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 static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc)
142 if (qdisc->flags & TCQ_F_BUILTIN)
144 return refcount_dec_if_one(&qdisc->refcnt);
147 /* Intended to be used by unlocked users, when concurrent qdisc release is
151 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
153 if (qdisc->flags & TCQ_F_BUILTIN)
155 if (refcount_inc_not_zero(&qdisc->refcnt))
160 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
161 * root_lock section, or provide their own memory barriers -- ordering
162 * against qdisc_run_begin/end() atomic bit operations.
164 static inline bool qdisc_is_running(struct Qdisc *qdisc)
166 if (qdisc->flags & TCQ_F_NOLOCK)
167 return spin_is_locked(&qdisc->seqlock);
168 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
171 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
173 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
176 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
178 return q->flags & TCQ_F_CPUSTATS;
181 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
183 if (qdisc_is_percpu_stats(qdisc))
184 return nolock_qdisc_is_empty(qdisc);
185 return !READ_ONCE(qdisc->q.qlen);
188 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
189 * the qdisc root lock acquired.
191 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
193 if (qdisc->flags & TCQ_F_NOLOCK) {
194 if (spin_trylock(&qdisc->seqlock))
197 /* No need to insist if the MISSED flag was already set.
198 * Note that test_and_set_bit() also gives us memory ordering
199 * guarantees wrt potential earlier enqueue() and below
200 * spin_trylock(), both of which are necessary to prevent races
202 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
205 /* Try to take the lock again to make sure that we will either
206 * grab it or the CPU that still has it will see MISSED set
207 * when testing it in qdisc_run_end()
209 return spin_trylock(&qdisc->seqlock);
211 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
214 static inline void qdisc_run_end(struct Qdisc *qdisc)
216 if (qdisc->flags & TCQ_F_NOLOCK) {
217 spin_unlock(&qdisc->seqlock);
219 /* spin_unlock() only has store-release semantic. The unlock
220 * and test_bit() ordering is a store-load ordering, so a full
221 * memory barrier is needed here.
225 if (unlikely(test_bit(__QDISC_STATE_MISSED,
227 __netif_schedule(qdisc);
229 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
233 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
235 return qdisc->flags & TCQ_F_ONETXQUEUE;
238 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
241 /* Non-BQL migrated drivers will return 0, too. */
242 return dql_avail(&txq->dql);
248 struct Qdisc_class_ops {
250 /* Child qdisc manipulation */
251 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
252 int (*graft)(struct Qdisc *, unsigned long cl,
253 struct Qdisc *, struct Qdisc **,
254 struct netlink_ext_ack *extack);
255 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
256 void (*qlen_notify)(struct Qdisc *, unsigned long);
258 /* Class manipulation routines */
259 unsigned long (*find)(struct Qdisc *, u32 classid);
260 int (*change)(struct Qdisc *, u32, u32,
261 struct nlattr **, unsigned long *,
262 struct netlink_ext_ack *);
263 int (*delete)(struct Qdisc *, unsigned long,
264 struct netlink_ext_ack *);
265 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
267 /* Filter manipulation */
268 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
270 struct netlink_ext_ack *extack);
271 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
273 void (*unbind_tcf)(struct Qdisc *, unsigned long);
275 /* rtnetlink specific */
276 int (*dump)(struct Qdisc *, unsigned long,
277 struct sk_buff *skb, struct tcmsg*);
278 int (*dump_stats)(struct Qdisc *, unsigned long,
282 /* Qdisc_class_ops flag values */
284 /* Implements API that doesn't require rtnl lock */
285 enum qdisc_class_ops_flags {
286 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
290 struct Qdisc_ops *next;
291 const struct Qdisc_class_ops *cl_ops;
294 unsigned int static_flags;
296 int (*enqueue)(struct sk_buff *skb,
298 struct sk_buff **to_free);
299 struct sk_buff * (*dequeue)(struct Qdisc *);
300 struct sk_buff * (*peek)(struct Qdisc *);
302 int (*init)(struct Qdisc *sch, struct nlattr *arg,
303 struct netlink_ext_ack *extack);
304 void (*reset)(struct Qdisc *);
305 void (*destroy)(struct Qdisc *);
306 int (*change)(struct Qdisc *sch,
308 struct netlink_ext_ack *extack);
309 void (*attach)(struct Qdisc *sch);
310 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
311 void (*change_real_num_tx)(struct Qdisc *sch,
312 unsigned int new_real_tx);
314 int (*dump)(struct Qdisc *, struct sk_buff *);
315 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
317 void (*ingress_block_set)(struct Qdisc *sch,
319 void (*egress_block_set)(struct Qdisc *sch,
321 u32 (*ingress_block_get)(struct Qdisc *sch);
322 u32 (*egress_block_get)(struct Qdisc *sch);
324 struct module *owner;
334 const struct tcf_proto *goto_tp;
341 struct tcf_proto_ops {
342 struct list_head head;
345 int (*classify)(struct sk_buff *,
346 const struct tcf_proto *,
347 struct tcf_result *);
348 int (*init)(struct tcf_proto*);
349 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
350 struct netlink_ext_ack *extack);
352 void* (*get)(struct tcf_proto*, u32 handle);
353 void (*put)(struct tcf_proto *tp, void *f);
354 int (*change)(struct net *net, struct sk_buff *,
355 struct tcf_proto*, unsigned long,
356 u32 handle, struct nlattr **,
358 struct netlink_ext_ack *);
359 int (*delete)(struct tcf_proto *tp, void *arg,
360 bool *last, bool rtnl_held,
361 struct netlink_ext_ack *);
362 bool (*delete_empty)(struct tcf_proto *tp);
363 void (*walk)(struct tcf_proto *tp,
364 struct tcf_walker *arg, bool rtnl_held);
365 int (*reoffload)(struct tcf_proto *tp, bool add,
366 flow_setup_cb_t *cb, void *cb_priv,
367 struct netlink_ext_ack *extack);
368 void (*hw_add)(struct tcf_proto *tp,
370 void (*hw_del)(struct tcf_proto *tp,
372 void (*bind_class)(void *, u32, unsigned long,
373 void *, unsigned long);
374 void * (*tmplt_create)(struct net *net,
375 struct tcf_chain *chain,
377 struct netlink_ext_ack *extack);
378 void (*tmplt_destroy)(void *tmplt_priv);
379 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp,
382 /* rtnetlink specific */
383 int (*dump)(struct net*, struct tcf_proto*, void *,
384 struct sk_buff *skb, struct tcmsg*,
386 int (*terse_dump)(struct net *net,
387 struct tcf_proto *tp, void *fh,
389 struct tcmsg *t, bool rtnl_held);
390 int (*tmplt_dump)(struct sk_buff *skb,
394 struct module *owner;
398 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
399 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
400 * conditions can occur when filters are inserted/deleted simultaneously.
402 enum tcf_proto_ops_flags {
403 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
407 /* Fast access part */
408 struct tcf_proto __rcu *next;
411 /* called under RCU BH lock*/
412 int (*classify)(struct sk_buff *,
413 const struct tcf_proto *,
414 struct tcf_result *);
420 const struct tcf_proto_ops *ops;
421 struct tcf_chain *chain;
422 /* Lock protects tcf_proto shared state and can be used by unlocked
423 * classifiers to protect their private data.
429 struct hlist_node destroy_ht_node;
432 struct qdisc_skb_cb {
434 unsigned int pkt_len;
435 u16 slave_dev_queue_mapping;
438 #define QDISC_CB_PRIV_LEN 20
439 unsigned char data[QDISC_CB_PRIV_LEN];
442 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
445 /* Protects filter_chain. */
446 struct mutex filter_chain_lock;
447 struct tcf_proto __rcu *filter_chain;
448 struct list_head list;
449 struct tcf_block *block;
450 u32 index; /* chain index */
452 unsigned int action_refcnt;
453 bool explicitly_created;
455 const struct tcf_proto_ops *tmplt_ops;
461 /* Lock protects tcf_block and lifetime-management data of chains
462 * attached to the block (refcnt, action_refcnt, explicitly_created).
465 struct list_head chain_list;
466 u32 index; /* block index for shared blocks */
467 u32 classid; /* which class this block belongs to */
471 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
472 struct flow_block flow_block;
473 struct list_head owner_list;
475 atomic_t offloadcnt; /* Number of oddloaded filters */
476 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
477 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
479 struct tcf_chain *chain;
480 struct list_head filter_chain_list;
483 DECLARE_HASHTABLE(proto_destroy_ht, 7);
484 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
487 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
489 return lockdep_is_held(&chain->filter_chain_lock);
492 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
494 return lockdep_is_held(&tp->lock);
497 #define tcf_chain_dereference(p, chain) \
498 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
500 #define tcf_proto_dereference(p, tp) \
501 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
503 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
505 struct qdisc_skb_cb *qcb;
507 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
508 BUILD_BUG_ON(sizeof(qcb->data) < sz);
511 static inline int qdisc_qlen(const struct Qdisc *q)
516 static inline int qdisc_qlen_sum(const struct Qdisc *q)
518 __u32 qlen = q->qstats.qlen;
521 if (qdisc_is_percpu_stats(q)) {
522 for_each_possible_cpu(i)
523 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
531 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
533 return (struct qdisc_skb_cb *)skb->cb;
536 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
538 return &qdisc->q.lock;
541 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
543 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
548 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
550 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
553 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
555 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
558 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
560 struct Qdisc *root = qdisc_root_sleeping(qdisc);
563 return qdisc_lock(root);
566 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
568 return qdisc->dev_queue->dev;
571 static inline void sch_tree_lock(struct Qdisc *q)
573 if (q->flags & TCQ_F_MQROOT)
574 spin_lock_bh(qdisc_lock(q));
576 spin_lock_bh(qdisc_root_sleeping_lock(q));
579 static inline void sch_tree_unlock(struct Qdisc *q)
581 if (q->flags & TCQ_F_MQROOT)
582 spin_unlock_bh(qdisc_lock(q));
584 spin_unlock_bh(qdisc_root_sleeping_lock(q));
587 extern struct Qdisc noop_qdisc;
588 extern struct Qdisc_ops noop_qdisc_ops;
589 extern struct Qdisc_ops pfifo_fast_ops;
590 extern struct Qdisc_ops mq_qdisc_ops;
591 extern struct Qdisc_ops noqueue_qdisc_ops;
592 extern const struct Qdisc_ops *default_qdisc_ops;
593 static inline const struct Qdisc_ops *
594 get_default_qdisc_ops(const struct net_device *dev, int ntx)
596 return ntx < dev->real_num_tx_queues ?
597 default_qdisc_ops : &pfifo_fast_ops;
600 struct Qdisc_class_common {
602 struct hlist_node hnode;
605 struct Qdisc_class_hash {
606 struct hlist_head *hash;
607 unsigned int hashsize;
608 unsigned int hashmask;
609 unsigned int hashelems;
612 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
619 static inline struct Qdisc_class_common *
620 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
622 struct Qdisc_class_common *cl;
628 h = qdisc_class_hash(id, hash->hashmask);
629 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
630 if (cl->classid == id)
636 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
638 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
640 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
643 int qdisc_class_hash_init(struct Qdisc_class_hash *);
644 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
645 struct Qdisc_class_common *);
646 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
647 struct Qdisc_class_common *);
648 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
649 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
651 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
652 void dev_qdisc_change_real_num_tx(struct net_device *dev,
653 unsigned int new_real_tx);
654 void dev_init_scheduler(struct net_device *dev);
655 void dev_shutdown(struct net_device *dev);
656 void dev_activate(struct net_device *dev);
657 void dev_deactivate(struct net_device *dev);
658 void dev_deactivate_many(struct list_head *head);
659 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
660 struct Qdisc *qdisc);
661 void qdisc_reset(struct Qdisc *qdisc);
662 void qdisc_destroy(struct Qdisc *qdisc);
663 void qdisc_put(struct Qdisc *qdisc);
664 void qdisc_put_unlocked(struct Qdisc *qdisc);
665 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
666 #ifdef CONFIG_NET_SCHED
667 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
669 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
670 struct Qdisc *new, struct Qdisc *old,
671 enum tc_setup_type type, void *type_data,
672 struct netlink_ext_ack *extack);
675 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
678 q->flags &= ~TCQ_F_OFFLOADED;
683 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
684 struct Qdisc *new, struct Qdisc *old,
685 enum tc_setup_type type, void *type_data,
686 struct netlink_ext_ack *extack)
690 void qdisc_offload_query_caps(struct net_device *dev,
691 enum tc_setup_type type,
692 void *caps, size_t caps_len);
693 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
694 const struct Qdisc_ops *ops,
695 struct netlink_ext_ack *extack);
696 void qdisc_free(struct Qdisc *qdisc);
697 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
698 const struct Qdisc_ops *ops, u32 parentid,
699 struct netlink_ext_ack *extack);
700 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
701 const struct qdisc_size_table *stab);
702 int skb_do_redirect(struct sk_buff *);
704 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
706 #ifdef CONFIG_NET_CLS_ACT
707 return skb->tc_at_ingress;
713 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
715 #ifdef CONFIG_NET_CLS_ACT
716 if (skb->tc_skip_classify) {
717 skb->tc_skip_classify = 0;
724 /* Reset all TX qdiscs greater than index of a device. */
725 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
729 for (; i < dev->num_tx_queues; i++) {
730 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
732 spin_lock_bh(qdisc_lock(qdisc));
734 spin_unlock_bh(qdisc_lock(qdisc));
739 /* Are all TX queues of the device empty? */
740 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
745 for (i = 0; i < dev->num_tx_queues; i++) {
746 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
747 const struct Qdisc *q = rcu_dereference(txq->qdisc);
749 if (!qdisc_is_empty(q)) {
758 /* Are any of the TX qdiscs changing? */
759 static inline bool qdisc_tx_changing(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);
766 if (rcu_access_pointer(txq->qdisc) !=
767 rcu_access_pointer(txq->qdisc_sleeping))
773 /* Is the device using the noop qdisc on all queues? */
774 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
778 for (i = 0; i < dev->num_tx_queues; i++) {
779 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
780 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
786 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
788 return qdisc_skb_cb(skb)->pkt_len;
791 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
792 enum net_xmit_qdisc_t {
793 __NET_XMIT_STOLEN = 0x00010000,
794 __NET_XMIT_BYPASS = 0x00020000,
797 #ifdef CONFIG_NET_CLS_ACT
798 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
800 #define net_xmit_drop_count(e) (1)
803 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
804 const struct Qdisc *sch)
806 #ifdef CONFIG_NET_SCHED
807 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
810 __qdisc_calculate_pkt_len(skb, stab);
814 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
815 struct sk_buff **to_free)
817 qdisc_calculate_pkt_len(skb, sch);
818 return sch->enqueue(skb, sch, to_free);
821 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
822 __u64 bytes, __u32 packets)
824 u64_stats_update_begin(&bstats->syncp);
825 u64_stats_add(&bstats->bytes, bytes);
826 u64_stats_add(&bstats->packets, packets);
827 u64_stats_update_end(&bstats->syncp);
830 static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
831 const struct sk_buff *skb)
833 _bstats_update(bstats,
835 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
838 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
839 const struct sk_buff *skb)
841 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
844 static inline void qdisc_bstats_update(struct Qdisc *sch,
845 const struct sk_buff *skb)
847 bstats_update(&sch->bstats, skb);
850 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
851 const struct sk_buff *skb)
853 sch->qstats.backlog -= qdisc_pkt_len(skb);
856 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
857 const struct sk_buff *skb)
859 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
862 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
863 const struct sk_buff *skb)
865 sch->qstats.backlog += qdisc_pkt_len(skb);
868 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
869 const struct sk_buff *skb)
871 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
874 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
876 this_cpu_inc(sch->cpu_qstats->qlen);
879 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
881 this_cpu_dec(sch->cpu_qstats->qlen);
884 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
886 this_cpu_inc(sch->cpu_qstats->requeues);
889 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
891 sch->qstats.drops += count;
894 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
899 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
901 qstats->overlimits++;
904 static inline void qdisc_qstats_drop(struct Qdisc *sch)
906 qstats_drop_inc(&sch->qstats);
909 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
911 this_cpu_inc(sch->cpu_qstats->drops);
914 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
916 sch->qstats.overlimits++;
919 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
921 __u32 qlen = qdisc_qlen_sum(sch);
923 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
926 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
929 struct gnet_stats_queue qstats = { 0 };
931 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
932 *qlen = qstats.qlen + qdisc_qlen(sch);
933 *backlog = qstats.backlog;
936 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
940 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
941 qdisc_tree_reduce_backlog(sch, qlen, backlog);
944 static inline void qdisc_purge_queue(struct Qdisc *sch)
948 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
950 qdisc_tree_reduce_backlog(sch, qlen, backlog);
953 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
954 struct qdisc_skb_head *qh)
956 struct sk_buff *last = qh->tail;
969 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
971 __qdisc_enqueue_tail(skb, &sch->q);
972 qdisc_qstats_backlog_inc(sch, skb);
973 return NET_XMIT_SUCCESS;
976 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
977 struct qdisc_skb_head *qh)
979 skb->next = qh->head;
987 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
989 struct sk_buff *skb = qh->head;
991 if (likely(skb != NULL)) {
992 qh->head = skb->next;
994 if (qh->head == NULL)
1002 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1004 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1006 if (likely(skb != NULL)) {
1007 qdisc_qstats_backlog_dec(sch, skb);
1008 qdisc_bstats_update(sch, skb);
1014 /* Instead of calling kfree_skb() while root qdisc lock is held,
1015 * queue the skb for future freeing at end of __dev_xmit_skb()
1017 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1019 skb->next = *to_free;
1023 static inline void __qdisc_drop_all(struct sk_buff *skb,
1024 struct sk_buff **to_free)
1027 skb->prev->next = *to_free;
1029 skb->next = *to_free;
1033 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1034 struct qdisc_skb_head *qh,
1035 struct sk_buff **to_free)
1037 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1039 if (likely(skb != NULL)) {
1040 unsigned int len = qdisc_pkt_len(skb);
1042 qdisc_qstats_backlog_dec(sch, skb);
1043 __qdisc_drop(skb, to_free);
1050 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1052 const struct qdisc_skb_head *qh = &sch->q;
1057 /* generic pseudo peek method for non-work-conserving qdisc */
1058 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1060 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1062 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1064 skb = sch->dequeue(sch);
1067 __skb_queue_head(&sch->gso_skb, skb);
1068 /* it's still part of the queue */
1069 qdisc_qstats_backlog_inc(sch, skb);
1077 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1078 struct sk_buff *skb)
1080 if (qdisc_is_percpu_stats(sch)) {
1081 qdisc_qstats_cpu_backlog_dec(sch, skb);
1082 qdisc_bstats_cpu_update(sch, skb);
1083 qdisc_qstats_cpu_qlen_dec(sch);
1085 qdisc_qstats_backlog_dec(sch, skb);
1086 qdisc_bstats_update(sch, skb);
1091 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1092 unsigned int pkt_len)
1094 if (qdisc_is_percpu_stats(sch)) {
1095 qdisc_qstats_cpu_qlen_inc(sch);
1096 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1098 sch->qstats.backlog += pkt_len;
1103 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1104 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1106 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1109 skb = __skb_dequeue(&sch->gso_skb);
1110 if (qdisc_is_percpu_stats(sch)) {
1111 qdisc_qstats_cpu_backlog_dec(sch, skb);
1112 qdisc_qstats_cpu_qlen_dec(sch);
1114 qdisc_qstats_backlog_dec(sch, skb);
1118 skb = sch->dequeue(sch);
1124 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1127 * We do not know the backlog in bytes of this list, it
1128 * is up to the caller to correct it
1132 rtnl_kfree_skbs(qh->head, qh->tail);
1140 static inline void qdisc_reset_queue(struct Qdisc *sch)
1142 __qdisc_reset_queue(&sch->q);
1145 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1146 struct Qdisc **pold)
1154 qdisc_purge_queue(old);
1155 sch_tree_unlock(sch);
1160 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1162 rtnl_kfree_skbs(skb, skb);
1163 qdisc_qstats_drop(sch);
1166 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1167 struct sk_buff **to_free)
1169 __qdisc_drop(skb, to_free);
1170 qdisc_qstats_cpu_drop(sch);
1172 return NET_XMIT_DROP;
1175 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1176 struct sk_buff **to_free)
1178 __qdisc_drop(skb, to_free);
1179 qdisc_qstats_drop(sch);
1181 return NET_XMIT_DROP;
1184 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1185 struct sk_buff **to_free)
1187 __qdisc_drop_all(skb, to_free);
1188 qdisc_qstats_drop(sch);
1190 return NET_XMIT_DROP;
1193 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1194 long it will take to send a packet given its size.
1196 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1198 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1201 slot >>= rtab->rate.cell_log;
1203 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1204 return rtab->data[slot];
1207 struct psched_ratecfg {
1208 u64 rate_bytes_ps; /* bytes per second */
1216 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1224 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1225 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1227 return ((u64)len * r->mult) >> r->shift;
1230 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1231 const struct tc_ratespec *conf,
1234 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1235 const struct psched_ratecfg *r)
1237 memset(res, 0, sizeof(*res));
1239 /* legacy struct tc_ratespec has a 32bit @rate field
1240 * Qdisc using 64bit rate should add new attributes
1241 * in order to maintain compatibility.
1243 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1245 res->overhead = r->overhead;
1247 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1250 struct psched_pktrate {
1251 u64 rate_pkts_ps; /* packets per second */
1256 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1257 unsigned int pkt_num)
1259 return ((u64)pkt_num * r->mult) >> r->shift;
1262 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1264 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1265 * The fast path only needs to access filter list and to update stats
1268 struct tcf_proto *filter_list;
1269 struct tcf_block *block;
1270 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1271 struct gnet_stats_queue __percpu *cpu_qstats;
1272 unsigned long rcu_state;
1275 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1276 const struct sk_buff *skb)
1278 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1281 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1283 this_cpu_inc(miniq->cpu_qstats->drops);
1286 struct mini_Qdisc_pair {
1287 struct mini_Qdisc miniq1;
1288 struct mini_Qdisc miniq2;
1289 struct mini_Qdisc __rcu **p_miniq;
1292 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1293 struct tcf_proto *tp_head);
1294 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1295 struct mini_Qdisc __rcu **p_miniq);
1296 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1297 struct tcf_block *block);
1299 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1301 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1303 /* Make sure qdisc is no longer in SCHED state. */
1304 static inline void qdisc_synchronize(const struct Qdisc *q)
1306 while (test_bit(__QDISC_STATE_SCHED, &q->state))