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 | |
24 | struct Qdisc_ops; |
25 | struct qdisc_walker; |
26 | struct tcf_walker; |
27 | struct module; |
28 | struct bpf_flow_keys; |
29 | |
30 | struct qdisc_rate_table { |
31 | struct tc_ratespec rate; |
32 | u32 data[256]; |
33 | struct qdisc_rate_table *next; |
34 | int refcnt; |
35 | }; |
36 | |
37 | enum qdisc_state_t { |
38 | __QDISC_STATE_SCHED, |
39 | __QDISC_STATE_DEACTIVATED, |
40 | __QDISC_STATE_MISSED, |
41 | __QDISC_STATE_DRAINING, |
42 | }; |
43 | |
44 | enum qdisc_state2_t { |
45 | /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly. |
46 | * Use qdisc_run_begin/end() or qdisc_is_running() instead. |
47 | */ |
48 | __QDISC_STATE2_RUNNING, |
49 | }; |
50 | |
51 | #define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED) |
52 | #define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING) |
53 | |
54 | #define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \ |
55 | QDISC_STATE_DRAINING) |
56 | |
57 | struct qdisc_size_table { |
58 | struct rcu_head rcu; |
59 | struct list_head list; |
60 | struct tc_sizespec szopts; |
61 | int refcnt; |
62 | u16 data[]; |
63 | }; |
64 | |
65 | /* similar to sk_buff_head, but skb->prev pointer is undefined. */ |
66 | struct qdisc_skb_head { |
67 | struct sk_buff *head; |
68 | struct sk_buff *tail; |
69 | __u32 qlen; |
70 | spinlock_t lock; |
71 | }; |
72 | |
73 | struct Qdisc { |
74 | int (*enqueue)(struct sk_buff *skb, |
75 | struct Qdisc *sch, |
76 | struct sk_buff **to_free); |
77 | struct sk_buff * (*dequeue)(struct Qdisc *sch); |
78 | unsigned int flags; |
79 | #define TCQ_F_BUILTIN 1 |
80 | #define TCQ_F_INGRESS 2 |
81 | #define TCQ_F_CAN_BYPASS 4 |
82 | #define TCQ_F_MQROOT 8 |
83 | #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for |
84 | * q->dev_queue : It can test |
85 | * netif_xmit_frozen_or_stopped() before |
86 | * dequeueing next packet. |
87 | * Its true for MQ/MQPRIO slaves, or non |
88 | * multiqueue device. |
89 | */ |
90 | #define TCQ_F_WARN_NONWC (1 << 16) |
91 | #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */ |
92 | #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy : |
93 | * qdisc_tree_decrease_qlen() should stop. |
94 | */ |
95 | #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */ |
96 | #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */ |
97 | #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */ |
98 | u32 limit; |
99 | const struct Qdisc_ops *ops; |
100 | struct qdisc_size_table __rcu *stab; |
101 | struct hlist_node hash; |
102 | u32 handle; |
103 | u32 parent; |
104 | |
105 | struct netdev_queue *dev_queue; |
106 | |
107 | struct net_rate_estimator __rcu *rate_est; |
108 | struct gnet_stats_basic_sync __percpu *cpu_bstats; |
109 | struct gnet_stats_queue __percpu *cpu_qstats; |
110 | int pad; |
111 | refcount_t refcnt; |
112 | |
113 | /* |
114 | * For performance sake on SMP, we put highly modified fields at the end |
115 | */ |
116 | struct sk_buff_head gso_skb ____cacheline_aligned_in_smp; |
117 | struct qdisc_skb_head q; |
118 | struct gnet_stats_basic_sync bstats; |
119 | struct gnet_stats_queue qstats; |
120 | int owner; |
121 | unsigned long state; |
122 | unsigned long state2; /* must be written under qdisc spinlock */ |
123 | struct Qdisc *next_sched; |
124 | struct sk_buff_head skb_bad_txq; |
125 | |
126 | spinlock_t busylock ____cacheline_aligned_in_smp; |
127 | spinlock_t seqlock; |
128 | |
129 | struct rcu_head rcu; |
130 | netdevice_tracker dev_tracker; |
131 | struct lock_class_key root_lock_key; |
132 | /* private data */ |
133 | long privdata[] ____cacheline_aligned; |
134 | }; |
135 | |
136 | static inline void qdisc_refcount_inc(struct Qdisc *qdisc) |
137 | { |
138 | if (qdisc->flags & TCQ_F_BUILTIN) |
139 | return; |
140 | refcount_inc(r: &qdisc->refcnt); |
141 | } |
142 | |
143 | static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc) |
144 | { |
145 | if (qdisc->flags & TCQ_F_BUILTIN) |
146 | return true; |
147 | return refcount_dec_if_one(r: &qdisc->refcnt); |
148 | } |
149 | |
150 | /* Intended to be used by unlocked users, when concurrent qdisc release is |
151 | * possible. |
152 | */ |
153 | |
154 | static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc) |
155 | { |
156 | if (qdisc->flags & TCQ_F_BUILTIN) |
157 | return qdisc; |
158 | if (refcount_inc_not_zero(r: &qdisc->refcnt)) |
159 | return qdisc; |
160 | return NULL; |
161 | } |
162 | |
163 | /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc |
164 | * root_lock section, or provide their own memory barriers -- ordering |
165 | * against qdisc_run_begin/end() atomic bit operations. |
166 | */ |
167 | static inline bool qdisc_is_running(struct Qdisc *qdisc) |
168 | { |
169 | if (qdisc->flags & TCQ_F_NOLOCK) |
170 | return spin_is_locked(lock: &qdisc->seqlock); |
171 | return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); |
172 | } |
173 | |
174 | static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc) |
175 | { |
176 | return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY); |
177 | } |
178 | |
179 | static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) |
180 | { |
181 | return q->flags & TCQ_F_CPUSTATS; |
182 | } |
183 | |
184 | static inline bool qdisc_is_empty(const struct Qdisc *qdisc) |
185 | { |
186 | if (qdisc_is_percpu_stats(q: qdisc)) |
187 | return nolock_qdisc_is_empty(qdisc); |
188 | return !READ_ONCE(qdisc->q.qlen); |
189 | } |
190 | |
191 | /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with |
192 | * the qdisc root lock acquired. |
193 | */ |
194 | static inline bool qdisc_run_begin(struct Qdisc *qdisc) |
195 | { |
196 | if (qdisc->flags & TCQ_F_NOLOCK) { |
197 | if (spin_trylock(lock: &qdisc->seqlock)) |
198 | return true; |
199 | |
200 | /* No need to insist if the MISSED flag was already set. |
201 | * Note that test_and_set_bit() also gives us memory ordering |
202 | * guarantees wrt potential earlier enqueue() and below |
203 | * spin_trylock(), both of which are necessary to prevent races |
204 | */ |
205 | if (test_and_set_bit(nr: __QDISC_STATE_MISSED, addr: &qdisc->state)) |
206 | return false; |
207 | |
208 | /* Try to take the lock again to make sure that we will either |
209 | * grab it or the CPU that still has it will see MISSED set |
210 | * when testing it in qdisc_run_end() |
211 | */ |
212 | return spin_trylock(lock: &qdisc->seqlock); |
213 | } |
214 | return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); |
215 | } |
216 | |
217 | static inline void qdisc_run_end(struct Qdisc *qdisc) |
218 | { |
219 | if (qdisc->flags & TCQ_F_NOLOCK) { |
220 | spin_unlock(lock: &qdisc->seqlock); |
221 | |
222 | /* spin_unlock() only has store-release semantic. The unlock |
223 | * and test_bit() ordering is a store-load ordering, so a full |
224 | * memory barrier is needed here. |
225 | */ |
226 | smp_mb(); |
227 | |
228 | if (unlikely(test_bit(__QDISC_STATE_MISSED, |
229 | &qdisc->state))) |
230 | __netif_schedule(q: qdisc); |
231 | } else { |
232 | __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); |
233 | } |
234 | } |
235 | |
236 | static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) |
237 | { |
238 | return qdisc->flags & TCQ_F_ONETXQUEUE; |
239 | } |
240 | |
241 | static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) |
242 | { |
243 | return netdev_queue_dql_avail(txq); |
244 | } |
245 | |
246 | struct Qdisc_class_ops { |
247 | unsigned int flags; |
248 | /* Child qdisc manipulation */ |
249 | struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); |
250 | int (*graft)(struct Qdisc *, unsigned long cl, |
251 | struct Qdisc *, struct Qdisc **, |
252 | struct netlink_ext_ack *extack); |
253 | struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); |
254 | void (*qlen_notify)(struct Qdisc *, unsigned long); |
255 | |
256 | /* Class manipulation routines */ |
257 | unsigned long (*find)(struct Qdisc *, u32 classid); |
258 | int (*change)(struct Qdisc *, u32, u32, |
259 | struct nlattr **, unsigned long *, |
260 | struct netlink_ext_ack *); |
261 | int (*delete)(struct Qdisc *, unsigned long, |
262 | struct netlink_ext_ack *); |
263 | void (*walk)(struct Qdisc *, struct qdisc_walker * arg); |
264 | |
265 | /* Filter manipulation */ |
266 | struct tcf_block * (*tcf_block)(struct Qdisc *sch, |
267 | unsigned long arg, |
268 | struct netlink_ext_ack *extack); |
269 | unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, |
270 | u32 classid); |
271 | void (*unbind_tcf)(struct Qdisc *, unsigned long); |
272 | |
273 | /* rtnetlink specific */ |
274 | int (*dump)(struct Qdisc *, unsigned long, |
275 | struct sk_buff *skb, struct tcmsg*); |
276 | int (*dump_stats)(struct Qdisc *, unsigned long, |
277 | struct gnet_dump *); |
278 | }; |
279 | |
280 | /* Qdisc_class_ops flag values */ |
281 | |
282 | /* Implements API that doesn't require rtnl lock */ |
283 | enum qdisc_class_ops_flags { |
284 | QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, |
285 | }; |
286 | |
287 | struct Qdisc_ops { |
288 | struct Qdisc_ops *next; |
289 | const struct Qdisc_class_ops *cl_ops; |
290 | char id[IFNAMSIZ]; |
291 | int priv_size; |
292 | unsigned int static_flags; |
293 | |
294 | int (*enqueue)(struct sk_buff *skb, |
295 | struct Qdisc *sch, |
296 | struct sk_buff **to_free); |
297 | struct sk_buff * (*dequeue)(struct Qdisc *); |
298 | struct sk_buff * (*peek)(struct Qdisc *); |
299 | |
300 | int (*init)(struct Qdisc *sch, struct nlattr *arg, |
301 | struct netlink_ext_ack *extack); |
302 | void (*reset)(struct Qdisc *); |
303 | void (*destroy)(struct Qdisc *); |
304 | int (*change)(struct Qdisc *sch, |
305 | struct nlattr *arg, |
306 | struct netlink_ext_ack *extack); |
307 | void (*attach)(struct Qdisc *sch); |
308 | int (*change_tx_queue_len)(struct Qdisc *, unsigned int); |
309 | void (*change_real_num_tx)(struct Qdisc *sch, |
310 | unsigned int new_real_tx); |
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 | struct tcf_result { |
326 | union { |
327 | struct { |
328 | unsigned long class; |
329 | u32 classid; |
330 | }; |
331 | const struct tcf_proto *goto_tp; |
332 | }; |
333 | }; |
334 | |
335 | struct tcf_chain; |
336 | |
337 | struct tcf_proto_ops { |
338 | struct list_head head; |
339 | char kind[IFNAMSIZ]; |
340 | |
341 | int (*classify)(struct sk_buff *, |
342 | const struct tcf_proto *, |
343 | struct tcf_result *); |
344 | int (*init)(struct tcf_proto*); |
345 | void (*destroy)(struct tcf_proto *tp, bool rtnl_held, |
346 | struct netlink_ext_ack *extack); |
347 | |
348 | void* (*get)(struct tcf_proto*, u32 handle); |
349 | void (*put)(struct tcf_proto *tp, void *f); |
350 | int (*change)(struct net *net, struct sk_buff *, |
351 | struct tcf_proto*, unsigned long, |
352 | u32 handle, struct nlattr **, |
353 | void **, u32, |
354 | struct netlink_ext_ack *); |
355 | int (*delete)(struct tcf_proto *tp, void *arg, |
356 | bool *last, bool rtnl_held, |
357 | struct netlink_ext_ack *); |
358 | bool (*delete_empty)(struct tcf_proto *tp); |
359 | void (*walk)(struct tcf_proto *tp, |
360 | struct tcf_walker *arg, bool rtnl_held); |
361 | int (*reoffload)(struct tcf_proto *tp, bool add, |
362 | flow_setup_cb_t *cb, void *cb_priv, |
363 | struct netlink_ext_ack *extack); |
364 | void (*hw_add)(struct tcf_proto *tp, |
365 | void *type_data); |
366 | void (*hw_del)(struct tcf_proto *tp, |
367 | void *type_data); |
368 | void (*bind_class)(void *, u32, unsigned long, |
369 | void *, unsigned long); |
370 | void * (*tmplt_create)(struct net *net, |
371 | struct tcf_chain *chain, |
372 | struct nlattr **tca, |
373 | struct netlink_ext_ack *extack); |
374 | void (*tmplt_destroy)(void *tmplt_priv); |
375 | void (*tmplt_reoffload)(struct tcf_chain *chain, |
376 | bool add, |
377 | flow_setup_cb_t *cb, |
378 | void *cb_priv); |
379 | struct tcf_exts * (*get_exts)(const struct tcf_proto *tp, |
380 | u32 handle); |
381 | |
382 | /* rtnetlink specific */ |
383 | int (*dump)(struct net*, struct tcf_proto*, void *, |
384 | struct sk_buff *skb, struct tcmsg*, |
385 | bool); |
386 | int (*terse_dump)(struct net *net, |
387 | struct tcf_proto *tp, void *fh, |
388 | struct sk_buff *skb, |
389 | struct tcmsg *t, bool rtnl_held); |
390 | int (*tmplt_dump)(struct sk_buff *skb, |
391 | struct net *net, |
392 | void *tmplt_priv); |
393 | |
394 | struct module *owner; |
395 | int flags; |
396 | }; |
397 | |
398 | /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags |
399 | * are expected to implement tcf_proto_ops->delete_empty(), otherwise race |
400 | * conditions can occur when filters are inserted/deleted simultaneously. |
401 | */ |
402 | enum tcf_proto_ops_flags { |
403 | TCF_PROTO_OPS_DOIT_UNLOCKED = 1, |
404 | }; |
405 | |
406 | struct tcf_proto { |
407 | /* Fast access part */ |
408 | struct tcf_proto __rcu *next; |
409 | void __rcu *root; |
410 | |
411 | /* called under RCU BH lock*/ |
412 | int (*classify)(struct sk_buff *, |
413 | const struct tcf_proto *, |
414 | struct tcf_result *); |
415 | __be16 protocol; |
416 | |
417 | /* All the rest */ |
418 | u32 prio; |
419 | void *data; |
420 | const struct tcf_proto_ops *ops; |
421 | struct tcf_chain *chain; |
422 | /* Lock protects tcf_proto shared state and can be used by unlocked |
423 | * classifiers to protect their private data. |
424 | */ |
425 | spinlock_t lock; |
426 | bool deleting; |
427 | bool counted; |
428 | bool usesw; |
429 | refcount_t refcnt; |
430 | struct rcu_head rcu; |
431 | struct hlist_node destroy_ht_node; |
432 | }; |
433 | |
434 | struct qdisc_skb_cb { |
435 | struct { |
436 | unsigned int pkt_len; |
437 | u16 slave_dev_queue_mapping; |
438 | u16 tc_classid; |
439 | }; |
440 | #define QDISC_CB_PRIV_LEN 20 |
441 | unsigned char data[QDISC_CB_PRIV_LEN]; |
442 | }; |
443 | |
444 | typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); |
445 | |
446 | struct tcf_chain { |
447 | /* Protects filter_chain. */ |
448 | struct mutex filter_chain_lock; |
449 | struct tcf_proto __rcu *filter_chain; |
450 | struct list_head list; |
451 | struct tcf_block *block; |
452 | u32 index; /* chain index */ |
453 | unsigned int refcnt; |
454 | unsigned int action_refcnt; |
455 | bool explicitly_created; |
456 | bool flushing; |
457 | const struct tcf_proto_ops *tmplt_ops; |
458 | void *tmplt_priv; |
459 | struct rcu_head rcu; |
460 | }; |
461 | |
462 | struct tcf_block { |
463 | struct xarray ports; /* datapath accessible */ |
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 useswcnt; |
479 | atomic_t offloadcnt; /* Number of oddloaded filters */ |
480 | unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ |
481 | unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ |
482 | struct { |
483 | struct tcf_chain *chain; |
484 | struct list_head filter_chain_list; |
485 | } chain0; |
486 | struct rcu_head rcu; |
487 | DECLARE_HASHTABLE(proto_destroy_ht, 7); |
488 | struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ |
489 | }; |
490 | |
491 | struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index); |
492 | |
493 | static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) |
494 | { |
495 | return lockdep_is_held(&chain->filter_chain_lock); |
496 | } |
497 | |
498 | static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) |
499 | { |
500 | return lockdep_is_held(&tp->lock); |
501 | } |
502 | |
503 | #define tcf_chain_dereference(p, chain) \ |
504 | rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) |
505 | |
506 | #define tcf_proto_dereference(p, tp) \ |
507 | rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) |
508 | |
509 | static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) |
510 | { |
511 | struct qdisc_skb_cb *qcb; |
512 | |
513 | BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); |
514 | BUILD_BUG_ON(sizeof(qcb->data) < sz); |
515 | } |
516 | |
517 | static inline int qdisc_qlen(const struct Qdisc *q) |
518 | { |
519 | return q->q.qlen; |
520 | } |
521 | |
522 | static inline int qdisc_qlen_sum(const struct Qdisc *q) |
523 | { |
524 | __u32 qlen = q->qstats.qlen; |
525 | int i; |
526 | |
527 | if (qdisc_is_percpu_stats(q)) { |
528 | for_each_possible_cpu(i) |
529 | qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; |
530 | } else { |
531 | qlen += q->q.qlen; |
532 | } |
533 | |
534 | return qlen; |
535 | } |
536 | |
537 | static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) |
538 | { |
539 | return (struct qdisc_skb_cb *)skb->cb; |
540 | } |
541 | |
542 | static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) |
543 | { |
544 | return &qdisc->q.lock; |
545 | } |
546 | |
547 | static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) |
548 | { |
549 | struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); |
550 | |
551 | return q; |
552 | } |
553 | |
554 | static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) |
555 | { |
556 | return rcu_dereference_bh(qdisc->dev_queue->qdisc); |
557 | } |
558 | |
559 | static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) |
560 | { |
561 | return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping); |
562 | } |
563 | |
564 | static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) |
565 | { |
566 | struct Qdisc *root = qdisc_root_sleeping(qdisc); |
567 | |
568 | ASSERT_RTNL(); |
569 | return qdisc_lock(qdisc: root); |
570 | } |
571 | |
572 | static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) |
573 | { |
574 | return qdisc->dev_queue->dev; |
575 | } |
576 | |
577 | static inline void sch_tree_lock(struct Qdisc *q) |
578 | { |
579 | if (q->flags & TCQ_F_MQROOT) |
580 | spin_lock_bh(lock: qdisc_lock(qdisc: q)); |
581 | else |
582 | spin_lock_bh(lock: qdisc_root_sleeping_lock(qdisc: q)); |
583 | } |
584 | |
585 | static inline void sch_tree_unlock(struct Qdisc *q) |
586 | { |
587 | if (q->flags & TCQ_F_MQROOT) |
588 | spin_unlock_bh(lock: qdisc_lock(qdisc: q)); |
589 | else |
590 | spin_unlock_bh(lock: qdisc_root_sleeping_lock(qdisc: q)); |
591 | } |
592 | |
593 | extern struct Qdisc noop_qdisc; |
594 | extern struct Qdisc_ops noop_qdisc_ops; |
595 | extern struct Qdisc_ops pfifo_fast_ops; |
596 | extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1]; |
597 | extern struct Qdisc_ops mq_qdisc_ops; |
598 | extern struct Qdisc_ops noqueue_qdisc_ops; |
599 | extern const struct Qdisc_ops *default_qdisc_ops; |
600 | static inline const struct Qdisc_ops * |
601 | get_default_qdisc_ops(const struct net_device *dev, int ntx) |
602 | { |
603 | return ntx < dev->real_num_tx_queues ? |
604 | default_qdisc_ops : &pfifo_fast_ops; |
605 | } |
606 | |
607 | struct Qdisc_class_common { |
608 | u32 classid; |
609 | unsigned int filter_cnt; |
610 | struct hlist_node hnode; |
611 | }; |
612 | |
613 | struct Qdisc_class_hash { |
614 | struct hlist_head *hash; |
615 | unsigned int hashsize; |
616 | unsigned int hashmask; |
617 | unsigned int hashelems; |
618 | }; |
619 | |
620 | static inline unsigned int qdisc_class_hash(u32 id, u32 mask) |
621 | { |
622 | id ^= id >> 8; |
623 | id ^= id >> 4; |
624 | return id & mask; |
625 | } |
626 | |
627 | static inline struct Qdisc_class_common * |
628 | qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) |
629 | { |
630 | struct Qdisc_class_common *cl; |
631 | unsigned int h; |
632 | |
633 | if (!id) |
634 | return NULL; |
635 | |
636 | h = qdisc_class_hash(id, mask: hash->hashmask); |
637 | hlist_for_each_entry(cl, &hash->hash[h], hnode) { |
638 | if (cl->classid == id) |
639 | return cl; |
640 | } |
641 | return NULL; |
642 | } |
643 | |
644 | static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl) |
645 | { |
646 | return cl->filter_cnt > 0; |
647 | } |
648 | |
649 | static inline void qdisc_class_get(struct Qdisc_class_common *cl) |
650 | { |
651 | unsigned int res; |
652 | |
653 | if (check_add_overflow(cl->filter_cnt, 1, &res)) |
654 | WARN(1, "Qdisc class overflow" ); |
655 | |
656 | cl->filter_cnt = res; |
657 | } |
658 | |
659 | static inline void qdisc_class_put(struct Qdisc_class_common *cl) |
660 | { |
661 | unsigned int res; |
662 | |
663 | if (check_sub_overflow(cl->filter_cnt, 1, &res)) |
664 | WARN(1, "Qdisc class underflow" ); |
665 | |
666 | cl->filter_cnt = res; |
667 | } |
668 | |
669 | static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) |
670 | { |
671 | u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; |
672 | |
673 | return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; |
674 | } |
675 | |
676 | int qdisc_class_hash_init(struct Qdisc_class_hash *); |
677 | void qdisc_class_hash_insert(struct Qdisc_class_hash *, |
678 | struct Qdisc_class_common *); |
679 | void qdisc_class_hash_remove(struct Qdisc_class_hash *, |
680 | struct Qdisc_class_common *); |
681 | void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); |
682 | void qdisc_class_hash_destroy(struct Qdisc_class_hash *); |
683 | |
684 | int dev_qdisc_change_tx_queue_len(struct net_device *dev); |
685 | void dev_qdisc_change_real_num_tx(struct net_device *dev, |
686 | unsigned int new_real_tx); |
687 | void dev_init_scheduler(struct net_device *dev); |
688 | void dev_shutdown(struct net_device *dev); |
689 | void dev_activate(struct net_device *dev); |
690 | void dev_deactivate(struct net_device *dev); |
691 | void dev_deactivate_many(struct list_head *head); |
692 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, |
693 | struct Qdisc *qdisc); |
694 | void qdisc_reset(struct Qdisc *qdisc); |
695 | void qdisc_destroy(struct Qdisc *qdisc); |
696 | void qdisc_put(struct Qdisc *qdisc); |
697 | void qdisc_put_unlocked(struct Qdisc *qdisc); |
698 | void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); |
699 | #ifdef CONFIG_NET_SCHED |
700 | int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, |
701 | void *type_data); |
702 | void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, |
703 | struct Qdisc *new, struct Qdisc *old, |
704 | enum tc_setup_type type, void *type_data, |
705 | struct netlink_ext_ack *extack); |
706 | #else |
707 | static inline int |
708 | qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, |
709 | void *type_data) |
710 | { |
711 | q->flags &= ~TCQ_F_OFFLOADED; |
712 | return 0; |
713 | } |
714 | |
715 | static inline void |
716 | qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, |
717 | struct Qdisc *new, struct Qdisc *old, |
718 | enum tc_setup_type type, void *type_data, |
719 | struct netlink_ext_ack *extack) |
720 | { |
721 | } |
722 | #endif |
723 | void qdisc_offload_query_caps(struct net_device *dev, |
724 | enum tc_setup_type type, |
725 | void *caps, size_t caps_len); |
726 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
727 | const struct Qdisc_ops *ops, |
728 | struct netlink_ext_ack *extack); |
729 | void qdisc_free(struct Qdisc *qdisc); |
730 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
731 | const struct Qdisc_ops *ops, u32 parentid, |
732 | struct netlink_ext_ack *extack); |
733 | void __qdisc_calculate_pkt_len(struct sk_buff *skb, |
734 | const struct qdisc_size_table *stab); |
735 | int skb_do_redirect(struct sk_buff *); |
736 | |
737 | static inline bool skb_at_tc_ingress(const struct sk_buff *skb) |
738 | { |
739 | #ifdef CONFIG_NET_XGRESS |
740 | return skb->tc_at_ingress; |
741 | #else |
742 | return false; |
743 | #endif |
744 | } |
745 | |
746 | static inline bool skb_skip_tc_classify(struct sk_buff *skb) |
747 | { |
748 | #ifdef CONFIG_NET_CLS_ACT |
749 | if (skb->tc_skip_classify) { |
750 | skb->tc_skip_classify = 0; |
751 | return true; |
752 | } |
753 | #endif |
754 | return false; |
755 | } |
756 | |
757 | /* Reset all TX qdiscs greater than index of a device. */ |
758 | static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) |
759 | { |
760 | struct Qdisc *qdisc; |
761 | |
762 | for (; i < dev->num_tx_queues; i++) { |
763 | qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); |
764 | if (qdisc) { |
765 | spin_lock_bh(lock: qdisc_lock(qdisc)); |
766 | qdisc_reset(qdisc); |
767 | spin_unlock_bh(lock: qdisc_lock(qdisc)); |
768 | } |
769 | } |
770 | } |
771 | |
772 | /* Are all TX queues of the device empty? */ |
773 | static inline bool qdisc_all_tx_empty(const struct net_device *dev) |
774 | { |
775 | unsigned int i; |
776 | |
777 | rcu_read_lock(); |
778 | for (i = 0; i < dev->num_tx_queues; i++) { |
779 | struct netdev_queue *txq = netdev_get_tx_queue(dev, index: i); |
780 | const struct Qdisc *q = rcu_dereference(txq->qdisc); |
781 | |
782 | if (!qdisc_is_empty(qdisc: q)) { |
783 | rcu_read_unlock(); |
784 | return false; |
785 | } |
786 | } |
787 | rcu_read_unlock(); |
788 | return true; |
789 | } |
790 | |
791 | /* Are any of the TX qdiscs changing? */ |
792 | static inline bool qdisc_tx_changing(const struct net_device *dev) |
793 | { |
794 | unsigned int i; |
795 | |
796 | for (i = 0; i < dev->num_tx_queues; i++) { |
797 | struct netdev_queue *txq = netdev_get_tx_queue(dev, index: i); |
798 | |
799 | if (rcu_access_pointer(txq->qdisc) != |
800 | rcu_access_pointer(txq->qdisc_sleeping)) |
801 | return true; |
802 | } |
803 | return false; |
804 | } |
805 | |
806 | /* "noqueue" qdisc identified by not having any enqueue, see noqueue_init() */ |
807 | static inline bool qdisc_txq_has_no_queue(const struct netdev_queue *txq) |
808 | { |
809 | struct Qdisc *qdisc = rcu_access_pointer(txq->qdisc); |
810 | |
811 | return qdisc->enqueue == NULL; |
812 | } |
813 | |
814 | /* Is the device using the noop qdisc on all queues? */ |
815 | static inline bool qdisc_tx_is_noop(const struct net_device *dev) |
816 | { |
817 | unsigned int i; |
818 | |
819 | for (i = 0; i < dev->num_tx_queues; i++) { |
820 | struct netdev_queue *txq = netdev_get_tx_queue(dev, index: i); |
821 | if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) |
822 | return false; |
823 | } |
824 | return true; |
825 | } |
826 | |
827 | static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) |
828 | { |
829 | return qdisc_skb_cb(skb)->pkt_len; |
830 | } |
831 | |
832 | /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ |
833 | enum net_xmit_qdisc_t { |
834 | __NET_XMIT_STOLEN = 0x00010000, |
835 | __NET_XMIT_BYPASS = 0x00020000, |
836 | }; |
837 | |
838 | #ifdef CONFIG_NET_CLS_ACT |
839 | #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) |
840 | #else |
841 | #define net_xmit_drop_count(e) (1) |
842 | #endif |
843 | |
844 | static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, |
845 | const struct Qdisc *sch) |
846 | { |
847 | #ifdef CONFIG_NET_SCHED |
848 | struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); |
849 | |
850 | if (stab) |
851 | __qdisc_calculate_pkt_len(skb, stab); |
852 | #endif |
853 | } |
854 | |
855 | static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
856 | struct sk_buff **to_free) |
857 | { |
858 | return sch->enqueue(skb, sch, to_free); |
859 | } |
860 | |
861 | static inline void _bstats_update(struct gnet_stats_basic_sync *bstats, |
862 | __u64 bytes, __u64 packets) |
863 | { |
864 | u64_stats_update_begin(syncp: &bstats->syncp); |
865 | u64_stats_add(p: &bstats->bytes, val: bytes); |
866 | u64_stats_add(p: &bstats->packets, val: packets); |
867 | u64_stats_update_end(syncp: &bstats->syncp); |
868 | } |
869 | |
870 | static inline void bstats_update(struct gnet_stats_basic_sync *bstats, |
871 | const struct sk_buff *skb) |
872 | { |
873 | _bstats_update(bstats, |
874 | bytes: qdisc_pkt_len(skb), |
875 | packets: skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); |
876 | } |
877 | |
878 | static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, |
879 | const struct sk_buff *skb) |
880 | { |
881 | bstats_update(this_cpu_ptr(sch->cpu_bstats), skb); |
882 | } |
883 | |
884 | static inline void qdisc_bstats_update(struct Qdisc *sch, |
885 | const struct sk_buff *skb) |
886 | { |
887 | bstats_update(bstats: &sch->bstats, skb); |
888 | } |
889 | |
890 | static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, |
891 | const struct sk_buff *skb) |
892 | { |
893 | sch->qstats.backlog -= qdisc_pkt_len(skb); |
894 | } |
895 | |
896 | static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, |
897 | const struct sk_buff *skb) |
898 | { |
899 | this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); |
900 | } |
901 | |
902 | static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, |
903 | const struct sk_buff *skb) |
904 | { |
905 | sch->qstats.backlog += qdisc_pkt_len(skb); |
906 | } |
907 | |
908 | static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, |
909 | const struct sk_buff *skb) |
910 | { |
911 | this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); |
912 | } |
913 | |
914 | static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) |
915 | { |
916 | this_cpu_inc(sch->cpu_qstats->qlen); |
917 | } |
918 | |
919 | static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) |
920 | { |
921 | this_cpu_dec(sch->cpu_qstats->qlen); |
922 | } |
923 | |
924 | static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) |
925 | { |
926 | this_cpu_inc(sch->cpu_qstats->requeues); |
927 | } |
928 | |
929 | static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) |
930 | { |
931 | sch->qstats.drops += count; |
932 | } |
933 | |
934 | static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) |
935 | { |
936 | qstats->drops++; |
937 | } |
938 | |
939 | static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) |
940 | { |
941 | qstats->overlimits++; |
942 | } |
943 | |
944 | static inline void qdisc_qstats_drop(struct Qdisc *sch) |
945 | { |
946 | qstats_drop_inc(qstats: &sch->qstats); |
947 | } |
948 | |
949 | static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) |
950 | { |
951 | this_cpu_inc(sch->cpu_qstats->drops); |
952 | } |
953 | |
954 | static inline void qdisc_qstats_overlimit(struct Qdisc *sch) |
955 | { |
956 | sch->qstats.overlimits++; |
957 | } |
958 | |
959 | static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) |
960 | { |
961 | __u32 qlen = qdisc_qlen_sum(q: sch); |
962 | |
963 | return gnet_stats_copy_queue(d, cpu_q: sch->cpu_qstats, q: &sch->qstats, qlen); |
964 | } |
965 | |
966 | static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, |
967 | __u32 *backlog) |
968 | { |
969 | struct gnet_stats_queue qstats = { 0 }; |
970 | |
971 | gnet_stats_add_queue(qstats: &qstats, cpu_q: sch->cpu_qstats, q: &sch->qstats); |
972 | *qlen = qstats.qlen + qdisc_qlen(q: sch); |
973 | *backlog = qstats.backlog; |
974 | } |
975 | |
976 | static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) |
977 | { |
978 | __u32 qlen, backlog; |
979 | |
980 | qdisc_qstats_qlen_backlog(sch, qlen: &qlen, backlog: &backlog); |
981 | qdisc_tree_reduce_backlog(qdisc: sch, n: qlen, len: backlog); |
982 | } |
983 | |
984 | static inline void qdisc_purge_queue(struct Qdisc *sch) |
985 | { |
986 | __u32 qlen, backlog; |
987 | |
988 | qdisc_qstats_qlen_backlog(sch, qlen: &qlen, backlog: &backlog); |
989 | qdisc_reset(qdisc: sch); |
990 | qdisc_tree_reduce_backlog(qdisc: sch, n: qlen, len: backlog); |
991 | } |
992 | |
993 | static inline void __qdisc_enqueue_tail(struct sk_buff *skb, |
994 | struct qdisc_skb_head *qh) |
995 | { |
996 | struct sk_buff *last = qh->tail; |
997 | |
998 | if (last) { |
999 | skb->next = NULL; |
1000 | last->next = skb; |
1001 | qh->tail = skb; |
1002 | } else { |
1003 | qh->tail = skb; |
1004 | qh->head = skb; |
1005 | } |
1006 | qh->qlen++; |
1007 | } |
1008 | |
1009 | static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) |
1010 | { |
1011 | __qdisc_enqueue_tail(skb, qh: &sch->q); |
1012 | qdisc_qstats_backlog_inc(sch, skb); |
1013 | return NET_XMIT_SUCCESS; |
1014 | } |
1015 | |
1016 | static inline void __qdisc_enqueue_head(struct sk_buff *skb, |
1017 | struct qdisc_skb_head *qh) |
1018 | { |
1019 | skb->next = qh->head; |
1020 | |
1021 | if (!qh->head) |
1022 | qh->tail = skb; |
1023 | qh->head = skb; |
1024 | qh->qlen++; |
1025 | } |
1026 | |
1027 | static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) |
1028 | { |
1029 | struct sk_buff *skb = qh->head; |
1030 | |
1031 | if (likely(skb != NULL)) { |
1032 | qh->head = skb->next; |
1033 | qh->qlen--; |
1034 | if (qh->head == NULL) |
1035 | qh->tail = NULL; |
1036 | skb->next = NULL; |
1037 | } |
1038 | |
1039 | return skb; |
1040 | } |
1041 | |
1042 | static inline struct sk_buff *qdisc_dequeue_internal(struct Qdisc *sch, bool direct) |
1043 | { |
1044 | struct sk_buff *skb; |
1045 | |
1046 | skb = __skb_dequeue(list: &sch->gso_skb); |
1047 | if (skb) { |
1048 | sch->q.qlen--; |
1049 | return skb; |
1050 | } |
1051 | if (direct) |
1052 | return __qdisc_dequeue_head(qh: &sch->q); |
1053 | else |
1054 | return sch->dequeue(sch); |
1055 | } |
1056 | |
1057 | static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) |
1058 | { |
1059 | struct sk_buff *skb = __qdisc_dequeue_head(qh: &sch->q); |
1060 | |
1061 | if (likely(skb != NULL)) { |
1062 | qdisc_qstats_backlog_dec(sch, skb); |
1063 | qdisc_bstats_update(sch, skb); |
1064 | } |
1065 | |
1066 | return skb; |
1067 | } |
1068 | |
1069 | struct tc_skb_cb { |
1070 | struct qdisc_skb_cb qdisc_cb; |
1071 | u32 drop_reason; |
1072 | |
1073 | u16 zone; /* Only valid if post_ct = true */ |
1074 | u16 mru; |
1075 | u8 post_ct:1; |
1076 | u8 post_ct_snat:1; |
1077 | u8 post_ct_dnat:1; |
1078 | }; |
1079 | |
1080 | static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb) |
1081 | { |
1082 | struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb; |
1083 | |
1084 | BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); |
1085 | return cb; |
1086 | } |
1087 | |
1088 | static inline enum skb_drop_reason |
1089 | tcf_get_drop_reason(const struct sk_buff *skb) |
1090 | { |
1091 | return tc_skb_cb(skb)->drop_reason; |
1092 | } |
1093 | |
1094 | static inline void tcf_set_drop_reason(const struct sk_buff *skb, |
1095 | enum skb_drop_reason reason) |
1096 | { |
1097 | tc_skb_cb(skb)->drop_reason = reason; |
1098 | } |
1099 | |
1100 | /* Instead of calling kfree_skb() while root qdisc lock is held, |
1101 | * queue the skb for future freeing at end of __dev_xmit_skb() |
1102 | */ |
1103 | static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) |
1104 | { |
1105 | skb->next = *to_free; |
1106 | *to_free = skb; |
1107 | } |
1108 | |
1109 | static inline void __qdisc_drop_all(struct sk_buff *skb, |
1110 | struct sk_buff **to_free) |
1111 | { |
1112 | if (skb->prev) |
1113 | skb->prev->next = *to_free; |
1114 | else |
1115 | skb->next = *to_free; |
1116 | *to_free = skb; |
1117 | } |
1118 | |
1119 | static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, |
1120 | struct qdisc_skb_head *qh, |
1121 | struct sk_buff **to_free) |
1122 | { |
1123 | struct sk_buff *skb = __qdisc_dequeue_head(qh); |
1124 | |
1125 | if (likely(skb != NULL)) { |
1126 | unsigned int len = qdisc_pkt_len(skb); |
1127 | |
1128 | qdisc_qstats_backlog_dec(sch, skb); |
1129 | __qdisc_drop(skb, to_free); |
1130 | return len; |
1131 | } |
1132 | |
1133 | return 0; |
1134 | } |
1135 | |
1136 | static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) |
1137 | { |
1138 | const struct qdisc_skb_head *qh = &sch->q; |
1139 | |
1140 | return qh->head; |
1141 | } |
1142 | |
1143 | /* generic pseudo peek method for non-work-conserving qdisc */ |
1144 | static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) |
1145 | { |
1146 | struct sk_buff *skb = skb_peek(list_: &sch->gso_skb); |
1147 | |
1148 | /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ |
1149 | if (!skb) { |
1150 | skb = sch->dequeue(sch); |
1151 | |
1152 | if (skb) { |
1153 | __skb_queue_head(list: &sch->gso_skb, newsk: skb); |
1154 | /* it's still part of the queue */ |
1155 | qdisc_qstats_backlog_inc(sch, skb); |
1156 | sch->q.qlen++; |
1157 | } |
1158 | } |
1159 | |
1160 | return skb; |
1161 | } |
1162 | |
1163 | static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, |
1164 | struct sk_buff *skb) |
1165 | { |
1166 | if (qdisc_is_percpu_stats(q: sch)) { |
1167 | qdisc_qstats_cpu_backlog_dec(sch, skb); |
1168 | qdisc_bstats_cpu_update(sch, skb); |
1169 | qdisc_qstats_cpu_qlen_dec(sch); |
1170 | } else { |
1171 | qdisc_qstats_backlog_dec(sch, skb); |
1172 | qdisc_bstats_update(sch, skb); |
1173 | sch->q.qlen--; |
1174 | } |
1175 | } |
1176 | |
1177 | static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, |
1178 | unsigned int pkt_len) |
1179 | { |
1180 | if (qdisc_is_percpu_stats(q: sch)) { |
1181 | qdisc_qstats_cpu_qlen_inc(sch); |
1182 | this_cpu_add(sch->cpu_qstats->backlog, pkt_len); |
1183 | } else { |
1184 | sch->qstats.backlog += pkt_len; |
1185 | sch->q.qlen++; |
1186 | } |
1187 | } |
1188 | |
1189 | /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ |
1190 | static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) |
1191 | { |
1192 | struct sk_buff *skb = skb_peek(list_: &sch->gso_skb); |
1193 | |
1194 | if (skb) { |
1195 | skb = __skb_dequeue(list: &sch->gso_skb); |
1196 | if (qdisc_is_percpu_stats(q: sch)) { |
1197 | qdisc_qstats_cpu_backlog_dec(sch, skb); |
1198 | qdisc_qstats_cpu_qlen_dec(sch); |
1199 | } else { |
1200 | qdisc_qstats_backlog_dec(sch, skb); |
1201 | sch->q.qlen--; |
1202 | } |
1203 | } else { |
1204 | skb = sch->dequeue(sch); |
1205 | } |
1206 | |
1207 | return skb; |
1208 | } |
1209 | |
1210 | static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) |
1211 | { |
1212 | /* |
1213 | * We do not know the backlog in bytes of this list, it |
1214 | * is up to the caller to correct it |
1215 | */ |
1216 | ASSERT_RTNL(); |
1217 | if (qh->qlen) { |
1218 | rtnl_kfree_skbs(head: qh->head, tail: qh->tail); |
1219 | |
1220 | qh->head = NULL; |
1221 | qh->tail = NULL; |
1222 | qh->qlen = 0; |
1223 | } |
1224 | } |
1225 | |
1226 | static inline void qdisc_reset_queue(struct Qdisc *sch) |
1227 | { |
1228 | __qdisc_reset_queue(qh: &sch->q); |
1229 | } |
1230 | |
1231 | static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, |
1232 | struct Qdisc **pold) |
1233 | { |
1234 | struct Qdisc *old; |
1235 | |
1236 | sch_tree_lock(q: sch); |
1237 | old = *pold; |
1238 | *pold = new; |
1239 | if (old != NULL) |
1240 | qdisc_purge_queue(sch: old); |
1241 | sch_tree_unlock(q: sch); |
1242 | |
1243 | return old; |
1244 | } |
1245 | |
1246 | static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) |
1247 | { |
1248 | rtnl_kfree_skbs(head: skb, tail: skb); |
1249 | qdisc_qstats_drop(sch); |
1250 | } |
1251 | |
1252 | static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, |
1253 | struct sk_buff **to_free) |
1254 | { |
1255 | __qdisc_drop(skb, to_free); |
1256 | qdisc_qstats_cpu_drop(sch); |
1257 | |
1258 | return NET_XMIT_DROP; |
1259 | } |
1260 | |
1261 | static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, |
1262 | struct sk_buff **to_free) |
1263 | { |
1264 | __qdisc_drop(skb, to_free); |
1265 | qdisc_qstats_drop(sch); |
1266 | |
1267 | return NET_XMIT_DROP; |
1268 | } |
1269 | |
1270 | static inline int qdisc_drop_reason(struct sk_buff *skb, struct Qdisc *sch, |
1271 | struct sk_buff **to_free, |
1272 | enum skb_drop_reason reason) |
1273 | { |
1274 | tcf_set_drop_reason(skb, reason); |
1275 | return qdisc_drop(skb, sch, to_free); |
1276 | } |
1277 | |
1278 | static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, |
1279 | struct sk_buff **to_free) |
1280 | { |
1281 | __qdisc_drop_all(skb, to_free); |
1282 | qdisc_qstats_drop(sch); |
1283 | |
1284 | return NET_XMIT_DROP; |
1285 | } |
1286 | |
1287 | struct psched_ratecfg { |
1288 | u64 rate_bytes_ps; /* bytes per second */ |
1289 | u32 mult; |
1290 | u16 overhead; |
1291 | u16 mpu; |
1292 | u8 linklayer; |
1293 | u8 shift; |
1294 | }; |
1295 | |
1296 | static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, |
1297 | unsigned int len) |
1298 | { |
1299 | len += r->overhead; |
1300 | |
1301 | if (len < r->mpu) |
1302 | len = r->mpu; |
1303 | |
1304 | if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) |
1305 | return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; |
1306 | |
1307 | return ((u64)len * r->mult) >> r->shift; |
1308 | } |
1309 | |
1310 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
1311 | const struct tc_ratespec *conf, |
1312 | u64 rate64); |
1313 | |
1314 | static inline void psched_ratecfg_getrate(struct tc_ratespec *res, |
1315 | const struct psched_ratecfg *r) |
1316 | { |
1317 | memset(res, 0, sizeof(*res)); |
1318 | |
1319 | /* legacy struct tc_ratespec has a 32bit @rate field |
1320 | * Qdisc using 64bit rate should add new attributes |
1321 | * in order to maintain compatibility. |
1322 | */ |
1323 | res->rate = min_t(u64, r->rate_bytes_ps, ~0U); |
1324 | |
1325 | res->overhead = r->overhead; |
1326 | res->mpu = r->mpu; |
1327 | res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); |
1328 | } |
1329 | |
1330 | struct psched_pktrate { |
1331 | u64 rate_pkts_ps; /* packets per second */ |
1332 | u32 mult; |
1333 | u8 shift; |
1334 | }; |
1335 | |
1336 | static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, |
1337 | unsigned int pkt_num) |
1338 | { |
1339 | return ((u64)pkt_num * r->mult) >> r->shift; |
1340 | } |
1341 | |
1342 | void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); |
1343 | |
1344 | /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. |
1345 | * The fast path only needs to access filter list and to update stats |
1346 | */ |
1347 | struct mini_Qdisc { |
1348 | struct tcf_proto *filter_list; |
1349 | struct tcf_block *block; |
1350 | struct gnet_stats_basic_sync __percpu *cpu_bstats; |
1351 | struct gnet_stats_queue __percpu *cpu_qstats; |
1352 | unsigned long rcu_state; |
1353 | }; |
1354 | |
1355 | static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, |
1356 | const struct sk_buff *skb) |
1357 | { |
1358 | bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb); |
1359 | } |
1360 | |
1361 | static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) |
1362 | { |
1363 | this_cpu_inc(miniq->cpu_qstats->drops); |
1364 | } |
1365 | |
1366 | struct mini_Qdisc_pair { |
1367 | struct mini_Qdisc miniq1; |
1368 | struct mini_Qdisc miniq2; |
1369 | struct mini_Qdisc __rcu **p_miniq; |
1370 | }; |
1371 | |
1372 | void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, |
1373 | struct tcf_proto *tp_head); |
1374 | void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, |
1375 | struct mini_Qdisc __rcu **p_miniq); |
1376 | void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, |
1377 | struct tcf_block *block); |
1378 | |
1379 | void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx); |
1380 | |
1381 | int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); |
1382 | |
1383 | /* Make sure qdisc is no longer in SCHED state. */ |
1384 | static inline void qdisc_synchronize(const struct Qdisc *q) |
1385 | { |
1386 | while (test_bit(__QDISC_STATE_SCHED, &q->state)) |
1387 | msleep(msecs: 1); |
1388 | } |
1389 | |
1390 | #endif |
1391 | |