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 /* If the MISSED flag is set, it means other thread has
167 * set the MISSED flag before second spin_trylock(), so
168 * we can return false here to avoid multi cpus doing
169 * the set_bit() and second spin_trylock() concurrently.
171 if (test_bit(__QDISC_STATE_MISSED, &qdisc->state))
174 /* Set the MISSED flag before the second spin_trylock(),
175 * if the second spin_trylock() return false, it means
176 * other cpu holding the lock will do dequeuing for us
177 * or it will see the MISSED flag set after releasing
178 * lock and reschedule the net_tx_action() to do the
181 set_bit(__QDISC_STATE_MISSED, &qdisc->state);
183 /* Retry again in case other CPU may not see the new flag
184 * after it releases the lock at the end of qdisc_run_end().
186 if (!spin_trylock(&qdisc->seqlock))
190 WRITE_ONCE(qdisc->empty, false);
192 } else if (qdisc_is_running(qdisc)) {
195 /* Variant of write_seqcount_begin() telling lockdep a trylock
198 raw_write_seqcount_begin(&qdisc->running);
199 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
203 static inline void qdisc_run_end(struct Qdisc *qdisc)
205 if (qdisc->flags & TCQ_F_NOLOCK) {
206 spin_unlock(&qdisc->seqlock);
208 if (unlikely(test_bit(__QDISC_STATE_MISSED,
210 clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
211 __netif_schedule(qdisc);
214 write_seqcount_end(&qdisc->running);
218 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
220 return qdisc->flags & TCQ_F_ONETXQUEUE;
223 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
226 /* Non-BQL migrated drivers will return 0, too. */
227 return dql_avail(&txq->dql);
233 struct Qdisc_class_ops {
235 /* Child qdisc manipulation */
236 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
237 int (*graft)(struct Qdisc *, unsigned long cl,
238 struct Qdisc *, struct Qdisc **,
239 struct netlink_ext_ack *extack);
240 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
241 void (*qlen_notify)(struct Qdisc *, unsigned long);
243 /* Class manipulation routines */
244 unsigned long (*find)(struct Qdisc *, u32 classid);
245 int (*change)(struct Qdisc *, u32, u32,
246 struct nlattr **, unsigned long *,
247 struct netlink_ext_ack *);
248 int (*delete)(struct Qdisc *, unsigned long,
249 struct netlink_ext_ack *);
250 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
252 /* Filter manipulation */
253 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
255 struct netlink_ext_ack *extack);
256 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
258 void (*unbind_tcf)(struct Qdisc *, unsigned long);
260 /* rtnetlink specific */
261 int (*dump)(struct Qdisc *, unsigned long,
262 struct sk_buff *skb, struct tcmsg*);
263 int (*dump_stats)(struct Qdisc *, unsigned long,
267 /* Qdisc_class_ops flag values */
269 /* Implements API that doesn't require rtnl lock */
270 enum qdisc_class_ops_flags {
271 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
275 struct Qdisc_ops *next;
276 const struct Qdisc_class_ops *cl_ops;
279 unsigned int static_flags;
281 int (*enqueue)(struct sk_buff *skb,
283 struct sk_buff **to_free);
284 struct sk_buff * (*dequeue)(struct Qdisc *);
285 struct sk_buff * (*peek)(struct Qdisc *);
287 int (*init)(struct Qdisc *sch, struct nlattr *arg,
288 struct netlink_ext_ack *extack);
289 void (*reset)(struct Qdisc *);
290 void (*destroy)(struct Qdisc *);
291 int (*change)(struct Qdisc *sch,
293 struct netlink_ext_ack *extack);
294 void (*attach)(struct Qdisc *sch);
295 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
297 int (*dump)(struct Qdisc *, struct sk_buff *);
298 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
300 void (*ingress_block_set)(struct Qdisc *sch,
302 void (*egress_block_set)(struct Qdisc *sch,
304 u32 (*ingress_block_get)(struct Qdisc *sch);
305 u32 (*egress_block_get)(struct Qdisc *sch);
307 struct module *owner;
317 const struct tcf_proto *goto_tp;
319 /* used in the skb_tc_reinsert function */
322 struct gnet_stats_queue *qstats;
329 struct tcf_proto_ops {
330 struct list_head head;
333 int (*classify)(struct sk_buff *,
334 const struct tcf_proto *,
335 struct tcf_result *);
336 int (*init)(struct tcf_proto*);
337 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
338 struct netlink_ext_ack *extack);
340 void* (*get)(struct tcf_proto*, u32 handle);
341 void (*put)(struct tcf_proto *tp, void *f);
342 int (*change)(struct net *net, struct sk_buff *,
343 struct tcf_proto*, unsigned long,
344 u32 handle, struct nlattr **,
346 struct netlink_ext_ack *);
347 int (*delete)(struct tcf_proto *tp, void *arg,
348 bool *last, bool rtnl_held,
349 struct netlink_ext_ack *);
350 bool (*delete_empty)(struct tcf_proto *tp);
351 void (*walk)(struct tcf_proto *tp,
352 struct tcf_walker *arg, bool rtnl_held);
353 int (*reoffload)(struct tcf_proto *tp, bool add,
354 flow_setup_cb_t *cb, void *cb_priv,
355 struct netlink_ext_ack *extack);
356 void (*hw_add)(struct tcf_proto *tp,
358 void (*hw_del)(struct tcf_proto *tp,
360 void (*bind_class)(void *, u32, unsigned long,
361 void *, unsigned long);
362 void * (*tmplt_create)(struct net *net,
363 struct tcf_chain *chain,
365 struct netlink_ext_ack *extack);
366 void (*tmplt_destroy)(void *tmplt_priv);
368 /* rtnetlink specific */
369 int (*dump)(struct net*, struct tcf_proto*, void *,
370 struct sk_buff *skb, struct tcmsg*,
372 int (*terse_dump)(struct net *net,
373 struct tcf_proto *tp, void *fh,
375 struct tcmsg *t, bool rtnl_held);
376 int (*tmplt_dump)(struct sk_buff *skb,
380 struct module *owner;
384 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
385 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
386 * conditions can occur when filters are inserted/deleted simultaneously.
388 enum tcf_proto_ops_flags {
389 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
393 /* Fast access part */
394 struct tcf_proto __rcu *next;
397 /* called under RCU BH lock*/
398 int (*classify)(struct sk_buff *,
399 const struct tcf_proto *,
400 struct tcf_result *);
406 const struct tcf_proto_ops *ops;
407 struct tcf_chain *chain;
408 /* Lock protects tcf_proto shared state and can be used by unlocked
409 * classifiers to protect their private data.
415 struct hlist_node destroy_ht_node;
418 struct qdisc_skb_cb {
420 unsigned int pkt_len;
421 u16 slave_dev_queue_mapping;
424 #define QDISC_CB_PRIV_LEN 20
425 unsigned char data[QDISC_CB_PRIV_LEN];
430 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
433 /* Protects filter_chain. */
434 struct mutex filter_chain_lock;
435 struct tcf_proto __rcu *filter_chain;
436 struct list_head list;
437 struct tcf_block *block;
438 u32 index; /* chain index */
440 unsigned int action_refcnt;
441 bool explicitly_created;
443 const struct tcf_proto_ops *tmplt_ops;
449 /* Lock protects tcf_block and lifetime-management data of chains
450 * attached to the block (refcnt, action_refcnt, explicitly_created).
453 struct list_head chain_list;
454 u32 index; /* block index for shared blocks */
455 u32 classid; /* which class this block belongs to */
459 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
460 struct flow_block flow_block;
461 struct list_head owner_list;
463 atomic_t offloadcnt; /* Number of oddloaded filters */
464 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
465 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
467 struct tcf_chain *chain;
468 struct list_head filter_chain_list;
471 DECLARE_HASHTABLE(proto_destroy_ht, 7);
472 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
475 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
477 return lockdep_is_held(&chain->filter_chain_lock);
480 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
482 return lockdep_is_held(&tp->lock);
485 #define tcf_chain_dereference(p, chain) \
486 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
488 #define tcf_proto_dereference(p, tp) \
489 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
491 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
493 struct qdisc_skb_cb *qcb;
495 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
496 BUILD_BUG_ON(sizeof(qcb->data) < sz);
499 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
501 return this_cpu_ptr(q->cpu_qstats)->qlen;
504 static inline int qdisc_qlen(const struct Qdisc *q)
509 static inline int qdisc_qlen_sum(const struct Qdisc *q)
511 __u32 qlen = q->qstats.qlen;
514 if (qdisc_is_percpu_stats(q)) {
515 for_each_possible_cpu(i)
516 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
524 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
526 return (struct qdisc_skb_cb *)skb->cb;
529 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
531 return &qdisc->q.lock;
534 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
536 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
541 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
543 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
546 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
548 return qdisc->dev_queue->qdisc_sleeping;
551 /* The qdisc root lock is a mechanism by which to top level
552 * of a qdisc tree can be locked from any qdisc node in the
553 * forest. This allows changing the configuration of some
554 * aspect of the qdisc tree while blocking out asynchronous
555 * qdisc access in the packet processing paths.
557 * It is only legal to do this when the root will not change
558 * on us. Otherwise we'll potentially lock the wrong qdisc
559 * root. This is enforced by holding the RTNL semaphore, which
560 * all users of this lock accessor must do.
562 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
564 struct Qdisc *root = qdisc_root(qdisc);
567 return qdisc_lock(root);
570 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
572 struct Qdisc *root = qdisc_root_sleeping(qdisc);
575 return qdisc_lock(root);
578 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
580 struct Qdisc *root = qdisc_root_sleeping(qdisc);
583 return &root->running;
586 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
588 return qdisc->dev_queue->dev;
591 static inline void sch_tree_lock(struct Qdisc *q)
593 if (q->flags & TCQ_F_MQROOT)
594 spin_lock_bh(qdisc_lock(q));
596 spin_lock_bh(qdisc_root_sleeping_lock(q));
599 static inline void sch_tree_unlock(struct Qdisc *q)
601 if (q->flags & TCQ_F_MQROOT)
602 spin_unlock_bh(qdisc_lock(q));
604 spin_unlock_bh(qdisc_root_sleeping_lock(q));
607 extern struct Qdisc noop_qdisc;
608 extern struct Qdisc_ops noop_qdisc_ops;
609 extern struct Qdisc_ops pfifo_fast_ops;
610 extern struct Qdisc_ops mq_qdisc_ops;
611 extern struct Qdisc_ops noqueue_qdisc_ops;
612 extern const struct Qdisc_ops *default_qdisc_ops;
613 static inline const struct Qdisc_ops *
614 get_default_qdisc_ops(const struct net_device *dev, int ntx)
616 return ntx < dev->real_num_tx_queues ?
617 default_qdisc_ops : &pfifo_fast_ops;
620 struct Qdisc_class_common {
622 struct hlist_node hnode;
625 struct Qdisc_class_hash {
626 struct hlist_head *hash;
627 unsigned int hashsize;
628 unsigned int hashmask;
629 unsigned int hashelems;
632 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
639 static inline struct Qdisc_class_common *
640 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
642 struct Qdisc_class_common *cl;
648 h = qdisc_class_hash(id, hash->hashmask);
649 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
650 if (cl->classid == id)
656 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
658 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
660 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
663 int qdisc_class_hash_init(struct Qdisc_class_hash *);
664 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
665 struct Qdisc_class_common *);
666 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
667 struct Qdisc_class_common *);
668 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
669 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
671 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
672 void dev_init_scheduler(struct net_device *dev);
673 void dev_shutdown(struct net_device *dev);
674 void dev_activate(struct net_device *dev);
675 void dev_deactivate(struct net_device *dev);
676 void dev_deactivate_many(struct list_head *head);
677 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
678 struct Qdisc *qdisc);
679 void qdisc_reset(struct Qdisc *qdisc);
680 void qdisc_put(struct Qdisc *qdisc);
681 void qdisc_put_unlocked(struct Qdisc *qdisc);
682 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
683 #ifdef CONFIG_NET_SCHED
684 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
686 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
687 struct Qdisc *new, struct Qdisc *old,
688 enum tc_setup_type type, void *type_data,
689 struct netlink_ext_ack *extack);
692 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
695 q->flags &= ~TCQ_F_OFFLOADED;
700 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
701 struct Qdisc *new, struct Qdisc *old,
702 enum tc_setup_type type, void *type_data,
703 struct netlink_ext_ack *extack)
707 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
708 const struct Qdisc_ops *ops,
709 struct netlink_ext_ack *extack);
710 void qdisc_free(struct Qdisc *qdisc);
711 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
712 const struct Qdisc_ops *ops, u32 parentid,
713 struct netlink_ext_ack *extack);
714 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
715 const struct qdisc_size_table *stab);
716 int skb_do_redirect(struct sk_buff *);
718 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
720 #ifdef CONFIG_NET_CLS_ACT
721 return skb->tc_at_ingress;
727 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
729 #ifdef CONFIG_NET_CLS_ACT
730 if (skb->tc_skip_classify) {
731 skb->tc_skip_classify = 0;
738 /* Reset all TX qdiscs greater than index of a device. */
739 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
743 for (; i < dev->num_tx_queues; i++) {
744 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
746 spin_lock_bh(qdisc_lock(qdisc));
748 spin_unlock_bh(qdisc_lock(qdisc));
753 /* Are all TX queues of the device empty? */
754 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
759 for (i = 0; i < dev->num_tx_queues; i++) {
760 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
761 const struct Qdisc *q = rcu_dereference(txq->qdisc);
763 if (!qdisc_is_empty(q)) {
772 /* Are any of the TX qdiscs changing? */
773 static inline bool qdisc_tx_changing(const struct net_device *dev)
777 for (i = 0; i < dev->num_tx_queues; i++) {
778 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
779 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
785 /* Is the device using the noop qdisc on all queues? */
786 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
790 for (i = 0; i < dev->num_tx_queues; i++) {
791 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
792 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
798 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
800 return qdisc_skb_cb(skb)->pkt_len;
803 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
804 enum net_xmit_qdisc_t {
805 __NET_XMIT_STOLEN = 0x00010000,
806 __NET_XMIT_BYPASS = 0x00020000,
809 #ifdef CONFIG_NET_CLS_ACT
810 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
812 #define net_xmit_drop_count(e) (1)
815 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
816 const struct Qdisc *sch)
818 #ifdef CONFIG_NET_SCHED
819 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
822 __qdisc_calculate_pkt_len(skb, stab);
826 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
827 struct sk_buff **to_free)
829 qdisc_calculate_pkt_len(skb, sch);
830 return sch->enqueue(skb, sch, to_free);
833 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
834 __u64 bytes, __u32 packets)
836 bstats->bytes += bytes;
837 bstats->packets += packets;
840 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
841 const struct sk_buff *skb)
843 _bstats_update(bstats,
845 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
848 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
849 __u64 bytes, __u32 packets)
851 u64_stats_update_begin(&bstats->syncp);
852 _bstats_update(&bstats->bstats, bytes, packets);
853 u64_stats_update_end(&bstats->syncp);
856 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
857 const struct sk_buff *skb)
859 u64_stats_update_begin(&bstats->syncp);
860 bstats_update(&bstats->bstats, skb);
861 u64_stats_update_end(&bstats->syncp);
864 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
865 const struct sk_buff *skb)
867 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
870 static inline void qdisc_bstats_update(struct Qdisc *sch,
871 const struct sk_buff *skb)
873 bstats_update(&sch->bstats, skb);
876 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
877 const struct sk_buff *skb)
879 sch->qstats.backlog -= qdisc_pkt_len(skb);
882 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
883 const struct sk_buff *skb)
885 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
888 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
889 const struct sk_buff *skb)
891 sch->qstats.backlog += qdisc_pkt_len(skb);
894 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
895 const struct sk_buff *skb)
897 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
900 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
902 this_cpu_inc(sch->cpu_qstats->qlen);
905 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
907 this_cpu_dec(sch->cpu_qstats->qlen);
910 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
912 this_cpu_inc(sch->cpu_qstats->requeues);
915 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
917 sch->qstats.drops += count;
920 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
925 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
927 qstats->overlimits++;
930 static inline void qdisc_qstats_drop(struct Qdisc *sch)
932 qstats_drop_inc(&sch->qstats);
935 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
937 this_cpu_inc(sch->cpu_qstats->drops);
940 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
942 sch->qstats.overlimits++;
945 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
947 __u32 qlen = qdisc_qlen_sum(sch);
949 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
952 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
955 struct gnet_stats_queue qstats = { 0 };
956 __u32 len = qdisc_qlen_sum(sch);
958 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
960 *backlog = qstats.backlog;
963 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
967 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
968 qdisc_tree_reduce_backlog(sch, qlen, backlog);
971 static inline void qdisc_purge_queue(struct Qdisc *sch)
975 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
977 qdisc_tree_reduce_backlog(sch, qlen, backlog);
980 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
987 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
988 struct qdisc_skb_head *qh)
990 struct sk_buff *last = qh->tail;
1003 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1005 __qdisc_enqueue_tail(skb, &sch->q);
1006 qdisc_qstats_backlog_inc(sch, skb);
1007 return NET_XMIT_SUCCESS;
1010 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1011 struct qdisc_skb_head *qh)
1013 skb->next = qh->head;
1021 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1023 struct sk_buff *skb = qh->head;
1025 if (likely(skb != NULL)) {
1026 qh->head = skb->next;
1028 if (qh->head == NULL)
1036 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1038 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1040 if (likely(skb != NULL)) {
1041 qdisc_qstats_backlog_dec(sch, skb);
1042 qdisc_bstats_update(sch, skb);
1048 /* Instead of calling kfree_skb() while root qdisc lock is held,
1049 * queue the skb for future freeing at end of __dev_xmit_skb()
1051 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1053 skb->next = *to_free;
1057 static inline void __qdisc_drop_all(struct sk_buff *skb,
1058 struct sk_buff **to_free)
1061 skb->prev->next = *to_free;
1063 skb->next = *to_free;
1067 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1068 struct qdisc_skb_head *qh,
1069 struct sk_buff **to_free)
1071 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1073 if (likely(skb != NULL)) {
1074 unsigned int len = qdisc_pkt_len(skb);
1076 qdisc_qstats_backlog_dec(sch, skb);
1077 __qdisc_drop(skb, to_free);
1084 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1086 const struct qdisc_skb_head *qh = &sch->q;
1091 /* generic pseudo peek method for non-work-conserving qdisc */
1092 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1094 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1096 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1098 skb = sch->dequeue(sch);
1101 __skb_queue_head(&sch->gso_skb, skb);
1102 /* it's still part of the queue */
1103 qdisc_qstats_backlog_inc(sch, skb);
1111 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1112 struct sk_buff *skb)
1114 if (qdisc_is_percpu_stats(sch)) {
1115 qdisc_qstats_cpu_backlog_dec(sch, skb);
1116 qdisc_bstats_cpu_update(sch, skb);
1117 qdisc_qstats_cpu_qlen_dec(sch);
1119 qdisc_qstats_backlog_dec(sch, skb);
1120 qdisc_bstats_update(sch, skb);
1125 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1126 unsigned int pkt_len)
1128 if (qdisc_is_percpu_stats(sch)) {
1129 qdisc_qstats_cpu_qlen_inc(sch);
1130 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1132 sch->qstats.backlog += pkt_len;
1137 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1138 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1140 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1143 skb = __skb_dequeue(&sch->gso_skb);
1144 if (qdisc_is_percpu_stats(sch)) {
1145 qdisc_qstats_cpu_backlog_dec(sch, skb);
1146 qdisc_qstats_cpu_qlen_dec(sch);
1148 qdisc_qstats_backlog_dec(sch, skb);
1152 skb = sch->dequeue(sch);
1158 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1161 * We do not know the backlog in bytes of this list, it
1162 * is up to the caller to correct it
1166 rtnl_kfree_skbs(qh->head, qh->tail);
1174 static inline void qdisc_reset_queue(struct Qdisc *sch)
1176 __qdisc_reset_queue(&sch->q);
1177 sch->qstats.backlog = 0;
1180 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1181 struct Qdisc **pold)
1189 qdisc_purge_queue(old);
1190 sch_tree_unlock(sch);
1195 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1197 rtnl_kfree_skbs(skb, skb);
1198 qdisc_qstats_drop(sch);
1201 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1202 struct sk_buff **to_free)
1204 __qdisc_drop(skb, to_free);
1205 qdisc_qstats_cpu_drop(sch);
1207 return NET_XMIT_DROP;
1210 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1211 struct sk_buff **to_free)
1213 __qdisc_drop(skb, to_free);
1214 qdisc_qstats_drop(sch);
1216 return NET_XMIT_DROP;
1219 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1220 struct sk_buff **to_free)
1222 __qdisc_drop_all(skb, to_free);
1223 qdisc_qstats_drop(sch);
1225 return NET_XMIT_DROP;
1228 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1229 long it will take to send a packet given its size.
1231 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1233 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1236 slot >>= rtab->rate.cell_log;
1238 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1239 return rtab->data[slot];
1242 struct psched_ratecfg {
1243 u64 rate_bytes_ps; /* bytes per second */
1250 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1255 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1256 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1258 return ((u64)len * r->mult) >> r->shift;
1261 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1262 const struct tc_ratespec *conf,
1265 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1266 const struct psched_ratecfg *r)
1268 memset(res, 0, sizeof(*res));
1270 /* legacy struct tc_ratespec has a 32bit @rate field
1271 * Qdisc using 64bit rate should add new attributes
1272 * in order to maintain compatibility.
1274 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1276 res->overhead = r->overhead;
1277 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1280 struct psched_pktrate {
1281 u64 rate_pkts_ps; /* packets per second */
1286 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1287 unsigned int pkt_num)
1289 return ((u64)pkt_num * r->mult) >> r->shift;
1292 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1294 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1295 * The fast path only needs to access filter list and to update stats
1298 struct tcf_proto *filter_list;
1299 struct tcf_block *block;
1300 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1301 struct gnet_stats_queue __percpu *cpu_qstats;
1302 struct rcu_head rcu;
1305 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1306 const struct sk_buff *skb)
1308 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1311 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1313 this_cpu_inc(miniq->cpu_qstats->drops);
1316 struct mini_Qdisc_pair {
1317 struct mini_Qdisc miniq1;
1318 struct mini_Qdisc miniq2;
1319 struct mini_Qdisc __rcu **p_miniq;
1322 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1323 struct tcf_proto *tp_head);
1324 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1325 struct mini_Qdisc __rcu **p_miniq);
1326 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1327 struct tcf_block *block);
1329 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));