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 __QDISC_STATE_MISSED,
40 __QDISC_STATE_DRAINING,
41};
42
43#define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED)
44#define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING)
45
46#define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \
47 QDISC_STATE_DRAINING)
48
49struct qdisc_size_table {
50 struct rcu_head rcu;
51 struct list_head list;
52 struct tc_sizespec szopts;
53 int refcnt;
54 u16 data[];
55};
56
57/* similar to sk_buff_head, but skb->prev pointer is undefined. */
58struct qdisc_skb_head {
59 struct sk_buff *head;
60 struct sk_buff *tail;
61 __u32 qlen;
62 spinlock_t lock;
63};
64
65struct Qdisc {
66 int (*enqueue)(struct sk_buff *skb,
67 struct Qdisc *sch,
68 struct sk_buff **to_free);
69 struct sk_buff * (*dequeue)(struct Qdisc *sch);
70 unsigned int flags;
71#define TCQ_F_BUILTIN 1
72#define TCQ_F_INGRESS 2
73#define TCQ_F_CAN_BYPASS 4
74#define TCQ_F_MQROOT 8
75#define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
76 * q->dev_queue : It can test
77 * netif_xmit_frozen_or_stopped() before
78 * dequeueing next packet.
79 * Its true for MQ/MQPRIO slaves, or non
80 * multiqueue device.
81 */
82#define TCQ_F_WARN_NONWC (1 << 16)
83#define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
84#define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
85 * qdisc_tree_decrease_qlen() should stop.
86 */
87#define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
88#define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
89#define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
90 u32 limit;
91 const struct Qdisc_ops *ops;
92 struct qdisc_size_table __rcu *stab;
93 struct hlist_node hash;
94 u32 handle;
95 u32 parent;
96
97 struct netdev_queue *dev_queue;
98
99 struct net_rate_estimator __rcu *rate_est;
100 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
101 struct gnet_stats_queue __percpu *cpu_qstats;
102 int pad;
103 refcount_t refcnt;
104
105 /*
106 * For performance sake on SMP, we put highly modified fields at the end
107 */
108 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp;
109 struct qdisc_skb_head q;
110 struct gnet_stats_basic_packed bstats;
111 seqcount_t running;
112 struct gnet_stats_queue qstats;
113 unsigned long state;
114 struct Qdisc *next_sched;
115 struct sk_buff_head skb_bad_txq;
116
117 spinlock_t busylock ____cacheline_aligned_in_smp;
118 spinlock_t seqlock;
119
120 struct rcu_head rcu;
121
122 /* private data */
123 long privdata[] ____cacheline_aligned;
124};
125
126static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
127{
128 if (qdisc->flags & TCQ_F_BUILTIN)
129 return;
130 refcount_inc(&qdisc->refcnt);
131}
132
133/* Intended to be used by unlocked users, when concurrent qdisc release is
134 * possible.
135 */
136
137static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
138{
139 if (qdisc->flags & TCQ_F_BUILTIN)
140 return qdisc;
141 if (refcount_inc_not_zero(&qdisc->refcnt))
142 return qdisc;
143 return NULL;
144}
145
146static inline bool qdisc_is_running(struct Qdisc *qdisc)
147{
148 if (qdisc->flags & TCQ_F_NOLOCK)
149 return spin_is_locked(&qdisc->seqlock);
150 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
151}
152
153static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
154{
155 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
156}
157
158static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
159{
160 return q->flags & TCQ_F_CPUSTATS;
161}
162
163static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
164{
165 if (qdisc_is_percpu_stats(qdisc))
166 return nolock_qdisc_is_empty(qdisc);
167 return !READ_ONCE(qdisc->q.qlen);
168}
169
170static inline bool qdisc_run_begin(struct Qdisc *qdisc)
171{
172 if (qdisc->flags & TCQ_F_NOLOCK) {
173 if (spin_trylock(&qdisc->seqlock))
174 return true;
175
176 /* Paired with smp_mb__after_atomic() to make sure
177 * STATE_MISSED checking is synchronized with clearing
178 * in pfifo_fast_dequeue().
179 */
180 smp_mb__before_atomic();
181
182 /* If the MISSED flag is set, it means other thread has
183 * set the MISSED flag before second spin_trylock(), so
184 * we can return false here to avoid multi cpus doing
185 * the set_bit() and second spin_trylock() concurrently.
186 */
187 if (test_bit(__QDISC_STATE_MISSED, &qdisc->state))
188 return false;
189
190 /* Set the MISSED flag before the second spin_trylock(),
191 * if the second spin_trylock() return false, it means
192 * other cpu holding the lock will do dequeuing for us
193 * or it will see the MISSED flag set after releasing
194 * lock and reschedule the net_tx_action() to do the
195 * dequeuing.
196 */
197 set_bit(__QDISC_STATE_MISSED, &qdisc->state);
198
199 /* spin_trylock() only has load-acquire semantic, so use
200 * smp_mb__after_atomic() to ensure STATE_MISSED is set
201 * before doing the second spin_trylock().
202 */
203 smp_mb__after_atomic();
204
205 /* Retry again in case other CPU may not see the new flag
206 * after it releases the lock at the end of qdisc_run_end().
207 */
208 return spin_trylock(&qdisc->seqlock);
209 } else if (qdisc_is_running(qdisc)) {
210 return false;
211 }
212 /* Variant of write_seqcount_begin() telling lockdep a trylock
213 * was attempted.
214 */
215 raw_write_seqcount_begin(&qdisc->running);
216 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
217 return true;
218}
219
220static inline void qdisc_run_end(struct Qdisc *qdisc)
221{
222 if (qdisc->flags & TCQ_F_NOLOCK) {
223 spin_unlock(&qdisc->seqlock);
224
225 if (unlikely(test_bit(__QDISC_STATE_MISSED,
226 &qdisc->state)))
227 __netif_schedule(qdisc);
228 } else {
229 write_seqcount_end(&qdisc->running);
230 }
231}
232
233static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
234{
235 return qdisc->flags & TCQ_F_ONETXQUEUE;
236}
237
238static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
239{
240#ifdef CONFIG_BQL
241 /* Non-BQL migrated drivers will return 0, too. */
242 return dql_avail(&txq->dql);
243#else
244 return 0;
245#endif
246}
247
248struct Qdisc_class_ops {
249 unsigned int flags;
250 /* Child qdisc manipulation */
251 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
252 int (*graft)(struct Qdisc *, unsigned long cl,
253 struct Qdisc *, struct Qdisc **,
254 struct netlink_ext_ack *extack);
255 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
256 void (*qlen_notify)(struct Qdisc *, unsigned long);
257
258 /* Class manipulation routines */
259 unsigned long (*find)(struct Qdisc *, u32 classid);
260 int (*change)(struct Qdisc *, u32, u32,
261 struct nlattr **, unsigned long *,
262 struct netlink_ext_ack *);
263 int (*delete)(struct Qdisc *, unsigned long,
264 struct netlink_ext_ack *);
265 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
266
267 /* Filter manipulation */
268 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
269 unsigned long arg,
270 struct netlink_ext_ack *extack);
271 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
272 u32 classid);
273 void (*unbind_tcf)(struct Qdisc *, unsigned long);
274
275 /* rtnetlink specific */
276 int (*dump)(struct Qdisc *, unsigned long,
277 struct sk_buff *skb, struct tcmsg*);
278 int (*dump_stats)(struct Qdisc *, unsigned long,
279 struct gnet_dump *);
280};
281
282/* Qdisc_class_ops flag values */
283
284/* Implements API that doesn't require rtnl lock */
285enum qdisc_class_ops_flags {
286 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
287};
288
289struct Qdisc_ops {
290 struct Qdisc_ops *next;
291 const struct Qdisc_class_ops *cl_ops;
292 char id[IFNAMSIZ];
293 int priv_size;
294 unsigned int static_flags;
295
296 int (*enqueue)(struct sk_buff *skb,
297 struct Qdisc *sch,
298 struct sk_buff **to_free);
299 struct sk_buff * (*dequeue)(struct Qdisc *);
300 struct sk_buff * (*peek)(struct Qdisc *);
301
302 int (*init)(struct Qdisc *sch, struct nlattr *arg,
303 struct netlink_ext_ack *extack);
304 void (*reset)(struct Qdisc *);
305 void (*destroy)(struct Qdisc *);
306 int (*change)(struct Qdisc *sch,
307 struct nlattr *arg,
308 struct netlink_ext_ack *extack);
309 void (*attach)(struct Qdisc *sch);
310 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
311
312 int (*dump)(struct Qdisc *, struct sk_buff *);
313 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
314
315 void (*ingress_block_set)(struct Qdisc *sch,
316 u32 block_index);
317 void (*egress_block_set)(struct Qdisc *sch,
318 u32 block_index);
319 u32 (*ingress_block_get)(struct Qdisc *sch);
320 u32 (*egress_block_get)(struct Qdisc *sch);
321
322 struct module *owner;
323};
324
325
326struct tcf_result {
327 union {
328 struct {
329 unsigned long class;
330 u32 classid;
331 };
332 const struct tcf_proto *goto_tp;
333
334 /* used in the skb_tc_reinsert function */
335 struct {
336 bool ingress;
337 struct gnet_stats_queue *qstats;
338 };
339 };
340};
341
342struct tcf_chain;
343
344struct tcf_proto_ops {
345 struct list_head head;
346 char kind[IFNAMSIZ];
347
348 int (*classify)(struct sk_buff *,
349 const struct tcf_proto *,
350 struct tcf_result *);
351 int (*init)(struct tcf_proto*);
352 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
353 struct netlink_ext_ack *extack);
354
355 void* (*get)(struct tcf_proto*, u32 handle);
356 void (*put)(struct tcf_proto *tp, void *f);
357 int (*change)(struct net *net, struct sk_buff *,
358 struct tcf_proto*, unsigned long,
359 u32 handle, struct nlattr **,
360 void **, bool, bool,
361 struct netlink_ext_ack *);
362 int (*delete)(struct tcf_proto *tp, void *arg,
363 bool *last, bool rtnl_held,
364 struct netlink_ext_ack *);
365 bool (*delete_empty)(struct tcf_proto *tp);
366 void (*walk)(struct tcf_proto *tp,
367 struct tcf_walker *arg, bool rtnl_held);
368 int (*reoffload)(struct tcf_proto *tp, bool add,
369 flow_setup_cb_t *cb, void *cb_priv,
370 struct netlink_ext_ack *extack);
371 void (*hw_add)(struct tcf_proto *tp,
372 void *type_data);
373 void (*hw_del)(struct tcf_proto *tp,
374 void *type_data);
375 void (*bind_class)(void *, u32, unsigned long,
376 void *, unsigned long);
377 void * (*tmplt_create)(struct net *net,
378 struct tcf_chain *chain,
379 struct nlattr **tca,
380 struct netlink_ext_ack *extack);
381 void (*tmplt_destroy)(void *tmplt_priv);
382
383 /* rtnetlink specific */
384 int (*dump)(struct net*, struct tcf_proto*, void *,
385 struct sk_buff *skb, struct tcmsg*,
386 bool);
387 int (*terse_dump)(struct net *net,
388 struct tcf_proto *tp, void *fh,
389 struct sk_buff *skb,
390 struct tcmsg *t, bool rtnl_held);
391 int (*tmplt_dump)(struct sk_buff *skb,
392 struct net *net,
393 void *tmplt_priv);
394
395 struct module *owner;
396 int flags;
397};
398
399/* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
400 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
401 * conditions can occur when filters are inserted/deleted simultaneously.
402 */
403enum tcf_proto_ops_flags {
404 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
405};
406
407struct tcf_proto {
408 /* Fast access part */
409 struct tcf_proto __rcu *next;
410 void __rcu *root;
411
412 /* called under RCU BH lock*/
413 int (*classify)(struct sk_buff *,
414 const struct tcf_proto *,
415 struct tcf_result *);
416 __be16 protocol;
417
418 /* All the rest */
419 u32 prio;
420 void *data;
421 const struct tcf_proto_ops *ops;
422 struct tcf_chain *chain;
423 /* Lock protects tcf_proto shared state and can be used by unlocked
424 * classifiers to protect their private data.
425 */
426 spinlock_t lock;
427 bool deleting;
428 refcount_t refcnt;
429 struct rcu_head rcu;
430 struct hlist_node destroy_ht_node;
431};
432
433struct qdisc_skb_cb {
434 struct {
435 unsigned int pkt_len;
436 u16 slave_dev_queue_mapping;
437 u16 tc_classid;
438 };
439#define QDISC_CB_PRIV_LEN 20
440 unsigned char data[QDISC_CB_PRIV_LEN];
441 u16 mru;
442 bool post_ct;
443};
444
445typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
446
447struct tcf_chain {
448 /* Protects filter_chain. */
449 struct mutex filter_chain_lock;
450 struct tcf_proto __rcu *filter_chain;
451 struct list_head list;
452 struct tcf_block *block;
453 u32 index; /* chain index */
454 unsigned int refcnt;
455 unsigned int action_refcnt;
456 bool explicitly_created;
457 bool flushing;
458 const struct tcf_proto_ops *tmplt_ops;
459 void *tmplt_priv;
460 struct rcu_head rcu;
461};
462
463struct tcf_block {
464 /* Lock protects tcf_block and lifetime-management data of chains
465 * attached to the block (refcnt, action_refcnt, explicitly_created).
466 */
467 struct mutex lock;
468 struct list_head chain_list;
469 u32 index; /* block index for shared blocks */
470 u32 classid; /* which class this block belongs to */
471 refcount_t refcnt;
472 struct net *net;
473 struct Qdisc *q;
474 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
475 struct flow_block flow_block;
476 struct list_head owner_list;
477 bool keep_dst;
478 atomic_t offloadcnt; /* Number of oddloaded filters */
479 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
480 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
481 struct {
482 struct tcf_chain *chain;
483 struct list_head filter_chain_list;
484 } chain0;
485 struct rcu_head rcu;
486 DECLARE_HASHTABLE(proto_destroy_ht, 7);
487 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
488};
489
490static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
491{
492 return lockdep_is_held(&chain->filter_chain_lock);
493}
494
495static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
496{
497 return lockdep_is_held(&tp->lock);
498}
499
500#define tcf_chain_dereference(p, chain) \
501 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
502
503#define tcf_proto_dereference(p, tp) \
504 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
505
506static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
507{
508 struct qdisc_skb_cb *qcb;
509
510 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
511 BUILD_BUG_ON(sizeof(qcb->data) < sz);
512}
513
514static inline int qdisc_qlen_cpu(const struct Qdisc *q)
515{
516 return this_cpu_ptr(q->cpu_qstats)->qlen;
517}
518
519static inline int qdisc_qlen(const struct Qdisc *q)
520{
521 return q->q.qlen;
522}
523
524static inline int qdisc_qlen_sum(const struct Qdisc *q)
525{
526 __u32 qlen = q->qstats.qlen;
527 int i;
528
529 if (qdisc_is_percpu_stats(q)) {
530 for_each_possible_cpu(i)
531 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
532 } else {
533 qlen += q->q.qlen;
534 }
535
536 return qlen;
537}
538
539static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
540{
541 return (struct qdisc_skb_cb *)skb->cb;
542}
543
544static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
545{
546 return &qdisc->q.lock;
547}
548
549static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
550{
551 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
552
553 return q;
554}
555
556static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
557{
558 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
559}
560
561static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
562{
563 return qdisc->dev_queue->qdisc_sleeping;
564}
565
566/* The qdisc root lock is a mechanism by which to top level
567 * of a qdisc tree can be locked from any qdisc node in the
568 * forest. This allows changing the configuration of some
569 * aspect of the qdisc tree while blocking out asynchronous
570 * qdisc access in the packet processing paths.
571 *
572 * It is only legal to do this when the root will not change
573 * on us. Otherwise we'll potentially lock the wrong qdisc
574 * root. This is enforced by holding the RTNL semaphore, which
575 * all users of this lock accessor must do.
576 */
577static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
578{
579 struct Qdisc *root = qdisc_root(qdisc);
580
581 ASSERT_RTNL();
582 return qdisc_lock(root);
583}
584
585static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
586{
587 struct Qdisc *root = qdisc_root_sleeping(qdisc);
588
589 ASSERT_RTNL();
590 return qdisc_lock(root);
591}
592
593static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
594{
595 struct Qdisc *root = qdisc_root_sleeping(qdisc);
596
597 ASSERT_RTNL();
598 return &root->running;
599}
600
601static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
602{
603 return qdisc->dev_queue->dev;
604}
605
606static inline void sch_tree_lock(struct Qdisc *q)
607{
608 if (q->flags & TCQ_F_MQROOT)
609 spin_lock_bh(qdisc_lock(q));
610 else
611 spin_lock_bh(qdisc_root_sleeping_lock(q));
612}
613
614static inline void sch_tree_unlock(struct Qdisc *q)
615{
616 if (q->flags & TCQ_F_MQROOT)
617 spin_unlock_bh(qdisc_lock(q));
618 else
619 spin_unlock_bh(qdisc_root_sleeping_lock(q));
620}
621
622extern struct Qdisc noop_qdisc;
623extern struct Qdisc_ops noop_qdisc_ops;
624extern struct Qdisc_ops pfifo_fast_ops;
625extern struct Qdisc_ops mq_qdisc_ops;
626extern struct Qdisc_ops noqueue_qdisc_ops;
627extern const struct Qdisc_ops *default_qdisc_ops;
628static inline const struct Qdisc_ops *
629get_default_qdisc_ops(const struct net_device *dev, int ntx)
630{
631 return ntx < dev->real_num_tx_queues ?
632 default_qdisc_ops : &pfifo_fast_ops;
633}
634
635struct Qdisc_class_common {
636 u32 classid;
637 struct hlist_node hnode;
638};
639
640struct Qdisc_class_hash {
641 struct hlist_head *hash;
642 unsigned int hashsize;
643 unsigned int hashmask;
644 unsigned int hashelems;
645};
646
647static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
648{
649 id ^= id >> 8;
650 id ^= id >> 4;
651 return id & mask;
652}
653
654static inline struct Qdisc_class_common *
655qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
656{
657 struct Qdisc_class_common *cl;
658 unsigned int h;
659
660 if (!id)
661 return NULL;
662
663 h = qdisc_class_hash(id, hash->hashmask);
664 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
665 if (cl->classid == id)
666 return cl;
667 }
668 return NULL;
669}
670
671static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
672{
673 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
674
675 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
676}
677
678int qdisc_class_hash_init(struct Qdisc_class_hash *);
679void qdisc_class_hash_insert(struct Qdisc_class_hash *,
680 struct Qdisc_class_common *);
681void qdisc_class_hash_remove(struct Qdisc_class_hash *,
682 struct Qdisc_class_common *);
683void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
684void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
685
686int dev_qdisc_change_tx_queue_len(struct net_device *dev);
687void dev_init_scheduler(struct net_device *dev);
688void dev_shutdown(struct net_device *dev);
689void dev_activate(struct net_device *dev);
690void dev_deactivate(struct net_device *dev);
691void dev_deactivate_many(struct list_head *head);
692struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
693 struct Qdisc *qdisc);
694void qdisc_reset(struct Qdisc *qdisc);
695void qdisc_put(struct Qdisc *qdisc);
696void qdisc_put_unlocked(struct Qdisc *qdisc);
697void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
698#ifdef CONFIG_NET_SCHED
699int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
700 void *type_data);
701void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
702 struct Qdisc *new, struct Qdisc *old,
703 enum tc_setup_type type, void *type_data,
704 struct netlink_ext_ack *extack);
705#else
706static inline int
707qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
708 void *type_data)
709{
710 q->flags &= ~TCQ_F_OFFLOADED;
711 return 0;
712}
713
714static inline void
715qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
716 struct Qdisc *new, struct Qdisc *old,
717 enum tc_setup_type type, void *type_data,
718 struct netlink_ext_ack *extack)
719{
720}
721#endif
722struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
723 const struct Qdisc_ops *ops,
724 struct netlink_ext_ack *extack);
725void qdisc_free(struct Qdisc *qdisc);
726struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
727 const struct Qdisc_ops *ops, u32 parentid,
728 struct netlink_ext_ack *extack);
729void __qdisc_calculate_pkt_len(struct sk_buff *skb,
730 const struct qdisc_size_table *stab);
731int skb_do_redirect(struct sk_buff *);
732
733static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
734{
735#ifdef CONFIG_NET_CLS_ACT
736 return skb->tc_at_ingress;
737#else
738 return false;
739#endif
740}
741
742static inline bool skb_skip_tc_classify(struct sk_buff *skb)
743{
744#ifdef CONFIG_NET_CLS_ACT
745 if (skb->tc_skip_classify) {
746 skb->tc_skip_classify = 0;
747 return true;
748 }
749#endif
750 return false;
751}
752
753/* Reset all TX qdiscs greater than index of a device. */
754static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
755{
756 struct Qdisc *qdisc;
757
758 for (; i < dev->num_tx_queues; i++) {
759 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
760 if (qdisc) {
761 spin_lock_bh(qdisc_lock(qdisc));
762 qdisc_reset(qdisc);
763 spin_unlock_bh(qdisc_lock(qdisc));
764 }
765 }
766}
767
768/* Are all TX queues of the device empty? */
769static inline bool qdisc_all_tx_empty(const struct net_device *dev)
770{
771 unsigned int i;
772
773 rcu_read_lock();
774 for (i = 0; i < dev->num_tx_queues; i++) {
775 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
776 const struct Qdisc *q = rcu_dereference(txq->qdisc);
777
778 if (!qdisc_is_empty(q)) {
779 rcu_read_unlock();
780 return false;
781 }
782 }
783 rcu_read_unlock();
784 return true;
785}
786
787/* Are any of the TX qdiscs changing? */
788static inline bool qdisc_tx_changing(const struct net_device *dev)
789{
790 unsigned int i;
791
792 for (i = 0; i < dev->num_tx_queues; i++) {
793 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
794 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
795 return true;
796 }
797 return false;
798}
799
800/* Is the device using the noop qdisc on all queues? */
801static inline bool qdisc_tx_is_noop(const struct net_device *dev)
802{
803 unsigned int i;
804
805 for (i = 0; i < dev->num_tx_queues; i++) {
806 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
807 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
808 return false;
809 }
810 return true;
811}
812
813static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
814{
815 return qdisc_skb_cb(skb)->pkt_len;
816}
817
818/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
819enum net_xmit_qdisc_t {
820 __NET_XMIT_STOLEN = 0x00010000,
821 __NET_XMIT_BYPASS = 0x00020000,
822};
823
824#ifdef CONFIG_NET_CLS_ACT
825#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
826#else
827#define net_xmit_drop_count(e) (1)
828#endif
829
830static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
831 const struct Qdisc *sch)
832{
833#ifdef CONFIG_NET_SCHED
834 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
835
836 if (stab)
837 __qdisc_calculate_pkt_len(skb, stab);
838#endif
839}
840
841static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
842 struct sk_buff **to_free)
843{
844 qdisc_calculate_pkt_len(skb, sch);
845 return sch->enqueue(skb, sch, to_free);
846}
847
848static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
849 __u64 bytes, __u32 packets)
850{
851 bstats->bytes += bytes;
852 bstats->packets += packets;
853}
854
855static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
856 const struct sk_buff *skb)
857{
858 _bstats_update(bstats,
859 qdisc_pkt_len(skb),
860 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
861}
862
863static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
864 __u64 bytes, __u32 packets)
865{
866 u64_stats_update_begin(&bstats->syncp);
867 _bstats_update(&bstats->bstats, bytes, packets);
868 u64_stats_update_end(&bstats->syncp);
869}
870
871static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
872 const struct sk_buff *skb)
873{
874 u64_stats_update_begin(&bstats->syncp);
875 bstats_update(&bstats->bstats, skb);
876 u64_stats_update_end(&bstats->syncp);
877}
878
879static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
880 const struct sk_buff *skb)
881{
882 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
883}
884
885static inline void qdisc_bstats_update(struct Qdisc *sch,
886 const struct sk_buff *skb)
887{
888 bstats_update(&sch->bstats, skb);
889}
890
891static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
892 const struct sk_buff *skb)
893{
894 sch->qstats.backlog -= qdisc_pkt_len(skb);
895}
896
897static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
898 const struct sk_buff *skb)
899{
900 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
901}
902
903static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
904 const struct sk_buff *skb)
905{
906 sch->qstats.backlog += qdisc_pkt_len(skb);
907}
908
909static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
910 const struct sk_buff *skb)
911{
912 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
913}
914
915static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
916{
917 this_cpu_inc(sch->cpu_qstats->qlen);
918}
919
920static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
921{
922 this_cpu_dec(sch->cpu_qstats->qlen);
923}
924
925static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
926{
927 this_cpu_inc(sch->cpu_qstats->requeues);
928}
929
930static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
931{
932 sch->qstats.drops += count;
933}
934
935static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
936{
937 qstats->drops++;
938}
939
940static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
941{
942 qstats->overlimits++;
943}
944
945static inline void qdisc_qstats_drop(struct Qdisc *sch)
946{
947 qstats_drop_inc(&sch->qstats);
948}
949
950static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
951{
952 this_cpu_inc(sch->cpu_qstats->drops);
953}
954
955static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
956{
957 sch->qstats.overlimits++;
958}
959
960static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
961{
962 __u32 qlen = qdisc_qlen_sum(sch);
963
964 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
965}
966
967static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen,
968 __u32 *backlog)
969{
970 struct gnet_stats_queue qstats = { 0 };
971 __u32 len = qdisc_qlen_sum(sch);
972
973 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len);
974 *qlen = qstats.qlen;
975 *backlog = qstats.backlog;
976}
977
978static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
979{
980 __u32 qlen, backlog;
981
982 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
983 qdisc_tree_reduce_backlog(sch, qlen, backlog);
984}
985
986static inline void qdisc_purge_queue(struct Qdisc *sch)
987{
988 __u32 qlen, backlog;
989
990 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
991 qdisc_reset(sch);
992 qdisc_tree_reduce_backlog(sch, qlen, backlog);
993}
994
995static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
996{
997 qh->head = NULL;
998 qh->tail = NULL;
999 qh->qlen = 0;
1000}
1001
1002static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
1003 struct qdisc_skb_head *qh)
1004{
1005 struct sk_buff *last = qh->tail;
1006
1007 if (last) {
1008 skb->next = NULL;
1009 last->next = skb;
1010 qh->tail = skb;
1011 } else {
1012 qh->tail = skb;
1013 qh->head = skb;
1014 }
1015 qh->qlen++;
1016}
1017
1018static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1019{
1020 __qdisc_enqueue_tail(skb, &sch->q);
1021 qdisc_qstats_backlog_inc(sch, skb);
1022 return NET_XMIT_SUCCESS;
1023}
1024
1025static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1026 struct qdisc_skb_head *qh)
1027{
1028 skb->next = qh->head;
1029
1030 if (!qh->head)
1031 qh->tail = skb;
1032 qh->head = skb;
1033 qh->qlen++;
1034}
1035
1036static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1037{
1038 struct sk_buff *skb = qh->head;
1039
1040 if (likely(skb != NULL)) {
1041 qh->head = skb->next;
1042 qh->qlen--;
1043 if (qh->head == NULL)
1044 qh->tail = NULL;
1045 skb->next = NULL;
1046 }
1047
1048 return skb;
1049}
1050
1051static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1052{
1053 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1054
1055 if (likely(skb != NULL)) {
1056 qdisc_qstats_backlog_dec(sch, skb);
1057 qdisc_bstats_update(sch, skb);
1058 }
1059
1060 return skb;
1061}
1062
1063/* Instead of calling kfree_skb() while root qdisc lock is held,
1064 * queue the skb for future freeing at end of __dev_xmit_skb()
1065 */
1066static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1067{
1068 skb->next = *to_free;
1069 *to_free = skb;
1070}
1071
1072static inline void __qdisc_drop_all(struct sk_buff *skb,
1073 struct sk_buff **to_free)
1074{
1075 if (skb->prev)
1076 skb->prev->next = *to_free;
1077 else
1078 skb->next = *to_free;
1079 *to_free = skb;
1080}
1081
1082static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1083 struct qdisc_skb_head *qh,
1084 struct sk_buff **to_free)
1085{
1086 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1087
1088 if (likely(skb != NULL)) {
1089 unsigned int len = qdisc_pkt_len(skb);
1090
1091 qdisc_qstats_backlog_dec(sch, skb);
1092 __qdisc_drop(skb, to_free);
1093 return len;
1094 }
1095
1096 return 0;
1097}
1098
1099static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1100{
1101 const struct qdisc_skb_head *qh = &sch->q;
1102
1103 return qh->head;
1104}
1105
1106/* generic pseudo peek method for non-work-conserving qdisc */
1107static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1108{
1109 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1110
1111 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1112 if (!skb) {
1113 skb = sch->dequeue(sch);
1114
1115 if (skb) {
1116 __skb_queue_head(&sch->gso_skb, skb);
1117 /* it's still part of the queue */
1118 qdisc_qstats_backlog_inc(sch, skb);
1119 sch->q.qlen++;
1120 }
1121 }
1122
1123 return skb;
1124}
1125
1126static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1127 struct sk_buff *skb)
1128{
1129 if (qdisc_is_percpu_stats(sch)) {
1130 qdisc_qstats_cpu_backlog_dec(sch, skb);
1131 qdisc_bstats_cpu_update(sch, skb);
1132 qdisc_qstats_cpu_qlen_dec(sch);
1133 } else {
1134 qdisc_qstats_backlog_dec(sch, skb);
1135 qdisc_bstats_update(sch, skb);
1136 sch->q.qlen--;
1137 }
1138}
1139
1140static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1141 unsigned int pkt_len)
1142{
1143 if (qdisc_is_percpu_stats(sch)) {
1144 qdisc_qstats_cpu_qlen_inc(sch);
1145 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1146 } else {
1147 sch->qstats.backlog += pkt_len;
1148 sch->q.qlen++;
1149 }
1150}
1151
1152/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1153static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1154{
1155 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1156
1157 if (skb) {
1158 skb = __skb_dequeue(&sch->gso_skb);
1159 if (qdisc_is_percpu_stats(sch)) {
1160 qdisc_qstats_cpu_backlog_dec(sch, skb);
1161 qdisc_qstats_cpu_qlen_dec(sch);
1162 } else {
1163 qdisc_qstats_backlog_dec(sch, skb);
1164 sch->q.qlen--;
1165 }
1166 } else {
1167 skb = sch->dequeue(sch);
1168 }
1169
1170 return skb;
1171}
1172
1173static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1174{
1175 /*
1176 * We do not know the backlog in bytes of this list, it
1177 * is up to the caller to correct it
1178 */
1179 ASSERT_RTNL();
1180 if (qh->qlen) {
1181 rtnl_kfree_skbs(qh->head, qh->tail);
1182
1183 qh->head = NULL;
1184 qh->tail = NULL;
1185 qh->qlen = 0;
1186 }
1187}
1188
1189static inline void qdisc_reset_queue(struct Qdisc *sch)
1190{
1191 __qdisc_reset_queue(&sch->q);
1192 sch->qstats.backlog = 0;
1193}
1194
1195static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1196 struct Qdisc **pold)
1197{
1198 struct Qdisc *old;
1199
1200 sch_tree_lock(sch);
1201 old = *pold;
1202 *pold = new;
1203 if (old != NULL)
1204 qdisc_purge_queue(old);
1205 sch_tree_unlock(sch);
1206
1207 return old;
1208}
1209
1210static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1211{
1212 rtnl_kfree_skbs(skb, skb);
1213 qdisc_qstats_drop(sch);
1214}
1215
1216static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1217 struct sk_buff **to_free)
1218{
1219 __qdisc_drop(skb, to_free);
1220 qdisc_qstats_cpu_drop(sch);
1221
1222 return NET_XMIT_DROP;
1223}
1224
1225static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1226 struct sk_buff **to_free)
1227{
1228 __qdisc_drop(skb, to_free);
1229 qdisc_qstats_drop(sch);
1230
1231 return NET_XMIT_DROP;
1232}
1233
1234static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1235 struct sk_buff **to_free)
1236{
1237 __qdisc_drop_all(skb, to_free);
1238 qdisc_qstats_drop(sch);
1239
1240 return NET_XMIT_DROP;
1241}
1242
1243/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1244 long it will take to send a packet given its size.
1245 */
1246static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1247{
1248 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1249 if (slot < 0)
1250 slot = 0;
1251 slot >>= rtab->rate.cell_log;
1252 if (slot > 255)
1253 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1254 return rtab->data[slot];
1255}
1256
1257struct psched_ratecfg {
1258 u64 rate_bytes_ps; /* bytes per second */
1259 u32 mult;
1260 u16 overhead;
1261 u8 linklayer;
1262 u8 shift;
1263};
1264
1265static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1266 unsigned int len)
1267{
1268 len += r->overhead;
1269
1270 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1271 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1272
1273 return ((u64)len * r->mult) >> r->shift;
1274}
1275
1276void psched_ratecfg_precompute(struct psched_ratecfg *r,
1277 const struct tc_ratespec *conf,
1278 u64 rate64);
1279
1280static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1281 const struct psched_ratecfg *r)
1282{
1283 memset(res, 0, sizeof(*res));
1284
1285 /* legacy struct tc_ratespec has a 32bit @rate field
1286 * Qdisc using 64bit rate should add new attributes
1287 * in order to maintain compatibility.
1288 */
1289 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1290
1291 res->overhead = r->overhead;
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_cpu __percpu *cpu_bstats;
1316 struct gnet_stats_queue __percpu *cpu_qstats;
1317 struct rcu_head rcu;
1318};
1319
1320static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1321 const struct sk_buff *skb)
1322{
1323 bstats_cpu_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
1344int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1345
1346#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