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#include <linux/xarray.h>
23
24struct Qdisc_ops;
25struct qdisc_walker;
26struct tcf_walker;
27struct module;
28struct bpf_flow_keys;
29
30struct qdisc_rate_table {
31 struct tc_ratespec rate;
32 u32 data[256];
33 struct qdisc_rate_table *next;
34 int refcnt;
35};
36
37enum qdisc_state_t {
38 __QDISC_STATE_SCHED,
39 __QDISC_STATE_DEACTIVATED,
40 __QDISC_STATE_MISSED,
41 __QDISC_STATE_DRAINING,
42};
43
44enum qdisc_state2_t {
45 /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly.
46 * Use qdisc_run_begin/end() or qdisc_is_running() instead.
47 */
48 __QDISC_STATE2_RUNNING,
49};
50
51#define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED)
52#define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING)
53
54#define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \
55 QDISC_STATE_DRAINING)
56
57struct qdisc_size_table {
58 struct rcu_head rcu;
59 struct list_head list;
60 struct tc_sizespec szopts;
61 int refcnt;
62 u16 data[];
63};
64
65/* similar to sk_buff_head, but skb->prev pointer is undefined. */
66struct qdisc_skb_head {
67 struct sk_buff *head;
68 struct sk_buff *tail;
69 __u32 qlen;
70 spinlock_t lock;
71};
72
73struct Qdisc {
74 int (*enqueue)(struct sk_buff *skb,
75 struct Qdisc *sch,
76 struct sk_buff **to_free);
77 struct sk_buff * (*dequeue)(struct Qdisc *sch);
78 unsigned int flags;
79#define TCQ_F_BUILTIN 1
80#define TCQ_F_INGRESS 2
81#define TCQ_F_CAN_BYPASS 4
82#define TCQ_F_MQROOT 8
83#define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
84 * q->dev_queue : It can test
85 * netif_xmit_frozen_or_stopped() before
86 * dequeueing next packet.
87 * Its true for MQ/MQPRIO slaves, or non
88 * multiqueue device.
89 */
90#define TCQ_F_WARN_NONWC (1 << 16)
91#define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
92#define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
93 * qdisc_tree_decrease_qlen() should stop.
94 */
95#define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
96#define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
97#define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
98 u32 limit;
99 const struct Qdisc_ops *ops;
100 struct qdisc_size_table __rcu *stab;
101 struct hlist_node hash;
102 u32 handle;
103 u32 parent;
104
105 struct netdev_queue *dev_queue;
106
107 struct net_rate_estimator __rcu *rate_est;
108 struct gnet_stats_basic_sync __percpu *cpu_bstats;
109 struct gnet_stats_queue __percpu *cpu_qstats;
110 int pad;
111 refcount_t refcnt;
112
113 /*
114 * For performance sake on SMP, we put highly modified fields at the end
115 */
116 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
117 struct qdisc_skb_head q;
118 struct gnet_stats_basic_sync bstats;
119 struct gnet_stats_queue qstats;
120 unsigned long state;
121 unsigned long state2; /* must be written under qdisc spinlock */
122 struct Qdisc *next_sched;
123 struct sk_buff_head skb_bad_txq;
124
125 spinlock_t busylock ____cacheline_aligned_in_smp;
126 spinlock_t seqlock;
127
128 struct rcu_head rcu;
129 netdevice_tracker dev_tracker;
130 /* private data */
131 long privdata[] ____cacheline_aligned;
132};
133
134static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
135{
136 if (qdisc->flags & TCQ_F_BUILTIN)
137 return;
138 refcount_inc(&qdisc->refcnt);
139}
140
141static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc)
142{
143 if (qdisc->flags & TCQ_F_BUILTIN)
144 return true;
145 return refcount_dec_if_one(&qdisc->refcnt);
146}
147
148/* Intended to be used by unlocked users, when concurrent qdisc release is
149 * possible.
150 */
151
152static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
153{
154 if (qdisc->flags & TCQ_F_BUILTIN)
155 return qdisc;
156 if (refcount_inc_not_zero(&qdisc->refcnt))
157 return qdisc;
158 return NULL;
159}
160
161/* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
162 * root_lock section, or provide their own memory barriers -- ordering
163 * against qdisc_run_begin/end() atomic bit operations.
164 */
165static inline bool qdisc_is_running(struct Qdisc *qdisc)
166{
167 if (qdisc->flags & TCQ_F_NOLOCK)
168 return spin_is_locked(&qdisc->seqlock);
169 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
170}
171
172static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
173{
174 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
175}
176
177static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
178{
179 return q->flags & TCQ_F_CPUSTATS;
180}
181
182static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
183{
184 if (qdisc_is_percpu_stats(qdisc))
185 return nolock_qdisc_is_empty(qdisc);
186 return !READ_ONCE(qdisc->q.qlen);
187}
188
189/* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
190 * the qdisc root lock acquired.
191 */
192static inline bool qdisc_run_begin(struct Qdisc *qdisc)
193{
194 if (qdisc->flags & TCQ_F_NOLOCK) {
195 if (spin_trylock(&qdisc->seqlock))
196 return true;
197
198 /* No need to insist if the MISSED flag was already set.
199 * Note that test_and_set_bit() also gives us memory ordering
200 * guarantees wrt potential earlier enqueue() and below
201 * spin_trylock(), both of which are necessary to prevent races
202 */
203 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
204 return false;
205
206 /* Try to take the lock again to make sure that we will either
207 * grab it or the CPU that still has it will see MISSED set
208 * when testing it in qdisc_run_end()
209 */
210 return spin_trylock(&qdisc->seqlock);
211 }
212 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
213}
214
215static inline void qdisc_run_end(struct Qdisc *qdisc)
216{
217 if (qdisc->flags & TCQ_F_NOLOCK) {
218 spin_unlock(&qdisc->seqlock);
219
220 /* spin_unlock() only has store-release semantic. The unlock
221 * and test_bit() ordering is a store-load ordering, so a full
222 * memory barrier is needed here.
223 */
224 smp_mb();
225
226 if (unlikely(test_bit(__QDISC_STATE_MISSED,
227 &qdisc->state)))
228 __netif_schedule(qdisc);
229 } else {
230 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
231 }
232}
233
234static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
235{
236 return qdisc->flags & TCQ_F_ONETXQUEUE;
237}
238
239static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
240{
241 return netdev_queue_dql_avail(txq);
242}
243
244struct Qdisc_class_ops {
245 unsigned int flags;
246 /* Child qdisc manipulation */
247 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
248 int (*graft)(struct Qdisc *, unsigned long cl,
249 struct Qdisc *, struct Qdisc **,
250 struct netlink_ext_ack *extack);
251 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
252 void (*qlen_notify)(struct Qdisc *, unsigned long);
253
254 /* Class manipulation routines */
255 unsigned long (*find)(struct Qdisc *, u32 classid);
256 int (*change)(struct Qdisc *, u32, u32,
257 struct nlattr **, unsigned long *,
258 struct netlink_ext_ack *);
259 int (*delete)(struct Qdisc *, unsigned long,
260 struct netlink_ext_ack *);
261 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
262
263 /* Filter manipulation */
264 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
265 unsigned long arg,
266 struct netlink_ext_ack *extack);
267 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
268 u32 classid);
269 void (*unbind_tcf)(struct Qdisc *, unsigned long);
270
271 /* rtnetlink specific */
272 int (*dump)(struct Qdisc *, unsigned long,
273 struct sk_buff *skb, struct tcmsg*);
274 int (*dump_stats)(struct Qdisc *, unsigned long,
275 struct gnet_dump *);
276};
277
278/* Qdisc_class_ops flag values */
279
280/* Implements API that doesn't require rtnl lock */
281enum qdisc_class_ops_flags {
282 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
283};
284
285struct Qdisc_ops {
286 struct Qdisc_ops *next;
287 const struct Qdisc_class_ops *cl_ops;
288 char id[IFNAMSIZ];
289 int priv_size;
290 unsigned int static_flags;
291
292 int (*enqueue)(struct sk_buff *skb,
293 struct Qdisc *sch,
294 struct sk_buff **to_free);
295 struct sk_buff * (*dequeue)(struct Qdisc *);
296 struct sk_buff * (*peek)(struct Qdisc *);
297
298 int (*init)(struct Qdisc *sch, struct nlattr *arg,
299 struct netlink_ext_ack *extack);
300 void (*reset)(struct Qdisc *);
301 void (*destroy)(struct Qdisc *);
302 int (*change)(struct Qdisc *sch,
303 struct nlattr *arg,
304 struct netlink_ext_ack *extack);
305 void (*attach)(struct Qdisc *sch);
306 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
307 void (*change_real_num_tx)(struct Qdisc *sch,
308 unsigned int new_real_tx);
309
310 int (*dump)(struct Qdisc *, struct sk_buff *);
311 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
312
313 void (*ingress_block_set)(struct Qdisc *sch,
314 u32 block_index);
315 void (*egress_block_set)(struct Qdisc *sch,
316 u32 block_index);
317 u32 (*ingress_block_get)(struct Qdisc *sch);
318 u32 (*egress_block_get)(struct Qdisc *sch);
319
320 struct module *owner;
321};
322
323struct tcf_result {
324 union {
325 struct {
326 unsigned long class;
327 u32 classid;
328 };
329 const struct tcf_proto *goto_tp;
330 };
331};
332
333struct tcf_chain;
334
335struct tcf_proto_ops {
336 struct list_head head;
337 char kind[IFNAMSIZ];
338
339 int (*classify)(struct sk_buff *,
340 const struct tcf_proto *,
341 struct tcf_result *);
342 int (*init)(struct tcf_proto*);
343 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
344 struct netlink_ext_ack *extack);
345
346 void* (*get)(struct tcf_proto*, u32 handle);
347 void (*put)(struct tcf_proto *tp, void *f);
348 int (*change)(struct net *net, struct sk_buff *,
349 struct tcf_proto*, unsigned long,
350 u32 handle, struct nlattr **,
351 void **, u32,
352 struct netlink_ext_ack *);
353 int (*delete)(struct tcf_proto *tp, void *arg,
354 bool *last, bool rtnl_held,
355 struct netlink_ext_ack *);
356 bool (*delete_empty)(struct tcf_proto *tp);
357 void (*walk)(struct tcf_proto *tp,
358 struct tcf_walker *arg, bool rtnl_held);
359 int (*reoffload)(struct tcf_proto *tp, bool add,
360 flow_setup_cb_t *cb, void *cb_priv,
361 struct netlink_ext_ack *extack);
362 void (*hw_add)(struct tcf_proto *tp,
363 void *type_data);
364 void (*hw_del)(struct tcf_proto *tp,
365 void *type_data);
366 void (*bind_class)(void *, u32, unsigned long,
367 void *, unsigned long);
368 void * (*tmplt_create)(struct net *net,
369 struct tcf_chain *chain,
370 struct nlattr **tca,
371 struct netlink_ext_ack *extack);
372 void (*tmplt_destroy)(void *tmplt_priv);
373 void (*tmplt_reoffload)(struct tcf_chain *chain,
374 bool add,
375 flow_setup_cb_t *cb,
376 void *cb_priv);
377 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp,
378 u32 handle);
379
380 /* rtnetlink specific */
381 int (*dump)(struct net*, struct tcf_proto*, void *,
382 struct sk_buff *skb, struct tcmsg*,
383 bool);
384 int (*terse_dump)(struct net *net,
385 struct tcf_proto *tp, void *fh,
386 struct sk_buff *skb,
387 struct tcmsg *t, bool rtnl_held);
388 int (*tmplt_dump)(struct sk_buff *skb,
389 struct net *net,
390 void *tmplt_priv);
391
392 struct module *owner;
393 int flags;
394};
395
396/* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
397 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
398 * conditions can occur when filters are inserted/deleted simultaneously.
399 */
400enum tcf_proto_ops_flags {
401 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
402};
403
404struct tcf_proto {
405 /* Fast access part */
406 struct tcf_proto __rcu *next;
407 void __rcu *root;
408
409 /* called under RCU BH lock*/
410 int (*classify)(struct sk_buff *,
411 const struct tcf_proto *,
412 struct tcf_result *);
413 __be16 protocol;
414
415 /* All the rest */
416 u32 prio;
417 void *data;
418 const struct tcf_proto_ops *ops;
419 struct tcf_chain *chain;
420 /* Lock protects tcf_proto shared state and can be used by unlocked
421 * classifiers to protect their private data.
422 */
423 spinlock_t lock;
424 bool deleting;
425 refcount_t refcnt;
426 struct rcu_head rcu;
427 struct hlist_node destroy_ht_node;
428};
429
430struct qdisc_skb_cb {
431 struct {
432 unsigned int pkt_len;
433 u16 slave_dev_queue_mapping;
434 u16 tc_classid;
435 };
436#define QDISC_CB_PRIV_LEN 20
437 unsigned char data[QDISC_CB_PRIV_LEN];
438};
439
440typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
441
442struct tcf_chain {
443 /* Protects filter_chain. */
444 struct mutex filter_chain_lock;
445 struct tcf_proto __rcu *filter_chain;
446 struct list_head list;
447 struct tcf_block *block;
448 u32 index; /* chain index */
449 unsigned int refcnt;
450 unsigned int action_refcnt;
451 bool explicitly_created;
452 bool flushing;
453 const struct tcf_proto_ops *tmplt_ops;
454 void *tmplt_priv;
455 struct rcu_head rcu;
456};
457
458struct tcf_block {
459 struct xarray ports; /* datapath accessible */
460 /* Lock protects tcf_block and lifetime-management data of chains
461 * attached to the block (refcnt, action_refcnt, explicitly_created).
462 */
463 struct mutex lock;
464 struct list_head chain_list;
465 u32 index; /* block index for shared blocks */
466 u32 classid; /* which class this block belongs to */
467 refcount_t refcnt;
468 struct net *net;
469 struct Qdisc *q;
470 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
471 struct flow_block flow_block;
472 struct list_head owner_list;
473 bool keep_dst;
474 atomic_t offloadcnt; /* Number of oddloaded filters */
475 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
476 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
477 struct {
478 struct tcf_chain *chain;
479 struct list_head filter_chain_list;
480 } chain0;
481 struct rcu_head rcu;
482 DECLARE_HASHTABLE(proto_destroy_ht, 7);
483 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
484};
485
486struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index);
487
488static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
489{
490 return lockdep_is_held(&chain->filter_chain_lock);
491}
492
493static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
494{
495 return lockdep_is_held(&tp->lock);
496}
497
498#define tcf_chain_dereference(p, chain) \
499 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
500
501#define tcf_proto_dereference(p, tp) \
502 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
503
504static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
505{
506 struct qdisc_skb_cb *qcb;
507
508 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
509 BUILD_BUG_ON(sizeof(qcb->data) < sz);
510}
511
512static inline int qdisc_qlen(const struct Qdisc *q)
513{
514 return q->q.qlen;
515}
516
517static inline int qdisc_qlen_sum(const struct Qdisc *q)
518{
519 __u32 qlen = q->qstats.qlen;
520 int i;
521
522 if (qdisc_is_percpu_stats(q)) {
523 for_each_possible_cpu(i)
524 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
525 } else {
526 qlen += q->q.qlen;
527 }
528
529 return qlen;
530}
531
532static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
533{
534 return (struct qdisc_skb_cb *)skb->cb;
535}
536
537static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
538{
539 return &qdisc->q.lock;
540}
541
542static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
543{
544 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
545
546 return q;
547}
548
549static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
550{
551 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
552}
553
554static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
555{
556 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
557}
558
559static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
560{
561 struct Qdisc *root = qdisc_root_sleeping(qdisc);
562
563 ASSERT_RTNL();
564 return qdisc_lock(root);
565}
566
567static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
568{
569 return qdisc->dev_queue->dev;
570}
571
572static inline void sch_tree_lock(struct Qdisc *q)
573{
574 if (q->flags & TCQ_F_MQROOT)
575 spin_lock_bh(qdisc_lock(q));
576 else
577 spin_lock_bh(qdisc_root_sleeping_lock(q));
578}
579
580static inline void sch_tree_unlock(struct Qdisc *q)
581{
582 if (q->flags & TCQ_F_MQROOT)
583 spin_unlock_bh(qdisc_lock(q));
584 else
585 spin_unlock_bh(qdisc_root_sleeping_lock(q));
586}
587
588extern struct Qdisc noop_qdisc;
589extern struct Qdisc_ops noop_qdisc_ops;
590extern struct Qdisc_ops pfifo_fast_ops;
591extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
592extern struct Qdisc_ops mq_qdisc_ops;
593extern struct Qdisc_ops noqueue_qdisc_ops;
594extern const struct Qdisc_ops *default_qdisc_ops;
595static inline const struct Qdisc_ops *
596get_default_qdisc_ops(const struct net_device *dev, int ntx)
597{
598 return ntx < dev->real_num_tx_queues ?
599 default_qdisc_ops : &pfifo_fast_ops;
600}
601
602struct Qdisc_class_common {
603 u32 classid;
604 unsigned int filter_cnt;
605 struct hlist_node hnode;
606};
607
608struct Qdisc_class_hash {
609 struct hlist_head *hash;
610 unsigned int hashsize;
611 unsigned int hashmask;
612 unsigned int hashelems;
613};
614
615static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
616{
617 id ^= id >> 8;
618 id ^= id >> 4;
619 return id & mask;
620}
621
622static inline struct Qdisc_class_common *
623qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
624{
625 struct Qdisc_class_common *cl;
626 unsigned int h;
627
628 if (!id)
629 return NULL;
630
631 h = qdisc_class_hash(id, hash->hashmask);
632 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
633 if (cl->classid == id)
634 return cl;
635 }
636 return NULL;
637}
638
639static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
640{
641 return cl->filter_cnt > 0;
642}
643
644static inline void qdisc_class_get(struct Qdisc_class_common *cl)
645{
646 unsigned int res;
647
648 if (check_add_overflow(cl->filter_cnt, 1, &res))
649 WARN(1, "Qdisc class overflow");
650
651 cl->filter_cnt = res;
652}
653
654static inline void qdisc_class_put(struct Qdisc_class_common *cl)
655{
656 unsigned int res;
657
658 if (check_sub_overflow(cl->filter_cnt, 1, &res))
659 WARN(1, "Qdisc class underflow");
660
661 cl->filter_cnt = res;
662}
663
664static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
665{
666 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
667
668 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
669}
670
671int qdisc_class_hash_init(struct Qdisc_class_hash *);
672void qdisc_class_hash_insert(struct Qdisc_class_hash *,
673 struct Qdisc_class_common *);
674void qdisc_class_hash_remove(struct Qdisc_class_hash *,
675 struct Qdisc_class_common *);
676void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
677void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
678
679int dev_qdisc_change_tx_queue_len(struct net_device *dev);
680void dev_qdisc_change_real_num_tx(struct net_device *dev,
681 unsigned int new_real_tx);
682void dev_init_scheduler(struct net_device *dev);
683void dev_shutdown(struct net_device *dev);
684void dev_activate(struct net_device *dev);
685void dev_deactivate(struct net_device *dev);
686void dev_deactivate_many(struct list_head *head);
687struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
688 struct Qdisc *qdisc);
689void qdisc_reset(struct Qdisc *qdisc);
690void qdisc_destroy(struct Qdisc *qdisc);
691void qdisc_put(struct Qdisc *qdisc);
692void qdisc_put_unlocked(struct Qdisc *qdisc);
693void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
694#ifdef CONFIG_NET_SCHED
695int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
696 void *type_data);
697void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
698 struct Qdisc *new, struct Qdisc *old,
699 enum tc_setup_type type, void *type_data,
700 struct netlink_ext_ack *extack);
701#else
702static inline int
703qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
704 void *type_data)
705{
706 q->flags &= ~TCQ_F_OFFLOADED;
707 return 0;
708}
709
710static inline void
711qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
712 struct Qdisc *new, struct Qdisc *old,
713 enum tc_setup_type type, void *type_data,
714 struct netlink_ext_ack *extack)
715{
716}
717#endif
718void qdisc_offload_query_caps(struct net_device *dev,
719 enum tc_setup_type type,
720 void *caps, size_t caps_len);
721struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
722 const struct Qdisc_ops *ops,
723 struct netlink_ext_ack *extack);
724void qdisc_free(struct Qdisc *qdisc);
725struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
726 const struct Qdisc_ops *ops, u32 parentid,
727 struct netlink_ext_ack *extack);
728void __qdisc_calculate_pkt_len(struct sk_buff *skb,
729 const struct qdisc_size_table *stab);
730int skb_do_redirect(struct sk_buff *);
731
732static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
733{
734#ifdef CONFIG_NET_XGRESS
735 return skb->tc_at_ingress;
736#else
737 return false;
738#endif
739}
740
741static inline bool skb_skip_tc_classify(struct sk_buff *skb)
742{
743#ifdef CONFIG_NET_CLS_ACT
744 if (skb->tc_skip_classify) {
745 skb->tc_skip_classify = 0;
746 return true;
747 }
748#endif
749 return false;
750}
751
752/* Reset all TX qdiscs greater than index of a device. */
753static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
754{
755 struct Qdisc *qdisc;
756
757 for (; i < dev->num_tx_queues; i++) {
758 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
759 if (qdisc) {
760 spin_lock_bh(qdisc_lock(qdisc));
761 qdisc_reset(qdisc);
762 spin_unlock_bh(qdisc_lock(qdisc));
763 }
764 }
765}
766
767/* Are all TX queues of the device empty? */
768static inline bool qdisc_all_tx_empty(const struct net_device *dev)
769{
770 unsigned int i;
771
772 rcu_read_lock();
773 for (i = 0; i < dev->num_tx_queues; i++) {
774 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
775 const struct Qdisc *q = rcu_dereference(txq->qdisc);
776
777 if (!qdisc_is_empty(q)) {
778 rcu_read_unlock();
779 return false;
780 }
781 }
782 rcu_read_unlock();
783 return true;
784}
785
786/* Are any of the TX qdiscs changing? */
787static inline bool qdisc_tx_changing(const struct net_device *dev)
788{
789 unsigned int i;
790
791 for (i = 0; i < dev->num_tx_queues; i++) {
792 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
793
794 if (rcu_access_pointer(txq->qdisc) !=
795 rcu_access_pointer(txq->qdisc_sleeping))
796 return true;
797 }
798 return false;
799}
800
801/* Is the device using the noop qdisc on all queues? */
802static inline bool qdisc_tx_is_noop(const struct net_device *dev)
803{
804 unsigned int i;
805
806 for (i = 0; i < dev->num_tx_queues; i++) {
807 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
808 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
809 return false;
810 }
811 return true;
812}
813
814static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
815{
816 return qdisc_skb_cb(skb)->pkt_len;
817}
818
819/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
820enum net_xmit_qdisc_t {
821 __NET_XMIT_STOLEN = 0x00010000,
822 __NET_XMIT_BYPASS = 0x00020000,
823};
824
825#ifdef CONFIG_NET_CLS_ACT
826#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
827#else
828#define net_xmit_drop_count(e) (1)
829#endif
830
831static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
832 const struct Qdisc *sch)
833{
834#ifdef CONFIG_NET_SCHED
835 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
836
837 if (stab)
838 __qdisc_calculate_pkt_len(skb, stab);
839#endif
840}
841
842static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
843 struct sk_buff **to_free)
844{
845 qdisc_calculate_pkt_len(skb, sch);
846 return sch->enqueue(skb, sch, to_free);
847}
848
849static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
850 __u64 bytes, __u32 packets)
851{
852 u64_stats_update_begin(&bstats->syncp);
853 u64_stats_add(&bstats->bytes, bytes);
854 u64_stats_add(&bstats->packets, packets);
855 u64_stats_update_end(&bstats->syncp);
856}
857
858static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
859 const struct sk_buff *skb)
860{
861 _bstats_update(bstats,
862 qdisc_pkt_len(skb),
863 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
864}
865
866static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
867 const struct sk_buff *skb)
868{
869 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
870}
871
872static inline void qdisc_bstats_update(struct Qdisc *sch,
873 const struct sk_buff *skb)
874{
875 bstats_update(&sch->bstats, skb);
876}
877
878static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
879 const struct sk_buff *skb)
880{
881 sch->qstats.backlog -= qdisc_pkt_len(skb);
882}
883
884static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
885 const struct sk_buff *skb)
886{
887 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
888}
889
890static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
891 const struct sk_buff *skb)
892{
893 sch->qstats.backlog += qdisc_pkt_len(skb);
894}
895
896static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
897 const struct sk_buff *skb)
898{
899 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
900}
901
902static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
903{
904 this_cpu_inc(sch->cpu_qstats->qlen);
905}
906
907static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
908{
909 this_cpu_dec(sch->cpu_qstats->qlen);
910}
911
912static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
913{
914 this_cpu_inc(sch->cpu_qstats->requeues);
915}
916
917static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
918{
919 sch->qstats.drops += count;
920}
921
922static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
923{
924 qstats->drops++;
925}
926
927static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
928{
929 qstats->overlimits++;
930}
931
932static inline void qdisc_qstats_drop(struct Qdisc *sch)
933{
934 qstats_drop_inc(&sch->qstats);
935}
936
937static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
938{
939 this_cpu_inc(sch->cpu_qstats->drops);
940}
941
942static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
943{
944 sch->qstats.overlimits++;
945}
946
947static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
948{
949 __u32 qlen = qdisc_qlen_sum(sch);
950
951 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
952}
953
954static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
955 __u32 *backlog)
956{
957 struct gnet_stats_queue qstats = { 0 };
958
959 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
960 *qlen = qstats.qlen + qdisc_qlen(sch);
961 *backlog = qstats.backlog;
962}
963
964static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
965{
966 __u32 qlen, backlog;
967
968 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
969 qdisc_tree_reduce_backlog(sch, qlen, backlog);
970}
971
972static inline void qdisc_purge_queue(struct Qdisc *sch)
973{
974 __u32 qlen, backlog;
975
976 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
977 qdisc_reset(sch);
978 qdisc_tree_reduce_backlog(sch, qlen, backlog);
979}
980
981static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
982 struct qdisc_skb_head *qh)
983{
984 struct sk_buff *last = qh->tail;
985
986 if (last) {
987 skb->next = NULL;
988 last->next = skb;
989 qh->tail = skb;
990 } else {
991 qh->tail = skb;
992 qh->head = skb;
993 }
994 qh->qlen++;
995}
996
997static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
998{
999 __qdisc_enqueue_tail(skb, &sch->q);
1000 qdisc_qstats_backlog_inc(sch, skb);
1001 return NET_XMIT_SUCCESS;
1002}
1003
1004static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1005 struct qdisc_skb_head *qh)
1006{
1007 skb->next = qh->head;
1008
1009 if (!qh->head)
1010 qh->tail = skb;
1011 qh->head = skb;
1012 qh->qlen++;
1013}
1014
1015static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1016{
1017 struct sk_buff *skb = qh->head;
1018
1019 if (likely(skb != NULL)) {
1020 qh->head = skb->next;
1021 qh->qlen--;
1022 if (qh->head == NULL)
1023 qh->tail = NULL;
1024 skb->next = NULL;
1025 }
1026
1027 return skb;
1028}
1029
1030static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1031{
1032 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1033
1034 if (likely(skb != NULL)) {
1035 qdisc_qstats_backlog_dec(sch, skb);
1036 qdisc_bstats_update(sch, skb);
1037 }
1038
1039 return skb;
1040}
1041
1042struct tc_skb_cb {
1043 struct qdisc_skb_cb qdisc_cb;
1044 u32 drop_reason;
1045
1046 u16 zone; /* Only valid if post_ct = true */
1047 u16 mru;
1048 u8 post_ct:1;
1049 u8 post_ct_snat:1;
1050 u8 post_ct_dnat:1;
1051};
1052
1053static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
1054{
1055 struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
1056
1057 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
1058 return cb;
1059}
1060
1061static inline enum skb_drop_reason
1062tcf_get_drop_reason(const struct sk_buff *skb)
1063{
1064 return tc_skb_cb(skb)->drop_reason;
1065}
1066
1067static inline void tcf_set_drop_reason(const struct sk_buff *skb,
1068 enum skb_drop_reason reason)
1069{
1070 tc_skb_cb(skb)->drop_reason = reason;
1071}
1072
1073/* Instead of calling kfree_skb() while root qdisc lock is held,
1074 * queue the skb for future freeing at end of __dev_xmit_skb()
1075 */
1076static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1077{
1078 skb->next = *to_free;
1079 *to_free = skb;
1080}
1081
1082static inline void __qdisc_drop_all(struct sk_buff *skb,
1083 struct sk_buff **to_free)
1084{
1085 if (skb->prev)
1086 skb->prev->next = *to_free;
1087 else
1088 skb->next = *to_free;
1089 *to_free = skb;
1090}
1091
1092static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1093 struct qdisc_skb_head *qh,
1094 struct sk_buff **to_free)
1095{
1096 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1097
1098 if (likely(skb != NULL)) {
1099 unsigned int len = qdisc_pkt_len(skb);
1100
1101 qdisc_qstats_backlog_dec(sch, skb);
1102 __qdisc_drop(skb, to_free);
1103 return len;
1104 }
1105
1106 return 0;
1107}
1108
1109static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1110{
1111 const struct qdisc_skb_head *qh = &sch->q;
1112
1113 return qh->head;
1114}
1115
1116/* generic pseudo peek method for non-work-conserving qdisc */
1117static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1118{
1119 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1120
1121 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1122 if (!skb) {
1123 skb = sch->dequeue(sch);
1124
1125 if (skb) {
1126 __skb_queue_head(&sch->gso_skb, skb);
1127 /* it's still part of the queue */
1128 qdisc_qstats_backlog_inc(sch, skb);
1129 sch->q.qlen++;
1130 }
1131 }
1132
1133 return skb;
1134}
1135
1136static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1137 struct sk_buff *skb)
1138{
1139 if (qdisc_is_percpu_stats(sch)) {
1140 qdisc_qstats_cpu_backlog_dec(sch, skb);
1141 qdisc_bstats_cpu_update(sch, skb);
1142 qdisc_qstats_cpu_qlen_dec(sch);
1143 } else {
1144 qdisc_qstats_backlog_dec(sch, skb);
1145 qdisc_bstats_update(sch, skb);
1146 sch->q.qlen--;
1147 }
1148}
1149
1150static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1151 unsigned int pkt_len)
1152{
1153 if (qdisc_is_percpu_stats(sch)) {
1154 qdisc_qstats_cpu_qlen_inc(sch);
1155 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1156 } else {
1157 sch->qstats.backlog += pkt_len;
1158 sch->q.qlen++;
1159 }
1160}
1161
1162/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1163static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1164{
1165 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1166
1167 if (skb) {
1168 skb = __skb_dequeue(&sch->gso_skb);
1169 if (qdisc_is_percpu_stats(sch)) {
1170 qdisc_qstats_cpu_backlog_dec(sch, skb);
1171 qdisc_qstats_cpu_qlen_dec(sch);
1172 } else {
1173 qdisc_qstats_backlog_dec(sch, skb);
1174 sch->q.qlen--;
1175 }
1176 } else {
1177 skb = sch->dequeue(sch);
1178 }
1179
1180 return skb;
1181}
1182
1183static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1184{
1185 /*
1186 * We do not know the backlog in bytes of this list, it
1187 * is up to the caller to correct it
1188 */
1189 ASSERT_RTNL();
1190 if (qh->qlen) {
1191 rtnl_kfree_skbs(qh->head, qh->tail);
1192
1193 qh->head = NULL;
1194 qh->tail = NULL;
1195 qh->qlen = 0;
1196 }
1197}
1198
1199static inline void qdisc_reset_queue(struct Qdisc *sch)
1200{
1201 __qdisc_reset_queue(&sch->q);
1202}
1203
1204static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1205 struct Qdisc **pold)
1206{
1207 struct Qdisc *old;
1208
1209 sch_tree_lock(sch);
1210 old = *pold;
1211 *pold = new;
1212 if (old != NULL)
1213 qdisc_purge_queue(old);
1214 sch_tree_unlock(sch);
1215
1216 return old;
1217}
1218
1219static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1220{
1221 rtnl_kfree_skbs(skb, skb);
1222 qdisc_qstats_drop(sch);
1223}
1224
1225static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1226 struct sk_buff **to_free)
1227{
1228 __qdisc_drop(skb, to_free);
1229 qdisc_qstats_cpu_drop(sch);
1230
1231 return NET_XMIT_DROP;
1232}
1233
1234static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1235 struct sk_buff **to_free)
1236{
1237 __qdisc_drop(skb, to_free);
1238 qdisc_qstats_drop(sch);
1239
1240 return NET_XMIT_DROP;
1241}
1242
1243static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1244 struct sk_buff **to_free)
1245{
1246 __qdisc_drop_all(skb, to_free);
1247 qdisc_qstats_drop(sch);
1248
1249 return NET_XMIT_DROP;
1250}
1251
1252struct psched_ratecfg {
1253 u64 rate_bytes_ps; /* bytes per second */
1254 u32 mult;
1255 u16 overhead;
1256 u16 mpu;
1257 u8 linklayer;
1258 u8 shift;
1259};
1260
1261static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1262 unsigned int len)
1263{
1264 len += r->overhead;
1265
1266 if (len < r->mpu)
1267 len = r->mpu;
1268
1269 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1270 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1271
1272 return ((u64)len * r->mult) >> r->shift;
1273}
1274
1275void psched_ratecfg_precompute(struct psched_ratecfg *r,
1276 const struct tc_ratespec *conf,
1277 u64 rate64);
1278
1279static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1280 const struct psched_ratecfg *r)
1281{
1282 memset(res, 0, sizeof(*res));
1283
1284 /* legacy struct tc_ratespec has a 32bit @rate field
1285 * Qdisc using 64bit rate should add new attributes
1286 * in order to maintain compatibility.
1287 */
1288 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1289
1290 res->overhead = r->overhead;
1291 res->mpu = r->mpu;
1292 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1293}
1294
1295struct psched_pktrate {
1296 u64 rate_pkts_ps; /* packets per second */
1297 u32 mult;
1298 u8 shift;
1299};
1300
1301static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1302 unsigned int pkt_num)
1303{
1304 return ((u64)pkt_num * r->mult) >> r->shift;
1305}
1306
1307void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1308
1309/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1310 * The fast path only needs to access filter list and to update stats
1311 */
1312struct mini_Qdisc {
1313 struct tcf_proto *filter_list;
1314 struct tcf_block *block;
1315 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1316 struct gnet_stats_queue __percpu *cpu_qstats;
1317 unsigned long rcu_state;
1318};
1319
1320static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1321 const struct sk_buff *skb)
1322{
1323 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1324}
1325
1326static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1327{
1328 this_cpu_inc(miniq->cpu_qstats->drops);
1329}
1330
1331struct mini_Qdisc_pair {
1332 struct mini_Qdisc miniq1;
1333 struct mini_Qdisc miniq2;
1334 struct mini_Qdisc __rcu **p_miniq;
1335};
1336
1337void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1338 struct tcf_proto *tp_head);
1339void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1340 struct mini_Qdisc __rcu **p_miniq);
1341void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1342 struct tcf_block *block);
1343
1344void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1345
1346int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1347
1348/* Make sure qdisc is no longer in SCHED state. */
1349static inline void qdisc_synchronize(const struct Qdisc *q)
1350{
1351 while (test_bit(__QDISC_STATE_SCHED, &q->state))
1352 msleep(1);
1353}
1354
1355#endif