Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef __NET_SCHED_GENERIC_H
3#define __NET_SCHED_GENERIC_H
4
5#include <linux/netdevice.h>
6#include <linux/types.h>
7#include <linux/rcupdate.h>
8#include <linux/pkt_sched.h>
9#include <linux/pkt_cls.h>
10#include <linux/percpu.h>
11#include <linux/dynamic_queue_limits.h>
12#include <linux/list.h>
13#include <linux/refcount.h>
14#include <linux/workqueue.h>
15#include <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 struct lock_class_key root_lock_key;
132 /* private data */
133 long privdata[] ____cacheline_aligned;
134};
135
136static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
137{
138 if (qdisc->flags & TCQ_F_BUILTIN)
139 return;
140 refcount_inc(&qdisc->refcnt);
141}
142
143static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc)
144{
145 if (qdisc->flags & TCQ_F_BUILTIN)
146 return true;
147 return refcount_dec_if_one(&qdisc->refcnt);
148}
149
150/* Intended to be used by unlocked users, when concurrent qdisc release is
151 * possible.
152 */
153
154static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
155{
156 if (qdisc->flags & TCQ_F_BUILTIN)
157 return qdisc;
158 if (refcount_inc_not_zero(&qdisc->refcnt))
159 return qdisc;
160 return NULL;
161}
162
163/* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
164 * root_lock section, or provide their own memory barriers -- ordering
165 * against qdisc_run_begin/end() atomic bit operations.
166 */
167static inline bool qdisc_is_running(struct Qdisc *qdisc)
168{
169 if (qdisc->flags & TCQ_F_NOLOCK)
170 return spin_is_locked(&qdisc->seqlock);
171 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
172}
173
174static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
175{
176 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
177}
178
179static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
180{
181 return q->flags & TCQ_F_CPUSTATS;
182}
183
184static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
185{
186 if (qdisc_is_percpu_stats(qdisc))
187 return nolock_qdisc_is_empty(qdisc);
188 return !READ_ONCE(qdisc->q.qlen);
189}
190
191/* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
192 * the qdisc root lock acquired.
193 */
194static inline bool qdisc_run_begin(struct Qdisc *qdisc)
195{
196 if (qdisc->flags & TCQ_F_NOLOCK) {
197 if (spin_trylock(&qdisc->seqlock))
198 return true;
199
200 /* No need to insist if the MISSED flag was already set.
201 * Note that test_and_set_bit() also gives us memory ordering
202 * guarantees wrt potential earlier enqueue() and below
203 * spin_trylock(), both of which are necessary to prevent races
204 */
205 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
206 return false;
207
208 /* Try to take the lock again to make sure that we will either
209 * grab it or the CPU that still has it will see MISSED set
210 * when testing it in qdisc_run_end()
211 */
212 return spin_trylock(&qdisc->seqlock);
213 }
214 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
215}
216
217static inline void qdisc_run_end(struct Qdisc *qdisc)
218{
219 if (qdisc->flags & TCQ_F_NOLOCK) {
220 spin_unlock(&qdisc->seqlock);
221
222 /* spin_unlock() only has store-release semantic. The unlock
223 * and test_bit() ordering is a store-load ordering, so a full
224 * memory barrier is needed here.
225 */
226 smp_mb();
227
228 if (unlikely(test_bit(__QDISC_STATE_MISSED,
229 &qdisc->state)))
230 __netif_schedule(qdisc);
231 } else {
232 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
233 }
234}
235
236static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
237{
238 return qdisc->flags & TCQ_F_ONETXQUEUE;
239}
240
241static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
242{
243 return netdev_queue_dql_avail(txq);
244}
245
246struct Qdisc_class_ops {
247 unsigned int flags;
248 /* Child qdisc manipulation */
249 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
250 int (*graft)(struct Qdisc *, unsigned long cl,
251 struct Qdisc *, struct Qdisc **,
252 struct netlink_ext_ack *extack);
253 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
254 void (*qlen_notify)(struct Qdisc *, unsigned long);
255
256 /* Class manipulation routines */
257 unsigned long (*find)(struct Qdisc *, u32 classid);
258 int (*change)(struct Qdisc *, u32, u32,
259 struct nlattr **, unsigned long *,
260 struct netlink_ext_ack *);
261 int (*delete)(struct Qdisc *, unsigned long,
262 struct netlink_ext_ack *);
263 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
264
265 /* Filter manipulation */
266 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
267 unsigned long arg,
268 struct netlink_ext_ack *extack);
269 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
270 u32 classid);
271 void (*unbind_tcf)(struct Qdisc *, unsigned long);
272
273 /* rtnetlink specific */
274 int (*dump)(struct Qdisc *, unsigned long,
275 struct sk_buff *skb, struct tcmsg*);
276 int (*dump_stats)(struct Qdisc *, unsigned long,
277 struct gnet_dump *);
278};
279
280/* Qdisc_class_ops flag values */
281
282/* Implements API that doesn't require rtnl lock */
283enum qdisc_class_ops_flags {
284 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
285};
286
287struct Qdisc_ops {
288 struct Qdisc_ops *next;
289 const struct Qdisc_class_ops *cl_ops;
290 char id[IFNAMSIZ];
291 int priv_size;
292 unsigned int static_flags;
293
294 int (*enqueue)(struct sk_buff *skb,
295 struct Qdisc *sch,
296 struct sk_buff **to_free);
297 struct sk_buff * (*dequeue)(struct Qdisc *);
298 struct sk_buff * (*peek)(struct Qdisc *);
299
300 int (*init)(struct Qdisc *sch, struct nlattr *arg,
301 struct netlink_ext_ack *extack);
302 void (*reset)(struct Qdisc *);
303 void (*destroy)(struct Qdisc *);
304 int (*change)(struct Qdisc *sch,
305 struct nlattr *arg,
306 struct netlink_ext_ack *extack);
307 void (*attach)(struct Qdisc *sch);
308 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
309 void (*change_real_num_tx)(struct Qdisc *sch,
310 unsigned int new_real_tx);
311
312 int (*dump)(struct Qdisc *, struct sk_buff *);
313 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
314
315 void (*ingress_block_set)(struct Qdisc *sch,
316 u32 block_index);
317 void (*egress_block_set)(struct Qdisc *sch,
318 u32 block_index);
319 u32 (*ingress_block_get)(struct Qdisc *sch);
320 u32 (*egress_block_get)(struct Qdisc *sch);
321
322 struct module *owner;
323};
324
325struct tcf_result {
326 union {
327 struct {
328 unsigned long class;
329 u32 classid;
330 };
331 const struct tcf_proto *goto_tp;
332 };
333};
334
335struct tcf_chain;
336
337struct tcf_proto_ops {
338 struct list_head head;
339 char kind[IFNAMSIZ];
340
341 int (*classify)(struct sk_buff *,
342 const struct tcf_proto *,
343 struct tcf_result *);
344 int (*init)(struct tcf_proto*);
345 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
346 struct netlink_ext_ack *extack);
347
348 void* (*get)(struct tcf_proto*, u32 handle);
349 void (*put)(struct tcf_proto *tp, void *f);
350 int (*change)(struct net *net, struct sk_buff *,
351 struct tcf_proto*, unsigned long,
352 u32 handle, struct nlattr **,
353 void **, u32,
354 struct netlink_ext_ack *);
355 int (*delete)(struct tcf_proto *tp, void *arg,
356 bool *last, bool rtnl_held,
357 struct netlink_ext_ack *);
358 bool (*delete_empty)(struct tcf_proto *tp);
359 void (*walk)(struct tcf_proto *tp,
360 struct tcf_walker *arg, bool rtnl_held);
361 int (*reoffload)(struct tcf_proto *tp, bool add,
362 flow_setup_cb_t *cb, void *cb_priv,
363 struct netlink_ext_ack *extack);
364 void (*hw_add)(struct tcf_proto *tp,
365 void *type_data);
366 void (*hw_del)(struct tcf_proto *tp,
367 void *type_data);
368 void (*bind_class)(void *, u32, unsigned long,
369 void *, unsigned long);
370 void * (*tmplt_create)(struct net *net,
371 struct tcf_chain *chain,
372 struct nlattr **tca,
373 struct netlink_ext_ack *extack);
374 void (*tmplt_destroy)(void *tmplt_priv);
375 void (*tmplt_reoffload)(struct tcf_chain *chain,
376 bool add,
377 flow_setup_cb_t *cb,
378 void *cb_priv);
379 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp,
380 u32 handle);
381
382 /* rtnetlink specific */
383 int (*dump)(struct net*, struct tcf_proto*, void *,
384 struct sk_buff *skb, struct tcmsg*,
385 bool);
386 int (*terse_dump)(struct net *net,
387 struct tcf_proto *tp, void *fh,
388 struct sk_buff *skb,
389 struct tcmsg *t, bool rtnl_held);
390 int (*tmplt_dump)(struct sk_buff *skb,
391 struct net *net,
392 void *tmplt_priv);
393
394 struct module *owner;
395 int flags;
396};
397
398/* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
399 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
400 * conditions can occur when filters are inserted/deleted simultaneously.
401 */
402enum tcf_proto_ops_flags {
403 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
404};
405
406struct tcf_proto {
407 /* Fast access part */
408 struct tcf_proto __rcu *next;
409 void __rcu *root;
410
411 /* called under RCU BH lock*/
412 int (*classify)(struct sk_buff *,
413 const struct tcf_proto *,
414 struct tcf_result *);
415 __be16 protocol;
416
417 /* All the rest */
418 u32 prio;
419 void *data;
420 const struct tcf_proto_ops *ops;
421 struct tcf_chain *chain;
422 /* Lock protects tcf_proto shared state and can be used by unlocked
423 * classifiers to protect their private data.
424 */
425 spinlock_t lock;
426 bool deleting;
427 bool counted;
428 bool usesw;
429 refcount_t refcnt;
430 struct rcu_head rcu;
431 struct hlist_node destroy_ht_node;
432};
433
434struct qdisc_skb_cb {
435 struct {
436 unsigned int pkt_len;
437 u16 slave_dev_queue_mapping;
438 u16 tc_classid;
439 };
440#define QDISC_CB_PRIV_LEN 20
441 unsigned char data[QDISC_CB_PRIV_LEN];
442};
443
444typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
445
446struct tcf_chain {
447 /* Protects filter_chain. */
448 struct mutex filter_chain_lock;
449 struct tcf_proto __rcu *filter_chain;
450 struct list_head list;
451 struct tcf_block *block;
452 u32 index; /* chain index */
453 unsigned int refcnt;
454 unsigned int action_refcnt;
455 bool explicitly_created;
456 bool flushing;
457 const struct tcf_proto_ops *tmplt_ops;
458 void *tmplt_priv;
459 struct rcu_head rcu;
460};
461
462struct tcf_block {
463 struct xarray ports; /* datapath accessible */
464 /* Lock protects tcf_block and lifetime-management data of chains
465 * attached to the block (refcnt, action_refcnt, explicitly_created).
466 */
467 struct mutex lock;
468 struct list_head chain_list;
469 u32 index; /* block index for shared blocks */
470 u32 classid; /* which class this block belongs to */
471 refcount_t refcnt;
472 struct net *net;
473 struct Qdisc *q;
474 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
475 struct flow_block flow_block;
476 struct list_head owner_list;
477 bool keep_dst;
478 atomic_t useswcnt;
479 atomic_t offloadcnt; /* Number of oddloaded filters */
480 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
481 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
482 struct {
483 struct tcf_chain *chain;
484 struct list_head filter_chain_list;
485 } chain0;
486 struct rcu_head rcu;
487 DECLARE_HASHTABLE(proto_destroy_ht, 7);
488 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
489};
490
491struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index);
492
493static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
494{
495 return lockdep_is_held(&chain->filter_chain_lock);
496}
497
498static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
499{
500 return lockdep_is_held(&tp->lock);
501}
502
503#define tcf_chain_dereference(p, chain) \
504 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
505
506#define tcf_proto_dereference(p, tp) \
507 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
508
509static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
510{
511 struct qdisc_skb_cb *qcb;
512
513 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
514 BUILD_BUG_ON(sizeof(qcb->data) < sz);
515}
516
517static inline int qdisc_qlen(const struct Qdisc *q)
518{
519 return q->q.qlen;
520}
521
522static inline int qdisc_qlen_sum(const struct Qdisc *q)
523{
524 __u32 qlen = q->qstats.qlen;
525 int i;
526
527 if (qdisc_is_percpu_stats(q)) {
528 for_each_possible_cpu(i)
529 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
530 } else {
531 qlen += q->q.qlen;
532 }
533
534 return qlen;
535}
536
537static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
538{
539 return (struct qdisc_skb_cb *)skb->cb;
540}
541
542static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
543{
544 return &qdisc->q.lock;
545}
546
547static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
548{
549 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
550
551 return q;
552}
553
554static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
555{
556 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
557}
558
559static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
560{
561 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
562}
563
564static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
565{
566 struct Qdisc *root = qdisc_root_sleeping(qdisc);
567
568 ASSERT_RTNL();
569 return qdisc_lock(root);
570}
571
572static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
573{
574 return qdisc->dev_queue->dev;
575}
576
577static inline void sch_tree_lock(struct Qdisc *q)
578{
579 if (q->flags & TCQ_F_MQROOT)
580 spin_lock_bh(qdisc_lock(q));
581 else
582 spin_lock_bh(qdisc_root_sleeping_lock(q));
583}
584
585static inline void sch_tree_unlock(struct Qdisc *q)
586{
587 if (q->flags & TCQ_F_MQROOT)
588 spin_unlock_bh(qdisc_lock(q));
589 else
590 spin_unlock_bh(qdisc_root_sleeping_lock(q));
591}
592
593extern struct Qdisc noop_qdisc;
594extern struct Qdisc_ops noop_qdisc_ops;
595extern struct Qdisc_ops pfifo_fast_ops;
596extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
597extern struct Qdisc_ops mq_qdisc_ops;
598extern struct Qdisc_ops noqueue_qdisc_ops;
599extern const struct Qdisc_ops *default_qdisc_ops;
600static inline const struct Qdisc_ops *
601get_default_qdisc_ops(const struct net_device *dev, int ntx)
602{
603 return ntx < dev->real_num_tx_queues ?
604 default_qdisc_ops : &pfifo_fast_ops;
605}
606
607struct Qdisc_class_common {
608 u32 classid;
609 unsigned int filter_cnt;
610 struct hlist_node hnode;
611};
612
613struct Qdisc_class_hash {
614 struct hlist_head *hash;
615 unsigned int hashsize;
616 unsigned int hashmask;
617 unsigned int hashelems;
618};
619
620static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
621{
622 id ^= id >> 8;
623 id ^= id >> 4;
624 return id & mask;
625}
626
627static inline struct Qdisc_class_common *
628qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
629{
630 struct Qdisc_class_common *cl;
631 unsigned int h;
632
633 if (!id)
634 return NULL;
635
636 h = qdisc_class_hash(id, hash->hashmask);
637 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
638 if (cl->classid == id)
639 return cl;
640 }
641 return NULL;
642}
643
644static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
645{
646 return cl->filter_cnt > 0;
647}
648
649static inline void qdisc_class_get(struct Qdisc_class_common *cl)
650{
651 unsigned int res;
652
653 if (check_add_overflow(cl->filter_cnt, 1, &res))
654 WARN(1, "Qdisc class overflow");
655
656 cl->filter_cnt = res;
657}
658
659static inline void qdisc_class_put(struct Qdisc_class_common *cl)
660{
661 unsigned int res;
662
663 if (check_sub_overflow(cl->filter_cnt, 1, &res))
664 WARN(1, "Qdisc class underflow");
665
666 cl->filter_cnt = res;
667}
668
669static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
670{
671 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
672
673 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
674}
675
676int qdisc_class_hash_init(struct Qdisc_class_hash *);
677void qdisc_class_hash_insert(struct Qdisc_class_hash *,
678 struct Qdisc_class_common *);
679void qdisc_class_hash_remove(struct Qdisc_class_hash *,
680 struct Qdisc_class_common *);
681void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
682void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
683
684int dev_qdisc_change_tx_queue_len(struct net_device *dev);
685void dev_qdisc_change_real_num_tx(struct net_device *dev,
686 unsigned int new_real_tx);
687void dev_init_scheduler(struct net_device *dev);
688void dev_shutdown(struct net_device *dev);
689void dev_activate(struct net_device *dev);
690void dev_deactivate(struct net_device *dev);
691void dev_deactivate_many(struct list_head *head);
692struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
693 struct Qdisc *qdisc);
694void qdisc_reset(struct Qdisc *qdisc);
695void qdisc_destroy(struct Qdisc *qdisc);
696void qdisc_put(struct Qdisc *qdisc);
697void qdisc_put_unlocked(struct Qdisc *qdisc);
698void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
699#ifdef CONFIG_NET_SCHED
700int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
701 void *type_data);
702void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
703 struct Qdisc *new, struct Qdisc *old,
704 enum tc_setup_type type, void *type_data,
705 struct netlink_ext_ack *extack);
706#else
707static inline int
708qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
709 void *type_data)
710{
711 q->flags &= ~TCQ_F_OFFLOADED;
712 return 0;
713}
714
715static inline void
716qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
717 struct Qdisc *new, struct Qdisc *old,
718 enum tc_setup_type type, void *type_data,
719 struct netlink_ext_ack *extack)
720{
721}
722#endif
723void qdisc_offload_query_caps(struct net_device *dev,
724 enum tc_setup_type type,
725 void *caps, size_t caps_len);
726struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
727 const struct Qdisc_ops *ops,
728 struct netlink_ext_ack *extack);
729void qdisc_free(struct Qdisc *qdisc);
730struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
731 const struct Qdisc_ops *ops, u32 parentid,
732 struct netlink_ext_ack *extack);
733void __qdisc_calculate_pkt_len(struct sk_buff *skb,
734 const struct qdisc_size_table *stab);
735int skb_do_redirect(struct sk_buff *);
736
737static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
738{
739#ifdef CONFIG_NET_XGRESS
740 return skb->tc_at_ingress;
741#else
742 return false;
743#endif
744}
745
746static inline bool skb_skip_tc_classify(struct sk_buff *skb)
747{
748#ifdef CONFIG_NET_CLS_ACT
749 if (skb->tc_skip_classify) {
750 skb->tc_skip_classify = 0;
751 return true;
752 }
753#endif
754 return false;
755}
756
757/* Reset all TX qdiscs greater than index of a device. */
758static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
759{
760 struct Qdisc *qdisc;
761
762 for (; i < dev->num_tx_queues; i++) {
763 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
764 if (qdisc) {
765 spin_lock_bh(qdisc_lock(qdisc));
766 qdisc_reset(qdisc);
767 spin_unlock_bh(qdisc_lock(qdisc));
768 }
769 }
770}
771
772/* Are all TX queues of the device empty? */
773static inline bool qdisc_all_tx_empty(const struct net_device *dev)
774{
775 unsigned int i;
776
777 rcu_read_lock();
778 for (i = 0; i < dev->num_tx_queues; i++) {
779 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
780 const struct Qdisc *q = rcu_dereference(txq->qdisc);
781
782 if (!qdisc_is_empty(q)) {
783 rcu_read_unlock();
784 return false;
785 }
786 }
787 rcu_read_unlock();
788 return true;
789}
790
791/* Are any of the TX qdiscs changing? */
792static inline bool qdisc_tx_changing(const struct net_device *dev)
793{
794 unsigned int i;
795
796 for (i = 0; i < dev->num_tx_queues; i++) {
797 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
798
799 if (rcu_access_pointer(txq->qdisc) !=
800 rcu_access_pointer(txq->qdisc_sleeping))
801 return true;
802 }
803 return false;
804}
805
806/* Is the device using the noop qdisc on all queues? */
807static inline bool qdisc_tx_is_noop(const struct net_device *dev)
808{
809 unsigned int i;
810
811 for (i = 0; i < dev->num_tx_queues; i++) {
812 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
813 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
814 return false;
815 }
816 return true;
817}
818
819static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
820{
821 return qdisc_skb_cb(skb)->pkt_len;
822}
823
824/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
825enum net_xmit_qdisc_t {
826 __NET_XMIT_STOLEN = 0x00010000,
827 __NET_XMIT_BYPASS = 0x00020000,
828};
829
830#ifdef CONFIG_NET_CLS_ACT
831#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
832#else
833#define net_xmit_drop_count(e) (1)
834#endif
835
836static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
837 const struct Qdisc *sch)
838{
839#ifdef CONFIG_NET_SCHED
840 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
841
842 if (stab)
843 __qdisc_calculate_pkt_len(skb, stab);
844#endif
845}
846
847static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
848 struct sk_buff **to_free)
849{
850 return sch->enqueue(skb, sch, to_free);
851}
852
853static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
854 __u64 bytes, __u64 packets)
855{
856 u64_stats_update_begin(&bstats->syncp);
857 u64_stats_add(&bstats->bytes, bytes);
858 u64_stats_add(&bstats->packets, packets);
859 u64_stats_update_end(&bstats->syncp);
860}
861
862static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
863 const struct sk_buff *skb)
864{
865 _bstats_update(bstats,
866 qdisc_pkt_len(skb),
867 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
868}
869
870static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
871 const struct sk_buff *skb)
872{
873 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
874}
875
876static inline void qdisc_bstats_update(struct Qdisc *sch,
877 const struct sk_buff *skb)
878{
879 bstats_update(&sch->bstats, skb);
880}
881
882static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
883 const struct sk_buff *skb)
884{
885 sch->qstats.backlog -= qdisc_pkt_len(skb);
886}
887
888static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
889 const struct sk_buff *skb)
890{
891 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
892}
893
894static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
895 const struct sk_buff *skb)
896{
897 sch->qstats.backlog += qdisc_pkt_len(skb);
898}
899
900static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
901 const struct sk_buff *skb)
902{
903 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
904}
905
906static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
907{
908 this_cpu_inc(sch->cpu_qstats->qlen);
909}
910
911static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
912{
913 this_cpu_dec(sch->cpu_qstats->qlen);
914}
915
916static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
917{
918 this_cpu_inc(sch->cpu_qstats->requeues);
919}
920
921static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
922{
923 sch->qstats.drops += count;
924}
925
926static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
927{
928 qstats->drops++;
929}
930
931static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
932{
933 qstats->overlimits++;
934}
935
936static inline void qdisc_qstats_drop(struct Qdisc *sch)
937{
938 qstats_drop_inc(&sch->qstats);
939}
940
941static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
942{
943 this_cpu_inc(sch->cpu_qstats->drops);
944}
945
946static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
947{
948 sch->qstats.overlimits++;
949}
950
951static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
952{
953 __u32 qlen = qdisc_qlen_sum(sch);
954
955 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
956}
957
958static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
959 __u32 *backlog)
960{
961 struct gnet_stats_queue qstats = { 0 };
962
963 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
964 *qlen = qstats.qlen + qdisc_qlen(sch);
965 *backlog = qstats.backlog;
966}
967
968static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
969{
970 __u32 qlen, backlog;
971
972 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
973 qdisc_tree_reduce_backlog(sch, qlen, backlog);
974}
975
976static inline void qdisc_purge_queue(struct Qdisc *sch)
977{
978 __u32 qlen, backlog;
979
980 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
981 qdisc_reset(sch);
982 qdisc_tree_reduce_backlog(sch, qlen, backlog);
983}
984
985static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
986 struct qdisc_skb_head *qh)
987{
988 struct sk_buff *last = qh->tail;
989
990 if (last) {
991 skb->next = NULL;
992 last->next = skb;
993 qh->tail = skb;
994 } else {
995 qh->tail = skb;
996 qh->head = skb;
997 }
998 qh->qlen++;
999}
1000
1001static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1002{
1003 __qdisc_enqueue_tail(skb, &sch->q);
1004 qdisc_qstats_backlog_inc(sch, skb);
1005 return NET_XMIT_SUCCESS;
1006}
1007
1008static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1009 struct qdisc_skb_head *qh)
1010{
1011 skb->next = qh->head;
1012
1013 if (!qh->head)
1014 qh->tail = skb;
1015 qh->head = skb;
1016 qh->qlen++;
1017}
1018
1019static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1020{
1021 struct sk_buff *skb = qh->head;
1022
1023 if (likely(skb != NULL)) {
1024 qh->head = skb->next;
1025 qh->qlen--;
1026 if (qh->head == NULL)
1027 qh->tail = NULL;
1028 skb->next = NULL;
1029 }
1030
1031 return skb;
1032}
1033
1034static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1035{
1036 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1037
1038 if (likely(skb != NULL)) {
1039 qdisc_qstats_backlog_dec(sch, skb);
1040 qdisc_bstats_update(sch, skb);
1041 }
1042
1043 return skb;
1044}
1045
1046struct tc_skb_cb {
1047 struct qdisc_skb_cb qdisc_cb;
1048 u32 drop_reason;
1049
1050 u16 zone; /* Only valid if post_ct = true */
1051 u16 mru;
1052 u8 post_ct:1;
1053 u8 post_ct_snat:1;
1054 u8 post_ct_dnat:1;
1055};
1056
1057static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
1058{
1059 struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
1060
1061 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
1062 return cb;
1063}
1064
1065static inline enum skb_drop_reason
1066tcf_get_drop_reason(const struct sk_buff *skb)
1067{
1068 return tc_skb_cb(skb)->drop_reason;
1069}
1070
1071static inline void tcf_set_drop_reason(const struct sk_buff *skb,
1072 enum skb_drop_reason reason)
1073{
1074 tc_skb_cb(skb)->drop_reason = reason;
1075}
1076
1077/* Instead of calling kfree_skb() while root qdisc lock is held,
1078 * queue the skb for future freeing at end of __dev_xmit_skb()
1079 */
1080static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1081{
1082 skb->next = *to_free;
1083 *to_free = skb;
1084}
1085
1086static inline void __qdisc_drop_all(struct sk_buff *skb,
1087 struct sk_buff **to_free)
1088{
1089 if (skb->prev)
1090 skb->prev->next = *to_free;
1091 else
1092 skb->next = *to_free;
1093 *to_free = skb;
1094}
1095
1096static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1097 struct qdisc_skb_head *qh,
1098 struct sk_buff **to_free)
1099{
1100 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1101
1102 if (likely(skb != NULL)) {
1103 unsigned int len = qdisc_pkt_len(skb);
1104
1105 qdisc_qstats_backlog_dec(sch, skb);
1106 __qdisc_drop(skb, to_free);
1107 return len;
1108 }
1109
1110 return 0;
1111}
1112
1113static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1114{
1115 const struct qdisc_skb_head *qh = &sch->q;
1116
1117 return qh->head;
1118}
1119
1120/* generic pseudo peek method for non-work-conserving qdisc */
1121static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1122{
1123 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1124
1125 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1126 if (!skb) {
1127 skb = sch->dequeue(sch);
1128
1129 if (skb) {
1130 __skb_queue_head(&sch->gso_skb, skb);
1131 /* it's still part of the queue */
1132 qdisc_qstats_backlog_inc(sch, skb);
1133 sch->q.qlen++;
1134 }
1135 }
1136
1137 return skb;
1138}
1139
1140static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1141 struct sk_buff *skb)
1142{
1143 if (qdisc_is_percpu_stats(sch)) {
1144 qdisc_qstats_cpu_backlog_dec(sch, skb);
1145 qdisc_bstats_cpu_update(sch, skb);
1146 qdisc_qstats_cpu_qlen_dec(sch);
1147 } else {
1148 qdisc_qstats_backlog_dec(sch, skb);
1149 qdisc_bstats_update(sch, skb);
1150 sch->q.qlen--;
1151 }
1152}
1153
1154static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1155 unsigned int pkt_len)
1156{
1157 if (qdisc_is_percpu_stats(sch)) {
1158 qdisc_qstats_cpu_qlen_inc(sch);
1159 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1160 } else {
1161 sch->qstats.backlog += pkt_len;
1162 sch->q.qlen++;
1163 }
1164}
1165
1166/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1167static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1168{
1169 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1170
1171 if (skb) {
1172 skb = __skb_dequeue(&sch->gso_skb);
1173 if (qdisc_is_percpu_stats(sch)) {
1174 qdisc_qstats_cpu_backlog_dec(sch, skb);
1175 qdisc_qstats_cpu_qlen_dec(sch);
1176 } else {
1177 qdisc_qstats_backlog_dec(sch, skb);
1178 sch->q.qlen--;
1179 }
1180 } else {
1181 skb = sch->dequeue(sch);
1182 }
1183
1184 return skb;
1185}
1186
1187static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1188{
1189 /*
1190 * We do not know the backlog in bytes of this list, it
1191 * is up to the caller to correct it
1192 */
1193 ASSERT_RTNL();
1194 if (qh->qlen) {
1195 rtnl_kfree_skbs(qh->head, qh->tail);
1196
1197 qh->head = NULL;
1198 qh->tail = NULL;
1199 qh->qlen = 0;
1200 }
1201}
1202
1203static inline void qdisc_reset_queue(struct Qdisc *sch)
1204{
1205 __qdisc_reset_queue(&sch->q);
1206}
1207
1208static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1209 struct Qdisc **pold)
1210{
1211 struct Qdisc *old;
1212
1213 sch_tree_lock(sch);
1214 old = *pold;
1215 *pold = new;
1216 if (old != NULL)
1217 qdisc_purge_queue(old);
1218 sch_tree_unlock(sch);
1219
1220 return old;
1221}
1222
1223static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1224{
1225 rtnl_kfree_skbs(skb, skb);
1226 qdisc_qstats_drop(sch);
1227}
1228
1229static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1230 struct sk_buff **to_free)
1231{
1232 __qdisc_drop(skb, to_free);
1233 qdisc_qstats_cpu_drop(sch);
1234
1235 return NET_XMIT_DROP;
1236}
1237
1238static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1239 struct sk_buff **to_free)
1240{
1241 __qdisc_drop(skb, to_free);
1242 qdisc_qstats_drop(sch);
1243
1244 return NET_XMIT_DROP;
1245}
1246
1247static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1248 struct sk_buff **to_free)
1249{
1250 __qdisc_drop_all(skb, to_free);
1251 qdisc_qstats_drop(sch);
1252
1253 return NET_XMIT_DROP;
1254}
1255
1256struct psched_ratecfg {
1257 u64 rate_bytes_ps; /* bytes per second */
1258 u32 mult;
1259 u16 overhead;
1260 u16 mpu;
1261 u8 linklayer;
1262 u8 shift;
1263};
1264
1265static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1266 unsigned int len)
1267{
1268 len += r->overhead;
1269
1270 if (len < r->mpu)
1271 len = r->mpu;
1272
1273 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1274 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1275
1276 return ((u64)len * r->mult) >> r->shift;
1277}
1278
1279void psched_ratecfg_precompute(struct psched_ratecfg *r,
1280 const struct tc_ratespec *conf,
1281 u64 rate64);
1282
1283static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1284 const struct psched_ratecfg *r)
1285{
1286 memset(res, 0, sizeof(*res));
1287
1288 /* legacy struct tc_ratespec has a 32bit @rate field
1289 * Qdisc using 64bit rate should add new attributes
1290 * in order to maintain compatibility.
1291 */
1292 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1293
1294 res->overhead = r->overhead;
1295 res->mpu = r->mpu;
1296 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1297}
1298
1299struct psched_pktrate {
1300 u64 rate_pkts_ps; /* packets per second */
1301 u32 mult;
1302 u8 shift;
1303};
1304
1305static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1306 unsigned int pkt_num)
1307{
1308 return ((u64)pkt_num * r->mult) >> r->shift;
1309}
1310
1311void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1312
1313/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1314 * The fast path only needs to access filter list and to update stats
1315 */
1316struct mini_Qdisc {
1317 struct tcf_proto *filter_list;
1318 struct tcf_block *block;
1319 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1320 struct gnet_stats_queue __percpu *cpu_qstats;
1321 unsigned long rcu_state;
1322};
1323
1324static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1325 const struct sk_buff *skb)
1326{
1327 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1328}
1329
1330static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1331{
1332 this_cpu_inc(miniq->cpu_qstats->drops);
1333}
1334
1335struct mini_Qdisc_pair {
1336 struct mini_Qdisc miniq1;
1337 struct mini_Qdisc miniq2;
1338 struct mini_Qdisc __rcu **p_miniq;
1339};
1340
1341void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1342 struct tcf_proto *tp_head);
1343void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1344 struct mini_Qdisc __rcu **p_miniq);
1345void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1346 struct tcf_block *block);
1347
1348void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1349
1350int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1351
1352/* Make sure qdisc is no longer in SCHED state. */
1353static inline void qdisc_synchronize(const struct Qdisc *q)
1354{
1355 while (test_bit(__QDISC_STATE_SCHED, &q->state))
1356 msleep(1);
1357}
1358
1359#endif
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 unsigned long state;
121 unsigned long state2; /* must be written under qdisc spinlock */
122 struct Qdisc *next_sched;
123 struct sk_buff_head skb_bad_txq;
124
125 spinlock_t busylock ____cacheline_aligned_in_smp;
126 spinlock_t seqlock;
127
128 struct rcu_head rcu;
129 netdevice_tracker dev_tracker;
130 /* private data */
131 long privdata[] ____cacheline_aligned;
132};
133
134static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
135{
136 if (qdisc->flags & TCQ_F_BUILTIN)
137 return;
138 refcount_inc(&qdisc->refcnt);
139}
140
141static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc)
142{
143 if (qdisc->flags & TCQ_F_BUILTIN)
144 return true;
145 return refcount_dec_if_one(&qdisc->refcnt);
146}
147
148/* Intended to be used by unlocked users, when concurrent qdisc release is
149 * possible.
150 */
151
152static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
153{
154 if (qdisc->flags & TCQ_F_BUILTIN)
155 return qdisc;
156 if (refcount_inc_not_zero(&qdisc->refcnt))
157 return qdisc;
158 return NULL;
159}
160
161/* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
162 * root_lock section, or provide their own memory barriers -- ordering
163 * against qdisc_run_begin/end() atomic bit operations.
164 */
165static inline bool qdisc_is_running(struct Qdisc *qdisc)
166{
167 if (qdisc->flags & TCQ_F_NOLOCK)
168 return spin_is_locked(&qdisc->seqlock);
169 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
170}
171
172static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
173{
174 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
175}
176
177static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
178{
179 return q->flags & TCQ_F_CPUSTATS;
180}
181
182static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
183{
184 if (qdisc_is_percpu_stats(qdisc))
185 return nolock_qdisc_is_empty(qdisc);
186 return !READ_ONCE(qdisc->q.qlen);
187}
188
189/* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
190 * the qdisc root lock acquired.
191 */
192static inline bool qdisc_run_begin(struct Qdisc *qdisc)
193{
194 if (qdisc->flags & TCQ_F_NOLOCK) {
195 if (spin_trylock(&qdisc->seqlock))
196 return true;
197
198 /* No need to insist if the MISSED flag was already set.
199 * Note that test_and_set_bit() also gives us memory ordering
200 * guarantees wrt potential earlier enqueue() and below
201 * spin_trylock(), both of which are necessary to prevent races
202 */
203 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
204 return false;
205
206 /* Try to take the lock again to make sure that we will either
207 * grab it or the CPU that still has it will see MISSED set
208 * when testing it in qdisc_run_end()
209 */
210 return spin_trylock(&qdisc->seqlock);
211 }
212 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
213}
214
215static inline void qdisc_run_end(struct Qdisc *qdisc)
216{
217 if (qdisc->flags & TCQ_F_NOLOCK) {
218 spin_unlock(&qdisc->seqlock);
219
220 /* spin_unlock() only has store-release semantic. The unlock
221 * and test_bit() ordering is a store-load ordering, so a full
222 * memory barrier is needed here.
223 */
224 smp_mb();
225
226 if (unlikely(test_bit(__QDISC_STATE_MISSED,
227 &qdisc->state)))
228 __netif_schedule(qdisc);
229 } else {
230 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
231 }
232}
233
234static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
235{
236 return qdisc->flags & TCQ_F_ONETXQUEUE;
237}
238
239static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
240{
241 return netdev_queue_dql_avail(txq);
242}
243
244struct Qdisc_class_ops {
245 unsigned int flags;
246 /* Child qdisc manipulation */
247 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
248 int (*graft)(struct Qdisc *, unsigned long cl,
249 struct Qdisc *, struct Qdisc **,
250 struct netlink_ext_ack *extack);
251 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
252 void (*qlen_notify)(struct Qdisc *, unsigned long);
253
254 /* Class manipulation routines */
255 unsigned long (*find)(struct Qdisc *, u32 classid);
256 int (*change)(struct Qdisc *, u32, u32,
257 struct nlattr **, unsigned long *,
258 struct netlink_ext_ack *);
259 int (*delete)(struct Qdisc *, unsigned long,
260 struct netlink_ext_ack *);
261 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
262
263 /* Filter manipulation */
264 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
265 unsigned long arg,
266 struct netlink_ext_ack *extack);
267 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
268 u32 classid);
269 void (*unbind_tcf)(struct Qdisc *, unsigned long);
270
271 /* rtnetlink specific */
272 int (*dump)(struct Qdisc *, unsigned long,
273 struct sk_buff *skb, struct tcmsg*);
274 int (*dump_stats)(struct Qdisc *, unsigned long,
275 struct gnet_dump *);
276};
277
278/* Qdisc_class_ops flag values */
279
280/* Implements API that doesn't require rtnl lock */
281enum qdisc_class_ops_flags {
282 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
283};
284
285struct Qdisc_ops {
286 struct Qdisc_ops *next;
287 const struct Qdisc_class_ops *cl_ops;
288 char id[IFNAMSIZ];
289 int priv_size;
290 unsigned int static_flags;
291
292 int (*enqueue)(struct sk_buff *skb,
293 struct Qdisc *sch,
294 struct sk_buff **to_free);
295 struct sk_buff * (*dequeue)(struct Qdisc *);
296 struct sk_buff * (*peek)(struct Qdisc *);
297
298 int (*init)(struct Qdisc *sch, struct nlattr *arg,
299 struct netlink_ext_ack *extack);
300 void (*reset)(struct Qdisc *);
301 void (*destroy)(struct Qdisc *);
302 int (*change)(struct Qdisc *sch,
303 struct nlattr *arg,
304 struct netlink_ext_ack *extack);
305 void (*attach)(struct Qdisc *sch);
306 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
307 void (*change_real_num_tx)(struct Qdisc *sch,
308 unsigned int new_real_tx);
309
310 int (*dump)(struct Qdisc *, struct sk_buff *);
311 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
312
313 void (*ingress_block_set)(struct Qdisc *sch,
314 u32 block_index);
315 void (*egress_block_set)(struct Qdisc *sch,
316 u32 block_index);
317 u32 (*ingress_block_get)(struct Qdisc *sch);
318 u32 (*egress_block_get)(struct Qdisc *sch);
319
320 struct module *owner;
321};
322
323struct tcf_result {
324 union {
325 struct {
326 unsigned long class;
327 u32 classid;
328 };
329 const struct tcf_proto *goto_tp;
330 };
331};
332
333struct tcf_chain;
334
335struct tcf_proto_ops {
336 struct list_head head;
337 char kind[IFNAMSIZ];
338
339 int (*classify)(struct sk_buff *,
340 const struct tcf_proto *,
341 struct tcf_result *);
342 int (*init)(struct tcf_proto*);
343 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
344 struct netlink_ext_ack *extack);
345
346 void* (*get)(struct tcf_proto*, u32 handle);
347 void (*put)(struct tcf_proto *tp, void *f);
348 int (*change)(struct net *net, struct sk_buff *,
349 struct tcf_proto*, unsigned long,
350 u32 handle, struct nlattr **,
351 void **, u32,
352 struct netlink_ext_ack *);
353 int (*delete)(struct tcf_proto *tp, void *arg,
354 bool *last, bool rtnl_held,
355 struct netlink_ext_ack *);
356 bool (*delete_empty)(struct tcf_proto *tp);
357 void (*walk)(struct tcf_proto *tp,
358 struct tcf_walker *arg, bool rtnl_held);
359 int (*reoffload)(struct tcf_proto *tp, bool add,
360 flow_setup_cb_t *cb, void *cb_priv,
361 struct netlink_ext_ack *extack);
362 void (*hw_add)(struct tcf_proto *tp,
363 void *type_data);
364 void (*hw_del)(struct tcf_proto *tp,
365 void *type_data);
366 void (*bind_class)(void *, u32, unsigned long,
367 void *, unsigned long);
368 void * (*tmplt_create)(struct net *net,
369 struct tcf_chain *chain,
370 struct nlattr **tca,
371 struct netlink_ext_ack *extack);
372 void (*tmplt_destroy)(void *tmplt_priv);
373 void (*tmplt_reoffload)(struct tcf_chain *chain,
374 bool add,
375 flow_setup_cb_t *cb,
376 void *cb_priv);
377 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp,
378 u32 handle);
379
380 /* rtnetlink specific */
381 int (*dump)(struct net*, struct tcf_proto*, void *,
382 struct sk_buff *skb, struct tcmsg*,
383 bool);
384 int (*terse_dump)(struct net *net,
385 struct tcf_proto *tp, void *fh,
386 struct sk_buff *skb,
387 struct tcmsg *t, bool rtnl_held);
388 int (*tmplt_dump)(struct sk_buff *skb,
389 struct net *net,
390 void *tmplt_priv);
391
392 struct module *owner;
393 int flags;
394};
395
396/* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
397 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
398 * conditions can occur when filters are inserted/deleted simultaneously.
399 */
400enum tcf_proto_ops_flags {
401 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
402};
403
404struct tcf_proto {
405 /* Fast access part */
406 struct tcf_proto __rcu *next;
407 void __rcu *root;
408
409 /* called under RCU BH lock*/
410 int (*classify)(struct sk_buff *,
411 const struct tcf_proto *,
412 struct tcf_result *);
413 __be16 protocol;
414
415 /* All the rest */
416 u32 prio;
417 void *data;
418 const struct tcf_proto_ops *ops;
419 struct tcf_chain *chain;
420 /* Lock protects tcf_proto shared state and can be used by unlocked
421 * classifiers to protect their private data.
422 */
423 spinlock_t lock;
424 bool deleting;
425 refcount_t refcnt;
426 struct rcu_head rcu;
427 struct hlist_node destroy_ht_node;
428};
429
430struct qdisc_skb_cb {
431 struct {
432 unsigned int pkt_len;
433 u16 slave_dev_queue_mapping;
434 u16 tc_classid;
435 };
436#define QDISC_CB_PRIV_LEN 20
437 unsigned char data[QDISC_CB_PRIV_LEN];
438};
439
440typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
441
442struct tcf_chain {
443 /* Protects filter_chain. */
444 struct mutex filter_chain_lock;
445 struct tcf_proto __rcu *filter_chain;
446 struct list_head list;
447 struct tcf_block *block;
448 u32 index; /* chain index */
449 unsigned int refcnt;
450 unsigned int action_refcnt;
451 bool explicitly_created;
452 bool flushing;
453 const struct tcf_proto_ops *tmplt_ops;
454 void *tmplt_priv;
455 struct rcu_head rcu;
456};
457
458struct tcf_block {
459 struct xarray ports; /* datapath accessible */
460 /* Lock protects tcf_block and lifetime-management data of chains
461 * attached to the block (refcnt, action_refcnt, explicitly_created).
462 */
463 struct mutex lock;
464 struct list_head chain_list;
465 u32 index; /* block index for shared blocks */
466 u32 classid; /* which class this block belongs to */
467 refcount_t refcnt;
468 struct net *net;
469 struct Qdisc *q;
470 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
471 struct flow_block flow_block;
472 struct list_head owner_list;
473 bool keep_dst;
474 atomic_t offloadcnt; /* Number of oddloaded filters */
475 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
476 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
477 struct {
478 struct tcf_chain *chain;
479 struct list_head filter_chain_list;
480 } chain0;
481 struct rcu_head rcu;
482 DECLARE_HASHTABLE(proto_destroy_ht, 7);
483 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
484};
485
486struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index);
487
488static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
489{
490 return lockdep_is_held(&chain->filter_chain_lock);
491}
492
493static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
494{
495 return lockdep_is_held(&tp->lock);
496}
497
498#define tcf_chain_dereference(p, chain) \
499 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
500
501#define tcf_proto_dereference(p, tp) \
502 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
503
504static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
505{
506 struct qdisc_skb_cb *qcb;
507
508 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
509 BUILD_BUG_ON(sizeof(qcb->data) < sz);
510}
511
512static inline int qdisc_qlen(const struct Qdisc *q)
513{
514 return q->q.qlen;
515}
516
517static inline int qdisc_qlen_sum(const struct Qdisc *q)
518{
519 __u32 qlen = q->qstats.qlen;
520 int i;
521
522 if (qdisc_is_percpu_stats(q)) {
523 for_each_possible_cpu(i)
524 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
525 } else {
526 qlen += q->q.qlen;
527 }
528
529 return qlen;
530}
531
532static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
533{
534 return (struct qdisc_skb_cb *)skb->cb;
535}
536
537static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
538{
539 return &qdisc->q.lock;
540}
541
542static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
543{
544 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
545
546 return q;
547}
548
549static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
550{
551 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
552}
553
554static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
555{
556 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
557}
558
559static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
560{
561 struct Qdisc *root = qdisc_root_sleeping(qdisc);
562
563 ASSERT_RTNL();
564 return qdisc_lock(root);
565}
566
567static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
568{
569 return qdisc->dev_queue->dev;
570}
571
572static inline void sch_tree_lock(struct Qdisc *q)
573{
574 if (q->flags & TCQ_F_MQROOT)
575 spin_lock_bh(qdisc_lock(q));
576 else
577 spin_lock_bh(qdisc_root_sleeping_lock(q));
578}
579
580static inline void sch_tree_unlock(struct Qdisc *q)
581{
582 if (q->flags & TCQ_F_MQROOT)
583 spin_unlock_bh(qdisc_lock(q));
584 else
585 spin_unlock_bh(qdisc_root_sleeping_lock(q));
586}
587
588extern struct Qdisc noop_qdisc;
589extern struct Qdisc_ops noop_qdisc_ops;
590extern struct Qdisc_ops pfifo_fast_ops;
591extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
592extern struct Qdisc_ops mq_qdisc_ops;
593extern struct Qdisc_ops noqueue_qdisc_ops;
594extern const struct Qdisc_ops *default_qdisc_ops;
595static inline const struct Qdisc_ops *
596get_default_qdisc_ops(const struct net_device *dev, int ntx)
597{
598 return ntx < dev->real_num_tx_queues ?
599 default_qdisc_ops : &pfifo_fast_ops;
600}
601
602struct Qdisc_class_common {
603 u32 classid;
604 unsigned int filter_cnt;
605 struct hlist_node hnode;
606};
607
608struct Qdisc_class_hash {
609 struct hlist_head *hash;
610 unsigned int hashsize;
611 unsigned int hashmask;
612 unsigned int hashelems;
613};
614
615static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
616{
617 id ^= id >> 8;
618 id ^= id >> 4;
619 return id & mask;
620}
621
622static inline struct Qdisc_class_common *
623qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
624{
625 struct Qdisc_class_common *cl;
626 unsigned int h;
627
628 if (!id)
629 return NULL;
630
631 h = qdisc_class_hash(id, hash->hashmask);
632 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
633 if (cl->classid == id)
634 return cl;
635 }
636 return NULL;
637}
638
639static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
640{
641 return cl->filter_cnt > 0;
642}
643
644static inline void qdisc_class_get(struct Qdisc_class_common *cl)
645{
646 unsigned int res;
647
648 if (check_add_overflow(cl->filter_cnt, 1, &res))
649 WARN(1, "Qdisc class overflow");
650
651 cl->filter_cnt = res;
652}
653
654static inline void qdisc_class_put(struct Qdisc_class_common *cl)
655{
656 unsigned int res;
657
658 if (check_sub_overflow(cl->filter_cnt, 1, &res))
659 WARN(1, "Qdisc class underflow");
660
661 cl->filter_cnt = res;
662}
663
664static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
665{
666 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
667
668 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
669}
670
671int qdisc_class_hash_init(struct Qdisc_class_hash *);
672void qdisc_class_hash_insert(struct Qdisc_class_hash *,
673 struct Qdisc_class_common *);
674void qdisc_class_hash_remove(struct Qdisc_class_hash *,
675 struct Qdisc_class_common *);
676void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
677void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
678
679int dev_qdisc_change_tx_queue_len(struct net_device *dev);
680void dev_qdisc_change_real_num_tx(struct net_device *dev,
681 unsigned int new_real_tx);
682void dev_init_scheduler(struct net_device *dev);
683void dev_shutdown(struct net_device *dev);
684void dev_activate(struct net_device *dev);
685void dev_deactivate(struct net_device *dev);
686void dev_deactivate_many(struct list_head *head);
687struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
688 struct Qdisc *qdisc);
689void qdisc_reset(struct Qdisc *qdisc);
690void qdisc_destroy(struct Qdisc *qdisc);
691void qdisc_put(struct Qdisc *qdisc);
692void qdisc_put_unlocked(struct Qdisc *qdisc);
693void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
694#ifdef CONFIG_NET_SCHED
695int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
696 void *type_data);
697void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
698 struct Qdisc *new, struct Qdisc *old,
699 enum tc_setup_type type, void *type_data,
700 struct netlink_ext_ack *extack);
701#else
702static inline int
703qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
704 void *type_data)
705{
706 q->flags &= ~TCQ_F_OFFLOADED;
707 return 0;
708}
709
710static inline void
711qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
712 struct Qdisc *new, struct Qdisc *old,
713 enum tc_setup_type type, void *type_data,
714 struct netlink_ext_ack *extack)
715{
716}
717#endif
718void qdisc_offload_query_caps(struct net_device *dev,
719 enum tc_setup_type type,
720 void *caps, size_t caps_len);
721struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
722 const struct Qdisc_ops *ops,
723 struct netlink_ext_ack *extack);
724void qdisc_free(struct Qdisc *qdisc);
725struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
726 const struct Qdisc_ops *ops, u32 parentid,
727 struct netlink_ext_ack *extack);
728void __qdisc_calculate_pkt_len(struct sk_buff *skb,
729 const struct qdisc_size_table *stab);
730int skb_do_redirect(struct sk_buff *);
731
732static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
733{
734#ifdef CONFIG_NET_XGRESS
735 return skb->tc_at_ingress;
736#else
737 return false;
738#endif
739}
740
741static inline bool skb_skip_tc_classify(struct sk_buff *skb)
742{
743#ifdef CONFIG_NET_CLS_ACT
744 if (skb->tc_skip_classify) {
745 skb->tc_skip_classify = 0;
746 return true;
747 }
748#endif
749 return false;
750}
751
752/* Reset all TX qdiscs greater than index of a device. */
753static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
754{
755 struct Qdisc *qdisc;
756
757 for (; i < dev->num_tx_queues; i++) {
758 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
759 if (qdisc) {
760 spin_lock_bh(qdisc_lock(qdisc));
761 qdisc_reset(qdisc);
762 spin_unlock_bh(qdisc_lock(qdisc));
763 }
764 }
765}
766
767/* Are all TX queues of the device empty? */
768static inline bool qdisc_all_tx_empty(const struct net_device *dev)
769{
770 unsigned int i;
771
772 rcu_read_lock();
773 for (i = 0; i < dev->num_tx_queues; i++) {
774 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
775 const struct Qdisc *q = rcu_dereference(txq->qdisc);
776
777 if (!qdisc_is_empty(q)) {
778 rcu_read_unlock();
779 return false;
780 }
781 }
782 rcu_read_unlock();
783 return true;
784}
785
786/* Are any of the TX qdiscs changing? */
787static inline bool qdisc_tx_changing(const struct net_device *dev)
788{
789 unsigned int i;
790
791 for (i = 0; i < dev->num_tx_queues; i++) {
792 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
793
794 if (rcu_access_pointer(txq->qdisc) !=
795 rcu_access_pointer(txq->qdisc_sleeping))
796 return true;
797 }
798 return false;
799}
800
801/* Is the device using the noop qdisc on all queues? */
802static inline bool qdisc_tx_is_noop(const struct net_device *dev)
803{
804 unsigned int i;
805
806 for (i = 0; i < dev->num_tx_queues; i++) {
807 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
808 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
809 return false;
810 }
811 return true;
812}
813
814static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
815{
816 return qdisc_skb_cb(skb)->pkt_len;
817}
818
819/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
820enum net_xmit_qdisc_t {
821 __NET_XMIT_STOLEN = 0x00010000,
822 __NET_XMIT_BYPASS = 0x00020000,
823};
824
825#ifdef CONFIG_NET_CLS_ACT
826#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
827#else
828#define net_xmit_drop_count(e) (1)
829#endif
830
831static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
832 const struct Qdisc *sch)
833{
834#ifdef CONFIG_NET_SCHED
835 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
836
837 if (stab)
838 __qdisc_calculate_pkt_len(skb, stab);
839#endif
840}
841
842static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
843 struct sk_buff **to_free)
844{
845 qdisc_calculate_pkt_len(skb, sch);
846 return sch->enqueue(skb, sch, to_free);
847}
848
849static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
850 __u64 bytes, __u32 packets)
851{
852 u64_stats_update_begin(&bstats->syncp);
853 u64_stats_add(&bstats->bytes, bytes);
854 u64_stats_add(&bstats->packets, packets);
855 u64_stats_update_end(&bstats->syncp);
856}
857
858static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
859 const struct sk_buff *skb)
860{
861 _bstats_update(bstats,
862 qdisc_pkt_len(skb),
863 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
864}
865
866static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
867 const struct sk_buff *skb)
868{
869 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
870}
871
872static inline void qdisc_bstats_update(struct Qdisc *sch,
873 const struct sk_buff *skb)
874{
875 bstats_update(&sch->bstats, skb);
876}
877
878static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
879 const struct sk_buff *skb)
880{
881 sch->qstats.backlog -= qdisc_pkt_len(skb);
882}
883
884static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
885 const struct sk_buff *skb)
886{
887 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
888}
889
890static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
891 const struct sk_buff *skb)
892{
893 sch->qstats.backlog += qdisc_pkt_len(skb);
894}
895
896static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
897 const struct sk_buff *skb)
898{
899 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
900}
901
902static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
903{
904 this_cpu_inc(sch->cpu_qstats->qlen);
905}
906
907static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
908{
909 this_cpu_dec(sch->cpu_qstats->qlen);
910}
911
912static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
913{
914 this_cpu_inc(sch->cpu_qstats->requeues);
915}
916
917static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
918{
919 sch->qstats.drops += count;
920}
921
922static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
923{
924 qstats->drops++;
925}
926
927static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
928{
929 qstats->overlimits++;
930}
931
932static inline void qdisc_qstats_drop(struct Qdisc *sch)
933{
934 qstats_drop_inc(&sch->qstats);
935}
936
937static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
938{
939 this_cpu_inc(sch->cpu_qstats->drops);
940}
941
942static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
943{
944 sch->qstats.overlimits++;
945}
946
947static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
948{
949 __u32 qlen = qdisc_qlen_sum(sch);
950
951 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
952}
953
954static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
955 __u32 *backlog)
956{
957 struct gnet_stats_queue qstats = { 0 };
958
959 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
960 *qlen = qstats.qlen + qdisc_qlen(sch);
961 *backlog = qstats.backlog;
962}
963
964static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
965{
966 __u32 qlen, backlog;
967
968 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
969 qdisc_tree_reduce_backlog(sch, qlen, backlog);
970}
971
972static inline void qdisc_purge_queue(struct Qdisc *sch)
973{
974 __u32 qlen, backlog;
975
976 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
977 qdisc_reset(sch);
978 qdisc_tree_reduce_backlog(sch, qlen, backlog);
979}
980
981static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
982 struct qdisc_skb_head *qh)
983{
984 struct sk_buff *last = qh->tail;
985
986 if (last) {
987 skb->next = NULL;
988 last->next = skb;
989 qh->tail = skb;
990 } else {
991 qh->tail = skb;
992 qh->head = skb;
993 }
994 qh->qlen++;
995}
996
997static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
998{
999 __qdisc_enqueue_tail(skb, &sch->q);
1000 qdisc_qstats_backlog_inc(sch, skb);
1001 return NET_XMIT_SUCCESS;
1002}
1003
1004static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1005 struct qdisc_skb_head *qh)
1006{
1007 skb->next = qh->head;
1008
1009 if (!qh->head)
1010 qh->tail = skb;
1011 qh->head = skb;
1012 qh->qlen++;
1013}
1014
1015static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1016{
1017 struct sk_buff *skb = qh->head;
1018
1019 if (likely(skb != NULL)) {
1020 qh->head = skb->next;
1021 qh->qlen--;
1022 if (qh->head == NULL)
1023 qh->tail = NULL;
1024 skb->next = NULL;
1025 }
1026
1027 return skb;
1028}
1029
1030static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1031{
1032 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1033
1034 if (likely(skb != NULL)) {
1035 qdisc_qstats_backlog_dec(sch, skb);
1036 qdisc_bstats_update(sch, skb);
1037 }
1038
1039 return skb;
1040}
1041
1042struct tc_skb_cb {
1043 struct qdisc_skb_cb qdisc_cb;
1044 u32 drop_reason;
1045
1046 u16 zone; /* Only valid if post_ct = true */
1047 u16 mru;
1048 u8 post_ct:1;
1049 u8 post_ct_snat:1;
1050 u8 post_ct_dnat:1;
1051};
1052
1053static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
1054{
1055 struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
1056
1057 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
1058 return cb;
1059}
1060
1061static inline enum skb_drop_reason
1062tcf_get_drop_reason(const struct sk_buff *skb)
1063{
1064 return tc_skb_cb(skb)->drop_reason;
1065}
1066
1067static inline void tcf_set_drop_reason(const struct sk_buff *skb,
1068 enum skb_drop_reason reason)
1069{
1070 tc_skb_cb(skb)->drop_reason = reason;
1071}
1072
1073/* Instead of calling kfree_skb() while root qdisc lock is held,
1074 * queue the skb for future freeing at end of __dev_xmit_skb()
1075 */
1076static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1077{
1078 skb->next = *to_free;
1079 *to_free = skb;
1080}
1081
1082static inline void __qdisc_drop_all(struct sk_buff *skb,
1083 struct sk_buff **to_free)
1084{
1085 if (skb->prev)
1086 skb->prev->next = *to_free;
1087 else
1088 skb->next = *to_free;
1089 *to_free = skb;
1090}
1091
1092static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1093 struct qdisc_skb_head *qh,
1094 struct sk_buff **to_free)
1095{
1096 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1097
1098 if (likely(skb != NULL)) {
1099 unsigned int len = qdisc_pkt_len(skb);
1100
1101 qdisc_qstats_backlog_dec(sch, skb);
1102 __qdisc_drop(skb, to_free);
1103 return len;
1104 }
1105
1106 return 0;
1107}
1108
1109static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1110{
1111 const struct qdisc_skb_head *qh = &sch->q;
1112
1113 return qh->head;
1114}
1115
1116/* generic pseudo peek method for non-work-conserving qdisc */
1117static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1118{
1119 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1120
1121 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1122 if (!skb) {
1123 skb = sch->dequeue(sch);
1124
1125 if (skb) {
1126 __skb_queue_head(&sch->gso_skb, skb);
1127 /* it's still part of the queue */
1128 qdisc_qstats_backlog_inc(sch, skb);
1129 sch->q.qlen++;
1130 }
1131 }
1132
1133 return skb;
1134}
1135
1136static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1137 struct sk_buff *skb)
1138{
1139 if (qdisc_is_percpu_stats(sch)) {
1140 qdisc_qstats_cpu_backlog_dec(sch, skb);
1141 qdisc_bstats_cpu_update(sch, skb);
1142 qdisc_qstats_cpu_qlen_dec(sch);
1143 } else {
1144 qdisc_qstats_backlog_dec(sch, skb);
1145 qdisc_bstats_update(sch, skb);
1146 sch->q.qlen--;
1147 }
1148}
1149
1150static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1151 unsigned int pkt_len)
1152{
1153 if (qdisc_is_percpu_stats(sch)) {
1154 qdisc_qstats_cpu_qlen_inc(sch);
1155 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1156 } else {
1157 sch->qstats.backlog += pkt_len;
1158 sch->q.qlen++;
1159 }
1160}
1161
1162/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1163static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1164{
1165 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1166
1167 if (skb) {
1168 skb = __skb_dequeue(&sch->gso_skb);
1169 if (qdisc_is_percpu_stats(sch)) {
1170 qdisc_qstats_cpu_backlog_dec(sch, skb);
1171 qdisc_qstats_cpu_qlen_dec(sch);
1172 } else {
1173 qdisc_qstats_backlog_dec(sch, skb);
1174 sch->q.qlen--;
1175 }
1176 } else {
1177 skb = sch->dequeue(sch);
1178 }
1179
1180 return skb;
1181}
1182
1183static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1184{
1185 /*
1186 * We do not know the backlog in bytes of this list, it
1187 * is up to the caller to correct it
1188 */
1189 ASSERT_RTNL();
1190 if (qh->qlen) {
1191 rtnl_kfree_skbs(qh->head, qh->tail);
1192
1193 qh->head = NULL;
1194 qh->tail = NULL;
1195 qh->qlen = 0;
1196 }
1197}
1198
1199static inline void qdisc_reset_queue(struct Qdisc *sch)
1200{
1201 __qdisc_reset_queue(&sch->q);
1202}
1203
1204static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1205 struct Qdisc **pold)
1206{
1207 struct Qdisc *old;
1208
1209 sch_tree_lock(sch);
1210 old = *pold;
1211 *pold = new;
1212 if (old != NULL)
1213 qdisc_purge_queue(old);
1214 sch_tree_unlock(sch);
1215
1216 return old;
1217}
1218
1219static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1220{
1221 rtnl_kfree_skbs(skb, skb);
1222 qdisc_qstats_drop(sch);
1223}
1224
1225static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1226 struct sk_buff **to_free)
1227{
1228 __qdisc_drop(skb, to_free);
1229 qdisc_qstats_cpu_drop(sch);
1230
1231 return NET_XMIT_DROP;
1232}
1233
1234static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1235 struct sk_buff **to_free)
1236{
1237 __qdisc_drop(skb, to_free);
1238 qdisc_qstats_drop(sch);
1239
1240 return NET_XMIT_DROP;
1241}
1242
1243static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1244 struct sk_buff **to_free)
1245{
1246 __qdisc_drop_all(skb, to_free);
1247 qdisc_qstats_drop(sch);
1248
1249 return NET_XMIT_DROP;
1250}
1251
1252struct psched_ratecfg {
1253 u64 rate_bytes_ps; /* bytes per second */
1254 u32 mult;
1255 u16 overhead;
1256 u16 mpu;
1257 u8 linklayer;
1258 u8 shift;
1259};
1260
1261static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1262 unsigned int len)
1263{
1264 len += r->overhead;
1265
1266 if (len < r->mpu)
1267 len = r->mpu;
1268
1269 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1270 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1271
1272 return ((u64)len * r->mult) >> r->shift;
1273}
1274
1275void psched_ratecfg_precompute(struct psched_ratecfg *r,
1276 const struct tc_ratespec *conf,
1277 u64 rate64);
1278
1279static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1280 const struct psched_ratecfg *r)
1281{
1282 memset(res, 0, sizeof(*res));
1283
1284 /* legacy struct tc_ratespec has a 32bit @rate field
1285 * Qdisc using 64bit rate should add new attributes
1286 * in order to maintain compatibility.
1287 */
1288 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1289
1290 res->overhead = r->overhead;
1291 res->mpu = r->mpu;
1292 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1293}
1294
1295struct psched_pktrate {
1296 u64 rate_pkts_ps; /* packets per second */
1297 u32 mult;
1298 u8 shift;
1299};
1300
1301static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1302 unsigned int pkt_num)
1303{
1304 return ((u64)pkt_num * r->mult) >> r->shift;
1305}
1306
1307void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1308
1309/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1310 * The fast path only needs to access filter list and to update stats
1311 */
1312struct mini_Qdisc {
1313 struct tcf_proto *filter_list;
1314 struct tcf_block *block;
1315 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1316 struct gnet_stats_queue __percpu *cpu_qstats;
1317 unsigned long rcu_state;
1318};
1319
1320static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1321 const struct sk_buff *skb)
1322{
1323 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1324}
1325
1326static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1327{
1328 this_cpu_inc(miniq->cpu_qstats->drops);
1329}
1330
1331struct mini_Qdisc_pair {
1332 struct mini_Qdisc miniq1;
1333 struct mini_Qdisc miniq2;
1334 struct mini_Qdisc __rcu **p_miniq;
1335};
1336
1337void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1338 struct tcf_proto *tp_head);
1339void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1340 struct mini_Qdisc __rcu **p_miniq);
1341void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1342 struct tcf_block *block);
1343
1344void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1345
1346int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1347
1348/* Make sure qdisc is no longer in SCHED state. */
1349static inline void qdisc_synchronize(const struct Qdisc *q)
1350{
1351 while (test_bit(__QDISC_STATE_SCHED, &q->state))
1352 msleep(1);
1353}
1354
1355#endif