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 #include <net/dropreason-qdisc.h>
24
25 struct Qdisc_ops;
26 struct qdisc_walker;
27 struct tcf_walker;
28 struct module;
29 struct bpf_flow_keys;
30 struct Qdisc;
31 struct netdev_queue;
32
33 struct qdisc_rate_table {
34 struct tc_ratespec rate;
35 u32 data[256];
36 struct qdisc_rate_table *next;
37 int refcnt;
38 };
39
40 enum qdisc_state_t {
41 __QDISC_STATE_SCHED,
42 __QDISC_STATE_DEACTIVATED,
43 __QDISC_STATE_MISSED,
44 __QDISC_STATE_DRAINING,
45 };
46
47 #define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED)
48 #define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING)
49
50 #define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \
51 QDISC_STATE_DRAINING)
52
53 struct qdisc_size_table {
54 struct rcu_head rcu;
55 struct list_head list;
56 struct tc_sizespec szopts;
57 int refcnt;
58 u16 data[];
59 };
60
61 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
62 struct qdisc_skb_head {
63 struct sk_buff *head;
64 struct sk_buff *tail;
65 __u32 qlen;
66 spinlock_t lock;
67 };
68
69 struct Qdisc {
70 int (*enqueue)(struct sk_buff *skb,
71 struct Qdisc *sch,
72 struct sk_buff **to_free);
73 struct sk_buff * (*dequeue)(struct Qdisc *sch);
74 unsigned int flags;
75 #define TCQ_F_BUILTIN 1
76 #define TCQ_F_INGRESS 2
77 #define TCQ_F_CAN_BYPASS 4
78 #define TCQ_F_MQROOT 8
79 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
80 * q->dev_queue : It can test
81 * netif_xmit_frozen_or_stopped() before
82 * dequeueing next packet.
83 * Its true for MQ/MQPRIO slaves, or non
84 * multiqueue device.
85 */
86 #define TCQ_F_WARN_NONWC (1 << 16)
87 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
88 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
89 * qdisc_tree_reduce_backlog() should stop.
90 */
91 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
92 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
93 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
94 #define TCQ_F_DEQUEUE_DROPS 0x400 /* ->dequeue() can drop packets in q->to_free */
95
96 u32 limit;
97 const struct Qdisc_ops *ops;
98 struct qdisc_size_table __rcu *stab;
99 struct hlist_node hash;
100 u32 handle;
101 u32 parent;
102 int depth;
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 /* Cache line potentially dirtied in dequeue() or __netif_reschedule(). */
113 __cacheline_group_begin(Qdisc_read_mostly) ____cacheline_aligned;
114 struct sk_buff_head gso_skb;
115 struct Qdisc *next_sched;
116 struct sk_buff_head skb_bad_txq;
117 __cacheline_group_end(Qdisc_read_mostly);
118
119 /* Fields dirtied in dequeue() fast path. */
120 __cacheline_group_begin(Qdisc_write) ____cacheline_aligned;
121 struct qdisc_skb_head q;
122 unsigned long state;
123 struct gnet_stats_basic_sync bstats;
124 bool running; /* must be written under qdisc spinlock */
125
126 /* Note : we only change qstats.backlog in fast path. */
127 struct gnet_stats_queue qstats;
128
129 struct sk_buff *to_free;
130 __cacheline_group_end(Qdisc_write);
131
132
133 atomic_long_t defer_count ____cacheline_aligned_in_smp;
134 struct llist_head defer_list;
135
136 spinlock_t seqlock;
137
138 struct rcu_head rcu;
139 netdevice_tracker dev_tracker;
140 struct lock_class_key root_lock_key;
141 /* private data */
142 long privdata[] ____cacheline_aligned;
143 };
144
qdisc_refcount_inc(struct Qdisc * qdisc)145 static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
146 {
147 if (qdisc->flags & TCQ_F_BUILTIN)
148 return;
149 refcount_inc(&qdisc->refcnt);
150 }
151
qdisc_refcount_dec_if_one(struct Qdisc * qdisc)152 static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc)
153 {
154 if (qdisc->flags & TCQ_F_BUILTIN)
155 return true;
156 return refcount_dec_if_one(&qdisc->refcnt);
157 }
158
159 /* Intended to be used by unlocked users, when concurrent qdisc release is
160 * possible.
161 */
162
qdisc_refcount_inc_nz(struct Qdisc * qdisc)163 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
164 {
165 if (qdisc->flags & TCQ_F_BUILTIN)
166 return qdisc;
167 if (refcount_inc_not_zero(&qdisc->refcnt))
168 return qdisc;
169 return NULL;
170 }
171
172 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
173 * root_lock section, or provide their own memory barriers -- ordering
174 * against qdisc_run_begin/end() atomic bit operations.
175 */
qdisc_is_running(struct Qdisc * qdisc)176 static inline bool qdisc_is_running(struct Qdisc *qdisc)
177 {
178 if (qdisc->flags & TCQ_F_NOLOCK)
179 return spin_is_locked(&qdisc->seqlock);
180 return READ_ONCE(qdisc->running);
181 }
182
nolock_qdisc_is_empty(const struct Qdisc * qdisc)183 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
184 {
185 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
186 }
187
qdisc_is_percpu_stats(const struct Qdisc * q)188 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
189 {
190 return q->flags & TCQ_F_CPUSTATS;
191 }
192
qdisc_is_empty(const struct Qdisc * qdisc)193 static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
194 {
195 if (qdisc_is_percpu_stats(qdisc))
196 return nolock_qdisc_is_empty(qdisc);
197 return !READ_ONCE(qdisc->q.qlen);
198 }
199
200 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
201 * the qdisc root lock acquired.
202 */
qdisc_run_begin(struct Qdisc * qdisc)203 static inline bool qdisc_run_begin(struct Qdisc *qdisc)
204 {
205 if (qdisc->flags & TCQ_F_NOLOCK) {
206 if (spin_trylock(&qdisc->seqlock))
207 return true;
208
209 /* No need to insist if the MISSED flag was already set.
210 * Note that test_and_set_bit() also gives us memory ordering
211 * guarantees wrt potential earlier enqueue() and below
212 * spin_trylock(), both of which are necessary to prevent races
213 */
214 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
215 return false;
216
217 /* Try to take the lock again to make sure that we will either
218 * grab it or the CPU that still has it will see MISSED set
219 * when testing it in qdisc_run_end()
220 */
221 return spin_trylock(&qdisc->seqlock);
222 }
223 if (READ_ONCE(qdisc->running))
224 return false;
225 WRITE_ONCE(qdisc->running, true);
226 return true;
227 }
228
qdisc_run_end(struct Qdisc * qdisc)229 static inline struct sk_buff *qdisc_run_end(struct Qdisc *qdisc)
230 {
231 struct sk_buff *to_free = NULL;
232
233 if (qdisc->flags & TCQ_F_NOLOCK) {
234 spin_unlock(&qdisc->seqlock);
235
236 /* spin_unlock() only has store-release semantic. The unlock
237 * and test_bit() ordering is a store-load ordering, so a full
238 * memory barrier is needed here.
239 */
240 smp_mb();
241
242 if (unlikely(test_bit(__QDISC_STATE_MISSED,
243 &qdisc->state)))
244 __netif_schedule(qdisc);
245 return NULL;
246 }
247
248 if (qdisc->flags & TCQ_F_DEQUEUE_DROPS) {
249 to_free = qdisc->to_free;
250 if (to_free)
251 qdisc->to_free = NULL;
252 }
253 WRITE_ONCE(qdisc->running, false);
254 return to_free;
255 }
256
qdisc_may_bulk(const struct Qdisc * qdisc)257 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
258 {
259 return qdisc->flags & TCQ_F_ONETXQUEUE;
260 }
261
qdisc_avail_bulklimit(const struct netdev_queue * txq)262 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
263 {
264 return netdev_queue_dql_avail(txq);
265 }
266
267 struct Qdisc_class_ops {
268 unsigned int flags;
269 /* Child qdisc manipulation */
270 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *);
271 int (*graft)(struct Qdisc *, unsigned long cl,
272 struct Qdisc *, struct Qdisc **,
273 struct netlink_ext_ack *extack);
274 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
275 void (*qlen_notify)(struct Qdisc *, unsigned long);
276
277 /* Class manipulation routines */
278 unsigned long (*find)(struct Qdisc *, u32 classid);
279 int (*change)(struct Qdisc *, u32, u32,
280 struct nlattr **, unsigned long *,
281 struct netlink_ext_ack *);
282 int (*delete)(struct Qdisc *, unsigned long,
283 struct netlink_ext_ack *);
284 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
285
286 /* Filter manipulation */
287 struct tcf_block * (*tcf_block)(struct Qdisc *sch,
288 unsigned long arg,
289 struct netlink_ext_ack *extack);
290 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long,
291 u32 classid);
292 void (*unbind_tcf)(struct Qdisc *, unsigned long);
293
294 /* rtnetlink specific */
295 int (*dump)(struct Qdisc *, unsigned long,
296 struct sk_buff *skb, struct tcmsg*);
297 int (*dump_stats)(struct Qdisc *, unsigned long,
298 struct gnet_dump *);
299 };
300
301 /* Qdisc_class_ops flag values */
302
303 /* Implements API that doesn't require rtnl lock */
304 enum qdisc_class_ops_flags {
305 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
306 };
307
308 struct Qdisc_ops {
309 struct Qdisc_ops *next;
310 const struct Qdisc_class_ops *cl_ops;
311 char id[IFNAMSIZ];
312 int priv_size;
313 unsigned int static_flags;
314
315 int (*enqueue)(struct sk_buff *skb,
316 struct Qdisc *sch,
317 struct sk_buff **to_free);
318 struct sk_buff * (*dequeue)(struct Qdisc *);
319 struct sk_buff * (*peek)(struct Qdisc *);
320
321 int (*init)(struct Qdisc *sch, struct nlattr *arg,
322 struct netlink_ext_ack *extack);
323 void (*reset)(struct Qdisc *);
324 void (*destroy)(struct Qdisc *);
325 int (*change)(struct Qdisc *sch,
326 struct nlattr *arg,
327 struct netlink_ext_ack *extack);
328 void (*attach)(struct Qdisc *sch);
329 int (*change_tx_queue_len)(struct Qdisc *, unsigned int);
330 void (*change_real_num_tx)(struct Qdisc *sch,
331 unsigned int new_real_tx);
332
333 int (*dump)(struct Qdisc *, struct sk_buff *);
334 int (*dump_stats)(struct Qdisc *, struct gnet_dump *);
335
336 void (*ingress_block_set)(struct Qdisc *sch,
337 u32 block_index);
338 void (*egress_block_set)(struct Qdisc *sch,
339 u32 block_index);
340 u32 (*ingress_block_get)(struct Qdisc *sch);
341 u32 (*egress_block_get)(struct Qdisc *sch);
342
343 struct module *owner;
344 };
345
346 struct tcf_result {
347 union {
348 struct {
349 unsigned long class;
350 u32 classid;
351 };
352 const struct tcf_proto *goto_tp;
353 };
354 };
355
356 struct tcf_chain;
357
358 struct tcf_proto_ops {
359 struct list_head head;
360 char kind[IFNAMSIZ];
361
362 int (*classify)(struct sk_buff *,
363 const struct tcf_proto *,
364 struct tcf_result *);
365 int (*init)(struct tcf_proto*);
366 void (*destroy)(struct tcf_proto *tp, bool rtnl_held,
367 struct netlink_ext_ack *extack);
368
369 void* (*get)(struct tcf_proto*, u32 handle);
370 void (*put)(struct tcf_proto *tp, void *f);
371 int (*change)(struct net *net, struct sk_buff *,
372 struct tcf_proto*, unsigned long,
373 u32 handle, struct nlattr **,
374 void **, u32,
375 struct netlink_ext_ack *);
376 int (*delete)(struct tcf_proto *tp, void *arg,
377 bool *last, bool rtnl_held,
378 struct netlink_ext_ack *);
379 bool (*delete_empty)(struct tcf_proto *tp);
380 void (*walk)(struct tcf_proto *tp,
381 struct tcf_walker *arg, bool rtnl_held);
382 int (*reoffload)(struct tcf_proto *tp, bool add,
383 flow_setup_cb_t *cb, void *cb_priv,
384 struct netlink_ext_ack *extack);
385 void (*hw_add)(struct tcf_proto *tp,
386 void *type_data);
387 void (*hw_del)(struct tcf_proto *tp,
388 void *type_data);
389 void (*bind_class)(void *, u32, unsigned long,
390 void *, unsigned long);
391 void * (*tmplt_create)(struct net *net,
392 struct tcf_chain *chain,
393 struct nlattr **tca,
394 struct netlink_ext_ack *extack);
395 void (*tmplt_destroy)(void *tmplt_priv);
396 void (*tmplt_reoffload)(struct tcf_chain *chain,
397 bool add,
398 flow_setup_cb_t *cb,
399 void *cb_priv);
400 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp,
401 u32 handle);
402
403 /* rtnetlink specific */
404 int (*dump)(struct net*, struct tcf_proto*, void *,
405 struct sk_buff *skb, struct tcmsg*,
406 bool);
407 int (*terse_dump)(struct net *net,
408 struct tcf_proto *tp, void *fh,
409 struct sk_buff *skb,
410 struct tcmsg *t, bool rtnl_held);
411 int (*tmplt_dump)(struct sk_buff *skb,
412 struct net *net,
413 void *tmplt_priv);
414
415 struct module *owner;
416 int flags;
417 };
418
419 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
420 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
421 * conditions can occur when filters are inserted/deleted simultaneously.
422 */
423 enum tcf_proto_ops_flags {
424 TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
425 };
426
427 struct tcf_proto {
428 /* Fast access part */
429 struct tcf_proto __rcu *next;
430 void __rcu *root;
431
432 /* called under RCU BH lock*/
433 int (*classify)(struct sk_buff *,
434 const struct tcf_proto *,
435 struct tcf_result *);
436 __be16 protocol;
437
438 /* All the rest */
439 u32 prio;
440 void *data;
441 const struct tcf_proto_ops *ops;
442 struct tcf_chain *chain;
443 /* Lock protects tcf_proto shared state and can be used by unlocked
444 * classifiers to protect their private data.
445 */
446 spinlock_t lock;
447 bool deleting;
448 bool counted;
449 bool usesw;
450 refcount_t refcnt;
451 struct rcu_head rcu;
452 struct hlist_node destroy_ht_node;
453 };
454
455 struct qdisc_skb_cb {
456 unsigned int pkt_len;
457 u16 pkt_segs;
458 u16 tc_classid;
459 #define QDISC_CB_PRIV_LEN 20
460 unsigned char data[QDISC_CB_PRIV_LEN];
461
462 u16 slave_dev_queue_mapping;
463 u8 post_ct:1;
464 u8 post_ct_snat:1;
465 u8 post_ct_dnat:1;
466 };
467
468 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
469
470 struct tcf_chain {
471 /* Protects filter_chain. */
472 struct mutex filter_chain_lock;
473 struct tcf_proto __rcu *filter_chain;
474 struct list_head list;
475 struct tcf_block *block;
476 u32 index; /* chain index */
477 unsigned int refcnt;
478 unsigned int action_refcnt;
479 bool explicitly_created;
480 bool flushing;
481 const struct tcf_proto_ops *tmplt_ops;
482 void *tmplt_priv;
483 struct rcu_head rcu;
484 };
485
486 struct tcf_block {
487 struct xarray ports; /* datapath accessible */
488 /* Lock protects tcf_block and lifetime-management data of chains
489 * attached to the block (refcnt, action_refcnt, explicitly_created).
490 */
491 struct mutex lock;
492 struct list_head chain_list;
493 u32 index; /* block index for shared blocks */
494 u32 classid; /* which class this block belongs to */
495 refcount_t refcnt;
496 struct net *net;
497 struct Qdisc *q;
498 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
499 struct flow_block flow_block;
500 struct list_head owner_list;
501 bool keep_dst;
502 atomic_t useswcnt;
503 atomic_t offloadcnt; /* Number of oddloaded filters */
504 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
505 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
506 struct {
507 struct tcf_chain *chain;
508 struct list_head filter_chain_list;
509 } chain0;
510 struct rcu_head rcu;
511 DECLARE_HASHTABLE(proto_destroy_ht, 7);
512 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
513 };
514
515 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index);
516
lockdep_tcf_chain_is_locked(struct tcf_chain * chain)517 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
518 {
519 return lockdep_is_held(&chain->filter_chain_lock);
520 }
521
lockdep_tcf_proto_is_locked(struct tcf_proto * tp)522 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
523 {
524 return lockdep_is_held(&tp->lock);
525 }
526
527 #define tcf_chain_dereference(p, chain) \
528 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
529
530 #define tcf_proto_dereference(p, tp) \
531 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
532
qdisc_cb_private_validate(const struct sk_buff * skb,int sz)533 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
534 {
535 struct qdisc_skb_cb *qcb;
536
537 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
538 BUILD_BUG_ON(sizeof(qcb->data) < sz);
539 }
540
qdisc_qlen(const struct Qdisc * q)541 static inline int qdisc_qlen(const struct Qdisc *q)
542 {
543 return q->q.qlen;
544 }
545
qdisc_qlen_lockless(const struct Qdisc * q)546 static inline int qdisc_qlen_lockless(const struct Qdisc *q)
547 {
548 return READ_ONCE(q->q.qlen);
549 }
550
qdisc_qlen_inc(struct Qdisc * q)551 static inline void qdisc_qlen_inc(struct Qdisc *q)
552 {
553 WRITE_ONCE(q->q.qlen, q->q.qlen + 1);
554 }
555
qdisc_qlen_dec(struct Qdisc * q)556 static inline void qdisc_qlen_dec(struct Qdisc *q)
557 {
558 WRITE_ONCE(q->q.qlen, q->q.qlen - 1);
559 }
560
qdisc_qlen_sum(const struct Qdisc * q)561 static inline int qdisc_qlen_sum(const struct Qdisc *q)
562 {
563 __u32 qlen = q->qstats.qlen;
564 int i;
565
566 if (qdisc_is_percpu_stats(q)) {
567 for_each_possible_cpu(i)
568 qlen += READ_ONCE(per_cpu_ptr(q->cpu_qstats, i)->qlen);
569 } else {
570 qlen += qdisc_qlen_lockless(q);
571 }
572
573 return qlen;
574 }
575
qdisc_skb_cb(const struct sk_buff * skb)576 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
577 {
578 return (struct qdisc_skb_cb *)skb->cb;
579 }
580
qdisc_lock(struct Qdisc * qdisc)581 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
582 {
583 return &qdisc->q.lock;
584 }
585
qdisc_root(const struct Qdisc * qdisc)586 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
587 {
588 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
589
590 return q;
591 }
592
qdisc_root_bh(const struct Qdisc * qdisc)593 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
594 {
595 return rcu_dereference_bh(qdisc->dev_queue->qdisc);
596 }
597
qdisc_root_sleeping(const struct Qdisc * qdisc)598 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
599 {
600 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping);
601 }
602
qdisc_root_sleeping_lock(const struct Qdisc * qdisc)603 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
604 {
605 struct Qdisc *root = qdisc_root_sleeping(qdisc);
606
607 ASSERT_RTNL();
608 return qdisc_lock(root);
609 }
610
qdisc_dev(const struct Qdisc * qdisc)611 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
612 {
613 return qdisc->dev_queue->dev;
614 }
615
sch_tree_lock(struct Qdisc * q)616 static inline void sch_tree_lock(struct Qdisc *q)
617 {
618 if (q->flags & TCQ_F_MQROOT)
619 spin_lock_bh(qdisc_lock(q));
620 else
621 spin_lock_bh(qdisc_root_sleeping_lock(q));
622 }
623
sch_tree_unlock(struct Qdisc * q)624 static inline void sch_tree_unlock(struct Qdisc *q)
625 {
626 if (q->flags & TCQ_F_MQROOT)
627 spin_unlock_bh(qdisc_lock(q));
628 else
629 spin_unlock_bh(qdisc_root_sleeping_lock(q));
630 }
631
632 extern struct Qdisc noop_qdisc;
633 extern struct Qdisc_ops noop_qdisc_ops;
634 extern struct Qdisc_ops pfifo_fast_ops;
635 extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1];
636 extern struct Qdisc_ops mq_qdisc_ops;
637 extern struct Qdisc_ops noqueue_qdisc_ops;
638 extern const struct Qdisc_ops *default_qdisc_ops;
639 static inline const struct Qdisc_ops *
get_default_qdisc_ops(const struct net_device * dev,int ntx)640 get_default_qdisc_ops(const struct net_device *dev, int ntx)
641 {
642 return ntx < dev->real_num_tx_queues ?
643 default_qdisc_ops : &pfifo_fast_ops;
644 }
645
646 struct Qdisc_class_common {
647 u32 classid;
648 unsigned int filter_cnt;
649 struct hlist_node hnode;
650 };
651
652 struct Qdisc_class_hash {
653 struct hlist_head *hash;
654 unsigned int hashsize;
655 unsigned int hashmask;
656 unsigned int hashelems;
657 };
658
qdisc_class_hash(u32 id,u32 mask)659 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
660 {
661 id ^= id >> 8;
662 id ^= id >> 4;
663 return id & mask;
664 }
665
666 static inline struct Qdisc_class_common *
qdisc_class_find(const struct Qdisc_class_hash * hash,u32 id)667 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
668 {
669 struct Qdisc_class_common *cl;
670 unsigned int h;
671
672 if (!id)
673 return NULL;
674
675 h = qdisc_class_hash(id, hash->hashmask);
676 hlist_for_each_entry(cl, &hash->hash[h], hnode) {
677 if (cl->classid == id)
678 return cl;
679 }
680 return NULL;
681 }
682
qdisc_class_in_use(const struct Qdisc_class_common * cl)683 static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl)
684 {
685 return cl->filter_cnt > 0;
686 }
687
qdisc_class_get(struct Qdisc_class_common * cl)688 static inline void qdisc_class_get(struct Qdisc_class_common *cl)
689 {
690 unsigned int res;
691
692 if (check_add_overflow(cl->filter_cnt, 1, &res))
693 WARN(1, "Qdisc class overflow");
694
695 cl->filter_cnt = res;
696 }
697
qdisc_class_put(struct Qdisc_class_common * cl)698 static inline void qdisc_class_put(struct Qdisc_class_common *cl)
699 {
700 unsigned int res;
701
702 if (check_sub_overflow(cl->filter_cnt, 1, &res))
703 WARN(1, "Qdisc class underflow");
704
705 cl->filter_cnt = res;
706 }
707
tc_classid_to_hwtc(struct net_device * dev,u32 classid)708 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
709 {
710 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
711
712 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
713 }
714
715 int qdisc_class_hash_init(struct Qdisc_class_hash *);
716 void qdisc_class_hash_insert(struct Qdisc_class_hash *,
717 struct Qdisc_class_common *);
718 void qdisc_class_hash_remove(struct Qdisc_class_hash *,
719 struct Qdisc_class_common *);
720 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
721 void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
722
723 int dev_qdisc_change_tx_queue_len(struct net_device *dev);
724 void dev_qdisc_change_real_num_tx(struct net_device *dev,
725 unsigned int new_real_tx);
726 void dev_init_scheduler(struct net_device *dev);
727 void dev_shutdown(struct net_device *dev);
728 void dev_activate(struct net_device *dev);
729 void dev_deactivate(struct net_device *dev, bool reset_needed);
730 void dev_deactivate_many(struct list_head *head, bool reset_needed);
731 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
732 struct Qdisc *qdisc);
733 void qdisc_reset(struct Qdisc *qdisc);
734 void qdisc_destroy(struct Qdisc *qdisc);
735 void qdisc_put(struct Qdisc *qdisc);
736 void qdisc_put_unlocked(struct Qdisc *qdisc);
737 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
738
dev_reset_queue(struct net_device * dev,struct netdev_queue * dev_queue,void * _unused)739 static inline void dev_reset_queue(struct net_device *dev,
740 struct netdev_queue *dev_queue,
741 void *_unused)
742 {
743 struct Qdisc *qdisc;
744 bool nolock;
745
746 qdisc = rtnl_dereference(dev_queue->qdisc_sleeping);
747 if (!qdisc)
748 return;
749
750 nolock = qdisc->flags & TCQ_F_NOLOCK;
751
752 if (nolock)
753 spin_lock_bh(&qdisc->seqlock);
754 spin_lock_bh(qdisc_lock(qdisc));
755
756 qdisc_reset(qdisc);
757
758 spin_unlock_bh(qdisc_lock(qdisc));
759 if (nolock) {
760 clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
761 clear_bit(__QDISC_STATE_DRAINING, &qdisc->state);
762 spin_unlock_bh(&qdisc->seqlock);
763 }
764 }
765
766 #ifdef CONFIG_NET_SCHED
767 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
768 void *type_data);
769 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
770 struct Qdisc *new, struct Qdisc *old,
771 enum tc_setup_type type, void *type_data,
772 struct netlink_ext_ack *extack);
773 #else
774 static inline int
qdisc_offload_dump_helper(struct Qdisc * q,enum tc_setup_type type,void * type_data)775 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
776 void *type_data)
777 {
778 q->flags &= ~TCQ_F_OFFLOADED;
779 return 0;
780 }
781
782 static inline void
qdisc_offload_graft_helper(struct net_device * dev,struct Qdisc * sch,struct Qdisc * new,struct Qdisc * old,enum tc_setup_type type,void * type_data,struct netlink_ext_ack * extack)783 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
784 struct Qdisc *new, struct Qdisc *old,
785 enum tc_setup_type type, void *type_data,
786 struct netlink_ext_ack *extack)
787 {
788 }
789 #endif
790 void qdisc_offload_query_caps(struct net_device *dev,
791 enum tc_setup_type type,
792 void *caps, size_t caps_len);
793 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
794 const struct Qdisc_ops *ops,
795 struct netlink_ext_ack *extack);
796 void qdisc_free(struct Qdisc *qdisc);
797 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
798 const struct Qdisc_ops *ops, u32 parentid,
799 struct netlink_ext_ack *extack);
800 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
801 const struct qdisc_size_table *stab);
802 int skb_do_redirect(struct sk_buff *);
803
skb_at_tc_ingress(const struct sk_buff * skb)804 static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
805 {
806 #ifdef CONFIG_NET_XGRESS
807 return skb->tc_at_ingress;
808 #else
809 return false;
810 #endif
811 }
812
skb_skip_tc_classify(struct sk_buff * skb)813 static inline bool skb_skip_tc_classify(struct sk_buff *skb)
814 {
815 #ifdef CONFIG_NET_CLS_ACT
816 if (skb->tc_skip_classify) {
817 skb->tc_skip_classify = 0;
818 return true;
819 }
820 #endif
821 return false;
822 }
823
824 /* Reset all TX qdiscs greater than index of a device. */
qdisc_reset_all_tx_gt(struct net_device * dev,unsigned int i)825 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
826 {
827 struct Qdisc *qdisc;
828 bool nolock;
829
830 for (; i < dev->num_tx_queues; i++) {
831 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
832 if (qdisc) {
833 nolock = qdisc->flags & TCQ_F_NOLOCK;
834
835 if (nolock)
836 spin_lock_bh(&qdisc->seqlock);
837 spin_lock_bh(qdisc_lock(qdisc));
838 qdisc_reset(qdisc);
839 spin_unlock_bh(qdisc_lock(qdisc));
840 if (nolock) {
841 clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
842 clear_bit(__QDISC_STATE_DRAINING, &qdisc->state);
843 spin_unlock_bh(&qdisc->seqlock);
844 }
845 }
846 }
847 }
848
849 /* Are all TX queues of the device empty? */
qdisc_all_tx_empty(const struct net_device * dev)850 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
851 {
852 unsigned int i;
853
854 rcu_read_lock();
855 for (i = 0; i < dev->num_tx_queues; i++) {
856 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
857 const struct Qdisc *q = rcu_dereference(txq->qdisc);
858
859 if (!qdisc_is_empty(q)) {
860 rcu_read_unlock();
861 return false;
862 }
863 }
864 rcu_read_unlock();
865 return true;
866 }
867
868 /* Are any of the TX qdiscs changing? */
qdisc_tx_changing(const struct net_device * dev)869 static inline bool qdisc_tx_changing(const struct net_device *dev)
870 {
871 unsigned int i;
872
873 for (i = 0; i < dev->num_tx_queues; i++) {
874 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
875
876 if (rcu_access_pointer(txq->qdisc) !=
877 rcu_access_pointer(txq->qdisc_sleeping))
878 return true;
879 }
880 return false;
881 }
882
883 /* "noqueue" qdisc identified by not having any enqueue, see noqueue_init() */
qdisc_txq_has_no_queue(const struct netdev_queue * txq)884 static inline bool qdisc_txq_has_no_queue(const struct netdev_queue *txq)
885 {
886 struct Qdisc *qdisc = rcu_access_pointer(txq->qdisc);
887
888 return qdisc->enqueue == NULL;
889 }
890
891 /* Is the device using the noop qdisc on all queues? */
qdisc_tx_is_noop(const struct net_device * dev)892 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
893 {
894 unsigned int i;
895
896 for (i = 0; i < dev->num_tx_queues; i++) {
897 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
898 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
899 return false;
900 }
901 return true;
902 }
903
qdisc_pkt_len(const struct sk_buff * skb)904 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
905 {
906 return qdisc_skb_cb(skb)->pkt_len;
907 }
908
qdisc_pkt_segs(const struct sk_buff * skb)909 static inline unsigned int qdisc_pkt_segs(const struct sk_buff *skb)
910 {
911 u32 pkt_segs = qdisc_skb_cb(skb)->pkt_segs;
912
913 DEBUG_NET_WARN_ON_ONCE(pkt_segs !=
914 (skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1));
915 return pkt_segs;
916 }
917
918 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
919 enum net_xmit_qdisc_t {
920 __NET_XMIT_STOLEN = 0x00010000,
921 __NET_XMIT_BYPASS = 0x00020000,
922 };
923
924 #ifdef CONFIG_NET_CLS_ACT
925 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
926 #else
927 #define net_xmit_drop_count(e) (1)
928 #endif
929
qdisc_calculate_pkt_len(struct sk_buff * skb,const struct Qdisc * sch)930 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
931 const struct Qdisc *sch)
932 {
933 #ifdef CONFIG_NET_SCHED
934 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
935
936 if (stab)
937 __qdisc_calculate_pkt_len(skb, stab);
938 #endif
939 }
940
qdisc_enqueue(struct sk_buff * skb,struct Qdisc * sch,struct sk_buff ** to_free)941 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
942 struct sk_buff **to_free)
943 {
944 return sch->enqueue(skb, sch, to_free);
945 }
946
_bstats_update(struct gnet_stats_basic_sync * bstats,__u64 bytes,__u64 packets)947 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
948 __u64 bytes, __u64 packets)
949 {
950 u64_stats_update_begin(&bstats->syncp);
951 u64_stats_add(&bstats->bytes, bytes);
952 u64_stats_add(&bstats->packets, packets);
953 u64_stats_update_end(&bstats->syncp);
954 }
955
_bstats_set(struct gnet_stats_basic_sync * bstats,u64 bytes,u64 packets)956 static inline void _bstats_set(struct gnet_stats_basic_sync *bstats,
957 u64 bytes, u64 packets)
958 {
959 u64_stats_update_begin(&bstats->syncp);
960 u64_stats_set(&bstats->bytes, bytes);
961 u64_stats_set(&bstats->packets, packets);
962 u64_stats_update_end(&bstats->syncp);
963 }
964
bstats_update(struct gnet_stats_basic_sync * bstats,const struct sk_buff * skb)965 static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
966 const struct sk_buff *skb)
967 {
968 _bstats_update(bstats, qdisc_pkt_len(skb), qdisc_pkt_segs(skb));
969 }
970
qdisc_bstats_cpu_update(struct Qdisc * sch,const struct sk_buff * skb)971 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
972 const struct sk_buff *skb)
973 {
974 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
975 }
976
qdisc_bstats_update(struct Qdisc * sch,const struct sk_buff * skb)977 static inline void qdisc_bstats_update(struct Qdisc *sch,
978 const struct sk_buff *skb)
979 {
980 bstats_update(&sch->bstats, skb);
981 }
982
qstats_backlog_sub(struct Qdisc * sch,u32 val)983 static inline void qstats_backlog_sub(struct Qdisc *sch, u32 val)
984 {
985 WRITE_ONCE(sch->qstats.backlog, sch->qstats.backlog - val);
986 }
987
qdisc_qstats_backlog_dec(struct Qdisc * sch,const struct sk_buff * skb)988 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
989 const struct sk_buff *skb)
990 {
991 qstats_backlog_sub(sch, qdisc_pkt_len(skb));
992 }
993
qdisc_qstats_cpu_backlog_dec(struct Qdisc * sch,const struct sk_buff * skb)994 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
995 const struct sk_buff *skb)
996 {
997 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
998 }
999
qstats_backlog_add(struct Qdisc * sch,u32 val)1000 static inline void qstats_backlog_add(struct Qdisc *sch, u32 val)
1001 {
1002 WRITE_ONCE(sch->qstats.backlog, sch->qstats.backlog + val);
1003 }
1004
qdisc_qstats_backlog_inc(struct Qdisc * sch,const struct sk_buff * skb)1005 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
1006 const struct sk_buff *skb)
1007 {
1008 qstats_backlog_add(sch, qdisc_pkt_len(skb));
1009 }
1010
qdisc_qstats_cpu_backlog_inc(struct Qdisc * sch,const struct sk_buff * skb)1011 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
1012 const struct sk_buff *skb)
1013 {
1014 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
1015 }
1016
qdisc_qstats_cpu_qlen_inc(struct Qdisc * sch)1017 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
1018 {
1019 this_cpu_inc(sch->cpu_qstats->qlen);
1020 }
1021
qdisc_qstats_cpu_qlen_dec(struct Qdisc * sch)1022 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
1023 {
1024 this_cpu_dec(sch->cpu_qstats->qlen);
1025 }
1026
qdisc_qstats_cpu_requeues_inc(struct Qdisc * sch)1027 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
1028 {
1029 this_cpu_inc(sch->cpu_qstats->requeues);
1030 }
1031
__qdisc_qstats_drop(struct Qdisc * sch,int count)1032 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
1033 {
1034 WRITE_ONCE(sch->qstats.drops, sch->qstats.drops + count);
1035 }
1036
qstats_drop_inc(struct gnet_stats_queue * qstats)1037 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
1038 {
1039 WRITE_ONCE(qstats->drops, qstats->drops + 1);
1040 }
1041
qstats_cpu_drop_inc(struct gnet_stats_queue __percpu * qstats)1042 static inline void qstats_cpu_drop_inc(struct gnet_stats_queue __percpu *qstats)
1043 {
1044 this_cpu_inc(qstats->drops);
1045 }
1046
qstats_cpu_overlimit_inc(struct gnet_stats_queue __percpu * qstats)1047 static inline void qstats_cpu_overlimit_inc(struct gnet_stats_queue __percpu *qstats)
1048 {
1049 this_cpu_inc(qstats->overlimits);
1050 }
1051
qdisc_qstats_drop(struct Qdisc * sch)1052 static inline void qdisc_qstats_drop(struct Qdisc *sch)
1053 {
1054 qstats_drop_inc(&sch->qstats);
1055 }
1056
qdisc_qstats_cpu_drop(struct Qdisc * sch)1057 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
1058 {
1059 this_cpu_inc(sch->cpu_qstats->drops);
1060 }
1061
qdisc_qstats_overlimit(struct Qdisc * sch)1062 static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
1063 {
1064 WRITE_ONCE(sch->qstats.overlimits, sch->qstats.overlimits + 1);
1065 }
1066
qdisc_qstats_copy(struct gnet_dump * d,const struct Qdisc * sch)1067 static inline int qdisc_qstats_copy(struct gnet_dump *d, const struct Qdisc *sch)
1068 {
1069 __u32 qlen = qdisc_qlen_sum(sch);
1070
1071 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
1072 }
1073
qdisc_qstats_qlen_backlog(const struct Qdisc * sch,u32 * qlen,u32 * backlog)1074 static inline void qdisc_qstats_qlen_backlog(const struct Qdisc *sch,
1075 u32 *qlen, u32 *backlog)
1076 {
1077 struct gnet_stats_queue qstats = { 0 };
1078
1079 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
1080 *qlen = qstats.qlen + qdisc_qlen_lockless(sch);
1081 *backlog = qstats.backlog;
1082 }
1083
qdisc_purge_queue(struct Qdisc * sch)1084 static inline void qdisc_purge_queue(struct Qdisc *sch)
1085 {
1086 __u32 qlen, backlog;
1087
1088 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
1089 qdisc_reset(sch);
1090 qdisc_tree_reduce_backlog(sch, qlen, backlog);
1091 }
1092
__qdisc_enqueue_tail(struct sk_buff * skb,struct qdisc_skb_head * qh)1093 static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
1094 struct qdisc_skb_head *qh)
1095 {
1096 struct sk_buff *last = qh->tail;
1097
1098 if (last) {
1099 skb->next = NULL;
1100 last->next = skb;
1101 qh->tail = skb;
1102 } else {
1103 qh->tail = skb;
1104 qh->head = skb;
1105 }
1106 WRITE_ONCE(qh->qlen, qh->qlen + 1);
1107 }
1108
qdisc_enqueue_tail(struct sk_buff * skb,struct Qdisc * sch)1109 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
1110 {
1111 __qdisc_enqueue_tail(skb, &sch->q);
1112 qdisc_qstats_backlog_inc(sch, skb);
1113 return NET_XMIT_SUCCESS;
1114 }
1115
__qdisc_enqueue_head(struct sk_buff * skb,struct qdisc_skb_head * qh)1116 static inline void __qdisc_enqueue_head(struct sk_buff *skb,
1117 struct qdisc_skb_head *qh)
1118 {
1119 skb->next = qh->head;
1120
1121 if (!qh->head)
1122 qh->tail = skb;
1123 qh->head = skb;
1124 WRITE_ONCE(qh->qlen, qh->qlen + 1);
1125 }
1126
__qdisc_dequeue_head(struct qdisc_skb_head * qh)1127 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
1128 {
1129 struct sk_buff *skb = qh->head;
1130
1131 if (likely(skb != NULL)) {
1132 qh->head = skb->next;
1133 WRITE_ONCE(qh->qlen, qh->qlen - 1);
1134 if (qh->head == NULL)
1135 qh->tail = NULL;
1136 skb->next = NULL;
1137 }
1138
1139 return skb;
1140 }
1141
qdisc_dequeue_internal(struct Qdisc * sch,bool direct)1142 static inline struct sk_buff *qdisc_dequeue_internal(struct Qdisc *sch, bool direct)
1143 {
1144 struct sk_buff *skb;
1145
1146 skb = __skb_dequeue(&sch->gso_skb);
1147 if (skb) {
1148 qdisc_qlen_dec(sch);
1149 qdisc_qstats_backlog_dec(sch, skb);
1150 return skb;
1151 }
1152 if (direct) {
1153 skb = __qdisc_dequeue_head(&sch->q);
1154 if (skb)
1155 qdisc_qstats_backlog_dec(sch, skb);
1156 return skb;
1157 } else {
1158 return sch->dequeue(sch);
1159 }
1160 }
1161
qdisc_dequeue_head(struct Qdisc * sch)1162 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
1163 {
1164 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
1165
1166 if (likely(skb != NULL)) {
1167 qdisc_qstats_backlog_dec(sch, skb);
1168 qdisc_bstats_update(sch, skb);
1169 }
1170
1171 return skb;
1172 }
1173
1174 struct tc_skb_cb {
1175 struct qdisc_skb_cb qdisc_cb;
1176 u32 drop_reason;
1177
1178 u16 zone; /* Only valid if qdisc_skb_cb(skb)->post_ct = true */
1179 u16 mru;
1180 };
1181
tc_skb_cb(const struct sk_buff * skb)1182 static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
1183 {
1184 struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
1185
1186 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
1187 return cb;
1188 }
1189
1190 /* TC classifier accessors - use enum skb_drop_reason */
1191 static inline enum skb_drop_reason
tcf_get_drop_reason(const struct sk_buff * skb)1192 tcf_get_drop_reason(const struct sk_buff *skb)
1193 {
1194 return (enum skb_drop_reason)tc_skb_cb(skb)->drop_reason;
1195 }
1196
tcf_set_drop_reason(const struct sk_buff * skb,enum skb_drop_reason reason)1197 static inline void tcf_set_drop_reason(const struct sk_buff *skb,
1198 enum skb_drop_reason reason)
1199 {
1200 tc_skb_cb(skb)->drop_reason = (enum qdisc_drop_reason)reason;
1201 }
1202
1203 /* Qdisc accessors - use enum qdisc_drop_reason */
1204 static inline enum qdisc_drop_reason
tcf_get_qdisc_drop_reason(const struct sk_buff * skb)1205 tcf_get_qdisc_drop_reason(const struct sk_buff *skb)
1206 {
1207 return tc_skb_cb(skb)->drop_reason;
1208 }
1209
tcf_set_qdisc_drop_reason(const struct sk_buff * skb,enum qdisc_drop_reason reason)1210 static inline void tcf_set_qdisc_drop_reason(const struct sk_buff *skb,
1211 enum qdisc_drop_reason reason)
1212 {
1213 tc_skb_cb(skb)->drop_reason = reason;
1214 }
1215
1216 void __tcf_kfree_skb_list(struct sk_buff *skb, struct Qdisc *q,
1217 struct netdev_queue *txq, struct net_device *dev);
1218
tcf_kfree_skb_list(struct sk_buff * skb,struct Qdisc * q,struct netdev_queue * txq,struct net_device * dev)1219 static inline void tcf_kfree_skb_list(struct sk_buff *skb, struct Qdisc *q,
1220 struct netdev_queue *txq,
1221 struct net_device *dev)
1222 {
1223 if (unlikely(skb))
1224 __tcf_kfree_skb_list(skb, q, txq, dev);
1225 }
1226
qdisc_dequeue_drop(struct Qdisc * q,struct sk_buff * skb,enum qdisc_drop_reason reason)1227 static inline void qdisc_dequeue_drop(struct Qdisc *q, struct sk_buff *skb,
1228 enum qdisc_drop_reason reason)
1229 {
1230 struct Qdisc *root;
1231
1232 DEBUG_NET_WARN_ON_ONCE(!(q->flags & TCQ_F_DEQUEUE_DROPS));
1233 DEBUG_NET_WARN_ON_ONCE(q->flags & TCQ_F_NOLOCK);
1234
1235 rcu_read_lock();
1236 root = qdisc_root_sleeping(q);
1237
1238 if (root->flags & TCQ_F_DEQUEUE_DROPS) {
1239 tcf_set_qdisc_drop_reason(skb, reason);
1240 skb->next = root->to_free;
1241 root->to_free = skb;
1242 } else {
1243 kfree_skb_reason(skb, (enum skb_drop_reason)reason);
1244 }
1245 rcu_read_unlock();
1246 }
1247
1248 /* Instead of calling kfree_skb() while root qdisc lock is held,
1249 * queue the skb for future freeing at end of __dev_xmit_skb()
1250 */
__qdisc_drop(struct sk_buff * skb,struct sk_buff ** to_free)1251 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1252 {
1253 skb->next = *to_free;
1254 *to_free = skb;
1255 }
1256
__qdisc_drop_all(struct sk_buff * skb,struct sk_buff ** to_free)1257 static inline void __qdisc_drop_all(struct sk_buff *skb,
1258 struct sk_buff **to_free)
1259 {
1260 if (skb->prev)
1261 skb->prev->next = *to_free;
1262 else
1263 skb->next = *to_free;
1264 *to_free = skb;
1265 }
1266
__qdisc_queue_drop_head(struct Qdisc * sch,struct qdisc_skb_head * qh,struct sk_buff ** to_free)1267 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1268 struct qdisc_skb_head *qh,
1269 struct sk_buff **to_free)
1270 {
1271 struct sk_buff *skb = __qdisc_dequeue_head(qh);
1272
1273 if (likely(skb != NULL)) {
1274 unsigned int len = qdisc_pkt_len(skb);
1275
1276 qdisc_qstats_backlog_dec(sch, skb);
1277 __qdisc_drop(skb, to_free);
1278 return len;
1279 }
1280
1281 return 0;
1282 }
1283
qdisc_peek_head(struct Qdisc * sch)1284 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1285 {
1286 const struct qdisc_skb_head *qh = &sch->q;
1287
1288 return qh->head;
1289 }
1290
1291 /* generic pseudo peek method for non-work-conserving qdisc */
qdisc_peek_dequeued(struct Qdisc * sch)1292 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1293 {
1294 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1295
1296 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1297 if (!skb) {
1298 skb = sch->dequeue(sch);
1299
1300 if (skb) {
1301 __skb_queue_head(&sch->gso_skb, skb);
1302 /* it's still part of the queue */
1303 qdisc_qstats_backlog_inc(sch, skb);
1304 qdisc_qlen_inc(sch);
1305 }
1306 }
1307
1308 return skb;
1309 }
1310
qdisc_update_stats_at_dequeue(struct Qdisc * sch,struct sk_buff * skb)1311 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1312 struct sk_buff *skb)
1313 {
1314 if (qdisc_is_percpu_stats(sch)) {
1315 qdisc_qstats_cpu_backlog_dec(sch, skb);
1316 qdisc_bstats_cpu_update(sch, skb);
1317 qdisc_qstats_cpu_qlen_dec(sch);
1318 } else {
1319 qdisc_qstats_backlog_dec(sch, skb);
1320 qdisc_bstats_update(sch, skb);
1321 qdisc_qlen_dec(sch);
1322 }
1323 }
1324
qdisc_update_stats_at_enqueue(struct Qdisc * sch,unsigned int pkt_len)1325 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1326 unsigned int pkt_len)
1327 {
1328 if (qdisc_is_percpu_stats(sch)) {
1329 qdisc_qstats_cpu_qlen_inc(sch);
1330 this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1331 } else {
1332 qstats_backlog_add(sch, pkt_len);
1333 qdisc_qlen_inc(sch);
1334 }
1335 }
1336
1337 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
qdisc_dequeue_peeked(struct Qdisc * sch)1338 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1339 {
1340 struct sk_buff *skb = skb_peek(&sch->gso_skb);
1341
1342 if (skb) {
1343 skb = __skb_dequeue(&sch->gso_skb);
1344 if (qdisc_is_percpu_stats(sch)) {
1345 qdisc_qstats_cpu_backlog_dec(sch, skb);
1346 qdisc_qstats_cpu_qlen_dec(sch);
1347 } else {
1348 qdisc_qstats_backlog_dec(sch, skb);
1349 qdisc_qlen_dec(sch);
1350 }
1351 } else {
1352 skb = sch->dequeue(sch);
1353 }
1354
1355 return skb;
1356 }
1357
__qdisc_reset_queue(struct qdisc_skb_head * qh)1358 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1359 {
1360 /*
1361 * We do not know the backlog in bytes of this list, it
1362 * is up to the caller to correct it
1363 */
1364 ASSERT_RTNL();
1365 if (qh->qlen) {
1366 rtnl_kfree_skbs(qh->head, qh->tail);
1367
1368 qh->head = NULL;
1369 qh->tail = NULL;
1370 WRITE_ONCE(qh->qlen, 0);
1371 }
1372 }
1373
qdisc_reset_queue(struct Qdisc * sch)1374 static inline void qdisc_reset_queue(struct Qdisc *sch)
1375 {
1376 __qdisc_reset_queue(&sch->q);
1377 }
1378
qdisc_replace(struct Qdisc * sch,struct Qdisc * new,struct Qdisc ** pold)1379 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1380 struct Qdisc **pold)
1381 {
1382 struct Qdisc *old;
1383
1384 sch_tree_lock(sch);
1385 old = *pold;
1386 *pold = new;
1387 if (old != NULL)
1388 qdisc_purge_queue(old);
1389 sch_tree_unlock(sch);
1390
1391 return old;
1392 }
1393
rtnl_qdisc_drop(struct sk_buff * skb,struct Qdisc * sch)1394 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1395 {
1396 rtnl_kfree_skbs(skb, skb);
1397 qdisc_qstats_drop(sch);
1398 }
1399
qdisc_drop_cpu(struct sk_buff * skb,struct Qdisc * sch,struct sk_buff ** to_free)1400 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1401 struct sk_buff **to_free)
1402 {
1403 __qdisc_drop(skb, to_free);
1404 qdisc_qstats_cpu_drop(sch);
1405
1406 return NET_XMIT_DROP;
1407 }
1408
qdisc_drop(struct sk_buff * skb,struct Qdisc * sch,struct sk_buff ** to_free)1409 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1410 struct sk_buff **to_free)
1411 {
1412 __qdisc_drop(skb, to_free);
1413 qdisc_qstats_drop(sch);
1414
1415 return NET_XMIT_DROP;
1416 }
1417
qdisc_drop_reason(struct sk_buff * skb,struct Qdisc * sch,struct sk_buff ** to_free,enum qdisc_drop_reason reason)1418 static inline int qdisc_drop_reason(struct sk_buff *skb, struct Qdisc *sch,
1419 struct sk_buff **to_free,
1420 enum qdisc_drop_reason reason)
1421 {
1422 tcf_set_qdisc_drop_reason(skb, reason);
1423 return qdisc_drop(skb, sch, to_free);
1424 }
1425
qdisc_drop_all(struct sk_buff * skb,struct Qdisc * sch,struct sk_buff ** to_free)1426 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1427 struct sk_buff **to_free)
1428 {
1429 __qdisc_drop_all(skb, to_free);
1430 qdisc_qstats_drop(sch);
1431
1432 return NET_XMIT_DROP;
1433 }
1434
1435 struct psched_ratecfg {
1436 u64 rate_bytes_ps; /* bytes per second */
1437 u32 mult;
1438 u16 overhead;
1439 u16 mpu;
1440 u8 linklayer;
1441 u8 shift;
1442 };
1443
psched_l2t_ns(const struct psched_ratecfg * r,unsigned int len)1444 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1445 unsigned int len)
1446 {
1447 len += r->overhead;
1448
1449 if (len < r->mpu)
1450 len = r->mpu;
1451
1452 if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1453 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1454
1455 return ((u64)len * r->mult) >> r->shift;
1456 }
1457
1458 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1459 const struct tc_ratespec *conf,
1460 u64 rate64);
1461
psched_ratecfg_getrate(struct tc_ratespec * res,const struct psched_ratecfg * r)1462 static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1463 const struct psched_ratecfg *r)
1464 {
1465 memset(res, 0, sizeof(*res));
1466
1467 /* legacy struct tc_ratespec has a 32bit @rate field
1468 * Qdisc using 64bit rate should add new attributes
1469 * in order to maintain compatibility.
1470 */
1471 res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1472
1473 res->overhead = r->overhead;
1474 res->mpu = r->mpu;
1475 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1476 }
1477
1478 struct psched_pktrate {
1479 u64 rate_pkts_ps; /* packets per second */
1480 u32 mult;
1481 u8 shift;
1482 };
1483
psched_pkt2t_ns(const struct psched_pktrate * r,unsigned int pkt_num)1484 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1485 unsigned int pkt_num)
1486 {
1487 return ((u64)pkt_num * r->mult) >> r->shift;
1488 }
1489
1490 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1491
1492 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1493 * The fast path only needs to access filter list and to update stats
1494 */
1495 struct mini_Qdisc {
1496 struct tcf_proto *filter_list;
1497 struct tcf_block *block;
1498 struct gnet_stats_basic_sync __percpu *cpu_bstats;
1499 struct gnet_stats_queue __percpu *cpu_qstats;
1500 unsigned long rcu_state;
1501 };
1502
mini_qdisc_bstats_cpu_update(struct mini_Qdisc * miniq,const struct sk_buff * skb)1503 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1504 const struct sk_buff *skb)
1505 {
1506 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1507 }
1508
mini_qdisc_qstats_cpu_drop(struct mini_Qdisc * miniq)1509 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1510 {
1511 this_cpu_inc(miniq->cpu_qstats->drops);
1512 }
1513
1514 struct mini_Qdisc_pair {
1515 struct mini_Qdisc miniq1;
1516 struct mini_Qdisc miniq2;
1517 struct mini_Qdisc __rcu **p_miniq;
1518 };
1519
1520 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1521 struct tcf_proto *tp_head);
1522 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1523 struct mini_Qdisc __rcu **p_miniq);
1524 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1525 struct tcf_block *block);
1526
mini_qdisc_pair_inited(struct mini_Qdisc_pair * miniqp)1527 static inline bool mini_qdisc_pair_inited(struct mini_Qdisc_pair *miniqp)
1528 {
1529 return !!miniqp->p_miniq;
1530 }
1531
1532 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1533
1534 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1535
1536 /* Make sure qdisc is no longer in SCHED state. */
qdisc_synchronize(const struct Qdisc * q)1537 static inline void qdisc_synchronize(const struct Qdisc *q)
1538 {
1539 while (test_bit(__QDISC_STATE_SCHED, &q->state))
1540 msleep(1);
1541 }
1542
1543 #endif
1544