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 /* used in the skb_tc_reinsert function */
330 struct {
331 bool ingress;
332 struct gnet_stats_queue *qstats;
333 };
334 };
335};
336
337struct tcf_chain;
338
339struct tcf_proto_ops {
340 struct list_head head;
341 char kind[IFNAMSIZ];
342
343 int (*classify)(struct sk_buff *,
344 const struct tcf_proto *,
345 struct tcf_result *);
346 int (*init)(struct tcf_proto*);
347 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
348 struct netlink_ext_ack *extack);
349
350 void* (*get)(struct tcf_proto*, u32 handle);
351 void (*put)(struct tcf_proto *tp, void *f);
352 int (*change)(struct net *net, struct sk_buff *,
353 struct tcf_proto*, unsigned long,
354 u32 handle, struct nlattr **,
355 void **, u32,
356 struct netlink_ext_ack *);
357 int (*delete)(struct tcf_proto *tp, void *arg,
358 bool *last, bool rtnl_held,
359 struct netlink_ext_ack *);
360 bool (*delete_empty)(struct tcf_proto *tp);
361 void (*walk)(struct tcf_proto *tp,
362 struct tcf_walker *arg, bool rtnl_held);
363 int (*reoffload)(struct tcf_proto *tp, bool add,
364 flow_setup_cb_t *cb, void *cb_priv,
365 struct netlink_ext_ack *extack);
366 void (*hw_add)(struct tcf_proto *tp,
367 void *type_data);
368 void (*hw_del)(struct tcf_proto *tp,
369 void *type_data);
370 void (*bind_class)(void *, u32, unsigned long,
371 void *, unsigned long);
372 void * (*tmplt_create)(struct net *net,
373 struct tcf_chain *chain,
374 struct nlattr **tca,
375 struct netlink_ext_ack *extack);
376 void (*tmplt_destroy)(void *tmplt_priv);
377
378 /* rtnetlink specific */
379 int (*dump)(struct net*, struct tcf_proto*, void *,
380 struct sk_buff *skb, struct tcmsg*,
381 bool);
382 int (*terse_dump)(struct net *net,
383 struct tcf_proto *tp, void *fh,
384 struct sk_buff *skb,
385 struct tcmsg *t, bool rtnl_held);
386 int (*tmplt_dump)(struct sk_buff *skb,
387 struct net *net,
388 void *tmplt_priv);
389
390 struct module *owner;
391 int flags;
392};
393
394/* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
395 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
396 * conditions can occur when filters are inserted/deleted simultaneously.
397 */
398enum tcf_proto_ops_flags {
399 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
400};
401
402struct tcf_proto {
403 /* Fast access part */
404 struct tcf_proto __rcu *next;
405 void __rcu *root;
406
407 /* called under RCU BH lock*/
408 int (*classify)(struct sk_buff *,
409 const struct tcf_proto *,
410 struct tcf_result *);
411 __be16 protocol;
412
413 /* All the rest */
414 u32 prio;
415 void *data;
416 const struct tcf_proto_ops *ops;
417 struct tcf_chain *chain;
418 /* Lock protects tcf_proto shared state and can be used by unlocked
419 * classifiers to protect their private data.
420 */
421 spinlock_t lock;
422 bool deleting;
423 refcount_t refcnt;
424 struct rcu_head rcu;
425 struct hlist_node destroy_ht_node;
426};
427
428struct qdisc_skb_cb {
429 struct {
430 unsigned int pkt_len;
431 u16 slave_dev_queue_mapping;
432 u16 tc_classid;
433 };
434#define QDISC_CB_PRIV_LEN 20
435 unsigned char data[QDISC_CB_PRIV_LEN];
436};
437
438typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
439
440struct tcf_chain {
441 /* Protects filter_chain. */
442 struct mutex filter_chain_lock;
443 struct tcf_proto __rcu *filter_chain;
444 struct list_head list;
445 struct tcf_block *block;
446 u32 index; /* chain index */
447 unsigned int refcnt;
448 unsigned int action_refcnt;
449 bool explicitly_created;
450 bool flushing;
451 const struct tcf_proto_ops *tmplt_ops;
452 void *tmplt_priv;
453 struct rcu_head rcu;
454};
455
456struct tcf_block {
457 /* Lock protects tcf_block and lifetime-management data of chains
458 * attached to the block (refcnt, action_refcnt, explicitly_created).
459 */
460 struct mutex lock;
461 struct list_head chain_list;
462 u32 index; /* block index for shared blocks */
463 u32 classid; /* which class this block belongs to */
464 refcount_t refcnt;
465 struct net *net;
466 struct Qdisc *q;
467 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
468 struct flow_block flow_block;
469 struct list_head owner_list;
470 bool keep_dst;
471 atomic_t offloadcnt; /* Number of oddloaded filters */
472 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
473 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
474 struct {
475 struct tcf_chain *chain;
476 struct list_head filter_chain_list;
477 } chain0;
478 struct rcu_head rcu;
479 DECLARE_HASHTABLE(proto_destroy_ht, 7);
480 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
481};
482
483static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
484{
485 return lockdep_is_held(&chain->filter_chain_lock);
486}
487
488static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
489{
490 return lockdep_is_held(&tp->lock);
491}
492
493#define tcf_chain_dereference(p, chain) \
494 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
495
496#define tcf_proto_dereference(p, tp) \
497 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
498
499static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
500{
501 struct qdisc_skb_cb *qcb;
502
503 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
504 BUILD_BUG_ON(sizeof(qcb->data) < sz);
505}
506
507static inline int qdisc_qlen(const struct Qdisc *q)
508{
509 return q->q.qlen;
510}
511
512static inline int qdisc_qlen_sum(const struct Qdisc *q)
513{
514 __u32 qlen = q->qstats.qlen;
515 int i;
516
517 if (qdisc_is_percpu_stats(q)) {
518 for_each_possible_cpu(i)
519 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
520 } else {
521 qlen += q->q.qlen;
522 }
523
524 return qlen;
525}
526
527static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
528{
529 return (struct qdisc_skb_cb *)skb->cb;
530}
531
532static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
533{
534 return &qdisc->q.lock;
535}
536
537static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
538{
539 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
540
541 return q;
542}
543
544static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
545{
546 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
547}
548
549static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
550{
551 return qdisc->dev_queue->qdisc_sleeping;
552}
553
554static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
555{
556 struct Qdisc *root = qdisc_root_sleeping(qdisc);
557
558 ASSERT_RTNL();
559 return qdisc_lock(root);
560}
561
562static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
563{
564 return qdisc->dev_queue->dev;
565}
566
567static inline void sch_tree_lock(struct Qdisc *q)
568{
569 if (q->flags & TCQ_F_MQROOT)
570 spin_lock_bh(qdisc_lock(q));
571 else
572 spin_lock_bh(qdisc_root_sleeping_lock(q));
573}
574
575static inline void sch_tree_unlock(struct Qdisc *q)
576{
577 if (q->flags & TCQ_F_MQROOT)
578 spin_unlock_bh(qdisc_lock(q));
579 else
580 spin_unlock_bh(qdisc_root_sleeping_lock(q));
581}
582
583extern struct Qdisc noop_qdisc;
584extern struct Qdisc_ops noop_qdisc_ops;
585extern struct Qdisc_ops pfifo_fast_ops;
586extern struct Qdisc_ops mq_qdisc_ops;
587extern struct Qdisc_ops noqueue_qdisc_ops;
588extern const struct Qdisc_ops *default_qdisc_ops;
589static inline const struct Qdisc_ops *
590get_default_qdisc_ops(const struct net_device *dev, int ntx)
591{
592 return ntx < dev->real_num_tx_queues ?
593 default_qdisc_ops : &pfifo_fast_ops;
594}
595
596struct Qdisc_class_common {
597 u32 classid;
598 struct hlist_node hnode;
599};
600
601struct Qdisc_class_hash {
602 struct hlist_head *hash;
603 unsigned int hashsize;
604 unsigned int hashmask;
605 unsigned int hashelems;
606};
607
608static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
609{
610 id ^= id >> 8;
611 id ^= id >> 4;
612 return id & mask;
613}
614
615static inline struct Qdisc_class_common *
616qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
617{
618 struct Qdisc_class_common *cl;
619 unsigned int h;
620
621 if (!id)
622 return NULL;
623
624 h = qdisc_class_hash(id, hash->hashmask);
625 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
626 if (cl->classid == id)
627 return cl;
628 }
629 return NULL;
630}
631
632static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
633{
634 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
635
636 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
637}
638
639int qdisc_class_hash_init(struct Qdisc_class_hash *);
640void qdisc_class_hash_insert(struct Qdisc_class_hash *,
641 struct Qdisc_class_common *);
642void qdisc_class_hash_remove(struct Qdisc_class_hash *,
643 struct Qdisc_class_common *);
644void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
645void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
646
647int dev_qdisc_change_tx_queue_len(struct net_device *dev);
648void dev_qdisc_change_real_num_tx(struct net_device *dev,
649 unsigned int new_real_tx);
650void dev_init_scheduler(struct net_device *dev);
651void dev_shutdown(struct net_device *dev);
652void dev_activate(struct net_device *dev);
653void dev_deactivate(struct net_device *dev);
654void dev_deactivate_many(struct list_head *head);
655struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
656 struct Qdisc *qdisc);
657void qdisc_reset(struct Qdisc *qdisc);
658void qdisc_put(struct Qdisc *qdisc);
659void qdisc_put_unlocked(struct Qdisc *qdisc);
660void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
661#ifdef CONFIG_NET_SCHED
662int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
663 void *type_data);
664void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
665 struct Qdisc *new, struct Qdisc *old,
666 enum tc_setup_type type, void *type_data,
667 struct netlink_ext_ack *extack);
668#else
669static inline int
670qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
671 void *type_data)
672{
673 q->flags &= ~TCQ_F_OFFLOADED;
674 return 0;
675}
676
677static inline void
678qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
679 struct Qdisc *new, struct Qdisc *old,
680 enum tc_setup_type type, void *type_data,
681 struct netlink_ext_ack *extack)
682{
683}
684#endif
685struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
686 const struct Qdisc_ops *ops,
687 struct netlink_ext_ack *extack);
688void qdisc_free(struct Qdisc *qdisc);
689struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
690 const struct Qdisc_ops *ops, u32 parentid,
691 struct netlink_ext_ack *extack);
692void __qdisc_calculate_pkt_len(struct sk_buff *skb,
693 const struct qdisc_size_table *stab);
694int skb_do_redirect(struct sk_buff *);
695
696static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
697{
698#ifdef CONFIG_NET_CLS_ACT
699 return skb->tc_at_ingress;
700#else
701 return false;
702#endif
703}
704
705static inline bool skb_skip_tc_classify(struct sk_buff *skb)
706{
707#ifdef CONFIG_NET_CLS_ACT
708 if (skb->tc_skip_classify) {
709 skb->tc_skip_classify = 0;
710 return true;
711 }
712#endif
713 return false;
714}
715
716/* Reset all TX qdiscs greater than index of a device. */
717static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
718{
719 struct Qdisc *qdisc;
720
721 for (; i < dev->num_tx_queues; i++) {
722 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
723 if (qdisc) {
724 spin_lock_bh(qdisc_lock(qdisc));
725 qdisc_reset(qdisc);
726 spin_unlock_bh(qdisc_lock(qdisc));
727 }
728 }
729}
730
731/* Are all TX queues of the device empty? */
732static inline bool qdisc_all_tx_empty(const struct net_device *dev)
733{
734 unsigned int i;
735
736 rcu_read_lock();
737 for (i = 0; i < dev->num_tx_queues; i++) {
738 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
739 const struct Qdisc *q = rcu_dereference(txq->qdisc);
740
741 if (!qdisc_is_empty(q)) {
742 rcu_read_unlock();
743 return false;
744 }
745 }
746 rcu_read_unlock();
747 return true;
748}
749
750/* Are any of the TX qdiscs changing? */
751static inline bool qdisc_tx_changing(const struct net_device *dev)
752{
753 unsigned int i;
754
755 for (i = 0; i < dev->num_tx_queues; i++) {
756 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
757 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
758 return true;
759 }
760 return false;
761}
762
763/* Is the device using the noop qdisc on all queues? */
764static inline bool qdisc_tx_is_noop(const struct net_device *dev)
765{
766 unsigned int i;
767
768 for (i = 0; i < dev->num_tx_queues; i++) {
769 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
770 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
771 return false;
772 }
773 return true;
774}
775
776static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
777{
778 return qdisc_skb_cb(skb)->pkt_len;
779}
780
781/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
782enum net_xmit_qdisc_t {
783 __NET_XMIT_STOLEN = 0x00010000,
784 __NET_XMIT_BYPASS = 0x00020000,
785};
786
787#ifdef CONFIG_NET_CLS_ACT
788#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
789#else
790#define net_xmit_drop_count(e) (1)
791#endif
792
793static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
794 const struct Qdisc *sch)
795{
796#ifdef CONFIG_NET_SCHED
797 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
798
799 if (stab)
800 __qdisc_calculate_pkt_len(skb, stab);
801#endif
802}
803
804static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
805 struct sk_buff **to_free)
806{
807 qdisc_calculate_pkt_len(skb, sch);
808 return sch->enqueue(skb, sch, to_free);
809}
810
811static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
812 __u64 bytes, __u32 packets)
813{
814 u64_stats_update_begin(&bstats->syncp);
815 u64_stats_add(&bstats->bytes, bytes);
816 u64_stats_add(&bstats->packets, packets);
817 u64_stats_update_end(&bstats->syncp);
818}
819
820static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
821 const struct sk_buff *skb)
822{
823 _bstats_update(bstats,
824 qdisc_pkt_len(skb),
825 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
826}
827
828static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
829 const struct sk_buff *skb)
830{
831 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
832}
833
834static inline void qdisc_bstats_update(struct Qdisc *sch,
835 const struct sk_buff *skb)
836{
837 bstats_update(&sch->bstats, skb);
838}
839
840static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
841 const struct sk_buff *skb)
842{
843 sch->qstats.backlog -= qdisc_pkt_len(skb);
844}
845
846static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
847 const struct sk_buff *skb)
848{
849 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
850}
851
852static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
853 const struct sk_buff *skb)
854{
855 sch->qstats.backlog += qdisc_pkt_len(skb);
856}
857
858static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
859 const struct sk_buff *skb)
860{
861 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
862}
863
864static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
865{
866 this_cpu_inc(sch->cpu_qstats->qlen);
867}
868
869static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
870{
871 this_cpu_dec(sch->cpu_qstats->qlen);
872}
873
874static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
875{
876 this_cpu_inc(sch->cpu_qstats->requeues);
877}
878
879static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
880{
881 sch->qstats.drops += count;
882}
883
884static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
885{
886 qstats->drops++;
887}
888
889static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
890{
891 qstats->overlimits++;
892}
893
894static inline void qdisc_qstats_drop(struct Qdisc *sch)
895{
896 qstats_drop_inc(&sch->qstats);
897}
898
899static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
900{
901 this_cpu_inc(sch->cpu_qstats->drops);
902}
903
904static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
905{
906 sch->qstats.overlimits++;
907}
908
909static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
910{
911 __u32 qlen = qdisc_qlen_sum(sch);
912
913 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
914}
915
916static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
917 __u32 *backlog)
918{
919 struct gnet_stats_queue qstats = { 0 };
920
921 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
922 *qlen = qstats.qlen + qdisc_qlen(sch);
923 *backlog = qstats.backlog;
924}
925
926static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
927{
928 __u32 qlen, backlog;
929
930 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
931 qdisc_tree_reduce_backlog(sch, qlen, backlog);
932}
933
934static inline void qdisc_purge_queue(struct Qdisc *sch)
935{
936 __u32 qlen, backlog;
937
938 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
939 qdisc_reset(sch);
940 qdisc_tree_reduce_backlog(sch, qlen, backlog);
941}
942
943static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
944{
945 qh->head = NULL;
946 qh->tail = NULL;
947 qh->qlen = 0;
948}
949
950static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
951 struct qdisc_skb_head *qh)
952{
953 struct sk_buff *last = qh->tail;
954
955 if (last) {
956 skb->next = NULL;
957 last->next = skb;
958 qh->tail = skb;
959 } else {
960 qh->tail = skb;
961 qh->head = skb;
962 }
963 qh->qlen++;
964}
965
966static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
967{
968 __qdisc_enqueue_tail(skb, &sch->q);
969 qdisc_qstats_backlog_inc(sch, skb);
970 return NET_XMIT_SUCCESS;
971}
972
973static inline void __qdisc_enqueue_head(struct sk_buff *skb,
974 struct qdisc_skb_head *qh)
975{
976 skb->next = qh->head;
977
978 if (!qh->head)
979 qh->tail = skb;
980 qh->head = skb;
981 qh->qlen++;
982}
983
984static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
985{
986 struct sk_buff *skb = qh->head;
987
988 if (likely(skb != NULL)) {
989 qh->head = skb->next;
990 qh->qlen--;
991 if (qh->head == NULL)
992 qh->tail = NULL;
993 skb->next = NULL;
994 }
995
996 return skb;
997}
998
999static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1000{
1001 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1002
1003 if (likely(skb != NULL)) {
1004 qdisc_qstats_backlog_dec(sch, skb);
1005 qdisc_bstats_update(sch, skb);
1006 }
1007
1008 return skb;
1009}
1010
1011/* Instead of calling kfree_skb() while root qdisc lock is held,
1012 * queue the skb for future freeing at end of __dev_xmit_skb()
1013 */
1014static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1015{
1016 skb->next = *to_free;
1017 *to_free = skb;
1018}
1019
1020static inline void __qdisc_drop_all(struct sk_buff *skb,
1021 struct sk_buff **to_free)
1022{
1023 if (skb->prev)
1024 skb->prev->next = *to_free;
1025 else
1026 skb->next = *to_free;
1027 *to_free = skb;
1028}
1029
1030static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1031 struct qdisc_skb_head *qh,
1032 struct sk_buff **to_free)
1033{
1034 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1035
1036 if (likely(skb != NULL)) {
1037 unsigned int len = qdisc_pkt_len(skb);
1038
1039 qdisc_qstats_backlog_dec(sch, skb);
1040 __qdisc_drop(skb, to_free);
1041 return len;
1042 }
1043
1044 return 0;
1045}
1046
1047static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1048{
1049 const struct qdisc_skb_head *qh = &sch->q;
1050
1051 return qh->head;
1052}
1053
1054/* generic pseudo peek method for non-work-conserving qdisc */
1055static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1056{
1057 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1058
1059 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1060 if (!skb) {
1061 skb = sch->dequeue(sch);
1062
1063 if (skb) {
1064 __skb_queue_head(&sch->gso_skb, skb);
1065 /* it's still part of the queue */
1066 qdisc_qstats_backlog_inc(sch, skb);
1067 sch->q.qlen++;
1068 }
1069 }
1070
1071 return skb;
1072}
1073
1074static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1075 struct sk_buff *skb)
1076{
1077 if (qdisc_is_percpu_stats(sch)) {
1078 qdisc_qstats_cpu_backlog_dec(sch, skb);
1079 qdisc_bstats_cpu_update(sch, skb);
1080 qdisc_qstats_cpu_qlen_dec(sch);
1081 } else {
1082 qdisc_qstats_backlog_dec(sch, skb);
1083 qdisc_bstats_update(sch, skb);
1084 sch->q.qlen--;
1085 }
1086}
1087
1088static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1089 unsigned int pkt_len)
1090{
1091 if (qdisc_is_percpu_stats(sch)) {
1092 qdisc_qstats_cpu_qlen_inc(sch);
1093 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1094 } else {
1095 sch->qstats.backlog += pkt_len;
1096 sch->q.qlen++;
1097 }
1098}
1099
1100/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1101static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1102{
1103 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1104
1105 if (skb) {
1106 skb = __skb_dequeue(&sch->gso_skb);
1107 if (qdisc_is_percpu_stats(sch)) {
1108 qdisc_qstats_cpu_backlog_dec(sch, skb);
1109 qdisc_qstats_cpu_qlen_dec(sch);
1110 } else {
1111 qdisc_qstats_backlog_dec(sch, skb);
1112 sch->q.qlen--;
1113 }
1114 } else {
1115 skb = sch->dequeue(sch);
1116 }
1117
1118 return skb;
1119}
1120
1121static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1122{
1123 /*
1124 * We do not know the backlog in bytes of this list, it
1125 * is up to the caller to correct it
1126 */
1127 ASSERT_RTNL();
1128 if (qh->qlen) {
1129 rtnl_kfree_skbs(qh->head, qh->tail);
1130
1131 qh->head = NULL;
1132 qh->tail = NULL;
1133 qh->qlen = 0;
1134 }
1135}
1136
1137static inline void qdisc_reset_queue(struct Qdisc *sch)
1138{
1139 __qdisc_reset_queue(&sch->q);
1140 sch->qstats.backlog = 0;
1141}
1142
1143static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1144 struct Qdisc **pold)
1145{
1146 struct Qdisc *old;
1147
1148 sch_tree_lock(sch);
1149 old = *pold;
1150 *pold = new;
1151 if (old != NULL)
1152 qdisc_purge_queue(old);
1153 sch_tree_unlock(sch);
1154
1155 return old;
1156}
1157
1158static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1159{
1160 rtnl_kfree_skbs(skb, skb);
1161 qdisc_qstats_drop(sch);
1162}
1163
1164static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1165 struct sk_buff **to_free)
1166{
1167 __qdisc_drop(skb, to_free);
1168 qdisc_qstats_cpu_drop(sch);
1169
1170 return NET_XMIT_DROP;
1171}
1172
1173static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1174 struct sk_buff **to_free)
1175{
1176 __qdisc_drop(skb, to_free);
1177 qdisc_qstats_drop(sch);
1178
1179 return NET_XMIT_DROP;
1180}
1181
1182static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1183 struct sk_buff **to_free)
1184{
1185 __qdisc_drop_all(skb, to_free);
1186 qdisc_qstats_drop(sch);
1187
1188 return NET_XMIT_DROP;
1189}
1190
1191/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1192 long it will take to send a packet given its size.
1193 */
1194static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1195{
1196 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1197 if (slot < 0)
1198 slot = 0;
1199 slot >>= rtab->rate.cell_log;
1200 if (slot > 255)
1201 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1202 return rtab->data[slot];
1203}
1204
1205struct psched_ratecfg {
1206 u64 rate_bytes_ps; /* bytes per second */
1207 u32 mult;
1208 u16 overhead;
1209 u16 mpu;
1210 u8 linklayer;
1211 u8 shift;
1212};
1213
1214static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1215 unsigned int len)
1216{
1217 len += r->overhead;
1218
1219 if (len < r->mpu)
1220 len = r->mpu;
1221
1222 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1223 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1224
1225 return ((u64)len * r->mult) >> r->shift;
1226}
1227
1228void psched_ratecfg_precompute(struct psched_ratecfg *r,
1229 const struct tc_ratespec *conf,
1230 u64 rate64);
1231
1232static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1233 const struct psched_ratecfg *r)
1234{
1235 memset(res, 0, sizeof(*res));
1236
1237 /* legacy struct tc_ratespec has a 32bit @rate field
1238 * Qdisc using 64bit rate should add new attributes
1239 * in order to maintain compatibility.
1240 */
1241 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1242
1243 res->overhead = r->overhead;
1244 res->mpu = r->mpu;
1245 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1246}
1247
1248struct psched_pktrate {
1249 u64 rate_pkts_ps; /* packets per second */
1250 u32 mult;
1251 u8 shift;
1252};
1253
1254static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1255 unsigned int pkt_num)
1256{
1257 return ((u64)pkt_num * r->mult) >> r->shift;
1258}
1259
1260void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1261
1262/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1263 * The fast path only needs to access filter list and to update stats
1264 */
1265struct mini_Qdisc {
1266 struct tcf_proto *filter_list;
1267 struct tcf_block *block;
1268 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1269 struct gnet_stats_queue __percpu *cpu_qstats;
1270 unsigned long rcu_state;
1271};
1272
1273static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1274 const struct sk_buff *skb)
1275{
1276 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1277}
1278
1279static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1280{
1281 this_cpu_inc(miniq->cpu_qstats->drops);
1282}
1283
1284struct mini_Qdisc_pair {
1285 struct mini_Qdisc miniq1;
1286 struct mini_Qdisc miniq2;
1287 struct mini_Qdisc __rcu **p_miniq;
1288};
1289
1290void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1291 struct tcf_proto *tp_head);
1292void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1293 struct mini_Qdisc __rcu **p_miniq);
1294void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1295 struct tcf_block *block);
1296
1297void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1298
1299int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1300
1301#endif
1302

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