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)
241 return netdev_queue_dql_avail(txq);
244 struct Qdisc_class_ops {
246 /* Child qdisc manipulation */
247 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
248 int (*graft)(struct Qdisc *, unsigned long cl,
249 struct Qdisc *, struct Qdisc **,
250 struct netlink_ext_ack *extack);
251 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
252 void (*qlen_notify)(struct Qdisc *, unsigned long);
254 /* Class manipulation routines */
255 unsigned long (*find)(struct Qdisc *, u32 classid);
256 int (*change)(struct Qdisc *, u32, u32,
257 struct nlattr **, unsigned long *,
258 struct netlink_ext_ack *);
259 int (*delete)(struct Qdisc *, unsigned long,
260 struct netlink_ext_ack *);
261 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
263 /* Filter manipulation */
264 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
266 struct netlink_ext_ack *extack);
267 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
269 void (*unbind_tcf)(struct Qdisc *, unsigned long);
271 /* rtnetlink specific */
272 int (*dump)(struct Qdisc *, unsigned long,
273 struct sk_buff *skb, struct tcmsg*);
274 int (*dump_stats)(struct Qdisc *, unsigned long,
278 /* Qdisc_class_ops flag values */
280 /* Implements API that doesn't require rtnl lock */
281 enum qdisc_class_ops_flags {
282 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
286 struct Qdisc_ops *next;
287 const struct Qdisc_class_ops *cl_ops;
290 unsigned int static_flags;
292 int (*enqueue)(struct sk_buff *skb,
294 struct sk_buff **to_free);
295 struct sk_buff * (*dequeue)(struct Qdisc *);
296 struct sk_buff * (*peek)(struct Qdisc *);
298 int (*init)(struct Qdisc *sch, struct nlattr *arg,
299 struct netlink_ext_ack *extack);
300 void (*reset)(struct Qdisc *);
301 void (*destroy)(struct Qdisc *);
302 int (*change)(struct Qdisc *sch,
304 struct netlink_ext_ack *extack);
305 void (*attach)(struct Qdisc *sch);
306 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
307 void (*change_real_num_tx)(struct Qdisc *sch,
308 unsigned int new_real_tx);
310 int (*dump)(struct Qdisc *, struct sk_buff *);
311 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
313 void (*ingress_block_set)(struct Qdisc *sch,
315 void (*egress_block_set)(struct Qdisc *sch,
317 u32 (*ingress_block_get)(struct Qdisc *sch);
318 u32 (*egress_block_get)(struct Qdisc *sch);
320 struct module *owner;
329 const struct tcf_proto *goto_tp;
335 struct tcf_proto_ops {
336 struct list_head head;
339 int (*classify)(struct sk_buff *,
340 const struct tcf_proto *,
341 struct tcf_result *);
342 int (*init)(struct tcf_proto*);
343 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
344 struct netlink_ext_ack *extack);
346 void* (*get)(struct tcf_proto*, u32 handle);
347 void (*put)(struct tcf_proto *tp, void *f);
348 int (*change)(struct net *net, struct sk_buff *,
349 struct tcf_proto*, unsigned long,
350 u32 handle, struct nlattr **,
352 struct netlink_ext_ack *);
353 int (*delete)(struct tcf_proto *tp, void *arg,
354 bool *last, bool rtnl_held,
355 struct netlink_ext_ack *);
356 bool (*delete_empty)(struct tcf_proto *tp);
357 void (*walk)(struct tcf_proto *tp,
358 struct tcf_walker *arg, bool rtnl_held);
359 int (*reoffload)(struct tcf_proto *tp, bool add,
360 flow_setup_cb_t *cb, void *cb_priv,
361 struct netlink_ext_ack *extack);
362 void (*hw_add)(struct tcf_proto *tp,
364 void (*hw_del)(struct tcf_proto *tp,
366 void (*bind_class)(void *, u32, unsigned long,
367 void *, unsigned long);
368 void * (*tmplt_create)(struct net *net,
369 struct tcf_chain *chain,
371 struct netlink_ext_ack *extack);
372 void (*tmplt_destroy)(void *tmplt_priv);
373 void (*tmplt_reoffload)(struct tcf_chain *chain,
377 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp,
380 /* rtnetlink specific */
381 int (*dump)(struct net*, struct tcf_proto*, void *,
382 struct sk_buff *skb, struct tcmsg*,
384 int (*terse_dump)(struct net *net,
385 struct tcf_proto *tp, void *fh,
387 struct tcmsg *t, bool rtnl_held);
388 int (*tmplt_dump)(struct sk_buff *skb,
392 struct module *owner;
396 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
397 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
398 * conditions can occur when filters are inserted/deleted simultaneously.
400 enum tcf_proto_ops_flags {
401 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
405 /* Fast access part */
406 struct tcf_proto __rcu *next;
409 /* called under RCU BH lock*/
410 int (*classify)(struct sk_buff *,
411 const struct tcf_proto *,
412 struct tcf_result *);
418 const struct tcf_proto_ops *ops;
419 struct tcf_chain *chain;
420 /* Lock protects tcf_proto shared state and can be used by unlocked
421 * classifiers to protect their private data.
427 struct hlist_node destroy_ht_node;
430 struct qdisc_skb_cb {
432 unsigned int pkt_len;
433 u16 slave_dev_queue_mapping;
436 #define QDISC_CB_PRIV_LEN 20
437 unsigned char data[QDISC_CB_PRIV_LEN];
440 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
443 /* Protects filter_chain. */
444 struct mutex filter_chain_lock;
445 struct tcf_proto __rcu *filter_chain;
446 struct list_head list;
447 struct tcf_block *block;
448 u32 index; /* chain index */
450 unsigned int action_refcnt;
451 bool explicitly_created;
453 const struct tcf_proto_ops *tmplt_ops;
459 struct xarray ports; /* datapath accessible */
460 /* Lock protects tcf_block and lifetime-management data of chains
461 * attached to the block (refcnt, action_refcnt, explicitly_created).
464 struct list_head chain_list;
465 u32 index; /* block index for shared blocks */
466 u32 classid; /* which class this block belongs to */
470 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
471 struct flow_block flow_block;
472 struct list_head owner_list;
474 atomic_t offloadcnt; /* Number of oddloaded filters */
475 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
476 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
478 struct tcf_chain *chain;
479 struct list_head filter_chain_list;
482 DECLARE_HASHTABLE(proto_destroy_ht, 7);
483 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
486 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index);
488 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
490 return lockdep_is_held(&chain->filter_chain_lock);
493 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
495 return lockdep_is_held(&tp->lock);
498 #define tcf_chain_dereference(p, chain) \
499 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
501 #define tcf_proto_dereference(p, tp) \
502 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
504 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
506 struct qdisc_skb_cb *qcb;
508 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
509 BUILD_BUG_ON(sizeof(qcb->data) < sz);
512 static inline int qdisc_qlen(const struct Qdisc *q)
517 static inline int qdisc_qlen_sum(const struct Qdisc *q)
519 __u32 qlen = q->qstats.qlen;
522 if (qdisc_is_percpu_stats(q)) {
523 for_each_possible_cpu(i)
524 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
532 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
534 return (struct qdisc_skb_cb *)skb->cb;
537 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
539 return &qdisc->q.lock;
542 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
544 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
549 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
551 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
554 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
556 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
559 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
561 struct Qdisc *root = qdisc_root_sleeping(qdisc);
564 return qdisc_lock(root);
567 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
569 return qdisc->dev_queue->dev;
572 static inline void sch_tree_lock(struct Qdisc *q)
574 if (q->flags & TCQ_F_MQROOT)
575 spin_lock_bh(qdisc_lock(q));
577 spin_lock_bh(qdisc_root_sleeping_lock(q));
580 static inline void sch_tree_unlock(struct Qdisc *q)
582 if (q->flags & TCQ_F_MQROOT)
583 spin_unlock_bh(qdisc_lock(q));
585 spin_unlock_bh(qdisc_root_sleeping_lock(q));
588 extern struct Qdisc noop_qdisc;
589 extern struct Qdisc_ops noop_qdisc_ops;
590 extern struct Qdisc_ops pfifo_fast_ops;
591 extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
592 extern struct Qdisc_ops mq_qdisc_ops;
593 extern struct Qdisc_ops noqueue_qdisc_ops;
594 extern const struct Qdisc_ops *default_qdisc_ops;
595 static inline const struct Qdisc_ops *
596 get_default_qdisc_ops(const struct net_device *dev, int ntx)
598 return ntx < dev->real_num_tx_queues ?
599 default_qdisc_ops : &pfifo_fast_ops;
602 struct Qdisc_class_common {
604 unsigned int filter_cnt;
605 struct hlist_node hnode;
608 struct Qdisc_class_hash {
609 struct hlist_head *hash;
610 unsigned int hashsize;
611 unsigned int hashmask;
612 unsigned int hashelems;
615 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
622 static inline struct Qdisc_class_common *
623 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
625 struct Qdisc_class_common *cl;
631 h = qdisc_class_hash(id, hash->hashmask);
632 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
633 if (cl->classid == id)
639 static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
641 return cl->filter_cnt > 0;
644 static inline void qdisc_class_get(struct Qdisc_class_common *cl)
648 if (check_add_overflow(cl->filter_cnt, 1, &res))
649 WARN(1, "Qdisc class overflow");
651 cl->filter_cnt = res;
654 static inline void qdisc_class_put(struct Qdisc_class_common *cl)
658 if (check_sub_overflow(cl->filter_cnt, 1, &res))
659 WARN(1, "Qdisc class underflow");
661 cl->filter_cnt = res;
664 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
666 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
668 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
671 int qdisc_class_hash_init(struct Qdisc_class_hash *);
672 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
673 struct Qdisc_class_common *);
674 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
675 struct Qdisc_class_common *);
676 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
677 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
679 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
680 void dev_qdisc_change_real_num_tx(struct net_device *dev,
681 unsigned int new_real_tx);
682 void dev_init_scheduler(struct net_device *dev);
683 void dev_shutdown(struct net_device *dev);
684 void dev_activate(struct net_device *dev);
685 void dev_deactivate(struct net_device *dev);
686 void dev_deactivate_many(struct list_head *head);
687 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
688 struct Qdisc *qdisc);
689 void qdisc_reset(struct Qdisc *qdisc);
690 void qdisc_destroy(struct Qdisc *qdisc);
691 void qdisc_put(struct Qdisc *qdisc);
692 void qdisc_put_unlocked(struct Qdisc *qdisc);
693 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
694 #ifdef CONFIG_NET_SCHED
695 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
697 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
698 struct Qdisc *new, struct Qdisc *old,
699 enum tc_setup_type type, void *type_data,
700 struct netlink_ext_ack *extack);
703 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
706 q->flags &= ~TCQ_F_OFFLOADED;
711 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
712 struct Qdisc *new, struct Qdisc *old,
713 enum tc_setup_type type, void *type_data,
714 struct netlink_ext_ack *extack)
718 void qdisc_offload_query_caps(struct net_device *dev,
719 enum tc_setup_type type,
720 void *caps, size_t caps_len);
721 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
722 const struct Qdisc_ops *ops,
723 struct netlink_ext_ack *extack);
724 void qdisc_free(struct Qdisc *qdisc);
725 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
726 const struct Qdisc_ops *ops, u32 parentid,
727 struct netlink_ext_ack *extack);
728 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
729 const struct qdisc_size_table *stab);
730 int skb_do_redirect(struct sk_buff *);
732 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
734 #ifdef CONFIG_NET_XGRESS
735 return skb->tc_at_ingress;
741 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
743 #ifdef CONFIG_NET_CLS_ACT
744 if (skb->tc_skip_classify) {
745 skb->tc_skip_classify = 0;
752 /* Reset all TX qdiscs greater than index of a device. */
753 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
757 for (; i < dev->num_tx_queues; i++) {
758 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
760 spin_lock_bh(qdisc_lock(qdisc));
762 spin_unlock_bh(qdisc_lock(qdisc));
767 /* Are all TX queues of the device empty? */
768 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
773 for (i = 0; i < dev->num_tx_queues; i++) {
774 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
775 const struct Qdisc *q = rcu_dereference(txq->qdisc);
777 if (!qdisc_is_empty(q)) {
786 /* Are any of the TX qdiscs changing? */
787 static inline bool qdisc_tx_changing(const struct net_device *dev)
791 for (i = 0; i < dev->num_tx_queues; i++) {
792 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
794 if (rcu_access_pointer(txq->qdisc) !=
795 rcu_access_pointer(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_enqueue_tail(struct sk_buff *skb,
982 struct qdisc_skb_head *qh)
984 struct sk_buff *last = qh->tail;
997 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
999 __qdisc_enqueue_tail(skb, &sch->q);
1000 qdisc_qstats_backlog_inc(sch, skb);
1001 return NET_XMIT_SUCCESS;
1004 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1005 struct qdisc_skb_head *qh)
1007 skb->next = qh->head;
1015 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1017 struct sk_buff *skb = qh->head;
1019 if (likely(skb != NULL)) {
1020 qh->head = skb->next;
1022 if (qh->head == NULL)
1030 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1032 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1034 if (likely(skb != NULL)) {
1035 qdisc_qstats_backlog_dec(sch, skb);
1036 qdisc_bstats_update(sch, skb);
1043 struct qdisc_skb_cb qdisc_cb;
1046 u16 zone; /* Only valid if post_ct = true */
1053 static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
1055 struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
1057 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
1061 static inline enum skb_drop_reason
1062 tcf_get_drop_reason(const struct sk_buff *skb)
1064 return tc_skb_cb(skb)->drop_reason;
1067 static inline void tcf_set_drop_reason(const struct sk_buff *skb,
1068 enum skb_drop_reason reason)
1070 tc_skb_cb(skb)->drop_reason = reason;
1073 /* Instead of calling kfree_skb() while root qdisc lock is held,
1074 * queue the skb for future freeing at end of __dev_xmit_skb()
1076 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1078 skb->next = *to_free;
1082 static inline void __qdisc_drop_all(struct sk_buff *skb,
1083 struct sk_buff **to_free)
1086 skb->prev->next = *to_free;
1088 skb->next = *to_free;
1092 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1093 struct qdisc_skb_head *qh,
1094 struct sk_buff **to_free)
1096 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1098 if (likely(skb != NULL)) {
1099 unsigned int len = qdisc_pkt_len(skb);
1101 qdisc_qstats_backlog_dec(sch, skb);
1102 __qdisc_drop(skb, to_free);
1109 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1111 const struct qdisc_skb_head *qh = &sch->q;
1116 /* generic pseudo peek method for non-work-conserving qdisc */
1117 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1119 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1121 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1123 skb = sch->dequeue(sch);
1126 __skb_queue_head(&sch->gso_skb, skb);
1127 /* it's still part of the queue */
1128 qdisc_qstats_backlog_inc(sch, skb);
1136 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1137 struct sk_buff *skb)
1139 if (qdisc_is_percpu_stats(sch)) {
1140 qdisc_qstats_cpu_backlog_dec(sch, skb);
1141 qdisc_bstats_cpu_update(sch, skb);
1142 qdisc_qstats_cpu_qlen_dec(sch);
1144 qdisc_qstats_backlog_dec(sch, skb);
1145 qdisc_bstats_update(sch, skb);
1150 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1151 unsigned int pkt_len)
1153 if (qdisc_is_percpu_stats(sch)) {
1154 qdisc_qstats_cpu_qlen_inc(sch);
1155 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1157 sch->qstats.backlog += pkt_len;
1162 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1163 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1165 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1168 skb = __skb_dequeue(&sch->gso_skb);
1169 if (qdisc_is_percpu_stats(sch)) {
1170 qdisc_qstats_cpu_backlog_dec(sch, skb);
1171 qdisc_qstats_cpu_qlen_dec(sch);
1173 qdisc_qstats_backlog_dec(sch, skb);
1177 skb = sch->dequeue(sch);
1183 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1186 * We do not know the backlog in bytes of this list, it
1187 * is up to the caller to correct it
1191 rtnl_kfree_skbs(qh->head, qh->tail);
1199 static inline void qdisc_reset_queue(struct Qdisc *sch)
1201 __qdisc_reset_queue(&sch->q);
1204 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1205 struct Qdisc **pold)
1213 qdisc_purge_queue(old);
1214 sch_tree_unlock(sch);
1219 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1221 rtnl_kfree_skbs(skb, skb);
1222 qdisc_qstats_drop(sch);
1225 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1226 struct sk_buff **to_free)
1228 __qdisc_drop(skb, to_free);
1229 qdisc_qstats_cpu_drop(sch);
1231 return NET_XMIT_DROP;
1234 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1235 struct sk_buff **to_free)
1237 __qdisc_drop(skb, to_free);
1238 qdisc_qstats_drop(sch);
1240 return NET_XMIT_DROP;
1243 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1244 struct sk_buff **to_free)
1246 __qdisc_drop_all(skb, to_free);
1247 qdisc_qstats_drop(sch);
1249 return NET_XMIT_DROP;
1252 struct psched_ratecfg {
1253 u64 rate_bytes_ps; /* bytes per second */
1261 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1269 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1270 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1272 return ((u64)len * r->mult) >> r->shift;
1275 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1276 const struct tc_ratespec *conf,
1279 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1280 const struct psched_ratecfg *r)
1282 memset(res, 0, sizeof(*res));
1284 /* legacy struct tc_ratespec has a 32bit @rate field
1285 * Qdisc using 64bit rate should add new attributes
1286 * in order to maintain compatibility.
1288 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1290 res->overhead = r->overhead;
1292 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1295 struct psched_pktrate {
1296 u64 rate_pkts_ps; /* packets per second */
1301 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1302 unsigned int pkt_num)
1304 return ((u64)pkt_num * r->mult) >> r->shift;
1307 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1309 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1310 * The fast path only needs to access filter list and to update stats
1313 struct tcf_proto *filter_list;
1314 struct tcf_block *block;
1315 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1316 struct gnet_stats_queue __percpu *cpu_qstats;
1317 unsigned long rcu_state;
1320 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1321 const struct sk_buff *skb)
1323 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1326 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1328 this_cpu_inc(miniq->cpu_qstats->drops);
1331 struct mini_Qdisc_pair {
1332 struct mini_Qdisc miniq1;
1333 struct mini_Qdisc miniq2;
1334 struct mini_Qdisc __rcu **p_miniq;
1337 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1338 struct tcf_proto *tp_head);
1339 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1340 struct mini_Qdisc __rcu **p_miniq);
1341 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1342 struct tcf_block *block);
1344 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1346 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1348 /* Make sure qdisc is no longer in SCHED state. */
1349 static inline void qdisc_synchronize(const struct Qdisc *q)
1351 while (test_bit(__QDISC_STATE_SCHED, &q->state))