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#include <linux/xarray.h>
23
24struct Qdisc_ops;
25struct qdisc_walker;
26struct tcf_walker;
27struct module;
28struct bpf_flow_keys;
29
30struct qdisc_rate_table {
31 struct tc_ratespec rate;
32 u32 data[256];
33 struct qdisc_rate_table *next;
34 int refcnt;
35};
36
37enum qdisc_state_t {
38 __QDISC_STATE_SCHED,
39 __QDISC_STATE_DEACTIVATED,
40 __QDISC_STATE_MISSED,
41 __QDISC_STATE_DRAINING,
42};
43
44enum qdisc_state2_t {
45 /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly.
46 * Use qdisc_run_begin/end() or qdisc_is_running() instead.
47 */
48 __QDISC_STATE2_RUNNING,
49};
50
51#define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED)
52#define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING)
53
54#define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \
55 QDISC_STATE_DRAINING)
56
57struct qdisc_size_table {
58 struct rcu_head rcu;
59 struct list_head list;
60 struct tc_sizespec szopts;
61 int refcnt;
62 u16 data[];
63};
64
65/* similar to sk_buff_head, but skb->prev pointer is undefined. */
66struct qdisc_skb_head {
67 struct sk_buff *head;
68 struct sk_buff *tail;
69 __u32 qlen;
70 spinlock_t lock;
71};
72
73struct Qdisc {
74 int (*enqueue)(struct sk_buff *skb,
75 struct Qdisc *sch,
76 struct sk_buff **to_free);
77 struct sk_buff * (*dequeue)(struct Qdisc *sch);
78 unsigned int flags;
79#define TCQ_F_BUILTIN 1
80#define TCQ_F_INGRESS 2
81#define TCQ_F_CAN_BYPASS 4
82#define TCQ_F_MQROOT 8
83#define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
84 * q->dev_queue : It can test
85 * netif_xmit_frozen_or_stopped() before
86 * dequeueing next packet.
87 * Its true for MQ/MQPRIO slaves, or non
88 * multiqueue device.
89 */
90#define TCQ_F_WARN_NONWC (1 << 16)
91#define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
92#define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
93 * qdisc_tree_decrease_qlen() should stop.
94 */
95#define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
96#define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
97#define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
98 u32 limit;
99 const struct Qdisc_ops *ops;
100 struct qdisc_size_table __rcu *stab;
101 struct hlist_node hash;
102 u32 handle;
103 u32 parent;
104
105 struct netdev_queue *dev_queue;
106
107 struct net_rate_estimator __rcu *rate_est;
108 struct gnet_stats_basic_sync __percpu *cpu_bstats;
109 struct gnet_stats_queue __percpu *cpu_qstats;
110 int pad;
111 refcount_t refcnt;
112
113 /*
114 * For performance sake on SMP, we put highly modified fields at the end
115 */
116 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
117 struct qdisc_skb_head q;
118 struct gnet_stats_basic_sync bstats;
119 struct gnet_stats_queue qstats;
120 int owner;
121 unsigned long state;
122 unsigned long state2; /* must be written under qdisc spinlock */
123 struct Qdisc *next_sched;
124 struct sk_buff_head skb_bad_txq;
125
126 spinlock_t busylock ____cacheline_aligned_in_smp;
127 spinlock_t seqlock;
128
129 struct rcu_head rcu;
130 netdevice_tracker dev_tracker;
131 /* private data */
132 long privdata[] ____cacheline_aligned;
133};
134
135static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
136{
137 if (qdisc->flags & TCQ_F_BUILTIN)
138 return;
139 refcount_inc(r: &qdisc->refcnt);
140}
141
142static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc)
143{
144 if (qdisc->flags & TCQ_F_BUILTIN)
145 return true;
146 return refcount_dec_if_one(r: &qdisc->refcnt);
147}
148
149/* Intended to be used by unlocked users, when concurrent qdisc release is
150 * possible.
151 */
152
153static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
154{
155 if (qdisc->flags & TCQ_F_BUILTIN)
156 return qdisc;
157 if (refcount_inc_not_zero(r: &qdisc->refcnt))
158 return qdisc;
159 return NULL;
160}
161
162/* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
163 * root_lock section, or provide their own memory barriers -- ordering
164 * against qdisc_run_begin/end() atomic bit operations.
165 */
166static inline bool qdisc_is_running(struct Qdisc *qdisc)
167{
168 if (qdisc->flags & TCQ_F_NOLOCK)
169 return spin_is_locked(lock: &qdisc->seqlock);
170 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
171}
172
173static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
174{
175 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
176}
177
178static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
179{
180 return q->flags & TCQ_F_CPUSTATS;
181}
182
183static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
184{
185 if (qdisc_is_percpu_stats(q: qdisc))
186 return nolock_qdisc_is_empty(qdisc);
187 return !READ_ONCE(qdisc->q.qlen);
188}
189
190/* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
191 * the qdisc root lock acquired.
192 */
193static inline bool qdisc_run_begin(struct Qdisc *qdisc)
194{
195 if (qdisc->flags & TCQ_F_NOLOCK) {
196 if (spin_trylock(lock: &qdisc->seqlock))
197 return true;
198
199 /* No need to insist if the MISSED flag was already set.
200 * Note that test_and_set_bit() also gives us memory ordering
201 * guarantees wrt potential earlier enqueue() and below
202 * spin_trylock(), both of which are necessary to prevent races
203 */
204 if (test_and_set_bit(nr: __QDISC_STATE_MISSED, addr: &qdisc->state))
205 return false;
206
207 /* Try to take the lock again to make sure that we will either
208 * grab it or the CPU that still has it will see MISSED set
209 * when testing it in qdisc_run_end()
210 */
211 return spin_trylock(lock: &qdisc->seqlock);
212 }
213 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
214}
215
216static inline void qdisc_run_end(struct Qdisc *qdisc)
217{
218 if (qdisc->flags & TCQ_F_NOLOCK) {
219 spin_unlock(lock: &qdisc->seqlock);
220
221 /* spin_unlock() only has store-release semantic. The unlock
222 * and test_bit() ordering is a store-load ordering, so a full
223 * memory barrier is needed here.
224 */
225 smp_mb();
226
227 if (unlikely(test_bit(__QDISC_STATE_MISSED,
228 &qdisc->state)))
229 __netif_schedule(q: qdisc);
230 } else {
231 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
232 }
233}
234
235static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
236{
237 return qdisc->flags & TCQ_F_ONETXQUEUE;
238}
239
240static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
241{
242 return netdev_queue_dql_avail(txq);
243}
244
245struct Qdisc_class_ops {
246 unsigned int flags;
247 /* Child qdisc manipulation */
248 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
249 int (*graft)(struct Qdisc *, unsigned long cl,
250 struct Qdisc *, struct Qdisc **,
251 struct netlink_ext_ack *extack);
252 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
253 void (*qlen_notify)(struct Qdisc *, unsigned long);
254
255 /* Class manipulation routines */
256 unsigned long (*find)(struct Qdisc *, u32 classid);
257 int (*change)(struct Qdisc *, u32, u32,
258 struct nlattr **, unsigned long *,
259 struct netlink_ext_ack *);
260 int (*delete)(struct Qdisc *, unsigned long,
261 struct netlink_ext_ack *);
262 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
263
264 /* Filter manipulation */
265 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
266 unsigned long arg,
267 struct netlink_ext_ack *extack);
268 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
269 u32 classid);
270 void (*unbind_tcf)(struct Qdisc *, unsigned long);
271
272 /* rtnetlink specific */
273 int (*dump)(struct Qdisc *, unsigned long,
274 struct sk_buff *skb, struct tcmsg*);
275 int (*dump_stats)(struct Qdisc *, unsigned long,
276 struct gnet_dump *);
277};
278
279/* Qdisc_class_ops flag values */
280
281/* Implements API that doesn't require rtnl lock */
282enum qdisc_class_ops_flags {
283 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
284};
285
286struct Qdisc_ops {
287 struct Qdisc_ops *next;
288 const struct Qdisc_class_ops *cl_ops;
289 char id[IFNAMSIZ];
290 int priv_size;
291 unsigned int static_flags;
292
293 int (*enqueue)(struct sk_buff *skb,
294 struct Qdisc *sch,
295 struct sk_buff **to_free);
296 struct sk_buff * (*dequeue)(struct Qdisc *);
297 struct sk_buff * (*peek)(struct Qdisc *);
298
299 int (*init)(struct Qdisc *sch, struct nlattr *arg,
300 struct netlink_ext_ack *extack);
301 void (*reset)(struct Qdisc *);
302 void (*destroy)(struct Qdisc *);
303 int (*change)(struct Qdisc *sch,
304 struct nlattr *arg,
305 struct netlink_ext_ack *extack);
306 void (*attach)(struct Qdisc *sch);
307 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
308 void (*change_real_num_tx)(struct Qdisc *sch,
309 unsigned int new_real_tx);
310
311 int (*dump)(struct Qdisc *, struct sk_buff *);
312 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
313
314 void (*ingress_block_set)(struct Qdisc *sch,
315 u32 block_index);
316 void (*egress_block_set)(struct Qdisc *sch,
317 u32 block_index);
318 u32 (*ingress_block_get)(struct Qdisc *sch);
319 u32 (*egress_block_get)(struct Qdisc *sch);
320
321 struct module *owner;
322};
323
324struct tcf_result {
325 union {
326 struct {
327 unsigned long class;
328 u32 classid;
329 };
330 const struct tcf_proto *goto_tp;
331 };
332};
333
334struct tcf_chain;
335
336struct tcf_proto_ops {
337 struct list_head head;
338 char kind[IFNAMSIZ];
339
340 int (*classify)(struct sk_buff *,
341 const struct tcf_proto *,
342 struct tcf_result *);
343 int (*init)(struct tcf_proto*);
344 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
345 struct netlink_ext_ack *extack);
346
347 void* (*get)(struct tcf_proto*, u32 handle);
348 void (*put)(struct tcf_proto *tp, void *f);
349 int (*change)(struct net *net, struct sk_buff *,
350 struct tcf_proto*, unsigned long,
351 u32 handle, struct nlattr **,
352 void **, u32,
353 struct netlink_ext_ack *);
354 int (*delete)(struct tcf_proto *tp, void *arg,
355 bool *last, bool rtnl_held,
356 struct netlink_ext_ack *);
357 bool (*delete_empty)(struct tcf_proto *tp);
358 void (*walk)(struct tcf_proto *tp,
359 struct tcf_walker *arg, bool rtnl_held);
360 int (*reoffload)(struct tcf_proto *tp, bool add,
361 flow_setup_cb_t *cb, void *cb_priv,
362 struct netlink_ext_ack *extack);
363 void (*hw_add)(struct tcf_proto *tp,
364 void *type_data);
365 void (*hw_del)(struct tcf_proto *tp,
366 void *type_data);
367 void (*bind_class)(void *, u32, unsigned long,
368 void *, unsigned long);
369 void * (*tmplt_create)(struct net *net,
370 struct tcf_chain *chain,
371 struct nlattr **tca,
372 struct netlink_ext_ack *extack);
373 void (*tmplt_destroy)(void *tmplt_priv);
374 void (*tmplt_reoffload)(struct tcf_chain *chain,
375 bool add,
376 flow_setup_cb_t *cb,
377 void *cb_priv);
378 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp,
379 u32 handle);
380
381 /* rtnetlink specific */
382 int (*dump)(struct net*, struct tcf_proto*, void *,
383 struct sk_buff *skb, struct tcmsg*,
384 bool);
385 int (*terse_dump)(struct net *net,
386 struct tcf_proto *tp, void *fh,
387 struct sk_buff *skb,
388 struct tcmsg *t, bool rtnl_held);
389 int (*tmplt_dump)(struct sk_buff *skb,
390 struct net *net,
391 void *tmplt_priv);
392
393 struct module *owner;
394 int flags;
395};
396
397/* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
398 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
399 * conditions can occur when filters are inserted/deleted simultaneously.
400 */
401enum tcf_proto_ops_flags {
402 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
403};
404
405struct tcf_proto {
406 /* Fast access part */
407 struct tcf_proto __rcu *next;
408 void __rcu *root;
409
410 /* called under RCU BH lock*/
411 int (*classify)(struct sk_buff *,
412 const struct tcf_proto *,
413 struct tcf_result *);
414 __be16 protocol;
415
416 /* All the rest */
417 u32 prio;
418 void *data;
419 const struct tcf_proto_ops *ops;
420 struct tcf_chain *chain;
421 /* Lock protects tcf_proto shared state and can be used by unlocked
422 * classifiers to protect their private data.
423 */
424 spinlock_t lock;
425 bool deleting;
426 refcount_t refcnt;
427 struct rcu_head rcu;
428 struct hlist_node destroy_ht_node;
429};
430
431struct qdisc_skb_cb {
432 struct {
433 unsigned int pkt_len;
434 u16 slave_dev_queue_mapping;
435 u16 tc_classid;
436 };
437#define QDISC_CB_PRIV_LEN 20
438 unsigned char data[QDISC_CB_PRIV_LEN];
439};
440
441typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
442
443struct tcf_chain {
444 /* Protects filter_chain. */
445 struct mutex filter_chain_lock;
446 struct tcf_proto __rcu *filter_chain;
447 struct list_head list;
448 struct tcf_block *block;
449 u32 index; /* chain index */
450 unsigned int refcnt;
451 unsigned int action_refcnt;
452 bool explicitly_created;
453 bool flushing;
454 const struct tcf_proto_ops *tmplt_ops;
455 void *tmplt_priv;
456 struct rcu_head rcu;
457};
458
459struct tcf_block {
460 struct xarray ports; /* datapath accessible */
461 /* Lock protects tcf_block and lifetime-management data of chains
462 * attached to the block (refcnt, action_refcnt, explicitly_created).
463 */
464 struct mutex lock;
465 struct list_head chain_list;
466 u32 index; /* block index for shared blocks */
467 u32 classid; /* which class this block belongs to */
468 refcount_t refcnt;
469 struct net *net;
470 struct Qdisc *q;
471 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
472 struct flow_block flow_block;
473 struct list_head owner_list;
474 bool keep_dst;
475 atomic_t offloadcnt; /* Number of oddloaded filters */
476 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
477 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
478 struct {
479 struct tcf_chain *chain;
480 struct list_head filter_chain_list;
481 } chain0;
482 struct rcu_head rcu;
483 DECLARE_HASHTABLE(proto_destroy_ht, 7);
484 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
485};
486
487struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index);
488
489static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
490{
491 return lockdep_is_held(&chain->filter_chain_lock);
492}
493
494static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
495{
496 return lockdep_is_held(&tp->lock);
497}
498
499#define tcf_chain_dereference(p, chain) \
500 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
501
502#define tcf_proto_dereference(p, tp) \
503 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
504
505static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
506{
507 struct qdisc_skb_cb *qcb;
508
509 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
510 BUILD_BUG_ON(sizeof(qcb->data) < sz);
511}
512
513static inline int qdisc_qlen(const struct Qdisc *q)
514{
515 return q->q.qlen;
516}
517
518static inline int qdisc_qlen_sum(const struct Qdisc *q)
519{
520 __u32 qlen = q->qstats.qlen;
521 int i;
522
523 if (qdisc_is_percpu_stats(q)) {
524 for_each_possible_cpu(i)
525 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
526 } else {
527 qlen += q->q.qlen;
528 }
529
530 return qlen;
531}
532
533static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
534{
535 return (struct qdisc_skb_cb *)skb->cb;
536}
537
538static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
539{
540 return &qdisc->q.lock;
541}
542
543static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
544{
545 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
546
547 return q;
548}
549
550static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
551{
552 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
553}
554
555static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
556{
557 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
558}
559
560static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
561{
562 struct Qdisc *root = qdisc_root_sleeping(qdisc);
563
564 ASSERT_RTNL();
565 return qdisc_lock(qdisc: root);
566}
567
568static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
569{
570 return qdisc->dev_queue->dev;
571}
572
573static inline void sch_tree_lock(struct Qdisc *q)
574{
575 if (q->flags & TCQ_F_MQROOT)
576 spin_lock_bh(lock: qdisc_lock(qdisc: q));
577 else
578 spin_lock_bh(lock: qdisc_root_sleeping_lock(qdisc: q));
579}
580
581static inline void sch_tree_unlock(struct Qdisc *q)
582{
583 if (q->flags & TCQ_F_MQROOT)
584 spin_unlock_bh(lock: qdisc_lock(qdisc: q));
585 else
586 spin_unlock_bh(lock: qdisc_root_sleeping_lock(qdisc: q));
587}
588
589extern struct Qdisc noop_qdisc;
590extern struct Qdisc_ops noop_qdisc_ops;
591extern struct Qdisc_ops pfifo_fast_ops;
592extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
593extern struct Qdisc_ops mq_qdisc_ops;
594extern struct Qdisc_ops noqueue_qdisc_ops;
595extern const struct Qdisc_ops *default_qdisc_ops;
596static inline const struct Qdisc_ops *
597get_default_qdisc_ops(const struct net_device *dev, int ntx)
598{
599 return ntx < dev->real_num_tx_queues ?
600 default_qdisc_ops : &pfifo_fast_ops;
601}
602
603struct Qdisc_class_common {
604 u32 classid;
605 unsigned int filter_cnt;
606 struct hlist_node hnode;
607};
608
609struct Qdisc_class_hash {
610 struct hlist_head *hash;
611 unsigned int hashsize;
612 unsigned int hashmask;
613 unsigned int hashelems;
614};
615
616static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
617{
618 id ^= id >> 8;
619 id ^= id >> 4;
620 return id & mask;
621}
622
623static inline struct Qdisc_class_common *
624qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
625{
626 struct Qdisc_class_common *cl;
627 unsigned int h;
628
629 if (!id)
630 return NULL;
631
632 h = qdisc_class_hash(id, mask: hash->hashmask);
633 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
634 if (cl->classid == id)
635 return cl;
636 }
637 return NULL;
638}
639
640static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
641{
642 return cl->filter_cnt > 0;
643}
644
645static inline void qdisc_class_get(struct Qdisc_class_common *cl)
646{
647 unsigned int res;
648
649 if (check_add_overflow(cl->filter_cnt, 1, &res))
650 WARN(1, "Qdisc class overflow");
651
652 cl->filter_cnt = res;
653}
654
655static inline void qdisc_class_put(struct Qdisc_class_common *cl)
656{
657 unsigned int res;
658
659 if (check_sub_overflow(cl->filter_cnt, 1, &res))
660 WARN(1, "Qdisc class underflow");
661
662 cl->filter_cnt = res;
663}
664
665static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
666{
667 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
668
669 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
670}
671
672int qdisc_class_hash_init(struct Qdisc_class_hash *);
673void qdisc_class_hash_insert(struct Qdisc_class_hash *,
674 struct Qdisc_class_common *);
675void qdisc_class_hash_remove(struct Qdisc_class_hash *,
676 struct Qdisc_class_common *);
677void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
678void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
679
680int dev_qdisc_change_tx_queue_len(struct net_device *dev);
681void dev_qdisc_change_real_num_tx(struct net_device *dev,
682 unsigned int new_real_tx);
683void dev_init_scheduler(struct net_device *dev);
684void dev_shutdown(struct net_device *dev);
685void dev_activate(struct net_device *dev);
686void dev_deactivate(struct net_device *dev);
687void dev_deactivate_many(struct list_head *head);
688struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
689 struct Qdisc *qdisc);
690void qdisc_reset(struct Qdisc *qdisc);
691void qdisc_destroy(struct Qdisc *qdisc);
692void qdisc_put(struct Qdisc *qdisc);
693void qdisc_put_unlocked(struct Qdisc *qdisc);
694void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
695#ifdef CONFIG_NET_SCHED
696int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
697 void *type_data);
698void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
699 struct Qdisc *new, struct Qdisc *old,
700 enum tc_setup_type type, void *type_data,
701 struct netlink_ext_ack *extack);
702#else
703static inline int
704qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
705 void *type_data)
706{
707 q->flags &= ~TCQ_F_OFFLOADED;
708 return 0;
709}
710
711static inline void
712qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
713 struct Qdisc *new, struct Qdisc *old,
714 enum tc_setup_type type, void *type_data,
715 struct netlink_ext_ack *extack)
716{
717}
718#endif
719void qdisc_offload_query_caps(struct net_device *dev,
720 enum tc_setup_type type,
721 void *caps, size_t caps_len);
722struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
723 const struct Qdisc_ops *ops,
724 struct netlink_ext_ack *extack);
725void qdisc_free(struct Qdisc *qdisc);
726struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
727 const struct Qdisc_ops *ops, u32 parentid,
728 struct netlink_ext_ack *extack);
729void __qdisc_calculate_pkt_len(struct sk_buff *skb,
730 const struct qdisc_size_table *stab);
731int skb_do_redirect(struct sk_buff *);
732
733static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
734{
735#ifdef CONFIG_NET_XGRESS
736 return skb->tc_at_ingress;
737#else
738 return false;
739#endif
740}
741
742static inline bool skb_skip_tc_classify(struct sk_buff *skb)
743{
744#ifdef CONFIG_NET_CLS_ACT
745 if (skb->tc_skip_classify) {
746 skb->tc_skip_classify = 0;
747 return true;
748 }
749#endif
750 return false;
751}
752
753/* Reset all TX qdiscs greater than index of a device. */
754static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
755{
756 struct Qdisc *qdisc;
757
758 for (; i < dev->num_tx_queues; i++) {
759 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
760 if (qdisc) {
761 spin_lock_bh(lock: qdisc_lock(qdisc));
762 qdisc_reset(qdisc);
763 spin_unlock_bh(lock: qdisc_lock(qdisc));
764 }
765 }
766}
767
768/* Are all TX queues of the device empty? */
769static inline bool qdisc_all_tx_empty(const struct net_device *dev)
770{
771 unsigned int i;
772
773 rcu_read_lock();
774 for (i = 0; i < dev->num_tx_queues; i++) {
775 struct netdev_queue *txq = netdev_get_tx_queue(dev, index: i);
776 const struct Qdisc *q = rcu_dereference(txq->qdisc);
777
778 if (!qdisc_is_empty(qdisc: q)) {
779 rcu_read_unlock();
780 return false;
781 }
782 }
783 rcu_read_unlock();
784 return true;
785}
786
787/* Are any of the TX qdiscs changing? */
788static inline bool qdisc_tx_changing(const struct net_device *dev)
789{
790 unsigned int i;
791
792 for (i = 0; i < dev->num_tx_queues; i++) {
793 struct netdev_queue *txq = netdev_get_tx_queue(dev, index: i);
794
795 if (rcu_access_pointer(txq->qdisc) !=
796 rcu_access_pointer(txq->qdisc_sleeping))
797 return true;
798 }
799 return false;
800}
801
802/* Is the device using the noop qdisc on all queues? */
803static inline bool qdisc_tx_is_noop(const struct net_device *dev)
804{
805 unsigned int i;
806
807 for (i = 0; i < dev->num_tx_queues; i++) {
808 struct netdev_queue *txq = netdev_get_tx_queue(dev, index: i);
809 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
810 return false;
811 }
812 return true;
813}
814
815static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
816{
817 return qdisc_skb_cb(skb)->pkt_len;
818}
819
820/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
821enum net_xmit_qdisc_t {
822 __NET_XMIT_STOLEN = 0x00010000,
823 __NET_XMIT_BYPASS = 0x00020000,
824};
825
826#ifdef CONFIG_NET_CLS_ACT
827#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
828#else
829#define net_xmit_drop_count(e) (1)
830#endif
831
832static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
833 const struct Qdisc *sch)
834{
835#ifdef CONFIG_NET_SCHED
836 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
837
838 if (stab)
839 __qdisc_calculate_pkt_len(skb, stab);
840#endif
841}
842
843static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
844 struct sk_buff **to_free)
845{
846 qdisc_calculate_pkt_len(skb, sch);
847 return sch->enqueue(skb, sch, to_free);
848}
849
850static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
851 __u64 bytes, __u32 packets)
852{
853 u64_stats_update_begin(syncp: &bstats->syncp);
854 u64_stats_add(p: &bstats->bytes, val: bytes);
855 u64_stats_add(p: &bstats->packets, val: packets);
856 u64_stats_update_end(syncp: &bstats->syncp);
857}
858
859static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
860 const struct sk_buff *skb)
861{
862 _bstats_update(bstats,
863 bytes: qdisc_pkt_len(skb),
864 packets: skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
865}
866
867static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
868 const struct sk_buff *skb)
869{
870 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
871}
872
873static inline void qdisc_bstats_update(struct Qdisc *sch,
874 const struct sk_buff *skb)
875{
876 bstats_update(bstats: &sch->bstats, skb);
877}
878
879static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
880 const struct sk_buff *skb)
881{
882 sch->qstats.backlog -= qdisc_pkt_len(skb);
883}
884
885static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
886 const struct sk_buff *skb)
887{
888 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
889}
890
891static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
892 const struct sk_buff *skb)
893{
894 sch->qstats.backlog += qdisc_pkt_len(skb);
895}
896
897static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
898 const struct sk_buff *skb)
899{
900 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
901}
902
903static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
904{
905 this_cpu_inc(sch->cpu_qstats->qlen);
906}
907
908static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
909{
910 this_cpu_dec(sch->cpu_qstats->qlen);
911}
912
913static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
914{
915 this_cpu_inc(sch->cpu_qstats->requeues);
916}
917
918static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
919{
920 sch->qstats.drops += count;
921}
922
923static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
924{
925 qstats->drops++;
926}
927
928static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
929{
930 qstats->overlimits++;
931}
932
933static inline void qdisc_qstats_drop(struct Qdisc *sch)
934{
935 qstats_drop_inc(qstats: &sch->qstats);
936}
937
938static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
939{
940 this_cpu_inc(sch->cpu_qstats->drops);
941}
942
943static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
944{
945 sch->qstats.overlimits++;
946}
947
948static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
949{
950 __u32 qlen = qdisc_qlen_sum(q: sch);
951
952 return gnet_stats_copy_queue(d, cpu_q: sch->cpu_qstats, q: &sch->qstats, qlen);
953}
954
955static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
956 __u32 *backlog)
957{
958 struct gnet_stats_queue qstats = { 0 };
959
960 gnet_stats_add_queue(qstats: &qstats, cpu_q: sch->cpu_qstats, q: &sch->qstats);
961 *qlen = qstats.qlen + qdisc_qlen(q: sch);
962 *backlog = qstats.backlog;
963}
964
965static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
966{
967 __u32 qlen, backlog;
968
969 qdisc_qstats_qlen_backlog(sch, qlen: &qlen, backlog: &backlog);
970 qdisc_tree_reduce_backlog(qdisc: sch, n: qlen, len: backlog);
971}
972
973static inline void qdisc_purge_queue(struct Qdisc *sch)
974{
975 __u32 qlen, backlog;
976
977 qdisc_qstats_qlen_backlog(sch, qlen: &qlen, backlog: &backlog);
978 qdisc_reset(qdisc: sch);
979 qdisc_tree_reduce_backlog(qdisc: sch, n: qlen, len: backlog);
980}
981
982static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
983 struct qdisc_skb_head *qh)
984{
985 struct sk_buff *last = qh->tail;
986
987 if (last) {
988 skb->next = NULL;
989 last->next = skb;
990 qh->tail = skb;
991 } else {
992 qh->tail = skb;
993 qh->head = skb;
994 }
995 qh->qlen++;
996}
997
998static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
999{
1000 __qdisc_enqueue_tail(skb, qh: &sch->q);
1001 qdisc_qstats_backlog_inc(sch, skb);
1002 return NET_XMIT_SUCCESS;
1003}
1004
1005static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1006 struct qdisc_skb_head *qh)
1007{
1008 skb->next = qh->head;
1009
1010 if (!qh->head)
1011 qh->tail = skb;
1012 qh->head = skb;
1013 qh->qlen++;
1014}
1015
1016static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1017{
1018 struct sk_buff *skb = qh->head;
1019
1020 if (likely(skb != NULL)) {
1021 qh->head = skb->next;
1022 qh->qlen--;
1023 if (qh->head == NULL)
1024 qh->tail = NULL;
1025 skb->next = NULL;
1026 }
1027
1028 return skb;
1029}
1030
1031static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1032{
1033 struct sk_buff *skb = __qdisc_dequeue_head(qh: &sch->q);
1034
1035 if (likely(skb != NULL)) {
1036 qdisc_qstats_backlog_dec(sch, skb);
1037 qdisc_bstats_update(sch, skb);
1038 }
1039
1040 return skb;
1041}
1042
1043struct tc_skb_cb {
1044 struct qdisc_skb_cb qdisc_cb;
1045 u32 drop_reason;
1046
1047 u16 zone; /* Only valid if post_ct = true */
1048 u16 mru;
1049 u8 post_ct:1;
1050 u8 post_ct_snat:1;
1051 u8 post_ct_dnat:1;
1052};
1053
1054static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
1055{
1056 struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
1057
1058 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
1059 return cb;
1060}
1061
1062static inline enum skb_drop_reason
1063tcf_get_drop_reason(const struct sk_buff *skb)
1064{
1065 return tc_skb_cb(skb)->drop_reason;
1066}
1067
1068static inline void tcf_set_drop_reason(const struct sk_buff *skb,
1069 enum skb_drop_reason reason)
1070{
1071 tc_skb_cb(skb)->drop_reason = reason;
1072}
1073
1074/* Instead of calling kfree_skb() while root qdisc lock is held,
1075 * queue the skb for future freeing at end of __dev_xmit_skb()
1076 */
1077static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1078{
1079 skb->next = *to_free;
1080 *to_free = skb;
1081}
1082
1083static inline void __qdisc_drop_all(struct sk_buff *skb,
1084 struct sk_buff **to_free)
1085{
1086 if (skb->prev)
1087 skb->prev->next = *to_free;
1088 else
1089 skb->next = *to_free;
1090 *to_free = skb;
1091}
1092
1093static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1094 struct qdisc_skb_head *qh,
1095 struct sk_buff **to_free)
1096{
1097 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1098
1099 if (likely(skb != NULL)) {
1100 unsigned int len = qdisc_pkt_len(skb);
1101
1102 qdisc_qstats_backlog_dec(sch, skb);
1103 __qdisc_drop(skb, to_free);
1104 return len;
1105 }
1106
1107 return 0;
1108}
1109
1110static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1111{
1112 const struct qdisc_skb_head *qh = &sch->q;
1113
1114 return qh->head;
1115}
1116
1117/* generic pseudo peek method for non-work-conserving qdisc */
1118static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1119{
1120 struct sk_buff *skb = skb_peek(list_: &sch->gso_skb);
1121
1122 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1123 if (!skb) {
1124 skb = sch->dequeue(sch);
1125
1126 if (skb) {
1127 __skb_queue_head(list: &sch->gso_skb, newsk: skb);
1128 /* it's still part of the queue */
1129 qdisc_qstats_backlog_inc(sch, skb);
1130 sch->q.qlen++;
1131 }
1132 }
1133
1134 return skb;
1135}
1136
1137static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1138 struct sk_buff *skb)
1139{
1140 if (qdisc_is_percpu_stats(q: sch)) {
1141 qdisc_qstats_cpu_backlog_dec(sch, skb);
1142 qdisc_bstats_cpu_update(sch, skb);
1143 qdisc_qstats_cpu_qlen_dec(sch);
1144 } else {
1145 qdisc_qstats_backlog_dec(sch, skb);
1146 qdisc_bstats_update(sch, skb);
1147 sch->q.qlen--;
1148 }
1149}
1150
1151static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1152 unsigned int pkt_len)
1153{
1154 if (qdisc_is_percpu_stats(q: sch)) {
1155 qdisc_qstats_cpu_qlen_inc(sch);
1156 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1157 } else {
1158 sch->qstats.backlog += pkt_len;
1159 sch->q.qlen++;
1160 }
1161}
1162
1163/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1164static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1165{
1166 struct sk_buff *skb = skb_peek(list_: &sch->gso_skb);
1167
1168 if (skb) {
1169 skb = __skb_dequeue(list: &sch->gso_skb);
1170 if (qdisc_is_percpu_stats(q: sch)) {
1171 qdisc_qstats_cpu_backlog_dec(sch, skb);
1172 qdisc_qstats_cpu_qlen_dec(sch);
1173 } else {
1174 qdisc_qstats_backlog_dec(sch, skb);
1175 sch->q.qlen--;
1176 }
1177 } else {
1178 skb = sch->dequeue(sch);
1179 }
1180
1181 return skb;
1182}
1183
1184static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1185{
1186 /*
1187 * We do not know the backlog in bytes of this list, it
1188 * is up to the caller to correct it
1189 */
1190 ASSERT_RTNL();
1191 if (qh->qlen) {
1192 rtnl_kfree_skbs(head: qh->head, tail: qh->tail);
1193
1194 qh->head = NULL;
1195 qh->tail = NULL;
1196 qh->qlen = 0;
1197 }
1198}
1199
1200static inline void qdisc_reset_queue(struct Qdisc *sch)
1201{
1202 __qdisc_reset_queue(qh: &sch->q);
1203}
1204
1205static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1206 struct Qdisc **pold)
1207{
1208 struct Qdisc *old;
1209
1210 sch_tree_lock(q: sch);
1211 old = *pold;
1212 *pold = new;
1213 if (old != NULL)
1214 qdisc_purge_queue(sch: old);
1215 sch_tree_unlock(q: sch);
1216
1217 return old;
1218}
1219
1220static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1221{
1222 rtnl_kfree_skbs(head: skb, tail: skb);
1223 qdisc_qstats_drop(sch);
1224}
1225
1226static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1227 struct sk_buff **to_free)
1228{
1229 __qdisc_drop(skb, to_free);
1230 qdisc_qstats_cpu_drop(sch);
1231
1232 return NET_XMIT_DROP;
1233}
1234
1235static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1236 struct sk_buff **to_free)
1237{
1238 __qdisc_drop(skb, to_free);
1239 qdisc_qstats_drop(sch);
1240
1241 return NET_XMIT_DROP;
1242}
1243
1244static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1245 struct sk_buff **to_free)
1246{
1247 __qdisc_drop_all(skb, to_free);
1248 qdisc_qstats_drop(sch);
1249
1250 return NET_XMIT_DROP;
1251}
1252
1253struct psched_ratecfg {
1254 u64 rate_bytes_ps; /* bytes per second */
1255 u32 mult;
1256 u16 overhead;
1257 u16 mpu;
1258 u8 linklayer;
1259 u8 shift;
1260};
1261
1262static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1263 unsigned int len)
1264{
1265 len += r->overhead;
1266
1267 if (len < r->mpu)
1268 len = r->mpu;
1269
1270 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1271 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1272
1273 return ((u64)len * r->mult) >> r->shift;
1274}
1275
1276void psched_ratecfg_precompute(struct psched_ratecfg *r,
1277 const struct tc_ratespec *conf,
1278 u64 rate64);
1279
1280static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1281 const struct psched_ratecfg *r)
1282{
1283 memset(res, 0, sizeof(*res));
1284
1285 /* legacy struct tc_ratespec has a 32bit @rate field
1286 * Qdisc using 64bit rate should add new attributes
1287 * in order to maintain compatibility.
1288 */
1289 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1290
1291 res->overhead = r->overhead;
1292 res->mpu = r->mpu;
1293 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1294}
1295
1296struct psched_pktrate {
1297 u64 rate_pkts_ps; /* packets per second */
1298 u32 mult;
1299 u8 shift;
1300};
1301
1302static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1303 unsigned int pkt_num)
1304{
1305 return ((u64)pkt_num * r->mult) >> r->shift;
1306}
1307
1308void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1309
1310/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1311 * The fast path only needs to access filter list and to update stats
1312 */
1313struct mini_Qdisc {
1314 struct tcf_proto *filter_list;
1315 struct tcf_block *block;
1316 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1317 struct gnet_stats_queue __percpu *cpu_qstats;
1318 unsigned long rcu_state;
1319};
1320
1321static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1322 const struct sk_buff *skb)
1323{
1324 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1325}
1326
1327static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1328{
1329 this_cpu_inc(miniq->cpu_qstats->drops);
1330}
1331
1332struct mini_Qdisc_pair {
1333 struct mini_Qdisc miniq1;
1334 struct mini_Qdisc miniq2;
1335 struct mini_Qdisc __rcu **p_miniq;
1336};
1337
1338void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1339 struct tcf_proto *tp_head);
1340void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1341 struct mini_Qdisc __rcu **p_miniq);
1342void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1343 struct tcf_block *block);
1344
1345void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1346
1347int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1348
1349/* Make sure qdisc is no longer in SCHED state. */
1350static inline void qdisc_synchronize(const struct Qdisc *q)
1351{
1352 while (test_bit(__QDISC_STATE_SCHED, &q->state))
1353 msleep(msecs: 1);
1354}
1355
1356#endif
1357

source code of linux/include/net/sch_generic.h