]> git.ipfire.org Git - thirdparty/linux.git/blob - include/net/sch_generic.h
io_uring: reset -EBUSY error when io sq thread is waken up
[thirdparty/linux.git] / include / net / sch_generic.h
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 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,
318 void *type_data);
319 void (*hw_del)(struct tcf_proto *tp,
320 void *type_data);
321 void (*bind_class)(void *, u32, unsigned long,
322 void *, unsigned long);
323 void * (*tmplt_create)(struct net *net,
324 struct tcf_chain *chain,
325 struct nlattr **tca,
326 struct netlink_ext_ack *extack);
327 void (*tmplt_destroy)(void *tmplt_priv);
328
329 /* rtnetlink specific */
330 int (*dump)(struct net*, struct tcf_proto*, void *,
331 struct sk_buff *skb, struct tcmsg*,
332 bool);
333 int (*tmplt_dump)(struct sk_buff *skb,
334 struct net *net,
335 void *tmplt_priv);
336
337 struct module *owner;
338 int flags;
339 };
340
341 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
342 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
343 * conditions can occur when filters are inserted/deleted simultaneously.
344 */
345 enum tcf_proto_ops_flags {
346 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
347 };
348
349 struct tcf_proto {
350 /* Fast access part */
351 struct tcf_proto __rcu *next;
352 void __rcu *root;
353
354 /* called under RCU BH lock*/
355 int (*classify)(struct sk_buff *,
356 const struct tcf_proto *,
357 struct tcf_result *);
358 __be16 protocol;
359
360 /* All the rest */
361 u32 prio;
362 void *data;
363 const struct tcf_proto_ops *ops;
364 struct tcf_chain *chain;
365 /* Lock protects tcf_proto shared state and can be used by unlocked
366 * classifiers to protect their private data.
367 */
368 spinlock_t lock;
369 bool deleting;
370 refcount_t refcnt;
371 struct rcu_head rcu;
372 struct hlist_node destroy_ht_node;
373 };
374
375 struct qdisc_skb_cb {
376 struct {
377 unsigned int pkt_len;
378 u16 slave_dev_queue_mapping;
379 u16 tc_classid;
380 };
381 #define QDISC_CB_PRIV_LEN 20
382 unsigned char data[QDISC_CB_PRIV_LEN];
383 };
384
385 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
386
387 struct tcf_chain {
388 /* Protects filter_chain. */
389 struct mutex filter_chain_lock;
390 struct tcf_proto __rcu *filter_chain;
391 struct list_head list;
392 struct tcf_block *block;
393 u32 index; /* chain index */
394 unsigned int refcnt;
395 unsigned int action_refcnt;
396 bool explicitly_created;
397 bool flushing;
398 const struct tcf_proto_ops *tmplt_ops;
399 void *tmplt_priv;
400 struct rcu_head rcu;
401 };
402
403 struct tcf_block {
404 /* Lock protects tcf_block and lifetime-management data of chains
405 * attached to the block (refcnt, action_refcnt, explicitly_created).
406 */
407 struct mutex lock;
408 struct list_head chain_list;
409 u32 index; /* block index for shared blocks */
410 refcount_t refcnt;
411 struct net *net;
412 struct Qdisc *q;
413 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
414 struct flow_block flow_block;
415 struct list_head owner_list;
416 bool keep_dst;
417 atomic_t offloadcnt; /* Number of oddloaded filters */
418 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
419 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
420 struct {
421 struct tcf_chain *chain;
422 struct list_head filter_chain_list;
423 } chain0;
424 struct rcu_head rcu;
425 DECLARE_HASHTABLE(proto_destroy_ht, 7);
426 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
427 };
428
429 #ifdef CONFIG_PROVE_LOCKING
430 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
431 {
432 return lockdep_is_held(&chain->filter_chain_lock);
433 }
434
435 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
436 {
437 return lockdep_is_held(&tp->lock);
438 }
439 #else
440 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain)
441 {
442 return true;
443 }
444
445 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
446 {
447 return true;
448 }
449 #endif /* #ifdef CONFIG_PROVE_LOCKING */
450
451 #define tcf_chain_dereference(p, chain) \
452 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
453
454 #define tcf_proto_dereference(p, tp) \
455 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
456
457 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
458 {
459 struct qdisc_skb_cb *qcb;
460
461 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
462 BUILD_BUG_ON(sizeof(qcb->data) < sz);
463 }
464
465 static inline int qdisc_qlen_cpu(const struct Qdisc *q)
466 {
467 return this_cpu_ptr(q->cpu_qstats)->qlen;
468 }
469
470 static inline int qdisc_qlen(const struct Qdisc *q)
471 {
472 return q->q.qlen;
473 }
474
475 static inline int qdisc_qlen_sum(const struct Qdisc *q)
476 {
477 __u32 qlen = q->qstats.qlen;
478 int i;
479
480 if (qdisc_is_percpu_stats(q)) {
481 for_each_possible_cpu(i)
482 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
483 } else {
484 qlen += q->q.qlen;
485 }
486
487 return qlen;
488 }
489
490 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
491 {
492 return (struct qdisc_skb_cb *)skb->cb;
493 }
494
495 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
496 {
497 return &qdisc->q.lock;
498 }
499
500 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
501 {
502 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
503
504 return q;
505 }
506
507 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
508 {
509 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
510 }
511
512 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
513 {
514 return qdisc->dev_queue->qdisc_sleeping;
515 }
516
517 /* The qdisc root lock is a mechanism by which to top level
518 * of a qdisc tree can be locked from any qdisc node in the
519 * forest. This allows changing the configuration of some
520 * aspect of the qdisc tree while blocking out asynchronous
521 * qdisc access in the packet processing paths.
522 *
523 * It is only legal to do this when the root will not change
524 * on us. Otherwise we'll potentially lock the wrong qdisc
525 * root. This is enforced by holding the RTNL semaphore, which
526 * all users of this lock accessor must do.
527 */
528 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
529 {
530 struct Qdisc *root = qdisc_root(qdisc);
531
532 ASSERT_RTNL();
533 return qdisc_lock(root);
534 }
535
536 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
537 {
538 struct Qdisc *root = qdisc_root_sleeping(qdisc);
539
540 ASSERT_RTNL();
541 return qdisc_lock(root);
542 }
543
544 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
545 {
546 struct Qdisc *root = qdisc_root_sleeping(qdisc);
547
548 ASSERT_RTNL();
549 return &root->running;
550 }
551
552 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
553 {
554 return qdisc->dev_queue->dev;
555 }
556
557 static inline void sch_tree_lock(const struct Qdisc *q)
558 {
559 spin_lock_bh(qdisc_root_sleeping_lock(q));
560 }
561
562 static inline void sch_tree_unlock(const struct Qdisc *q)
563 {
564 spin_unlock_bh(qdisc_root_sleeping_lock(q));
565 }
566
567 extern struct Qdisc noop_qdisc;
568 extern struct Qdisc_ops noop_qdisc_ops;
569 extern struct Qdisc_ops pfifo_fast_ops;
570 extern struct Qdisc_ops mq_qdisc_ops;
571 extern struct Qdisc_ops noqueue_qdisc_ops;
572 extern const struct Qdisc_ops *default_qdisc_ops;
573 static inline const struct Qdisc_ops *
574 get_default_qdisc_ops(const struct net_device *dev, int ntx)
575 {
576 return ntx < dev->real_num_tx_queues ?
577 default_qdisc_ops : &pfifo_fast_ops;
578 }
579
580 struct Qdisc_class_common {
581 u32 classid;
582 struct hlist_node hnode;
583 };
584
585 struct Qdisc_class_hash {
586 struct hlist_head *hash;
587 unsigned int hashsize;
588 unsigned int hashmask;
589 unsigned int hashelems;
590 };
591
592 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
593 {
594 id ^= id >> 8;
595 id ^= id >> 4;
596 return id & mask;
597 }
598
599 static inline struct Qdisc_class_common *
600 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
601 {
602 struct Qdisc_class_common *cl;
603 unsigned int h;
604
605 if (!id)
606 return NULL;
607
608 h = qdisc_class_hash(id, hash->hashmask);
609 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
610 if (cl->classid == id)
611 return cl;
612 }
613 return NULL;
614 }
615
616 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
617 {
618 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
619
620 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
621 }
622
623 int qdisc_class_hash_init(struct Qdisc_class_hash *);
624 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
625 struct Qdisc_class_common *);
626 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
627 struct Qdisc_class_common *);
628 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
629 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
630
631 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
632 void dev_init_scheduler(struct net_device *dev);
633 void dev_shutdown(struct net_device *dev);
634 void dev_activate(struct net_device *dev);
635 void dev_deactivate(struct net_device *dev);
636 void dev_deactivate_many(struct list_head *head);
637 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
638 struct Qdisc *qdisc);
639 void qdisc_reset(struct Qdisc *qdisc);
640 void qdisc_put(struct Qdisc *qdisc);
641 void qdisc_put_unlocked(struct Qdisc *qdisc);
642 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
643 #ifdef CONFIG_NET_SCHED
644 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
645 void *type_data);
646 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
647 struct Qdisc *new, struct Qdisc *old,
648 enum tc_setup_type type, void *type_data,
649 struct netlink_ext_ack *extack);
650 #else
651 static inline int
652 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
653 void *type_data)
654 {
655 q->flags &= ~TCQ_F_OFFLOADED;
656 return 0;
657 }
658
659 static inline void
660 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
661 struct Qdisc *new, struct Qdisc *old,
662 enum tc_setup_type type, void *type_data,
663 struct netlink_ext_ack *extack)
664 {
665 }
666 #endif
667 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
668 const struct Qdisc_ops *ops,
669 struct netlink_ext_ack *extack);
670 void qdisc_free(struct Qdisc *qdisc);
671 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
672 const struct Qdisc_ops *ops, u32 parentid,
673 struct netlink_ext_ack *extack);
674 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
675 const struct qdisc_size_table *stab);
676 int skb_do_redirect(struct sk_buff *);
677
678 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
679 {
680 #ifdef CONFIG_NET_CLS_ACT
681 return skb->tc_at_ingress;
682 #else
683 return false;
684 #endif
685 }
686
687 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
688 {
689 #ifdef CONFIG_NET_CLS_ACT
690 if (skb->tc_skip_classify) {
691 skb->tc_skip_classify = 0;
692 return true;
693 }
694 #endif
695 return false;
696 }
697
698 /* Reset all TX qdiscs greater than index of a device. */
699 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
700 {
701 struct Qdisc *qdisc;
702
703 for (; i < dev->num_tx_queues; i++) {
704 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
705 if (qdisc) {
706 spin_lock_bh(qdisc_lock(qdisc));
707 qdisc_reset(qdisc);
708 spin_unlock_bh(qdisc_lock(qdisc));
709 }
710 }
711 }
712
713 static inline void qdisc_reset_all_tx(struct net_device *dev)
714 {
715 qdisc_reset_all_tx_gt(dev, 0);
716 }
717
718 /* Are all TX queues of the device empty? */
719 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
720 {
721 unsigned int i;
722
723 rcu_read_lock();
724 for (i = 0; i < dev->num_tx_queues; i++) {
725 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
726 const struct Qdisc *q = rcu_dereference(txq->qdisc);
727
728 if (!qdisc_is_empty(q)) {
729 rcu_read_unlock();
730 return false;
731 }
732 }
733 rcu_read_unlock();
734 return true;
735 }
736
737 /* Are any of the TX qdiscs changing? */
738 static inline bool qdisc_tx_changing(const struct net_device *dev)
739 {
740 unsigned int i;
741
742 for (i = 0; i < dev->num_tx_queues; i++) {
743 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
744 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
745 return true;
746 }
747 return false;
748 }
749
750 /* Is the device using the noop qdisc on all queues? */
751 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
752 {
753 unsigned int i;
754
755 for (i = 0; i < dev->num_tx_queues; i++) {
756 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
757 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
758 return false;
759 }
760 return true;
761 }
762
763 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
764 {
765 return qdisc_skb_cb(skb)->pkt_len;
766 }
767
768 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
769 enum net_xmit_qdisc_t {
770 __NET_XMIT_STOLEN = 0x00010000,
771 __NET_XMIT_BYPASS = 0x00020000,
772 };
773
774 #ifdef CONFIG_NET_CLS_ACT
775 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
776 #else
777 #define net_xmit_drop_count(e) (1)
778 #endif
779
780 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
781 const struct Qdisc *sch)
782 {
783 #ifdef CONFIG_NET_SCHED
784 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
785
786 if (stab)
787 __qdisc_calculate_pkt_len(skb, stab);
788 #endif
789 }
790
791 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
792 struct sk_buff **to_free)
793 {
794 qdisc_calculate_pkt_len(skb, sch);
795 return sch->enqueue(skb, sch, to_free);
796 }
797
798 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
799 __u64 bytes, __u32 packets)
800 {
801 bstats->bytes += bytes;
802 bstats->packets += packets;
803 }
804
805 static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
806 const struct sk_buff *skb)
807 {
808 _bstats_update(bstats,
809 qdisc_pkt_len(skb),
810 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
811 }
812
813 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
814 __u64 bytes, __u32 packets)
815 {
816 u64_stats_update_begin(&bstats->syncp);
817 _bstats_update(&bstats->bstats, bytes, packets);
818 u64_stats_update_end(&bstats->syncp);
819 }
820
821 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
822 const struct sk_buff *skb)
823 {
824 u64_stats_update_begin(&bstats->syncp);
825 bstats_update(&bstats->bstats, skb);
826 u64_stats_update_end(&bstats->syncp);
827 }
828
829 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
830 const struct sk_buff *skb)
831 {
832 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
833 }
834
835 static inline void qdisc_bstats_update(struct Qdisc *sch,
836 const struct sk_buff *skb)
837 {
838 bstats_update(&sch->bstats, skb);
839 }
840
841 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
842 const struct sk_buff *skb)
843 {
844 sch->qstats.backlog -= qdisc_pkt_len(skb);
845 }
846
847 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
848 const struct sk_buff *skb)
849 {
850 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
851 }
852
853 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
854 const struct sk_buff *skb)
855 {
856 sch->qstats.backlog += qdisc_pkt_len(skb);
857 }
858
859 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
860 const struct sk_buff *skb)
861 {
862 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
863 }
864
865 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
866 {
867 this_cpu_inc(sch->cpu_qstats->qlen);
868 }
869
870 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
871 {
872 this_cpu_dec(sch->cpu_qstats->qlen);
873 }
874
875 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
876 {
877 this_cpu_inc(sch->cpu_qstats->requeues);
878 }
879
880 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
881 {
882 sch->qstats.drops += count;
883 }
884
885 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
886 {
887 qstats->drops++;
888 }
889
890 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
891 {
892 qstats->overlimits++;
893 }
894
895 static inline void qdisc_qstats_drop(struct Qdisc *sch)
896 {
897 qstats_drop_inc(&sch->qstats);
898 }
899
900 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
901 {
902 this_cpu_inc(sch->cpu_qstats->drops);
903 }
904
905 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
906 {
907 sch->qstats.overlimits++;
908 }
909
910 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
911 {
912 __u32 qlen = qdisc_qlen_sum(sch);
913
914 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
915 }
916
917 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
918 __u32 *backlog)
919 {
920 struct gnet_stats_queue qstats = { 0 };
921 __u32 len = qdisc_qlen_sum(sch);
922
923 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
924 *qlen = qstats.qlen;
925 *backlog = qstats.backlog;
926 }
927
928 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
929 {
930 __u32 qlen, backlog;
931
932 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
933 qdisc_tree_reduce_backlog(sch, qlen, backlog);
934 }
935
936 static inline void qdisc_purge_queue(struct Qdisc *sch)
937 {
938 __u32 qlen, backlog;
939
940 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
941 qdisc_reset(sch);
942 qdisc_tree_reduce_backlog(sch, qlen, backlog);
943 }
944
945 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
946 {
947 qh->head = NULL;
948 qh->tail = NULL;
949 qh->qlen = 0;
950 }
951
952 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
953 struct qdisc_skb_head *qh)
954 {
955 struct sk_buff *last = qh->tail;
956
957 if (last) {
958 skb->next = NULL;
959 last->next = skb;
960 qh->tail = skb;
961 } else {
962 qh->tail = skb;
963 qh->head = skb;
964 }
965 qh->qlen++;
966 }
967
968 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
969 {
970 __qdisc_enqueue_tail(skb, &sch->q);
971 qdisc_qstats_backlog_inc(sch, skb);
972 return NET_XMIT_SUCCESS;
973 }
974
975 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
976 struct qdisc_skb_head *qh)
977 {
978 skb->next = qh->head;
979
980 if (!qh->head)
981 qh->tail = skb;
982 qh->head = skb;
983 qh->qlen++;
984 }
985
986 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
987 {
988 struct sk_buff *skb = qh->head;
989
990 if (likely(skb != NULL)) {
991 qh->head = skb->next;
992 qh->qlen--;
993 if (qh->head == NULL)
994 qh->tail = NULL;
995 skb->next = NULL;
996 }
997
998 return skb;
999 }
1000
1001 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1002 {
1003 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1004
1005 if (likely(skb != NULL)) {
1006 qdisc_qstats_backlog_dec(sch, skb);
1007 qdisc_bstats_update(sch, skb);
1008 }
1009
1010 return skb;
1011 }
1012
1013 /* Instead of calling kfree_skb() while root qdisc lock is held,
1014 * queue the skb for future freeing at end of __dev_xmit_skb()
1015 */
1016 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1017 {
1018 skb->next = *to_free;
1019 *to_free = skb;
1020 }
1021
1022 static inline void __qdisc_drop_all(struct sk_buff *skb,
1023 struct sk_buff **to_free)
1024 {
1025 if (skb->prev)
1026 skb->prev->next = *to_free;
1027 else
1028 skb->next = *to_free;
1029 *to_free = skb;
1030 }
1031
1032 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1033 struct qdisc_skb_head *qh,
1034 struct sk_buff **to_free)
1035 {
1036 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1037
1038 if (likely(skb != NULL)) {
1039 unsigned int len = qdisc_pkt_len(skb);
1040
1041 qdisc_qstats_backlog_dec(sch, skb);
1042 __qdisc_drop(skb, to_free);
1043 return len;
1044 }
1045
1046 return 0;
1047 }
1048
1049 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
1050 struct sk_buff **to_free)
1051 {
1052 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
1053 }
1054
1055 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1056 {
1057 const struct qdisc_skb_head *qh = &sch->q;
1058
1059 return qh->head;
1060 }
1061
1062 /* generic pseudo peek method for non-work-conserving qdisc */
1063 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1064 {
1065 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1066
1067 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1068 if (!skb) {
1069 skb = sch->dequeue(sch);
1070
1071 if (skb) {
1072 __skb_queue_head(&sch->gso_skb, skb);
1073 /* it's still part of the queue */
1074 qdisc_qstats_backlog_inc(sch, skb);
1075 sch->q.qlen++;
1076 }
1077 }
1078
1079 return skb;
1080 }
1081
1082 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1083 struct sk_buff *skb)
1084 {
1085 if (qdisc_is_percpu_stats(sch)) {
1086 qdisc_qstats_cpu_backlog_dec(sch, skb);
1087 qdisc_bstats_cpu_update(sch, skb);
1088 qdisc_qstats_cpu_qlen_dec(sch);
1089 } else {
1090 qdisc_qstats_backlog_dec(sch, skb);
1091 qdisc_bstats_update(sch, skb);
1092 sch->q.qlen--;
1093 }
1094 }
1095
1096 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1097 unsigned int pkt_len)
1098 {
1099 if (qdisc_is_percpu_stats(sch)) {
1100 qdisc_qstats_cpu_qlen_inc(sch);
1101 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1102 } else {
1103 sch->qstats.backlog += pkt_len;
1104 sch->q.qlen++;
1105 }
1106 }
1107
1108 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1109 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1110 {
1111 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1112
1113 if (skb) {
1114 skb = __skb_dequeue(&sch->gso_skb);
1115 if (qdisc_is_percpu_stats(sch)) {
1116 qdisc_qstats_cpu_backlog_dec(sch, skb);
1117 qdisc_qstats_cpu_qlen_dec(sch);
1118 } else {
1119 qdisc_qstats_backlog_dec(sch, skb);
1120 sch->q.qlen--;
1121 }
1122 } else {
1123 skb = sch->dequeue(sch);
1124 }
1125
1126 return skb;
1127 }
1128
1129 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1130 {
1131 /*
1132 * We do not know the backlog in bytes of this list, it
1133 * is up to the caller to correct it
1134 */
1135 ASSERT_RTNL();
1136 if (qh->qlen) {
1137 rtnl_kfree_skbs(qh->head, qh->tail);
1138
1139 qh->head = NULL;
1140 qh->tail = NULL;
1141 qh->qlen = 0;
1142 }
1143 }
1144
1145 static inline void qdisc_reset_queue(struct Qdisc *sch)
1146 {
1147 __qdisc_reset_queue(&sch->q);
1148 sch->qstats.backlog = 0;
1149 }
1150
1151 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1152 struct Qdisc **pold)
1153 {
1154 struct Qdisc *old;
1155
1156 sch_tree_lock(sch);
1157 old = *pold;
1158 *pold = new;
1159 if (old != NULL)
1160 qdisc_tree_flush_backlog(old);
1161 sch_tree_unlock(sch);
1162
1163 return old;
1164 }
1165
1166 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1167 {
1168 rtnl_kfree_skbs(skb, skb);
1169 qdisc_qstats_drop(sch);
1170 }
1171
1172 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1173 struct sk_buff **to_free)
1174 {
1175 __qdisc_drop(skb, to_free);
1176 qdisc_qstats_cpu_drop(sch);
1177
1178 return NET_XMIT_DROP;
1179 }
1180
1181 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1182 struct sk_buff **to_free)
1183 {
1184 __qdisc_drop(skb, to_free);
1185 qdisc_qstats_drop(sch);
1186
1187 return NET_XMIT_DROP;
1188 }
1189
1190 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1191 struct sk_buff **to_free)
1192 {
1193 __qdisc_drop_all(skb, to_free);
1194 qdisc_qstats_drop(sch);
1195
1196 return NET_XMIT_DROP;
1197 }
1198
1199 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1200 long it will take to send a packet given its size.
1201 */
1202 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1203 {
1204 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1205 if (slot < 0)
1206 slot = 0;
1207 slot >>= rtab->rate.cell_log;
1208 if (slot > 255)
1209 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1210 return rtab->data[slot];
1211 }
1212
1213 struct psched_ratecfg {
1214 u64 rate_bytes_ps; /* bytes per second */
1215 u32 mult;
1216 u16 overhead;
1217 u8 linklayer;
1218 u8 shift;
1219 };
1220
1221 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1222 unsigned int len)
1223 {
1224 len += r->overhead;
1225
1226 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1227 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1228
1229 return ((u64)len * r->mult) >> r->shift;
1230 }
1231
1232 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1233 const struct tc_ratespec *conf,
1234 u64 rate64);
1235
1236 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1237 const struct psched_ratecfg *r)
1238 {
1239 memset(res, 0, sizeof(*res));
1240
1241 /* legacy struct tc_ratespec has a 32bit @rate field
1242 * Qdisc using 64bit rate should add new attributes
1243 * in order to maintain compatibility.
1244 */
1245 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1246
1247 res->overhead = r->overhead;
1248 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1249 }
1250
1251 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1252 * The fast path only needs to access filter list and to update stats
1253 */
1254 struct mini_Qdisc {
1255 struct tcf_proto *filter_list;
1256 struct tcf_block *block;
1257 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1258 struct gnet_stats_queue __percpu *cpu_qstats;
1259 struct rcu_head rcu;
1260 };
1261
1262 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1263 const struct sk_buff *skb)
1264 {
1265 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1266 }
1267
1268 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1269 {
1270 this_cpu_inc(miniq->cpu_qstats->drops);
1271 }
1272
1273 struct mini_Qdisc_pair {
1274 struct mini_Qdisc miniq1;
1275 struct mini_Qdisc miniq2;
1276 struct mini_Qdisc __rcu **p_miniq;
1277 };
1278
1279 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1280 struct tcf_proto *tp_head);
1281 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1282 struct mini_Qdisc __rcu **p_miniq);
1283 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1284 struct tcf_block *block);
1285
1286 static inline int skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res)
1287 {
1288 return res->ingress ? netif_receive_skb(skb) : dev_queue_xmit(skb);
1289 }
1290
1291 #endif