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 /* Intended to be used by unlocked users, when concurrent qdisc release is
144 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
146 if (qdisc->flags & TCQ_F_BUILTIN)
148 if (refcount_inc_not_zero(&qdisc->refcnt))
153 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
154 * root_lock section, or provide their own memory barriers -- ordering
155 * against qdisc_run_begin/end() atomic bit operations.
157 static inline bool qdisc_is_running(struct Qdisc *qdisc)
159 if (qdisc->flags & TCQ_F_NOLOCK)
160 return spin_is_locked(&qdisc->seqlock);
161 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
164 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
166 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
169 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
171 return q->flags & TCQ_F_CPUSTATS;
174 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
176 if (qdisc_is_percpu_stats(qdisc))
177 return nolock_qdisc_is_empty(qdisc);
178 return !READ_ONCE(qdisc->q.qlen);
181 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
182 * the qdisc root lock acquired.
184 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
186 if (qdisc->flags & TCQ_F_NOLOCK) {
187 if (spin_trylock(&qdisc->seqlock))
190 /* Paired with smp_mb__after_atomic() to make sure
191 * STATE_MISSED checking is synchronized with clearing
192 * in pfifo_fast_dequeue().
194 smp_mb__before_atomic();
196 /* If the MISSED flag is set, it means other thread has
197 * set the MISSED flag before second spin_trylock(), so
198 * we can return false here to avoid multi cpus doing
199 * the set_bit() and second spin_trylock() concurrently.
201 if (test_bit(__QDISC_STATE_MISSED, &qdisc->state))
204 /* Set the MISSED flag before the second spin_trylock(),
205 * if the second spin_trylock() return false, it means
206 * other cpu holding the lock will do dequeuing for us
207 * or it will see the MISSED flag set after releasing
208 * lock and reschedule the net_tx_action() to do the
211 set_bit(__QDISC_STATE_MISSED, &qdisc->state);
213 /* spin_trylock() only has load-acquire semantic, so use
214 * smp_mb__after_atomic() to ensure STATE_MISSED is set
215 * before doing the second spin_trylock().
217 smp_mb__after_atomic();
219 /* Retry again in case other CPU may not see the new flag
220 * after it releases the lock at the end of qdisc_run_end().
222 return spin_trylock(&qdisc->seqlock);
224 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
227 static inline void qdisc_run_end(struct Qdisc *qdisc)
229 if (qdisc->flags & TCQ_F_NOLOCK) {
230 spin_unlock(&qdisc->seqlock);
232 if (unlikely(test_bit(__QDISC_STATE_MISSED,
234 __netif_schedule(qdisc);
236 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
240 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
242 return qdisc->flags & TCQ_F_ONETXQUEUE;
245 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
248 /* Non-BQL migrated drivers will return 0, too. */
249 return dql_avail(&txq->dql);
255 struct Qdisc_class_ops {
257 /* Child qdisc manipulation */
258 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
259 int (*graft)(struct Qdisc *, unsigned long cl,
260 struct Qdisc *, struct Qdisc **,
261 struct netlink_ext_ack *extack);
262 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
263 void (*qlen_notify)(struct Qdisc *, unsigned long);
265 /* Class manipulation routines */
266 unsigned long (*find)(struct Qdisc *, u32 classid);
267 int (*change)(struct Qdisc *, u32, u32,
268 struct nlattr **, unsigned long *,
269 struct netlink_ext_ack *);
270 int (*delete)(struct Qdisc *, unsigned long,
271 struct netlink_ext_ack *);
272 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
274 /* Filter manipulation */
275 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
277 struct netlink_ext_ack *extack);
278 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
280 void (*unbind_tcf)(struct Qdisc *, unsigned long);
282 /* rtnetlink specific */
283 int (*dump)(struct Qdisc *, unsigned long,
284 struct sk_buff *skb, struct tcmsg*);
285 int (*dump_stats)(struct Qdisc *, unsigned long,
289 /* Qdisc_class_ops flag values */
291 /* Implements API that doesn't require rtnl lock */
292 enum qdisc_class_ops_flags {
293 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
297 struct Qdisc_ops *next;
298 const struct Qdisc_class_ops *cl_ops;
301 unsigned int static_flags;
303 int (*enqueue)(struct sk_buff *skb,
305 struct sk_buff **to_free);
306 struct sk_buff * (*dequeue)(struct Qdisc *);
307 struct sk_buff * (*peek)(struct Qdisc *);
309 int (*init)(struct Qdisc *sch, struct nlattr *arg,
310 struct netlink_ext_ack *extack);
311 void (*reset)(struct Qdisc *);
312 void (*destroy)(struct Qdisc *);
313 int (*change)(struct Qdisc *sch,
315 struct netlink_ext_ack *extack);
316 void (*attach)(struct Qdisc *sch);
317 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
318 void (*change_real_num_tx)(struct Qdisc *sch,
319 unsigned int new_real_tx);
321 int (*dump)(struct Qdisc *, struct sk_buff *);
322 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
324 void (*ingress_block_set)(struct Qdisc *sch,
326 void (*egress_block_set)(struct Qdisc *sch,
328 u32 (*ingress_block_get)(struct Qdisc *sch);
329 u32 (*egress_block_get)(struct Qdisc *sch);
331 struct module *owner;
341 const struct tcf_proto *goto_tp;
343 /* used in the skb_tc_reinsert function */
346 struct gnet_stats_queue *qstats;
353 struct tcf_proto_ops {
354 struct list_head head;
357 int (*classify)(struct sk_buff *,
358 const struct tcf_proto *,
359 struct tcf_result *);
360 int (*init)(struct tcf_proto*);
361 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
362 struct netlink_ext_ack *extack);
364 void* (*get)(struct tcf_proto*, u32 handle);
365 void (*put)(struct tcf_proto *tp, void *f);
366 int (*change)(struct net *net, struct sk_buff *,
367 struct tcf_proto*, unsigned long,
368 u32 handle, struct nlattr **,
370 struct netlink_ext_ack *);
371 int (*delete)(struct tcf_proto *tp, void *arg,
372 bool *last, bool rtnl_held,
373 struct netlink_ext_ack *);
374 bool (*delete_empty)(struct tcf_proto *tp);
375 void (*walk)(struct tcf_proto *tp,
376 struct tcf_walker *arg, bool rtnl_held);
377 int (*reoffload)(struct tcf_proto *tp, bool add,
378 flow_setup_cb_t *cb, void *cb_priv,
379 struct netlink_ext_ack *extack);
380 void (*hw_add)(struct tcf_proto *tp,
382 void (*hw_del)(struct tcf_proto *tp,
384 void (*bind_class)(void *, u32, unsigned long,
385 void *, unsigned long);
386 void * (*tmplt_create)(struct net *net,
387 struct tcf_chain *chain,
389 struct netlink_ext_ack *extack);
390 void (*tmplt_destroy)(void *tmplt_priv);
392 /* rtnetlink specific */
393 int (*dump)(struct net*, struct tcf_proto*, void *,
394 struct sk_buff *skb, struct tcmsg*,
396 int (*terse_dump)(struct net *net,
397 struct tcf_proto *tp, void *fh,
399 struct tcmsg *t, bool rtnl_held);
400 int (*tmplt_dump)(struct sk_buff *skb,
404 struct module *owner;
408 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
409 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
410 * conditions can occur when filters are inserted/deleted simultaneously.
412 enum tcf_proto_ops_flags {
413 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
417 /* Fast access part */
418 struct tcf_proto __rcu *next;
421 /* called under RCU BH lock*/
422 int (*classify)(struct sk_buff *,
423 const struct tcf_proto *,
424 struct tcf_result *);
430 const struct tcf_proto_ops *ops;
431 struct tcf_chain *chain;
432 /* Lock protects tcf_proto shared state and can be used by unlocked
433 * classifiers to protect their private data.
439 struct hlist_node destroy_ht_node;
442 struct qdisc_skb_cb {
444 unsigned int pkt_len;
445 u16 slave_dev_queue_mapping;
448 #define QDISC_CB_PRIV_LEN 20
449 unsigned char data[QDISC_CB_PRIV_LEN];
452 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
455 /* Protects filter_chain. */
456 struct mutex filter_chain_lock;
457 struct tcf_proto __rcu *filter_chain;
458 struct list_head list;
459 struct tcf_block *block;
460 u32 index; /* chain index */
462 unsigned int action_refcnt;
463 bool explicitly_created;
465 const struct tcf_proto_ops *tmplt_ops;
471 /* Lock protects tcf_block and lifetime-management data of chains
472 * attached to the block (refcnt, action_refcnt, explicitly_created).
475 struct list_head chain_list;
476 u32 index; /* block index for shared blocks */
477 u32 classid; /* which class this block belongs to */
481 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
482 struct flow_block flow_block;
483 struct list_head owner_list;
485 atomic_t offloadcnt; /* Number of oddloaded filters */
486 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
487 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
489 struct tcf_chain *chain;
490 struct list_head filter_chain_list;
493 DECLARE_HASHTABLE(proto_destroy_ht, 7);
494 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
497 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
499 return lockdep_is_held(&chain->filter_chain_lock);
502 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
504 return lockdep_is_held(&tp->lock);
507 #define tcf_chain_dereference(p, chain) \
508 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
510 #define tcf_proto_dereference(p, tp) \
511 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
513 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
515 struct qdisc_skb_cb *qcb;
517 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
518 BUILD_BUG_ON(sizeof(qcb->data) < sz);
521 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
523 return this_cpu_ptr(q->cpu_qstats)->qlen;
526 static inline int qdisc_qlen(const struct Qdisc *q)
531 static inline int qdisc_qlen_sum(const struct Qdisc *q)
533 __u32 qlen = q->qstats.qlen;
536 if (qdisc_is_percpu_stats(q)) {
537 for_each_possible_cpu(i)
538 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
546 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
548 return (struct qdisc_skb_cb *)skb->cb;
551 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
553 return &qdisc->q.lock;
556 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
558 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
563 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
565 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
568 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
570 return qdisc->dev_queue->qdisc_sleeping;
573 /* The qdisc root lock is a mechanism by which to top level
574 * of a qdisc tree can be locked from any qdisc node in the
575 * forest. This allows changing the configuration of some
576 * aspect of the qdisc tree while blocking out asynchronous
577 * qdisc access in the packet processing paths.
579 * It is only legal to do this when the root will not change
580 * on us. Otherwise we'll potentially lock the wrong qdisc
581 * root. This is enforced by holding the RTNL semaphore, which
582 * all users of this lock accessor must do.
584 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
586 struct Qdisc *root = qdisc_root(qdisc);
589 return qdisc_lock(root);
592 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
594 struct Qdisc *root = qdisc_root_sleeping(qdisc);
597 return qdisc_lock(root);
600 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
602 return qdisc->dev_queue->dev;
605 static inline void sch_tree_lock(struct Qdisc *q)
607 if (q->flags & TCQ_F_MQROOT)
608 spin_lock_bh(qdisc_lock(q));
610 spin_lock_bh(qdisc_root_sleeping_lock(q));
613 static inline void sch_tree_unlock(struct Qdisc *q)
615 if (q->flags & TCQ_F_MQROOT)
616 spin_unlock_bh(qdisc_lock(q));
618 spin_unlock_bh(qdisc_root_sleeping_lock(q));
621 extern struct Qdisc noop_qdisc;
622 extern struct Qdisc_ops noop_qdisc_ops;
623 extern struct Qdisc_ops pfifo_fast_ops;
624 extern struct Qdisc_ops mq_qdisc_ops;
625 extern struct Qdisc_ops noqueue_qdisc_ops;
626 extern const struct Qdisc_ops *default_qdisc_ops;
627 static inline const struct Qdisc_ops *
628 get_default_qdisc_ops(const struct net_device *dev, int ntx)
630 return ntx < dev->real_num_tx_queues ?
631 default_qdisc_ops : &pfifo_fast_ops;
634 struct Qdisc_class_common {
636 struct hlist_node hnode;
639 struct Qdisc_class_hash {
640 struct hlist_head *hash;
641 unsigned int hashsize;
642 unsigned int hashmask;
643 unsigned int hashelems;
646 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
653 static inline struct Qdisc_class_common *
654 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
656 struct Qdisc_class_common *cl;
662 h = qdisc_class_hash(id, hash->hashmask);
663 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
664 if (cl->classid == id)
670 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
672 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
674 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
677 int qdisc_class_hash_init(struct Qdisc_class_hash *);
678 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
679 struct Qdisc_class_common *);
680 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
681 struct Qdisc_class_common *);
682 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
683 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
685 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
686 void dev_qdisc_change_real_num_tx(struct net_device *dev,
687 unsigned int new_real_tx);
688 void dev_init_scheduler(struct net_device *dev);
689 void dev_shutdown(struct net_device *dev);
690 void dev_activate(struct net_device *dev);
691 void dev_deactivate(struct net_device *dev);
692 void dev_deactivate_many(struct list_head *head);
693 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
694 struct Qdisc *qdisc);
695 void qdisc_reset(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 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_CLS_ACT
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);
795 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
801 /* Is the device using the noop qdisc on all queues? */
802 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
806 for (i = 0; i < dev->num_tx_queues; i++) {
807 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
808 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
814 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
816 return qdisc_skb_cb(skb)->pkt_len;
819 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
820 enum net_xmit_qdisc_t {
821 __NET_XMIT_STOLEN = 0x00010000,
822 __NET_XMIT_BYPASS = 0x00020000,
825 #ifdef CONFIG_NET_CLS_ACT
826 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
828 #define net_xmit_drop_count(e) (1)
831 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
832 const struct Qdisc *sch)
834 #ifdef CONFIG_NET_SCHED
835 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
838 __qdisc_calculate_pkt_len(skb, stab);
842 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
843 struct sk_buff **to_free)
845 qdisc_calculate_pkt_len(skb, sch);
846 return sch->enqueue(skb, sch, to_free);
849 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
850 __u64 bytes, __u32 packets)
852 u64_stats_update_begin(&bstats->syncp);
853 u64_stats_add(&bstats->bytes, bytes);
854 u64_stats_add(&bstats->packets, packets);
855 u64_stats_update_end(&bstats->syncp);
858 static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
859 const struct sk_buff *skb)
861 _bstats_update(bstats,
863 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
866 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
867 const struct sk_buff *skb)
869 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
872 static inline void qdisc_bstats_update(struct Qdisc *sch,
873 const struct sk_buff *skb)
875 bstats_update(&sch->bstats, skb);
878 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
879 const struct sk_buff *skb)
881 sch->qstats.backlog -= qdisc_pkt_len(skb);
884 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
885 const struct sk_buff *skb)
887 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
890 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
891 const struct sk_buff *skb)
893 sch->qstats.backlog += qdisc_pkt_len(skb);
896 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
897 const struct sk_buff *skb)
899 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
902 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
904 this_cpu_inc(sch->cpu_qstats->qlen);
907 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
909 this_cpu_dec(sch->cpu_qstats->qlen);
912 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
914 this_cpu_inc(sch->cpu_qstats->requeues);
917 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
919 sch->qstats.drops += count;
922 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
927 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
929 qstats->overlimits++;
932 static inline void qdisc_qstats_drop(struct Qdisc *sch)
934 qstats_drop_inc(&sch->qstats);
937 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
939 this_cpu_inc(sch->cpu_qstats->drops);
942 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
944 sch->qstats.overlimits++;
947 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
949 __u32 qlen = qdisc_qlen_sum(sch);
951 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
954 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
957 struct gnet_stats_queue qstats = { 0 };
959 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
960 *qlen = qstats.qlen + qdisc_qlen(sch);
961 *backlog = qstats.backlog;
964 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
968 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
969 qdisc_tree_reduce_backlog(sch, qlen, backlog);
972 static inline void qdisc_purge_queue(struct Qdisc *sch)
976 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
978 qdisc_tree_reduce_backlog(sch, qlen, backlog);
981 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
988 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
989 struct qdisc_skb_head *qh)
991 struct sk_buff *last = qh->tail;
1004 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1006 __qdisc_enqueue_tail(skb, &sch->q);
1007 qdisc_qstats_backlog_inc(sch, skb);
1008 return NET_XMIT_SUCCESS;
1011 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1012 struct qdisc_skb_head *qh)
1014 skb->next = qh->head;
1022 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1024 struct sk_buff *skb = qh->head;
1026 if (likely(skb != NULL)) {
1027 qh->head = skb->next;
1029 if (qh->head == NULL)
1037 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1039 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1041 if (likely(skb != NULL)) {
1042 qdisc_qstats_backlog_dec(sch, skb);
1043 qdisc_bstats_update(sch, skb);
1049 /* Instead of calling kfree_skb() while root qdisc lock is held,
1050 * queue the skb for future freeing at end of __dev_xmit_skb()
1052 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1054 skb->next = *to_free;
1058 static inline void __qdisc_drop_all(struct sk_buff *skb,
1059 struct sk_buff **to_free)
1062 skb->prev->next = *to_free;
1064 skb->next = *to_free;
1068 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1069 struct qdisc_skb_head *qh,
1070 struct sk_buff **to_free)
1072 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1074 if (likely(skb != NULL)) {
1075 unsigned int len = qdisc_pkt_len(skb);
1077 qdisc_qstats_backlog_dec(sch, skb);
1078 __qdisc_drop(skb, to_free);
1085 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1087 const struct qdisc_skb_head *qh = &sch->q;
1092 /* generic pseudo peek method for non-work-conserving qdisc */
1093 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1095 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1097 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1099 skb = sch->dequeue(sch);
1102 __skb_queue_head(&sch->gso_skb, skb);
1103 /* it's still part of the queue */
1104 qdisc_qstats_backlog_inc(sch, skb);
1112 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1113 struct sk_buff *skb)
1115 if (qdisc_is_percpu_stats(sch)) {
1116 qdisc_qstats_cpu_backlog_dec(sch, skb);
1117 qdisc_bstats_cpu_update(sch, skb);
1118 qdisc_qstats_cpu_qlen_dec(sch);
1120 qdisc_qstats_backlog_dec(sch, skb);
1121 qdisc_bstats_update(sch, skb);
1126 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1127 unsigned int pkt_len)
1129 if (qdisc_is_percpu_stats(sch)) {
1130 qdisc_qstats_cpu_qlen_inc(sch);
1131 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1133 sch->qstats.backlog += pkt_len;
1138 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1139 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1141 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1144 skb = __skb_dequeue(&sch->gso_skb);
1145 if (qdisc_is_percpu_stats(sch)) {
1146 qdisc_qstats_cpu_backlog_dec(sch, skb);
1147 qdisc_qstats_cpu_qlen_dec(sch);
1149 qdisc_qstats_backlog_dec(sch, skb);
1153 skb = sch->dequeue(sch);
1159 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1162 * We do not know the backlog in bytes of this list, it
1163 * is up to the caller to correct it
1167 rtnl_kfree_skbs(qh->head, qh->tail);
1175 static inline void qdisc_reset_queue(struct Qdisc *sch)
1177 __qdisc_reset_queue(&sch->q);
1178 sch->qstats.backlog = 0;
1181 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1182 struct Qdisc **pold)
1190 qdisc_purge_queue(old);
1191 sch_tree_unlock(sch);
1196 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1198 rtnl_kfree_skbs(skb, skb);
1199 qdisc_qstats_drop(sch);
1202 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1203 struct sk_buff **to_free)
1205 __qdisc_drop(skb, to_free);
1206 qdisc_qstats_cpu_drop(sch);
1208 return NET_XMIT_DROP;
1211 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1212 struct sk_buff **to_free)
1214 __qdisc_drop(skb, to_free);
1215 qdisc_qstats_drop(sch);
1217 return NET_XMIT_DROP;
1220 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1221 struct sk_buff **to_free)
1223 __qdisc_drop_all(skb, to_free);
1224 qdisc_qstats_drop(sch);
1226 return NET_XMIT_DROP;
1229 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1230 long it will take to send a packet given its size.
1232 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1234 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1237 slot >>= rtab->rate.cell_log;
1239 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1240 return rtab->data[slot];
1243 struct psched_ratecfg {
1244 u64 rate_bytes_ps; /* bytes per second */
1252 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1260 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1261 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1263 return ((u64)len * r->mult) >> r->shift;
1266 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1267 const struct tc_ratespec *conf,
1270 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1271 const struct psched_ratecfg *r)
1273 memset(res, 0, sizeof(*res));
1275 /* legacy struct tc_ratespec has a 32bit @rate field
1276 * Qdisc using 64bit rate should add new attributes
1277 * in order to maintain compatibility.
1279 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1281 res->overhead = r->overhead;
1283 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1286 struct psched_pktrate {
1287 u64 rate_pkts_ps; /* packets per second */
1292 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1293 unsigned int pkt_num)
1295 return ((u64)pkt_num * r->mult) >> r->shift;
1298 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1300 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1301 * The fast path only needs to access filter list and to update stats
1304 struct tcf_proto *filter_list;
1305 struct tcf_block *block;
1306 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1307 struct gnet_stats_queue __percpu *cpu_qstats;
1308 unsigned long rcu_state;
1311 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1312 const struct sk_buff *skb)
1314 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1317 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1319 this_cpu_inc(miniq->cpu_qstats->drops);
1322 struct mini_Qdisc_pair {
1323 struct mini_Qdisc miniq1;
1324 struct mini_Qdisc miniq2;
1325 struct mini_Qdisc __rcu **p_miniq;
1328 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1329 struct tcf_proto *tp_head);
1330 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1331 struct mini_Qdisc __rcu **p_miniq);
1332 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1333 struct tcf_block *block);
1335 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1337 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));