?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
split dev_queue
[cor.git] / include / net / sch_generic.h
blob144f264ea394d18d92d1824926abf2caa3245e36
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 <linux/mutex.h>
16 #include <linux/rwsem.h>
17 #include <linux/atomic.h>
18 #include <linux/hashtable.h>
19 #include <net/gen_stats.h>
20 #include <net/rtnetlink.h>
21 #include <net/flow_offload.h>
22
23 struct Qdisc_ops;
24 struct qdisc_walker;
25 struct tcf_walker;
26 struct module;
27 struct bpf_flow_keys;
28
29 struct qdisc_rate_table {
30 struct tc_ratespec rate;
31 u32 data[256];
32 struct qdisc_rate_table *next;
33 int refcnt;
34 };
35
36 enum qdisc_state_t {
37 __QDISC_STATE_SCHED,
38 __QDISC_STATE_DEACTIVATED,
39 };
40
41 struct qdisc_size_table {
42 struct rcu_head rcu;
43 struct list_head list;
44 struct tc_sizespec szopts;
45 int refcnt;
46 u16 data[];
47 };
48
49 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
50 struct qdisc_skb_head {
51 struct sk_buff *head;
52 struct sk_buff *tail;
53 __u32 qlen;
54 spinlock_t lock;
55 };
56
57 struct Qdisc {
58 int (*enqueue)(struct sk_buff *skb,
59 struct Qdisc *sch,
60 struct sk_buff **to_free);
61 struct sk_buff * (*dequeue)(struct Qdisc *sch);
62 unsigned int flags;
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
72 * multiqueue device.
73 */
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.
78 */
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 */
82 u32 limit;
83 const struct Qdisc_ops *ops;
84 struct qdisc_size_table __rcu *stab;
85 struct hlist_node hash;
86 u32 handle;
87 u32 parent;
88
89 struct netdev_queue *dev_queue;
90
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;
94 int padded;
95 refcount_t refcnt;
96
97 /*
98 * For performance sake on SMP, we put highly modified fields at the end
99 */
100 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
101 struct qdisc_skb_head q;
102 struct gnet_stats_basic_packed bstats;
103 seqcount_t running;
104 struct gnet_stats_queue qstats;
105 unsigned long state;
106 struct Qdisc *next_sched;
107 struct sk_buff_head skb_bad_txq;
108
109 spinlock_t busylock ____cacheline_aligned_in_smp;
110 spinlock_t seqlock;
111
112 /* for NOLOCK qdisc, true if there are no enqueued skbs */
113 bool empty;
114 struct rcu_head rcu;
115 };
116
117 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
118 {
119 if (qdisc->flags & TCQ_F_BUILTIN)
120 return;
121 refcount_inc(&qdisc->refcnt);
122 }
123
124 /* Intended to be used by unlocked users, when concurrent qdisc release is
125 * possible.
126 */
127
128 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
129 {
130 if (qdisc->flags & TCQ_F_BUILTIN)
131 return qdisc;
132 if (refcount_inc_not_zero(&qdisc->refcnt))
133 return qdisc;
134 return NULL;
135 }
136
137 static inline bool qdisc_is_running(struct Qdisc *qdisc)
138 {
139 if (qdisc->flags & TCQ_F_NOLOCK)
140 return spin_is_locked(&qdisc->seqlock);
141 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
142 }
143
144 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
145 {
146 return q->flags & TCQ_F_CPUSTATS;
147 }
148
149 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
150 {
151 if (qdisc_is_percpu_stats(qdisc))
152 return READ_ONCE(qdisc->empty);
153 return !READ_ONCE(qdisc->q.qlen);
154 }
155
156 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
157 {
158 if (qdisc->flags & TCQ_F_NOLOCK) {
159 if (!spin_trylock(&qdisc->seqlock))
160 return false;
161 WRITE_ONCE(qdisc->empty, false);
162 } else if (qdisc_is_running(qdisc)) {
163 return false;
164 }
165 /* Variant of write_seqcount_begin() telling lockdep a trylock
166 * was attempted.
167 */
168 raw_write_seqcount_begin(&qdisc->running);
169 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
170 return true;
171 }
172
173 static inline void qdisc_run_end(struct Qdisc *qdisc)
174 {
175 write_seqcount_end(&qdisc->running);
176 if (qdisc->flags & TCQ_F_NOLOCK)
177 spin_unlock(&qdisc->seqlock);
178 }
179
180 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
181 {
182 return qdisc->flags & TCQ_F_ONETXQUEUE;
183 }
184
185 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
186 {
187 #ifdef CONFIG_BQL
188 /* Non-BQL migrated drivers will return 0, too. */
189 return dql_avail(&txq->dql);
190 #else
191 return 0;
192 #endif
193 }
194
195 struct Qdisc_class_ops {
196 unsigned int flags;
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);
204
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);
212
213 /* Filter manipulation */
214 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
215 unsigned long arg,
216 struct netlink_ext_ack *extack);
217 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
218 u32 classid);
219 void (*unbind_tcf)(struct Qdisc *, unsigned long);
220
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,
225 struct gnet_dump *);
226 };
227
228 /* Qdisc_class_ops flag values */
229
230 /* Implements API that doesn't require rtnl lock */
231 enum qdisc_class_ops_flags {
232 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
233 };
234
235 struct Qdisc_ops {
236 struct Qdisc_ops *next;
237 const struct Qdisc_class_ops *cl_ops;
238 char id[IFNAMSIZ];
239 int priv_size;
240 unsigned int static_flags;
241
242 int (*enqueue)(struct sk_buff *skb,
243 struct Qdisc *sch,
244 struct sk_buff **to_free);
245 struct sk_buff * (*dequeue)(struct Qdisc *);
246 struct sk_buff * (*peek)(struct Qdisc *);
247
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,
253 struct nlattr *arg,
254 struct netlink_ext_ack *extack);
255 void (*attach)(struct Qdisc *sch);
256 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
257
258 int (*dump)(struct Qdisc *, struct sk_buff *);
259 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
260
261 void (*ingress_block_set)(struct Qdisc *sch,
262 u32 block_index);
263 void (*egress_block_set)(struct Qdisc *sch,
264 u32 block_index);
265 u32 (*ingress_block_get)(struct Qdisc *sch);
266 u32 (*egress_block_get)(struct Qdisc *sch);
267
268 struct module *owner;
269 };
270
271
272 struct tcf_result {
273 union {
274 struct {
275 unsigned long class;
276 u32 classid;
277 };
278 const struct tcf_proto *goto_tp;
279
280 /* used in the skb_tc_reinsert function */
281 struct {
282 bool ingress;
283 struct gnet_stats_queue *qstats;
284 };
285 };
286 };
287
288 struct tcf_chain;
289
290 struct tcf_proto_ops {
291 struct list_head head;
292 char kind[IFNAMSIZ];
293
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);
300
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 **,
306 void **, bool, bool,
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 flow_setup_cb_t *cb, void *cb_priv,
315 struct netlink_ext_ack *extack);
316 void (*hw_add)(struct tcf_proto *tp,
317 void *type_data);
318 void (*hw_del)(struct tcf_proto *tp,
319 void *type_data);
320 void (*bind_class)(void *, u32, unsigned long);
321 void * (*tmplt_create)(struct net *net,
322 struct tcf_chain *chain,
323 struct nlattr **tca,
324 struct netlink_ext_ack *extack);
325 void (*tmplt_destroy)(void *tmplt_priv);
326
327 /* rtnetlink specific */
328 int (*dump)(struct net*, struct tcf_proto*, void *,
329 struct sk_buff *skb, struct tcmsg*,
330 bool);
331 int (*tmplt_dump)(struct sk_buff *skb,
332 struct net *net,
333 void *tmplt_priv);
334
335 struct module *owner;
336 int flags;
337 };
338
339 enum tcf_proto_ops_flags {
340 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
341 };
342
343 struct tcf_proto {
344 /* Fast access part */
345 struct tcf_proto __rcu *next;
346 void __rcu *root;
347
348 /* called under RCU BH lock*/
349 int (*classify)(struct sk_buff *,
350 const struct tcf_proto *,
351 struct tcf_result *);
352 __be16 protocol;
353
354 /* All the rest */
355 u32 prio;
356 void *data;
357 const struct tcf_proto_ops *ops;
358 struct tcf_chain *chain;
359 /* Lock protects tcf_proto shared state and can be used by unlocked
360 * classifiers to protect their private data.
361 */
362 spinlock_t lock;
363 bool deleting;
364 refcount_t refcnt;
365 struct rcu_head rcu;
366 struct hlist_node destroy_ht_node;
367 };
368
369 struct qdisc_skb_cb {
370 struct {
371 unsigned int pkt_len;
372 u16 slave_dev_queue_mapping;
373 u16 tc_classid;
374 };
375 #define QDISC_CB_PRIV_LEN 20
376 unsigned char data[QDISC_CB_PRIV_LEN];
377 };
378
379 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
380
381 struct tcf_chain {
382 /* Protects filter_chain. */
383 struct mutex filter_chain_lock;
384 struct tcf_proto __rcu *filter_chain;
385 struct list_head list;
386 struct tcf_block *block;
387 u32 index; /* chain index */
388 unsigned int refcnt;
389 unsigned int action_refcnt;
390 bool explicitly_created;
391 bool flushing;
392 const struct tcf_proto_ops *tmplt_ops;
393 void *tmplt_priv;
394 struct rcu_head rcu;
395 };
396
397 struct tcf_block {
398 /* Lock protects tcf_block and lifetime-management data of chains
399 * attached to the block (refcnt, action_refcnt, explicitly_created).
400 */
401 struct mutex lock;
402 struct list_head chain_list;
403 u32 index; /* block index for shared blocks */
404 refcount_t refcnt;
405 struct net *net;
406 struct Qdisc *q;
407 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
408 struct flow_block flow_block;
409 struct list_head owner_list;
410 bool keep_dst;
411 atomic_t offloadcnt; /* Number of oddloaded filters */
412 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
413 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
414 struct {
415 struct tcf_chain *chain;
416 struct list_head filter_chain_list;
417 } chain0;
418 struct rcu_head rcu;
419 DECLARE_HASHTABLE(proto_destroy_ht, 7);
420 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
421 };
422
423 #ifdef CONFIG_PROVE_LOCKING
424 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
425 {
426 return lockdep_is_held(&chain->filter_chain_lock);
427 }
428
429 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
430 {
431 return lockdep_is_held(&tp->lock);
432 }
433 #else
434 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain)
435 {
436 return true;
437 }
438
439 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
440 {
441 return true;
442 }
443 #endif /* #ifdef CONFIG_PROVE_LOCKING */
444
445 #define tcf_chain_dereference(p, chain) \
446 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
447
448 #define tcf_proto_dereference(p, tp) \
449 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
450
451 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
452 {
453 struct qdisc_skb_cb *qcb;
454
455 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
456 BUILD_BUG_ON(sizeof(qcb->data) < sz);
457 }
458
459 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
460 {
461 return this_cpu_ptr(q->cpu_qstats)->qlen;
462 }
463
464 static inline int qdisc_qlen(const struct Qdisc *q)
465 {
466 return q->q.qlen;
467 }
468
469 static inline int qdisc_qlen_sum(const struct Qdisc *q)
470 {
471 __u32 qlen = q->qstats.qlen;
472 int i;
473
474 if (qdisc_is_percpu_stats(q)) {
475 for_each_possible_cpu(i)
476 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
477 } else {
478 qlen += q->q.qlen;
479 }
480
481 return qlen;
482 }
483
484 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
485 {
486 return (struct qdisc_skb_cb *)skb->cb;
487 }
488
489 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
490 {
491 return &qdisc->q.lock;
492 }
493
494 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
495 {
496 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
497
498 return q;
499 }
500
501 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
502 {
503 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
504 }
505
506 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
507 {
508 return qdisc->dev_queue->qdisc_sleeping;
509 }
510
511 /* The qdisc root lock is a mechanism by which to top level
512 * of a qdisc tree can be locked from any qdisc node in the
513 * forest. This allows changing the configuration of some
514 * aspect of the qdisc tree while blocking out asynchronous
515 * qdisc access in the packet processing paths.
516 *
517 * It is only legal to do this when the root will not change
518 * on us. Otherwise we'll potentially lock the wrong qdisc
519 * root. This is enforced by holding the RTNL semaphore, which
520 * all users of this lock accessor must do.
521 */
522 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
523 {
524 struct Qdisc *root = qdisc_root(qdisc);
525
526 ASSERT_RTNL();
527 return qdisc_lock(root);
528 }
529
530 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
531 {
532 struct Qdisc *root = qdisc_root_sleeping(qdisc);
533
534 ASSERT_RTNL();
535 return qdisc_lock(root);
536 }
537
538 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
539 {
540 struct Qdisc *root = qdisc_root_sleeping(qdisc);
541
542 ASSERT_RTNL();
543 return &root->running;
544 }
545
546 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
547 {
548 return qdisc->dev_queue->dev;
549 }
550
551 static inline void sch_tree_lock(const struct Qdisc *q)
552 {
553 spin_lock_bh(qdisc_root_sleeping_lock(q));
554 }
555
556 static inline void sch_tree_unlock(const struct Qdisc *q)
557 {
558 spin_unlock_bh(qdisc_root_sleeping_lock(q));
559 }
560
561 extern struct Qdisc noop_qdisc;
562 extern struct Qdisc_ops noop_qdisc_ops;
563 extern struct Qdisc_ops pfifo_fast_ops;
564 extern struct Qdisc_ops mq_qdisc_ops;
565 extern struct Qdisc_ops noqueue_qdisc_ops;
566 extern const struct Qdisc_ops *default_qdisc_ops;
567 static inline const struct Qdisc_ops *
568 get_default_qdisc_ops(const struct net_device *dev, int ntx)
569 {
570 return ntx < dev->real_num_tx_queues ?
571 default_qdisc_ops : &pfifo_fast_ops;
572 }
573
574 struct Qdisc_class_common {
575 u32 classid;
576 struct hlist_node hnode;
577 };
578
579 struct Qdisc_class_hash {
580 struct hlist_head *hash;
581 unsigned int hashsize;
582 unsigned int hashmask;
583 unsigned int hashelems;
584 };
585
586 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
587 {
588 id ^= id >> 8;
589 id ^= id >> 4;
590 return id & mask;
591 }
592
593 static inline struct Qdisc_class_common *
594 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
595 {
596 struct Qdisc_class_common *cl;
597 unsigned int h;
598
599 if (!id)
600 return NULL;
601
602 h = qdisc_class_hash(id, hash->hashmask);
603 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
604 if (cl->classid == id)
605 return cl;
606 }
607 return NULL;
608 }
609
610 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
611 {
612 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
613
614 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
615 }
616
617 int qdisc_class_hash_init(struct Qdisc_class_hash *);
618 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
619 struct Qdisc_class_common *);
620 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
621 struct Qdisc_class_common *);
622 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
623 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
624
625 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
626 void dev_init_scheduler(struct net_device *dev);
627 void dev_shutdown(struct net_device *dev);
628 void dev_activate(struct net_device *dev);
629 void dev_deactivate(struct net_device *dev);
630 void dev_deactivate_many(struct list_head *head);
631 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
632 struct Qdisc *qdisc);
633 void qdisc_reset(struct Qdisc *qdisc);
634 void qdisc_put(struct Qdisc *qdisc);
635 void qdisc_put_unlocked(struct Qdisc *qdisc);
636 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
637 #ifdef CONFIG_NET_SCHED
638 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
639 void *type_data);
640 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
641 struct Qdisc *new, struct Qdisc *old,
642 enum tc_setup_type type, void *type_data,
643 struct netlink_ext_ack *extack);
644 #else
645 static inline int
646 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
647 void *type_data)
648 {
649 q->flags &= ~TCQ_F_OFFLOADED;
650 return 0;
651 }
652
653 static inline void
654 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
655 struct Qdisc *new, struct Qdisc *old,
656 enum tc_setup_type type, void *type_data,
657 struct netlink_ext_ack *extack)
658 {
659 }
660 #endif
661 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
662 const struct Qdisc_ops *ops,
663 struct netlink_ext_ack *extack);
664 void qdisc_free(struct Qdisc *qdisc);
665 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
666 const struct Qdisc_ops *ops, u32 parentid,
667 struct netlink_ext_ack *extack);
668 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
669 const struct qdisc_size_table *stab);
670 int skb_do_redirect(struct sk_buff *);
671
672 static inline void skb_reset_tc(struct sk_buff *skb)
673 {
674 #ifdef CONFIG_NET_CLS_ACT
675 skb->tc_redirected = 0;
676 #endif
677 }
678
679 static inline bool skb_is_tc_redirected(const struct sk_buff *skb)
680 {
681 #ifdef CONFIG_NET_CLS_ACT
682 return skb->tc_redirected;
683 #else
684 return false;
685 #endif
686 }
687
688 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
689 {
690 #ifdef CONFIG_NET_CLS_ACT
691 return skb->tc_at_ingress;
692 #else
693 return false;
694 #endif
695 }
696
697 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
698 {
699 #ifdef CONFIG_NET_CLS_ACT
700 if (skb->tc_skip_classify) {
701 skb->tc_skip_classify = 0;
702 return true;
703 }
704 #endif
705 return false;
706 }
707
708 /* Reset all TX qdiscs greater than index of a device. */
709 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
710 {
711 struct Qdisc *qdisc;
712
713 for (; i < dev->num_tx_queues; i++) {
714 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
715 if (qdisc) {
716 spin_lock_bh(qdisc_lock(qdisc));
717 qdisc_reset(qdisc);
718 spin_unlock_bh(qdisc_lock(qdisc));
719 }
720 }
721 }
722
723 static inline void qdisc_reset_all_tx(struct net_device *dev)
724 {
725 qdisc_reset_all_tx_gt(dev, 0);
726 }
727
728 /* Are all TX queues of the device empty? */
729 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
730 {
731 unsigned int i;
732
733 rcu_read_lock();
734 for (i = 0; i < dev->num_tx_queues; i++) {
735 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
736 const struct Qdisc *q = rcu_dereference(txq->qdisc);
737
738 if (!qdisc_is_empty(q)) {
739 rcu_read_unlock();
740 return false;
741 }
742 }
743 rcu_read_unlock();
744 return true;
745 }
746
747 /* Are any of the TX qdiscs changing? */
748 static inline bool qdisc_tx_changing(const struct net_device *dev)
749 {
750 unsigned int i;
751
752 for (i = 0; i < dev->num_tx_queues; i++) {
753 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
754 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
755 return true;
756 }
757 return false;
758 }
759
760 /* Is the device using the noop qdisc on all queues? */
761 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
762 {
763 unsigned int i;
764
765 for (i = 0; i < dev->num_tx_queues; i++) {
766 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
767 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
768 return false;
769 }
770 return true;
771 }
772
773 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
774 {
775 return qdisc_skb_cb(skb)->pkt_len;
776 }
777
778 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
779 enum net_xmit_qdisc_t {
780 __NET_XMIT_STOLEN = 0x00010000,
781 __NET_XMIT_BYPASS = 0x00020000,
782 };
783
784 #ifdef CONFIG_NET_CLS_ACT
785 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
786 #else
787 #define net_xmit_drop_count(e) (1)
788 #endif
789
790 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
791 const struct Qdisc *sch)
792 {
793 #ifdef CONFIG_NET_SCHED
794 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
795
796 if (stab)
797 __qdisc_calculate_pkt_len(skb, stab);
798 #endif
799 }
800
801 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
802 struct sk_buff **to_free)
803 {
804 qdisc_calculate_pkt_len(skb, sch);
805 return sch->enqueue(skb, sch, to_free);
806 }
807
808 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
809 __u64 bytes, __u32 packets)
810 {
811 bstats->bytes += bytes;
812 bstats->packets += packets;
813 }
814
815 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
816 const struct sk_buff *skb)
817 {
818 _bstats_update(bstats,
819 qdisc_pkt_len(skb),
820 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
821 }
822
823 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
824 __u64 bytes, __u32 packets)
825 {
826 u64_stats_update_begin(&bstats->syncp);
827 _bstats_update(&bstats->bstats, bytes, packets);
828 u64_stats_update_end(&bstats->syncp);
829 }
830
831 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
832 const struct sk_buff *skb)
833 {
834 u64_stats_update_begin(&bstats->syncp);
835 bstats_update(&bstats->bstats, skb);
836 u64_stats_update_end(&bstats->syncp);
837 }
838
839 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
840 const struct sk_buff *skb)
841 {
842 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
843 }
844
845 static inline void qdisc_bstats_update(struct Qdisc *sch,
846 const struct sk_buff *skb)
847 {
848 bstats_update(&sch->bstats, skb);
849 }
850
851 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
852 const struct sk_buff *skb)
853 {
854 sch->qstats.backlog -= qdisc_pkt_len(skb);
855 }
856
857 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
858 const struct sk_buff *skb)
859 {
860 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
861 }
862
863 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
864 const struct sk_buff *skb)
865 {
866 sch->qstats.backlog += qdisc_pkt_len(skb);
867 }
868
869 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
870 const struct sk_buff *skb)
871 {
872 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
873 }
874
875 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
876 {
877 this_cpu_inc(sch->cpu_qstats->qlen);
878 }
879
880 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
881 {
882 this_cpu_dec(sch->cpu_qstats->qlen);
883 }
884
885 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
886 {
887 this_cpu_inc(sch->cpu_qstats->requeues);
888 }
889
890 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
891 {
892 sch->qstats.drops += count;
893 }
894
895 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
896 {
897 qstats->drops++;
898 }
899
900 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
901 {
902 qstats->overlimits++;
903 }
904
905 static inline void qdisc_qstats_drop(struct Qdisc *sch)
906 {
907 qstats_drop_inc(&sch->qstats);
908 }
909
910 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
911 {
912 this_cpu_inc(sch->cpu_qstats->drops);
913 }
914
915 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
916 {
917 sch->qstats.overlimits++;
918 }
919
920 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
921 {
922 __u32 qlen = qdisc_qlen_sum(sch);
923
924 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
925 }
926
927 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
928 __u32 *backlog)
929 {
930 struct gnet_stats_queue qstats = { 0 };
931 __u32 len = qdisc_qlen_sum(sch);
932
933 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
934 *qlen = qstats.qlen;
935 *backlog = qstats.backlog;
936 }
937
938 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
939 {
940 __u32 qlen, backlog;
941
942 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
943 qdisc_tree_reduce_backlog(sch, qlen, backlog);
944 }
945
946 static inline void qdisc_purge_queue(struct Qdisc *sch)
947 {
948 __u32 qlen, backlog;
949
950 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
951 qdisc_reset(sch);
952 qdisc_tree_reduce_backlog(sch, qlen, backlog);
953 }
954
955 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
956 {
957 qh->head = NULL;
958 qh->tail = NULL;
959 qh->qlen = 0;
960 }
961
962 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
963 struct qdisc_skb_head *qh)
964 {
965 struct sk_buff *last = qh->tail;
966
967 if (last) {
968 skb->next = NULL;
969 last->next = skb;
970 qh->tail = skb;
971 } else {
972 qh->tail = skb;
973 qh->head = skb;
974 }
975 qh->qlen++;
976 }
977
978 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
979 {
980 __qdisc_enqueue_tail(skb, &sch->q);
981 qdisc_qstats_backlog_inc(sch, skb);
982 return NET_XMIT_SUCCESS;
983 }
984
985 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
986 struct qdisc_skb_head *qh)
987 {
988 skb->next = qh->head;
989
990 if (!qh->head)
991 qh->tail = skb;
992 qh->head = skb;
993 qh->qlen++;
994 }
995
996 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
997 {
998 struct sk_buff *skb = qh->head;
999
1000 if (likely(skb != NULL)) {
1001 qh->head = skb->next;
1002 qh->qlen--;
1003 if (qh->head == NULL)
1004 qh->tail = NULL;
1005 skb->next = NULL;
1006 }
1007
1008 return skb;
1009 }
1010
1011 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1012 {
1013 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1014
1015 if (likely(skb != NULL)) {
1016 qdisc_qstats_backlog_dec(sch, skb);
1017 qdisc_bstats_update(sch, skb);
1018 }
1019
1020 return skb;
1021 }
1022
1023 /* Instead of calling kfree_skb() while root qdisc lock is held,
1024 * queue the skb for future freeing at end of __dev_xmit_skb()
1025 */
1026 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1027 {
1028 skb->next = *to_free;
1029 *to_free = skb;
1030 }
1031
1032 static inline void __qdisc_drop_all(struct sk_buff *skb,
1033 struct sk_buff **to_free)
1034 {
1035 if (skb->prev)
1036 skb->prev->next = *to_free;
1037 else
1038 skb->next = *to_free;
1039 *to_free = skb;
1040 }
1041
1042 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1043 struct qdisc_skb_head *qh,
1044 struct sk_buff **to_free)
1045 {
1046 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1047
1048 if (likely(skb != NULL)) {
1049 unsigned int len = qdisc_pkt_len(skb);
1050
1051 qdisc_qstats_backlog_dec(sch, skb);
1052 __qdisc_drop(skb, to_free);
1053 return len;
1054 }
1055
1056 return 0;
1057 }
1058
1059 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
1060 struct sk_buff **to_free)
1061 {
1062 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
1063 }
1064
1065 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1066 {
1067 const struct qdisc_skb_head *qh = &sch->q;
1068
1069 return qh->head;
1070 }
1071
1072 /* generic pseudo peek method for non-work-conserving qdisc */
1073 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1074 {
1075 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1076
1077 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1078 if (!skb) {
1079 skb = sch->dequeue(sch);
1080
1081 if (skb) {
1082 __skb_queue_head(&sch->gso_skb, skb);
1083 /* it's still part of the queue */
1084 qdisc_qstats_backlog_inc(sch, skb);
1085 sch->q.qlen++;
1086 }
1087 }
1088
1089 return skb;
1090 }
1091
1092 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1093 struct sk_buff *skb)
1094 {
1095 if (qdisc_is_percpu_stats(sch)) {
1096 qdisc_qstats_cpu_backlog_dec(sch, skb);
1097 qdisc_bstats_cpu_update(sch, skb);
1098 qdisc_qstats_cpu_qlen_dec(sch);
1099 } else {
1100 qdisc_qstats_backlog_dec(sch, skb);
1101 qdisc_bstats_update(sch, skb);
1102 sch->q.qlen--;
1103 }
1104 }
1105
1106 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1107 unsigned int pkt_len)
1108 {
1109 if (qdisc_is_percpu_stats(sch)) {
1110 qdisc_qstats_cpu_qlen_inc(sch);
1111 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1112 } else {
1113 sch->qstats.backlog += pkt_len;
1114 sch->q.qlen++;
1115 }
1116 }
1117
1118 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1119 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1120 {
1121 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1122
1123 if (skb) {
1124 skb = __skb_dequeue(&sch->gso_skb);
1125 if (qdisc_is_percpu_stats(sch)) {
1126 qdisc_qstats_cpu_backlog_dec(sch, skb);
1127 qdisc_qstats_cpu_qlen_dec(sch);
1128 } else {
1129 qdisc_qstats_backlog_dec(sch, skb);
1130 sch->q.qlen--;
1131 }
1132 } else {
1133 skb = sch->dequeue(sch);
1134 }
1135
1136 return skb;
1137 }
1138
1139 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1140 {
1141 /*
1142 * We do not know the backlog in bytes of this list, it
1143 * is up to the caller to correct it
1144 */
1145 ASSERT_RTNL();
1146 if (qh->qlen) {
1147 rtnl_kfree_skbs(qh->head, qh->tail);
1148
1149 qh->head = NULL;
1150 qh->tail = NULL;
1151 qh->qlen = 0;
1152 }
1153 }
1154
1155 static inline void qdisc_reset_queue(struct Qdisc *sch)
1156 {
1157 __qdisc_reset_queue(&sch->q);
1158 sch->qstats.backlog = 0;
1159 }
1160
1161 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1162 struct Qdisc **pold)
1163 {
1164 struct Qdisc *old;
1165
1166 sch_tree_lock(sch);
1167 old = *pold;
1168 *pold = new;
1169 if (old != NULL)
1170 qdisc_tree_flush_backlog(old);
1171 sch_tree_unlock(sch);
1172
1173 return old;
1174 }
1175
1176 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1177 {
1178 rtnl_kfree_skbs(skb, skb);
1179 qdisc_qstats_drop(sch);
1180 }
1181
1182 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1183 struct sk_buff **to_free)
1184 {
1185 __qdisc_drop(skb, to_free);
1186 qdisc_qstats_cpu_drop(sch);
1187
1188 return NET_XMIT_DROP;
1189 }
1190
1191 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1192 struct sk_buff **to_free)
1193 {
1194 __qdisc_drop(skb, to_free);
1195 qdisc_qstats_drop(sch);
1196
1197 return NET_XMIT_DROP;
1198 }
1199
1200 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1201 struct sk_buff **to_free)
1202 {
1203 __qdisc_drop_all(skb, to_free);
1204 qdisc_qstats_drop(sch);
1205
1206 return NET_XMIT_DROP;
1207 }
1208
1209 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1210 long it will take to send a packet given its size.
1211 */
1212 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1213 {
1214 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1215 if (slot < 0)
1216 slot = 0;
1217 slot >>= rtab->rate.cell_log;
1218 if (slot > 255)
1219 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1220 return rtab->data[slot];
1221 }
1222
1223 struct psched_ratecfg {
1224 u64 rate_bytes_ps; /* bytes per second */
1225 u32 mult;
1226 u16 overhead;
1227 u8 linklayer;
1228 u8 shift;
1229 };
1230
1231 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1232 unsigned int len)
1233 {
1234 len += r->overhead;
1235
1236 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1237 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1238
1239 return ((u64)len * r->mult) >> r->shift;
1240 }
1241
1242 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1243 const struct tc_ratespec *conf,
1244 u64 rate64);
1245
1246 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1247 const struct psched_ratecfg *r)
1248 {
1249 memset(res, 0, sizeof(*res));
1250
1251 /* legacy struct tc_ratespec has a 32bit @rate field
1252 * Qdisc using 64bit rate should add new attributes
1253 * in order to maintain compatibility.
1254 */
1255 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1256
1257 res->overhead = r->overhead;
1258 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1259 }
1260
1261 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1262 * The fast path only needs to access filter list and to update stats
1263 */
1264 struct mini_Qdisc {
1265 struct tcf_proto *filter_list;
1266 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1267 struct gnet_stats_queue __percpu *cpu_qstats;
1268 struct rcu_head rcu;
1269 };
1270
1271 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1272 const struct sk_buff *skb)
1273 {
1274 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1275 }
1276
1277 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1278 {
1279 this_cpu_inc(miniq->cpu_qstats->drops);
1280 }
1281
1282 struct mini_Qdisc_pair {
1283 struct mini_Qdisc miniq1;
1284 struct mini_Qdisc miniq2;
1285 struct mini_Qdisc __rcu **p_miniq;
1286 };
1287
1288 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1289 struct tcf_proto *tp_head);
1290 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1291 struct mini_Qdisc __rcu **p_miniq);
1292
1293 static inline int skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res)
1294 {
1295 return res->ingress ? netif_receive_skb(skb) : dev_queue_xmit(skb);
1296 }
1297
1298 #endif
connection oriented routing