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 /* Paired with smp_mb__after_atomic() to make sure
167 * STATE_MISSED checking is synchronized with clearing
168 * in pfifo_fast_dequeue().
170 smp_mb__before_atomic();
172 /* If the MISSED flag is set, it means other thread has
173 * set the MISSED flag before second spin_trylock(), so
174 * we can return false here to avoid multi cpus doing
175 * the set_bit() and second spin_trylock() concurrently.
177 if (test_bit(__QDISC_STATE_MISSED, &qdisc->state))
180 /* Set the MISSED flag before the second spin_trylock(),
181 * if the second spin_trylock() return false, it means
182 * other cpu holding the lock will do dequeuing for us
183 * or it will see the MISSED flag set after releasing
184 * lock and reschedule the net_tx_action() to do the
187 set_bit(__QDISC_STATE_MISSED, &qdisc->state);
189 /* spin_trylock() only has load-acquire semantic, so use
190 * smp_mb__after_atomic() to ensure STATE_MISSED is set
191 * before doing the second spin_trylock().
193 smp_mb__after_atomic();
195 /* Retry again in case other CPU may not see the new flag
196 * after it releases the lock at the end of qdisc_run_end().
198 if (!spin_trylock(&qdisc->seqlock))
202 WRITE_ONCE(qdisc->empty, false);
203 } else if (qdisc_is_running(qdisc)) {
206 /* Variant of write_seqcount_begin() telling lockdep a trylock
209 raw_write_seqcount_begin(&qdisc->running);
210 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
214 static inline void qdisc_run_end(struct Qdisc *qdisc)
216 write_seqcount_end(&qdisc->running);
217 if (qdisc->flags & TCQ_F_NOLOCK) {
218 spin_unlock(&qdisc->seqlock);
220 if (unlikely(test_bit(__QDISC_STATE_MISSED,
222 clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
223 __netif_schedule(qdisc);
228 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
230 return qdisc->flags & TCQ_F_ONETXQUEUE;
233 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
236 /* Non-BQL migrated drivers will return 0, too. */
237 return dql_avail(&txq->dql);
243 struct Qdisc_class_ops {
245 /* Child qdisc manipulation */
246 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
247 int (*graft)(struct Qdisc *, unsigned long cl,
248 struct Qdisc *, struct Qdisc **,
249 struct netlink_ext_ack *extack);
250 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
251 void (*qlen_notify)(struct Qdisc *, unsigned long);
253 /* Class manipulation routines */
254 unsigned long (*find)(struct Qdisc *, u32 classid);
255 int (*change)(struct Qdisc *, u32, u32,
256 struct nlattr **, unsigned long *,
257 struct netlink_ext_ack *);
258 int (*delete)(struct Qdisc *, unsigned long,
259 struct netlink_ext_ack *);
260 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
262 /* Filter manipulation */
263 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
265 struct netlink_ext_ack *extack);
266 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
268 void (*unbind_tcf)(struct Qdisc *, unsigned long);
270 /* rtnetlink specific */
271 int (*dump)(struct Qdisc *, unsigned long,
272 struct sk_buff *skb, struct tcmsg*);
273 int (*dump_stats)(struct Qdisc *, unsigned long,
277 /* Qdisc_class_ops flag values */
279 /* Implements API that doesn't require rtnl lock */
280 enum qdisc_class_ops_flags {
281 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
285 struct Qdisc_ops *next;
286 const struct Qdisc_class_ops *cl_ops;
289 unsigned int static_flags;
291 int (*enqueue)(struct sk_buff *skb,
293 struct sk_buff **to_free);
294 struct sk_buff * (*dequeue)(struct Qdisc *);
295 struct sk_buff * (*peek)(struct Qdisc *);
297 int (*init)(struct Qdisc *sch, struct nlattr *arg,
298 struct netlink_ext_ack *extack);
299 void (*reset)(struct Qdisc *);
300 void (*destroy)(struct Qdisc *);
301 int (*change)(struct Qdisc *sch,
303 struct netlink_ext_ack *extack);
304 void (*attach)(struct Qdisc *sch);
305 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
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;
329 /* used in the skb_tc_reinsert function */
332 struct gnet_stats_queue *qstats;
339 struct tcf_proto_ops {
340 struct list_head head;
343 int (*classify)(struct sk_buff *,
344 const struct tcf_proto *,
345 struct tcf_result *);
346 int (*init)(struct tcf_proto*);
347 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
348 struct netlink_ext_ack *extack);
350 void* (*get)(struct tcf_proto*, u32 handle);
351 void (*put)(struct tcf_proto *tp, void *f);
352 int (*change)(struct net *net, struct sk_buff *,
353 struct tcf_proto*, unsigned long,
354 u32 handle, struct nlattr **,
356 struct netlink_ext_ack *);
357 int (*delete)(struct tcf_proto *tp, void *arg,
358 bool *last, bool rtnl_held,
359 struct netlink_ext_ack *);
360 bool (*delete_empty)(struct tcf_proto *tp);
361 void (*walk)(struct tcf_proto *tp,
362 struct tcf_walker *arg, bool rtnl_held);
363 int (*reoffload)(struct tcf_proto *tp, bool add,
364 flow_setup_cb_t *cb, void *cb_priv,
365 struct netlink_ext_ack *extack);
366 void (*hw_add)(struct tcf_proto *tp,
368 void (*hw_del)(struct tcf_proto *tp,
370 void (*bind_class)(void *, u32, unsigned long,
371 void *, unsigned long);
372 void * (*tmplt_create)(struct net *net,
373 struct tcf_chain *chain,
375 struct netlink_ext_ack *extack);
376 void (*tmplt_destroy)(void *tmplt_priv);
378 /* rtnetlink specific */
379 int (*dump)(struct net*, struct tcf_proto*, void *,
380 struct sk_buff *skb, struct tcmsg*,
382 int (*terse_dump)(struct net *net,
383 struct tcf_proto *tp, void *fh,
385 struct tcmsg *t, bool rtnl_held);
386 int (*tmplt_dump)(struct sk_buff *skb,
390 struct module *owner;
394 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
395 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
396 * conditions can occur when filters are inserted/deleted simultaneously.
398 enum tcf_proto_ops_flags {
399 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
403 /* Fast access part */
404 struct tcf_proto __rcu *next;
407 /* called under RCU BH lock*/
408 int (*classify)(struct sk_buff *,
409 const struct tcf_proto *,
410 struct tcf_result *);
416 const struct tcf_proto_ops *ops;
417 struct tcf_chain *chain;
418 /* Lock protects tcf_proto shared state and can be used by unlocked
419 * classifiers to protect their private data.
425 struct hlist_node destroy_ht_node;
428 struct qdisc_skb_cb {
430 unsigned int pkt_len;
431 u16 slave_dev_queue_mapping;
434 #define QDISC_CB_PRIV_LEN 20
435 unsigned char data[QDISC_CB_PRIV_LEN];
440 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
443 /* Protects filter_chain. */
444 struct mutex filter_chain_lock;
445 struct tcf_proto __rcu *filter_chain;
446 struct list_head list;
447 struct tcf_block *block;
448 u32 index; /* chain index */
450 unsigned int action_refcnt;
451 bool explicitly_created;
453 const struct tcf_proto_ops *tmplt_ops;
459 /* Lock protects tcf_block and lifetime-management data of chains
460 * attached to the block (refcnt, action_refcnt, explicitly_created).
463 struct list_head chain_list;
464 u32 index; /* block index for shared blocks */
465 u32 classid; /* which class this block belongs to */
469 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
470 struct flow_block flow_block;
471 struct list_head owner_list;
473 atomic_t offloadcnt; /* Number of oddloaded filters */
474 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
475 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
477 struct tcf_chain *chain;
478 struct list_head filter_chain_list;
481 DECLARE_HASHTABLE(proto_destroy_ht, 7);
482 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
485 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
487 return lockdep_is_held(&chain->filter_chain_lock);
490 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
492 return lockdep_is_held(&tp->lock);
495 #define tcf_chain_dereference(p, chain) \
496 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
498 #define tcf_proto_dereference(p, tp) \
499 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
501 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
503 struct qdisc_skb_cb *qcb;
505 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
506 BUILD_BUG_ON(sizeof(qcb->data) < sz);
509 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
511 return this_cpu_ptr(q->cpu_qstats)->qlen;
514 static inline int qdisc_qlen(const struct Qdisc *q)
519 static inline int qdisc_qlen_sum(const struct Qdisc *q)
521 __u32 qlen = q->qstats.qlen;
524 if (qdisc_is_percpu_stats(q)) {
525 for_each_possible_cpu(i)
526 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
534 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
536 return (struct qdisc_skb_cb *)skb->cb;
539 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
541 return &qdisc->q.lock;
544 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
546 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
551 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
553 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
556 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
558 return qdisc->dev_queue->qdisc_sleeping;
561 /* The qdisc root lock is a mechanism by which to top level
562 * of a qdisc tree can be locked from any qdisc node in the
563 * forest. This allows changing the configuration of some
564 * aspect of the qdisc tree while blocking out asynchronous
565 * qdisc access in the packet processing paths.
567 * It is only legal to do this when the root will not change
568 * on us. Otherwise we'll potentially lock the wrong qdisc
569 * root. This is enforced by holding the RTNL semaphore, which
570 * all users of this lock accessor must do.
572 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
574 struct Qdisc *root = qdisc_root(qdisc);
577 return qdisc_lock(root);
580 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
582 struct Qdisc *root = qdisc_root_sleeping(qdisc);
585 return qdisc_lock(root);
588 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
590 struct Qdisc *root = qdisc_root_sleeping(qdisc);
593 return &root->running;
596 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
598 return qdisc->dev_queue->dev;
601 static inline void sch_tree_lock(struct Qdisc *q)
603 if (q->flags & TCQ_F_MQROOT)
604 spin_lock_bh(qdisc_lock(q));
606 spin_lock_bh(qdisc_root_sleeping_lock(q));
609 static inline void sch_tree_unlock(struct Qdisc *q)
611 if (q->flags & TCQ_F_MQROOT)
612 spin_unlock_bh(qdisc_lock(q));
614 spin_unlock_bh(qdisc_root_sleeping_lock(q));
617 extern struct Qdisc noop_qdisc;
618 extern struct Qdisc_ops noop_qdisc_ops;
619 extern struct Qdisc_ops pfifo_fast_ops;
620 extern struct Qdisc_ops mq_qdisc_ops;
621 extern struct Qdisc_ops noqueue_qdisc_ops;
622 extern const struct Qdisc_ops *default_qdisc_ops;
623 static inline const struct Qdisc_ops *
624 get_default_qdisc_ops(const struct net_device *dev, int ntx)
626 return ntx < dev->real_num_tx_queues ?
627 default_qdisc_ops : &pfifo_fast_ops;
630 struct Qdisc_class_common {
632 struct hlist_node hnode;
635 struct Qdisc_class_hash {
636 struct hlist_head *hash;
637 unsigned int hashsize;
638 unsigned int hashmask;
639 unsigned int hashelems;
642 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
649 static inline struct Qdisc_class_common *
650 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
652 struct Qdisc_class_common *cl;
658 h = qdisc_class_hash(id, hash->hashmask);
659 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
660 if (cl->classid == id)
666 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
668 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
670 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
673 int qdisc_class_hash_init(struct Qdisc_class_hash *);
674 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
675 struct Qdisc_class_common *);
676 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
677 struct Qdisc_class_common *);
678 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
679 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
681 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
682 void dev_init_scheduler(struct net_device *dev);
683 void dev_shutdown(struct net_device *dev);
684 void dev_activate(struct net_device *dev);
685 void dev_deactivate(struct net_device *dev);
686 void dev_deactivate_many(struct list_head *head);
687 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
688 struct Qdisc *qdisc);
689 void qdisc_reset(struct Qdisc *qdisc);
690 void qdisc_put(struct Qdisc *qdisc);
691 void qdisc_put_unlocked(struct Qdisc *qdisc);
692 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
693 #ifdef CONFIG_NET_SCHED
694 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
696 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
697 struct Qdisc *new, struct Qdisc *old,
698 enum tc_setup_type type, void *type_data,
699 struct netlink_ext_ack *extack);
702 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
705 q->flags &= ~TCQ_F_OFFLOADED;
710 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
711 struct Qdisc *new, struct Qdisc *old,
712 enum tc_setup_type type, void *type_data,
713 struct netlink_ext_ack *extack)
717 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
718 const struct Qdisc_ops *ops,
719 struct netlink_ext_ack *extack);
720 void qdisc_free(struct Qdisc *qdisc);
721 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
722 const struct Qdisc_ops *ops, u32 parentid,
723 struct netlink_ext_ack *extack);
724 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
725 const struct qdisc_size_table *stab);
726 int skb_do_redirect(struct sk_buff *);
728 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
730 #ifdef CONFIG_NET_CLS_ACT
731 return skb->tc_at_ingress;
737 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
739 #ifdef CONFIG_NET_CLS_ACT
740 if (skb->tc_skip_classify) {
741 skb->tc_skip_classify = 0;
748 /* Reset all TX qdiscs greater than index of a device. */
749 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
753 for (; i < dev->num_tx_queues; i++) {
754 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
756 spin_lock_bh(qdisc_lock(qdisc));
758 spin_unlock_bh(qdisc_lock(qdisc));
763 /* Are all TX queues of the device empty? */
764 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
769 for (i = 0; i < dev->num_tx_queues; i++) {
770 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
771 const struct Qdisc *q = rcu_dereference(txq->qdisc);
773 if (!qdisc_is_empty(q)) {
782 /* Are any of the TX qdiscs changing? */
783 static inline bool qdisc_tx_changing(const struct net_device *dev)
787 for (i = 0; i < dev->num_tx_queues; i++) {
788 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
789 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
795 /* Is the device using the noop qdisc on all queues? */
796 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
800 for (i = 0; i < dev->num_tx_queues; i++) {
801 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
802 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
808 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
810 return qdisc_skb_cb(skb)->pkt_len;
813 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
814 enum net_xmit_qdisc_t {
815 __NET_XMIT_STOLEN = 0x00010000,
816 __NET_XMIT_BYPASS = 0x00020000,
819 #ifdef CONFIG_NET_CLS_ACT
820 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
822 #define net_xmit_drop_count(e) (1)
825 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
826 const struct Qdisc *sch)
828 #ifdef CONFIG_NET_SCHED
829 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
832 __qdisc_calculate_pkt_len(skb, stab);
836 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
837 struct sk_buff **to_free)
839 qdisc_calculate_pkt_len(skb, sch);
840 return sch->enqueue(skb, sch, to_free);
843 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
844 __u64 bytes, __u32 packets)
846 bstats->bytes += bytes;
847 bstats->packets += packets;
850 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
851 const struct sk_buff *skb)
853 _bstats_update(bstats,
855 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
858 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
859 __u64 bytes, __u32 packets)
861 u64_stats_update_begin(&bstats->syncp);
862 _bstats_update(&bstats->bstats, bytes, packets);
863 u64_stats_update_end(&bstats->syncp);
866 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
867 const struct sk_buff *skb)
869 u64_stats_update_begin(&bstats->syncp);
870 bstats_update(&bstats->bstats, skb);
871 u64_stats_update_end(&bstats->syncp);
874 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
875 const struct sk_buff *skb)
877 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
880 static inline void qdisc_bstats_update(struct Qdisc *sch,
881 const struct sk_buff *skb)
883 bstats_update(&sch->bstats, skb);
886 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
887 const struct sk_buff *skb)
889 sch->qstats.backlog -= qdisc_pkt_len(skb);
892 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
893 const struct sk_buff *skb)
895 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
898 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
899 const struct sk_buff *skb)
901 sch->qstats.backlog += qdisc_pkt_len(skb);
904 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
905 const struct sk_buff *skb)
907 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
910 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
912 this_cpu_inc(sch->cpu_qstats->qlen);
915 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
917 this_cpu_dec(sch->cpu_qstats->qlen);
920 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
922 this_cpu_inc(sch->cpu_qstats->requeues);
925 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
927 sch->qstats.drops += count;
930 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
935 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
937 qstats->overlimits++;
940 static inline void qdisc_qstats_drop(struct Qdisc *sch)
942 qstats_drop_inc(&sch->qstats);
945 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
947 this_cpu_inc(sch->cpu_qstats->drops);
950 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
952 sch->qstats.overlimits++;
955 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
957 __u32 qlen = qdisc_qlen_sum(sch);
959 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
962 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
965 struct gnet_stats_queue qstats = { 0 };
966 __u32 len = qdisc_qlen_sum(sch);
968 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
970 *backlog = qstats.backlog;
973 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
977 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
978 qdisc_tree_reduce_backlog(sch, qlen, backlog);
981 static inline void qdisc_purge_queue(struct Qdisc *sch)
985 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
987 qdisc_tree_reduce_backlog(sch, qlen, backlog);
990 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
997 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
998 struct qdisc_skb_head *qh)
1000 struct sk_buff *last = qh->tail;
1013 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1015 __qdisc_enqueue_tail(skb, &sch->q);
1016 qdisc_qstats_backlog_inc(sch, skb);
1017 return NET_XMIT_SUCCESS;
1020 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1021 struct qdisc_skb_head *qh)
1023 skb->next = qh->head;
1031 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1033 struct sk_buff *skb = qh->head;
1035 if (likely(skb != NULL)) {
1036 qh->head = skb->next;
1038 if (qh->head == NULL)
1046 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1048 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1050 if (likely(skb != NULL)) {
1051 qdisc_qstats_backlog_dec(sch, skb);
1052 qdisc_bstats_update(sch, skb);
1058 /* Instead of calling kfree_skb() while root qdisc lock is held,
1059 * queue the skb for future freeing at end of __dev_xmit_skb()
1061 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1063 skb->next = *to_free;
1067 static inline void __qdisc_drop_all(struct sk_buff *skb,
1068 struct sk_buff **to_free)
1071 skb->prev->next = *to_free;
1073 skb->next = *to_free;
1077 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1078 struct qdisc_skb_head *qh,
1079 struct sk_buff **to_free)
1081 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1083 if (likely(skb != NULL)) {
1084 unsigned int len = qdisc_pkt_len(skb);
1086 qdisc_qstats_backlog_dec(sch, skb);
1087 __qdisc_drop(skb, to_free);
1094 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1096 const struct qdisc_skb_head *qh = &sch->q;
1101 /* generic pseudo peek method for non-work-conserving qdisc */
1102 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1104 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1106 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1108 skb = sch->dequeue(sch);
1111 __skb_queue_head(&sch->gso_skb, skb);
1112 /* it's still part of the queue */
1113 qdisc_qstats_backlog_inc(sch, skb);
1121 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1122 struct sk_buff *skb)
1124 if (qdisc_is_percpu_stats(sch)) {
1125 qdisc_qstats_cpu_backlog_dec(sch, skb);
1126 qdisc_bstats_cpu_update(sch, skb);
1127 qdisc_qstats_cpu_qlen_dec(sch);
1129 qdisc_qstats_backlog_dec(sch, skb);
1130 qdisc_bstats_update(sch, skb);
1135 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1136 unsigned int pkt_len)
1138 if (qdisc_is_percpu_stats(sch)) {
1139 qdisc_qstats_cpu_qlen_inc(sch);
1140 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1142 sch->qstats.backlog += pkt_len;
1147 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1148 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1150 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1153 skb = __skb_dequeue(&sch->gso_skb);
1154 if (qdisc_is_percpu_stats(sch)) {
1155 qdisc_qstats_cpu_backlog_dec(sch, skb);
1156 qdisc_qstats_cpu_qlen_dec(sch);
1158 qdisc_qstats_backlog_dec(sch, skb);
1162 skb = sch->dequeue(sch);
1168 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1171 * We do not know the backlog in bytes of this list, it
1172 * is up to the caller to correct it
1176 rtnl_kfree_skbs(qh->head, qh->tail);
1184 static inline void qdisc_reset_queue(struct Qdisc *sch)
1186 __qdisc_reset_queue(&sch->q);
1187 sch->qstats.backlog = 0;
1190 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1191 struct Qdisc **pold)
1199 qdisc_purge_queue(old);
1200 sch_tree_unlock(sch);
1205 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1207 rtnl_kfree_skbs(skb, skb);
1208 qdisc_qstats_drop(sch);
1211 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1212 struct sk_buff **to_free)
1214 __qdisc_drop(skb, to_free);
1215 qdisc_qstats_cpu_drop(sch);
1217 return NET_XMIT_DROP;
1220 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1221 struct sk_buff **to_free)
1223 __qdisc_drop(skb, to_free);
1224 qdisc_qstats_drop(sch);
1226 return NET_XMIT_DROP;
1229 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1230 struct sk_buff **to_free)
1232 __qdisc_drop_all(skb, to_free);
1233 qdisc_qstats_drop(sch);
1235 return NET_XMIT_DROP;
1238 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1239 long it will take to send a packet given its size.
1241 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1243 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1246 slot >>= rtab->rate.cell_log;
1248 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1249 return rtab->data[slot];
1252 struct psched_ratecfg {
1253 u64 rate_bytes_ps; /* bytes per second */
1260 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1265 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1266 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1268 return ((u64)len * r->mult) >> r->shift;
1271 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1272 const struct tc_ratespec *conf,
1275 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1276 const struct psched_ratecfg *r)
1278 memset(res, 0, sizeof(*res));
1280 /* legacy struct tc_ratespec has a 32bit @rate field
1281 * Qdisc using 64bit rate should add new attributes
1282 * in order to maintain compatibility.
1284 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1286 res->overhead = r->overhead;
1287 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1290 struct psched_pktrate {
1291 u64 rate_pkts_ps; /* packets per second */
1296 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1297 unsigned int pkt_num)
1299 return ((u64)pkt_num * r->mult) >> r->shift;
1302 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1304 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1305 * The fast path only needs to access filter list and to update stats
1308 struct tcf_proto *filter_list;
1309 struct tcf_block *block;
1310 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1311 struct gnet_stats_queue __percpu *cpu_qstats;
1312 struct rcu_head rcu;
1315 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1316 const struct sk_buff *skb)
1318 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1321 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1323 this_cpu_inc(miniq->cpu_qstats->drops);
1326 struct mini_Qdisc_pair {
1327 struct mini_Qdisc miniq1;
1328 struct mini_Qdisc miniq2;
1329 struct mini_Qdisc __rcu **p_miniq;
1332 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1333 struct tcf_proto *tp_head);
1334 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1335 struct mini_Qdisc __rcu **p_miniq);
1336 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1337 struct tcf_block *block);
1339 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));