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 <net/gen_stats.h>
17#include <net/rtnetlink.h>
18
19struct Qdisc_ops;
20struct qdisc_walker;
21struct tcf_walker;
22struct module;
23struct bpf_flow_keys;
24
25typedef int tc_setup_cb_t(enum tc_setup_type type,
26 void *type_data, void *cb_priv);
27
28typedef int tc_indr_block_bind_cb_t(struct net_device *dev, void *cb_priv,
29 enum tc_setup_type type, void *type_data);
30
31struct qdisc_rate_table {
32 struct tc_ratespec rate;
33 u32 data[256];
34 struct qdisc_rate_table *next;
35 int refcnt;
36};
37
38enum qdisc_state_t {
39 __QDISC_STATE_SCHED,
40 __QDISC_STATE_DEACTIVATED,
41};
42
43struct qdisc_size_table {
44 struct rcu_head rcu;
45 struct list_head list;
46 struct tc_sizespec szopts;
47 int refcnt;
48 u16 data[];
49};
50
51/* similar to sk_buff_head, but skb->prev pointer is undefined. */
52struct qdisc_skb_head {
53 struct sk_buff *head;
54 struct sk_buff *tail;
55 union {
56 u32 qlen;
57 atomic_t atomic_qlen;
58 };
59 spinlock_t lock;
60};
61
62struct Qdisc {
63 int (*enqueue)(struct sk_buff *skb,
64 struct Qdisc *sch,
65 struct sk_buff **to_free);
66 struct sk_buff * (*dequeue)(struct Qdisc *sch);
67 unsigned int flags;
68#define TCQ_F_BUILTIN 1
69#define TCQ_F_INGRESS 2
70#define TCQ_F_CAN_BYPASS 4
71#define TCQ_F_MQROOT 8
72#define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
73 * q->dev_queue : It can test
74 * netif_xmit_frozen_or_stopped() before
75 * dequeueing next packet.
76 * Its true for MQ/MQPRIO slaves, or non
77 * multiqueue device.
78 */
79#define TCQ_F_WARN_NONWC (1 << 16)
80#define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
81#define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
82 * qdisc_tree_decrease_qlen() should stop.
83 */
84#define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
85#define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
86#define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
87 u32 limit;
88 const struct Qdisc_ops *ops;
89 struct qdisc_size_table __rcu *stab;
90 struct hlist_node hash;
91 u32 handle;
92 u32 parent;
93
94 struct netdev_queue *dev_queue;
95
96 struct net_rate_estimator __rcu *rate_est;
97 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
98 struct gnet_stats_queue __percpu *cpu_qstats;
99 int padded;
100 refcount_t refcnt;
101
102 /*
103 * For performance sake on SMP, we put highly modified fields at the end
104 */
105 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
106 struct qdisc_skb_head q;
107 struct gnet_stats_basic_packed bstats;
108 seqcount_t running;
109 struct gnet_stats_queue qstats;
110 unsigned long state;
111 struct Qdisc *next_sched;
112 struct sk_buff_head skb_bad_txq;
113
114 spinlock_t busylock ____cacheline_aligned_in_smp;
115 spinlock_t seqlock;
116 struct rcu_head rcu;
117};
118
119static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
120{
121 if (qdisc->flags & TCQ_F_BUILTIN)
122 return;
123 refcount_inc(&qdisc->refcnt);
124}
125
126/* Intended to be used by unlocked users, when concurrent qdisc release is
127 * possible.
128 */
129
130static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
131{
132 if (qdisc->flags & TCQ_F_BUILTIN)
133 return qdisc;
134 if (refcount_inc_not_zero(&qdisc->refcnt))
135 return qdisc;
136 return NULL;
137}
138
139static inline bool qdisc_is_running(struct Qdisc *qdisc)
140{
141 if (qdisc->flags & TCQ_F_NOLOCK)
142 return spin_is_locked(&qdisc->seqlock);
143 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
144}
145
146static inline bool qdisc_run_begin(struct Qdisc *qdisc)
147{
148 if (qdisc->flags & TCQ_F_NOLOCK) {
149 if (!spin_trylock(&qdisc->seqlock))
150 return false;
151 } else if (qdisc_is_running(qdisc)) {
152 return false;
153 }
154 /* Variant of write_seqcount_begin() telling lockdep a trylock
155 * was attempted.
156 */
157 raw_write_seqcount_begin(&qdisc->running);
158 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
159 return true;
160}
161
162static inline void qdisc_run_end(struct Qdisc *qdisc)
163{
164 write_seqcount_end(&qdisc->running);
165 if (qdisc->flags & TCQ_F_NOLOCK)
166 spin_unlock(&qdisc->seqlock);
167}
168
169static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
170{
171 return qdisc->flags & TCQ_F_ONETXQUEUE;
172}
173
174static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
175{
176#ifdef CONFIG_BQL
177 /* Non-BQL migrated drivers will return 0, too. */
178 return dql_avail(&txq->dql);
179#else
180 return 0;
181#endif
182}
183
184struct Qdisc_class_ops {
185 unsigned int flags;
186 /* Child qdisc manipulation */
187 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
188 int (*graft)(struct Qdisc *, unsigned long cl,
189 struct Qdisc *, struct Qdisc **,
190 struct netlink_ext_ack *extack);
191 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
192 void (*qlen_notify)(struct Qdisc *, unsigned long);
193
194 /* Class manipulation routines */
195 unsigned long (*find)(struct Qdisc *, u32 classid);
196 int (*change)(struct Qdisc *, u32, u32,
197 struct nlattr **, unsigned long *,
198 struct netlink_ext_ack *);
199 int (*delete)(struct Qdisc *, unsigned long);
200 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
201
202 /* Filter manipulation */
203 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
204 unsigned long arg,
205 struct netlink_ext_ack *extack);
206 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
207 u32 classid);
208 void (*unbind_tcf)(struct Qdisc *, unsigned long);
209
210 /* rtnetlink specific */
211 int (*dump)(struct Qdisc *, unsigned long,
212 struct sk_buff *skb, struct tcmsg*);
213 int (*dump_stats)(struct Qdisc *, unsigned long,
214 struct gnet_dump *);
215};
216
217/* Qdisc_class_ops flag values */
218
219/* Implements API that doesn't require rtnl lock */
220enum qdisc_class_ops_flags {
221 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
222};
223
224struct Qdisc_ops {
225 struct Qdisc_ops *next;
226 const struct Qdisc_class_ops *cl_ops;
227 char id[IFNAMSIZ];
228 int priv_size;
229 unsigned int static_flags;
230
231 int (*enqueue)(struct sk_buff *skb,
232 struct Qdisc *sch,
233 struct sk_buff **to_free);
234 struct sk_buff * (*dequeue)(struct Qdisc *);
235 struct sk_buff * (*peek)(struct Qdisc *);
236
237 int (*init)(struct Qdisc *sch, struct nlattr *arg,
238 struct netlink_ext_ack *extack);
239 void (*reset)(struct Qdisc *);
240 void (*destroy)(struct Qdisc *);
241 int (*change)(struct Qdisc *sch,
242 struct nlattr *arg,
243 struct netlink_ext_ack *extack);
244 void (*attach)(struct Qdisc *sch);
245 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
246
247 int (*dump)(struct Qdisc *, struct sk_buff *);
248 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
249
250 void (*ingress_block_set)(struct Qdisc *sch,
251 u32 block_index);
252 void (*egress_block_set)(struct Qdisc *sch,
253 u32 block_index);
254 u32 (*ingress_block_get)(struct Qdisc *sch);
255 u32 (*egress_block_get)(struct Qdisc *sch);
256
257 struct module *owner;
258};
259
260
261struct tcf_result {
262 union {
263 struct {
264 unsigned long class;
265 u32 classid;
266 };
267 const struct tcf_proto *goto_tp;
268
269 /* used by the TC_ACT_REINSERT action */
270 struct {
271 bool ingress;
272 struct gnet_stats_queue *qstats;
273 };
274 };
275};
276
277struct tcf_chain;
278
279struct tcf_proto_ops {
280 struct list_head head;
281 char kind[IFNAMSIZ];
282
283 int (*classify)(struct sk_buff *,
284 const struct tcf_proto *,
285 struct tcf_result *);
286 int (*init)(struct tcf_proto*);
287 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
288 struct netlink_ext_ack *extack);
289
290 void* (*get)(struct tcf_proto*, u32 handle);
291 void (*put)(struct tcf_proto *tp, void *f);
292 int (*change)(struct net *net, struct sk_buff *,
293 struct tcf_proto*, unsigned long,
294 u32 handle, struct nlattr **,
295 void **, bool, bool,
296 struct netlink_ext_ack *);
297 int (*delete)(struct tcf_proto *tp, void *arg,
298 bool *last, bool rtnl_held,
299 struct netlink_ext_ack *);
300 void (*walk)(struct tcf_proto *tp,
301 struct tcf_walker *arg, bool rtnl_held);
302 int (*reoffload)(struct tcf_proto *tp, bool add,
303 tc_setup_cb_t *cb, void *cb_priv,
304 struct netlink_ext_ack *extack);
305 void (*bind_class)(void *, u32, unsigned long);
306 void * (*tmplt_create)(struct net *net,
307 struct tcf_chain *chain,
308 struct nlattr **tca,
309 struct netlink_ext_ack *extack);
310 void (*tmplt_destroy)(void *tmplt_priv);
311
312 /* rtnetlink specific */
313 int (*dump)(struct net*, struct tcf_proto*, void *,
314 struct sk_buff *skb, struct tcmsg*,
315 bool);
316 int (*tmplt_dump)(struct sk_buff *skb,
317 struct net *net,
318 void *tmplt_priv);
319
320 struct module *owner;
321 int flags;
322};
323
324enum tcf_proto_ops_flags {
325 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
326};
327
328struct tcf_proto {
329 /* Fast access part */
330 struct tcf_proto __rcu *next;
331 void __rcu *root;
332
333 /* called under RCU BH lock*/
334 int (*classify)(struct sk_buff *,
335 const struct tcf_proto *,
336 struct tcf_result *);
337 __be16 protocol;
338
339 /* All the rest */
340 u32 prio;
341 void *data;
342 const struct tcf_proto_ops *ops;
343 struct tcf_chain *chain;
344 /* Lock protects tcf_proto shared state and can be used by unlocked
345 * classifiers to protect their private data.
346 */
347 spinlock_t lock;
348 bool deleting;
349 refcount_t refcnt;
350 struct rcu_head rcu;
351};
352
353struct qdisc_skb_cb {
354 union {
355 struct {
356 unsigned int pkt_len;
357 u16 slave_dev_queue_mapping;
358 u16 tc_classid;
359 };
360 struct bpf_flow_keys *flow_keys;
361 };
362#define QDISC_CB_PRIV_LEN 20
363 unsigned char data[QDISC_CB_PRIV_LEN];
364};
365
366typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
367
368struct tcf_chain {
369 /* Protects filter_chain. */
370 struct mutex filter_chain_lock;
371 struct tcf_proto __rcu *filter_chain;
372 struct list_head list;
373 struct tcf_block *block;
374 u32 index; /* chain index */
375 unsigned int refcnt;
376 unsigned int action_refcnt;
377 bool explicitly_created;
378 bool flushing;
379 const struct tcf_proto_ops *tmplt_ops;
380 void *tmplt_priv;
381};
382
383struct tcf_block {
384 /* Lock protects tcf_block and lifetime-management data of chains
385 * attached to the block (refcnt, action_refcnt, explicitly_created).
386 */
387 struct mutex lock;
388 struct list_head chain_list;
389 u32 index; /* block index for shared blocks */
390 refcount_t refcnt;
391 struct net *net;
392 struct Qdisc *q;
393 struct list_head cb_list;
394 struct list_head owner_list;
395 bool keep_dst;
396 unsigned int offloadcnt; /* Number of oddloaded filters */
397 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
398 struct {
399 struct tcf_chain *chain;
400 struct list_head filter_chain_list;
401 } chain0;
402 struct rcu_head rcu;
403};
404
405#ifdef CONFIG_PROVE_LOCKING
406static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
407{
408 return lockdep_is_held(&chain->filter_chain_lock);
409}
410
411static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
412{
413 return lockdep_is_held(&tp->lock);
414}
415#else
416static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain)
417{
418 return true;
419}
420
421static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
422{
423 return true;
424}
425#endif /* #ifdef CONFIG_PROVE_LOCKING */
426
427#define tcf_chain_dereference(p, chain) \
428 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
429
430#define tcf_proto_dereference(p, tp) \
431 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
432
433static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
434{
435 if (*flags & TCA_CLS_FLAGS_IN_HW)
436 return;
437 *flags |= TCA_CLS_FLAGS_IN_HW;
438 block->offloadcnt++;
439}
440
441static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
442{
443 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
444 return;
445 *flags &= ~TCA_CLS_FLAGS_IN_HW;
446 block->offloadcnt--;
447}
448
449static inline void
450tc_cls_offload_cnt_update(struct tcf_block *block, u32 *cnt,
451 u32 *flags, bool add)
452{
453 if (add) {
454 if (!*cnt)
455 tcf_block_offload_inc(block, flags);
456 (*cnt)++;
457 } else {
458 (*cnt)--;
459 if (!*cnt)
460 tcf_block_offload_dec(block, flags);
461 }
462}
463
464static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
465{
466 struct qdisc_skb_cb *qcb;
467
468 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
469 BUILD_BUG_ON(sizeof(qcb->data) < sz);
470}
471
472static inline int qdisc_qlen(const struct Qdisc *q)
473{
474 return q->q.qlen;
475}
476
477static inline u32 qdisc_qlen_sum(const struct Qdisc *q)
478{
479 u32 qlen = q->qstats.qlen;
480
481 if (q->flags & TCQ_F_NOLOCK)
482 qlen += atomic_read(&q->q.atomic_qlen);
483 else
484 qlen += q->q.qlen;
485
486 return qlen;
487}
488
489static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
490{
491 return (struct qdisc_skb_cb *)skb->cb;
492}
493
494static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
495{
496 return &qdisc->q.lock;
497}
498
499static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
500{
501 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
502
503 return q;
504}
505
506static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
507{
508 return qdisc->dev_queue->qdisc_sleeping;
509}
510
511/* The qdisc root lock is a mechanism by which to top level
512 * of a qdisc tree can be locked from any qdisc node in the
513 * forest. This allows changing the configuration of some
514 * aspect of the qdisc tree while blocking out asynchronous
515 * qdisc access in the packet processing paths.
516 *
517 * It is only legal to do this when the root will not change
518 * on us. Otherwise we'll potentially lock the wrong qdisc
519 * root. This is enforced by holding the RTNL semaphore, which
520 * all users of this lock accessor must do.
521 */
522static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
523{
524 struct Qdisc *root = qdisc_root(qdisc);
525
526 ASSERT_RTNL();
527 return qdisc_lock(root);
528}
529
530static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
531{
532 struct Qdisc *root = qdisc_root_sleeping(qdisc);
533
534 ASSERT_RTNL();
535 return qdisc_lock(root);
536}
537
538static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
539{
540 struct Qdisc *root = qdisc_root_sleeping(qdisc);
541
542 ASSERT_RTNL();
543 return &root->running;
544}
545
546static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
547{
548 return qdisc->dev_queue->dev;
549}
550
551static inline void sch_tree_lock(const struct Qdisc *q)
552{
553 spin_lock_bh(qdisc_root_sleeping_lock(q));
554}
555
556static inline void sch_tree_unlock(const struct Qdisc *q)
557{
558 spin_unlock_bh(qdisc_root_sleeping_lock(q));
559}
560
561extern struct Qdisc noop_qdisc;
562extern struct Qdisc_ops noop_qdisc_ops;
563extern struct Qdisc_ops pfifo_fast_ops;
564extern struct Qdisc_ops mq_qdisc_ops;
565extern struct Qdisc_ops noqueue_qdisc_ops;
566extern const struct Qdisc_ops *default_qdisc_ops;
567static inline const struct Qdisc_ops *
568get_default_qdisc_ops(const struct net_device *dev, int ntx)
569{
570 return ntx < dev->real_num_tx_queues ?
571 default_qdisc_ops : &pfifo_fast_ops;
572}
573
574struct Qdisc_class_common {
575 u32 classid;
576 struct hlist_node hnode;
577};
578
579struct Qdisc_class_hash {
580 struct hlist_head *hash;
581 unsigned int hashsize;
582 unsigned int hashmask;
583 unsigned int hashelems;
584};
585
586static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
587{
588 id ^= id >> 8;
589 id ^= id >> 4;
590 return id & mask;
591}
592
593static inline struct Qdisc_class_common *
594qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
595{
596 struct Qdisc_class_common *cl;
597 unsigned int h;
598
599 if (!id)
600 return NULL;
601
602 h = qdisc_class_hash(id, hash->hashmask);
603 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
604 if (cl->classid == id)
605 return cl;
606 }
607 return NULL;
608}
609
610static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
611{
612 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
613
614 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
615}
616
617int qdisc_class_hash_init(struct Qdisc_class_hash *);
618void qdisc_class_hash_insert(struct Qdisc_class_hash *,
619 struct Qdisc_class_common *);
620void qdisc_class_hash_remove(struct Qdisc_class_hash *,
621 struct Qdisc_class_common *);
622void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
623void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
624
625int dev_qdisc_change_tx_queue_len(struct net_device *dev);
626void dev_init_scheduler(struct net_device *dev);
627void dev_shutdown(struct net_device *dev);
628void dev_activate(struct net_device *dev);
629void dev_deactivate(struct net_device *dev);
630void dev_deactivate_many(struct list_head *head);
631struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
632 struct Qdisc *qdisc);
633void qdisc_reset(struct Qdisc *qdisc);
634void qdisc_put(struct Qdisc *qdisc);
635void qdisc_put_unlocked(struct Qdisc *qdisc);
636void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
637#ifdef CONFIG_NET_SCHED
638int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
639 void *type_data);
640void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
641 struct Qdisc *new, struct Qdisc *old,
642 enum tc_setup_type type, void *type_data,
643 struct netlink_ext_ack *extack);
644#else
645static inline int
646qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
647 void *type_data)
648{
649 q->flags &= ~TCQ_F_OFFLOADED;
650 return 0;
651}
652
653static inline void
654qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
655 struct Qdisc *new, struct Qdisc *old,
656 enum tc_setup_type type, void *type_data,
657 struct netlink_ext_ack *extack)
658{
659}
660#endif
661struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
662 const struct Qdisc_ops *ops,
663 struct netlink_ext_ack *extack);
664void qdisc_free(struct Qdisc *qdisc);
665struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
666 const struct Qdisc_ops *ops, u32 parentid,
667 struct netlink_ext_ack *extack);
668void __qdisc_calculate_pkt_len(struct sk_buff *skb,
669 const struct qdisc_size_table *stab);
670int skb_do_redirect(struct sk_buff *);
671
672static inline void skb_reset_tc(struct sk_buff *skb)
673{
674#ifdef CONFIG_NET_CLS_ACT
675 skb->tc_redirected = 0;
676#endif
677}
678
679static inline bool skb_is_tc_redirected(const struct sk_buff *skb)
680{
681#ifdef CONFIG_NET_CLS_ACT
682 return skb->tc_redirected;
683#else
684 return false;
685#endif
686}
687
688static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
689{
690#ifdef CONFIG_NET_CLS_ACT
691 return skb->tc_at_ingress;
692#else
693 return false;
694#endif
695}
696
697static inline bool skb_skip_tc_classify(struct sk_buff *skb)
698{
699#ifdef CONFIG_NET_CLS_ACT
700 if (skb->tc_skip_classify) {
701 skb->tc_skip_classify = 0;
702 return true;
703 }
704#endif
705 return false;
706}
707
708/* Reset all TX qdiscs greater than index of a device. */
709static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
710{
711 struct Qdisc *qdisc;
712
713 for (; i < dev->num_tx_queues; i++) {
714 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
715 if (qdisc) {
716 spin_lock_bh(qdisc_lock(qdisc));
717 qdisc_reset(qdisc);
718 spin_unlock_bh(qdisc_lock(qdisc));
719 }
720 }
721}
722
723static inline void qdisc_reset_all_tx(struct net_device *dev)
724{
725 qdisc_reset_all_tx_gt(dev, 0);
726}
727
728/* Are all TX queues of the device empty? */
729static inline bool qdisc_all_tx_empty(const struct net_device *dev)
730{
731 unsigned int i;
732
733 rcu_read_lock();
734 for (i = 0; i < dev->num_tx_queues; i++) {
735 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
736 const struct Qdisc *q = rcu_dereference(txq->qdisc);
737
738 if (q->q.qlen) {
739 rcu_read_unlock();
740 return false;
741 }
742 }
743 rcu_read_unlock();
744 return true;
745}
746
747/* Are any of the TX qdiscs changing? */
748static inline bool qdisc_tx_changing(const struct net_device *dev)
749{
750 unsigned int i;
751
752 for (i = 0; i < dev->num_tx_queues; i++) {
753 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
754 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
755 return true;
756 }
757 return false;
758}
759
760/* Is the device using the noop qdisc on all queues? */
761static inline bool qdisc_tx_is_noop(const struct net_device *dev)
762{
763 unsigned int i;
764
765 for (i = 0; i < dev->num_tx_queues; i++) {
766 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
767 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
768 return false;
769 }
770 return true;
771}
772
773static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
774{
775 return qdisc_skb_cb(skb)->pkt_len;
776}
777
778/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
779enum net_xmit_qdisc_t {
780 __NET_XMIT_STOLEN = 0x00010000,
781 __NET_XMIT_BYPASS = 0x00020000,
782};
783
784#ifdef CONFIG_NET_CLS_ACT
785#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
786#else
787#define net_xmit_drop_count(e) (1)
788#endif
789
790static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
791 const struct Qdisc *sch)
792{
793#ifdef CONFIG_NET_SCHED
794 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
795
796 if (stab)
797 __qdisc_calculate_pkt_len(skb, stab);
798#endif
799}
800
801static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
802 struct sk_buff **to_free)
803{
804 qdisc_calculate_pkt_len(skb, sch);
805 return sch->enqueue(skb, sch, to_free);
806}
807
808static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
809{
810 return q->flags & TCQ_F_CPUSTATS;
811}
812
813static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
814 __u64 bytes, __u32 packets)
815{
816 bstats->bytes += bytes;
817 bstats->packets += packets;
818}
819
820static inline void bstats_update(struct gnet_stats_basic_packed *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 _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
829 __u64 bytes, __u32 packets)
830{
831 u64_stats_update_begin(&bstats->syncp);
832 _bstats_update(&bstats->bstats, bytes, packets);
833 u64_stats_update_end(&bstats->syncp);
834}
835
836static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
837 const struct sk_buff *skb)
838{
839 u64_stats_update_begin(&bstats->syncp);
840 bstats_update(&bstats->bstats, skb);
841 u64_stats_update_end(&bstats->syncp);
842}
843
844static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
845 const struct sk_buff *skb)
846{
847 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
848}
849
850static inline void qdisc_bstats_update(struct Qdisc *sch,
851 const struct sk_buff *skb)
852{
853 bstats_update(&sch->bstats, skb);
854}
855
856static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
857 const struct sk_buff *skb)
858{
859 sch->qstats.backlog -= qdisc_pkt_len(skb);
860}
861
862static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
863 const struct sk_buff *skb)
864{
865 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
866}
867
868static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
869 const struct sk_buff *skb)
870{
871 sch->qstats.backlog += qdisc_pkt_len(skb);
872}
873
874static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
875 const struct sk_buff *skb)
876{
877 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
878}
879
880static inline void qdisc_qstats_atomic_qlen_inc(struct Qdisc *sch)
881{
882 atomic_inc(&sch->q.atomic_qlen);
883}
884
885static inline void qdisc_qstats_atomic_qlen_dec(struct Qdisc *sch)
886{
887 atomic_dec(&sch->q.atomic_qlen);
888}
889
890static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
891{
892 this_cpu_inc(sch->cpu_qstats->requeues);
893}
894
895static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
896{
897 sch->qstats.drops += count;
898}
899
900static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
901{
902 qstats->drops++;
903}
904
905static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
906{
907 qstats->overlimits++;
908}
909
910static inline void qdisc_qstats_drop(struct Qdisc *sch)
911{
912 qstats_drop_inc(&sch->qstats);
913}
914
915static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
916{
917 this_cpu_inc(sch->cpu_qstats->drops);
918}
919
920static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
921{
922 sch->qstats.overlimits++;
923}
924
925static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
926{
927 qh->head = NULL;
928 qh->tail = NULL;
929 qh->qlen = 0;
930}
931
932static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
933 struct qdisc_skb_head *qh)
934{
935 struct sk_buff *last = qh->tail;
936
937 if (last) {
938 skb->next = NULL;
939 last->next = skb;
940 qh->tail = skb;
941 } else {
942 qh->tail = skb;
943 qh->head = skb;
944 }
945 qh->qlen++;
946}
947
948static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
949{
950 __qdisc_enqueue_tail(skb, &sch->q);
951 qdisc_qstats_backlog_inc(sch, skb);
952 return NET_XMIT_SUCCESS;
953}
954
955static inline void __qdisc_enqueue_head(struct sk_buff *skb,
956 struct qdisc_skb_head *qh)
957{
958 skb->next = qh->head;
959
960 if (!qh->head)
961 qh->tail = skb;
962 qh->head = skb;
963 qh->qlen++;
964}
965
966static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
967{
968 struct sk_buff *skb = qh->head;
969
970 if (likely(skb != NULL)) {
971 qh->head = skb->next;
972 qh->qlen--;
973 if (qh->head == NULL)
974 qh->tail = NULL;
975 skb->next = NULL;
976 }
977
978 return skb;
979}
980
981static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
982{
983 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
984
985 if (likely(skb != NULL)) {
986 qdisc_qstats_backlog_dec(sch, skb);
987 qdisc_bstats_update(sch, skb);
988 }
989
990 return skb;
991}
992
993/* Instead of calling kfree_skb() while root qdisc lock is held,
994 * queue the skb for future freeing at end of __dev_xmit_skb()
995 */
996static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
997{
998 skb->next = *to_free;
999 *to_free = skb;
1000}
1001
1002static inline void __qdisc_drop_all(struct sk_buff *skb,
1003 struct sk_buff **to_free)
1004{
1005 if (skb->prev)
1006 skb->prev->next = *to_free;
1007 else
1008 skb->next = *to_free;
1009 *to_free = skb;
1010}
1011
1012static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1013 struct qdisc_skb_head *qh,
1014 struct sk_buff **to_free)
1015{
1016 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1017
1018 if (likely(skb != NULL)) {
1019 unsigned int len = qdisc_pkt_len(skb);
1020
1021 qdisc_qstats_backlog_dec(sch, skb);
1022 __qdisc_drop(skb, to_free);
1023 return len;
1024 }
1025
1026 return 0;
1027}
1028
1029static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
1030 struct sk_buff **to_free)
1031{
1032 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
1033}
1034
1035static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1036{
1037 const struct qdisc_skb_head *qh = &sch->q;
1038
1039 return qh->head;
1040}
1041
1042/* generic pseudo peek method for non-work-conserving qdisc */
1043static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1044{
1045 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1046
1047 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1048 if (!skb) {
1049 skb = sch->dequeue(sch);
1050
1051 if (skb) {
1052 __skb_queue_head(&sch->gso_skb, skb);
1053 /* it's still part of the queue */
1054 qdisc_qstats_backlog_inc(sch, skb);
1055 sch->q.qlen++;
1056 }
1057 }
1058
1059 return skb;
1060}
1061
1062/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1063static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1064{
1065 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1066
1067 if (skb) {
1068 skb = __skb_dequeue(&sch->gso_skb);
1069 qdisc_qstats_backlog_dec(sch, skb);
1070 sch->q.qlen--;
1071 } else {
1072 skb = sch->dequeue(sch);
1073 }
1074
1075 return skb;
1076}
1077
1078static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1079{
1080 /*
1081 * We do not know the backlog in bytes of this list, it
1082 * is up to the caller to correct it
1083 */
1084 ASSERT_RTNL();
1085 if (qh->qlen) {
1086 rtnl_kfree_skbs(qh->head, qh->tail);
1087
1088 qh->head = NULL;
1089 qh->tail = NULL;
1090 qh->qlen = 0;
1091 }
1092}
1093
1094static inline void qdisc_reset_queue(struct Qdisc *sch)
1095{
1096 __qdisc_reset_queue(&sch->q);
1097 sch->qstats.backlog = 0;
1098}
1099
1100static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1101 struct Qdisc **pold)
1102{
1103 struct Qdisc *old;
1104
1105 sch_tree_lock(sch);
1106 old = *pold;
1107 *pold = new;
1108 if (old != NULL) {
1109 unsigned int qlen = old->q.qlen;
1110 unsigned int backlog = old->qstats.backlog;
1111
1112 qdisc_reset(old);
1113 qdisc_tree_reduce_backlog(old, qlen, backlog);
1114 }
1115 sch_tree_unlock(sch);
1116
1117 return old;
1118}
1119
1120static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1121{
1122 rtnl_kfree_skbs(skb, skb);
1123 qdisc_qstats_drop(sch);
1124}
1125
1126static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1127 struct sk_buff **to_free)
1128{
1129 __qdisc_drop(skb, to_free);
1130 qdisc_qstats_cpu_drop(sch);
1131
1132 return NET_XMIT_DROP;
1133}
1134
1135static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1136 struct sk_buff **to_free)
1137{
1138 __qdisc_drop(skb, to_free);
1139 qdisc_qstats_drop(sch);
1140
1141 return NET_XMIT_DROP;
1142}
1143
1144static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1145 struct sk_buff **to_free)
1146{
1147 __qdisc_drop_all(skb, to_free);
1148 qdisc_qstats_drop(sch);
1149
1150 return NET_XMIT_DROP;
1151}
1152
1153/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1154 long it will take to send a packet given its size.
1155 */
1156static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1157{
1158 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1159 if (slot < 0)
1160 slot = 0;
1161 slot >>= rtab->rate.cell_log;
1162 if (slot > 255)
1163 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1164 return rtab->data[slot];
1165}
1166
1167struct psched_ratecfg {
1168 u64 rate_bytes_ps; /* bytes per second */
1169 u32 mult;
1170 u16 overhead;
1171 u8 linklayer;
1172 u8 shift;
1173};
1174
1175static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1176 unsigned int len)
1177{
1178 len += r->overhead;
1179
1180 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1181 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1182
1183 return ((u64)len * r->mult) >> r->shift;
1184}
1185
1186void psched_ratecfg_precompute(struct psched_ratecfg *r,
1187 const struct tc_ratespec *conf,
1188 u64 rate64);
1189
1190static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1191 const struct psched_ratecfg *r)
1192{
1193 memset(res, 0, sizeof(*res));
1194
1195 /* legacy struct tc_ratespec has a 32bit @rate field
1196 * Qdisc using 64bit rate should add new attributes
1197 * in order to maintain compatibility.
1198 */
1199 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1200
1201 res->overhead = r->overhead;
1202 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1203}
1204
1205/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1206 * The fast path only needs to access filter list and to update stats
1207 */
1208struct mini_Qdisc {
1209 struct tcf_proto *filter_list;
1210 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1211 struct gnet_stats_queue __percpu *cpu_qstats;
1212 struct rcu_head rcu;
1213};
1214
1215static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1216 const struct sk_buff *skb)
1217{
1218 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1219}
1220
1221static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1222{
1223 this_cpu_inc(miniq->cpu_qstats->drops);
1224}
1225
1226struct mini_Qdisc_pair {
1227 struct mini_Qdisc miniq1;
1228 struct mini_Qdisc miniq2;
1229 struct mini_Qdisc __rcu **p_miniq;
1230};
1231
1232void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1233 struct tcf_proto *tp_head);
1234void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1235 struct mini_Qdisc __rcu **p_miniq);
1236
1237static inline void skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res)
1238{
1239 struct gnet_stats_queue *stats = res->qstats;
1240 int ret;
1241
1242 if (res->ingress)
1243 ret = netif_receive_skb(skb);
1244 else
1245 ret = dev_queue_xmit(skb);
1246 if (ret && stats)
1247 qstats_overlimit_inc(res->qstats);
1248}
1249
1250#endif
1251