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,
43 #define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED)
44 #define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING)
46 #define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \
49 struct qdisc_size_table {
51 struct list_head list;
52 struct tc_sizespec szopts;
57 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
58 struct qdisc_skb_head {
66 int (*enqueue)(struct sk_buff *skb,
68 struct sk_buff **to_free);
69 struct sk_buff * (*dequeue)(struct Qdisc *sch);
71 #define TCQ_F_BUILTIN 1
72 #define TCQ_F_INGRESS 2
73 #define TCQ_F_CAN_BYPASS 4
74 #define TCQ_F_MQROOT 8
75 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
76 * q->dev_queue : It can test
77 * netif_xmit_frozen_or_stopped() before
78 * dequeueing next packet.
79 * Its true for MQ/MQPRIO slaves, or non
82 #define TCQ_F_WARN_NONWC (1 << 16)
83 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
84 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
85 * qdisc_tree_decrease_qlen() should stop.
87 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
88 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
89 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
91 const struct Qdisc_ops *ops;
92 struct qdisc_size_table __rcu *stab;
93 struct hlist_node hash;
97 struct netdev_queue *dev_queue;
99 struct net_rate_estimator __rcu *rate_est;
100 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
101 struct gnet_stats_queue __percpu *cpu_qstats;
106 * For performance sake on SMP, we put highly modified fields at the end
108 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
109 struct qdisc_skb_head q;
110 struct gnet_stats_basic_packed bstats;
112 struct gnet_stats_queue qstats;
114 struct Qdisc *next_sched;
115 struct sk_buff_head skb_bad_txq;
117 spinlock_t busylock ____cacheline_aligned_in_smp;
123 long privdata[] ____cacheline_aligned;
126 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
128 if (qdisc->flags & TCQ_F_BUILTIN)
130 refcount_inc(&qdisc->refcnt);
133 /* Intended to be used by unlocked users, when concurrent qdisc release is
137 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
139 if (qdisc->flags & TCQ_F_BUILTIN)
141 if (refcount_inc_not_zero(&qdisc->refcnt))
146 static inline bool qdisc_is_running(struct Qdisc *qdisc)
148 if (qdisc->flags & TCQ_F_NOLOCK)
149 return spin_is_locked(&qdisc->seqlock);
150 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
153 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
155 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
158 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
160 return q->flags & TCQ_F_CPUSTATS;
163 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
165 if (qdisc_is_percpu_stats(qdisc))
166 return nolock_qdisc_is_empty(qdisc);
167 return !READ_ONCE(qdisc->q.qlen);
170 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
172 if (qdisc->flags & TCQ_F_NOLOCK) {
173 if (spin_trylock(&qdisc->seqlock))
176 /* No need to insist if the MISSED flag was already set.
177 * Note that test_and_set_bit() also gives us memory ordering
178 * guarantees wrt potential earlier enqueue() and below
179 * spin_trylock(), both of which are necessary to prevent races
181 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
184 /* Try to take the lock again to make sure that we will either
185 * grab it or the CPU that still has it will see MISSED set
186 * when testing it in qdisc_run_end()
188 return spin_trylock(&qdisc->seqlock);
189 } else if (qdisc_is_running(qdisc)) {
192 /* Variant of write_seqcount_begin() telling lockdep a trylock
195 raw_write_seqcount_begin(&qdisc->running);
196 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
200 static inline void qdisc_run_end(struct Qdisc *qdisc)
202 if (qdisc->flags & TCQ_F_NOLOCK) {
203 spin_unlock(&qdisc->seqlock);
205 /* spin_unlock() only has store-release semantic. The unlock
206 * and test_bit() ordering is a store-load ordering, so a full
207 * memory barrier is needed here.
211 if (unlikely(test_bit(__QDISC_STATE_MISSED,
213 __netif_schedule(qdisc);
215 write_seqcount_end(&qdisc->running);
219 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
221 return qdisc->flags & TCQ_F_ONETXQUEUE;
224 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
227 /* Non-BQL migrated drivers will return 0, too. */
228 return dql_avail(&txq->dql);
234 struct Qdisc_class_ops {
236 /* Child qdisc manipulation */
237 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
238 int (*graft)(struct Qdisc *, unsigned long cl,
239 struct Qdisc *, struct Qdisc **,
240 struct netlink_ext_ack *extack);
241 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
242 void (*qlen_notify)(struct Qdisc *, unsigned long);
244 /* Class manipulation routines */
245 unsigned long (*find)(struct Qdisc *, u32 classid);
246 int (*change)(struct Qdisc *, u32, u32,
247 struct nlattr **, unsigned long *,
248 struct netlink_ext_ack *);
249 int (*delete)(struct Qdisc *, unsigned long,
250 struct netlink_ext_ack *);
251 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
253 /* Filter manipulation */
254 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
256 struct netlink_ext_ack *extack);
257 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
259 void (*unbind_tcf)(struct Qdisc *, unsigned long);
261 /* rtnetlink specific */
262 int (*dump)(struct Qdisc *, unsigned long,
263 struct sk_buff *skb, struct tcmsg*);
264 int (*dump_stats)(struct Qdisc *, unsigned long,
268 /* Qdisc_class_ops flag values */
270 /* Implements API that doesn't require rtnl lock */
271 enum qdisc_class_ops_flags {
272 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
276 struct Qdisc_ops *next;
277 const struct Qdisc_class_ops *cl_ops;
280 unsigned int static_flags;
282 int (*enqueue)(struct sk_buff *skb,
284 struct sk_buff **to_free);
285 struct sk_buff * (*dequeue)(struct Qdisc *);
286 struct sk_buff * (*peek)(struct Qdisc *);
288 int (*init)(struct Qdisc *sch, struct nlattr *arg,
289 struct netlink_ext_ack *extack);
290 void (*reset)(struct Qdisc *);
291 void (*destroy)(struct Qdisc *);
292 int (*change)(struct Qdisc *sch,
294 struct netlink_ext_ack *extack);
295 void (*attach)(struct Qdisc *sch);
296 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
297 void (*change_real_num_tx)(struct Qdisc *sch,
298 unsigned int new_real_tx);
300 int (*dump)(struct Qdisc *, struct sk_buff *);
301 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
303 void (*ingress_block_set)(struct Qdisc *sch,
305 void (*egress_block_set)(struct Qdisc *sch,
307 u32 (*ingress_block_get)(struct Qdisc *sch);
308 u32 (*egress_block_get)(struct Qdisc *sch);
310 struct module *owner;
320 const struct tcf_proto *goto_tp;
322 /* used in the skb_tc_reinsert function */
325 struct gnet_stats_queue *qstats;
332 struct tcf_proto_ops {
333 struct list_head head;
336 int (*classify)(struct sk_buff *,
337 const struct tcf_proto *,
338 struct tcf_result *);
339 int (*init)(struct tcf_proto*);
340 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
341 struct netlink_ext_ack *extack);
343 void* (*get)(struct tcf_proto*, u32 handle);
344 void (*put)(struct tcf_proto *tp, void *f);
345 int (*change)(struct net *net, struct sk_buff *,
346 struct tcf_proto*, unsigned long,
347 u32 handle, struct nlattr **,
349 struct netlink_ext_ack *);
350 int (*delete)(struct tcf_proto *tp, void *arg,
351 bool *last, bool rtnl_held,
352 struct netlink_ext_ack *);
353 bool (*delete_empty)(struct tcf_proto *tp);
354 void (*walk)(struct tcf_proto *tp,
355 struct tcf_walker *arg, bool rtnl_held);
356 int (*reoffload)(struct tcf_proto *tp, bool add,
357 flow_setup_cb_t *cb, void *cb_priv,
358 struct netlink_ext_ack *extack);
359 void (*hw_add)(struct tcf_proto *tp,
361 void (*hw_del)(struct tcf_proto *tp,
363 void (*bind_class)(void *, u32, unsigned long,
364 void *, unsigned long);
365 void * (*tmplt_create)(struct net *net,
366 struct tcf_chain *chain,
368 struct netlink_ext_ack *extack);
369 void (*tmplt_destroy)(void *tmplt_priv);
371 /* rtnetlink specific */
372 int (*dump)(struct net*, struct tcf_proto*, void *,
373 struct sk_buff *skb, struct tcmsg*,
375 int (*terse_dump)(struct net *net,
376 struct tcf_proto *tp, void *fh,
378 struct tcmsg *t, bool rtnl_held);
379 int (*tmplt_dump)(struct sk_buff *skb,
383 struct module *owner;
387 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
388 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
389 * conditions can occur when filters are inserted/deleted simultaneously.
391 enum tcf_proto_ops_flags {
392 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
396 /* Fast access part */
397 struct tcf_proto __rcu *next;
400 /* called under RCU BH lock*/
401 int (*classify)(struct sk_buff *,
402 const struct tcf_proto *,
403 struct tcf_result *);
409 const struct tcf_proto_ops *ops;
410 struct tcf_chain *chain;
411 /* Lock protects tcf_proto shared state and can be used by unlocked
412 * classifiers to protect their private data.
418 struct hlist_node destroy_ht_node;
421 struct qdisc_skb_cb {
423 unsigned int pkt_len;
424 u16 slave_dev_queue_mapping;
427 #define QDISC_CB_PRIV_LEN 20
428 unsigned char data[QDISC_CB_PRIV_LEN];
431 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
434 /* Protects filter_chain. */
435 struct mutex filter_chain_lock;
436 struct tcf_proto __rcu *filter_chain;
437 struct list_head list;
438 struct tcf_block *block;
439 u32 index; /* chain index */
441 unsigned int action_refcnt;
442 bool explicitly_created;
444 const struct tcf_proto_ops *tmplt_ops;
450 /* Lock protects tcf_block and lifetime-management data of chains
451 * attached to the block (refcnt, action_refcnt, explicitly_created).
454 struct list_head chain_list;
455 u32 index; /* block index for shared blocks */
456 u32 classid; /* which class this block belongs to */
460 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
461 struct flow_block flow_block;
462 struct list_head owner_list;
464 atomic_t offloadcnt; /* Number of oddloaded filters */
465 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
466 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
468 struct tcf_chain *chain;
469 struct list_head filter_chain_list;
472 DECLARE_HASHTABLE(proto_destroy_ht, 7);
473 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
476 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
478 return lockdep_is_held(&chain->filter_chain_lock);
481 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
483 return lockdep_is_held(&tp->lock);
486 #define tcf_chain_dereference(p, chain) \
487 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
489 #define tcf_proto_dereference(p, tp) \
490 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
492 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
494 struct qdisc_skb_cb *qcb;
496 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
497 BUILD_BUG_ON(sizeof(qcb->data) < sz);
500 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
502 return this_cpu_ptr(q->cpu_qstats)->qlen;
505 static inline int qdisc_qlen(const struct Qdisc *q)
510 static inline int qdisc_qlen_sum(const struct Qdisc *q)
512 __u32 qlen = q->qstats.qlen;
515 if (qdisc_is_percpu_stats(q)) {
516 for_each_possible_cpu(i)
517 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
525 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
527 return (struct qdisc_skb_cb *)skb->cb;
530 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
532 return &qdisc->q.lock;
535 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
537 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
542 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
544 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
547 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
549 return qdisc->dev_queue->qdisc_sleeping;
552 /* The qdisc root lock is a mechanism by which to top level
553 * of a qdisc tree can be locked from any qdisc node in the
554 * forest. This allows changing the configuration of some
555 * aspect of the qdisc tree while blocking out asynchronous
556 * qdisc access in the packet processing paths.
558 * It is only legal to do this when the root will not change
559 * on us. Otherwise we'll potentially lock the wrong qdisc
560 * root. This is enforced by holding the RTNL semaphore, which
561 * all users of this lock accessor must do.
563 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
565 struct Qdisc *root = qdisc_root(qdisc);
568 return qdisc_lock(root);
571 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
573 struct Qdisc *root = qdisc_root_sleeping(qdisc);
576 return qdisc_lock(root);
579 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
581 struct Qdisc *root = qdisc_root_sleeping(qdisc);
584 return &root->running;
587 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
589 return qdisc->dev_queue->dev;
592 static inline void sch_tree_lock(struct Qdisc *q)
594 if (q->flags & TCQ_F_MQROOT)
595 spin_lock_bh(qdisc_lock(q));
597 spin_lock_bh(qdisc_root_sleeping_lock(q));
600 static inline void sch_tree_unlock(struct Qdisc *q)
602 if (q->flags & TCQ_F_MQROOT)
603 spin_unlock_bh(qdisc_lock(q));
605 spin_unlock_bh(qdisc_root_sleeping_lock(q));
608 extern struct Qdisc noop_qdisc;
609 extern struct Qdisc_ops noop_qdisc_ops;
610 extern struct Qdisc_ops pfifo_fast_ops;
611 extern struct Qdisc_ops mq_qdisc_ops;
612 extern struct Qdisc_ops noqueue_qdisc_ops;
613 extern const struct Qdisc_ops *default_qdisc_ops;
614 static inline const struct Qdisc_ops *
615 get_default_qdisc_ops(const struct net_device *dev, int ntx)
617 return ntx < dev->real_num_tx_queues ?
618 default_qdisc_ops : &pfifo_fast_ops;
621 struct Qdisc_class_common {
623 struct hlist_node hnode;
626 struct Qdisc_class_hash {
627 struct hlist_head *hash;
628 unsigned int hashsize;
629 unsigned int hashmask;
630 unsigned int hashelems;
633 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
640 static inline struct Qdisc_class_common *
641 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
643 struct Qdisc_class_common *cl;
649 h = qdisc_class_hash(id, hash->hashmask);
650 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
651 if (cl->classid == id)
657 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
659 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
661 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
664 int qdisc_class_hash_init(struct Qdisc_class_hash *);
665 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
666 struct Qdisc_class_common *);
667 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
668 struct Qdisc_class_common *);
669 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
670 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
672 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
673 void dev_qdisc_change_real_num_tx(struct net_device *dev,
674 unsigned int new_real_tx);
675 void dev_init_scheduler(struct net_device *dev);
676 void dev_shutdown(struct net_device *dev);
677 void dev_activate(struct net_device *dev);
678 void dev_deactivate(struct net_device *dev);
679 void dev_deactivate_many(struct list_head *head);
680 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
681 struct Qdisc *qdisc);
682 void qdisc_reset(struct Qdisc *qdisc);
683 void qdisc_put(struct Qdisc *qdisc);
684 void qdisc_put_unlocked(struct Qdisc *qdisc);
685 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
686 #ifdef CONFIG_NET_SCHED
687 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
689 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
690 struct Qdisc *new, struct Qdisc *old,
691 enum tc_setup_type type, void *type_data,
692 struct netlink_ext_ack *extack);
695 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
698 q->flags &= ~TCQ_F_OFFLOADED;
703 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
704 struct Qdisc *new, struct Qdisc *old,
705 enum tc_setup_type type, void *type_data,
706 struct netlink_ext_ack *extack)
710 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
711 const struct Qdisc_ops *ops,
712 struct netlink_ext_ack *extack);
713 void qdisc_free(struct Qdisc *qdisc);
714 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
715 const struct Qdisc_ops *ops, u32 parentid,
716 struct netlink_ext_ack *extack);
717 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
718 const struct qdisc_size_table *stab);
719 int skb_do_redirect(struct sk_buff *);
721 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
723 #ifdef CONFIG_NET_CLS_ACT
724 return skb->tc_at_ingress;
730 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
732 #ifdef CONFIG_NET_CLS_ACT
733 if (skb->tc_skip_classify) {
734 skb->tc_skip_classify = 0;
741 /* Reset all TX qdiscs greater than index of a device. */
742 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
746 for (; i < dev->num_tx_queues; i++) {
747 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
749 spin_lock_bh(qdisc_lock(qdisc));
751 spin_unlock_bh(qdisc_lock(qdisc));
756 /* Are all TX queues of the device empty? */
757 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
762 for (i = 0; i < dev->num_tx_queues; i++) {
763 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
764 const struct Qdisc *q = rcu_dereference(txq->qdisc);
766 if (!qdisc_is_empty(q)) {
775 /* Are any of the TX qdiscs changing? */
776 static inline bool qdisc_tx_changing(const struct net_device *dev)
780 for (i = 0; i < dev->num_tx_queues; i++) {
781 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
782 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
788 /* Is the device using the noop qdisc on all queues? */
789 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
793 for (i = 0; i < dev->num_tx_queues; i++) {
794 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
795 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
801 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
803 return qdisc_skb_cb(skb)->pkt_len;
806 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
807 enum net_xmit_qdisc_t {
808 __NET_XMIT_STOLEN = 0x00010000,
809 __NET_XMIT_BYPASS = 0x00020000,
812 #ifdef CONFIG_NET_CLS_ACT
813 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
815 #define net_xmit_drop_count(e) (1)
818 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
819 const struct Qdisc *sch)
821 #ifdef CONFIG_NET_SCHED
822 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
825 __qdisc_calculate_pkt_len(skb, stab);
829 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
830 struct sk_buff **to_free)
832 qdisc_calculate_pkt_len(skb, sch);
833 return sch->enqueue(skb, sch, to_free);
836 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
837 __u64 bytes, __u32 packets)
839 bstats->bytes += bytes;
840 bstats->packets += packets;
843 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
844 const struct sk_buff *skb)
846 _bstats_update(bstats,
848 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
851 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
852 __u64 bytes, __u32 packets)
854 u64_stats_update_begin(&bstats->syncp);
855 _bstats_update(&bstats->bstats, bytes, packets);
856 u64_stats_update_end(&bstats->syncp);
859 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
860 const struct sk_buff *skb)
862 u64_stats_update_begin(&bstats->syncp);
863 bstats_update(&bstats->bstats, skb);
864 u64_stats_update_end(&bstats->syncp);
867 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
868 const struct sk_buff *skb)
870 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
873 static inline void qdisc_bstats_update(struct Qdisc *sch,
874 const struct sk_buff *skb)
876 bstats_update(&sch->bstats, skb);
879 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
880 const struct sk_buff *skb)
882 sch->qstats.backlog -= qdisc_pkt_len(skb);
885 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
886 const struct sk_buff *skb)
888 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
891 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
892 const struct sk_buff *skb)
894 sch->qstats.backlog += qdisc_pkt_len(skb);
897 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
898 const struct sk_buff *skb)
900 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
903 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
905 this_cpu_inc(sch->cpu_qstats->qlen);
908 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
910 this_cpu_dec(sch->cpu_qstats->qlen);
913 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
915 this_cpu_inc(sch->cpu_qstats->requeues);
918 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
920 sch->qstats.drops += count;
923 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
928 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
930 qstats->overlimits++;
933 static inline void qdisc_qstats_drop(struct Qdisc *sch)
935 qstats_drop_inc(&sch->qstats);
938 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
940 this_cpu_inc(sch->cpu_qstats->drops);
943 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
945 sch->qstats.overlimits++;
948 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
950 __u32 qlen = qdisc_qlen_sum(sch);
952 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
955 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
958 struct gnet_stats_queue qstats = { 0 };
959 __u32 len = qdisc_qlen_sum(sch);
961 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
963 *backlog = qstats.backlog;
966 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
970 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
971 qdisc_tree_reduce_backlog(sch, qlen, backlog);
974 static inline void qdisc_purge_queue(struct Qdisc *sch)
978 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
980 qdisc_tree_reduce_backlog(sch, qlen, backlog);
983 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
990 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
991 struct qdisc_skb_head *qh)
993 struct sk_buff *last = qh->tail;
1006 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1008 __qdisc_enqueue_tail(skb, &sch->q);
1009 qdisc_qstats_backlog_inc(sch, skb);
1010 return NET_XMIT_SUCCESS;
1013 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1014 struct qdisc_skb_head *qh)
1016 skb->next = qh->head;
1024 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1026 struct sk_buff *skb = qh->head;
1028 if (likely(skb != NULL)) {
1029 qh->head = skb->next;
1031 if (qh->head == NULL)
1039 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1041 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1043 if (likely(skb != NULL)) {
1044 qdisc_qstats_backlog_dec(sch, skb);
1045 qdisc_bstats_update(sch, skb);
1051 /* Instead of calling kfree_skb() while root qdisc lock is held,
1052 * queue the skb for future freeing at end of __dev_xmit_skb()
1054 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1056 skb->next = *to_free;
1060 static inline void __qdisc_drop_all(struct sk_buff *skb,
1061 struct sk_buff **to_free)
1064 skb->prev->next = *to_free;
1066 skb->next = *to_free;
1070 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1071 struct qdisc_skb_head *qh,
1072 struct sk_buff **to_free)
1074 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1076 if (likely(skb != NULL)) {
1077 unsigned int len = qdisc_pkt_len(skb);
1079 qdisc_qstats_backlog_dec(sch, skb);
1080 __qdisc_drop(skb, to_free);
1087 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1089 const struct qdisc_skb_head *qh = &sch->q;
1094 /* generic pseudo peek method for non-work-conserving qdisc */
1095 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1097 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1099 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1101 skb = sch->dequeue(sch);
1104 __skb_queue_head(&sch->gso_skb, skb);
1105 /* it's still part of the queue */
1106 qdisc_qstats_backlog_inc(sch, skb);
1114 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1115 struct sk_buff *skb)
1117 if (qdisc_is_percpu_stats(sch)) {
1118 qdisc_qstats_cpu_backlog_dec(sch, skb);
1119 qdisc_bstats_cpu_update(sch, skb);
1120 qdisc_qstats_cpu_qlen_dec(sch);
1122 qdisc_qstats_backlog_dec(sch, skb);
1123 qdisc_bstats_update(sch, skb);
1128 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1129 unsigned int pkt_len)
1131 if (qdisc_is_percpu_stats(sch)) {
1132 qdisc_qstats_cpu_qlen_inc(sch);
1133 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1135 sch->qstats.backlog += pkt_len;
1140 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1141 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1143 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1146 skb = __skb_dequeue(&sch->gso_skb);
1147 if (qdisc_is_percpu_stats(sch)) {
1148 qdisc_qstats_cpu_backlog_dec(sch, skb);
1149 qdisc_qstats_cpu_qlen_dec(sch);
1151 qdisc_qstats_backlog_dec(sch, skb);
1155 skb = sch->dequeue(sch);
1161 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1164 * We do not know the backlog in bytes of this list, it
1165 * is up to the caller to correct it
1169 rtnl_kfree_skbs(qh->head, qh->tail);
1177 static inline void qdisc_reset_queue(struct Qdisc *sch)
1179 __qdisc_reset_queue(&sch->q);
1182 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1183 struct Qdisc **pold)
1191 qdisc_purge_queue(old);
1192 sch_tree_unlock(sch);
1197 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1199 rtnl_kfree_skbs(skb, skb);
1200 qdisc_qstats_drop(sch);
1203 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1204 struct sk_buff **to_free)
1206 __qdisc_drop(skb, to_free);
1207 qdisc_qstats_cpu_drop(sch);
1209 return NET_XMIT_DROP;
1212 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1213 struct sk_buff **to_free)
1215 __qdisc_drop(skb, to_free);
1216 qdisc_qstats_drop(sch);
1218 return NET_XMIT_DROP;
1221 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1222 struct sk_buff **to_free)
1224 __qdisc_drop_all(skb, to_free);
1225 qdisc_qstats_drop(sch);
1227 return NET_XMIT_DROP;
1230 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1231 long it will take to send a packet given its size.
1233 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1235 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1238 slot >>= rtab->rate.cell_log;
1240 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1241 return rtab->data[slot];
1244 struct psched_ratecfg {
1245 u64 rate_bytes_ps; /* bytes per second */
1253 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1261 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1262 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1264 return ((u64)len * r->mult) >> r->shift;
1267 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1268 const struct tc_ratespec *conf,
1271 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1272 const struct psched_ratecfg *r)
1274 memset(res, 0, sizeof(*res));
1276 /* legacy struct tc_ratespec has a 32bit @rate field
1277 * Qdisc using 64bit rate should add new attributes
1278 * in order to maintain compatibility.
1280 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1282 res->overhead = r->overhead;
1284 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1287 struct psched_pktrate {
1288 u64 rate_pkts_ps; /* packets per second */
1293 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1294 unsigned int pkt_num)
1296 return ((u64)pkt_num * r->mult) >> r->shift;
1299 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1301 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1302 * The fast path only needs to access filter list and to update stats
1305 struct tcf_proto *filter_list;
1306 struct tcf_block *block;
1307 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1308 struct gnet_stats_queue __percpu *cpu_qstats;
1309 struct rcu_head rcu;
1312 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1313 const struct sk_buff *skb)
1315 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1318 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1320 this_cpu_inc(miniq->cpu_qstats->drops);
1323 struct mini_Qdisc_pair {
1324 struct mini_Qdisc miniq1;
1325 struct mini_Qdisc miniq2;
1326 struct mini_Qdisc __rcu **p_miniq;
1329 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1330 struct tcf_proto *tp_head);
1331 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1332 struct mini_Qdisc __rcu **p_miniq);
1333 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1334 struct tcf_block *block);
1336 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1338 /* Make sure qdisc is no longer in SCHED state. */
1339 static inline void qdisc_synchronize(const struct Qdisc *q)
1341 while (test_bit(__QDISC_STATE_SCHED, &q->state))