Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef __NET_SCHED_GENERIC_H
3#define __NET_SCHED_GENERIC_H
4
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 <net/gen_stats.h>
16#include <net/rtnetlink.h>
17
18struct Qdisc_ops;
19struct qdisc_walker;
20struct tcf_walker;
21struct module;
22
23struct qdisc_rate_table {
24 struct tc_ratespec rate;
25 u32 data[256];
26 struct qdisc_rate_table *next;
27 int refcnt;
28};
29
30enum qdisc_state_t {
31 __QDISC_STATE_SCHED,
32 __QDISC_STATE_DEACTIVATED,
33 __QDISC_STATE_RUNNING,
34};
35
36struct qdisc_size_table {
37 struct rcu_head rcu;
38 struct list_head list;
39 struct tc_sizespec szopts;
40 int refcnt;
41 u16 data[];
42};
43
44/* similar to sk_buff_head, but skb->prev pointer is undefined. */
45struct qdisc_skb_head {
46 struct sk_buff *head;
47 struct sk_buff *tail;
48 __u32 qlen;
49 spinlock_t lock;
50};
51
52struct Qdisc {
53 int (*enqueue)(struct sk_buff *skb,
54 struct Qdisc *sch,
55 struct sk_buff **to_free);
56 struct sk_buff * (*dequeue)(struct Qdisc *sch);
57 unsigned int flags;
58#define TCQ_F_BUILTIN 1
59#define TCQ_F_INGRESS 2
60#define TCQ_F_CAN_BYPASS 4
61#define TCQ_F_MQROOT 8
62#define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
63 * q->dev_queue : It can test
64 * netif_xmit_frozen_or_stopped() before
65 * dequeueing next packet.
66 * Its true for MQ/MQPRIO slaves, or non
67 * multiqueue device.
68 */
69#define TCQ_F_WARN_NONWC (1 << 16)
70#define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
71#define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
72 * qdisc_tree_decrease_qlen() should stop.
73 */
74#define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
75#define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
76#define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
77 u32 limit;
78 const struct Qdisc_ops *ops;
79 struct qdisc_size_table __rcu *stab;
80 struct hlist_node hash;
81 u32 handle;
82 u32 parent;
83
84 struct netdev_queue *dev_queue;
85
86 struct net_rate_estimator __rcu *rate_est;
87 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
88 struct gnet_stats_queue __percpu *cpu_qstats;
89
90 /*
91 * For performance sake on SMP, we put highly modified fields at the end
92 */
93 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
94 struct qdisc_skb_head q;
95 struct gnet_stats_basic_packed bstats;
96 seqcount_t running;
97 struct gnet_stats_queue qstats;
98 unsigned long state;
99 struct Qdisc *next_sched;
100 struct sk_buff_head skb_bad_txq;
101 int padded;
102 refcount_t refcnt;
103
104 spinlock_t busylock ____cacheline_aligned_in_smp;
105};
106
107static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
108{
109 if (qdisc->flags & TCQ_F_BUILTIN)
110 return;
111 refcount_inc(&qdisc->refcnt);
112}
113
114static inline bool qdisc_is_running(const struct Qdisc *qdisc)
115{
116 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
117}
118
119static inline bool qdisc_run_begin(struct Qdisc *qdisc)
120{
121 if (qdisc_is_running(qdisc))
122 return false;
123 /* Variant of write_seqcount_begin() telling lockdep a trylock
124 * was attempted.
125 */
126 raw_write_seqcount_begin(&qdisc->running);
127 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
128 return true;
129}
130
131static inline void qdisc_run_end(struct Qdisc *qdisc)
132{
133 write_seqcount_end(&qdisc->running);
134}
135
136static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
137{
138 return qdisc->flags & TCQ_F_ONETXQUEUE;
139}
140
141static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
142{
143#ifdef CONFIG_BQL
144 /* Non-BQL migrated drivers will return 0, too. */
145 return dql_avail(&txq->dql);
146#else
147 return 0;
148#endif
149}
150
151struct Qdisc_class_ops {
152 /* Child qdisc manipulation */
153 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
154 int (*graft)(struct Qdisc *, unsigned long cl,
155 struct Qdisc *, struct Qdisc **,
156 struct netlink_ext_ack *extack);
157 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
158 void (*qlen_notify)(struct Qdisc *, unsigned long);
159
160 /* Class manipulation routines */
161 unsigned long (*find)(struct Qdisc *, u32 classid);
162 int (*change)(struct Qdisc *, u32, u32,
163 struct nlattr **, unsigned long *,
164 struct netlink_ext_ack *);
165 int (*delete)(struct Qdisc *, unsigned long);
166 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
167
168 /* Filter manipulation */
169 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
170 unsigned long arg,
171 struct netlink_ext_ack *extack);
172 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
173 u32 classid);
174 void (*unbind_tcf)(struct Qdisc *, unsigned long);
175
176 /* rtnetlink specific */
177 int (*dump)(struct Qdisc *, unsigned long,
178 struct sk_buff *skb, struct tcmsg*);
179 int (*dump_stats)(struct Qdisc *, unsigned long,
180 struct gnet_dump *);
181};
182
183struct Qdisc_ops {
184 struct Qdisc_ops *next;
185 const struct Qdisc_class_ops *cl_ops;
186 char id[IFNAMSIZ];
187 int priv_size;
188 unsigned int static_flags;
189
190 int (*enqueue)(struct sk_buff *skb,
191 struct Qdisc *sch,
192 struct sk_buff **to_free);
193 struct sk_buff * (*dequeue)(struct Qdisc *);
194 struct sk_buff * (*peek)(struct Qdisc *);
195
196 int (*init)(struct Qdisc *sch, struct nlattr *arg,
197 struct netlink_ext_ack *extack);
198 void (*reset)(struct Qdisc *);
199 void (*destroy)(struct Qdisc *);
200 int (*change)(struct Qdisc *sch,
201 struct nlattr *arg,
202 struct netlink_ext_ack *extack);
203 void (*attach)(struct Qdisc *sch);
204 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
205
206 int (*dump)(struct Qdisc *, struct sk_buff *);
207 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
208
209 void (*ingress_block_set)(struct Qdisc *sch,
210 u32 block_index);
211 void (*egress_block_set)(struct Qdisc *sch,
212 u32 block_index);
213 u32 (*ingress_block_get)(struct Qdisc *sch);
214 u32 (*egress_block_get)(struct Qdisc *sch);
215
216 struct module *owner;
217};
218
219
220struct tcf_result {
221 union {
222 struct {
223 unsigned long class;
224 u32 classid;
225 };
226 const struct tcf_proto *goto_tp;
227 };
228};
229
230struct tcf_proto_ops {
231 struct list_head head;
232 char kind[IFNAMSIZ];
233
234 int (*classify)(struct sk_buff *,
235 const struct tcf_proto *,
236 struct tcf_result *);
237 int (*init)(struct tcf_proto*);
238 void (*destroy)(struct tcf_proto *tp,
239 struct netlink_ext_ack *extack);
240
241 void* (*get)(struct tcf_proto*, u32 handle);
242 int (*change)(struct net *net, struct sk_buff *,
243 struct tcf_proto*, unsigned long,
244 u32 handle, struct nlattr **,
245 void **, bool,
246 struct netlink_ext_ack *);
247 int (*delete)(struct tcf_proto *tp, void *arg,
248 bool *last,
249 struct netlink_ext_ack *);
250 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
251 void (*bind_class)(void *, u32, unsigned long);
252
253 /* rtnetlink specific */
254 int (*dump)(struct net*, struct tcf_proto*, void *,
255 struct sk_buff *skb, struct tcmsg*);
256
257 struct module *owner;
258};
259
260struct tcf_proto {
261 /* Fast access part */
262 struct tcf_proto __rcu *next;
263 void __rcu *root;
264 int (*classify)(struct sk_buff *,
265 const struct tcf_proto *,
266 struct tcf_result *);
267 __be16 protocol;
268
269 /* All the rest */
270 u32 prio;
271 void *data;
272 const struct tcf_proto_ops *ops;
273 struct tcf_chain *chain;
274 struct rcu_head rcu;
275};
276
277struct qdisc_skb_cb {
278 unsigned int pkt_len;
279 u16 slave_dev_queue_mapping;
280 u16 tc_classid;
281#define QDISC_CB_PRIV_LEN 20
282 unsigned char data[QDISC_CB_PRIV_LEN];
283};
284
285typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
286
287struct tcf_chain {
288 struct tcf_proto __rcu *filter_chain;
289 struct list_head filter_chain_list;
290 struct list_head list;
291 struct tcf_block *block;
292 u32 index; /* chain index */
293 unsigned int refcnt;
294};
295
296struct tcf_block {
297 struct list_head chain_list;
298 u32 index; /* block index for shared blocks */
299 unsigned int refcnt;
300 struct net *net;
301 struct Qdisc *q;
302 struct list_head cb_list;
303 struct list_head owner_list;
304 bool keep_dst;
305 unsigned int offloadcnt; /* Number of oddloaded filters */
306 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
307};
308
309static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
310{
311 if (*flags & TCA_CLS_FLAGS_IN_HW)
312 return;
313 *flags |= TCA_CLS_FLAGS_IN_HW;
314 block->offloadcnt++;
315}
316
317static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
318{
319 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
320 return;
321 *flags &= ~TCA_CLS_FLAGS_IN_HW;
322 block->offloadcnt--;
323}
324
325static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
326{
327 struct qdisc_skb_cb *qcb;
328
329 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
330 BUILD_BUG_ON(sizeof(qcb->data) < sz);
331}
332
333static inline int qdisc_qlen_cpu(const struct Qdisc *q)
334{
335 return this_cpu_ptr(q->cpu_qstats)->qlen;
336}
337
338static inline int qdisc_qlen(const struct Qdisc *q)
339{
340 return q->q.qlen;
341}
342
343static inline int qdisc_qlen_sum(const struct Qdisc *q)
344{
345 __u32 qlen = 0;
346 int i;
347
348 if (q->flags & TCQ_F_NOLOCK) {
349 for_each_possible_cpu(i)
350 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
351 } else {
352 qlen = q->q.qlen;
353 }
354
355 return qlen;
356}
357
358static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
359{
360 return (struct qdisc_skb_cb *)skb->cb;
361}
362
363static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
364{
365 return &qdisc->q.lock;
366}
367
368static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
369{
370 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
371
372 return q;
373}
374
375static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
376{
377 return qdisc->dev_queue->qdisc_sleeping;
378}
379
380/* The qdisc root lock is a mechanism by which to top level
381 * of a qdisc tree can be locked from any qdisc node in the
382 * forest. This allows changing the configuration of some
383 * aspect of the qdisc tree while blocking out asynchronous
384 * qdisc access in the packet processing paths.
385 *
386 * It is only legal to do this when the root will not change
387 * on us. Otherwise we'll potentially lock the wrong qdisc
388 * root. This is enforced by holding the RTNL semaphore, which
389 * all users of this lock accessor must do.
390 */
391static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
392{
393 struct Qdisc *root = qdisc_root(qdisc);
394
395 ASSERT_RTNL();
396 return qdisc_lock(root);
397}
398
399static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
400{
401 struct Qdisc *root = qdisc_root_sleeping(qdisc);
402
403 ASSERT_RTNL();
404 return qdisc_lock(root);
405}
406
407static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
408{
409 struct Qdisc *root = qdisc_root_sleeping(qdisc);
410
411 ASSERT_RTNL();
412 return &root->running;
413}
414
415static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
416{
417 return qdisc->dev_queue->dev;
418}
419
420static inline void sch_tree_lock(const struct Qdisc *q)
421{
422 spin_lock_bh(qdisc_root_sleeping_lock(q));
423}
424
425static inline void sch_tree_unlock(const struct Qdisc *q)
426{
427 spin_unlock_bh(qdisc_root_sleeping_lock(q));
428}
429
430extern struct Qdisc noop_qdisc;
431extern struct Qdisc_ops noop_qdisc_ops;
432extern struct Qdisc_ops pfifo_fast_ops;
433extern struct Qdisc_ops mq_qdisc_ops;
434extern struct Qdisc_ops noqueue_qdisc_ops;
435extern const struct Qdisc_ops *default_qdisc_ops;
436static inline const struct Qdisc_ops *
437get_default_qdisc_ops(const struct net_device *dev, int ntx)
438{
439 return ntx < dev->real_num_tx_queues ?
440 default_qdisc_ops : &pfifo_fast_ops;
441}
442
443struct Qdisc_class_common {
444 u32 classid;
445 struct hlist_node hnode;
446};
447
448struct Qdisc_class_hash {
449 struct hlist_head *hash;
450 unsigned int hashsize;
451 unsigned int hashmask;
452 unsigned int hashelems;
453};
454
455static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
456{
457 id ^= id >> 8;
458 id ^= id >> 4;
459 return id & mask;
460}
461
462static inline struct Qdisc_class_common *
463qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
464{
465 struct Qdisc_class_common *cl;
466 unsigned int h;
467
468 if (!id)
469 return NULL;
470
471 h = qdisc_class_hash(id, hash->hashmask);
472 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
473 if (cl->classid == id)
474 return cl;
475 }
476 return NULL;
477}
478
479static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
480{
481 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
482
483 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
484}
485
486int qdisc_class_hash_init(struct Qdisc_class_hash *);
487void qdisc_class_hash_insert(struct Qdisc_class_hash *,
488 struct Qdisc_class_common *);
489void qdisc_class_hash_remove(struct Qdisc_class_hash *,
490 struct Qdisc_class_common *);
491void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
492void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
493
494int dev_qdisc_change_tx_queue_len(struct net_device *dev);
495void dev_init_scheduler(struct net_device *dev);
496void dev_shutdown(struct net_device *dev);
497void dev_activate(struct net_device *dev);
498void dev_deactivate(struct net_device *dev);
499void dev_deactivate_many(struct list_head *head);
500struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
501 struct Qdisc *qdisc);
502void qdisc_reset(struct Qdisc *qdisc);
503void qdisc_destroy(struct Qdisc *qdisc);
504void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
505 unsigned int len);
506struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
507 const struct Qdisc_ops *ops,
508 struct netlink_ext_ack *extack);
509void qdisc_free(struct Qdisc *qdisc);
510struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
511 const struct Qdisc_ops *ops, u32 parentid,
512 struct netlink_ext_ack *extack);
513void __qdisc_calculate_pkt_len(struct sk_buff *skb,
514 const struct qdisc_size_table *stab);
515int skb_do_redirect(struct sk_buff *);
516
517static inline void skb_reset_tc(struct sk_buff *skb)
518{
519#ifdef CONFIG_NET_CLS_ACT
520 skb->tc_redirected = 0;
521#endif
522}
523
524static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
525{
526#ifdef CONFIG_NET_CLS_ACT
527 return skb->tc_at_ingress;
528#else
529 return false;
530#endif
531}
532
533static inline bool skb_skip_tc_classify(struct sk_buff *skb)
534{
535#ifdef CONFIG_NET_CLS_ACT
536 if (skb->tc_skip_classify) {
537 skb->tc_skip_classify = 0;
538 return true;
539 }
540#endif
541 return false;
542}
543
544/* Reset all TX qdiscs greater than index of a device. */
545static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
546{
547 struct Qdisc *qdisc;
548
549 for (; i < dev->num_tx_queues; i++) {
550 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
551 if (qdisc) {
552 spin_lock_bh(qdisc_lock(qdisc));
553 qdisc_reset(qdisc);
554 spin_unlock_bh(qdisc_lock(qdisc));
555 }
556 }
557}
558
559static inline void qdisc_reset_all_tx(struct net_device *dev)
560{
561 qdisc_reset_all_tx_gt(dev, 0);
562}
563
564/* Are all TX queues of the device empty? */
565static inline bool qdisc_all_tx_empty(const struct net_device *dev)
566{
567 unsigned int i;
568
569 rcu_read_lock();
570 for (i = 0; i < dev->num_tx_queues; i++) {
571 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
572 const struct Qdisc *q = rcu_dereference(txq->qdisc);
573
574 if (q->q.qlen) {
575 rcu_read_unlock();
576 return false;
577 }
578 }
579 rcu_read_unlock();
580 return true;
581}
582
583/* Are any of the TX qdiscs changing? */
584static inline bool qdisc_tx_changing(const struct net_device *dev)
585{
586 unsigned int i;
587
588 for (i = 0; i < dev->num_tx_queues; i++) {
589 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
590 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
591 return true;
592 }
593 return false;
594}
595
596/* Is the device using the noop qdisc on all queues? */
597static inline bool qdisc_tx_is_noop(const struct net_device *dev)
598{
599 unsigned int i;
600
601 for (i = 0; i < dev->num_tx_queues; i++) {
602 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
603 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
604 return false;
605 }
606 return true;
607}
608
609static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
610{
611 return qdisc_skb_cb(skb)->pkt_len;
612}
613
614/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
615enum net_xmit_qdisc_t {
616 __NET_XMIT_STOLEN = 0x00010000,
617 __NET_XMIT_BYPASS = 0x00020000,
618};
619
620#ifdef CONFIG_NET_CLS_ACT
621#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
622#else
623#define net_xmit_drop_count(e) (1)
624#endif
625
626static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
627 const struct Qdisc *sch)
628{
629#ifdef CONFIG_NET_SCHED
630 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
631
632 if (stab)
633 __qdisc_calculate_pkt_len(skb, stab);
634#endif
635}
636
637static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
638 struct sk_buff **to_free)
639{
640 qdisc_calculate_pkt_len(skb, sch);
641 return sch->enqueue(skb, sch, to_free);
642}
643
644static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
645{
646 return q->flags & TCQ_F_CPUSTATS;
647}
648
649static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
650 __u64 bytes, __u32 packets)
651{
652 bstats->bytes += bytes;
653 bstats->packets += packets;
654}
655
656static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
657 const struct sk_buff *skb)
658{
659 _bstats_update(bstats,
660 qdisc_pkt_len(skb),
661 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
662}
663
664static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
665 __u64 bytes, __u32 packets)
666{
667 u64_stats_update_begin(&bstats->syncp);
668 _bstats_update(&bstats->bstats, bytes, packets);
669 u64_stats_update_end(&bstats->syncp);
670}
671
672static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
673 const struct sk_buff *skb)
674{
675 u64_stats_update_begin(&bstats->syncp);
676 bstats_update(&bstats->bstats, skb);
677 u64_stats_update_end(&bstats->syncp);
678}
679
680static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
681 const struct sk_buff *skb)
682{
683 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
684}
685
686static inline void qdisc_bstats_update(struct Qdisc *sch,
687 const struct sk_buff *skb)
688{
689 bstats_update(&sch->bstats, skb);
690}
691
692static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
693 const struct sk_buff *skb)
694{
695 sch->qstats.backlog -= qdisc_pkt_len(skb);
696}
697
698static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
699 const struct sk_buff *skb)
700{
701 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
702}
703
704static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
705 const struct sk_buff *skb)
706{
707 sch->qstats.backlog += qdisc_pkt_len(skb);
708}
709
710static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
711 const struct sk_buff *skb)
712{
713 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
714}
715
716static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
717{
718 this_cpu_inc(sch->cpu_qstats->qlen);
719}
720
721static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
722{
723 this_cpu_dec(sch->cpu_qstats->qlen);
724}
725
726static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
727{
728 this_cpu_inc(sch->cpu_qstats->requeues);
729}
730
731static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
732{
733 sch->qstats.drops += count;
734}
735
736static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
737{
738 qstats->drops++;
739}
740
741static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
742{
743 qstats->overlimits++;
744}
745
746static inline void qdisc_qstats_drop(struct Qdisc *sch)
747{
748 qstats_drop_inc(&sch->qstats);
749}
750
751static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
752{
753 this_cpu_inc(sch->cpu_qstats->drops);
754}
755
756static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
757{
758 sch->qstats.overlimits++;
759}
760
761static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
762{
763 qh->head = NULL;
764 qh->tail = NULL;
765 qh->qlen = 0;
766}
767
768static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
769 struct qdisc_skb_head *qh)
770{
771 struct sk_buff *last = qh->tail;
772
773 if (last) {
774 skb->next = NULL;
775 last->next = skb;
776 qh->tail = skb;
777 } else {
778 qh->tail = skb;
779 qh->head = skb;
780 }
781 qh->qlen++;
782 qdisc_qstats_backlog_inc(sch, skb);
783
784 return NET_XMIT_SUCCESS;
785}
786
787static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
788{
789 return __qdisc_enqueue_tail(skb, sch, &sch->q);
790}
791
792static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
793{
794 struct sk_buff *skb = qh->head;
795
796 if (likely(skb != NULL)) {
797 qh->head = skb->next;
798 qh->qlen--;
799 if (qh->head == NULL)
800 qh->tail = NULL;
801 skb->next = NULL;
802 }
803
804 return skb;
805}
806
807static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
808{
809 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
810
811 if (likely(skb != NULL)) {
812 qdisc_qstats_backlog_dec(sch, skb);
813 qdisc_bstats_update(sch, skb);
814 }
815
816 return skb;
817}
818
819/* Instead of calling kfree_skb() while root qdisc lock is held,
820 * queue the skb for future freeing at end of __dev_xmit_skb()
821 */
822static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
823{
824 skb->next = *to_free;
825 *to_free = skb;
826}
827
828static inline void __qdisc_drop_all(struct sk_buff *skb,
829 struct sk_buff **to_free)
830{
831 if (skb->prev)
832 skb->prev->next = *to_free;
833 else
834 skb->next = *to_free;
835 *to_free = skb;
836}
837
838static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
839 struct qdisc_skb_head *qh,
840 struct sk_buff **to_free)
841{
842 struct sk_buff *skb = __qdisc_dequeue_head(qh);
843
844 if (likely(skb != NULL)) {
845 unsigned int len = qdisc_pkt_len(skb);
846
847 qdisc_qstats_backlog_dec(sch, skb);
848 __qdisc_drop(skb, to_free);
849 return len;
850 }
851
852 return 0;
853}
854
855static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
856 struct sk_buff **to_free)
857{
858 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
859}
860
861static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
862{
863 const struct qdisc_skb_head *qh = &sch->q;
864
865 return qh->head;
866}
867
868/* generic pseudo peek method for non-work-conserving qdisc */
869static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
870{
871 struct sk_buff *skb = skb_peek(&sch->gso_skb);
872
873 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
874 if (!skb) {
875 skb = sch->dequeue(sch);
876
877 if (skb) {
878 __skb_queue_head(&sch->gso_skb, skb);
879 /* it's still part of the queue */
880 qdisc_qstats_backlog_inc(sch, skb);
881 sch->q.qlen++;
882 }
883 }
884
885 return skb;
886}
887
888/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
889static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
890{
891 struct sk_buff *skb = skb_peek(&sch->gso_skb);
892
893 if (skb) {
894 skb = __skb_dequeue(&sch->gso_skb);
895 qdisc_qstats_backlog_dec(sch, skb);
896 sch->q.qlen--;
897 } else {
898 skb = sch->dequeue(sch);
899 }
900
901 return skb;
902}
903
904static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
905{
906 /*
907 * We do not know the backlog in bytes of this list, it
908 * is up to the caller to correct it
909 */
910 ASSERT_RTNL();
911 if (qh->qlen) {
912 rtnl_kfree_skbs(qh->head, qh->tail);
913
914 qh->head = NULL;
915 qh->tail = NULL;
916 qh->qlen = 0;
917 }
918}
919
920static inline void qdisc_reset_queue(struct Qdisc *sch)
921{
922 __qdisc_reset_queue(&sch->q);
923 sch->qstats.backlog = 0;
924}
925
926static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
927 struct Qdisc **pold)
928{
929 struct Qdisc *old;
930
931 sch_tree_lock(sch);
932 old = *pold;
933 *pold = new;
934 if (old != NULL) {
935 unsigned int qlen = old->q.qlen;
936 unsigned int backlog = old->qstats.backlog;
937
938 qdisc_reset(old);
939 qdisc_tree_reduce_backlog(old, qlen, backlog);
940 }
941 sch_tree_unlock(sch);
942
943 return old;
944}
945
946static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
947{
948 rtnl_kfree_skbs(skb, skb);
949 qdisc_qstats_drop(sch);
950}
951
952static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
953 struct sk_buff **to_free)
954{
955 __qdisc_drop(skb, to_free);
956 qdisc_qstats_cpu_drop(sch);
957
958 return NET_XMIT_DROP;
959}
960
961static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
962 struct sk_buff **to_free)
963{
964 __qdisc_drop(skb, to_free);
965 qdisc_qstats_drop(sch);
966
967 return NET_XMIT_DROP;
968}
969
970static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
971 struct sk_buff **to_free)
972{
973 __qdisc_drop_all(skb, to_free);
974 qdisc_qstats_drop(sch);
975
976 return NET_XMIT_DROP;
977}
978
979/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
980 long it will take to send a packet given its size.
981 */
982static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
983{
984 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
985 if (slot < 0)
986 slot = 0;
987 slot >>= rtab->rate.cell_log;
988 if (slot > 255)
989 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
990 return rtab->data[slot];
991}
992
993struct psched_ratecfg {
994 u64 rate_bytes_ps; /* bytes per second */
995 u32 mult;
996 u16 overhead;
997 u8 linklayer;
998 u8 shift;
999};
1000
1001static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1002 unsigned int len)
1003{
1004 len += r->overhead;
1005
1006 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1007 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1008
1009 return ((u64)len * r->mult) >> r->shift;
1010}
1011
1012void psched_ratecfg_precompute(struct psched_ratecfg *r,
1013 const struct tc_ratespec *conf,
1014 u64 rate64);
1015
1016static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1017 const struct psched_ratecfg *r)
1018{
1019 memset(res, 0, sizeof(*res));
1020
1021 /* legacy struct tc_ratespec has a 32bit @rate field
1022 * Qdisc using 64bit rate should add new attributes
1023 * in order to maintain compatibility.
1024 */
1025 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1026
1027 res->overhead = r->overhead;
1028 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1029}
1030
1031/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1032 * The fast path only needs to access filter list and to update stats
1033 */
1034struct mini_Qdisc {
1035 struct tcf_proto *filter_list;
1036 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1037 struct gnet_stats_queue __percpu *cpu_qstats;
1038 struct rcu_head rcu;
1039};
1040
1041static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1042 const struct sk_buff *skb)
1043{
1044 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1045}
1046
1047static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1048{
1049 this_cpu_inc(miniq->cpu_qstats->drops);
1050}
1051
1052struct mini_Qdisc_pair {
1053 struct mini_Qdisc miniq1;
1054 struct mini_Qdisc miniq2;
1055 struct mini_Qdisc __rcu **p_miniq;
1056};
1057
1058void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1059 struct tcf_proto *tp_head);
1060void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1061 struct mini_Qdisc __rcu **p_miniq);
1062
1063#endif
1#ifndef __NET_SCHED_GENERIC_H
2#define __NET_SCHED_GENERIC_H
3
4#include <linux/netdevice.h>
5#include <linux/types.h>
6#include <linux/rcupdate.h>
7#include <linux/pkt_sched.h>
8#include <linux/pkt_cls.h>
9#include <linux/percpu.h>
10#include <linux/dynamic_queue_limits.h>
11#include <net/gen_stats.h>
12#include <net/rtnetlink.h>
13
14struct Qdisc_ops;
15struct qdisc_walker;
16struct tcf_walker;
17struct module;
18
19struct qdisc_rate_table {
20 struct tc_ratespec rate;
21 u32 data[256];
22 struct qdisc_rate_table *next;
23 int refcnt;
24};
25
26enum qdisc_state_t {
27 __QDISC_STATE_SCHED,
28 __QDISC_STATE_DEACTIVATED,
29};
30
31struct qdisc_size_table {
32 struct rcu_head rcu;
33 struct list_head list;
34 struct tc_sizespec szopts;
35 int refcnt;
36 u16 data[];
37};
38
39/* similar to sk_buff_head, but skb->prev pointer is undefined. */
40struct qdisc_skb_head {
41 struct sk_buff *head;
42 struct sk_buff *tail;
43 __u32 qlen;
44 spinlock_t lock;
45};
46
47struct Qdisc {
48 int (*enqueue)(struct sk_buff *skb,
49 struct Qdisc *sch,
50 struct sk_buff **to_free);
51 struct sk_buff * (*dequeue)(struct Qdisc *sch);
52 unsigned int flags;
53#define TCQ_F_BUILTIN 1
54#define TCQ_F_INGRESS 2
55#define TCQ_F_CAN_BYPASS 4
56#define TCQ_F_MQROOT 8
57#define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
58 * q->dev_queue : It can test
59 * netif_xmit_frozen_or_stopped() before
60 * dequeueing next packet.
61 * Its true for MQ/MQPRIO slaves, or non
62 * multiqueue device.
63 */
64#define TCQ_F_WARN_NONWC (1 << 16)
65#define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
66#define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
67 * qdisc_tree_decrease_qlen() should stop.
68 */
69 u32 limit;
70 const struct Qdisc_ops *ops;
71 struct qdisc_size_table __rcu *stab;
72 struct hlist_node hash;
73 u32 handle;
74 u32 parent;
75 void *u32_node;
76
77 struct netdev_queue *dev_queue;
78
79 struct net_rate_estimator __rcu *rate_est;
80 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
81 struct gnet_stats_queue __percpu *cpu_qstats;
82
83 /*
84 * For performance sake on SMP, we put highly modified fields at the end
85 */
86 struct sk_buff *gso_skb ____cacheline_aligned_in_smp;
87 struct qdisc_skb_head q;
88 struct gnet_stats_basic_packed bstats;
89 seqcount_t running;
90 struct gnet_stats_queue qstats;
91 unsigned long state;
92 struct Qdisc *next_sched;
93 struct sk_buff *skb_bad_txq;
94 struct rcu_head rcu_head;
95 int padded;
96 atomic_t refcnt;
97
98 spinlock_t busylock ____cacheline_aligned_in_smp;
99};
100
101static inline bool qdisc_is_running(const struct Qdisc *qdisc)
102{
103 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
104}
105
106static inline bool qdisc_run_begin(struct Qdisc *qdisc)
107{
108 if (qdisc_is_running(qdisc))
109 return false;
110 /* Variant of write_seqcount_begin() telling lockdep a trylock
111 * was attempted.
112 */
113 raw_write_seqcount_begin(&qdisc->running);
114 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
115 return true;
116}
117
118static inline void qdisc_run_end(struct Qdisc *qdisc)
119{
120 write_seqcount_end(&qdisc->running);
121}
122
123static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
124{
125 return qdisc->flags & TCQ_F_ONETXQUEUE;
126}
127
128static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
129{
130#ifdef CONFIG_BQL
131 /* Non-BQL migrated drivers will return 0, too. */
132 return dql_avail(&txq->dql);
133#else
134 return 0;
135#endif
136}
137
138struct Qdisc_class_ops {
139 /* Child qdisc manipulation */
140 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
141 int (*graft)(struct Qdisc *, unsigned long cl,
142 struct Qdisc *, struct Qdisc **);
143 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
144 void (*qlen_notify)(struct Qdisc *, unsigned long);
145
146 /* Class manipulation routines */
147 unsigned long (*get)(struct Qdisc *, u32 classid);
148 void (*put)(struct Qdisc *, unsigned long);
149 int (*change)(struct Qdisc *, u32, u32,
150 struct nlattr **, unsigned long *);
151 int (*delete)(struct Qdisc *, unsigned long);
152 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
153
154 /* Filter manipulation */
155 struct tcf_proto __rcu ** (*tcf_chain)(struct Qdisc *, unsigned long);
156 bool (*tcf_cl_offload)(u32 classid);
157 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
158 u32 classid);
159 void (*unbind_tcf)(struct Qdisc *, unsigned long);
160
161 /* rtnetlink specific */
162 int (*dump)(struct Qdisc *, unsigned long,
163 struct sk_buff *skb, struct tcmsg*);
164 int (*dump_stats)(struct Qdisc *, unsigned long,
165 struct gnet_dump *);
166};
167
168struct Qdisc_ops {
169 struct Qdisc_ops *next;
170 const struct Qdisc_class_ops *cl_ops;
171 char id[IFNAMSIZ];
172 int priv_size;
173
174 int (*enqueue)(struct sk_buff *skb,
175 struct Qdisc *sch,
176 struct sk_buff **to_free);
177 struct sk_buff * (*dequeue)(struct Qdisc *);
178 struct sk_buff * (*peek)(struct Qdisc *);
179
180 int (*init)(struct Qdisc *, struct nlattr *arg);
181 void (*reset)(struct Qdisc *);
182 void (*destroy)(struct Qdisc *);
183 int (*change)(struct Qdisc *, struct nlattr *arg);
184 void (*attach)(struct Qdisc *);
185
186 int (*dump)(struct Qdisc *, struct sk_buff *);
187 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
188
189 struct module *owner;
190};
191
192
193struct tcf_result {
194 unsigned long class;
195 u32 classid;
196};
197
198struct tcf_proto_ops {
199 struct list_head head;
200 char kind[IFNAMSIZ];
201
202 int (*classify)(struct sk_buff *,
203 const struct tcf_proto *,
204 struct tcf_result *);
205 int (*init)(struct tcf_proto*);
206 bool (*destroy)(struct tcf_proto*, bool);
207
208 unsigned long (*get)(struct tcf_proto*, u32 handle);
209 int (*change)(struct net *net, struct sk_buff *,
210 struct tcf_proto*, unsigned long,
211 u32 handle, struct nlattr **,
212 unsigned long *, bool);
213 int (*delete)(struct tcf_proto*, unsigned long);
214 void (*walk)(struct tcf_proto*, struct tcf_walker *arg);
215
216 /* rtnetlink specific */
217 int (*dump)(struct net*, struct tcf_proto*, unsigned long,
218 struct sk_buff *skb, struct tcmsg*);
219
220 struct module *owner;
221};
222
223struct tcf_proto {
224 /* Fast access part */
225 struct tcf_proto __rcu *next;
226 void __rcu *root;
227 int (*classify)(struct sk_buff *,
228 const struct tcf_proto *,
229 struct tcf_result *);
230 __be16 protocol;
231
232 /* All the rest */
233 u32 prio;
234 u32 classid;
235 struct Qdisc *q;
236 void *data;
237 const struct tcf_proto_ops *ops;
238 struct rcu_head rcu;
239};
240
241struct qdisc_skb_cb {
242 unsigned int pkt_len;
243 u16 slave_dev_queue_mapping;
244 u16 tc_classid;
245#define QDISC_CB_PRIV_LEN 20
246 unsigned char data[QDISC_CB_PRIV_LEN];
247};
248
249static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
250{
251 struct qdisc_skb_cb *qcb;
252
253 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
254 BUILD_BUG_ON(sizeof(qcb->data) < sz);
255}
256
257static inline int qdisc_qlen(const struct Qdisc *q)
258{
259 return q->q.qlen;
260}
261
262static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
263{
264 return (struct qdisc_skb_cb *)skb->cb;
265}
266
267static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
268{
269 return &qdisc->q.lock;
270}
271
272static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
273{
274 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
275
276 return q;
277}
278
279static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
280{
281 return qdisc->dev_queue->qdisc_sleeping;
282}
283
284/* The qdisc root lock is a mechanism by which to top level
285 * of a qdisc tree can be locked from any qdisc node in the
286 * forest. This allows changing the configuration of some
287 * aspect of the qdisc tree while blocking out asynchronous
288 * qdisc access in the packet processing paths.
289 *
290 * It is only legal to do this when the root will not change
291 * on us. Otherwise we'll potentially lock the wrong qdisc
292 * root. This is enforced by holding the RTNL semaphore, which
293 * all users of this lock accessor must do.
294 */
295static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
296{
297 struct Qdisc *root = qdisc_root(qdisc);
298
299 ASSERT_RTNL();
300 return qdisc_lock(root);
301}
302
303static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
304{
305 struct Qdisc *root = qdisc_root_sleeping(qdisc);
306
307 ASSERT_RTNL();
308 return qdisc_lock(root);
309}
310
311static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
312{
313 struct Qdisc *root = qdisc_root_sleeping(qdisc);
314
315 ASSERT_RTNL();
316 return &root->running;
317}
318
319static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
320{
321 return qdisc->dev_queue->dev;
322}
323
324static inline void sch_tree_lock(const struct Qdisc *q)
325{
326 spin_lock_bh(qdisc_root_sleeping_lock(q));
327}
328
329static inline void sch_tree_unlock(const struct Qdisc *q)
330{
331 spin_unlock_bh(qdisc_root_sleeping_lock(q));
332}
333
334#define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
335#define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
336
337extern struct Qdisc noop_qdisc;
338extern struct Qdisc_ops noop_qdisc_ops;
339extern struct Qdisc_ops pfifo_fast_ops;
340extern struct Qdisc_ops mq_qdisc_ops;
341extern struct Qdisc_ops noqueue_qdisc_ops;
342extern const struct Qdisc_ops *default_qdisc_ops;
343static inline const struct Qdisc_ops *
344get_default_qdisc_ops(const struct net_device *dev, int ntx)
345{
346 return ntx < dev->real_num_tx_queues ?
347 default_qdisc_ops : &pfifo_fast_ops;
348}
349
350struct Qdisc_class_common {
351 u32 classid;
352 struct hlist_node hnode;
353};
354
355struct Qdisc_class_hash {
356 struct hlist_head *hash;
357 unsigned int hashsize;
358 unsigned int hashmask;
359 unsigned int hashelems;
360};
361
362static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
363{
364 id ^= id >> 8;
365 id ^= id >> 4;
366 return id & mask;
367}
368
369static inline struct Qdisc_class_common *
370qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
371{
372 struct Qdisc_class_common *cl;
373 unsigned int h;
374
375 h = qdisc_class_hash(id, hash->hashmask);
376 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
377 if (cl->classid == id)
378 return cl;
379 }
380 return NULL;
381}
382
383int qdisc_class_hash_init(struct Qdisc_class_hash *);
384void qdisc_class_hash_insert(struct Qdisc_class_hash *,
385 struct Qdisc_class_common *);
386void qdisc_class_hash_remove(struct Qdisc_class_hash *,
387 struct Qdisc_class_common *);
388void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
389void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
390
391void dev_init_scheduler(struct net_device *dev);
392void dev_shutdown(struct net_device *dev);
393void dev_activate(struct net_device *dev);
394void dev_deactivate(struct net_device *dev);
395void dev_deactivate_many(struct list_head *head);
396struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
397 struct Qdisc *qdisc);
398void qdisc_reset(struct Qdisc *qdisc);
399void qdisc_destroy(struct Qdisc *qdisc);
400void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
401 unsigned int len);
402struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
403 const struct Qdisc_ops *ops);
404struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
405 const struct Qdisc_ops *ops, u32 parentid);
406void __qdisc_calculate_pkt_len(struct sk_buff *skb,
407 const struct qdisc_size_table *stab);
408bool tcf_destroy(struct tcf_proto *tp, bool force);
409void tcf_destroy_chain(struct tcf_proto __rcu **fl);
410int skb_do_redirect(struct sk_buff *);
411
412static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
413{
414#ifdef CONFIG_NET_CLS_ACT
415 return G_TC_AT(skb->tc_verd) & AT_INGRESS;
416#else
417 return false;
418#endif
419}
420
421/* Reset all TX qdiscs greater then index of a device. */
422static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
423{
424 struct Qdisc *qdisc;
425
426 for (; i < dev->num_tx_queues; i++) {
427 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
428 if (qdisc) {
429 spin_lock_bh(qdisc_lock(qdisc));
430 qdisc_reset(qdisc);
431 spin_unlock_bh(qdisc_lock(qdisc));
432 }
433 }
434}
435
436static inline void qdisc_reset_all_tx(struct net_device *dev)
437{
438 qdisc_reset_all_tx_gt(dev, 0);
439}
440
441/* Are all TX queues of the device empty? */
442static inline bool qdisc_all_tx_empty(const struct net_device *dev)
443{
444 unsigned int i;
445
446 rcu_read_lock();
447 for (i = 0; i < dev->num_tx_queues; i++) {
448 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
449 const struct Qdisc *q = rcu_dereference(txq->qdisc);
450
451 if (q->q.qlen) {
452 rcu_read_unlock();
453 return false;
454 }
455 }
456 rcu_read_unlock();
457 return true;
458}
459
460/* Are any of the TX qdiscs changing? */
461static inline bool qdisc_tx_changing(const struct net_device *dev)
462{
463 unsigned int i;
464
465 for (i = 0; i < dev->num_tx_queues; i++) {
466 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
467 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
468 return true;
469 }
470 return false;
471}
472
473/* Is the device using the noop qdisc on all queues? */
474static inline bool qdisc_tx_is_noop(const struct net_device *dev)
475{
476 unsigned int i;
477
478 for (i = 0; i < dev->num_tx_queues; i++) {
479 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
480 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
481 return false;
482 }
483 return true;
484}
485
486static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
487{
488 return qdisc_skb_cb(skb)->pkt_len;
489}
490
491/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
492enum net_xmit_qdisc_t {
493 __NET_XMIT_STOLEN = 0x00010000,
494 __NET_XMIT_BYPASS = 0x00020000,
495};
496
497#ifdef CONFIG_NET_CLS_ACT
498#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
499#else
500#define net_xmit_drop_count(e) (1)
501#endif
502
503static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
504 const struct Qdisc *sch)
505{
506#ifdef CONFIG_NET_SCHED
507 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
508
509 if (stab)
510 __qdisc_calculate_pkt_len(skb, stab);
511#endif
512}
513
514static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
515 struct sk_buff **to_free)
516{
517 qdisc_calculate_pkt_len(skb, sch);
518 return sch->enqueue(skb, sch, to_free);
519}
520
521static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
522{
523 return q->flags & TCQ_F_CPUSTATS;
524}
525
526static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
527 __u64 bytes, __u32 packets)
528{
529 bstats->bytes += bytes;
530 bstats->packets += packets;
531}
532
533static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
534 const struct sk_buff *skb)
535{
536 _bstats_update(bstats,
537 qdisc_pkt_len(skb),
538 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
539}
540
541static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
542 __u64 bytes, __u32 packets)
543{
544 u64_stats_update_begin(&bstats->syncp);
545 _bstats_update(&bstats->bstats, bytes, packets);
546 u64_stats_update_end(&bstats->syncp);
547}
548
549static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
550 const struct sk_buff *skb)
551{
552 u64_stats_update_begin(&bstats->syncp);
553 bstats_update(&bstats->bstats, skb);
554 u64_stats_update_end(&bstats->syncp);
555}
556
557static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
558 const struct sk_buff *skb)
559{
560 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
561}
562
563static inline void qdisc_bstats_update(struct Qdisc *sch,
564 const struct sk_buff *skb)
565{
566 bstats_update(&sch->bstats, skb);
567}
568
569static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
570 const struct sk_buff *skb)
571{
572 sch->qstats.backlog -= qdisc_pkt_len(skb);
573}
574
575static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
576 const struct sk_buff *skb)
577{
578 sch->qstats.backlog += qdisc_pkt_len(skb);
579}
580
581static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
582{
583 sch->qstats.drops += count;
584}
585
586static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
587{
588 qstats->drops++;
589}
590
591static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
592{
593 qstats->overlimits++;
594}
595
596static inline void qdisc_qstats_drop(struct Qdisc *sch)
597{
598 qstats_drop_inc(&sch->qstats);
599}
600
601static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
602{
603 this_cpu_inc(sch->cpu_qstats->drops);
604}
605
606static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
607{
608 sch->qstats.overlimits++;
609}
610
611static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
612{
613 qh->head = NULL;
614 qh->tail = NULL;
615 qh->qlen = 0;
616}
617
618static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
619 struct qdisc_skb_head *qh)
620{
621 struct sk_buff *last = qh->tail;
622
623 if (last) {
624 skb->next = NULL;
625 last->next = skb;
626 qh->tail = skb;
627 } else {
628 qh->tail = skb;
629 qh->head = skb;
630 }
631 qh->qlen++;
632 qdisc_qstats_backlog_inc(sch, skb);
633
634 return NET_XMIT_SUCCESS;
635}
636
637static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
638{
639 return __qdisc_enqueue_tail(skb, sch, &sch->q);
640}
641
642static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
643{
644 struct sk_buff *skb = qh->head;
645
646 if (likely(skb != NULL)) {
647 qh->head = skb->next;
648 qh->qlen--;
649 if (qh->head == NULL)
650 qh->tail = NULL;
651 skb->next = NULL;
652 }
653
654 return skb;
655}
656
657static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
658{
659 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
660
661 if (likely(skb != NULL)) {
662 qdisc_qstats_backlog_dec(sch, skb);
663 qdisc_bstats_update(sch, skb);
664 }
665
666 return skb;
667}
668
669/* Instead of calling kfree_skb() while root qdisc lock is held,
670 * queue the skb for future freeing at end of __dev_xmit_skb()
671 */
672static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
673{
674 skb->next = *to_free;
675 *to_free = skb;
676}
677
678static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
679 struct qdisc_skb_head *qh,
680 struct sk_buff **to_free)
681{
682 struct sk_buff *skb = __qdisc_dequeue_head(qh);
683
684 if (likely(skb != NULL)) {
685 unsigned int len = qdisc_pkt_len(skb);
686
687 qdisc_qstats_backlog_dec(sch, skb);
688 __qdisc_drop(skb, to_free);
689 return len;
690 }
691
692 return 0;
693}
694
695static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
696 struct sk_buff **to_free)
697{
698 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
699}
700
701static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
702{
703 const struct qdisc_skb_head *qh = &sch->q;
704
705 return qh->head;
706}
707
708/* generic pseudo peek method for non-work-conserving qdisc */
709static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
710{
711 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
712 if (!sch->gso_skb) {
713 sch->gso_skb = sch->dequeue(sch);
714 if (sch->gso_skb) {
715 /* it's still part of the queue */
716 qdisc_qstats_backlog_inc(sch, sch->gso_skb);
717 sch->q.qlen++;
718 }
719 }
720
721 return sch->gso_skb;
722}
723
724/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
725static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
726{
727 struct sk_buff *skb = sch->gso_skb;
728
729 if (skb) {
730 sch->gso_skb = NULL;
731 qdisc_qstats_backlog_dec(sch, skb);
732 sch->q.qlen--;
733 } else {
734 skb = sch->dequeue(sch);
735 }
736
737 return skb;
738}
739
740static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
741{
742 /*
743 * We do not know the backlog in bytes of this list, it
744 * is up to the caller to correct it
745 */
746 ASSERT_RTNL();
747 if (qh->qlen) {
748 rtnl_kfree_skbs(qh->head, qh->tail);
749
750 qh->head = NULL;
751 qh->tail = NULL;
752 qh->qlen = 0;
753 }
754}
755
756static inline void qdisc_reset_queue(struct Qdisc *sch)
757{
758 __qdisc_reset_queue(&sch->q);
759 sch->qstats.backlog = 0;
760}
761
762static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
763 struct Qdisc **pold)
764{
765 struct Qdisc *old;
766
767 sch_tree_lock(sch);
768 old = *pold;
769 *pold = new;
770 if (old != NULL) {
771 qdisc_tree_reduce_backlog(old, old->q.qlen, old->qstats.backlog);
772 qdisc_reset(old);
773 }
774 sch_tree_unlock(sch);
775
776 return old;
777}
778
779static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
780{
781 rtnl_kfree_skbs(skb, skb);
782 qdisc_qstats_drop(sch);
783}
784
785
786static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
787 struct sk_buff **to_free)
788{
789 __qdisc_drop(skb, to_free);
790 qdisc_qstats_drop(sch);
791
792 return NET_XMIT_DROP;
793}
794
795/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
796 long it will take to send a packet given its size.
797 */
798static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
799{
800 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
801 if (slot < 0)
802 slot = 0;
803 slot >>= rtab->rate.cell_log;
804 if (slot > 255)
805 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
806 return rtab->data[slot];
807}
808
809struct psched_ratecfg {
810 u64 rate_bytes_ps; /* bytes per second */
811 u32 mult;
812 u16 overhead;
813 u8 linklayer;
814 u8 shift;
815};
816
817static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
818 unsigned int len)
819{
820 len += r->overhead;
821
822 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
823 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
824
825 return ((u64)len * r->mult) >> r->shift;
826}
827
828void psched_ratecfg_precompute(struct psched_ratecfg *r,
829 const struct tc_ratespec *conf,
830 u64 rate64);
831
832static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
833 const struct psched_ratecfg *r)
834{
835 memset(res, 0, sizeof(*res));
836
837 /* legacy struct tc_ratespec has a 32bit @rate field
838 * Qdisc using 64bit rate should add new attributes
839 * in order to maintain compatibility.
840 */
841 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
842
843 res->overhead = r->overhead;
844 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
845}
846
847#endif