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 <net/gen_stats.h>
19 #include <net/rtnetlink.h>
20 #include <net/flow_offload.h>
28 struct qdisc_rate_table {
29 struct tc_ratespec rate;
31 struct qdisc_rate_table *next;
37 __QDISC_STATE_DEACTIVATED,
40 struct qdisc_size_table {
42 struct list_head list;
43 struct tc_sizespec szopts;
48 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
49 struct qdisc_skb_head {
57 int (*enqueue)(struct sk_buff *skb,
59 struct sk_buff **to_free);
60 struct sk_buff * (*dequeue)(struct Qdisc *sch);
62 #define TCQ_F_BUILTIN 1
63 #define TCQ_F_INGRESS 2
64 #define TCQ_F_CAN_BYPASS 4
65 #define TCQ_F_MQROOT 8
66 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
67 * q->dev_queue : It can test
68 * netif_xmit_frozen_or_stopped() before
69 * dequeueing next packet.
70 * Its true for MQ/MQPRIO slaves, or non
73 #define TCQ_F_WARN_NONWC (1 << 16)
74 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
75 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
76 * qdisc_tree_decrease_qlen() should stop.
78 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
79 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
80 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
82 const struct Qdisc_ops *ops;
83 struct qdisc_size_table __rcu *stab;
84 struct hlist_node hash;
88 struct netdev_queue *dev_queue;
90 struct net_rate_estimator __rcu *rate_est;
91 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
92 struct gnet_stats_queue __percpu *cpu_qstats;
97 * For performance sake on SMP, we put highly modified fields at the end
99 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
100 struct qdisc_skb_head q;
101 struct gnet_stats_basic_packed bstats;
103 struct gnet_stats_queue qstats;
105 struct Qdisc *next_sched;
106 struct sk_buff_head skb_bad_txq;
108 spinlock_t busylock ____cacheline_aligned_in_smp;
111 /* for NOLOCK qdisc, true if there are no enqueued skbs */
116 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
118 if (qdisc->flags & TCQ_F_BUILTIN)
120 refcount_inc(&qdisc->refcnt);
123 /* Intended to be used by unlocked users, when concurrent qdisc release is
127 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
129 if (qdisc->flags & TCQ_F_BUILTIN)
131 if (refcount_inc_not_zero(&qdisc->refcnt))
136 static inline bool qdisc_is_running(struct Qdisc *qdisc)
138 if (qdisc->flags & TCQ_F_NOLOCK)
139 return spin_is_locked(&qdisc->seqlock);
140 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
143 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
145 return q->flags & TCQ_F_CPUSTATS;
148 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
150 if (qdisc_is_percpu_stats(qdisc))
152 return !qdisc->q.qlen;
155 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
157 if (qdisc->flags & TCQ_F_NOLOCK) {
158 if (!spin_trylock(&qdisc->seqlock))
160 qdisc->empty = false;
161 } else if (qdisc_is_running(qdisc)) {
164 /* Variant of write_seqcount_begin() telling lockdep a trylock
167 raw_write_seqcount_begin(&qdisc->running);
168 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
172 static inline void qdisc_run_end(struct Qdisc *qdisc)
174 write_seqcount_end(&qdisc->running);
175 if (qdisc->flags & TCQ_F_NOLOCK)
176 spin_unlock(&qdisc->seqlock);
179 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
181 return qdisc->flags & TCQ_F_ONETXQUEUE;
184 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
187 /* Non-BQL migrated drivers will return 0, too. */
188 return dql_avail(&txq->dql);
194 struct Qdisc_class_ops {
196 /* Child qdisc manipulation */
197 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
198 int (*graft)(struct Qdisc *, unsigned long cl,
199 struct Qdisc *, struct Qdisc **,
200 struct netlink_ext_ack *extack);
201 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
202 void (*qlen_notify)(struct Qdisc *, unsigned long);
204 /* Class manipulation routines */
205 unsigned long (*find)(struct Qdisc *, u32 classid);
206 int (*change)(struct Qdisc *, u32, u32,
207 struct nlattr **, unsigned long *,
208 struct netlink_ext_ack *);
209 int (*delete)(struct Qdisc *, unsigned long);
210 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
212 /* Filter manipulation */
213 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
215 struct netlink_ext_ack *extack);
216 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
218 void (*unbind_tcf)(struct Qdisc *, unsigned long);
220 /* rtnetlink specific */
221 int (*dump)(struct Qdisc *, unsigned long,
222 struct sk_buff *skb, struct tcmsg*);
223 int (*dump_stats)(struct Qdisc *, unsigned long,
227 /* Qdisc_class_ops flag values */
229 /* Implements API that doesn't require rtnl lock */
230 enum qdisc_class_ops_flags {
231 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
235 struct Qdisc_ops *next;
236 const struct Qdisc_class_ops *cl_ops;
239 unsigned int static_flags;
241 int (*enqueue)(struct sk_buff *skb,
243 struct sk_buff **to_free);
244 struct sk_buff * (*dequeue)(struct Qdisc *);
245 struct sk_buff * (*peek)(struct Qdisc *);
247 int (*init)(struct Qdisc *sch, struct nlattr *arg,
248 struct netlink_ext_ack *extack);
249 void (*reset)(struct Qdisc *);
250 void (*destroy)(struct Qdisc *);
251 int (*change)(struct Qdisc *sch,
253 struct netlink_ext_ack *extack);
254 void (*attach)(struct Qdisc *sch);
255 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
257 int (*dump)(struct Qdisc *, struct sk_buff *);
258 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
260 void (*ingress_block_set)(struct Qdisc *sch,
262 void (*egress_block_set)(struct Qdisc *sch,
264 u32 (*ingress_block_get)(struct Qdisc *sch);
265 u32 (*egress_block_get)(struct Qdisc *sch);
267 struct module *owner;
277 const struct tcf_proto *goto_tp;
279 /* used in the skb_tc_reinsert function */
282 struct gnet_stats_queue *qstats;
289 struct tcf_proto_ops {
290 struct list_head head;
293 int (*classify)(struct sk_buff *,
294 const struct tcf_proto *,
295 struct tcf_result *);
296 int (*init)(struct tcf_proto*);
297 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
298 struct netlink_ext_ack *extack);
300 void* (*get)(struct tcf_proto*, u32 handle);
301 void (*put)(struct tcf_proto *tp, void *f);
302 int (*change)(struct net *net, struct sk_buff *,
303 struct tcf_proto*, unsigned long,
304 u32 handle, struct nlattr **,
306 struct netlink_ext_ack *);
307 int (*delete)(struct tcf_proto *tp, void *arg,
308 bool *last, bool rtnl_held,
309 struct netlink_ext_ack *);
310 void (*walk)(struct tcf_proto *tp,
311 struct tcf_walker *arg, bool rtnl_held);
312 int (*reoffload)(struct tcf_proto *tp, bool add,
313 flow_setup_cb_t *cb, void *cb_priv,
314 struct netlink_ext_ack *extack);
315 void (*hw_add)(struct tcf_proto *tp,
317 void (*hw_del)(struct tcf_proto *tp,
319 void (*bind_class)(void *, u32, unsigned long);
320 void * (*tmplt_create)(struct net *net,
321 struct tcf_chain *chain,
323 struct netlink_ext_ack *extack);
324 void (*tmplt_destroy)(void *tmplt_priv);
326 /* rtnetlink specific */
327 int (*dump)(struct net*, struct tcf_proto*, void *,
328 struct sk_buff *skb, struct tcmsg*,
330 int (*tmplt_dump)(struct sk_buff *skb,
334 struct module *owner;
338 enum tcf_proto_ops_flags {
339 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
343 /* Fast access part */
344 struct tcf_proto __rcu *next;
347 /* called under RCU BH lock*/
348 int (*classify)(struct sk_buff *,
349 const struct tcf_proto *,
350 struct tcf_result *);
356 const struct tcf_proto_ops *ops;
357 struct tcf_chain *chain;
358 /* Lock protects tcf_proto shared state and can be used by unlocked
359 * classifiers to protect their private data.
367 struct qdisc_skb_cb {
369 unsigned int pkt_len;
370 u16 slave_dev_queue_mapping;
373 #define QDISC_CB_PRIV_LEN 20
374 unsigned char data[QDISC_CB_PRIV_LEN];
377 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
380 /* Protects filter_chain. */
381 struct mutex filter_chain_lock;
382 struct tcf_proto __rcu *filter_chain;
383 struct list_head list;
384 struct tcf_block *block;
385 u32 index; /* chain index */
387 unsigned int action_refcnt;
388 bool explicitly_created;
390 const struct tcf_proto_ops *tmplt_ops;
396 /* Lock protects tcf_block and lifetime-management data of chains
397 * attached to the block (refcnt, action_refcnt, explicitly_created).
400 struct list_head chain_list;
401 u32 index; /* block index for shared blocks */
405 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
406 struct flow_block flow_block;
407 struct list_head owner_list;
409 atomic_t offloadcnt; /* Number of oddloaded filters */
410 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
411 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
413 struct tcf_chain *chain;
414 struct list_head filter_chain_list;
419 #ifdef CONFIG_PROVE_LOCKING
420 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
422 return lockdep_is_held(&chain->filter_chain_lock);
425 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
427 return lockdep_is_held(&tp->lock);
430 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain)
435 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
439 #endif /* #ifdef CONFIG_PROVE_LOCKING */
441 #define tcf_chain_dereference(p, chain) \
442 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
444 #define tcf_proto_dereference(p, tp) \
445 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
447 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
449 struct qdisc_skb_cb *qcb;
451 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
452 BUILD_BUG_ON(sizeof(qcb->data) < sz);
455 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
457 return this_cpu_ptr(q->cpu_qstats)->qlen;
460 static inline int qdisc_qlen(const struct Qdisc *q)
465 static inline int qdisc_qlen_sum(const struct Qdisc *q)
467 __u32 qlen = q->qstats.qlen;
470 if (qdisc_is_percpu_stats(q)) {
471 for_each_possible_cpu(i)
472 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
480 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
482 return (struct qdisc_skb_cb *)skb->cb;
485 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
487 return &qdisc->q.lock;
490 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
492 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
497 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
499 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
502 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
504 return qdisc->dev_queue->qdisc_sleeping;
507 /* The qdisc root lock is a mechanism by which to top level
508 * of a qdisc tree can be locked from any qdisc node in the
509 * forest. This allows changing the configuration of some
510 * aspect of the qdisc tree while blocking out asynchronous
511 * qdisc access in the packet processing paths.
513 * It is only legal to do this when the root will not change
514 * on us. Otherwise we'll potentially lock the wrong qdisc
515 * root. This is enforced by holding the RTNL semaphore, which
516 * all users of this lock accessor must do.
518 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
520 struct Qdisc *root = qdisc_root(qdisc);
523 return qdisc_lock(root);
526 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
528 struct Qdisc *root = qdisc_root_sleeping(qdisc);
531 return qdisc_lock(root);
534 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
536 struct Qdisc *root = qdisc_root_sleeping(qdisc);
539 return &root->running;
542 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
544 return qdisc->dev_queue->dev;
547 static inline void sch_tree_lock(const struct Qdisc *q)
549 spin_lock_bh(qdisc_root_sleeping_lock(q));
552 static inline void sch_tree_unlock(const struct Qdisc *q)
554 spin_unlock_bh(qdisc_root_sleeping_lock(q));
557 extern struct Qdisc noop_qdisc;
558 extern struct Qdisc_ops noop_qdisc_ops;
559 extern struct Qdisc_ops pfifo_fast_ops;
560 extern struct Qdisc_ops mq_qdisc_ops;
561 extern struct Qdisc_ops noqueue_qdisc_ops;
562 extern const struct Qdisc_ops *default_qdisc_ops;
563 static inline const struct Qdisc_ops *
564 get_default_qdisc_ops(const struct net_device *dev, int ntx)
566 return ntx < dev->real_num_tx_queues ?
567 default_qdisc_ops : &pfifo_fast_ops;
570 struct Qdisc_class_common {
572 struct hlist_node hnode;
575 struct Qdisc_class_hash {
576 struct hlist_head *hash;
577 unsigned int hashsize;
578 unsigned int hashmask;
579 unsigned int hashelems;
582 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
589 static inline struct Qdisc_class_common *
590 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
592 struct Qdisc_class_common *cl;
598 h = qdisc_class_hash(id, hash->hashmask);
599 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
600 if (cl->classid == id)
606 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
608 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
610 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
613 int qdisc_class_hash_init(struct Qdisc_class_hash *);
614 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
615 struct Qdisc_class_common *);
616 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
617 struct Qdisc_class_common *);
618 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
619 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
621 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
622 void dev_init_scheduler(struct net_device *dev);
623 void dev_shutdown(struct net_device *dev);
624 void dev_activate(struct net_device *dev);
625 void dev_deactivate(struct net_device *dev);
626 void dev_deactivate_many(struct list_head *head);
627 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
628 struct Qdisc *qdisc);
629 void qdisc_reset(struct Qdisc *qdisc);
630 void qdisc_put(struct Qdisc *qdisc);
631 void qdisc_put_unlocked(struct Qdisc *qdisc);
632 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
633 #ifdef CONFIG_NET_SCHED
634 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
636 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
637 struct Qdisc *new, struct Qdisc *old,
638 enum tc_setup_type type, void *type_data,
639 struct netlink_ext_ack *extack);
642 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
645 q->flags &= ~TCQ_F_OFFLOADED;
650 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
651 struct Qdisc *new, struct Qdisc *old,
652 enum tc_setup_type type, void *type_data,
653 struct netlink_ext_ack *extack)
657 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
658 const struct Qdisc_ops *ops,
659 struct netlink_ext_ack *extack);
660 void qdisc_free(struct Qdisc *qdisc);
661 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
662 const struct Qdisc_ops *ops, u32 parentid,
663 struct netlink_ext_ack *extack);
664 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
665 const struct qdisc_size_table *stab);
666 int skb_do_redirect(struct sk_buff *);
668 static inline void skb_reset_tc(struct sk_buff *skb)
670 #ifdef CONFIG_NET_CLS_ACT
671 skb->tc_redirected = 0;
675 static inline bool skb_is_tc_redirected(const struct sk_buff *skb)
677 #ifdef CONFIG_NET_CLS_ACT
678 return skb->tc_redirected;
684 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
686 #ifdef CONFIG_NET_CLS_ACT
687 return skb->tc_at_ingress;
693 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
695 #ifdef CONFIG_NET_CLS_ACT
696 if (skb->tc_skip_classify) {
697 skb->tc_skip_classify = 0;
704 /* Reset all TX qdiscs greater than index of a device. */
705 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
709 for (; i < dev->num_tx_queues; i++) {
710 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
712 spin_lock_bh(qdisc_lock(qdisc));
714 spin_unlock_bh(qdisc_lock(qdisc));
719 static inline void qdisc_reset_all_tx(struct net_device *dev)
721 qdisc_reset_all_tx_gt(dev, 0);
724 /* Are all TX queues of the device empty? */
725 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
730 for (i = 0; i < dev->num_tx_queues; i++) {
731 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
732 const struct Qdisc *q = rcu_dereference(txq->qdisc);
734 if (!qdisc_is_empty(q)) {
743 /* Are any of the TX qdiscs changing? */
744 static inline bool qdisc_tx_changing(const struct net_device *dev)
748 for (i = 0; i < dev->num_tx_queues; i++) {
749 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
750 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
756 /* Is the device using the noop qdisc on all queues? */
757 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
761 for (i = 0; i < dev->num_tx_queues; i++) {
762 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
763 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
769 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
771 return qdisc_skb_cb(skb)->pkt_len;
774 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
775 enum net_xmit_qdisc_t {
776 __NET_XMIT_STOLEN = 0x00010000,
777 __NET_XMIT_BYPASS = 0x00020000,
780 #ifdef CONFIG_NET_CLS_ACT
781 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
783 #define net_xmit_drop_count(e) (1)
786 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
787 const struct Qdisc *sch)
789 #ifdef CONFIG_NET_SCHED
790 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
793 __qdisc_calculate_pkt_len(skb, stab);
797 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
798 struct sk_buff **to_free)
800 qdisc_calculate_pkt_len(skb, sch);
801 return sch->enqueue(skb, sch, to_free);
804 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
805 __u64 bytes, __u32 packets)
807 bstats->bytes += bytes;
808 bstats->packets += packets;
811 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
812 const struct sk_buff *skb)
814 _bstats_update(bstats,
816 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
819 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
820 __u64 bytes, __u32 packets)
822 u64_stats_update_begin(&bstats->syncp);
823 _bstats_update(&bstats->bstats, bytes, packets);
824 u64_stats_update_end(&bstats->syncp);
827 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
828 const struct sk_buff *skb)
830 u64_stats_update_begin(&bstats->syncp);
831 bstats_update(&bstats->bstats, skb);
832 u64_stats_update_end(&bstats->syncp);
835 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
836 const struct sk_buff *skb)
838 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
841 static inline void qdisc_bstats_update(struct Qdisc *sch,
842 const struct sk_buff *skb)
844 bstats_update(&sch->bstats, skb);
847 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
848 const struct sk_buff *skb)
850 sch->qstats.backlog -= qdisc_pkt_len(skb);
853 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
854 const struct sk_buff *skb)
856 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
859 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
860 const struct sk_buff *skb)
862 sch->qstats.backlog += qdisc_pkt_len(skb);
865 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
866 const struct sk_buff *skb)
868 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
871 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
873 this_cpu_inc(sch->cpu_qstats->qlen);
876 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
878 this_cpu_dec(sch->cpu_qstats->qlen);
881 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
883 this_cpu_inc(sch->cpu_qstats->requeues);
886 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
888 sch->qstats.drops += count;
891 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
896 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
898 qstats->overlimits++;
901 static inline void qdisc_qstats_drop(struct Qdisc *sch)
903 qstats_drop_inc(&sch->qstats);
906 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
908 this_cpu_inc(sch->cpu_qstats->drops);
911 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
913 sch->qstats.overlimits++;
916 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
918 __u32 qlen = qdisc_qlen_sum(sch);
920 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
923 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
926 struct gnet_stats_queue qstats = { 0 };
927 __u32 len = qdisc_qlen_sum(sch);
929 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
931 *backlog = qstats.backlog;
934 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
938 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
939 qdisc_tree_reduce_backlog(sch, qlen, backlog);
942 static inline void qdisc_purge_queue(struct Qdisc *sch)
946 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
948 qdisc_tree_reduce_backlog(sch, qlen, backlog);
951 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
958 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
959 struct qdisc_skb_head *qh)
961 struct sk_buff *last = qh->tail;
974 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
976 __qdisc_enqueue_tail(skb, &sch->q);
977 qdisc_qstats_backlog_inc(sch, skb);
978 return NET_XMIT_SUCCESS;
981 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
982 struct qdisc_skb_head *qh)
984 skb->next = qh->head;
992 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
994 struct sk_buff *skb = qh->head;
996 if (likely(skb != NULL)) {
997 qh->head = skb->next;
999 if (qh->head == NULL)
1007 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1009 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1011 if (likely(skb != NULL)) {
1012 qdisc_qstats_backlog_dec(sch, skb);
1013 qdisc_bstats_update(sch, skb);
1019 /* Instead of calling kfree_skb() while root qdisc lock is held,
1020 * queue the skb for future freeing at end of __dev_xmit_skb()
1022 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1024 skb->next = *to_free;
1028 static inline void __qdisc_drop_all(struct sk_buff *skb,
1029 struct sk_buff **to_free)
1032 skb->prev->next = *to_free;
1034 skb->next = *to_free;
1038 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1039 struct qdisc_skb_head *qh,
1040 struct sk_buff **to_free)
1042 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1044 if (likely(skb != NULL)) {
1045 unsigned int len = qdisc_pkt_len(skb);
1047 qdisc_qstats_backlog_dec(sch, skb);
1048 __qdisc_drop(skb, to_free);
1055 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
1056 struct sk_buff **to_free)
1058 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
1061 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1063 const struct qdisc_skb_head *qh = &sch->q;
1068 /* generic pseudo peek method for non-work-conserving qdisc */
1069 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1071 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1073 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1075 skb = sch->dequeue(sch);
1078 __skb_queue_head(&sch->gso_skb, skb);
1079 /* it's still part of the queue */
1080 qdisc_qstats_backlog_inc(sch, skb);
1088 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1089 struct sk_buff *skb)
1091 if (qdisc_is_percpu_stats(sch)) {
1092 qdisc_qstats_cpu_backlog_dec(sch, skb);
1093 qdisc_bstats_cpu_update(sch, skb);
1094 qdisc_qstats_cpu_qlen_dec(sch);
1096 qdisc_qstats_backlog_dec(sch, skb);
1097 qdisc_bstats_update(sch, skb);
1102 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1103 unsigned int pkt_len)
1105 if (qdisc_is_percpu_stats(sch)) {
1106 qdisc_qstats_cpu_qlen_inc(sch);
1107 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1109 sch->qstats.backlog += pkt_len;
1114 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1115 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1117 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1120 skb = __skb_dequeue(&sch->gso_skb);
1121 if (qdisc_is_percpu_stats(sch)) {
1122 qdisc_qstats_cpu_backlog_dec(sch, skb);
1123 qdisc_qstats_cpu_qlen_dec(sch);
1125 qdisc_qstats_backlog_dec(sch, skb);
1129 skb = sch->dequeue(sch);
1135 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1138 * We do not know the backlog in bytes of this list, it
1139 * is up to the caller to correct it
1143 rtnl_kfree_skbs(qh->head, qh->tail);
1151 static inline void qdisc_reset_queue(struct Qdisc *sch)
1153 __qdisc_reset_queue(&sch->q);
1154 sch->qstats.backlog = 0;
1157 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1158 struct Qdisc **pold)
1166 qdisc_tree_flush_backlog(old);
1167 sch_tree_unlock(sch);
1172 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1174 rtnl_kfree_skbs(skb, skb);
1175 qdisc_qstats_drop(sch);
1178 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1179 struct sk_buff **to_free)
1181 __qdisc_drop(skb, to_free);
1182 qdisc_qstats_cpu_drop(sch);
1184 return NET_XMIT_DROP;
1187 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1188 struct sk_buff **to_free)
1190 __qdisc_drop(skb, to_free);
1191 qdisc_qstats_drop(sch);
1193 return NET_XMIT_DROP;
1196 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1197 struct sk_buff **to_free)
1199 __qdisc_drop_all(skb, to_free);
1200 qdisc_qstats_drop(sch);
1202 return NET_XMIT_DROP;
1205 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1206 long it will take to send a packet given its size.
1208 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1210 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1213 slot >>= rtab->rate.cell_log;
1215 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1216 return rtab->data[slot];
1219 struct psched_ratecfg {
1220 u64 rate_bytes_ps; /* bytes per second */
1227 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1232 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1233 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1235 return ((u64)len * r->mult) >> r->shift;
1238 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1239 const struct tc_ratespec *conf,
1242 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1243 const struct psched_ratecfg *r)
1245 memset(res, 0, sizeof(*res));
1247 /* legacy struct tc_ratespec has a 32bit @rate field
1248 * Qdisc using 64bit rate should add new attributes
1249 * in order to maintain compatibility.
1251 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1253 res->overhead = r->overhead;
1254 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1257 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1258 * The fast path only needs to access filter list and to update stats
1261 struct tcf_proto *filter_list;
1262 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1263 struct gnet_stats_queue __percpu *cpu_qstats;
1264 struct rcu_head rcu;
1267 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1268 const struct sk_buff *skb)
1270 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1273 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1275 this_cpu_inc(miniq->cpu_qstats->drops);
1278 struct mini_Qdisc_pair {
1279 struct mini_Qdisc miniq1;
1280 struct mini_Qdisc miniq2;
1281 struct mini_Qdisc __rcu **p_miniq;
1284 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1285 struct tcf_proto *tp_head);
1286 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1287 struct mini_Qdisc __rcu **p_miniq);
1289 static inline void skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res)
1291 struct gnet_stats_queue *stats = res->qstats;
1295 ret = netif_receive_skb(skb);
1297 ret = dev_queue_xmit(skb);
1299 qstats_overlimit_inc(res->qstats);