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
23struct Qdisc_ops;
24struct qdisc_walker;
25struct tcf_walker;
26struct module;
27struct bpf_flow_keys;
28
29struct qdisc_rate_table {
30 struct tc_ratespec rate;
31 u32 data[256];
32 struct qdisc_rate_table *next;
33 int refcnt;
34};
35
36enum qdisc_state_t {
37 __QDISC_STATE_SCHED,
38 __QDISC_STATE_DEACTIVATED,
39 __QDISC_STATE_MISSED,
40 __QDISC_STATE_DRAINING,
41};
42
43enum qdisc_state2_t {
44 /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly.
45 * Use qdisc_run_begin/end() or qdisc_is_running() instead.
46 */
47 __QDISC_STATE2_RUNNING,
48};
49
50#define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED)
51#define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING)
52
53#define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \
54 QDISC_STATE_DRAINING)
55
56struct qdisc_size_table {
57 struct rcu_head rcu;
58 struct list_head list;
59 struct tc_sizespec szopts;
60 int refcnt;
61 u16 data[];
62};
63
64/* similar to sk_buff_head, but skb->prev pointer is undefined. */
65struct qdisc_skb_head {
66 struct sk_buff *head;
67 struct sk_buff *tail;
68 __u32 qlen;
69 spinlock_t lock;
70};
71
72struct Qdisc {
73 int (*enqueue)(struct sk_buff *skb,
74 struct Qdisc *sch,
75 struct sk_buff **to_free);
76 struct sk_buff * (*dequeue)(struct Qdisc *sch);
77 unsigned int flags;
78#define TCQ_F_BUILTIN 1
79#define TCQ_F_INGRESS 2
80#define TCQ_F_CAN_BYPASS 4
81#define TCQ_F_MQROOT 8
82#define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
83 * q->dev_queue : It can test
84 * netif_xmit_frozen_or_stopped() before
85 * dequeueing next packet.
86 * Its true for MQ/MQPRIO slaves, or non
87 * multiqueue device.
88 */
89#define TCQ_F_WARN_NONWC (1 << 16)
90#define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
91#define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
92 * qdisc_tree_decrease_qlen() should stop.
93 */
94#define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
95#define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
96#define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
97 u32 limit;
98 const struct Qdisc_ops *ops;
99 struct qdisc_size_table __rcu *stab;
100 struct hlist_node hash;
101 u32 handle;
102 u32 parent;
103
104 struct netdev_queue *dev_queue;
105
106 struct net_rate_estimator __rcu *rate_est;
107 struct gnet_stats_basic_sync __percpu *cpu_bstats;
108 struct gnet_stats_queue __percpu *cpu_qstats;
109 int pad;
110 refcount_t refcnt;
111
112 /*
113 * For performance sake on SMP, we put highly modified fields at the end
114 */
115 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
116 struct qdisc_skb_head q;
117 struct gnet_stats_basic_sync bstats;
118 struct gnet_stats_queue qstats;
119 unsigned long state;
120 unsigned long state2; /* must be written under qdisc spinlock */
121 struct Qdisc *next_sched;
122 struct sk_buff_head skb_bad_txq;
123
124 spinlock_t busylock ____cacheline_aligned_in_smp;
125 spinlock_t seqlock;
126
127 struct rcu_head rcu;
128 netdevice_tracker dev_tracker;
129 /* private data */
130 long privdata[] ____cacheline_aligned;
131};
132
133static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
134{
135 if (qdisc->flags & TCQ_F_BUILTIN)
136 return;
137 refcount_inc(&qdisc->refcnt);
138}
139
140/* Intended to be used by unlocked users, when concurrent qdisc release is
141 * possible.
142 */
143
144static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
145{
146 if (qdisc->flags & TCQ_F_BUILTIN)
147 return qdisc;
148 if (refcount_inc_not_zero(&qdisc->refcnt))
149 return qdisc;
150 return NULL;
151}
152
153/* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
154 * root_lock section, or provide their own memory barriers -- ordering
155 * against qdisc_run_begin/end() atomic bit operations.
156 */
157static inline bool qdisc_is_running(struct Qdisc *qdisc)
158{
159 if (qdisc->flags & TCQ_F_NOLOCK)
160 return spin_is_locked(&qdisc->seqlock);
161 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
162}
163
164static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
165{
166 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
167}
168
169static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
170{
171 return q->flags & TCQ_F_CPUSTATS;
172}
173
174static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
175{
176 if (qdisc_is_percpu_stats(qdisc))
177 return nolock_qdisc_is_empty(qdisc);
178 return !READ_ONCE(qdisc->q.qlen);
179}
180
181/* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
182 * the qdisc root lock acquired.
183 */
184static inline bool qdisc_run_begin(struct Qdisc *qdisc)
185{
186 if (qdisc->flags & TCQ_F_NOLOCK) {
187 if (spin_trylock(&qdisc->seqlock))
188 return true;
189
190 /* No need to insist if the MISSED flag was already set.
191 * Note that test_and_set_bit() also gives us memory ordering
192 * guarantees wrt potential earlier enqueue() and below
193 * spin_trylock(), both of which are necessary to prevent races
194 */
195 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
196 return false;
197
198 /* Try to take the lock again to make sure that we will either
199 * grab it or the CPU that still has it will see MISSED set
200 * when testing it in qdisc_run_end()
201 */
202 return spin_trylock(&qdisc->seqlock);
203 }
204 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
205}
206
207static inline void qdisc_run_end(struct Qdisc *qdisc)
208{
209 if (qdisc->flags & TCQ_F_NOLOCK) {
210 spin_unlock(&qdisc->seqlock);
211
212 /* spin_unlock() only has store-release semantic. The unlock
213 * and test_bit() ordering is a store-load ordering, so a full
214 * memory barrier is needed here.
215 */
216 smp_mb();
217
218 if (unlikely(test_bit(__QDISC_STATE_MISSED,
219 &qdisc->state)))
220 __netif_schedule(qdisc);
221 } else {
222 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
223 }
224}
225
226static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
227{
228 return qdisc->flags & TCQ_F_ONETXQUEUE;
229}
230
231static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
232{
233#ifdef CONFIG_BQL
234 /* Non-BQL migrated drivers will return 0, too. */
235 return dql_avail(&txq->dql);
236#else
237 return 0;
238#endif
239}
240
241struct Qdisc_class_ops {
242 unsigned int flags;
243 /* Child qdisc manipulation */
244 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
245 int (*graft)(struct Qdisc *, unsigned long cl,
246 struct Qdisc *, struct Qdisc **,
247 struct netlink_ext_ack *extack);
248 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
249 void (*qlen_notify)(struct Qdisc *, unsigned long);
250
251 /* Class manipulation routines */
252 unsigned long (*find)(struct Qdisc *, u32 classid);
253 int (*change)(struct Qdisc *, u32, u32,
254 struct nlattr **, unsigned long *,
255 struct netlink_ext_ack *);
256 int (*delete)(struct Qdisc *, unsigned long,
257 struct netlink_ext_ack *);
258 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
259
260 /* Filter manipulation */
261 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
262 unsigned long arg,
263 struct netlink_ext_ack *extack);
264 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
265 u32 classid);
266 void (*unbind_tcf)(struct Qdisc *, unsigned long);
267
268 /* rtnetlink specific */
269 int (*dump)(struct Qdisc *, unsigned long,
270 struct sk_buff *skb, struct tcmsg*);
271 int (*dump_stats)(struct Qdisc *, unsigned long,
272 struct gnet_dump *);
273};
274
275/* Qdisc_class_ops flag values */
276
277/* Implements API that doesn't require rtnl lock */
278enum qdisc_class_ops_flags {
279 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
280};
281
282struct Qdisc_ops {
283 struct Qdisc_ops *next;
284 const struct Qdisc_class_ops *cl_ops;
285 char id[IFNAMSIZ];
286 int priv_size;
287 unsigned int static_flags;
288
289 int (*enqueue)(struct sk_buff *skb,
290 struct Qdisc *sch,
291 struct sk_buff **to_free);
292 struct sk_buff * (*dequeue)(struct Qdisc *);
293 struct sk_buff * (*peek)(struct Qdisc *);
294
295 int (*init)(struct Qdisc *sch, struct nlattr *arg,
296 struct netlink_ext_ack *extack);
297 void (*reset)(struct Qdisc *);
298 void (*destroy)(struct Qdisc *);
299 int (*change)(struct Qdisc *sch,
300 struct nlattr *arg,
301 struct netlink_ext_ack *extack);
302 void (*attach)(struct Qdisc *sch);
303 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
304 void (*change_real_num_tx)(struct Qdisc *sch,
305 unsigned int new_real_tx);
306
307 int (*dump)(struct Qdisc *, struct sk_buff *);
308 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
309
310 void (*ingress_block_set)(struct Qdisc *sch,
311 u32 block_index);
312 void (*egress_block_set)(struct Qdisc *sch,
313 u32 block_index);
314 u32 (*ingress_block_get)(struct Qdisc *sch);
315 u32 (*egress_block_get)(struct Qdisc *sch);
316
317 struct module *owner;
318};
319
320
321struct tcf_result {
322 union {
323 struct {
324 unsigned long class;
325 u32 classid;
326 };
327 const struct tcf_proto *goto_tp;
328
329 };
330};
331
332struct tcf_chain;
333
334struct tcf_proto_ops {
335 struct list_head head;
336 char kind[IFNAMSIZ];
337
338 int (*classify)(struct sk_buff *,
339 const struct tcf_proto *,
340 struct tcf_result *);
341 int (*init)(struct tcf_proto*);
342 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
343 struct netlink_ext_ack *extack);
344
345 void* (*get)(struct tcf_proto*, u32 handle);
346 void (*put)(struct tcf_proto *tp, void *f);
347 int (*change)(struct net *net, struct sk_buff *,
348 struct tcf_proto*, unsigned long,
349 u32 handle, struct nlattr **,
350 void **, u32,
351 struct netlink_ext_ack *);
352 int (*delete)(struct tcf_proto *tp, void *arg,
353 bool *last, bool rtnl_held,
354 struct netlink_ext_ack *);
355 bool (*delete_empty)(struct tcf_proto *tp);
356 void (*walk)(struct tcf_proto *tp,
357 struct tcf_walker *arg, bool rtnl_held);
358 int (*reoffload)(struct tcf_proto *tp, bool add,
359 flow_setup_cb_t *cb, void *cb_priv,
360 struct netlink_ext_ack *extack);
361 void (*hw_add)(struct tcf_proto *tp,
362 void *type_data);
363 void (*hw_del)(struct tcf_proto *tp,
364 void *type_data);
365 void (*bind_class)(void *, u32, unsigned long,
366 void *, unsigned long);
367 void * (*tmplt_create)(struct net *net,
368 struct tcf_chain *chain,
369 struct nlattr **tca,
370 struct netlink_ext_ack *extack);
371 void (*tmplt_destroy)(void *tmplt_priv);
372
373 /* rtnetlink specific */
374 int (*dump)(struct net*, struct tcf_proto*, void *,
375 struct sk_buff *skb, struct tcmsg*,
376 bool);
377 int (*terse_dump)(struct net *net,
378 struct tcf_proto *tp, void *fh,
379 struct sk_buff *skb,
380 struct tcmsg *t, bool rtnl_held);
381 int (*tmplt_dump)(struct sk_buff *skb,
382 struct net *net,
383 void *tmplt_priv);
384
385 struct module *owner;
386 int flags;
387};
388
389/* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
390 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
391 * conditions can occur when filters are inserted/deleted simultaneously.
392 */
393enum tcf_proto_ops_flags {
394 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
395};
396
397struct tcf_proto {
398 /* Fast access part */
399 struct tcf_proto __rcu *next;
400 void __rcu *root;
401
402 /* called under RCU BH lock*/
403 int (*classify)(struct sk_buff *,
404 const struct tcf_proto *,
405 struct tcf_result *);
406 __be16 protocol;
407
408 /* All the rest */
409 u32 prio;
410 void *data;
411 const struct tcf_proto_ops *ops;
412 struct tcf_chain *chain;
413 /* Lock protects tcf_proto shared state and can be used by unlocked
414 * classifiers to protect their private data.
415 */
416 spinlock_t lock;
417 bool deleting;
418 refcount_t refcnt;
419 struct rcu_head rcu;
420 struct hlist_node destroy_ht_node;
421};
422
423struct qdisc_skb_cb {
424 struct {
425 unsigned int pkt_len;
426 u16 slave_dev_queue_mapping;
427 u16 tc_classid;
428 };
429#define QDISC_CB_PRIV_LEN 20
430 unsigned char data[QDISC_CB_PRIV_LEN];
431};
432
433typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
434
435struct tcf_chain {
436 /* Protects filter_chain. */
437 struct mutex filter_chain_lock;
438 struct tcf_proto __rcu *filter_chain;
439 struct list_head list;
440 struct tcf_block *block;
441 u32 index; /* chain index */
442 unsigned int refcnt;
443 unsigned int action_refcnt;
444 bool explicitly_created;
445 bool flushing;
446 const struct tcf_proto_ops *tmplt_ops;
447 void *tmplt_priv;
448 struct rcu_head rcu;
449};
450
451struct tcf_block {
452 /* Lock protects tcf_block and lifetime-management data of chains
453 * attached to the block (refcnt, action_refcnt, explicitly_created).
454 */
455 struct mutex lock;
456 struct list_head chain_list;
457 u32 index; /* block index for shared blocks */
458 u32 classid; /* which class this block belongs to */
459 refcount_t refcnt;
460 struct net *net;
461 struct Qdisc *q;
462 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
463 struct flow_block flow_block;
464 struct list_head owner_list;
465 bool keep_dst;
466 atomic_t offloadcnt; /* Number of oddloaded filters */
467 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
468 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
469 struct {
470 struct tcf_chain *chain;
471 struct list_head filter_chain_list;
472 } chain0;
473 struct rcu_head rcu;
474 DECLARE_HASHTABLE(proto_destroy_ht, 7);
475 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
476};
477
478static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
479{
480 return lockdep_is_held(&chain->filter_chain_lock);
481}
482
483static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
484{
485 return lockdep_is_held(&tp->lock);
486}
487
488#define tcf_chain_dereference(p, chain) \
489 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
490
491#define tcf_proto_dereference(p, tp) \
492 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
493
494static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
495{
496 struct qdisc_skb_cb *qcb;
497
498 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
499 BUILD_BUG_ON(sizeof(qcb->data) < sz);
500}
501
502static inline int qdisc_qlen(const struct Qdisc *q)
503{
504 return q->q.qlen;
505}
506
507static inline int qdisc_qlen_sum(const struct Qdisc *q)
508{
509 __u32 qlen = q->qstats.qlen;
510 int i;
511
512 if (qdisc_is_percpu_stats(q)) {
513 for_each_possible_cpu(i)
514 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
515 } else {
516 qlen += q->q.qlen;
517 }
518
519 return qlen;
520}
521
522static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
523{
524 return (struct qdisc_skb_cb *)skb->cb;
525}
526
527static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
528{
529 return &qdisc->q.lock;
530}
531
532static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
533{
534 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
535
536 return q;
537}
538
539static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
540{
541 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
542}
543
544static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
545{
546 return qdisc->dev_queue->qdisc_sleeping;
547}
548
549static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
550{
551 struct Qdisc *root = qdisc_root_sleeping(qdisc);
552
553 ASSERT_RTNL();
554 return qdisc_lock(root);
555}
556
557static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
558{
559 return qdisc->dev_queue->dev;
560}
561
562static inline void sch_tree_lock(struct Qdisc *q)
563{
564 if (q->flags & TCQ_F_MQROOT)
565 spin_lock_bh(qdisc_lock(q));
566 else
567 spin_lock_bh(qdisc_root_sleeping_lock(q));
568}
569
570static inline void sch_tree_unlock(struct Qdisc *q)
571{
572 if (q->flags & TCQ_F_MQROOT)
573 spin_unlock_bh(qdisc_lock(q));
574 else
575 spin_unlock_bh(qdisc_root_sleeping_lock(q));
576}
577
578extern struct Qdisc noop_qdisc;
579extern struct Qdisc_ops noop_qdisc_ops;
580extern struct Qdisc_ops pfifo_fast_ops;
581extern struct Qdisc_ops mq_qdisc_ops;
582extern struct Qdisc_ops noqueue_qdisc_ops;
583extern const struct Qdisc_ops *default_qdisc_ops;
584static inline const struct Qdisc_ops *
585get_default_qdisc_ops(const struct net_device *dev, int ntx)
586{
587 return ntx < dev->real_num_tx_queues ?
588 default_qdisc_ops : &pfifo_fast_ops;
589}
590
591struct Qdisc_class_common {
592 u32 classid;
593 struct hlist_node hnode;
594};
595
596struct Qdisc_class_hash {
597 struct hlist_head *hash;
598 unsigned int hashsize;
599 unsigned int hashmask;
600 unsigned int hashelems;
601};
602
603static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
604{
605 id ^= id >> 8;
606 id ^= id >> 4;
607 return id & mask;
608}
609
610static inline struct Qdisc_class_common *
611qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
612{
613 struct Qdisc_class_common *cl;
614 unsigned int h;
615
616 if (!id)
617 return NULL;
618
619 h = qdisc_class_hash(id, hash->hashmask);
620 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
621 if (cl->classid == id)
622 return cl;
623 }
624 return NULL;
625}
626
627static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
628{
629 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
630
631 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
632}
633
634int qdisc_class_hash_init(struct Qdisc_class_hash *);
635void qdisc_class_hash_insert(struct Qdisc_class_hash *,
636 struct Qdisc_class_common *);
637void qdisc_class_hash_remove(struct Qdisc_class_hash *,
638 struct Qdisc_class_common *);
639void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
640void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
641
642int dev_qdisc_change_tx_queue_len(struct net_device *dev);
643void dev_qdisc_change_real_num_tx(struct net_device *dev,
644 unsigned int new_real_tx);
645void dev_init_scheduler(struct net_device *dev);
646void dev_shutdown(struct net_device *dev);
647void dev_activate(struct net_device *dev);
648void dev_deactivate(struct net_device *dev);
649void dev_deactivate_many(struct list_head *head);
650struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
651 struct Qdisc *qdisc);
652void qdisc_reset(struct Qdisc *qdisc);
653void qdisc_put(struct Qdisc *qdisc);
654void qdisc_put_unlocked(struct Qdisc *qdisc);
655void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
656#ifdef CONFIG_NET_SCHED
657int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
658 void *type_data);
659void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
660 struct Qdisc *new, struct Qdisc *old,
661 enum tc_setup_type type, void *type_data,
662 struct netlink_ext_ack *extack);
663#else
664static inline int
665qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
666 void *type_data)
667{
668 q->flags &= ~TCQ_F_OFFLOADED;
669 return 0;
670}
671
672static inline void
673qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
674 struct Qdisc *new, struct Qdisc *old,
675 enum tc_setup_type type, void *type_data,
676 struct netlink_ext_ack *extack)
677{
678}
679#endif
680void qdisc_offload_query_caps(struct net_device *dev,
681 enum tc_setup_type type,
682 void *caps, size_t caps_len);
683struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
684 const struct Qdisc_ops *ops,
685 struct netlink_ext_ack *extack);
686void qdisc_free(struct Qdisc *qdisc);
687struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
688 const struct Qdisc_ops *ops, u32 parentid,
689 struct netlink_ext_ack *extack);
690void __qdisc_calculate_pkt_len(struct sk_buff *skb,
691 const struct qdisc_size_table *stab);
692int skb_do_redirect(struct sk_buff *);
693
694static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
695{
696#ifdef CONFIG_NET_CLS_ACT
697 return skb->tc_at_ingress;
698#else
699 return false;
700#endif
701}
702
703static inline bool skb_skip_tc_classify(struct sk_buff *skb)
704{
705#ifdef CONFIG_NET_CLS_ACT
706 if (skb->tc_skip_classify) {
707 skb->tc_skip_classify = 0;
708 return true;
709 }
710#endif
711 return false;
712}
713
714/* Reset all TX qdiscs greater than index of a device. */
715static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
716{
717 struct Qdisc *qdisc;
718
719 for (; i < dev->num_tx_queues; i++) {
720 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
721 if (qdisc) {
722 spin_lock_bh(qdisc_lock(qdisc));
723 qdisc_reset(qdisc);
724 spin_unlock_bh(qdisc_lock(qdisc));
725 }
726 }
727}
728
729/* Are all TX queues of the device empty? */
730static inline bool qdisc_all_tx_empty(const struct net_device *dev)
731{
732 unsigned int i;
733
734 rcu_read_lock();
735 for (i = 0; i < dev->num_tx_queues; i++) {
736 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
737 const struct Qdisc *q = rcu_dereference(txq->qdisc);
738
739 if (!qdisc_is_empty(q)) {
740 rcu_read_unlock();
741 return false;
742 }
743 }
744 rcu_read_unlock();
745 return true;
746}
747
748/* Are any of the TX qdiscs changing? */
749static inline bool qdisc_tx_changing(const struct net_device *dev)
750{
751 unsigned int i;
752
753 for (i = 0; i < dev->num_tx_queues; i++) {
754 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
755 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
756 return true;
757 }
758 return false;
759}
760
761/* Is the device using the noop qdisc on all queues? */
762static inline bool qdisc_tx_is_noop(const struct net_device *dev)
763{
764 unsigned int i;
765
766 for (i = 0; i < dev->num_tx_queues; i++) {
767 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
768 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
769 return false;
770 }
771 return true;
772}
773
774static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
775{
776 return qdisc_skb_cb(skb)->pkt_len;
777}
778
779/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
780enum net_xmit_qdisc_t {
781 __NET_XMIT_STOLEN = 0x00010000,
782 __NET_XMIT_BYPASS = 0x00020000,
783};
784
785#ifdef CONFIG_NET_CLS_ACT
786#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
787#else
788#define net_xmit_drop_count(e) (1)
789#endif
790
791static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
792 const struct Qdisc *sch)
793{
794#ifdef CONFIG_NET_SCHED
795 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
796
797 if (stab)
798 __qdisc_calculate_pkt_len(skb, stab);
799#endif
800}
801
802static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
803 struct sk_buff **to_free)
804{
805 qdisc_calculate_pkt_len(skb, sch);
806 return sch->enqueue(skb, sch, to_free);
807}
808
809static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
810 __u64 bytes, __u32 packets)
811{
812 u64_stats_update_begin(&bstats->syncp);
813 u64_stats_add(&bstats->bytes, bytes);
814 u64_stats_add(&bstats->packets, packets);
815 u64_stats_update_end(&bstats->syncp);
816}
817
818static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
819 const struct sk_buff *skb)
820{
821 _bstats_update(bstats,
822 qdisc_pkt_len(skb),
823 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
824}
825
826static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
827 const struct sk_buff *skb)
828{
829 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
830}
831
832static inline void qdisc_bstats_update(struct Qdisc *sch,
833 const struct sk_buff *skb)
834{
835 bstats_update(&sch->bstats, skb);
836}
837
838static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
839 const struct sk_buff *skb)
840{
841 sch->qstats.backlog -= qdisc_pkt_len(skb);
842}
843
844static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
845 const struct sk_buff *skb)
846{
847 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
848}
849
850static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
851 const struct sk_buff *skb)
852{
853 sch->qstats.backlog += qdisc_pkt_len(skb);
854}
855
856static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
857 const struct sk_buff *skb)
858{
859 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
860}
861
862static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
863{
864 this_cpu_inc(sch->cpu_qstats->qlen);
865}
866
867static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
868{
869 this_cpu_dec(sch->cpu_qstats->qlen);
870}
871
872static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
873{
874 this_cpu_inc(sch->cpu_qstats->requeues);
875}
876
877static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
878{
879 sch->qstats.drops += count;
880}
881
882static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
883{
884 qstats->drops++;
885}
886
887static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
888{
889 qstats->overlimits++;
890}
891
892static inline void qdisc_qstats_drop(struct Qdisc *sch)
893{
894 qstats_drop_inc(&sch->qstats);
895}
896
897static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
898{
899 this_cpu_inc(sch->cpu_qstats->drops);
900}
901
902static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
903{
904 sch->qstats.overlimits++;
905}
906
907static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
908{
909 __u32 qlen = qdisc_qlen_sum(sch);
910
911 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
912}
913
914static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
915 __u32 *backlog)
916{
917 struct gnet_stats_queue qstats = { 0 };
918
919 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
920 *qlen = qstats.qlen + qdisc_qlen(sch);
921 *backlog = qstats.backlog;
922}
923
924static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
925{
926 __u32 qlen, backlog;
927
928 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
929 qdisc_tree_reduce_backlog(sch, qlen, backlog);
930}
931
932static inline void qdisc_purge_queue(struct Qdisc *sch)
933{
934 __u32 qlen, backlog;
935
936 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
937 qdisc_reset(sch);
938 qdisc_tree_reduce_backlog(sch, qlen, backlog);
939}
940
941static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
942 struct qdisc_skb_head *qh)
943{
944 struct sk_buff *last = qh->tail;
945
946 if (last) {
947 skb->next = NULL;
948 last->next = skb;
949 qh->tail = skb;
950 } else {
951 qh->tail = skb;
952 qh->head = skb;
953 }
954 qh->qlen++;
955}
956
957static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
958{
959 __qdisc_enqueue_tail(skb, &sch->q);
960 qdisc_qstats_backlog_inc(sch, skb);
961 return NET_XMIT_SUCCESS;
962}
963
964static inline void __qdisc_enqueue_head(struct sk_buff *skb,
965 struct qdisc_skb_head *qh)
966{
967 skb->next = qh->head;
968
969 if (!qh->head)
970 qh->tail = skb;
971 qh->head = skb;
972 qh->qlen++;
973}
974
975static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
976{
977 struct sk_buff *skb = qh->head;
978
979 if (likely(skb != NULL)) {
980 qh->head = skb->next;
981 qh->qlen--;
982 if (qh->head == NULL)
983 qh->tail = NULL;
984 skb->next = NULL;
985 }
986
987 return skb;
988}
989
990static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
991{
992 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
993
994 if (likely(skb != NULL)) {
995 qdisc_qstats_backlog_dec(sch, skb);
996 qdisc_bstats_update(sch, skb);
997 }
998
999 return skb;
1000}
1001
1002/* Instead of calling kfree_skb() while root qdisc lock is held,
1003 * queue the skb for future freeing at end of __dev_xmit_skb()
1004 */
1005static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1006{
1007 skb->next = *to_free;
1008 *to_free = skb;
1009}
1010
1011static inline void __qdisc_drop_all(struct sk_buff *skb,
1012 struct sk_buff **to_free)
1013{
1014 if (skb->prev)
1015 skb->prev->next = *to_free;
1016 else
1017 skb->next = *to_free;
1018 *to_free = skb;
1019}
1020
1021static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1022 struct qdisc_skb_head *qh,
1023 struct sk_buff **to_free)
1024{
1025 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1026
1027 if (likely(skb != NULL)) {
1028 unsigned int len = qdisc_pkt_len(skb);
1029
1030 qdisc_qstats_backlog_dec(sch, skb);
1031 __qdisc_drop(skb, to_free);
1032 return len;
1033 }
1034
1035 return 0;
1036}
1037
1038static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1039{
1040 const struct qdisc_skb_head *qh = &sch->q;
1041
1042 return qh->head;
1043}
1044
1045/* generic pseudo peek method for non-work-conserving qdisc */
1046static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1047{
1048 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1049
1050 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1051 if (!skb) {
1052 skb = sch->dequeue(sch);
1053
1054 if (skb) {
1055 __skb_queue_head(&sch->gso_skb, skb);
1056 /* it's still part of the queue */
1057 qdisc_qstats_backlog_inc(sch, skb);
1058 sch->q.qlen++;
1059 }
1060 }
1061
1062 return skb;
1063}
1064
1065static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1066 struct sk_buff *skb)
1067{
1068 if (qdisc_is_percpu_stats(sch)) {
1069 qdisc_qstats_cpu_backlog_dec(sch, skb);
1070 qdisc_bstats_cpu_update(sch, skb);
1071 qdisc_qstats_cpu_qlen_dec(sch);
1072 } else {
1073 qdisc_qstats_backlog_dec(sch, skb);
1074 qdisc_bstats_update(sch, skb);
1075 sch->q.qlen--;
1076 }
1077}
1078
1079static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1080 unsigned int pkt_len)
1081{
1082 if (qdisc_is_percpu_stats(sch)) {
1083 qdisc_qstats_cpu_qlen_inc(sch);
1084 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1085 } else {
1086 sch->qstats.backlog += pkt_len;
1087 sch->q.qlen++;
1088 }
1089}
1090
1091/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1092static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1093{
1094 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1095
1096 if (skb) {
1097 skb = __skb_dequeue(&sch->gso_skb);
1098 if (qdisc_is_percpu_stats(sch)) {
1099 qdisc_qstats_cpu_backlog_dec(sch, skb);
1100 qdisc_qstats_cpu_qlen_dec(sch);
1101 } else {
1102 qdisc_qstats_backlog_dec(sch, skb);
1103 sch->q.qlen--;
1104 }
1105 } else {
1106 skb = sch->dequeue(sch);
1107 }
1108
1109 return skb;
1110}
1111
1112static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1113{
1114 /*
1115 * We do not know the backlog in bytes of this list, it
1116 * is up to the caller to correct it
1117 */
1118 ASSERT_RTNL();
1119 if (qh->qlen) {
1120 rtnl_kfree_skbs(qh->head, qh->tail);
1121
1122 qh->head = NULL;
1123 qh->tail = NULL;
1124 qh->qlen = 0;
1125 }
1126}
1127
1128static inline void qdisc_reset_queue(struct Qdisc *sch)
1129{
1130 __qdisc_reset_queue(&sch->q);
1131}
1132
1133static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1134 struct Qdisc **pold)
1135{
1136 struct Qdisc *old;
1137
1138 sch_tree_lock(sch);
1139 old = *pold;
1140 *pold = new;
1141 if (old != NULL)
1142 qdisc_purge_queue(old);
1143 sch_tree_unlock(sch);
1144
1145 return old;
1146}
1147
1148static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1149{
1150 rtnl_kfree_skbs(skb, skb);
1151 qdisc_qstats_drop(sch);
1152}
1153
1154static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1155 struct sk_buff **to_free)
1156{
1157 __qdisc_drop(skb, to_free);
1158 qdisc_qstats_cpu_drop(sch);
1159
1160 return NET_XMIT_DROP;
1161}
1162
1163static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1164 struct sk_buff **to_free)
1165{
1166 __qdisc_drop(skb, to_free);
1167 qdisc_qstats_drop(sch);
1168
1169 return NET_XMIT_DROP;
1170}
1171
1172static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1173 struct sk_buff **to_free)
1174{
1175 __qdisc_drop_all(skb, to_free);
1176 qdisc_qstats_drop(sch);
1177
1178 return NET_XMIT_DROP;
1179}
1180
1181/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1182 long it will take to send a packet given its size.
1183 */
1184static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1185{
1186 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1187 if (slot < 0)
1188 slot = 0;
1189 slot >>= rtab->rate.cell_log;
1190 if (slot > 255)
1191 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1192 return rtab->data[slot];
1193}
1194
1195struct psched_ratecfg {
1196 u64 rate_bytes_ps; /* bytes per second */
1197 u32 mult;
1198 u16 overhead;
1199 u16 mpu;
1200 u8 linklayer;
1201 u8 shift;
1202};
1203
1204static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1205 unsigned int len)
1206{
1207 len += r->overhead;
1208
1209 if (len < r->mpu)
1210 len = r->mpu;
1211
1212 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1213 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1214
1215 return ((u64)len * r->mult) >> r->shift;
1216}
1217
1218void psched_ratecfg_precompute(struct psched_ratecfg *r,
1219 const struct tc_ratespec *conf,
1220 u64 rate64);
1221
1222static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1223 const struct psched_ratecfg *r)
1224{
1225 memset(res, 0, sizeof(*res));
1226
1227 /* legacy struct tc_ratespec has a 32bit @rate field
1228 * Qdisc using 64bit rate should add new attributes
1229 * in order to maintain compatibility.
1230 */
1231 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1232
1233 res->overhead = r->overhead;
1234 res->mpu = r->mpu;
1235 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1236}
1237
1238struct psched_pktrate {
1239 u64 rate_pkts_ps; /* packets per second */
1240 u32 mult;
1241 u8 shift;
1242};
1243
1244static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1245 unsigned int pkt_num)
1246{
1247 return ((u64)pkt_num * r->mult) >> r->shift;
1248}
1249
1250void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1251
1252/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1253 * The fast path only needs to access filter list and to update stats
1254 */
1255struct mini_Qdisc {
1256 struct tcf_proto *filter_list;
1257 struct tcf_block *block;
1258 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1259 struct gnet_stats_queue __percpu *cpu_qstats;
1260 unsigned long rcu_state;
1261};
1262
1263static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1264 const struct sk_buff *skb)
1265{
1266 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1267}
1268
1269static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1270{
1271 this_cpu_inc(miniq->cpu_qstats->drops);
1272}
1273
1274struct mini_Qdisc_pair {
1275 struct mini_Qdisc miniq1;
1276 struct mini_Qdisc miniq2;
1277 struct mini_Qdisc __rcu **p_miniq;
1278};
1279
1280void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1281 struct tcf_proto *tp_head);
1282void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1283 struct mini_Qdisc __rcu **p_miniq);
1284void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1285 struct tcf_block *block);
1286
1287void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1288
1289int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1290
1291/* Make sure qdisc is no longer in SCHED state. */
1292static inline void qdisc_synchronize(const struct Qdisc *q)
1293{
1294 while (test_bit(__QDISC_STATE_SCHED, &q->state))
1295 msleep(1);
1296}
1297
1298#endif