struct Qdisc_class_common *);
641 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
642 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
644 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
645 void dev_qdisc_change_real_num_tx(struct net_device *dev,
646 unsigned int new_real_tx);
647 void dev_init_scheduler(struct net_device *dev);
648 void dev_shutdown(struct net_device *dev);
649 void dev_activate(struct net_device *dev);
650 void dev_deactivate(struct net_device *dev);
651 void dev_deactivate_many(struct list_head *head);
652 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
653 struct Qdisc *qdisc);
654 void qdisc_reset(struct Qdisc *qdisc);
655 void qdisc_put(struct Qdisc *qdisc);
656 void qdisc_put_unlocked(struct Qdisc *qdisc);
657 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
658 #ifdef CONFIG_NET_SCHED
659 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
661 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
662 struct Qdisc *new, struct Qdisc *old,
663 enum tc_setup_type type, void *type_data,
664 struct netlink_ext_ack *extack);
667 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
670 q->flags &= ~TCQ_F_OFFLOADED;
675 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
676 struct Qdisc *new, struct Qdisc *old,
677 enum tc_setup_type type, void *type_data,
678 struct netlink_ext_ack *extack)
682 void qdisc_offload_query_caps(struct net_device *dev,
683 enum tc_setup_type type,
684 void *caps, size_t caps_len);
685 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
686 const struct Qdisc_ops *ops,
687 struct netlink_ext_ack *extack);
688 void qdisc_free(struct Qdisc *qdisc);
689 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
690 const struct Qdisc_ops *ops, u32 parentid,
691 struct netlink_ext_ack *extack);
692 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
693 const struct qdisc_size_table *stab);
694 int skb_do_redirect(struct sk_buff *);
696 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
698 #ifdef CONFIG_NET_CLS_ACT
699 return skb->tc_at_ingress;
705 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
707 #ifdef CONFIG_NET_CLS_ACT
708 if (skb->tc_skip_classify) {
709 skb->tc_skip_classify = 0;
716 /* Reset all TX qdiscs greater than index of a device. */
717 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
721 for (; i < dev->num_tx_queues; i++) {
722 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
724 spin_lock_bh(qdisc_lock(qdisc));
726 spin_unlock_bh(qdisc_lock(qdisc));
731 /* Are all TX queues of the device empty? */
732 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
737 for (i = 0; i < dev->num_tx_queues; i++) {
738 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
739 const struct Qdisc *q = rcu_dereference(txq->qdisc);
741 if (!qdisc_is_empty(q)) {
750 /* Are any of the TX qdiscs changing? */
751 static inline bool qdisc_tx_changing(const struct net_device *dev)
755 for (i = 0; i < dev->num_tx_queues; i++) {
756 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
757 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
763 /* Is the device using the noop qdisc on all queues? */
764 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
768 for (i = 0; i < dev->num_tx_queues; i++) {
769 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
770 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
776 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
778 return qdisc_skb_cb(skb)->pkt_len;
781 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
782 enum net_xmit_qdisc_t {
783 __NET_XMIT_STOLEN = 0x00010000,
784 __NET_XMIT_BYPASS = 0x00020000,
787 #ifdef CONFIG_NET_CLS_ACT
788 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
790 #define net_xmit_drop_count(e) (1)
793 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
794 const struct Qdisc *sch)
796 #ifdef CONFIG_NET_SCHED
797 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
800 __qdisc_calculate_pkt_len(skb, stab);
804 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
805 struct sk_buff **to_free)
807 qdisc_calculate_pkt_len(skb, sch);
808 return sch->enqueue(skb, sch, to_free);
811 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
812 __u64 bytes, __u32 packets)
814 u64_stats_update_begin(&bstats->syncp);
815 u64_stats_add(&bstats->bytes, bytes);
816 u64_stats_add(&bstats->packets, packets);
817 u64_stats_update_end(&bstats->syncp);
820 static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
821 const struct sk_buff *skb)
823 _bstats_update(bstats,
825 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
828 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
829 const struct sk_buff *skb)
831 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
834 static inline void qdisc_bstats_update(struct Qdisc *sch,
835 const struct sk_buff *skb)
837 bstats_update(&sch->bstats, skb);
840 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
841 const struct sk_buff *skb)
843 sch->qstats.backlog -= qdisc_pkt_len(skb);
846 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
847 const struct sk_buff *skb)
849 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
852 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
853 const struct sk_buff *skb)
855 sch->qstats.backlog += qdisc_pkt_len(skb);
858 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
859 const struct sk_buff *skb)
861 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
864 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
866 this_cpu_inc(sch->cpu_qstats->qlen);
869 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
871 this_cpu_dec(sch->cpu_qstats->qlen);
874 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
876 this_cpu_inc(sch->cpu_qstats->requeues);
879 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
881 sch->qstats.drops += count;
884 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
889 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
891 qstats->overlimits++;
894 static inline void qdisc_qstats_drop(struct Qdisc *sch)
896 qstats_drop_inc(&sch->qstats);
899 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
901 this_cpu_inc(sch->cpu_qstats->drops);
904 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
906 sch->qstats.overlimits++;
909 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
911 __u32 qlen = qdisc_qlen_sum(sch);
913 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
916 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
919 struct gnet_stats_queue qstats = { 0 };
921 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
922 *qlen = qstats.qlen + qdisc_qlen(sch);
923 *backlog = qstats.backlog;
926 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
930 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
931 qdisc_tree_reduce_backlog(sch, qlen, backlog);
934 static inline void qdisc_purge_queue(struct Qdisc *sch)
938 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
940 qdisc_tree_reduce_backlog(sch, qlen, backlog);
943 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
944 struct qdisc_skb_head *qh)
946 struct sk_buff *last = qh->tail;
959 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
961 __qdisc_enqueue_tail(skb, &sch->q);
962 qdisc_qstats_backlog_inc(sch, skb);
963 return NET_XMIT_SUCCESS;
966 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
967 struct qdisc_skb_head *qh)
969 skb->next = qh->head;
977 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
979 struct sk_buff *skb = qh->head;
981 if (likely(skb != NULL)) {
982 qh->head = skb->next;
984 if (qh->head == NULL)
992 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
994 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
996 if (likely(skb != NULL)) {
997 qdisc_qstats_backlog_dec(sch, skb);
998 qdisc_bstats_update(sch, skb);
1004 /* Instead of calling kfree_skb() while root qdisc lock is held,
1005 * queue the skb for future freeing at end of __dev_xmit_skb()
1007 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1009 skb->next = *to_free;
1013 static inline void __qdisc_drop_all(struct sk_buff *skb,
1014 struct sk_buff **to_free)
1017 skb->prev->next = *to_free;
1019 skb->next = *to_free;
1023 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1024 struct qdisc_skb_head *qh,
1025 struct sk_buff **to_free)
1027 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1029 if (likely(skb != NULL)) {
1030 unsigned int len = qdisc_pkt_len(skb);
1032 qdisc_qstats_backlog_dec(sch, skb);
1033 __qdisc_drop(skb, to_free);
1040 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1042 const struct qdisc_skb_head *qh = &sch->q;
1047 /* generic pseudo peek method for non-work-conserving qdisc */
1048 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1050 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1052 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1054 skb = sch->dequeue(sch);
1057 __skb_queue_head(&sch->gso_skb, skb);
1058 /* it's still part of the queue */
1059 qdisc_qstats_backlog_inc(sch, skb);
1067 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1068 struct sk_buff *skb)
1070 if (qdisc_is_percpu_stats(sch)) {
1071 qdisc_qstats_cpu_backlog_dec(sch, skb);
1072 qdisc_bstats_cpu_update(sch, skb);
1073 qdisc_qstats_cpu_qlen_dec(sch);
1075 qdisc_qstats_backlog_dec(sch, skb);
1076 qdisc_bstats_update(sch, skb);
1081 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1082 unsigned int pkt_len)
1084 if (qdisc_is_percpu_stats(sch)) {
1085 qdisc_qstats_cpu_qlen_inc(sch);
1086 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1088 sch->qstats.backlog += pkt_len;
1093 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1094 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1096 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1099 skb = __skb_dequeue(&sch->gso_skb);
1100 if (qdisc_is_percpu_stats(sch)) {
1101 qdisc_qstats_cpu_backlog_dec(sch, skb);
1102 qdisc_qstats_cpu_qlen_dec(sch);
1104 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);
1135 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1136 struct Qdisc **pold)
1144 qdisc_purge_queue(old);
1145 sch_tree_unlock(sch);
1150 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1152 rtnl_kfree_skbs(skb, skb);
1153 qdisc_qstats_drop(sch);
1156 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1157 struct sk_buff **to_free)
1159 __qdisc_drop(skb, to_free);
1160 qdisc_qstats_cpu_drop(sch);
1162 return NET_XMIT_DROP;
1165 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1166 struct sk_buff **to_free)
1168 __qdisc_drop(skb, to_free);
1169 qdisc_qstats_drop(sch);
1171 return NET_XMIT_DROP;
1174 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1175 struct sk_buff **to_free)
1177 __qdisc_drop_all(skb, to_free);
1178 qdisc_qstats_drop(sch);
1180 return NET_XMIT_DROP;
1183 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1184 long it will take to send a packet given its size.
1186 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1188 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1191 slot >>= rtab->rate.cell_log;
1193 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1194 return rtab->data[slot];
1197 struct psched_ratecfg {
1198 u64 rate_bytes_ps; /* bytes per second */
1206 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1214 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1215 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1217 return ((u64)len * r->mult) >> r->shift;
1220 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1221 const struct tc_ratespec *conf,
1224 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1225 const struct psched_ratecfg *r)
1227 memset(res, 0, sizeof(*res));
1229 /* legacy struct tc_ratespec has a 32bit @rate field
1230 * Qdisc using 64bit rate should add new attributes
1231 * in order to maintain compatibility.
1233 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1235 res->overhead = r->overhead;
1237 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1240 struct psched_pktrate {
1241 u64 rate_pkts_ps; /* packets per second */
1246 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1247 unsigned int pkt_num)
1249 return ((u64)pkt_num * r->mult) >> r->shift;
1252 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1254 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1255 * The fast path only needs to access filter list and to update stats
1258 struct tcf_proto *filter_list;
1259 struct tcf_block *block;
1260 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1261 struct gnet_stats_queue __percpu *cpu_qstats;
1262 unsigned long rcu_state;
1265 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1266 const struct sk_buff *skb)
1268 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1271 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1273 this_cpu_inc(miniq->cpu_qstats->drops);
1276 struct mini_Qdisc_pair {
1277 struct mini_Qdisc miniq1;
1278 struct mini_Qdisc miniq2;
1279 struct mini_Qdisc __rcu **p_miniq;
1282 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1283 struct tcf_proto *tp_head);
1284 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1285 struct mini_Qdisc __rcu **p_miniq);
1286 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1287 struct tcf_block *block);
1289 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1291 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1293 /* Make sure qdisc is no longer in SCHED state. */
1294 static inline void qdisc_synchronize(const struct Qdisc *q)
1296 while (test_bit(__QDISC_STATE_SCHED, &q->state))