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>
22 #include <linux/xarray.h>
30 struct qdisc_rate_table {
31 struct tc_ratespec rate;
33 struct qdisc_rate_table *next;
39 __QDISC_STATE_DEACTIVATED,
41 __QDISC_STATE_DRAINING,
45 /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly.
46 * Use qdisc_run_begin/end() or qdisc_is_running() instead.
48 __QDISC_STATE2_RUNNING,
51 #define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED)
52 #define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING)
54 #define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \
57 struct qdisc_size_table {
59 struct list_head list;
60 struct tc_sizespec szopts;
65 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
66 struct qdisc_skb_head {
74 int (*enqueue)(struct sk_buff *skb,
76 struct sk_buff **to_free);
77 struct sk_buff * (*dequeue)(struct Qdisc *sch);
79 #define TCQ_F_BUILTIN 1
80 #define TCQ_F_INGRESS 2
81 #define TCQ_F_CAN_BYPASS 4
82 #define TCQ_F_MQROOT 8
83 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
84 * q->dev_queue : It can test
85 * netif_xmit_frozen_or_stopped() before
86 * dequeueing next packet.
87 * Its true for MQ/MQPRIO slaves, or non
90 #define TCQ_F_WARN_NONWC (1 << 16)
91 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
92 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
93 * qdisc_tree_decrease_qlen() should stop.
95 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
96 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
97 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
99 const struct Qdisc_ops *ops;
100 struct qdisc_size_table __rcu *stab;
101 struct hlist_node hash;
105 struct netdev_queue *dev_queue;
107 struct net_rate_estimator __rcu *rate_est;
108 struct gnet_stats_basic_sync __percpu *cpu_bstats;
109 struct gnet_stats_queue __percpu *cpu_qstats;
114 * For performance sake on SMP, we put highly modified fields at the end
116 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
117 struct qdisc_skb_head q;
118 struct gnet_stats_basic_sync bstats;
119 struct gnet_stats_queue qstats;
121 unsigned long state2; /* must be written under qdisc spinlock */
122 struct Qdisc *next_sched;
123 struct sk_buff_head skb_bad_txq;
125 spinlock_t busylock ____cacheline_aligned_in_smp;
129 netdevice_tracker dev_tracker;
131 long privdata[] ____cacheline_aligned;
134 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
136 if (qdisc->flags & TCQ_F_BUILTIN)
138 refcount_inc(&qdisc->refcnt);
141 static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc)
143 if (qdisc->flags & TCQ_F_BUILTIN)
145 return refcount_dec_if_one(&qdisc->refcnt);
148 /* Intended to be used by unlocked users, when concurrent qdisc release is
152 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
154 if (qdisc->flags & TCQ_F_BUILTIN)
156 if (refcount_inc_not_zero(&qdisc->refcnt))
161 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
162 * root_lock section, or provide their own memory barriers -- ordering
163 * against qdisc_run_begin/end() atomic bit operations.
165 static inline bool qdisc_is_running(struct Qdisc *qdisc)
167 if (qdisc->flags & TCQ_F_NOLOCK)
168 return spin_is_locked(&qdisc->seqlock);
169 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
172 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
174 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
177 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
179 return q->flags & TCQ_F_CPUSTATS;
182 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
184 if (qdisc_is_percpu_stats(qdisc))
185 return nolock_qdisc_is_empty(qdisc);
186 return !READ_ONCE(qdisc->q.qlen);
189 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
190 * the qdisc root lock acquired.
192 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
194 if (qdisc->flags & TCQ_F_NOLOCK) {
195 if (spin_trylock(&qdisc->seqlock))
198 /* No need to insist if the MISSED flag was already set.
199 * Note that test_and_set_bit() also gives us memory ordering
200 * guarantees wrt potential earlier enqueue() and below
201 * spin_trylock(), both of which are necessary to prevent races
203 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
206 /* Try to take the lock again to make sure that we will either
207 * grab it or the CPU that still has it will see MISSED set
208 * when testing it in qdisc_run_end()
210 return spin_trylock(&qdisc->seqlock);
212 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
215 static inline void qdisc_run_end(struct Qdisc *qdisc)
217 if (qdisc->flags & TCQ_F_NOLOCK) {
218 spin_unlock(&qdisc->seqlock);
220 /* spin_unlock() only has store-release semantic. The unlock
221 * and test_bit() ordering is a store-load ordering, so a full
222 * memory barrier is needed here.
226 if (unlikely(test_bit(__QDISC_STATE_MISSED,
228 __netif_schedule(qdisc);
230 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
234 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
236 return qdisc->flags & TCQ_F_ONETXQUEUE;
239 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
242 /* Non-BQL migrated drivers will return 0, too. */
243 return dql_avail(&txq->dql);
249 struct Qdisc_class_ops {
251 /* Child qdisc manipulation */
252 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
253 int (*graft)(struct Qdisc *, unsigned long cl,
254 struct Qdisc *, struct Qdisc **,
255 struct netlink_ext_ack *extack);
256 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
257 void (*qlen_notify)(struct Qdisc *, unsigned long);
259 /* Class manipulation routines */
260 unsigned long (*find)(struct Qdisc *, u32 classid);
261 int (*change)(struct Qdisc *, u32, u32,
262 struct nlattr **, unsigned long *,
263 struct netlink_ext_ack *);
264 int (*delete)(struct Qdisc *, unsigned long,
265 struct netlink_ext_ack *);
266 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
268 /* Filter manipulation */
269 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
271 struct netlink_ext_ack *extack);
272 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
274 void (*unbind_tcf)(struct Qdisc *, unsigned long);
276 /* rtnetlink specific */
277 int (*dump)(struct Qdisc *, unsigned long,
278 struct sk_buff *skb, struct tcmsg*);
279 int (*dump_stats)(struct Qdisc *, unsigned long,
283 /* Qdisc_class_ops flag values */
285 /* Implements API that doesn't require rtnl lock */
286 enum qdisc_class_ops_flags {
287 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
291 struct Qdisc_ops *next;
292 const struct Qdisc_class_ops *cl_ops;
295 unsigned int static_flags;
297 int (*enqueue)(struct sk_buff *skb,
299 struct sk_buff **to_free);
300 struct sk_buff * (*dequeue)(struct Qdisc *);
301 struct sk_buff * (*peek)(struct Qdisc *);
303 int (*init)(struct Qdisc *sch, struct nlattr *arg,
304 struct netlink_ext_ack *extack);
305 void (*reset)(struct Qdisc *);
306 void (*destroy)(struct Qdisc *);
307 int (*change)(struct Qdisc *sch,
309 struct netlink_ext_ack *extack);
310 void (*attach)(struct Qdisc *sch);
311 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
312 void (*change_real_num_tx)(struct Qdisc *sch,
313 unsigned int new_real_tx);
315 int (*dump)(struct Qdisc *, struct sk_buff *);
316 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
318 void (*ingress_block_set)(struct Qdisc *sch,
320 void (*egress_block_set)(struct Qdisc *sch,
322 u32 (*ingress_block_get)(struct Qdisc *sch);
323 u32 (*egress_block_get)(struct Qdisc *sch);
325 struct module *owner;
334 const struct tcf_proto *goto_tp;
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 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp,
381 /* rtnetlink specific */
382 int (*dump)(struct net*, struct tcf_proto*, void *,
383 struct sk_buff *skb, struct tcmsg*,
385 int (*terse_dump)(struct net *net,
386 struct tcf_proto *tp, void *fh,
388 struct tcmsg *t, bool rtnl_held);
389 int (*tmplt_dump)(struct sk_buff *skb,
393 struct module *owner;
397 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
398 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
399 * conditions can occur when filters are inserted/deleted simultaneously.
401 enum tcf_proto_ops_flags {
402 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
406 /* Fast access part */
407 struct tcf_proto __rcu *next;
410 /* called under RCU BH lock*/
411 int (*classify)(struct sk_buff *,
412 const struct tcf_proto *,
413 struct tcf_result *);
419 const struct tcf_proto_ops *ops;
420 struct tcf_chain *chain;
421 /* Lock protects tcf_proto shared state and can be used by unlocked
422 * classifiers to protect their private data.
428 struct hlist_node destroy_ht_node;
431 struct qdisc_skb_cb {
433 unsigned int pkt_len;
434 u16 slave_dev_queue_mapping;
437 #define QDISC_CB_PRIV_LEN 20
438 unsigned char data[QDISC_CB_PRIV_LEN];
441 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
444 /* Protects filter_chain. */
445 struct mutex filter_chain_lock;
446 struct tcf_proto __rcu *filter_chain;
447 struct list_head list;
448 struct tcf_block *block;
449 u32 index; /* chain index */
451 unsigned int action_refcnt;
452 bool explicitly_created;
454 const struct tcf_proto_ops *tmplt_ops;
460 struct xarray ports; /* datapath accessible */
461 /* Lock protects tcf_block and lifetime-management data of chains
462 * attached to the block (refcnt, action_refcnt, explicitly_created).
465 struct list_head chain_list;
466 u32 index; /* block index for shared blocks */
467 u32 classid; /* which class this block belongs to */
471 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
472 struct flow_block flow_block;
473 struct list_head owner_list;
475 atomic_t offloadcnt; /* Number of oddloaded filters */
476 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
477 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
479 struct tcf_chain *chain;
480 struct list_head filter_chain_list;
483 DECLARE_HASHTABLE(proto_destroy_ht, 7);
484 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
487 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index);
489 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
491 return lockdep_is_held(&chain->filter_chain_lock);
494 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
496 return lockdep_is_held(&tp->lock);
499 #define tcf_chain_dereference(p, chain) \
500 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
502 #define tcf_proto_dereference(p, tp) \
503 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
505 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
507 struct qdisc_skb_cb *qcb;
509 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
510 BUILD_BUG_ON(sizeof(qcb->data) < sz);
513 static inline int qdisc_qlen(const struct Qdisc *q)
518 static inline int qdisc_qlen_sum(const struct Qdisc *q)
520 __u32 qlen = q->qstats.qlen;
523 if (qdisc_is_percpu_stats(q)) {
524 for_each_possible_cpu(i)
525 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
533 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
535 return (struct qdisc_skb_cb *)skb->cb;
538 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
540 return &qdisc->q.lock;
543 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
545 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
550 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
552 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
555 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
557 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
560 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
562 struct Qdisc *root = qdisc_root_sleeping(qdisc);
565 return qdisc_lock(root);
568 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
570 return qdisc->dev_queue->dev;
573 static inline void sch_tree_lock(struct Qdisc *q)
575 if (q->flags & TCQ_F_MQROOT)
576 spin_lock_bh(qdisc_lock(q));
578 spin_lock_bh(qdisc_root_sleeping_lock(q));
581 static inline void sch_tree_unlock(struct Qdisc *q)
583 if (q->flags & TCQ_F_MQROOT)
584 spin_unlock_bh(qdisc_lock(q));
586 spin_unlock_bh(qdisc_root_sleeping_lock(q));
589 extern struct Qdisc noop_qdisc;
590 extern struct Qdisc_ops noop_qdisc_ops;
591 extern struct Qdisc_ops pfifo_fast_ops;
592 extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
593 extern struct Qdisc_ops mq_qdisc_ops;
594 extern struct Qdisc_ops noqueue_qdisc_ops;
595 extern const struct Qdisc_ops *default_qdisc_ops;
596 static inline const struct Qdisc_ops *
597 get_default_qdisc_ops(const struct net_device *dev, int ntx)
599 return ntx < dev->real_num_tx_queues ?
600 default_qdisc_ops : &pfifo_fast_ops;
603 struct Qdisc_class_common {
605 unsigned int filter_cnt;
606 struct hlist_node hnode;
609 struct Qdisc_class_hash {
610 struct hlist_head *hash;
611 unsigned int hashsize;
612 unsigned int hashmask;
613 unsigned int hashelems;
616 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
623 static inline struct Qdisc_class_common *
624 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
626 struct Qdisc_class_common *cl;
632 h = qdisc_class_hash(id, hash->hashmask);
633 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
634 if (cl->classid == id)
640 static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
642 return cl->filter_cnt > 0;
645 static inline void qdisc_class_get(struct Qdisc_class_common *cl)
649 if (check_add_overflow(cl->filter_cnt, 1, &res))
650 WARN(1, "Qdisc class overflow");
652 cl->filter_cnt = res;
655 static inline void qdisc_class_put(struct Qdisc_class_common *cl)
659 if (check_sub_overflow(cl->filter_cnt, 1, &res))
660 WARN(1, "Qdisc class underflow");
662 cl->filter_cnt = res;
665 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
667 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
669 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
672 int qdisc_class_hash_init(struct Qdisc_class_hash *);
673 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
674 struct Qdisc_class_common *);
675 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
676 struct Qdisc_class_common *);
677 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
678 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
680 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
681 void dev_qdisc_change_real_num_tx(struct net_device *dev,
682 unsigned int new_real_tx);
683 void dev_init_scheduler(struct net_device *dev);
684 void dev_shutdown(struct net_device *dev);
685 void dev_activate(struct net_device *dev);
686 void dev_deactivate(struct net_device *dev);
687 void dev_deactivate_many(struct list_head *head);
688 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
689 struct Qdisc *qdisc);
690 void qdisc_reset(struct Qdisc *qdisc);
691 void qdisc_destroy(struct Qdisc *qdisc);
692 void qdisc_put(struct Qdisc *qdisc);
693 void qdisc_put_unlocked(struct Qdisc *qdisc);
694 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
695 #ifdef CONFIG_NET_SCHED
696 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
698 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
699 struct Qdisc *new, struct Qdisc *old,
700 enum tc_setup_type type, void *type_data,
701 struct netlink_ext_ack *extack);
704 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
707 q->flags &= ~TCQ_F_OFFLOADED;
712 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
713 struct Qdisc *new, struct Qdisc *old,
714 enum tc_setup_type type, void *type_data,
715 struct netlink_ext_ack *extack)
719 void qdisc_offload_query_caps(struct net_device *dev,
720 enum tc_setup_type type,
721 void *caps, size_t caps_len);
722 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
723 const struct Qdisc_ops *ops,
724 struct netlink_ext_ack *extack);
725 void qdisc_free(struct Qdisc *qdisc);
726 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
727 const struct Qdisc_ops *ops, u32 parentid,
728 struct netlink_ext_ack *extack);
729 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
730 const struct qdisc_size_table *stab);
731 int skb_do_redirect(struct sk_buff *);
733 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
735 #ifdef CONFIG_NET_XGRESS
736 return skb->tc_at_ingress;
742 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
744 #ifdef CONFIG_NET_CLS_ACT
745 if (skb->tc_skip_classify) {
746 skb->tc_skip_classify = 0;
753 /* Reset all TX qdiscs greater than index of a device. */
754 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
758 for (; i < dev->num_tx_queues; i++) {
759 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
761 spin_lock_bh(qdisc_lock(qdisc));
763 spin_unlock_bh(qdisc_lock(qdisc));
768 /* Are all TX queues of the device empty? */
769 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
774 for (i = 0; i < dev->num_tx_queues; i++) {
775 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
776 const struct Qdisc *q = rcu_dereference(txq->qdisc);
778 if (!qdisc_is_empty(q)) {
787 /* Are any of the TX qdiscs changing? */
788 static inline bool qdisc_tx_changing(const struct net_device *dev)
792 for (i = 0; i < dev->num_tx_queues; i++) {
793 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
795 if (rcu_access_pointer(txq->qdisc) !=
796 rcu_access_pointer(txq->qdisc_sleeping))
802 /* Is the device using the noop qdisc on all queues? */
803 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
807 for (i = 0; i < dev->num_tx_queues; i++) {
808 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
809 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
815 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
817 return qdisc_skb_cb(skb)->pkt_len;
820 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
821 enum net_xmit_qdisc_t {
822 __NET_XMIT_STOLEN = 0x00010000,
823 __NET_XMIT_BYPASS = 0x00020000,
826 #ifdef CONFIG_NET_CLS_ACT
827 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
829 #define net_xmit_drop_count(e) (1)
832 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
833 const struct Qdisc *sch)
835 #ifdef CONFIG_NET_SCHED
836 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
839 __qdisc_calculate_pkt_len(skb, stab);
843 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
844 struct sk_buff **to_free)
846 qdisc_calculate_pkt_len(skb, sch);
847 return sch->enqueue(skb, sch, to_free);
850 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
851 __u64 bytes, __u32 packets)
853 u64_stats_update_begin(&bstats->syncp);
854 u64_stats_add(&bstats->bytes, bytes);
855 u64_stats_add(&bstats->packets, packets);
856 u64_stats_update_end(&bstats->syncp);
859 static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
860 const struct sk_buff *skb)
862 _bstats_update(bstats,
864 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
867 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
868 const struct sk_buff *skb)
870 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
873 static inline void qdisc_bstats_update(struct Qdisc *sch,
874 const struct sk_buff *skb)
876 bstats_update(&sch->bstats, skb);
879 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
880 const struct sk_buff *skb)
882 sch->qstats.backlog -= qdisc_pkt_len(skb);
885 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
886 const struct sk_buff *skb)
888 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
891 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
892 const struct sk_buff *skb)
894 sch->qstats.backlog += qdisc_pkt_len(skb);
897 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
898 const struct sk_buff *skb)
900 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
903 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
905 this_cpu_inc(sch->cpu_qstats->qlen);
908 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
910 this_cpu_dec(sch->cpu_qstats->qlen);
913 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
915 this_cpu_inc(sch->cpu_qstats->requeues);
918 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
920 sch->qstats.drops += count;
923 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
928 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
930 qstats->overlimits++;
933 static inline void qdisc_qstats_drop(struct Qdisc *sch)
935 qstats_drop_inc(&sch->qstats);
938 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
940 this_cpu_inc(sch->cpu_qstats->drops);
943 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
945 sch->qstats.overlimits++;
948 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
950 __u32 qlen = qdisc_qlen_sum(sch);
952 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
955 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
958 struct gnet_stats_queue qstats = { 0 };
960 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
961 *qlen = qstats.qlen + qdisc_qlen(sch);
962 *backlog = qstats.backlog;
965 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
969 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
970 qdisc_tree_reduce_backlog(sch, qlen, backlog);
973 static inline void qdisc_purge_queue(struct Qdisc *sch)
977 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
979 qdisc_tree_reduce_backlog(sch, qlen, backlog);
982 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
983 struct qdisc_skb_head *qh)
985 struct sk_buff *last = qh->tail;
998 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1000 __qdisc_enqueue_tail(skb, &sch->q);
1001 qdisc_qstats_backlog_inc(sch, skb);
1002 return NET_XMIT_SUCCESS;
1005 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1006 struct qdisc_skb_head *qh)
1008 skb->next = qh->head;
1016 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1018 struct sk_buff *skb = qh->head;
1020 if (likely(skb != NULL)) {
1021 qh->head = skb->next;
1023 if (qh->head == NULL)
1031 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1033 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1035 if (likely(skb != NULL)) {
1036 qdisc_qstats_backlog_dec(sch, skb);
1037 qdisc_bstats_update(sch, skb);
1044 struct qdisc_skb_cb qdisc_cb;
1047 u16 zone; /* Only valid if post_ct = true */
1054 static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
1056 struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
1058 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
1062 static inline enum skb_drop_reason
1063 tcf_get_drop_reason(const struct sk_buff *skb)
1065 return tc_skb_cb(skb)->drop_reason;
1068 static inline void tcf_set_drop_reason(const struct sk_buff *skb,
1069 enum skb_drop_reason reason)
1071 tc_skb_cb(skb)->drop_reason = reason;
1074 /* Instead of calling kfree_skb() while root qdisc lock is held,
1075 * queue the skb for future freeing at end of __dev_xmit_skb()
1077 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1079 skb->next = *to_free;
1083 static inline void __qdisc_drop_all(struct sk_buff *skb,
1084 struct sk_buff **to_free)
1087 skb->prev->next = *to_free;
1089 skb->next = *to_free;
1093 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1094 struct qdisc_skb_head *qh,
1095 struct sk_buff **to_free)
1097 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1099 if (likely(skb != NULL)) {
1100 unsigned int len = qdisc_pkt_len(skb);
1102 qdisc_qstats_backlog_dec(sch, skb);
1103 __qdisc_drop(skb, to_free);
1110 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1112 const struct qdisc_skb_head *qh = &sch->q;
1117 /* generic pseudo peek method for non-work-conserving qdisc */
1118 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1120 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1122 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1124 skb = sch->dequeue(sch);
1127 __skb_queue_head(&sch->gso_skb, skb);
1128 /* it's still part of the queue */
1129 qdisc_qstats_backlog_inc(sch, skb);
1137 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1138 struct sk_buff *skb)
1140 if (qdisc_is_percpu_stats(sch)) {
1141 qdisc_qstats_cpu_backlog_dec(sch, skb);
1142 qdisc_bstats_cpu_update(sch, skb);
1143 qdisc_qstats_cpu_qlen_dec(sch);
1145 qdisc_qstats_backlog_dec(sch, skb);
1146 qdisc_bstats_update(sch, skb);
1151 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1152 unsigned int pkt_len)
1154 if (qdisc_is_percpu_stats(sch)) {
1155 qdisc_qstats_cpu_qlen_inc(sch);
1156 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1158 sch->qstats.backlog += pkt_len;
1163 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1164 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1166 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1169 skb = __skb_dequeue(&sch->gso_skb);
1170 if (qdisc_is_percpu_stats(sch)) {
1171 qdisc_qstats_cpu_backlog_dec(sch, skb);
1172 qdisc_qstats_cpu_qlen_dec(sch);
1174 qdisc_qstats_backlog_dec(sch, skb);
1178 skb = sch->dequeue(sch);
1184 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1187 * We do not know the backlog in bytes of this list, it
1188 * is up to the caller to correct it
1192 rtnl_kfree_skbs(qh->head, qh->tail);
1200 static inline void qdisc_reset_queue(struct Qdisc *sch)
1202 __qdisc_reset_queue(&sch->q);
1205 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1206 struct Qdisc **pold)
1214 qdisc_purge_queue(old);
1215 sch_tree_unlock(sch);
1220 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1222 rtnl_kfree_skbs(skb, skb);
1223 qdisc_qstats_drop(sch);
1226 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1227 struct sk_buff **to_free)
1229 __qdisc_drop(skb, to_free);
1230 qdisc_qstats_cpu_drop(sch);
1232 return NET_XMIT_DROP;
1235 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1236 struct sk_buff **to_free)
1238 __qdisc_drop(skb, to_free);
1239 qdisc_qstats_drop(sch);
1241 return NET_XMIT_DROP;
1244 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1245 struct sk_buff **to_free)
1247 __qdisc_drop_all(skb, to_free);
1248 qdisc_qstats_drop(sch);
1250 return NET_XMIT_DROP;
1253 struct psched_ratecfg {
1254 u64 rate_bytes_ps; /* bytes per second */
1262 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1270 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1271 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1273 return ((u64)len * r->mult) >> r->shift;
1276 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1277 const struct tc_ratespec *conf,
1280 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1281 const struct psched_ratecfg *r)
1283 memset(res, 0, sizeof(*res));
1285 /* legacy struct tc_ratespec has a 32bit @rate field
1286 * Qdisc using 64bit rate should add new attributes
1287 * in order to maintain compatibility.
1289 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1291 res->overhead = r->overhead;
1293 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1296 struct psched_pktrate {
1297 u64 rate_pkts_ps; /* packets per second */
1302 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1303 unsigned int pkt_num)
1305 return ((u64)pkt_num * r->mult) >> r->shift;
1308 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1310 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1311 * The fast path only needs to access filter list and to update stats
1314 struct tcf_proto *filter_list;
1315 struct tcf_block *block;
1316 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1317 struct gnet_stats_queue __percpu *cpu_qstats;
1318 unsigned long rcu_state;
1321 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1322 const struct sk_buff *skb)
1324 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1327 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1329 this_cpu_inc(miniq->cpu_qstats->drops);
1332 struct mini_Qdisc_pair {
1333 struct mini_Qdisc miniq1;
1334 struct mini_Qdisc miniq2;
1335 struct mini_Qdisc __rcu **p_miniq;
1338 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1339 struct tcf_proto *tp_head);
1340 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1341 struct mini_Qdisc __rcu **p_miniq);
1342 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1343 struct tcf_block *block);
1345 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1347 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1349 /* Make sure qdisc is no longer in SCHED state. */
1350 static inline void qdisc_synchronize(const struct Qdisc *q)
1352 while (test_bit(__QDISC_STATE_SCHED, &q->state))