1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __NET_SCHED_GENERIC_H
3 #define __NET_SCHED_GENERIC_H
5 #include <linux/netdevice.h>
6 #include <linux/types.h>
7 #include <linux/rcupdate.h>
8 #include <linux/pkt_sched.h>
9 #include <linux/pkt_cls.h>
10 #include <linux/percpu.h>
11 #include <linux/dynamic_queue_limits.h>
12 #include <linux/list.h>
13 #include <linux/refcount.h>
14 #include <linux/workqueue.h>
15 #include <linux/mutex.h>
16 #include <linux/rwsem.h>
17 #include <linux/atomic.h>
18 #include <linux/hashtable.h>
19 #include <net/gen_stats.h>
20 #include <net/rtnetlink.h>
21 #include <net/flow_offload.h>
29 struct qdisc_rate_table {
30 struct tc_ratespec rate;
32 struct qdisc_rate_table *next;
38 __QDISC_STATE_DEACTIVATED,
40 __QDISC_STATE_DRAINING,
44 /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly.
45 * Use qdisc_run_begin/end() or qdisc_is_running() instead.
47 __QDISC_STATE2_RUNNING,
50 #define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED)
51 #define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING)
53 #define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \
56 struct qdisc_size_table {
58 struct list_head list;
59 struct tc_sizespec szopts;
64 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
65 struct qdisc_skb_head {
73 int (*enqueue)(struct sk_buff *skb,
75 struct sk_buff **to_free);
76 struct sk_buff * (*dequeue)(struct Qdisc *sch);
78 #define TCQ_F_BUILTIN 1
79 #define TCQ_F_INGRESS 2
80 #define TCQ_F_CAN_BYPASS 4
81 #define TCQ_F_MQROOT 8
82 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
83 * q->dev_queue : It can test
84 * netif_xmit_frozen_or_stopped() before
85 * dequeueing next packet.
86 * Its true for MQ/MQPRIO slaves, or non
89 #define TCQ_F_WARN_NONWC (1 << 16)
90 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
91 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
92 * qdisc_tree_decrease_qlen() should stop.
94 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
95 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
96 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
98 const struct Qdisc_ops *ops;
99 struct qdisc_size_table __rcu *stab;
100 struct hlist_node hash;
104 struct netdev_queue *dev_queue;
106 struct net_rate_estimator __rcu *rate_est;
107 struct gnet_stats_basic_sync __percpu *cpu_bstats;
108 struct gnet_stats_queue __percpu *cpu_qstats;
113 * For performance sake on SMP, we put highly modified fields at the end
115 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
116 struct qdisc_skb_head q;
117 struct gnet_stats_basic_sync bstats;
118 struct gnet_stats_queue qstats;
120 unsigned long state2; /* must be written under qdisc spinlock */
121 struct Qdisc *next_sched;
122 struct sk_buff_head skb_bad_txq;
124 spinlock_t busylock ____cacheline_aligned_in_smp;
128 netdevice_tracker dev_tracker;
130 long privdata[] ____cacheline_aligned;
133 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
135 if (qdisc->flags & TCQ_F_BUILTIN)
137 refcount_inc(&qdisc->refcnt);
140 static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc)
142 if (qdisc->flags & TCQ_F_BUILTIN)
144 return refcount_dec_if_one(&qdisc->refcnt);
147 /* Intended to be used by unlocked users, when concurrent qdisc release is
151 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
153 if (qdisc->flags & TCQ_F_BUILTIN)
155 if (refcount_inc_not_zero(&qdisc->refcnt))
160 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
161 * root_lock section, or provide their own memory barriers -- ordering
162 * against qdisc_run_begin/end() atomic bit operations.
164 static inline bool qdisc_is_running(struct Qdisc *qdisc)
166 if (qdisc->flags & TCQ_F_NOLOCK)
167 return spin_is_locked(&qdisc->seqlock);
168 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
171 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
173 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
176 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
178 return q->flags & TCQ_F_CPUSTATS;
181 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
183 if (qdisc_is_percpu_stats(qdisc))
184 return nolock_qdisc_is_empty(qdisc);
185 return !READ_ONCE(qdisc->q.qlen);
188 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
189 * the qdisc root lock acquired.
191 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
193 if (qdisc->flags & TCQ_F_NOLOCK) {
194 if (spin_trylock(&qdisc->seqlock))
197 /* No need to insist if the MISSED flag was already set.
198 * Note that test_and_set_bit() also gives us memory ordering
199 * guarantees wrt potential earlier enqueue() and below
200 * spin_trylock(), both of which are necessary to prevent races
202 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
205 /* Try to take the lock again to make sure that we will either
206 * grab it or the CPU that still has it will see MISSED set
207 * when testing it in qdisc_run_end()
209 return spin_trylock(&qdisc->seqlock);
211 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
214 static inline void qdisc_run_end(struct Qdisc *qdisc)
216 if (qdisc->flags & TCQ_F_NOLOCK) {
217 spin_unlock(&qdisc->seqlock);
219 /* spin_unlock() only has store-release semantic. The unlock
220 * and test_bit() ordering is a store-load ordering, so a full
221 * memory barrier is needed here.
225 if (unlikely(test_bit(__QDISC_STATE_MISSED,
227 __netif_schedule(qdisc);
229 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
233 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
235 return qdisc->flags & TCQ_F_ONETXQUEUE;
238 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
241 /* Non-BQL migrated drivers will return 0, too. */
242 return dql_avail(&txq->dql);
248 struct Qdisc_class_ops {
250 /* Child qdisc manipulation */
251 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
252 int (*graft)(struct Qdisc *, unsigned long cl,
253 struct Qdisc *, struct Qdisc **,
254 struct netlink_ext_ack *extack);
255 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
256 void (*qlen_notify)(struct Qdisc *, unsigned long);
258 /* Class manipulation routines */
259 unsigned long (*find)(struct Qdisc *, u32 classid);
260 int (*change)(struct Qdisc *, u32, u32,
261 struct nlattr **, unsigned long *,
262 struct netlink_ext_ack *);
263 int (*delete)(struct Qdisc *, unsigned long,
264 struct netlink_ext_ack *);
265 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
267 /* Filter manipulation */
268 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
270 struct netlink_ext_ack *extack);
271 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
273 void (*unbind_tcf)(struct Qdisc *, unsigned long);
275 /* rtnetlink specific */
276 int (*dump)(struct Qdisc *, unsigned long,
277 struct sk_buff *skb, struct tcmsg*);
278 int (*dump_stats)(struct Qdisc *, unsigned long,
282 /* Qdisc_class_ops flag values */
284 /* Implements API that doesn't require rtnl lock */
285 enum qdisc_class_ops_flags {
286 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
290 struct Qdisc_ops *next;
291 const struct Qdisc_class_ops *cl_ops;
294 unsigned int static_flags;
296 int (*enqueue)(struct sk_buff *skb,
298 struct sk_buff **to_free);
299 struct sk_buff * (*dequeue)(struct Qdisc *);
300 struct sk_buff * (*peek)(struct Qdisc *);
302 int (*init)(struct Qdisc *sch, struct nlattr *arg,
303 struct netlink_ext_ack *extack);
304 void (*reset)(struct Qdisc *);
305 void (*destroy)(struct Qdisc *);
306 int (*change)(struct Qdisc *sch,
308 struct netlink_ext_ack *extack);
309 void (*attach)(struct Qdisc *sch);
310 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
311 void (*change_real_num_tx)(struct Qdisc *sch,
312 unsigned int new_real_tx);
314 int (*dump)(struct Qdisc *, struct sk_buff *);
315 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
317 void (*ingress_block_set)(struct Qdisc *sch,
319 void (*egress_block_set)(struct Qdisc *sch,
321 u32 (*ingress_block_get)(struct Qdisc *sch);
322 u32 (*egress_block_get)(struct Qdisc *sch);
324 struct module *owner;
333 const struct tcf_proto *goto_tp;
335 enum skb_drop_reason drop_reason;
340 struct tcf_proto_ops {
341 struct list_head head;
344 int (*classify)(struct sk_buff *,
345 const struct tcf_proto *,
346 struct tcf_result *);
347 int (*init)(struct tcf_proto*);
348 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
349 struct netlink_ext_ack *extack);
351 void* (*get)(struct tcf_proto*, u32 handle);
352 void (*put)(struct tcf_proto *tp, void *f);
353 int (*change)(struct net *net, struct sk_buff *,
354 struct tcf_proto*, unsigned long,
355 u32 handle, struct nlattr **,
357 struct netlink_ext_ack *);
358 int (*delete)(struct tcf_proto *tp, void *arg,
359 bool *last, bool rtnl_held,
360 struct netlink_ext_ack *);
361 bool (*delete_empty)(struct tcf_proto *tp);
362 void (*walk)(struct tcf_proto *tp,
363 struct tcf_walker *arg, bool rtnl_held);
364 int (*reoffload)(struct tcf_proto *tp, bool add,
365 flow_setup_cb_t *cb, void *cb_priv,
366 struct netlink_ext_ack *extack);
367 void (*hw_add)(struct tcf_proto *tp,
369 void (*hw_del)(struct tcf_proto *tp,
371 void (*bind_class)(void *, u32, unsigned long,
372 void *, unsigned long);
373 void * (*tmplt_create)(struct net *net,
374 struct tcf_chain *chain,
376 struct netlink_ext_ack *extack);
377 void (*tmplt_destroy)(void *tmplt_priv);
378 void (*tmplt_reoffload)(struct tcf_chain *chain,
382 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp,
385 /* rtnetlink specific */
386 int (*dump)(struct net*, struct tcf_proto*, void *,
387 struct sk_buff *skb, struct tcmsg*,
389 int (*terse_dump)(struct net *net,
390 struct tcf_proto *tp, void *fh,
392 struct tcmsg *t, bool rtnl_held);
393 int (*tmplt_dump)(struct sk_buff *skb,
397 struct module *owner;
401 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
402 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
403 * conditions can occur when filters are inserted/deleted simultaneously.
405 enum tcf_proto_ops_flags {
406 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
410 /* Fast access part */
411 struct tcf_proto __rcu *next;
414 /* called under RCU BH lock*/
415 int (*classify)(struct sk_buff *,
416 const struct tcf_proto *,
417 struct tcf_result *);
423 const struct tcf_proto_ops *ops;
424 struct tcf_chain *chain;
425 /* Lock protects tcf_proto shared state and can be used by unlocked
426 * classifiers to protect their private data.
432 struct hlist_node destroy_ht_node;
435 struct qdisc_skb_cb {
437 unsigned int pkt_len;
438 u16 slave_dev_queue_mapping;
441 #define QDISC_CB_PRIV_LEN 20
442 unsigned char data[QDISC_CB_PRIV_LEN];
445 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
448 /* Protects filter_chain. */
449 struct mutex filter_chain_lock;
450 struct tcf_proto __rcu *filter_chain;
451 struct list_head list;
452 struct tcf_block *block;
453 u32 index; /* chain index */
455 unsigned int action_refcnt;
456 bool explicitly_created;
458 const struct tcf_proto_ops *tmplt_ops;
464 /* Lock protects tcf_block and lifetime-management data of chains
465 * attached to the block (refcnt, action_refcnt, explicitly_created).
468 struct list_head chain_list;
469 u32 index; /* block index for shared blocks */
470 u32 classid; /* which class this block belongs to */
474 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
475 struct flow_block flow_block;
476 struct list_head owner_list;
478 atomic_t offloadcnt; /* Number of oddloaded filters */
479 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
480 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
482 struct tcf_chain *chain;
483 struct list_head filter_chain_list;
486 DECLARE_HASHTABLE(proto_destroy_ht, 7);
487 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
490 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
492 return lockdep_is_held(&chain->filter_chain_lock);
495 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
497 return lockdep_is_held(&tp->lock);
500 #define tcf_chain_dereference(p, chain) \
501 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
503 #define tcf_proto_dereference(p, tp) \
504 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
506 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
508 struct qdisc_skb_cb *qcb;
510 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
511 BUILD_BUG_ON(sizeof(qcb->data) < sz);
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 rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
561 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
563 struct Qdisc *root = qdisc_root_sleeping(qdisc);
566 return qdisc_lock(root);
569 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
571 return qdisc->dev_queue->dev;
574 static inline void sch_tree_lock(struct Qdisc *q)
576 if (q->flags & TCQ_F_MQROOT)
577 spin_lock_bh(qdisc_lock(q));
579 spin_lock_bh(qdisc_root_sleeping_lock(q));
582 static inline void sch_tree_unlock(struct Qdisc *q)
584 if (q->flags & TCQ_F_MQROOT)
585 spin_unlock_bh(qdisc_lock(q));
587 spin_unlock_bh(qdisc_root_sleeping_lock(q));
590 extern struct Qdisc noop_qdisc;
591 extern struct Qdisc_ops noop_qdisc_ops;
592 extern struct Qdisc_ops pfifo_fast_ops;
593 extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
594 extern struct Qdisc_ops mq_qdisc_ops;
595 extern struct Qdisc_ops noqueue_qdisc_ops;
596 extern const struct Qdisc_ops *default_qdisc_ops;
597 static inline const struct Qdisc_ops *
598 get_default_qdisc_ops(const struct net_device *dev, int ntx)
600 return ntx < dev->real_num_tx_queues ?
601 default_qdisc_ops : &pfifo_fast_ops;
604 struct Qdisc_class_common {
606 unsigned int filter_cnt;
607 struct hlist_node hnode;
610 struct Qdisc_class_hash {
611 struct hlist_head *hash;
612 unsigned int hashsize;
613 unsigned int hashmask;
614 unsigned int hashelems;
617 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
624 static inline struct Qdisc_class_common *
625 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
627 struct Qdisc_class_common *cl;
633 h = qdisc_class_hash(id, hash->hashmask);
634 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
635 if (cl->classid == id)
641 static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
643 return cl->filter_cnt > 0;
646 static inline void qdisc_class_get(struct Qdisc_class_common *cl)
650 if (check_add_overflow(cl->filter_cnt, 1, &res))
651 WARN(1, "Qdisc class overflow");
653 cl->filter_cnt = res;
656 static inline void qdisc_class_put(struct Qdisc_class_common *cl)
660 if (check_sub_overflow(cl->filter_cnt, 1, &res))
661 WARN(1, "Qdisc class underflow");
663 cl->filter_cnt = res;
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_qdisc_change_real_num_tx(struct net_device *dev,
683 unsigned int new_real_tx);
684 void dev_init_scheduler(struct net_device *dev);
685 void dev_shutdown(struct net_device *dev);
686 void dev_activate(struct net_device *dev);
687 void dev_deactivate(struct net_device *dev);
688 void dev_deactivate_many(struct list_head *head);
689 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
690 struct Qdisc *qdisc);
691 void qdisc_reset(struct Qdisc *qdisc);
692 void qdisc_destroy(struct Qdisc *qdisc);
693 void qdisc_put(struct Qdisc *qdisc);
694 void qdisc_put_unlocked(struct Qdisc *qdisc);
695 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
696 #ifdef CONFIG_NET_SCHED
697 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
699 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
700 struct Qdisc *new, struct Qdisc *old,
701 enum tc_setup_type type, void *type_data,
702 struct netlink_ext_ack *extack);
705 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
708 q->flags &= ~TCQ_F_OFFLOADED;
713 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
714 struct Qdisc *new, struct Qdisc *old,
715 enum tc_setup_type type, void *type_data,
716 struct netlink_ext_ack *extack)
720 void qdisc_offload_query_caps(struct net_device *dev,
721 enum tc_setup_type type,
722 void *caps, size_t caps_len);
723 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
724 const struct Qdisc_ops *ops,
725 struct netlink_ext_ack *extack);
726 void qdisc_free(struct Qdisc *qdisc);
727 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
728 const struct Qdisc_ops *ops, u32 parentid,
729 struct netlink_ext_ack *extack);
730 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
731 const struct qdisc_size_table *stab);
732 int skb_do_redirect(struct sk_buff *);
734 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
736 #ifdef CONFIG_NET_XGRESS
737 return skb->tc_at_ingress;
743 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
745 #ifdef CONFIG_NET_CLS_ACT
746 if (skb->tc_skip_classify) {
747 skb->tc_skip_classify = 0;
754 /* Reset all TX qdiscs greater than index of a device. */
755 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
759 for (; i < dev->num_tx_queues; i++) {
760 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
762 spin_lock_bh(qdisc_lock(qdisc));
764 spin_unlock_bh(qdisc_lock(qdisc));
769 /* Are all TX queues of the device empty? */
770 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
775 for (i = 0; i < dev->num_tx_queues; i++) {
776 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
777 const struct Qdisc *q = rcu_dereference(txq->qdisc);
779 if (!qdisc_is_empty(q)) {
788 /* Are any of the TX qdiscs changing? */
789 static inline bool qdisc_tx_changing(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);
796 if (rcu_access_pointer(txq->qdisc) !=
797 rcu_access_pointer(txq->qdisc_sleeping))
803 /* Is the device using the noop qdisc on all queues? */
804 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
808 for (i = 0; i < dev->num_tx_queues; i++) {
809 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
810 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
816 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
818 return qdisc_skb_cb(skb)->pkt_len;
821 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
822 enum net_xmit_qdisc_t {
823 __NET_XMIT_STOLEN = 0x00010000,
824 __NET_XMIT_BYPASS = 0x00020000,
827 #ifdef CONFIG_NET_CLS_ACT
828 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
830 #define net_xmit_drop_count(e) (1)
833 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
834 const struct Qdisc *sch)
836 #ifdef CONFIG_NET_SCHED
837 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
840 __qdisc_calculate_pkt_len(skb, stab);
844 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
845 struct sk_buff **to_free)
847 qdisc_calculate_pkt_len(skb, sch);
848 return sch->enqueue(skb, sch, to_free);
851 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
852 __u64 bytes, __u32 packets)
854 u64_stats_update_begin(&bstats->syncp);
855 u64_stats_add(&bstats->bytes, bytes);
856 u64_stats_add(&bstats->packets, packets);
857 u64_stats_update_end(&bstats->syncp);
860 static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
861 const struct sk_buff *skb)
863 _bstats_update(bstats,
865 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
868 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
869 const struct sk_buff *skb)
871 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
874 static inline void qdisc_bstats_update(struct Qdisc *sch,
875 const struct sk_buff *skb)
877 bstats_update(&sch->bstats, skb);
880 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
881 const struct sk_buff *skb)
883 sch->qstats.backlog -= qdisc_pkt_len(skb);
886 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
887 const struct sk_buff *skb)
889 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
892 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
893 const struct sk_buff *skb)
895 sch->qstats.backlog += qdisc_pkt_len(skb);
898 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
899 const struct sk_buff *skb)
901 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
904 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
906 this_cpu_inc(sch->cpu_qstats->qlen);
909 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
911 this_cpu_dec(sch->cpu_qstats->qlen);
914 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
916 this_cpu_inc(sch->cpu_qstats->requeues);
919 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
921 sch->qstats.drops += count;
924 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
929 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
931 qstats->overlimits++;
934 static inline void qdisc_qstats_drop(struct Qdisc *sch)
936 qstats_drop_inc(&sch->qstats);
939 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
941 this_cpu_inc(sch->cpu_qstats->drops);
944 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
946 sch->qstats.overlimits++;
949 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
951 __u32 qlen = qdisc_qlen_sum(sch);
953 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
956 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
959 struct gnet_stats_queue qstats = { 0 };
961 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
962 *qlen = qstats.qlen + qdisc_qlen(sch);
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_enqueue_tail(struct sk_buff *skb,
984 struct qdisc_skb_head *qh)
986 struct sk_buff *last = qh->tail;
999 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1001 __qdisc_enqueue_tail(skb, &sch->q);
1002 qdisc_qstats_backlog_inc(sch, skb);
1003 return NET_XMIT_SUCCESS;
1006 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1007 struct qdisc_skb_head *qh)
1009 skb->next = qh->head;
1017 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1019 struct sk_buff *skb = qh->head;
1021 if (likely(skb != NULL)) {
1022 qh->head = skb->next;
1024 if (qh->head == NULL)
1032 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1034 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1036 if (likely(skb != NULL)) {
1037 qdisc_qstats_backlog_dec(sch, skb);
1038 qdisc_bstats_update(sch, skb);
1044 /* Instead of calling kfree_skb() while root qdisc lock is held,
1045 * queue the skb for future freeing at end of __dev_xmit_skb()
1047 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1049 skb->next = *to_free;
1053 static inline void __qdisc_drop_all(struct sk_buff *skb,
1054 struct sk_buff **to_free)
1057 skb->prev->next = *to_free;
1059 skb->next = *to_free;
1063 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1064 struct qdisc_skb_head *qh,
1065 struct sk_buff **to_free)
1067 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1069 if (likely(skb != NULL)) {
1070 unsigned int len = qdisc_pkt_len(skb);
1072 qdisc_qstats_backlog_dec(sch, skb);
1073 __qdisc_drop(skb, to_free);
1080 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1082 const struct qdisc_skb_head *qh = &sch->q;
1087 /* generic pseudo peek method for non-work-conserving qdisc */
1088 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1090 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1092 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1094 skb = sch->dequeue(sch);
1097 __skb_queue_head(&sch->gso_skb, skb);
1098 /* it's still part of the queue */
1099 qdisc_qstats_backlog_inc(sch, skb);
1107 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1108 struct sk_buff *skb)
1110 if (qdisc_is_percpu_stats(sch)) {
1111 qdisc_qstats_cpu_backlog_dec(sch, skb);
1112 qdisc_bstats_cpu_update(sch, skb);
1113 qdisc_qstats_cpu_qlen_dec(sch);
1115 qdisc_qstats_backlog_dec(sch, skb);
1116 qdisc_bstats_update(sch, skb);
1121 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1122 unsigned int pkt_len)
1124 if (qdisc_is_percpu_stats(sch)) {
1125 qdisc_qstats_cpu_qlen_inc(sch);
1126 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1128 sch->qstats.backlog += pkt_len;
1133 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1134 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1136 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1139 skb = __skb_dequeue(&sch->gso_skb);
1140 if (qdisc_is_percpu_stats(sch)) {
1141 qdisc_qstats_cpu_backlog_dec(sch, skb);
1142 qdisc_qstats_cpu_qlen_dec(sch);
1144 qdisc_qstats_backlog_dec(sch, skb);
1148 skb = sch->dequeue(sch);
1154 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1157 * We do not know the backlog in bytes of this list, it
1158 * is up to the caller to correct it
1162 rtnl_kfree_skbs(qh->head, qh->tail);
1170 static inline void qdisc_reset_queue(struct Qdisc *sch)
1172 __qdisc_reset_queue(&sch->q);
1175 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1176 struct Qdisc **pold)
1184 qdisc_purge_queue(old);
1185 sch_tree_unlock(sch);
1190 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1192 rtnl_kfree_skbs(skb, skb);
1193 qdisc_qstats_drop(sch);
1196 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1197 struct sk_buff **to_free)
1199 __qdisc_drop(skb, to_free);
1200 qdisc_qstats_cpu_drop(sch);
1202 return NET_XMIT_DROP;
1205 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1206 struct sk_buff **to_free)
1208 __qdisc_drop(skb, to_free);
1209 qdisc_qstats_drop(sch);
1211 return NET_XMIT_DROP;
1214 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1215 struct sk_buff **to_free)
1217 __qdisc_drop_all(skb, to_free);
1218 qdisc_qstats_drop(sch);
1220 return NET_XMIT_DROP;
1223 struct psched_ratecfg {
1224 u64 rate_bytes_ps; /* bytes per second */
1232 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1240 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1241 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1243 return ((u64)len * r->mult) >> r->shift;
1246 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1247 const struct tc_ratespec *conf,
1250 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1251 const struct psched_ratecfg *r)
1253 memset(res, 0, sizeof(*res));
1255 /* legacy struct tc_ratespec has a 32bit @rate field
1256 * Qdisc using 64bit rate should add new attributes
1257 * in order to maintain compatibility.
1259 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1261 res->overhead = r->overhead;
1263 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1266 struct psched_pktrate {
1267 u64 rate_pkts_ps; /* packets per second */
1272 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1273 unsigned int pkt_num)
1275 return ((u64)pkt_num * r->mult) >> r->shift;
1278 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1280 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1281 * The fast path only needs to access filter list and to update stats
1284 struct tcf_proto *filter_list;
1285 struct tcf_block *block;
1286 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1287 struct gnet_stats_queue __percpu *cpu_qstats;
1288 unsigned long rcu_state;
1291 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1292 const struct sk_buff *skb)
1294 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1297 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1299 this_cpu_inc(miniq->cpu_qstats->drops);
1302 struct mini_Qdisc_pair {
1303 struct mini_Qdisc miniq1;
1304 struct mini_Qdisc miniq2;
1305 struct mini_Qdisc __rcu **p_miniq;
1308 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1309 struct tcf_proto *tp_head);
1310 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1311 struct mini_Qdisc __rcu **p_miniq);
1312 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1313 struct tcf_block *block);
1315 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1317 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1319 /* Make sure qdisc is no longer in SCHED state. */
1320 static inline void qdisc_synchronize(const struct Qdisc *q)
1322 while (test_bit(__QDISC_STATE_SCHED, &q->state))