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,
41 struct qdisc_size_table {
43 struct list_head list;
44 struct tc_sizespec szopts;
49 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
50 struct qdisc_skb_head {
58 int (*enqueue)(struct sk_buff *skb,
60 struct sk_buff **to_free);
61 struct sk_buff * (*dequeue)(struct Qdisc *sch);
63 #define TCQ_F_BUILTIN 1
64 #define TCQ_F_INGRESS 2
65 #define TCQ_F_CAN_BYPASS 4
66 #define TCQ_F_MQROOT 8
67 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
68 * q->dev_queue : It can test
69 * netif_xmit_frozen_or_stopped() before
70 * dequeueing next packet.
71 * Its true for MQ/MQPRIO slaves, or non
74 #define TCQ_F_WARN_NONWC (1 << 16)
75 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
76 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
77 * qdisc_tree_decrease_qlen() should stop.
79 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
80 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
81 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
83 const struct Qdisc_ops *ops;
84 struct qdisc_size_table __rcu *stab;
85 struct hlist_node hash;
89 struct netdev_queue *dev_queue;
91 struct net_rate_estimator __rcu *rate_est;
92 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
93 struct gnet_stats_queue __percpu *cpu_qstats;
98 * For performance sake on SMP, we put highly modified fields at the end
100 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
101 struct qdisc_skb_head q;
102 struct gnet_stats_basic_packed bstats;
104 struct gnet_stats_queue qstats;
106 struct Qdisc *next_sched;
107 struct sk_buff_head skb_bad_txq;
109 spinlock_t busylock ____cacheline_aligned_in_smp;
112 /* for NOLOCK qdisc, true if there are no enqueued skbs */
117 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
119 if (qdisc->flags & TCQ_F_BUILTIN)
121 refcount_inc(&qdisc->refcnt);
124 /* Intended to be used by unlocked users, when concurrent qdisc release is
128 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
130 if (qdisc->flags & TCQ_F_BUILTIN)
132 if (refcount_inc_not_zero(&qdisc->refcnt))
137 static inline bool qdisc_is_running(struct Qdisc *qdisc)
139 if (qdisc->flags & TCQ_F_NOLOCK)
140 return spin_is_locked(&qdisc->seqlock);
141 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
144 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
146 return q->flags & TCQ_F_CPUSTATS;
149 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
151 if (qdisc_is_percpu_stats(qdisc))
152 return READ_ONCE(qdisc->empty);
153 return !READ_ONCE(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 WRITE_ONCE(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 in the skb_tc_reinsert function */
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 bool (*delete_empty)(struct tcf_proto *tp);
312 void (*walk)(struct tcf_proto *tp,
313 struct tcf_walker *arg, bool rtnl_held);
314 int (*reoffload)(struct tcf_proto *tp, bool add,
315 flow_setup_cb_t *cb, void *cb_priv,
316 struct netlink_ext_ack *extack);
317 void (*hw_add)(struct tcf_proto *tp,
319 void (*hw_del)(struct tcf_proto *tp,
321 void (*bind_class)(void *, u32, unsigned long);
322 void * (*tmplt_create)(struct net *net,
323 struct tcf_chain *chain,
325 struct netlink_ext_ack *extack);
326 void (*tmplt_destroy)(void *tmplt_priv);
328 /* rtnetlink specific */
329 int (*dump)(struct net*, struct tcf_proto*, void *,
330 struct sk_buff *skb, struct tcmsg*,
332 int (*tmplt_dump)(struct sk_buff *skb,
336 struct module *owner;
340 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
341 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
342 * conditions can occur when filters are inserted/deleted simultaneously.
344 enum tcf_proto_ops_flags {
345 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
349 /* Fast access part */
350 struct tcf_proto __rcu *next;
353 /* called under RCU BH lock*/
354 int (*classify)(struct sk_buff *,
355 const struct tcf_proto *,
356 struct tcf_result *);
362 const struct tcf_proto_ops *ops;
363 struct tcf_chain *chain;
364 /* Lock protects tcf_proto shared state and can be used by unlocked
365 * classifiers to protect their private data.
371 struct hlist_node destroy_ht_node;
374 struct qdisc_skb_cb {
376 unsigned int pkt_len;
377 u16 slave_dev_queue_mapping;
380 #define QDISC_CB_PRIV_LEN 20
381 unsigned char data[QDISC_CB_PRIV_LEN];
384 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
387 /* Protects filter_chain. */
388 struct mutex filter_chain_lock;
389 struct tcf_proto __rcu *filter_chain;
390 struct list_head list;
391 struct tcf_block *block;
392 u32 index; /* chain index */
394 unsigned int action_refcnt;
395 bool explicitly_created;
397 const struct tcf_proto_ops *tmplt_ops;
403 /* Lock protects tcf_block and lifetime-management data of chains
404 * attached to the block (refcnt, action_refcnt, explicitly_created).
407 struct list_head chain_list;
408 u32 index; /* block index for shared blocks */
412 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
413 struct flow_block flow_block;
414 struct list_head owner_list;
416 atomic_t offloadcnt; /* Number of oddloaded filters */
417 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
418 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
420 struct tcf_chain *chain;
421 struct list_head filter_chain_list;
424 DECLARE_HASHTABLE(proto_destroy_ht, 7);
425 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
428 #ifdef CONFIG_PROVE_LOCKING
429 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
431 return lockdep_is_held(&chain->filter_chain_lock);
434 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
436 return lockdep_is_held(&tp->lock);
439 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain)
444 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
448 #endif /* #ifdef CONFIG_PROVE_LOCKING */
450 #define tcf_chain_dereference(p, chain) \
451 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
453 #define tcf_proto_dereference(p, tp) \
454 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
456 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
458 struct qdisc_skb_cb *qcb;
460 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
461 BUILD_BUG_ON(sizeof(qcb->data) < sz);
464 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
466 return this_cpu_ptr(q->cpu_qstats)->qlen;
469 static inline int qdisc_qlen(const struct Qdisc *q)
474 static inline int qdisc_qlen_sum(const struct Qdisc *q)
476 __u32 qlen = q->qstats.qlen;
479 if (qdisc_is_percpu_stats(q)) {
480 for_each_possible_cpu(i)
481 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
489 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
491 return (struct qdisc_skb_cb *)skb->cb;
494 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
496 return &qdisc->q.lock;
499 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
501 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
506 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
508 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
511 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
513 return qdisc->dev_queue->qdisc_sleeping;
516 /* The qdisc root lock is a mechanism by which to top level
517 * of a qdisc tree can be locked from any qdisc node in the
518 * forest. This allows changing the configuration of some
519 * aspect of the qdisc tree while blocking out asynchronous
520 * qdisc access in the packet processing paths.
522 * It is only legal to do this when the root will not change
523 * on us. Otherwise we'll potentially lock the wrong qdisc
524 * root. This is enforced by holding the RTNL semaphore, which
525 * all users of this lock accessor must do.
527 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
529 struct Qdisc *root = qdisc_root(qdisc);
532 return qdisc_lock(root);
535 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
537 struct Qdisc *root = qdisc_root_sleeping(qdisc);
540 return qdisc_lock(root);
543 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
545 struct Qdisc *root = qdisc_root_sleeping(qdisc);
548 return &root->running;
551 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
553 return qdisc->dev_queue->dev;
556 static inline void sch_tree_lock(const struct Qdisc *q)
558 spin_lock_bh(qdisc_root_sleeping_lock(q));
561 static inline void sch_tree_unlock(const struct Qdisc *q)
563 spin_unlock_bh(qdisc_root_sleeping_lock(q));
566 extern struct Qdisc noop_qdisc;
567 extern struct Qdisc_ops noop_qdisc_ops;
568 extern struct Qdisc_ops pfifo_fast_ops;
569 extern struct Qdisc_ops mq_qdisc_ops;
570 extern struct Qdisc_ops noqueue_qdisc_ops;
571 extern const struct Qdisc_ops *default_qdisc_ops;
572 static inline const struct Qdisc_ops *
573 get_default_qdisc_ops(const struct net_device *dev, int ntx)
575 return ntx < dev->real_num_tx_queues ?
576 default_qdisc_ops : &pfifo_fast_ops;
579 struct Qdisc_class_common {
581 struct hlist_node hnode;
584 struct Qdisc_class_hash {
585 struct hlist_head *hash;
586 unsigned int hashsize;
587 unsigned int hashmask;
588 unsigned int hashelems;
591 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
598 static inline struct Qdisc_class_common *
599 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
601 struct Qdisc_class_common *cl;
607 h = qdisc_class_hash(id, hash->hashmask);
608 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
609 if (cl->classid == id)
615 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
617 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
619 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
622 int qdisc_class_hash_init(struct Qdisc_class_hash *);
623 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
624 struct Qdisc_class_common *);
625 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
626 struct Qdisc_class_common *);
627 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
628 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
630 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
631 void dev_init_scheduler(struct net_device *dev);
632 void dev_shutdown(struct net_device *dev);
633 void dev_activate(struct net_device *dev);
634 void dev_deactivate(struct net_device *dev);
635 void dev_deactivate_many(struct list_head *head);
636 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
637 struct Qdisc *qdisc);
638 void qdisc_reset(struct Qdisc *qdisc);
639 void qdisc_put(struct Qdisc *qdisc);
640 void qdisc_put_unlocked(struct Qdisc *qdisc);
641 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
642 #ifdef CONFIG_NET_SCHED
643 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
645 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
646 struct Qdisc *new, struct Qdisc *old,
647 enum tc_setup_type type, void *type_data,
648 struct netlink_ext_ack *extack);
651 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
654 q->flags &= ~TCQ_F_OFFLOADED;
659 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
660 struct Qdisc *new, struct Qdisc *old,
661 enum tc_setup_type type, void *type_data,
662 struct netlink_ext_ack *extack)
666 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
667 const struct Qdisc_ops *ops,
668 struct netlink_ext_ack *extack);
669 void qdisc_free(struct Qdisc *qdisc);
670 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
671 const struct Qdisc_ops *ops, u32 parentid,
672 struct netlink_ext_ack *extack);
673 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
674 const struct qdisc_size_table *stab);
675 int skb_do_redirect(struct sk_buff *);
677 static inline void skb_reset_tc(struct sk_buff *skb)
679 #ifdef CONFIG_NET_CLS_ACT
680 skb->tc_redirected = 0;
684 static inline bool skb_is_tc_redirected(const struct sk_buff *skb)
686 #ifdef CONFIG_NET_CLS_ACT
687 return skb->tc_redirected;
693 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
695 #ifdef CONFIG_NET_CLS_ACT
696 return skb->tc_at_ingress;
702 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
704 #ifdef CONFIG_NET_CLS_ACT
705 if (skb->tc_skip_classify) {
706 skb->tc_skip_classify = 0;
713 /* Reset all TX qdiscs greater than index of a device. */
714 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
718 for (; i < dev->num_tx_queues; i++) {
719 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
721 spin_lock_bh(qdisc_lock(qdisc));
723 spin_unlock_bh(qdisc_lock(qdisc));
728 static inline void qdisc_reset_all_tx(struct net_device *dev)
730 qdisc_reset_all_tx_gt(dev, 0);
733 /* Are all TX queues of the device empty? */
734 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
739 for (i = 0; i < dev->num_tx_queues; i++) {
740 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
741 const struct Qdisc *q = rcu_dereference(txq->qdisc);
743 if (!qdisc_is_empty(q)) {
752 /* Are any of the TX qdiscs changing? */
753 static inline bool qdisc_tx_changing(const struct net_device *dev)
757 for (i = 0; i < dev->num_tx_queues; i++) {
758 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
759 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
765 /* Is the device using the noop qdisc on all queues? */
766 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
770 for (i = 0; i < dev->num_tx_queues; i++) {
771 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
772 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
778 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
780 return qdisc_skb_cb(skb)->pkt_len;
783 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
784 enum net_xmit_qdisc_t {
785 __NET_XMIT_STOLEN = 0x00010000,
786 __NET_XMIT_BYPASS = 0x00020000,
789 #ifdef CONFIG_NET_CLS_ACT
790 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
792 #define net_xmit_drop_count(e) (1)
795 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
796 const struct Qdisc *sch)
798 #ifdef CONFIG_NET_SCHED
799 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
802 __qdisc_calculate_pkt_len(skb, stab);
806 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
807 struct sk_buff **to_free)
809 qdisc_calculate_pkt_len(skb, sch);
810 return sch->enqueue(skb, sch, to_free);
813 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
814 __u64 bytes, __u32 packets)
816 bstats->bytes += bytes;
817 bstats->packets += packets;
820 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
821 const struct sk_buff *skb)
823 _bstats_update(bstats,
825 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
828 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
829 __u64 bytes, __u32 packets)
831 u64_stats_update_begin(&bstats->syncp);
832 _bstats_update(&bstats->bstats, bytes, packets);
833 u64_stats_update_end(&bstats->syncp);
836 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
837 const struct sk_buff *skb)
839 u64_stats_update_begin(&bstats->syncp);
840 bstats_update(&bstats->bstats, skb);
841 u64_stats_update_end(&bstats->syncp);
844 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
845 const struct sk_buff *skb)
847 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
850 static inline void qdisc_bstats_update(struct Qdisc *sch,
851 const struct sk_buff *skb)
853 bstats_update(&sch->bstats, skb);
856 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
857 const struct sk_buff *skb)
859 sch->qstats.backlog -= qdisc_pkt_len(skb);
862 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
863 const struct sk_buff *skb)
865 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
868 static inline void qdisc_qstats_backlog_inc(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_inc(struct Qdisc *sch,
875 const struct sk_buff *skb)
877 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
880 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
882 this_cpu_inc(sch->cpu_qstats->qlen);
885 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
887 this_cpu_dec(sch->cpu_qstats->qlen);
890 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
892 this_cpu_inc(sch->cpu_qstats->requeues);
895 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
897 sch->qstats.drops += count;
900 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
905 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
907 qstats->overlimits++;
910 static inline void qdisc_qstats_drop(struct Qdisc *sch)
912 qstats_drop_inc(&sch->qstats);
915 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
917 this_cpu_inc(sch->cpu_qstats->drops);
920 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
922 sch->qstats.overlimits++;
925 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
927 __u32 qlen = qdisc_qlen_sum(sch);
929 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
932 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
935 struct gnet_stats_queue qstats = { 0 };
936 __u32 len = qdisc_qlen_sum(sch);
938 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
940 *backlog = qstats.backlog;
943 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
947 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
948 qdisc_tree_reduce_backlog(sch, qlen, backlog);
951 static inline void qdisc_purge_queue(struct Qdisc *sch)
955 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
957 qdisc_tree_reduce_backlog(sch, qlen, backlog);
960 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
967 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
968 struct qdisc_skb_head *qh)
970 struct sk_buff *last = qh->tail;
983 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
985 __qdisc_enqueue_tail(skb, &sch->q);
986 qdisc_qstats_backlog_inc(sch, skb);
987 return NET_XMIT_SUCCESS;
990 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
991 struct qdisc_skb_head *qh)
993 skb->next = qh->head;
1001 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1003 struct sk_buff *skb = qh->head;
1005 if (likely(skb != NULL)) {
1006 qh->head = skb->next;
1008 if (qh->head == NULL)
1016 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1018 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1020 if (likely(skb != NULL)) {
1021 qdisc_qstats_backlog_dec(sch, skb);
1022 qdisc_bstats_update(sch, skb);
1028 /* Instead of calling kfree_skb() while root qdisc lock is held,
1029 * queue the skb for future freeing at end of __dev_xmit_skb()
1031 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1033 skb->next = *to_free;
1037 static inline void __qdisc_drop_all(struct sk_buff *skb,
1038 struct sk_buff **to_free)
1041 skb->prev->next = *to_free;
1043 skb->next = *to_free;
1047 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1048 struct qdisc_skb_head *qh,
1049 struct sk_buff **to_free)
1051 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1053 if (likely(skb != NULL)) {
1054 unsigned int len = qdisc_pkt_len(skb);
1056 qdisc_qstats_backlog_dec(sch, skb);
1057 __qdisc_drop(skb, to_free);
1064 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
1065 struct sk_buff **to_free)
1067 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
1070 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1072 const struct qdisc_skb_head *qh = &sch->q;
1077 /* generic pseudo peek method for non-work-conserving qdisc */
1078 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1080 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1082 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1084 skb = sch->dequeue(sch);
1087 __skb_queue_head(&sch->gso_skb, skb);
1088 /* it's still part of the queue */
1089 qdisc_qstats_backlog_inc(sch, skb);
1097 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1098 struct sk_buff *skb)
1100 if (qdisc_is_percpu_stats(sch)) {
1101 qdisc_qstats_cpu_backlog_dec(sch, skb);
1102 qdisc_bstats_cpu_update(sch, skb);
1103 qdisc_qstats_cpu_qlen_dec(sch);
1105 qdisc_qstats_backlog_dec(sch, skb);
1106 qdisc_bstats_update(sch, skb);
1111 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1112 unsigned int pkt_len)
1114 if (qdisc_is_percpu_stats(sch)) {
1115 qdisc_qstats_cpu_qlen_inc(sch);
1116 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1118 sch->qstats.backlog += pkt_len;
1123 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1124 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1126 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1129 skb = __skb_dequeue(&sch->gso_skb);
1130 if (qdisc_is_percpu_stats(sch)) {
1131 qdisc_qstats_cpu_backlog_dec(sch, skb);
1132 qdisc_qstats_cpu_qlen_dec(sch);
1134 qdisc_qstats_backlog_dec(sch, skb);
1138 skb = sch->dequeue(sch);
1144 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1147 * We do not know the backlog in bytes of this list, it
1148 * is up to the caller to correct it
1152 rtnl_kfree_skbs(qh->head, qh->tail);
1160 static inline void qdisc_reset_queue(struct Qdisc *sch)
1162 __qdisc_reset_queue(&sch->q);
1163 sch->qstats.backlog = 0;
1166 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1167 struct Qdisc **pold)
1175 qdisc_tree_flush_backlog(old);
1176 sch_tree_unlock(sch);
1181 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1183 rtnl_kfree_skbs(skb, skb);
1184 qdisc_qstats_drop(sch);
1187 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1188 struct sk_buff **to_free)
1190 __qdisc_drop(skb, to_free);
1191 qdisc_qstats_cpu_drop(sch);
1193 return NET_XMIT_DROP;
1196 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1197 struct sk_buff **to_free)
1199 __qdisc_drop(skb, to_free);
1200 qdisc_qstats_drop(sch);
1202 return NET_XMIT_DROP;
1205 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1206 struct sk_buff **to_free)
1208 __qdisc_drop_all(skb, to_free);
1209 qdisc_qstats_drop(sch);
1211 return NET_XMIT_DROP;
1214 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1215 long it will take to send a packet given its size.
1217 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1219 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1222 slot >>= rtab->rate.cell_log;
1224 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1225 return rtab->data[slot];
1228 struct psched_ratecfg {
1229 u64 rate_bytes_ps; /* bytes per second */
1236 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1241 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1242 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1244 return ((u64)len * r->mult) >> r->shift;
1247 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1248 const struct tc_ratespec *conf,
1251 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1252 const struct psched_ratecfg *r)
1254 memset(res, 0, sizeof(*res));
1256 /* legacy struct tc_ratespec has a 32bit @rate field
1257 * Qdisc using 64bit rate should add new attributes
1258 * in order to maintain compatibility.
1260 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1262 res->overhead = r->overhead;
1263 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1266 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1267 * The fast path only needs to access filter list and to update stats
1270 struct tcf_proto *filter_list;
1271 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1272 struct gnet_stats_queue __percpu *cpu_qstats;
1273 struct rcu_head rcu;
1276 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1277 const struct sk_buff *skb)
1279 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1282 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1284 this_cpu_inc(miniq->cpu_qstats->drops);
1287 struct mini_Qdisc_pair {
1288 struct mini_Qdisc miniq1;
1289 struct mini_Qdisc miniq2;
1290 struct mini_Qdisc __rcu **p_miniq;
1293 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1294 struct tcf_proto *tp_head);
1295 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1296 struct mini_Qdisc __rcu **p_miniq);
1298 static inline int skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res)
1300 return res->ingress ? netif_receive_skb(skb) : dev_queue_xmit(skb);