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
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 qdisc->empty;
153 return !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 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 void (*walk)(struct tcf_proto *tp,
312 struct tcf_walker *arg, bool rtnl_held);
313 int (*reoffload)(struct tcf_proto *tp, bool add,
314 flow_setup_cb_t *cb, void *cb_priv,
315 struct netlink_ext_ack *extack);
316 void (*hw_add)(struct tcf_proto *tp,
317 void *type_data);
318 void (*hw_del)(struct tcf_proto *tp,
319 void *type_data);
320 void (*bind_class)(void *, u32, unsigned long);
321 void * (*tmplt_create)(struct net *net,
322 struct tcf_chain *chain,
323 struct nlattr **tca,
324 struct netlink_ext_ack *extack);
325 void (*tmplt_destroy)(void *tmplt_priv);
326
327 /* rtnetlink specific */
328 int (*dump)(struct net*, struct tcf_proto*, void *,
329 struct sk_buff *skb, struct tcmsg*,
330 bool);
331 int (*tmplt_dump)(struct sk_buff *skb,
332 struct net *net,
333 void *tmplt_priv);
334
335 struct module *owner;
336 int flags;
337};
338
339enum tcf_proto_ops_flags {
340 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
341};
342
343struct tcf_proto {
344 /* Fast access part */
345 struct tcf_proto __rcu *next;
346 void __rcu *root;
347
348 /* called under RCU BH lock*/
349 int (*classify)(struct sk_buff *,
350 const struct tcf_proto *,
351 struct tcf_result *);
352 __be16 protocol;
353
354 /* All the rest */
355 u32 prio;
356 void *data;
357 const struct tcf_proto_ops *ops;
358 struct tcf_chain *chain;
359 /* Lock protects tcf_proto shared state and can be used by unlocked
360 * classifiers to protect their private data.
361 */
362 spinlock_t lock;
363 bool deleting;
364 refcount_t refcnt;
365 struct rcu_head rcu;
366 struct hlist_node destroy_ht_node;
367};
368
369struct qdisc_skb_cb {
370 struct {
371 unsigned int pkt_len;
372 u16 slave_dev_queue_mapping;
373 u16 tc_classid;
374 };
375#define QDISC_CB_PRIV_LEN 20
376 unsigned char data[QDISC_CB_PRIV_LEN];
377};
378
379typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
380
381struct tcf_chain {
382 /* Protects filter_chain. */
383 struct mutex filter_chain_lock;
384 struct tcf_proto __rcu *filter_chain;
385 struct list_head list;
386 struct tcf_block *block;
387 u32 index; /* chain index */
388 unsigned int refcnt;
389 unsigned int action_refcnt;
390 bool explicitly_created;
391 bool flushing;
392 const struct tcf_proto_ops *tmplt_ops;
393 void *tmplt_priv;
394 struct rcu_head rcu;
395};
396
397struct tcf_block {
398 /* Lock protects tcf_block and lifetime-management data of chains
399 * attached to the block (refcnt, action_refcnt, explicitly_created).
400 */
401 struct mutex lock;
402 struct list_head chain_list;
403 u32 index; /* block index for shared blocks */
404 refcount_t refcnt;
405 struct net *net;
406 struct Qdisc *q;
407 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
408 struct flow_block flow_block;
409 struct list_head owner_list;
410 bool keep_dst;
411 atomic_t offloadcnt; /* Number of oddloaded filters */
412 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
413 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
414 struct {
415 struct tcf_chain *chain;
416 struct list_head filter_chain_list;
417 } chain0;
418 struct rcu_head rcu;
419 DECLARE_HASHTABLE(proto_destroy_ht, 7);
420 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
421};
422
423#ifdef CONFIG_PROVE_LOCKING
424static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
425{
426 return lockdep_is_held(&chain->filter_chain_lock);
427}
428
429static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
430{
431 return lockdep_is_held(&tp->lock);
432}
433#else
434static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain)
435{
436 return true;
437}
438
439static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
440{
441 return true;
442}
443#endif /* #ifdef CONFIG_PROVE_LOCKING */
444
445#define tcf_chain_dereference(p, chain) \
446 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
447
448#define tcf_proto_dereference(p, tp) \
449 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
450
451static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
452{
453 struct qdisc_skb_cb *qcb;
454
455 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
456 BUILD_BUG_ON(sizeof(qcb->data) < sz);
457}
458
459static inline int qdisc_qlen_cpu(const struct Qdisc *q)
460{
461 return this_cpu_ptr(q->cpu_qstats)->qlen;
462}
463
464static inline int qdisc_qlen(const struct Qdisc *q)
465{
466 return q->q.qlen;
467}
468
469static inline int qdisc_qlen_sum(const struct Qdisc *q)
470{
471 __u32 qlen = q->qstats.qlen;
472 int i;
473
474 if (qdisc_is_percpu_stats(q)) {
475 for_each_possible_cpu(i)
476 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
477 } else {
478 qlen += q->q.qlen;
479 }
480
481 return qlen;
482}
483
484static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
485{
486 return (struct qdisc_skb_cb *)skb->cb;
487}
488
489static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
490{
491 return &qdisc->q.lock;
492}
493
494static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
495{
496 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
497
498 return q;
499}
500
501static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
502{
503 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
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 (!qdisc_is_empty(q)) {
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 void _bstats_update(struct gnet_stats_basic_packed *bstats,
809 __u64 bytes, __u32 packets)
810{
811 bstats->bytes += bytes;
812 bstats->packets += packets;
813}
814
815static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
816 const struct sk_buff *skb)
817{
818 _bstats_update(bstats,
819 qdisc_pkt_len(skb),
820 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
821}
822
823static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
824 __u64 bytes, __u32 packets)
825{
826 u64_stats_update_begin(&bstats->syncp);
827 _bstats_update(&bstats->bstats, bytes, packets);
828 u64_stats_update_end(&bstats->syncp);
829}
830
831static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
832 const struct sk_buff *skb)
833{
834 u64_stats_update_begin(&bstats->syncp);
835 bstats_update(&bstats->bstats, skb);
836 u64_stats_update_end(&bstats->syncp);
837}
838
839static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
840 const struct sk_buff *skb)
841{
842 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
843}
844
845static inline void qdisc_bstats_update(struct Qdisc *sch,
846 const struct sk_buff *skb)
847{
848 bstats_update(&sch->bstats, skb);
849}
850
851static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
852 const struct sk_buff *skb)
853{
854 sch->qstats.backlog -= qdisc_pkt_len(skb);
855}
856
857static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
858 const struct sk_buff *skb)
859{
860 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
861}
862
863static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
864 const struct sk_buff *skb)
865{
866 sch->qstats.backlog += qdisc_pkt_len(skb);
867}
868
869static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
870 const struct sk_buff *skb)
871{
872 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
873}
874
875static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
876{
877 this_cpu_inc(sch->cpu_qstats->qlen);
878}
879
880static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
881{
882 this_cpu_dec(sch->cpu_qstats->qlen);
883}
884
885static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
886{
887 this_cpu_inc(sch->cpu_qstats->requeues);
888}
889
890static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
891{
892 sch->qstats.drops += count;
893}
894
895static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
896{
897 qstats->drops++;
898}
899
900static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
901{
902 qstats->overlimits++;
903}
904
905static inline void qdisc_qstats_drop(struct Qdisc *sch)
906{
907 qstats_drop_inc(&sch->qstats);
908}
909
910static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
911{
912 this_cpu_inc(sch->cpu_qstats->drops);
913}
914
915static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
916{
917 sch->qstats.overlimits++;
918}
919
920static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
921{
922 __u32 qlen = qdisc_qlen_sum(sch);
923
924 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
925}
926
927static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
928 __u32 *backlog)
929{
930 struct gnet_stats_queue qstats = { 0 };
931 __u32 len = qdisc_qlen_sum(sch);
932
933 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
934 *qlen = qstats.qlen;
935 *backlog = qstats.backlog;
936}
937
938static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
939{
940 __u32 qlen, backlog;
941
942 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
943 qdisc_tree_reduce_backlog(sch, qlen, backlog);
944}
945
946static inline void qdisc_purge_queue(struct Qdisc *sch)
947{
948 __u32 qlen, backlog;
949
950 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
951 qdisc_reset(sch);
952 qdisc_tree_reduce_backlog(sch, qlen, backlog);
953}
954
955static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
956{
957 qh->head = NULL;
958 qh->tail = NULL;
959 qh->qlen = 0;
960}
961
962static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
963 struct qdisc_skb_head *qh)
964{
965 struct sk_buff *last = qh->tail;
966
967 if (last) {
968 skb->next = NULL;
969 last->next = skb;
970 qh->tail = skb;
971 } else {
972 qh->tail = skb;
973 qh->head = skb;
974 }
975 qh->qlen++;
976}
977
978static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
979{
980 __qdisc_enqueue_tail(skb, &sch->q);
981 qdisc_qstats_backlog_inc(sch, skb);
982 return NET_XMIT_SUCCESS;
983}
984
985static inline void __qdisc_enqueue_head(struct sk_buff *skb,
986 struct qdisc_skb_head *qh)
987{
988 skb->next = qh->head;
989
990 if (!qh->head)
991 qh->tail = skb;
992 qh->head = skb;
993 qh->qlen++;
994}
995
996static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
997{
998 struct sk_buff *skb = qh->head;
999
1000 if (likely(skb != NULL)) {
1001 qh->head = skb->next;
1002 qh->qlen--;
1003 if (qh->head == NULL)
1004 qh->tail = NULL;
1005 skb->next = NULL;
1006 }
1007
1008 return skb;
1009}
1010
1011static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1012{
1013 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1014
1015 if (likely(skb != NULL)) {
1016 qdisc_qstats_backlog_dec(sch, skb);
1017 qdisc_bstats_update(sch, skb);
1018 }
1019
1020 return skb;
1021}
1022
1023/* Instead of calling kfree_skb() while root qdisc lock is held,
1024 * queue the skb for future freeing at end of __dev_xmit_skb()
1025 */
1026static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1027{
1028 skb->next = *to_free;
1029 *to_free = skb;
1030}
1031
1032static inline void __qdisc_drop_all(struct sk_buff *skb,
1033 struct sk_buff **to_free)
1034{
1035 if (skb->prev)
1036 skb->prev->next = *to_free;
1037 else
1038 skb->next = *to_free;
1039 *to_free = skb;
1040}
1041
1042static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1043 struct qdisc_skb_head *qh,
1044 struct sk_buff **to_free)
1045{
1046 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1047
1048 if (likely(skb != NULL)) {
1049 unsigned int len = qdisc_pkt_len(skb);
1050
1051 qdisc_qstats_backlog_dec(sch, skb);
1052 __qdisc_drop(skb, to_free);
1053 return len;
1054 }
1055
1056 return 0;
1057}
1058
1059static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
1060 struct sk_buff **to_free)
1061{
1062 return __qdisc_queue_drop_head(sch, &sch->q, to_free);
1063}
1064
1065static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1066{
1067 const struct qdisc_skb_head *qh = &sch->q;
1068
1069 return qh->head;
1070}
1071
1072/* generic pseudo peek method for non-work-conserving qdisc */
1073static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1074{
1075 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1076
1077 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1078 if (!skb) {
1079 skb = sch->dequeue(sch);
1080
1081 if (skb) {
1082 __skb_queue_head(&sch->gso_skb, skb);
1083 /* it's still part of the queue */
1084 qdisc_qstats_backlog_inc(sch, skb);
1085 sch->q.qlen++;
1086 }
1087 }
1088
1089 return skb;
1090}
1091
1092static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1093 struct sk_buff *skb)
1094{
1095 if (qdisc_is_percpu_stats(sch)) {
1096 qdisc_qstats_cpu_backlog_dec(sch, skb);
1097 qdisc_bstats_cpu_update(sch, skb);
1098 qdisc_qstats_cpu_qlen_dec(sch);
1099 } else {
1100 qdisc_qstats_backlog_dec(sch, skb);
1101 qdisc_bstats_update(sch, skb);
1102 sch->q.qlen--;
1103 }
1104}
1105
1106static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1107 unsigned int pkt_len)
1108{
1109 if (qdisc_is_percpu_stats(sch)) {
1110 qdisc_qstats_cpu_qlen_inc(sch);
1111 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1112 } else {
1113 sch->qstats.backlog += pkt_len;
1114 sch->q.qlen++;
1115 }
1116}
1117
1118/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1119static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1120{
1121 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1122
1123 if (skb) {
1124 skb = __skb_dequeue(&sch->gso_skb);
1125 if (qdisc_is_percpu_stats(sch)) {
1126 qdisc_qstats_cpu_backlog_dec(sch, skb);
1127 qdisc_qstats_cpu_qlen_dec(sch);
1128 } else {
1129 qdisc_qstats_backlog_dec(sch, skb);
1130 sch->q.qlen--;
1131 }
1132 } else {
1133 skb = sch->dequeue(sch);
1134 }
1135
1136 return skb;
1137}
1138
1139static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1140{
1141 /*
1142 * We do not know the backlog in bytes of this list, it
1143 * is up to the caller to correct it
1144 */
1145 ASSERT_RTNL();
1146 if (qh->qlen) {
1147 rtnl_kfree_skbs(qh->head, qh->tail);
1148
1149 qh->head = NULL;
1150 qh->tail = NULL;
1151 qh->qlen = 0;
1152 }
1153}
1154
1155static inline void qdisc_reset_queue(struct Qdisc *sch)
1156{
1157 __qdisc_reset_queue(&sch->q);
1158 sch->qstats.backlog = 0;
1159}
1160
1161static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1162 struct Qdisc **pold)
1163{
1164 struct Qdisc *old;
1165
1166 sch_tree_lock(sch);
1167 old = *pold;
1168 *pold = new;
1169 if (old != NULL)
1170 qdisc_tree_flush_backlog(old);
1171 sch_tree_unlock(sch);
1172
1173 return old;
1174}
1175
1176static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1177{
1178 rtnl_kfree_skbs(skb, skb);
1179 qdisc_qstats_drop(sch);
1180}
1181
1182static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1183 struct sk_buff **to_free)
1184{
1185 __qdisc_drop(skb, to_free);
1186 qdisc_qstats_cpu_drop(sch);
1187
1188 return NET_XMIT_DROP;
1189}
1190
1191static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1192 struct sk_buff **to_free)
1193{
1194 __qdisc_drop(skb, to_free);
1195 qdisc_qstats_drop(sch);
1196
1197 return NET_XMIT_DROP;
1198}
1199
1200static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1201 struct sk_buff **to_free)
1202{
1203 __qdisc_drop_all(skb, to_free);
1204 qdisc_qstats_drop(sch);
1205
1206 return NET_XMIT_DROP;
1207}
1208
1209/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1210 long it will take to send a packet given its size.
1211 */
1212static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1213{
1214 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1215 if (slot < 0)
1216 slot = 0;
1217 slot >>= rtab->rate.cell_log;
1218 if (slot > 255)
1219 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1220 return rtab->data[slot];
1221}
1222
1223struct psched_ratecfg {
1224 u64 rate_bytes_ps; /* bytes per second */
1225 u32 mult;
1226 u16 overhead;
1227 u8 linklayer;
1228 u8 shift;
1229};
1230
1231static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1232 unsigned int len)
1233{
1234 len += r->overhead;
1235
1236 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1237 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1238
1239 return ((u64)len * r->mult) >> r->shift;
1240}
1241
1242void psched_ratecfg_precompute(struct psched_ratecfg *r,
1243 const struct tc_ratespec *conf,
1244 u64 rate64);
1245
1246static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1247 const struct psched_ratecfg *r)
1248{
1249 memset(res, 0, sizeof(*res));
1250
1251 /* legacy struct tc_ratespec has a 32bit @rate field
1252 * Qdisc using 64bit rate should add new attributes
1253 * in order to maintain compatibility.
1254 */
1255 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1256
1257 res->overhead = r->overhead;
1258 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1259}
1260
1261/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1262 * The fast path only needs to access filter list and to update stats
1263 */
1264struct mini_Qdisc {
1265 struct tcf_proto *filter_list;
1266 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
1267 struct gnet_stats_queue __percpu *cpu_qstats;
1268 struct rcu_head rcu;
1269};
1270
1271static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1272 const struct sk_buff *skb)
1273{
1274 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1275}
1276
1277static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1278{
1279 this_cpu_inc(miniq->cpu_qstats->drops);
1280}
1281
1282struct mini_Qdisc_pair {
1283 struct mini_Qdisc miniq1;
1284 struct mini_Qdisc miniq2;
1285 struct mini_Qdisc __rcu **p_miniq;
1286};
1287
1288void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1289 struct tcf_proto *tp_head);
1290void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1291 struct mini_Qdisc __rcu **p_miniq);
1292
1293static inline void skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res)
1294{
1295 struct gnet_stats_queue *stats = res->qstats;
1296 int ret;
1297
1298 if (res->ingress)
1299 ret = netif_receive_skb(skb);
1300 else
1301 ret = dev_queue_xmit(skb);
1302 if (ret && stats)
1303 qstats_overlimit_inc(res->qstats);
1304}
1305
1306#endif
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef __NET_SCHED_GENERIC_H
3#define __NET_SCHED_GENERIC_H
4
5#include <linux/netdevice.h>
6#include <linux/types.h>
7#include <linux/rcupdate.h>
8#include <linux/pkt_sched.h>
9#include <linux/pkt_cls.h>
10#include <linux/percpu.h>
11#include <linux/dynamic_queue_limits.h>
12#include <linux/list.h>
13#include <linux/refcount.h>
14#include <linux/workqueue.h>
15#include <linux/mutex.h>
16#include <linux/rwsem.h>
17#include <linux/atomic.h>
18#include <linux/hashtable.h>
19#include <net/gen_stats.h>
20#include <net/rtnetlink.h>
21#include <net/flow_offload.h>
22
23struct Qdisc_ops;
24struct qdisc_walker;
25struct tcf_walker;
26struct module;
27struct bpf_flow_keys;
28
29struct qdisc_rate_table {
30 struct tc_ratespec rate;
31 u32 data[256];
32 struct qdisc_rate_table *next;
33 int refcnt;
34};
35
36enum qdisc_state_t {
37 __QDISC_STATE_SCHED,
38 __QDISC_STATE_DEACTIVATED,
39 __QDISC_STATE_MISSED,
40 __QDISC_STATE_DRAINING,
41};
42
43enum qdisc_state2_t {
44 /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly.
45 * Use qdisc_run_begin/end() or qdisc_is_running() instead.
46 */
47 __QDISC_STATE2_RUNNING,
48};
49
50#define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED)
51#define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING)
52
53#define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \
54 QDISC_STATE_DRAINING)
55
56struct qdisc_size_table {
57 struct rcu_head rcu;
58 struct list_head list;
59 struct tc_sizespec szopts;
60 int refcnt;
61 u16 data[];
62};
63
64/* similar to sk_buff_head, but skb->prev pointer is undefined. */
65struct qdisc_skb_head {
66 struct sk_buff *head;
67 struct sk_buff *tail;
68 __u32 qlen;
69 spinlock_t lock;
70};
71
72struct Qdisc {
73 int (*enqueue)(struct sk_buff *skb,
74 struct Qdisc *sch,
75 struct sk_buff **to_free);
76 struct sk_buff * (*dequeue)(struct Qdisc *sch);
77 unsigned int flags;
78#define TCQ_F_BUILTIN 1
79#define TCQ_F_INGRESS 2
80#define TCQ_F_CAN_BYPASS 4
81#define TCQ_F_MQROOT 8
82#define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
83 * q->dev_queue : It can test
84 * netif_xmit_frozen_or_stopped() before
85 * dequeueing next packet.
86 * Its true for MQ/MQPRIO slaves, or non
87 * multiqueue device.
88 */
89#define TCQ_F_WARN_NONWC (1 << 16)
90#define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
91#define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
92 * qdisc_tree_decrease_qlen() should stop.
93 */
94#define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
95#define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
96#define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
97 u32 limit;
98 const struct Qdisc_ops *ops;
99 struct qdisc_size_table __rcu *stab;
100 struct hlist_node hash;
101 u32 handle;
102 u32 parent;
103
104 struct netdev_queue *dev_queue;
105
106 struct net_rate_estimator __rcu *rate_est;
107 struct gnet_stats_basic_sync __percpu *cpu_bstats;
108 struct gnet_stats_queue __percpu *cpu_qstats;
109 int pad;
110 refcount_t refcnt;
111
112 /*
113 * For performance sake on SMP, we put highly modified fields at the end
114 */
115 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
116 struct qdisc_skb_head q;
117 struct gnet_stats_basic_sync bstats;
118 struct gnet_stats_queue qstats;
119 unsigned long state;
120 unsigned long state2; /* must be written under qdisc spinlock */
121 struct Qdisc *next_sched;
122 struct sk_buff_head skb_bad_txq;
123
124 spinlock_t busylock ____cacheline_aligned_in_smp;
125 spinlock_t seqlock;
126
127 struct rcu_head rcu;
128 netdevice_tracker dev_tracker;
129 /* private data */
130 long privdata[] ____cacheline_aligned;
131};
132
133static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
134{
135 if (qdisc->flags & TCQ_F_BUILTIN)
136 return;
137 refcount_inc(&qdisc->refcnt);
138}
139
140/* Intended to be used by unlocked users, when concurrent qdisc release is
141 * possible.
142 */
143
144static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
145{
146 if (qdisc->flags & TCQ_F_BUILTIN)
147 return qdisc;
148 if (refcount_inc_not_zero(&qdisc->refcnt))
149 return qdisc;
150 return NULL;
151}
152
153/* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
154 * root_lock section, or provide their own memory barriers -- ordering
155 * against qdisc_run_begin/end() atomic bit operations.
156 */
157static inline bool qdisc_is_running(struct Qdisc *qdisc)
158{
159 if (qdisc->flags & TCQ_F_NOLOCK)
160 return spin_is_locked(&qdisc->seqlock);
161 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
162}
163
164static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
165{
166 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
167}
168
169static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
170{
171 return q->flags & TCQ_F_CPUSTATS;
172}
173
174static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
175{
176 if (qdisc_is_percpu_stats(qdisc))
177 return nolock_qdisc_is_empty(qdisc);
178 return !READ_ONCE(qdisc->q.qlen);
179}
180
181/* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
182 * the qdisc root lock acquired.
183 */
184static inline bool qdisc_run_begin(struct Qdisc *qdisc)
185{
186 if (qdisc->flags & TCQ_F_NOLOCK) {
187 if (spin_trylock(&qdisc->seqlock))
188 return true;
189
190 /* No need to insist if the MISSED flag was already set.
191 * Note that test_and_set_bit() also gives us memory ordering
192 * guarantees wrt potential earlier enqueue() and below
193 * spin_trylock(), both of which are necessary to prevent races
194 */
195 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
196 return false;
197
198 /* Try to take the lock again to make sure that we will either
199 * grab it or the CPU that still has it will see MISSED set
200 * when testing it in qdisc_run_end()
201 */
202 return spin_trylock(&qdisc->seqlock);
203 }
204 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
205}
206
207static inline void qdisc_run_end(struct Qdisc *qdisc)
208{
209 if (qdisc->flags & TCQ_F_NOLOCK) {
210 spin_unlock(&qdisc->seqlock);
211
212 /* spin_unlock() only has store-release semantic. The unlock
213 * and test_bit() ordering is a store-load ordering, so a full
214 * memory barrier is needed here.
215 */
216 smp_mb();
217
218 if (unlikely(test_bit(__QDISC_STATE_MISSED,
219 &qdisc->state)))
220 __netif_schedule(qdisc);
221 } else {
222 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
223 }
224}
225
226static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
227{
228 return qdisc->flags & TCQ_F_ONETXQUEUE;
229}
230
231static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
232{
233#ifdef CONFIG_BQL
234 /* Non-BQL migrated drivers will return 0, too. */
235 return dql_avail(&txq->dql);
236#else
237 return 0;
238#endif
239}
240
241struct Qdisc_class_ops {
242 unsigned int flags;
243 /* Child qdisc manipulation */
244 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
245 int (*graft)(struct Qdisc *, unsigned long cl,
246 struct Qdisc *, struct Qdisc **,
247 struct netlink_ext_ack *extack);
248 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
249 void (*qlen_notify)(struct Qdisc *, unsigned long);
250
251 /* Class manipulation routines */
252 unsigned long (*find)(struct Qdisc *, u32 classid);
253 int (*change)(struct Qdisc *, u32, u32,
254 struct nlattr **, unsigned long *,
255 struct netlink_ext_ack *);
256 int (*delete)(struct Qdisc *, unsigned long,
257 struct netlink_ext_ack *);
258 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
259
260 /* Filter manipulation */
261 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
262 unsigned long arg,
263 struct netlink_ext_ack *extack);
264 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
265 u32 classid);
266 void (*unbind_tcf)(struct Qdisc *, unsigned long);
267
268 /* rtnetlink specific */
269 int (*dump)(struct Qdisc *, unsigned long,
270 struct sk_buff *skb, struct tcmsg*);
271 int (*dump_stats)(struct Qdisc *, unsigned long,
272 struct gnet_dump *);
273};
274
275/* Qdisc_class_ops flag values */
276
277/* Implements API that doesn't require rtnl lock */
278enum qdisc_class_ops_flags {
279 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
280};
281
282struct Qdisc_ops {
283 struct Qdisc_ops *next;
284 const struct Qdisc_class_ops *cl_ops;
285 char id[IFNAMSIZ];
286 int priv_size;
287 unsigned int static_flags;
288
289 int (*enqueue)(struct sk_buff *skb,
290 struct Qdisc *sch,
291 struct sk_buff **to_free);
292 struct sk_buff * (*dequeue)(struct Qdisc *);
293 struct sk_buff * (*peek)(struct Qdisc *);
294
295 int (*init)(struct Qdisc *sch, struct nlattr *arg,
296 struct netlink_ext_ack *extack);
297 void (*reset)(struct Qdisc *);
298 void (*destroy)(struct Qdisc *);
299 int (*change)(struct Qdisc *sch,
300 struct nlattr *arg,
301 struct netlink_ext_ack *extack);
302 void (*attach)(struct Qdisc *sch);
303 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
304 void (*change_real_num_tx)(struct Qdisc *sch,
305 unsigned int new_real_tx);
306
307 int (*dump)(struct Qdisc *, struct sk_buff *);
308 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
309
310 void (*ingress_block_set)(struct Qdisc *sch,
311 u32 block_index);
312 void (*egress_block_set)(struct Qdisc *sch,
313 u32 block_index);
314 u32 (*ingress_block_get)(struct Qdisc *sch);
315 u32 (*egress_block_get)(struct Qdisc *sch);
316
317 struct module *owner;
318};
319
320
321struct tcf_result {
322 union {
323 struct {
324 unsigned long class;
325 u32 classid;
326 };
327 const struct tcf_proto *goto_tp;
328
329 };
330};
331
332struct tcf_chain;
333
334struct tcf_proto_ops {
335 struct list_head head;
336 char kind[IFNAMSIZ];
337
338 int (*classify)(struct sk_buff *,
339 const struct tcf_proto *,
340 struct tcf_result *);
341 int (*init)(struct tcf_proto*);
342 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
343 struct netlink_ext_ack *extack);
344
345 void* (*get)(struct tcf_proto*, u32 handle);
346 void (*put)(struct tcf_proto *tp, void *f);
347 int (*change)(struct net *net, struct sk_buff *,
348 struct tcf_proto*, unsigned long,
349 u32 handle, struct nlattr **,
350 void **, u32,
351 struct netlink_ext_ack *);
352 int (*delete)(struct tcf_proto *tp, void *arg,
353 bool *last, bool rtnl_held,
354 struct netlink_ext_ack *);
355 bool (*delete_empty)(struct tcf_proto *tp);
356 void (*walk)(struct tcf_proto *tp,
357 struct tcf_walker *arg, bool rtnl_held);
358 int (*reoffload)(struct tcf_proto *tp, bool add,
359 flow_setup_cb_t *cb, void *cb_priv,
360 struct netlink_ext_ack *extack);
361 void (*hw_add)(struct tcf_proto *tp,
362 void *type_data);
363 void (*hw_del)(struct tcf_proto *tp,
364 void *type_data);
365 void (*bind_class)(void *, u32, unsigned long,
366 void *, unsigned long);
367 void * (*tmplt_create)(struct net *net,
368 struct tcf_chain *chain,
369 struct nlattr **tca,
370 struct netlink_ext_ack *extack);
371 void (*tmplt_destroy)(void *tmplt_priv);
372
373 /* rtnetlink specific */
374 int (*dump)(struct net*, struct tcf_proto*, void *,
375 struct sk_buff *skb, struct tcmsg*,
376 bool);
377 int (*terse_dump)(struct net *net,
378 struct tcf_proto *tp, void *fh,
379 struct sk_buff *skb,
380 struct tcmsg *t, bool rtnl_held);
381 int (*tmplt_dump)(struct sk_buff *skb,
382 struct net *net,
383 void *tmplt_priv);
384
385 struct module *owner;
386 int flags;
387};
388
389/* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
390 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
391 * conditions can occur when filters are inserted/deleted simultaneously.
392 */
393enum tcf_proto_ops_flags {
394 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
395};
396
397struct tcf_proto {
398 /* Fast access part */
399 struct tcf_proto __rcu *next;
400 void __rcu *root;
401
402 /* called under RCU BH lock*/
403 int (*classify)(struct sk_buff *,
404 const struct tcf_proto *,
405 struct tcf_result *);
406 __be16 protocol;
407
408 /* All the rest */
409 u32 prio;
410 void *data;
411 const struct tcf_proto_ops *ops;
412 struct tcf_chain *chain;
413 /* Lock protects tcf_proto shared state and can be used by unlocked
414 * classifiers to protect their private data.
415 */
416 spinlock_t lock;
417 bool deleting;
418 refcount_t refcnt;
419 struct rcu_head rcu;
420 struct hlist_node destroy_ht_node;
421};
422
423struct qdisc_skb_cb {
424 struct {
425 unsigned int pkt_len;
426 u16 slave_dev_queue_mapping;
427 u16 tc_classid;
428 };
429#define QDISC_CB_PRIV_LEN 20
430 unsigned char data[QDISC_CB_PRIV_LEN];
431};
432
433typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
434
435struct tcf_chain {
436 /* Protects filter_chain. */
437 struct mutex filter_chain_lock;
438 struct tcf_proto __rcu *filter_chain;
439 struct list_head list;
440 struct tcf_block *block;
441 u32 index; /* chain index */
442 unsigned int refcnt;
443 unsigned int action_refcnt;
444 bool explicitly_created;
445 bool flushing;
446 const struct tcf_proto_ops *tmplt_ops;
447 void *tmplt_priv;
448 struct rcu_head rcu;
449};
450
451struct tcf_block {
452 /* Lock protects tcf_block and lifetime-management data of chains
453 * attached to the block (refcnt, action_refcnt, explicitly_created).
454 */
455 struct mutex lock;
456 struct list_head chain_list;
457 u32 index; /* block index for shared blocks */
458 u32 classid; /* which class this block belongs to */
459 refcount_t refcnt;
460 struct net *net;
461 struct Qdisc *q;
462 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
463 struct flow_block flow_block;
464 struct list_head owner_list;
465 bool keep_dst;
466 atomic_t offloadcnt; /* Number of oddloaded filters */
467 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
468 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
469 struct {
470 struct tcf_chain *chain;
471 struct list_head filter_chain_list;
472 } chain0;
473 struct rcu_head rcu;
474 DECLARE_HASHTABLE(proto_destroy_ht, 7);
475 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
476};
477
478static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
479{
480 return lockdep_is_held(&chain->filter_chain_lock);
481}
482
483static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
484{
485 return lockdep_is_held(&tp->lock);
486}
487
488#define tcf_chain_dereference(p, chain) \
489 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
490
491#define tcf_proto_dereference(p, tp) \
492 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
493
494static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
495{
496 struct qdisc_skb_cb *qcb;
497
498 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
499 BUILD_BUG_ON(sizeof(qcb->data) < sz);
500}
501
502static inline int qdisc_qlen(const struct Qdisc *q)
503{
504 return q->q.qlen;
505}
506
507static inline int qdisc_qlen_sum(const struct Qdisc *q)
508{
509 __u32 qlen = q->qstats.qlen;
510 int i;
511
512 if (qdisc_is_percpu_stats(q)) {
513 for_each_possible_cpu(i)
514 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
515 } else {
516 qlen += q->q.qlen;
517 }
518
519 return qlen;
520}
521
522static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
523{
524 return (struct qdisc_skb_cb *)skb->cb;
525}
526
527static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
528{
529 return &qdisc->q.lock;
530}
531
532static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
533{
534 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
535
536 return q;
537}
538
539static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
540{
541 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
542}
543
544static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
545{
546 return qdisc->dev_queue->qdisc_sleeping;
547}
548
549static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
550{
551 struct Qdisc *root = qdisc_root_sleeping(qdisc);
552
553 ASSERT_RTNL();
554 return qdisc_lock(root);
555}
556
557static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
558{
559 return qdisc->dev_queue->dev;
560}
561
562static inline void sch_tree_lock(struct Qdisc *q)
563{
564 if (q->flags & TCQ_F_MQROOT)
565 spin_lock_bh(qdisc_lock(q));
566 else
567 spin_lock_bh(qdisc_root_sleeping_lock(q));
568}
569
570static inline void sch_tree_unlock(struct Qdisc *q)
571{
572 if (q->flags & TCQ_F_MQROOT)
573 spin_unlock_bh(qdisc_lock(q));
574 else
575 spin_unlock_bh(qdisc_root_sleeping_lock(q));
576}
577
578extern struct Qdisc noop_qdisc;
579extern struct Qdisc_ops noop_qdisc_ops;
580extern struct Qdisc_ops pfifo_fast_ops;
581extern struct Qdisc_ops mq_qdisc_ops;
582extern struct Qdisc_ops noqueue_qdisc_ops;
583extern const struct Qdisc_ops *default_qdisc_ops;
584static inline const struct Qdisc_ops *
585get_default_qdisc_ops(const struct net_device *dev, int ntx)
586{
587 return ntx < dev->real_num_tx_queues ?
588 default_qdisc_ops : &pfifo_fast_ops;
589}
590
591struct Qdisc_class_common {
592 u32 classid;
593 struct hlist_node hnode;
594};
595
596struct Qdisc_class_hash {
597 struct hlist_head *hash;
598 unsigned int hashsize;
599 unsigned int hashmask;
600 unsigned int hashelems;
601};
602
603static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
604{
605 id ^= id >> 8;
606 id ^= id >> 4;
607 return id & mask;
608}
609
610static inline struct Qdisc_class_common *
611qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
612{
613 struct Qdisc_class_common *cl;
614 unsigned int h;
615
616 if (!id)
617 return NULL;
618
619 h = qdisc_class_hash(id, hash->hashmask);
620 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
621 if (cl->classid == id)
622 return cl;
623 }
624 return NULL;
625}
626
627static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
628{
629 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
630
631 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
632}
633
634int qdisc_class_hash_init(struct Qdisc_class_hash *);
635void qdisc_class_hash_insert(struct Qdisc_class_hash *,
636 struct Qdisc_class_common *);
637void qdisc_class_hash_remove(struct Qdisc_class_hash *,
638 struct Qdisc_class_common *);
639void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
640void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
641
642int dev_qdisc_change_tx_queue_len(struct net_device *dev);
643void dev_qdisc_change_real_num_tx(struct net_device *dev,
644 unsigned int new_real_tx);
645void dev_init_scheduler(struct net_device *dev);
646void dev_shutdown(struct net_device *dev);
647void dev_activate(struct net_device *dev);
648void dev_deactivate(struct net_device *dev);
649void dev_deactivate_many(struct list_head *head);
650struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
651 struct Qdisc *qdisc);
652void qdisc_reset(struct Qdisc *qdisc);
653void qdisc_put(struct Qdisc *qdisc);
654void qdisc_put_unlocked(struct Qdisc *qdisc);
655void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
656#ifdef CONFIG_NET_SCHED
657int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
658 void *type_data);
659void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
660 struct Qdisc *new, struct Qdisc *old,
661 enum tc_setup_type type, void *type_data,
662 struct netlink_ext_ack *extack);
663#else
664static inline int
665qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
666 void *type_data)
667{
668 q->flags &= ~TCQ_F_OFFLOADED;
669 return 0;
670}
671
672static inline void
673qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
674 struct Qdisc *new, struct Qdisc *old,
675 enum tc_setup_type type, void *type_data,
676 struct netlink_ext_ack *extack)
677{
678}
679#endif
680void qdisc_offload_query_caps(struct net_device *dev,
681 enum tc_setup_type type,
682 void *caps, size_t caps_len);
683struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
684 const struct Qdisc_ops *ops,
685 struct netlink_ext_ack *extack);
686void qdisc_free(struct Qdisc *qdisc);
687struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
688 const struct Qdisc_ops *ops, u32 parentid,
689 struct netlink_ext_ack *extack);
690void __qdisc_calculate_pkt_len(struct sk_buff *skb,
691 const struct qdisc_size_table *stab);
692int skb_do_redirect(struct sk_buff *);
693
694static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
695{
696#ifdef CONFIG_NET_CLS_ACT
697 return skb->tc_at_ingress;
698#else
699 return false;
700#endif
701}
702
703static inline bool skb_skip_tc_classify(struct sk_buff *skb)
704{
705#ifdef CONFIG_NET_CLS_ACT
706 if (skb->tc_skip_classify) {
707 skb->tc_skip_classify = 0;
708 return true;
709 }
710#endif
711 return false;
712}
713
714/* Reset all TX qdiscs greater than index of a device. */
715static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
716{
717 struct Qdisc *qdisc;
718
719 for (; i < dev->num_tx_queues; i++) {
720 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
721 if (qdisc) {
722 spin_lock_bh(qdisc_lock(qdisc));
723 qdisc_reset(qdisc);
724 spin_unlock_bh(qdisc_lock(qdisc));
725 }
726 }
727}
728
729/* Are all TX queues of the device empty? */
730static inline bool qdisc_all_tx_empty(const struct net_device *dev)
731{
732 unsigned int i;
733
734 rcu_read_lock();
735 for (i = 0; i < dev->num_tx_queues; i++) {
736 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
737 const struct Qdisc *q = rcu_dereference(txq->qdisc);
738
739 if (!qdisc_is_empty(q)) {
740 rcu_read_unlock();
741 return false;
742 }
743 }
744 rcu_read_unlock();
745 return true;
746}
747
748/* Are any of the TX qdiscs changing? */
749static inline bool qdisc_tx_changing(const struct net_device *dev)
750{
751 unsigned int i;
752
753 for (i = 0; i < dev->num_tx_queues; i++) {
754 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
755 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
756 return true;
757 }
758 return false;
759}
760
761/* Is the device using the noop qdisc on all queues? */
762static inline bool qdisc_tx_is_noop(const struct net_device *dev)
763{
764 unsigned int i;
765
766 for (i = 0; i < dev->num_tx_queues; i++) {
767 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
768 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
769 return false;
770 }
771 return true;
772}
773
774static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
775{
776 return qdisc_skb_cb(skb)->pkt_len;
777}
778
779/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
780enum net_xmit_qdisc_t {
781 __NET_XMIT_STOLEN = 0x00010000,
782 __NET_XMIT_BYPASS = 0x00020000,
783};
784
785#ifdef CONFIG_NET_CLS_ACT
786#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
787#else
788#define net_xmit_drop_count(e) (1)
789#endif
790
791static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
792 const struct Qdisc *sch)
793{
794#ifdef CONFIG_NET_SCHED
795 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
796
797 if (stab)
798 __qdisc_calculate_pkt_len(skb, stab);
799#endif
800}
801
802static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
803 struct sk_buff **to_free)
804{
805 qdisc_calculate_pkt_len(skb, sch);
806 return sch->enqueue(skb, sch, to_free);
807}
808
809static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
810 __u64 bytes, __u32 packets)
811{
812 u64_stats_update_begin(&bstats->syncp);
813 u64_stats_add(&bstats->bytes, bytes);
814 u64_stats_add(&bstats->packets, packets);
815 u64_stats_update_end(&bstats->syncp);
816}
817
818static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
819 const struct sk_buff *skb)
820{
821 _bstats_update(bstats,
822 qdisc_pkt_len(skb),
823 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
824}
825
826static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
827 const struct sk_buff *skb)
828{
829 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
830}
831
832static inline void qdisc_bstats_update(struct Qdisc *sch,
833 const struct sk_buff *skb)
834{
835 bstats_update(&sch->bstats, skb);
836}
837
838static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
839 const struct sk_buff *skb)
840{
841 sch->qstats.backlog -= qdisc_pkt_len(skb);
842}
843
844static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
845 const struct sk_buff *skb)
846{
847 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
848}
849
850static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
851 const struct sk_buff *skb)
852{
853 sch->qstats.backlog += qdisc_pkt_len(skb);
854}
855
856static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
857 const struct sk_buff *skb)
858{
859 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
860}
861
862static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
863{
864 this_cpu_inc(sch->cpu_qstats->qlen);
865}
866
867static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
868{
869 this_cpu_dec(sch->cpu_qstats->qlen);
870}
871
872static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
873{
874 this_cpu_inc(sch->cpu_qstats->requeues);
875}
876
877static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
878{
879 sch->qstats.drops += count;
880}
881
882static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
883{
884 qstats->drops++;
885}
886
887static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
888{
889 qstats->overlimits++;
890}
891
892static inline void qdisc_qstats_drop(struct Qdisc *sch)
893{
894 qstats_drop_inc(&sch->qstats);
895}
896
897static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
898{
899 this_cpu_inc(sch->cpu_qstats->drops);
900}
901
902static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
903{
904 sch->qstats.overlimits++;
905}
906
907static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
908{
909 __u32 qlen = qdisc_qlen_sum(sch);
910
911 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
912}
913
914static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
915 __u32 *backlog)
916{
917 struct gnet_stats_queue qstats = { 0 };
918
919 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
920 *qlen = qstats.qlen + qdisc_qlen(sch);
921 *backlog = qstats.backlog;
922}
923
924static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
925{
926 __u32 qlen, backlog;
927
928 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
929 qdisc_tree_reduce_backlog(sch, qlen, backlog);
930}
931
932static inline void qdisc_purge_queue(struct Qdisc *sch)
933{
934 __u32 qlen, backlog;
935
936 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
937 qdisc_reset(sch);
938 qdisc_tree_reduce_backlog(sch, qlen, backlog);
939}
940
941static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
942 struct qdisc_skb_head *qh)
943{
944 struct sk_buff *last = qh->tail;
945
946 if (last) {
947 skb->next = NULL;
948 last->next = skb;
949 qh->tail = skb;
950 } else {
951 qh->tail = skb;
952 qh->head = skb;
953 }
954 qh->qlen++;
955}
956
957static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
958{
959 __qdisc_enqueue_tail(skb, &sch->q);
960 qdisc_qstats_backlog_inc(sch, skb);
961 return NET_XMIT_SUCCESS;
962}
963
964static inline void __qdisc_enqueue_head(struct sk_buff *skb,
965 struct qdisc_skb_head *qh)
966{
967 skb->next = qh->head;
968
969 if (!qh->head)
970 qh->tail = skb;
971 qh->head = skb;
972 qh->qlen++;
973}
974
975static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
976{
977 struct sk_buff *skb = qh->head;
978
979 if (likely(skb != NULL)) {
980 qh->head = skb->next;
981 qh->qlen--;
982 if (qh->head == NULL)
983 qh->tail = NULL;
984 skb->next = NULL;
985 }
986
987 return skb;
988}
989
990static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
991{
992 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
993
994 if (likely(skb != NULL)) {
995 qdisc_qstats_backlog_dec(sch, skb);
996 qdisc_bstats_update(sch, skb);
997 }
998
999 return skb;
1000}
1001
1002/* Instead of calling kfree_skb() while root qdisc lock is held,
1003 * queue the skb for future freeing at end of __dev_xmit_skb()
1004 */
1005static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1006{
1007 skb->next = *to_free;
1008 *to_free = skb;
1009}
1010
1011static inline void __qdisc_drop_all(struct sk_buff *skb,
1012 struct sk_buff **to_free)
1013{
1014 if (skb->prev)
1015 skb->prev->next = *to_free;
1016 else
1017 skb->next = *to_free;
1018 *to_free = skb;
1019}
1020
1021static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1022 struct qdisc_skb_head *qh,
1023 struct sk_buff **to_free)
1024{
1025 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1026
1027 if (likely(skb != NULL)) {
1028 unsigned int len = qdisc_pkt_len(skb);
1029
1030 qdisc_qstats_backlog_dec(sch, skb);
1031 __qdisc_drop(skb, to_free);
1032 return len;
1033 }
1034
1035 return 0;
1036}
1037
1038static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1039{
1040 const struct qdisc_skb_head *qh = &sch->q;
1041
1042 return qh->head;
1043}
1044
1045/* generic pseudo peek method for non-work-conserving qdisc */
1046static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1047{
1048 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1049
1050 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1051 if (!skb) {
1052 skb = sch->dequeue(sch);
1053
1054 if (skb) {
1055 __skb_queue_head(&sch->gso_skb, skb);
1056 /* it's still part of the queue */
1057 qdisc_qstats_backlog_inc(sch, skb);
1058 sch->q.qlen++;
1059 }
1060 }
1061
1062 return skb;
1063}
1064
1065static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1066 struct sk_buff *skb)
1067{
1068 if (qdisc_is_percpu_stats(sch)) {
1069 qdisc_qstats_cpu_backlog_dec(sch, skb);
1070 qdisc_bstats_cpu_update(sch, skb);
1071 qdisc_qstats_cpu_qlen_dec(sch);
1072 } else {
1073 qdisc_qstats_backlog_dec(sch, skb);
1074 qdisc_bstats_update(sch, skb);
1075 sch->q.qlen--;
1076 }
1077}
1078
1079static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1080 unsigned int pkt_len)
1081{
1082 if (qdisc_is_percpu_stats(sch)) {
1083 qdisc_qstats_cpu_qlen_inc(sch);
1084 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1085 } else {
1086 sch->qstats.backlog += pkt_len;
1087 sch->q.qlen++;
1088 }
1089}
1090
1091/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1092static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1093{
1094 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1095
1096 if (skb) {
1097 skb = __skb_dequeue(&sch->gso_skb);
1098 if (qdisc_is_percpu_stats(sch)) {
1099 qdisc_qstats_cpu_backlog_dec(sch, skb);
1100 qdisc_qstats_cpu_qlen_dec(sch);
1101 } else {
1102 qdisc_qstats_backlog_dec(sch, skb);
1103 sch->q.qlen--;
1104 }
1105 } else {
1106 skb = sch->dequeue(sch);
1107 }
1108
1109 return skb;
1110}
1111
1112static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1113{
1114 /*
1115 * We do not know the backlog in bytes of this list, it
1116 * is up to the caller to correct it
1117 */
1118 ASSERT_RTNL();
1119 if (qh->qlen) {
1120 rtnl_kfree_skbs(qh->head, qh->tail);
1121
1122 qh->head = NULL;
1123 qh->tail = NULL;
1124 qh->qlen = 0;
1125 }
1126}
1127
1128static inline void qdisc_reset_queue(struct Qdisc *sch)
1129{
1130 __qdisc_reset_queue(&sch->q);
1131}
1132
1133static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1134 struct Qdisc **pold)
1135{
1136 struct Qdisc *old;
1137
1138 sch_tree_lock(sch);
1139 old = *pold;
1140 *pold = new;
1141 if (old != NULL)
1142 qdisc_purge_queue(old);
1143 sch_tree_unlock(sch);
1144
1145 return old;
1146}
1147
1148static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1149{
1150 rtnl_kfree_skbs(skb, skb);
1151 qdisc_qstats_drop(sch);
1152}
1153
1154static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1155 struct sk_buff **to_free)
1156{
1157 __qdisc_drop(skb, to_free);
1158 qdisc_qstats_cpu_drop(sch);
1159
1160 return NET_XMIT_DROP;
1161}
1162
1163static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1164 struct sk_buff **to_free)
1165{
1166 __qdisc_drop(skb, to_free);
1167 qdisc_qstats_drop(sch);
1168
1169 return NET_XMIT_DROP;
1170}
1171
1172static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1173 struct sk_buff **to_free)
1174{
1175 __qdisc_drop_all(skb, to_free);
1176 qdisc_qstats_drop(sch);
1177
1178 return NET_XMIT_DROP;
1179}
1180
1181/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1182 long it will take to send a packet given its size.
1183 */
1184static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1185{
1186 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1187 if (slot < 0)
1188 slot = 0;
1189 slot >>= rtab->rate.cell_log;
1190 if (slot > 255)
1191 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1192 return rtab->data[slot];
1193}
1194
1195struct psched_ratecfg {
1196 u64 rate_bytes_ps; /* bytes per second */
1197 u32 mult;
1198 u16 overhead;
1199 u16 mpu;
1200 u8 linklayer;
1201 u8 shift;
1202};
1203
1204static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1205 unsigned int len)
1206{
1207 len += r->overhead;
1208
1209 if (len < r->mpu)
1210 len = r->mpu;
1211
1212 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1213 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1214
1215 return ((u64)len * r->mult) >> r->shift;
1216}
1217
1218void psched_ratecfg_precompute(struct psched_ratecfg *r,
1219 const struct tc_ratespec *conf,
1220 u64 rate64);
1221
1222static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1223 const struct psched_ratecfg *r)
1224{
1225 memset(res, 0, sizeof(*res));
1226
1227 /* legacy struct tc_ratespec has a 32bit @rate field
1228 * Qdisc using 64bit rate should add new attributes
1229 * in order to maintain compatibility.
1230 */
1231 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1232
1233 res->overhead = r->overhead;
1234 res->mpu = r->mpu;
1235 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1236}
1237
1238struct psched_pktrate {
1239 u64 rate_pkts_ps; /* packets per second */
1240 u32 mult;
1241 u8 shift;
1242};
1243
1244static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1245 unsigned int pkt_num)
1246{
1247 return ((u64)pkt_num * r->mult) >> r->shift;
1248}
1249
1250void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1251
1252/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1253 * The fast path only needs to access filter list and to update stats
1254 */
1255struct mini_Qdisc {
1256 struct tcf_proto *filter_list;
1257 struct tcf_block *block;
1258 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1259 struct gnet_stats_queue __percpu *cpu_qstats;
1260 unsigned long rcu_state;
1261};
1262
1263static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1264 const struct sk_buff *skb)
1265{
1266 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1267}
1268
1269static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1270{
1271 this_cpu_inc(miniq->cpu_qstats->drops);
1272}
1273
1274struct mini_Qdisc_pair {
1275 struct mini_Qdisc miniq1;
1276 struct mini_Qdisc miniq2;
1277 struct mini_Qdisc __rcu **p_miniq;
1278};
1279
1280void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1281 struct tcf_proto *tp_head);
1282void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1283 struct mini_Qdisc __rcu **p_miniq);
1284void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1285 struct tcf_block *block);
1286
1287void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1288
1289int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1290
1291/* Make sure qdisc is no longer in SCHED state. */
1292static inline void qdisc_synchronize(const struct Qdisc *q)
1293{
1294 while (test_bit(__QDISC_STATE_SCHED, &q->state))
1295 msleep(1);
1296}
1297
1298#endif