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 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 struct xarray ports; /* datapath accessible */
465 /* Lock protects tcf_block and lifetime-management data of chains
466 * attached to the block (refcnt, action_refcnt, explicitly_created).
469 struct list_head chain_list;
470 u32 index; /* block index for shared blocks */
471 u32 classid; /* which class this block belongs to */
475 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
476 struct flow_block flow_block;
477 struct list_head owner_list;
479 atomic_t offloadcnt; /* Number of oddloaded filters */
480 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
481 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
483 struct tcf_chain *chain;
484 struct list_head filter_chain_list;
487 DECLARE_HASHTABLE(proto_destroy_ht, 7);
488 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
491 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index);
493 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
495 return lockdep_is_held(&chain->filter_chain_lock);
498 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
500 return lockdep_is_held(&tp->lock);
503 #define tcf_chain_dereference(p, chain) \
504 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
506 #define tcf_proto_dereference(p, tp) \
507 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
509 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
511 struct qdisc_skb_cb *qcb;
513 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
514 BUILD_BUG_ON(sizeof(qcb->data) < sz);
517 static inline int qdisc_qlen(const struct Qdisc *q)
522 static inline int qdisc_qlen_sum(const struct Qdisc *q)
524 __u32 qlen = q->qstats.qlen;
527 if (qdisc_is_percpu_stats(q)) {
528 for_each_possible_cpu(i)
529 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
537 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
539 return (struct qdisc_skb_cb *)skb->cb;
542 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
544 return &qdisc->q.lock;
547 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
549 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
554 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
556 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
559 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
561 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
564 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
566 struct Qdisc *root = qdisc_root_sleeping(qdisc);
569 return qdisc_lock(root);
572 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
574 return qdisc->dev_queue->dev;
577 static inline void sch_tree_lock(struct Qdisc *q)
579 if (q->flags & TCQ_F_MQROOT)
580 spin_lock_bh(qdisc_lock(q));
582 spin_lock_bh(qdisc_root_sleeping_lock(q));
585 static inline void sch_tree_unlock(struct Qdisc *q)
587 if (q->flags & TCQ_F_MQROOT)
588 spin_unlock_bh(qdisc_lock(q));
590 spin_unlock_bh(qdisc_root_sleeping_lock(q));
593 extern struct Qdisc noop_qdisc;
594 extern struct Qdisc_ops noop_qdisc_ops;
595 extern struct Qdisc_ops pfifo_fast_ops;
596 extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
597 extern struct Qdisc_ops mq_qdisc_ops;
598 extern struct Qdisc_ops noqueue_qdisc_ops;
599 extern const struct Qdisc_ops *default_qdisc_ops;
600 static inline const struct Qdisc_ops *
601 get_default_qdisc_ops(const struct net_device *dev, int ntx)
603 return ntx < dev->real_num_tx_queues ?
604 default_qdisc_ops : &pfifo_fast_ops;
607 struct Qdisc_class_common {
609 unsigned int filter_cnt;
610 struct hlist_node hnode;
613 struct Qdisc_class_hash {
614 struct hlist_head *hash;
615 unsigned int hashsize;
616 unsigned int hashmask;
617 unsigned int hashelems;
620 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
627 static inline struct Qdisc_class_common *
628 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
630 struct Qdisc_class_common *cl;
636 h = qdisc_class_hash(id, hash->hashmask);
637 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
638 if (cl->classid == id)
644 static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
646 return cl->filter_cnt > 0;
649 static inline void qdisc_class_get(struct Qdisc_class_common *cl)
653 if (check_add_overflow(cl->filter_cnt, 1, &res))
654 WARN(1, "Qdisc class overflow");
656 cl->filter_cnt = res;
659 static inline void qdisc_class_put(struct Qdisc_class_common *cl)
663 if (check_sub_overflow(cl->filter_cnt, 1, &res))
664 WARN(1, "Qdisc class underflow");
666 cl->filter_cnt = res;
669 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
671 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
673 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
676 int qdisc_class_hash_init(struct Qdisc_class_hash *);
677 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
678 struct Qdisc_class_common *);
679 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
680 struct Qdisc_class_common *);
681 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
682 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
684 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
685 void dev_qdisc_change_real_num_tx(struct net_device *dev,
686 unsigned int new_real_tx);
687 void dev_init_scheduler(struct net_device *dev);
688 void dev_shutdown(struct net_device *dev);
689 void dev_activate(struct net_device *dev);
690 void dev_deactivate(struct net_device *dev);
691 void dev_deactivate_many(struct list_head *head);
692 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
693 struct Qdisc *qdisc);
694 void qdisc_reset(struct Qdisc *qdisc);
695 void qdisc_destroy(struct Qdisc *qdisc);
696 void qdisc_put(struct Qdisc *qdisc);
697 void qdisc_put_unlocked(struct Qdisc *qdisc);
698 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
699 #ifdef CONFIG_NET_SCHED
700 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
702 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
703 struct Qdisc *new, struct Qdisc *old,
704 enum tc_setup_type type, void *type_data,
705 struct netlink_ext_ack *extack);
708 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
711 q->flags &= ~TCQ_F_OFFLOADED;
716 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
717 struct Qdisc *new, struct Qdisc *old,
718 enum tc_setup_type type, void *type_data,
719 struct netlink_ext_ack *extack)
723 void qdisc_offload_query_caps(struct net_device *dev,
724 enum tc_setup_type type,
725 void *caps, size_t caps_len);
726 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
727 const struct Qdisc_ops *ops,
728 struct netlink_ext_ack *extack);
729 void qdisc_free(struct Qdisc *qdisc);
730 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
731 const struct Qdisc_ops *ops, u32 parentid,
732 struct netlink_ext_ack *extack);
733 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
734 const struct qdisc_size_table *stab);
735 int skb_do_redirect(struct sk_buff *);
737 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
739 #ifdef CONFIG_NET_XGRESS
740 return skb->tc_at_ingress;
746 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
748 #ifdef CONFIG_NET_CLS_ACT
749 if (skb->tc_skip_classify) {
750 skb->tc_skip_classify = 0;
757 /* Reset all TX qdiscs greater than index of a device. */
758 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
762 for (; i < dev->num_tx_queues; i++) {
763 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
765 spin_lock_bh(qdisc_lock(qdisc));
767 spin_unlock_bh(qdisc_lock(qdisc));
772 /* Are all TX queues of the device empty? */
773 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
778 for (i = 0; i < dev->num_tx_queues; i++) {
779 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
780 const struct Qdisc *q = rcu_dereference(txq->qdisc);
782 if (!qdisc_is_empty(q)) {
791 /* Are any of the TX qdiscs changing? */
792 static inline bool qdisc_tx_changing(const struct net_device *dev)
796 for (i = 0; i < dev->num_tx_queues; i++) {
797 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
799 if (rcu_access_pointer(txq->qdisc) !=
800 rcu_access_pointer(txq->qdisc_sleeping))
806 /* Is the device using the noop qdisc on all queues? */
807 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
811 for (i = 0; i < dev->num_tx_queues; i++) {
812 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
813 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
819 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
821 return qdisc_skb_cb(skb)->pkt_len;
824 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
825 enum net_xmit_qdisc_t {
826 __NET_XMIT_STOLEN = 0x00010000,
827 __NET_XMIT_BYPASS = 0x00020000,
830 #ifdef CONFIG_NET_CLS_ACT
831 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
833 #define net_xmit_drop_count(e) (1)
836 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
837 const struct Qdisc *sch)
839 #ifdef CONFIG_NET_SCHED
840 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
843 __qdisc_calculate_pkt_len(skb, stab);
847 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
848 struct sk_buff **to_free)
850 qdisc_calculate_pkt_len(skb, sch);
851 return sch->enqueue(skb, sch, to_free);
854 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
855 __u64 bytes, __u32 packets)
857 u64_stats_update_begin(&bstats->syncp);
858 u64_stats_add(&bstats->bytes, bytes);
859 u64_stats_add(&bstats->packets, packets);
860 u64_stats_update_end(&bstats->syncp);
863 static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
864 const struct sk_buff *skb)
866 _bstats_update(bstats,
868 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
871 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
872 const struct sk_buff *skb)
874 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
877 static inline void qdisc_bstats_update(struct Qdisc *sch,
878 const struct sk_buff *skb)
880 bstats_update(&sch->bstats, skb);
883 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
884 const struct sk_buff *skb)
886 sch->qstats.backlog -= qdisc_pkt_len(skb);
889 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
890 const struct sk_buff *skb)
892 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
895 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
896 const struct sk_buff *skb)
898 sch->qstats.backlog += qdisc_pkt_len(skb);
901 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
902 const struct sk_buff *skb)
904 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
907 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
909 this_cpu_inc(sch->cpu_qstats->qlen);
912 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
914 this_cpu_dec(sch->cpu_qstats->qlen);
917 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
919 this_cpu_inc(sch->cpu_qstats->requeues);
922 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
924 sch->qstats.drops += count;
927 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
932 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
934 qstats->overlimits++;
937 static inline void qdisc_qstats_drop(struct Qdisc *sch)
939 qstats_drop_inc(&sch->qstats);
942 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
944 this_cpu_inc(sch->cpu_qstats->drops);
947 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
949 sch->qstats.overlimits++;
952 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
954 __u32 qlen = qdisc_qlen_sum(sch);
956 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
959 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
962 struct gnet_stats_queue qstats = { 0 };
964 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
965 *qlen = qstats.qlen + qdisc_qlen(sch);
966 *backlog = qstats.backlog;
969 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
973 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
974 qdisc_tree_reduce_backlog(sch, qlen, backlog);
977 static inline void qdisc_purge_queue(struct Qdisc *sch)
981 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
983 qdisc_tree_reduce_backlog(sch, qlen, backlog);
986 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
987 struct qdisc_skb_head *qh)
989 struct sk_buff *last = qh->tail;
1002 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1004 __qdisc_enqueue_tail(skb, &sch->q);
1005 qdisc_qstats_backlog_inc(sch, skb);
1006 return NET_XMIT_SUCCESS;
1009 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1010 struct qdisc_skb_head *qh)
1012 skb->next = qh->head;
1020 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1022 struct sk_buff *skb = qh->head;
1024 if (likely(skb != NULL)) {
1025 qh->head = skb->next;
1027 if (qh->head == NULL)
1035 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1037 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1039 if (likely(skb != NULL)) {
1040 qdisc_qstats_backlog_dec(sch, skb);
1041 qdisc_bstats_update(sch, skb);
1048 struct qdisc_skb_cb qdisc_cb;
1051 u16 zone; /* Only valid if post_ct = true */
1058 static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
1060 struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
1062 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
1066 static inline enum skb_drop_reason
1067 tcf_get_drop_reason(const struct sk_buff *skb)
1069 return tc_skb_cb(skb)->drop_reason;
1072 static inline void tcf_set_drop_reason(const struct sk_buff *skb,
1073 enum skb_drop_reason reason)
1075 tc_skb_cb(skb)->drop_reason = reason;
1078 /* Instead of calling kfree_skb() while root qdisc lock is held,
1079 * queue the skb for future freeing at end of __dev_xmit_skb()
1081 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1083 skb->next = *to_free;
1087 static inline void __qdisc_drop_all(struct sk_buff *skb,
1088 struct sk_buff **to_free)
1091 skb->prev->next = *to_free;
1093 skb->next = *to_free;
1097 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1098 struct qdisc_skb_head *qh,
1099 struct sk_buff **to_free)
1101 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1103 if (likely(skb != NULL)) {
1104 unsigned int len = qdisc_pkt_len(skb);
1106 qdisc_qstats_backlog_dec(sch, skb);
1107 __qdisc_drop(skb, to_free);
1114 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1116 const struct qdisc_skb_head *qh = &sch->q;
1121 /* generic pseudo peek method for non-work-conserving qdisc */
1122 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1124 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1126 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1128 skb = sch->dequeue(sch);
1131 __skb_queue_head(&sch->gso_skb, skb);
1132 /* it's still part of the queue */
1133 qdisc_qstats_backlog_inc(sch, skb);
1141 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1142 struct sk_buff *skb)
1144 if (qdisc_is_percpu_stats(sch)) {
1145 qdisc_qstats_cpu_backlog_dec(sch, skb);
1146 qdisc_bstats_cpu_update(sch, skb);
1147 qdisc_qstats_cpu_qlen_dec(sch);
1149 qdisc_qstats_backlog_dec(sch, skb);
1150 qdisc_bstats_update(sch, skb);
1155 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1156 unsigned int pkt_len)
1158 if (qdisc_is_percpu_stats(sch)) {
1159 qdisc_qstats_cpu_qlen_inc(sch);
1160 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1162 sch->qstats.backlog += pkt_len;
1167 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1168 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1170 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1173 skb = __skb_dequeue(&sch->gso_skb);
1174 if (qdisc_is_percpu_stats(sch)) {
1175 qdisc_qstats_cpu_backlog_dec(sch, skb);
1176 qdisc_qstats_cpu_qlen_dec(sch);
1178 qdisc_qstats_backlog_dec(sch, skb);
1182 skb = sch->dequeue(sch);
1188 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1191 * We do not know the backlog in bytes of this list, it
1192 * is up to the caller to correct it
1196 rtnl_kfree_skbs(qh->head, qh->tail);
1204 static inline void qdisc_reset_queue(struct Qdisc *sch)
1206 __qdisc_reset_queue(&sch->q);
1209 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1210 struct Qdisc **pold)
1218 qdisc_purge_queue(old);
1219 sch_tree_unlock(sch);
1224 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1226 rtnl_kfree_skbs(skb, skb);
1227 qdisc_qstats_drop(sch);
1230 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1231 struct sk_buff **to_free)
1233 __qdisc_drop(skb, to_free);
1234 qdisc_qstats_cpu_drop(sch);
1236 return NET_XMIT_DROP;
1239 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1240 struct sk_buff **to_free)
1242 __qdisc_drop(skb, to_free);
1243 qdisc_qstats_drop(sch);
1245 return NET_XMIT_DROP;
1248 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1249 struct sk_buff **to_free)
1251 __qdisc_drop_all(skb, to_free);
1252 qdisc_qstats_drop(sch);
1254 return NET_XMIT_DROP;
1257 struct psched_ratecfg {
1258 u64 rate_bytes_ps; /* bytes per second */
1266 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1274 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1275 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1277 return ((u64)len * r->mult) >> r->shift;
1280 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1281 const struct tc_ratespec *conf,
1284 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1285 const struct psched_ratecfg *r)
1287 memset(res, 0, sizeof(*res));
1289 /* legacy struct tc_ratespec has a 32bit @rate field
1290 * Qdisc using 64bit rate should add new attributes
1291 * in order to maintain compatibility.
1293 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1295 res->overhead = r->overhead;
1297 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1300 struct psched_pktrate {
1301 u64 rate_pkts_ps; /* packets per second */
1306 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1307 unsigned int pkt_num)
1309 return ((u64)pkt_num * r->mult) >> r->shift;
1312 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1314 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1315 * The fast path only needs to access filter list and to update stats
1318 struct tcf_proto *filter_list;
1319 struct tcf_block *block;
1320 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1321 struct gnet_stats_queue __percpu *cpu_qstats;
1322 unsigned long rcu_state;
1325 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1326 const struct sk_buff *skb)
1328 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1331 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1333 this_cpu_inc(miniq->cpu_qstats->drops);
1336 struct mini_Qdisc_pair {
1337 struct mini_Qdisc miniq1;
1338 struct mini_Qdisc miniq2;
1339 struct mini_Qdisc __rcu **p_miniq;
1342 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1343 struct tcf_proto *tp_head);
1344 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1345 struct mini_Qdisc __rcu **p_miniq);
1346 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1347 struct tcf_block *block);
1349 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1351 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1353 /* Make sure qdisc is no longer in SCHED state. */
1354 static inline void qdisc_synchronize(const struct Qdisc *q)
1356 while (test_bit(__QDISC_STATE_SCHED, &q->state))