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