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 <net/gen_stats.h>
17 #include <net/rtnetlink.h>
25 typedef int tc_setup_cb_t(enum tc_setup_type type,
26 void *type_data, void *cb_priv);
28 typedef int tc_indr_block_bind_cb_t(struct net_device *dev, void *cb_priv,
29 enum tc_setup_type type, void *type_data);
31 struct qdisc_rate_table {
32 struct tc_ratespec rate;
34 struct qdisc_rate_table *next;
40 __QDISC_STATE_DEACTIVATED,
43 struct qdisc_size_table {
45 struct list_head list;
46 struct tc_sizespec szopts;
51 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
52 struct qdisc_skb_head {
63 int (*enqueue)(struct sk_buff *skb,
65 struct sk_buff **to_free);
66 struct sk_buff * (*dequeue)(struct Qdisc *sch);
68 #define TCQ_F_BUILTIN 1
69 #define TCQ_F_INGRESS 2
70 #define TCQ_F_CAN_BYPASS 4
71 #define TCQ_F_MQROOT 8
72 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
73 * q->dev_queue : It can test
74 * netif_xmit_frozen_or_stopped() before
75 * dequeueing next packet.
76 * Its true for MQ/MQPRIO slaves, or non
79 #define TCQ_F_WARN_NONWC (1 << 16)
80 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
81 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
82 * qdisc_tree_decrease_qlen() should stop.
84 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
85 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
86 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
88 const struct Qdisc_ops *ops;
89 struct qdisc_size_table __rcu *stab;
90 struct hlist_node hash;
94 struct netdev_queue *dev_queue;
96 struct net_rate_estimator __rcu *rate_est;
97 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
98 struct gnet_stats_queue __percpu *cpu_qstats;
103 * For performance sake on SMP, we put highly modified fields at the end
105 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
106 struct qdisc_skb_head q;
107 struct gnet_stats_basic_packed bstats;
109 struct gnet_stats_queue qstats;
111 struct Qdisc *next_sched;
112 struct sk_buff_head skb_bad_txq;
114 spinlock_t busylock ____cacheline_aligned_in_smp;
117 /* for NOLOCK qdisc, true if there are no enqueued skbs */
122 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
124 if (qdisc->flags & TCQ_F_BUILTIN)
126 refcount_inc(&qdisc->refcnt);
129 /* Intended to be used by unlocked users, when concurrent qdisc release is
133 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
135 if (qdisc->flags & TCQ_F_BUILTIN)
137 if (refcount_inc_not_zero(&qdisc->refcnt))
142 static inline bool qdisc_is_running(struct Qdisc *qdisc)
144 if (qdisc->flags & TCQ_F_NOLOCK)
145 return spin_is_locked(&qdisc->seqlock);
146 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
149 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
151 if (qdisc->flags & TCQ_F_NOLOCK)
153 return !qdisc->q.qlen;
156 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
158 if (qdisc->flags & TCQ_F_NOLOCK) {
159 if (!spin_trylock(&qdisc->seqlock))
161 qdisc->empty = false;
162 } else if (qdisc_is_running(qdisc)) {
165 /* Variant of write_seqcount_begin() telling lockdep a trylock
168 raw_write_seqcount_begin(&qdisc->running);
169 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
173 static inline void qdisc_run_end(struct Qdisc *qdisc)
175 write_seqcount_end(&qdisc->running);
176 if (qdisc->flags & TCQ_F_NOLOCK)
177 spin_unlock(&qdisc->seqlock);
180 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
182 return qdisc->flags & TCQ_F_ONETXQUEUE;
185 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
188 /* Non-BQL migrated drivers will return 0, too. */
189 return dql_avail(&txq->dql);
195 struct Qdisc_class_ops {
197 /* Child qdisc manipulation */
198 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
199 int (*graft)(struct Qdisc *, unsigned long cl,
200 struct Qdisc *, struct Qdisc **,
201 struct netlink_ext_ack *extack);
202 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
203 void (*qlen_notify)(struct Qdisc *, unsigned long);
205 /* Class manipulation routines */
206 unsigned long (*find)(struct Qdisc *, u32 classid);
207 int (*change)(struct Qdisc *, u32, u32,
208 struct nlattr **, unsigned long *,
209 struct netlink_ext_ack *);
210 int (*delete)(struct Qdisc *, unsigned long);
211 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
213 /* Filter manipulation */
214 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
216 struct netlink_ext_ack *extack);
217 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
219 void (*unbind_tcf)(struct Qdisc *, unsigned long);
221 /* rtnetlink specific */
222 int (*dump)(struct Qdisc *, unsigned long,
223 struct sk_buff *skb, struct tcmsg*);
224 int (*dump_stats)(struct Qdisc *, unsigned long,
228 /* Qdisc_class_ops flag values */
230 /* Implements API that doesn't require rtnl lock */
231 enum qdisc_class_ops_flags {
232 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
236 struct Qdisc_ops *next;
237 const struct Qdisc_class_ops *cl_ops;
240 unsigned int static_flags;
242 int (*enqueue)(struct sk_buff *skb,
244 struct sk_buff **to_free);
245 struct sk_buff * (*dequeue)(struct Qdisc *);
246 struct sk_buff * (*peek)(struct Qdisc *);
248 int (*init)(struct Qdisc *sch, struct nlattr *arg,
249 struct netlink_ext_ack *extack);
250 void (*reset)(struct Qdisc *);
251 void (*destroy)(struct Qdisc *);
252 int (*change)(struct Qdisc *sch,
254 struct netlink_ext_ack *extack);
255 void (*attach)(struct Qdisc *sch);
256 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
258 int (*dump)(struct Qdisc *, struct sk_buff *);
259 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
261 void (*ingress_block_set)(struct Qdisc *sch,
263 void (*egress_block_set)(struct Qdisc *sch,
265 u32 (*ingress_block_get)(struct Qdisc *sch);
266 u32 (*egress_block_get)(struct Qdisc *sch);
268 struct module *owner;
278 const struct tcf_proto *goto_tp;
280 /* used by the TC_ACT_REINSERT action */
283 struct gnet_stats_queue *qstats;
290 struct tcf_proto_ops {
291 struct list_head head;
294 int (*classify)(struct sk_buff *,
295 const struct tcf_proto *,
296 struct tcf_result *);
297 int (*init)(struct tcf_proto*);
298 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
299 struct netlink_ext_ack *extack);
301 void* (*get)(struct tcf_proto*, u32 handle);
302 void (*put)(struct tcf_proto *tp, void *f);
303 int (*change)(struct net *net, struct sk_buff *,
304 struct tcf_proto*, unsigned long,
305 u32 handle, struct nlattr **,
307 struct netlink_ext_ack *);
308 int (*delete)(struct tcf_proto *tp, void *arg,
309 bool *last, bool rtnl_held,
310 struct netlink_ext_ack *);
311 void (*walk)(struct tcf_proto *tp,
312 struct tcf_walker *arg, bool rtnl_held);
313 int (*reoffload)(struct tcf_proto *tp, bool add,
314 tc_setup_cb_t *cb, void *cb_priv,
315 struct netlink_ext_ack *extack);
316 void (*bind_class)(void *, u32, unsigned long);
317 void * (*tmplt_create)(struct net *net,
318 struct tcf_chain *chain,
320 struct netlink_ext_ack *extack);
321 void (*tmplt_destroy)(void *tmplt_priv);
323 /* rtnetlink specific */
324 int (*dump)(struct net*, struct tcf_proto*, void *,
325 struct sk_buff *skb, struct tcmsg*,
327 int (*tmplt_dump)(struct sk_buff *skb,
331 struct module *owner;
335 enum tcf_proto_ops_flags {
336 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
340 /* Fast access part */
341 struct tcf_proto __rcu *next;
344 /* called under RCU BH lock*/
345 int (*classify)(struct sk_buff *,
346 const struct tcf_proto *,
347 struct tcf_result *);
353 const struct tcf_proto_ops *ops;
354 struct tcf_chain *chain;
355 /* Lock protects tcf_proto shared state and can be used by unlocked
356 * classifiers to protect their private data.
364 struct qdisc_skb_cb {
367 unsigned int pkt_len;
368 u16 slave_dev_queue_mapping;
371 struct bpf_flow_keys *flow_keys;
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 list_head cb_list;
406 struct list_head owner_list;
408 unsigned int offloadcnt; /* Number of oddloaded filters */
409 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
411 struct tcf_chain *chain;
412 struct list_head filter_chain_list;
417 #ifdef CONFIG_PROVE_LOCKING
418 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
420 return lockdep_is_held(&chain->filter_chain_lock);
423 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
425 return lockdep_is_held(&tp->lock);
428 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain)
433 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
437 #endif /* #ifdef CONFIG_PROVE_LOCKING */
439 #define tcf_chain_dereference(p, chain) \
440 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
442 #define tcf_proto_dereference(p, tp) \
443 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
445 static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
447 if (*flags & TCA_CLS_FLAGS_IN_HW)
449 *flags |= TCA_CLS_FLAGS_IN_HW;
453 static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
455 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
457 *flags &= ~TCA_CLS_FLAGS_IN_HW;
462 tc_cls_offload_cnt_update(struct tcf_block *block, u32 *cnt,
463 u32 *flags, bool add)
467 tcf_block_offload_inc(block, flags);
472 tcf_block_offload_dec(block, flags);
476 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
478 struct qdisc_skb_cb *qcb;
480 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
481 BUILD_BUG_ON(sizeof(qcb->data) < sz);
484 static inline int qdisc_qlen(const struct Qdisc *q)
489 static inline u32 qdisc_qlen_sum(const struct Qdisc *q)
491 u32 qlen = q->qstats.qlen;
493 if (q->flags & TCQ_F_NOLOCK)
494 qlen += atomic_read(&q->q.atomic_qlen);
501 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
503 return (struct qdisc_skb_cb *)skb->cb;
506 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
508 return &qdisc->q.lock;
511 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
513 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
518 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
520 return qdisc->dev_queue->qdisc_sleeping;
523 /* The qdisc root lock is a mechanism by which to top level
524 * of a qdisc tree can be locked from any qdisc node in the
525 * forest. This allows changing the configuration of some
526 * aspect of the qdisc tree while blocking out asynchronous
527 * qdisc access in the packet processing paths.
529 * It is only legal to do this when the root will not change
530 * on us. Otherwise we'll potentially lock the wrong qdisc
531 * root. This is enforced by holding the RTNL semaphore, which
532 * all users of this lock accessor must do.
534 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
536 struct Qdisc *root = qdisc_root(qdisc);
539 return qdisc_lock(root);
542 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
544 struct Qdisc *root = qdisc_root_sleeping(qdisc);
547 return qdisc_lock(root);
550 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
552 struct Qdisc *root = qdisc_root_sleeping(qdisc);
555 return &root->running;
558 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
560 return qdisc->dev_queue->dev;
563 static inline void sch_tree_lock(const struct Qdisc *q)
565 spin_lock_bh(qdisc_root_sleeping_lock(q));
568 static inline void sch_tree_unlock(const struct Qdisc *q)
570 spin_unlock_bh(qdisc_root_sleeping_lock(q));
573 extern struct Qdisc noop_qdisc;
574 extern struct Qdisc_ops noop_qdisc_ops;
575 extern struct Qdisc_ops pfifo_fast_ops;
576 extern struct Qdisc_ops mq_qdisc_ops;
577 extern struct Qdisc_ops noqueue_qdisc_ops;
578 extern const struct Qdisc_ops *default_qdisc_ops;
579 static inline const struct Qdisc_ops *
580 get_default_qdisc_ops(const struct net_device *dev, int ntx)
582 return ntx < dev->real_num_tx_queues ?
583 default_qdisc_ops : &pfifo_fast_ops;
586 struct Qdisc_class_common {
588 struct hlist_node hnode;
591 struct Qdisc_class_hash {
592 struct hlist_head *hash;
593 unsigned int hashsize;
594 unsigned int hashmask;
595 unsigned int hashelems;
598 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
605 static inline struct Qdisc_class_common *
606 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
608 struct Qdisc_class_common *cl;
614 h = qdisc_class_hash(id, hash->hashmask);
615 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
616 if (cl->classid == id)
622 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
624 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
626 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
629 int qdisc_class_hash_init(struct Qdisc_class_hash *);
630 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
631 struct Qdisc_class_common *);
632 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
633 struct Qdisc_class_common *);
634 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
635 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
637 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
638 void dev_init_scheduler(struct net_device *dev);
639 void dev_shutdown(struct net_device *dev);
640 void dev_activate(struct net_device *dev);
641 void dev_deactivate(struct net_device *dev);
642 void dev_deactivate_many(struct list_head *head);
643 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
644 struct Qdisc *qdisc);
645 void qdisc_reset(struct Qdisc *qdisc);
646 void qdisc_put(struct Qdisc *qdisc);
647 void qdisc_put_unlocked(struct Qdisc *qdisc);
648 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
649 #ifdef CONFIG_NET_SCHED
650 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
652 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
653 struct Qdisc *new, struct Qdisc *old,
654 enum tc_setup_type type, void *type_data,
655 struct netlink_ext_ack *extack);
658 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
661 q->flags &= ~TCQ_F_OFFLOADED;
666 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
667 struct Qdisc *new, struct Qdisc *old,
668 enum tc_setup_type type, void *type_data,
669 struct netlink_ext_ack *extack)
673 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
674 const struct Qdisc_ops *ops,
675 struct netlink_ext_ack *extack);
676 void qdisc_free(struct Qdisc *qdisc);
677 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
678 const struct Qdisc_ops *ops, u32 parentid,
679 struct netlink_ext_ack *extack);
680 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
681 const struct qdisc_size_table *stab);
682 int skb_do_redirect(struct sk_buff *);
684 static inline void skb_reset_tc(struct sk_buff *skb)
686 #ifdef CONFIG_NET_CLS_ACT
687 skb->tc_redirected = 0;
691 static inline bool skb_is_tc_redirected(const struct sk_buff *skb)
693 #ifdef CONFIG_NET_CLS_ACT
694 return skb->tc_redirected;
700 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
702 #ifdef CONFIG_NET_CLS_ACT
703 return skb->tc_at_ingress;
709 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
711 #ifdef CONFIG_NET_CLS_ACT
712 if (skb->tc_skip_classify) {
713 skb->tc_skip_classify = 0;
720 /* Reset all TX qdiscs greater than index of a device. */
721 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
725 for (; i < dev->num_tx_queues; i++) {
726 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
728 spin_lock_bh(qdisc_lock(qdisc));
730 spin_unlock_bh(qdisc_lock(qdisc));
735 static inline void qdisc_reset_all_tx(struct net_device *dev)
737 qdisc_reset_all_tx_gt(dev, 0);
740 /* Are all TX queues of the device empty? */
741 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
746 for (i = 0; i < dev->num_tx_queues; i++) {
747 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
748 const struct Qdisc *q = rcu_dereference(txq->qdisc);
750 if (!qdisc_is_empty(q)) {
759 /* Are any of the TX qdiscs changing? */
760 static inline bool qdisc_tx_changing(const struct net_device *dev)
764 for (i = 0; i < dev->num_tx_queues; i++) {
765 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
766 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
772 /* Is the device using the noop qdisc on all queues? */
773 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
777 for (i = 0; i < dev->num_tx_queues; i++) {
778 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
779 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
785 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
787 return qdisc_skb_cb(skb)->pkt_len;
790 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
791 enum net_xmit_qdisc_t {
792 __NET_XMIT_STOLEN = 0x00010000,
793 __NET_XMIT_BYPASS = 0x00020000,
796 #ifdef CONFIG_NET_CLS_ACT
797 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
799 #define net_xmit_drop_count(e) (1)
802 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
803 const struct Qdisc *sch)
805 #ifdef CONFIG_NET_SCHED
806 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
809 __qdisc_calculate_pkt_len(skb, stab);
813 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
814 struct sk_buff **to_free)
816 qdisc_calculate_pkt_len(skb, sch);
817 return sch->enqueue(skb, sch, to_free);
820 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
822 return q->flags & TCQ_F_CPUSTATS;
825 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
826 __u64 bytes, __u32 packets)
828 bstats->bytes += bytes;
829 bstats->packets += packets;
832 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
833 const struct sk_buff *skb)
835 _bstats_update(bstats,
837 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
840 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
841 __u64 bytes, __u32 packets)
843 u64_stats_update_begin(&bstats->syncp);
844 _bstats_update(&bstats->bstats, bytes, packets);
845 u64_stats_update_end(&bstats->syncp);
848 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
849 const struct sk_buff *skb)
851 u64_stats_update_begin(&bstats->syncp);
852 bstats_update(&bstats->bstats, skb);
853 u64_stats_update_end(&bstats->syncp);
856 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
857 const struct sk_buff *skb)
859 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
862 static inline void qdisc_bstats_update(struct Qdisc *sch,
863 const struct sk_buff *skb)
865 bstats_update(&sch->bstats, skb);
868 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
869 const struct sk_buff *skb)
871 sch->qstats.backlog -= qdisc_pkt_len(skb);
874 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
875 const struct sk_buff *skb)
877 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
880 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
881 const struct sk_buff *skb)
883 sch->qstats.backlog += qdisc_pkt_len(skb);
886 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
887 const struct sk_buff *skb)
889 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
892 static inline void qdisc_qstats_atomic_qlen_inc(struct Qdisc *sch)
894 atomic_inc(&sch->q.atomic_qlen);
897 static inline void qdisc_qstats_atomic_qlen_dec(struct Qdisc *sch)
899 atomic_dec(&sch->q.atomic_qlen);
902 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
904 this_cpu_inc(sch->cpu_qstats->requeues);
907 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
909 sch->qstats.drops += count;
912 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
917 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
919 qstats->overlimits++;
922 static inline void qdisc_qstats_drop(struct Qdisc *sch)
924 qstats_drop_inc(&sch->qstats);
927 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
929 this_cpu_inc(sch->cpu_qstats->drops);
932 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
934 sch->qstats.overlimits++;
937 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
939 __u32 qlen = qdisc_qlen_sum(sch);
941 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
944 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
947 struct gnet_stats_queue qstats = { 0 };
948 __u32 len = qdisc_qlen_sum(sch);
950 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
952 *backlog = qstats.backlog;
955 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
959 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
960 qdisc_tree_reduce_backlog(sch, qlen, backlog);
963 static inline void qdisc_purge_queue(struct Qdisc *sch)
967 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
969 qdisc_tree_reduce_backlog(sch, qlen, backlog);
972 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
979 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
980 struct qdisc_skb_head *qh)
982 struct sk_buff *last = qh->tail;
995 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
997 __qdisc_enqueue_tail(skb, &sch->q);
998 qdisc_qstats_backlog_inc(sch, skb);
999 return NET_XMIT_SUCCESS;
1002 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1003 struct qdisc_skb_head *qh)
1005 skb->next = qh->head;
1013 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1015 struct sk_buff *skb = qh->head;
1017 if (likely(skb != NULL)) {
1018 qh->head = skb->next;
1020 if (qh->head == NULL)
1028 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1030 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1032 if (likely(skb != NULL)) {
1033 qdisc_qstats_backlog_dec(sch, skb);
1034 qdisc_bstats_update(sch, skb);
1040 /* Instead of calling kfree_skb() while root qdisc lock is held,
1041 * queue the skb for future freeing at end of __dev_xmit_skb()
1043 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1045 skb->next = *to_free;
1049 static inline void __qdisc_drop_all(struct sk_buff *skb,
1050 struct sk_buff **to_free)
1053 skb->prev->next = *to_free;
1055 skb->next = *to_free;
1059 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1060 struct qdisc_skb_head *qh,
1061 struct sk_buff **to_free)
1063 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1065 if (likely(skb != NULL)) {
1066 unsigned int len = qdisc_pkt_len(skb);
1068 qdisc_qstats_backlog_dec(sch, skb);
1069 __qdisc_drop(skb, to_free);
1076 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
1077 struct sk_buff **to_free)
1079 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
1082 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1084 const struct qdisc_skb_head *qh = &sch->q;
1089 /* generic pseudo peek method for non-work-conserving qdisc */
1090 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1092 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1094 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1096 skb = sch->dequeue(sch);
1099 __skb_queue_head(&sch->gso_skb, skb);
1100 /* it's still part of the queue */
1101 qdisc_qstats_backlog_inc(sch, skb);
1109 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1110 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1112 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1115 skb = __skb_dequeue(&sch->gso_skb);
1116 qdisc_qstats_backlog_dec(sch, skb);
1119 skb = sch->dequeue(sch);
1125 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1128 * We do not know the backlog in bytes of this list, it
1129 * is up to the caller to correct it
1133 rtnl_kfree_skbs(qh->head, qh->tail);
1141 static inline void qdisc_reset_queue(struct Qdisc *sch)
1143 __qdisc_reset_queue(&sch->q);
1144 sch->qstats.backlog = 0;
1147 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1148 struct Qdisc **pold)
1156 qdisc_tree_flush_backlog(old);
1157 sch_tree_unlock(sch);
1162 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1164 rtnl_kfree_skbs(skb, skb);
1165 qdisc_qstats_drop(sch);
1168 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1169 struct sk_buff **to_free)
1171 __qdisc_drop(skb, to_free);
1172 qdisc_qstats_cpu_drop(sch);
1174 return NET_XMIT_DROP;
1177 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1178 struct sk_buff **to_free)
1180 __qdisc_drop(skb, to_free);
1181 qdisc_qstats_drop(sch);
1183 return NET_XMIT_DROP;
1186 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1187 struct sk_buff **to_free)
1189 __qdisc_drop_all(skb, to_free);
1190 qdisc_qstats_drop(sch);
1192 return NET_XMIT_DROP;
1195 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1196 long it will take to send a packet given its size.
1198 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1200 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1203 slot >>= rtab->rate.cell_log;
1205 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1206 return rtab->data[slot];
1209 struct psched_ratecfg {
1210 u64 rate_bytes_ps; /* bytes per second */
1217 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1222 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1223 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1225 return ((u64)len * r->mult) >> r->shift;
1228 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1229 const struct tc_ratespec *conf,
1232 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1233 const struct psched_ratecfg *r)
1235 memset(res, 0, sizeof(*res));
1237 /* legacy struct tc_ratespec has a 32bit @rate field
1238 * Qdisc using 64bit rate should add new attributes
1239 * in order to maintain compatibility.
1241 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1243 res->overhead = r->overhead;
1244 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1247 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1248 * The fast path only needs to access filter list and to update stats
1251 struct tcf_proto *filter_list;
1252 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1253 struct gnet_stats_queue __percpu *cpu_qstats;
1254 struct rcu_head rcu;
1257 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1258 const struct sk_buff *skb)
1260 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1263 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1265 this_cpu_inc(miniq->cpu_qstats->drops);
1268 struct mini_Qdisc_pair {
1269 struct mini_Qdisc miniq1;
1270 struct mini_Qdisc miniq2;
1271 struct mini_Qdisc __rcu **p_miniq;
1274 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1275 struct tcf_proto *tp_head);
1276 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1277 struct mini_Qdisc __rcu **p_miniq);
1279 static inline void skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res)
1281 struct gnet_stats_queue *stats = res->qstats;
1285 ret = netif_receive_skb(skb);
1287 ret = dev_queue_xmit(skb);
1289 qstats_overlimit_inc(res->qstats);