1/* SPDX-License-Identifier: GPL-2.0 */ 2#ifndef __NET_SCHED_GENERIC_H 3#define __NET_SCHED_GENERIC_H 4 5#include <linux/netdevice.h> 6#include <linux/types.h> 7#include <linux/rcupdate.h> 8#include <linux/pkt_sched.h> 9#include <linux/pkt_cls.h> 10#include <linux/percpu.h> 11#include <linux/dynamic_queue_limits.h> 12#include <linux/list.h> 13#include <linux/refcount.h> 14#include <linux/workqueue.h> 15#include <linux/mutex.h> 16#include <linux/rwsem.h> 17#include <linux/atomic.h> 18#include <linux/hashtable.h> 19#include <net/gen_stats.h> 20#include <net/rtnetlink.h> 21#include <net/flow_offload.h> 22 23struct Qdisc_ops; 24struct qdisc_walker; 25struct tcf_walker; 26struct module; 27struct bpf_flow_keys; 28 29struct qdisc_rate_table { 30 struct tc_ratespec rate; 31 u32 data[256]; 32 struct qdisc_rate_table *next; 33 int refcnt; 34}; 35 36enum qdisc_state_t { 37 __QDISC_STATE_SCHED, 38 __QDISC_STATE_DEACTIVATED, 39 __QDISC_STATE_MISSED, 40 __QDISC_STATE_DRAINING, 41}; 42 43enum qdisc_state2_t { 44 /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly. 45 * Use qdisc_run_begin/end() or qdisc_is_running() instead. 46 */ 47 __QDISC_STATE2_RUNNING, 48}; 49 50#define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED) 51#define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING) 52 53#define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \ 54 QDISC_STATE_DRAINING) 55 56struct qdisc_size_table { 57 struct rcu_head rcu; 58 struct list_head list; 59 struct tc_sizespec szopts; 60 int refcnt; 61 u16 data[]; 62}; 63 64/* similar to sk_buff_head, but skb->prev pointer is undefined. */ 65struct qdisc_skb_head { 66 struct sk_buff *head; 67 struct sk_buff *tail; 68 __u32 qlen; 69 spinlock_t lock; 70}; 71 72struct Qdisc { 73 int (*enqueue)(struct sk_buff *skb, 74 struct Qdisc *sch, 75 struct sk_buff **to_free); 76 struct sk_buff * (*dequeue)(struct Qdisc *sch); 77 unsigned int flags; 78#define TCQ_F_BUILTIN 1 79#define TCQ_F_INGRESS 2 80#define TCQ_F_CAN_BYPASS 4 81#define TCQ_F_MQROOT 8 82#define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for 83 * q->dev_queue : It can test 84 * netif_xmit_frozen_or_stopped() before 85 * dequeueing next packet. 86 * Its true for MQ/MQPRIO slaves, or non 87 * multiqueue device. 88 */ 89#define TCQ_F_WARN_NONWC (1 << 16) 90#define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */ 91#define TCQ_F_NOPARENT 0x40 /* root of its hierarchy : 92 * qdisc_tree_decrease_qlen() should stop. 93 */ 94#define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */ 95#define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */ 96#define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */ 97 u32 limit; 98 const struct Qdisc_ops *ops; 99 struct qdisc_size_table __rcu *stab; 100 struct hlist_node hash; 101 u32 handle; 102 u32 parent; 103 104 struct netdev_queue *dev_queue; 105 106 struct net_rate_estimator __rcu *rate_est; 107 struct gnet_stats_basic_sync __percpu *cpu_bstats; 108 struct gnet_stats_queue __percpu *cpu_qstats; 109 int pad; 110 refcount_t refcnt; 111 112 /* 113 * For performance sake on SMP, we put highly modified fields at the end 114 */ 115 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp; 116 struct qdisc_skb_head q; 117 struct gnet_stats_basic_sync bstats; 118 struct gnet_stats_queue qstats; 119 unsigned long state; 120 unsigned long state2; /* must be written under qdisc spinlock */ 121 struct Qdisc *next_sched; 122 struct sk_buff_head skb_bad_txq; 123 124 spinlock_t busylock ____cacheline_aligned_in_smp; 125 spinlock_t seqlock; 126 127 struct rcu_head rcu; 128 netdevice_tracker dev_tracker; 129 /* private data */ 130 long privdata[] ____cacheline_aligned; 131}; 132 133static inline void qdisc_refcount_inc(struct Qdisc *qdisc) 134{ 135 if (qdisc->flags & TCQ_F_BUILTIN) 136 return; 137 refcount_inc(&qdisc->refcnt); 138} 139 140static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc) 141{ 142 if (qdisc->flags & TCQ_F_BUILTIN) 143 return true; 144 return refcount_dec_if_one(&qdisc->refcnt); 145} 146 147/* Intended to be used by unlocked users, when concurrent qdisc release is 148 * possible. 149 */ 150 151static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc) 152{ 153 if (qdisc->flags & TCQ_F_BUILTIN) 154 return qdisc; 155 if (refcount_inc_not_zero(&qdisc->refcnt)) 156 return qdisc; 157 return NULL; 158} 159 160/* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc 161 * root_lock section, or provide their own memory barriers -- ordering 162 * against qdisc_run_begin/end() atomic bit operations. 163 */ 164static inline bool qdisc_is_running(struct Qdisc *qdisc) 165{ 166 if (qdisc->flags & TCQ_F_NOLOCK) 167 return spin_is_locked(&qdisc->seqlock); 168 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 169} 170 171static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc) 172{ 173 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY); 174} 175 176static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) 177{ 178 return q->flags & TCQ_F_CPUSTATS; 179} 180 181static inline bool qdisc_is_empty(const struct Qdisc *qdisc) 182{ 183 if (qdisc_is_percpu_stats(qdisc)) 184 return nolock_qdisc_is_empty(qdisc); 185 return !READ_ONCE(qdisc->q.qlen); 186} 187 188/* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with 189 * the qdisc root lock acquired. 190 */ 191static inline bool qdisc_run_begin(struct Qdisc *qdisc) 192{ 193 if (qdisc->flags & TCQ_F_NOLOCK) { 194 if (spin_trylock(&qdisc->seqlock)) 195 return true; 196 197 /* No need to insist if the MISSED flag was already set. 198 * Note that test_and_set_bit() also gives us memory ordering 199 * guarantees wrt potential earlier enqueue() and below 200 * spin_trylock(), both of which are necessary to prevent races 201 */ 202 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state)) 203 return false; 204 205 /* Try to take the lock again to make sure that we will either 206 * grab it or the CPU that still has it will see MISSED set 207 * when testing it in qdisc_run_end() 208 */ 209 return spin_trylock(&qdisc->seqlock); 210 } 211 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 212} 213 214static inline void qdisc_run_end(struct Qdisc *qdisc) 215{ 216 if (qdisc->flags & TCQ_F_NOLOCK) { 217 spin_unlock(&qdisc->seqlock); 218 219 /* spin_unlock() only has store-release semantic. The unlock 220 * and test_bit() ordering is a store-load ordering, so a full 221 * memory barrier is needed here. 222 */ 223 smp_mb(); 224 225 if (unlikely(test_bit(__QDISC_STATE_MISSED, 226 &qdisc->state))) 227 __netif_schedule(qdisc); 228 } else { 229 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 230 } 231} 232 233static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 234{ 235 return qdisc->flags & TCQ_F_ONETXQUEUE; 236} 237 238static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) 239{ 240#ifdef CONFIG_BQL 241 /* Non-BQL migrated drivers will return 0, too. */ 242 return dql_avail(&txq->dql); 243#else 244 return 0; 245#endif 246} 247 248struct Qdisc_class_ops { 249 unsigned int flags; 250 /* Child qdisc manipulation */ 251 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 252 int (*graft)(struct Qdisc *, unsigned long cl, 253 struct Qdisc *, struct Qdisc **, 254 struct netlink_ext_ack *extack); 255 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 256 void (*qlen_notify)(struct Qdisc *, unsigned long); 257 258 /* Class manipulation routines */ 259 unsigned long (*find)(struct Qdisc *, u32 classid); 260 int (*change)(struct Qdisc *, u32, u32, 261 struct nlattr **, unsigned long *, 262 struct netlink_ext_ack *); 263 int (*delete)(struct Qdisc *, unsigned long, 264 struct netlink_ext_ack *); 265 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 266 267 /* Filter manipulation */ 268 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 269 unsigned long arg, 270 struct netlink_ext_ack *extack); 271 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 272 u32 classid); 273 void (*unbind_tcf)(struct Qdisc *, unsigned long); 274 275 /* rtnetlink specific */ 276 int (*dump)(struct Qdisc *, unsigned long, 277 struct sk_buff *skb, struct tcmsg*); 278 int (*dump_stats)(struct Qdisc *, unsigned long, 279 struct gnet_dump *); 280}; 281 282/* Qdisc_class_ops flag values */ 283 284/* Implements API that doesn't require rtnl lock */ 285enum qdisc_class_ops_flags { 286 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 287}; 288 289struct Qdisc_ops { 290 struct Qdisc_ops *next; 291 const struct Qdisc_class_ops *cl_ops; 292 char id[IFNAMSIZ]; 293 int priv_size; 294 unsigned int static_flags; 295 296 int (*enqueue)(struct sk_buff *skb, 297 struct Qdisc *sch, 298 struct sk_buff **to_free); 299 struct sk_buff * (*dequeue)(struct Qdisc *); 300 struct sk_buff * (*peek)(struct Qdisc *); 301 302 int (*init)(struct Qdisc *sch, struct nlattr *arg, 303 struct netlink_ext_ack *extack); 304 void (*reset)(struct Qdisc *); 305 void (*destroy)(struct Qdisc *); 306 int (*change)(struct Qdisc *sch, 307 struct nlattr *arg, 308 struct netlink_ext_ack *extack); 309 void (*attach)(struct Qdisc *sch); 310 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 311 void (*change_real_num_tx)(struct Qdisc *sch, 312 unsigned int new_real_tx); 313 314 int (*dump)(struct Qdisc *, struct sk_buff *); 315 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 316 317 void (*ingress_block_set)(struct Qdisc *sch, 318 u32 block_index); 319 void (*egress_block_set)(struct Qdisc *sch, 320 u32 block_index); 321 u32 (*ingress_block_get)(struct Qdisc *sch); 322 u32 (*egress_block_get)(struct Qdisc *sch); 323 324 struct module *owner; 325}; 326 327 328struct tcf_result { 329 union { 330 struct { 331 unsigned long class; 332 u32 classid; 333 }; 334 const struct tcf_proto *goto_tp; 335 336 }; 337}; 338 339struct tcf_chain; 340 341struct tcf_proto_ops { 342 struct list_head head; 343 char kind[IFNAMSIZ]; 344 345 int (*classify)(struct sk_buff *, 346 const struct tcf_proto *, 347 struct tcf_result *); 348 int (*init)(struct tcf_proto*); 349 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 350 struct netlink_ext_ack *extack); 351 352 void* (*get)(struct tcf_proto*, u32 handle); 353 void (*put)(struct tcf_proto *tp, void *f); 354 int (*change)(struct net *net, struct sk_buff *, 355 struct tcf_proto*, unsigned long, 356 u32 handle, struct nlattr **, 357 void **, u32, 358 struct netlink_ext_ack *); 359 int (*delete)(struct tcf_proto *tp, void *arg, 360 bool *last, bool rtnl_held, 361 struct netlink_ext_ack *); 362 bool (*delete_empty)(struct tcf_proto *tp); 363 void (*walk)(struct tcf_proto *tp, 364 struct tcf_walker *arg, bool rtnl_held); 365 int (*reoffload)(struct tcf_proto *tp, bool add, 366 flow_setup_cb_t *cb, void *cb_priv, 367 struct netlink_ext_ack *extack); 368 void (*hw_add)(struct tcf_proto *tp, 369 void *type_data); 370 void (*hw_del)(struct tcf_proto *tp, 371 void *type_data); 372 void (*bind_class)(void *, u32, unsigned long, 373 void *, unsigned long); 374 void * (*tmplt_create)(struct net *net, 375 struct tcf_chain *chain, 376 struct nlattr **tca, 377 struct netlink_ext_ack *extack); 378 void (*tmplt_destroy)(void *tmplt_priv); 379 void (*tmplt_reoffload)(struct tcf_chain *chain, 380 bool add, 381 flow_setup_cb_t *cb, 382 void *cb_priv); 383 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp, 384 u32 handle); 385 386 /* rtnetlink specific */ 387 int (*dump)(struct net*, struct tcf_proto*, void *, 388 struct sk_buff *skb, struct tcmsg*, 389 bool); 390 int (*terse_dump)(struct net *net, 391 struct tcf_proto *tp, void *fh, 392 struct sk_buff *skb, 393 struct tcmsg *t, bool rtnl_held); 394 int (*tmplt_dump)(struct sk_buff *skb, 395 struct net *net, 396 void *tmplt_priv); 397 398 struct module *owner; 399 int flags; 400}; 401 402/* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 403 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 404 * conditions can occur when filters are inserted/deleted simultaneously. 405 */ 406enum tcf_proto_ops_flags { 407 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 408}; 409 410struct tcf_proto { 411 /* Fast access part */ 412 struct tcf_proto __rcu *next; 413 void __rcu *root; 414 415 /* called under RCU BH lock*/ 416 int (*classify)(struct sk_buff *, 417 const struct tcf_proto *, 418 struct tcf_result *); 419 __be16 protocol; 420 421 /* All the rest */ 422 u32 prio; 423 void *data; 424 const struct tcf_proto_ops *ops; 425 struct tcf_chain *chain; 426 /* Lock protects tcf_proto shared state and can be used by unlocked 427 * classifiers to protect their private data. 428 */ 429 spinlock_t lock; 430 bool deleting; 431 refcount_t refcnt; 432 struct rcu_head rcu; 433 struct hlist_node destroy_ht_node; 434}; 435 436struct qdisc_skb_cb { 437 struct { 438 unsigned int pkt_len; 439 u16 slave_dev_queue_mapping; 440 u16 tc_classid; 441 }; 442#define QDISC_CB_PRIV_LEN 20 443 unsigned char data[QDISC_CB_PRIV_LEN]; 444}; 445 446typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 447 448struct tcf_chain { 449 /* Protects filter_chain. */ 450 struct mutex filter_chain_lock; 451 struct tcf_proto __rcu *filter_chain; 452 struct list_head list; 453 struct tcf_block *block; 454 u32 index; /* chain index */ 455 unsigned int refcnt; 456 unsigned int action_refcnt; 457 bool explicitly_created; 458 bool flushing; 459 const struct tcf_proto_ops *tmplt_ops; 460 void *tmplt_priv; 461 struct rcu_head rcu; 462}; 463 464struct tcf_block { 465 /* Lock protects tcf_block and lifetime-management data of chains 466 * attached to the block (refcnt, action_refcnt, explicitly_created). 467 */ 468 struct mutex lock; 469 struct list_head chain_list; 470 u32 index; /* block index for shared blocks */ 471 u32 classid; /* which class this block belongs to */ 472 refcount_t refcnt; 473 struct net *net; 474 struct Qdisc *q; 475 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 476 struct flow_block flow_block; 477 struct list_head owner_list; 478 bool keep_dst; 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 491static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 492{ 493 return lockdep_is_held(&chain->filter_chain_lock); 494} 495 496static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 497{ 498 return lockdep_is_held(&tp->lock); 499} 500 501#define tcf_chain_dereference(p, chain) \ 502 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 503 504#define tcf_proto_dereference(p, tp) \ 505 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 506 507static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 508{ 509 struct qdisc_skb_cb *qcb; 510 511 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 512 BUILD_BUG_ON(sizeof(qcb->data) < sz); 513} 514 515static inline int qdisc_qlen(const struct Qdisc *q) 516{ 517 return q->q.qlen; 518} 519 520static inline int qdisc_qlen_sum(const struct Qdisc *q) 521{ 522 __u32 qlen = q->qstats.qlen; 523 int i; 524 525 if (qdisc_is_percpu_stats(q)) { 526 for_each_possible_cpu(i) 527 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 528 } else { 529 qlen += q->q.qlen; 530 } 531 532 return qlen; 533} 534 535static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 536{ 537 return (struct qdisc_skb_cb *)skb->cb; 538} 539 540static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 541{ 542 return &qdisc->q.lock; 543} 544 545static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 546{ 547 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 548 549 return q; 550} 551 552static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 553{ 554 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 555} 556 557static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 558{ 559 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping); 560} 561 562static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 563{ 564 struct Qdisc *root = qdisc_root_sleeping(qdisc); 565 566 ASSERT_RTNL(); 567 return qdisc_lock(root); 568} 569 570static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 571{ 572 return qdisc->dev_queue->dev; 573} 574 575static inline void sch_tree_lock(struct Qdisc *q) 576{ 577 if (q->flags & TCQ_F_MQROOT) 578 spin_lock_bh(qdisc_lock(q)); 579 else 580 spin_lock_bh(qdisc_root_sleeping_lock(q)); 581} 582 583static inline void sch_tree_unlock(struct Qdisc *q) 584{ 585 if (q->flags & TCQ_F_MQROOT) 586 spin_unlock_bh(qdisc_lock(q)); 587 else 588 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 589} 590 591extern struct Qdisc noop_qdisc; 592extern struct Qdisc_ops noop_qdisc_ops; 593extern struct Qdisc_ops pfifo_fast_ops; 594extern struct Qdisc_ops mq_qdisc_ops; 595extern struct Qdisc_ops noqueue_qdisc_ops; 596extern const struct Qdisc_ops *default_qdisc_ops; 597static inline const struct Qdisc_ops * 598get_default_qdisc_ops(const struct net_device *dev, int ntx) 599{ 600 return ntx < dev->real_num_tx_queues ? 601 default_qdisc_ops : &pfifo_fast_ops; 602} 603 604struct Qdisc_class_common { 605 u32 classid; 606 unsigned int filter_cnt; 607 struct hlist_node hnode; 608}; 609 610struct Qdisc_class_hash { 611 struct hlist_head *hash; 612 unsigned int hashsize; 613 unsigned int hashmask; 614 unsigned int hashelems; 615}; 616 617static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 618{ 619 id ^= id >> 8; 620 id ^= id >> 4; 621 return id & mask; 622} 623 624static inline struct Qdisc_class_common * 625qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 626{ 627 struct Qdisc_class_common *cl; 628 unsigned int h; 629 630 if (!id) 631 return NULL; 632 633 h = qdisc_class_hash(id, hash->hashmask); 634 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 635 if (cl->classid == id) 636 return cl; 637 } 638 return NULL; 639} 640 641static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl) 642{ 643 return cl->filter_cnt > 0; 644} 645 646static inline void qdisc_class_get(struct Qdisc_class_common *cl) 647{ 648 unsigned int res; 649 650 if (check_add_overflow(cl->filter_cnt, 1, &res)) 651 WARN(1, "Qdisc class overflow"); 652 653 cl->filter_cnt = res; 654} 655 656static inline void qdisc_class_put(struct Qdisc_class_common *cl) 657{ 658 unsigned int res; 659 660 if (check_sub_overflow(cl->filter_cnt, 1, &res)) 661 WARN(1, "Qdisc class underflow"); 662 663 cl->filter_cnt = res; 664} 665 666static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 667{ 668 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 669 670 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 671} 672 673int qdisc_class_hash_init(struct Qdisc_class_hash *); 674void qdisc_class_hash_insert(struct Qdisc_class_hash *, 675 struct Qdisc_class_common *); 676void qdisc_class_hash_remove(struct Qdisc_class_hash *, 677 struct Qdisc_class_common *); 678void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 679void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 680 681int dev_qdisc_change_tx_queue_len(struct net_device *dev); 682void dev_qdisc_change_real_num_tx(struct net_device *dev, 683 unsigned int new_real_tx); 684void dev_init_scheduler(struct net_device *dev); 685void dev_shutdown(struct net_device *dev); 686void dev_activate(struct net_device *dev); 687void dev_deactivate(struct net_device *dev); 688void dev_deactivate_many(struct list_head *head); 689struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 690 struct Qdisc *qdisc); 691void qdisc_reset(struct Qdisc *qdisc); 692void qdisc_destroy(struct Qdisc *qdisc); 693void qdisc_put(struct Qdisc *qdisc); 694void qdisc_put_unlocked(struct Qdisc *qdisc); 695void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 696#ifdef CONFIG_NET_SCHED 697int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 698 void *type_data); 699void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 700 struct Qdisc *new, struct Qdisc *old, 701 enum tc_setup_type type, void *type_data, 702 struct netlink_ext_ack *extack); 703#else 704static inline int 705qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 706 void *type_data) 707{ 708 q->flags &= ~TCQ_F_OFFLOADED; 709 return 0; 710} 711 712static inline void 713qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 714 struct Qdisc *new, struct Qdisc *old, 715 enum tc_setup_type type, void *type_data, 716 struct netlink_ext_ack *extack) 717{ 718} 719#endif 720void qdisc_offload_query_caps(struct net_device *dev, 721 enum tc_setup_type type, 722 void *caps, size_t caps_len); 723struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 724 const struct Qdisc_ops *ops, 725 struct netlink_ext_ack *extack); 726void qdisc_free(struct Qdisc *qdisc); 727struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 728 const struct Qdisc_ops *ops, u32 parentid, 729 struct netlink_ext_ack *extack); 730void __qdisc_calculate_pkt_len(struct sk_buff *skb, 731 const struct qdisc_size_table *stab); 732int skb_do_redirect(struct sk_buff *); 733 734static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 735{ 736#ifdef CONFIG_NET_XGRESS 737 return skb->tc_at_ingress; 738#else 739 return false; 740#endif 741} 742 743static inline bool skb_skip_tc_classify(struct sk_buff *skb) 744{ 745#ifdef CONFIG_NET_CLS_ACT 746 if (skb->tc_skip_classify) { 747 skb->tc_skip_classify = 0; 748 return true; 749 } 750#endif 751 return false; 752} 753 754/* Reset all TX qdiscs greater than index of a device. */ 755static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 756{ 757 struct Qdisc *qdisc; 758 759 for (; i < dev->num_tx_queues; i++) { 760 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 761 if (qdisc) { 762 spin_lock_bh(qdisc_lock(qdisc)); 763 qdisc_reset(qdisc); 764 spin_unlock_bh(qdisc_lock(qdisc)); 765 } 766 } 767} 768 769/* Are all TX queues of the device empty? */ 770static inline bool qdisc_all_tx_empty(const struct net_device *dev) 771{ 772 unsigned int i; 773 774 rcu_read_lock(); 775 for (i = 0; i < dev->num_tx_queues; i++) { 776 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 777 const struct Qdisc *q = rcu_dereference(txq->qdisc); 778 779 if (!qdisc_is_empty(q)) { 780 rcu_read_unlock(); 781 return false; 782 } 783 } 784 rcu_read_unlock(); 785 return true; 786} 787 788/* Are any of the TX qdiscs changing? */ 789static inline bool qdisc_tx_changing(const struct net_device *dev) 790{ 791 unsigned int i; 792 793 for (i = 0; i < dev->num_tx_queues; i++) { 794 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 795 796 if (rcu_access_pointer(txq->qdisc) != 797 rcu_access_pointer(txq->qdisc_sleeping)) 798 return true; 799 } 800 return false; 801} 802 803/* Is the device using the noop qdisc on all queues? */ 804static inline bool qdisc_tx_is_noop(const struct net_device *dev) 805{ 806 unsigned int i; 807 808 for (i = 0; i < dev->num_tx_queues; i++) { 809 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 810 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 811 return false; 812 } 813 return true; 814} 815 816static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 817{ 818 return qdisc_skb_cb(skb)->pkt_len; 819} 820 821/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 822enum net_xmit_qdisc_t { 823 __NET_XMIT_STOLEN = 0x00010000, 824 __NET_XMIT_BYPASS = 0x00020000, 825}; 826 827#ifdef CONFIG_NET_CLS_ACT 828#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 829#else 830#define net_xmit_drop_count(e) (1) 831#endif 832 833static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 834 const struct Qdisc *sch) 835{ 836#ifdef CONFIG_NET_SCHED 837 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 838 839 if (stab) 840 __qdisc_calculate_pkt_len(skb, stab); 841#endif 842} 843 844static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 845 struct sk_buff **to_free) 846{ 847 qdisc_calculate_pkt_len(skb, sch); 848 return sch->enqueue(skb, sch, to_free); 849} 850 851static inline void _bstats_update(struct gnet_stats_basic_sync *bstats, 852 __u64 bytes, __u32 packets) 853{ 854 u64_stats_update_begin(&bstats->syncp); 855 u64_stats_add(&bstats->bytes, bytes); 856 u64_stats_add(&bstats->packets, packets); 857 u64_stats_update_end(&bstats->syncp); 858} 859 860static inline void bstats_update(struct gnet_stats_basic_sync *bstats, 861 const struct sk_buff *skb) 862{ 863 _bstats_update(bstats, 864 qdisc_pkt_len(skb), 865 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 866} 867 868static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 869 const struct sk_buff *skb) 870{ 871 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb); 872} 873 874static inline void qdisc_bstats_update(struct Qdisc *sch, 875 const struct sk_buff *skb) 876{ 877 bstats_update(&sch->bstats, skb); 878} 879 880static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 881 const struct sk_buff *skb) 882{ 883 sch->qstats.backlog -= qdisc_pkt_len(skb); 884} 885 886static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 887 const struct sk_buff *skb) 888{ 889 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 890} 891 892static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 893 const struct sk_buff *skb) 894{ 895 sch->qstats.backlog += qdisc_pkt_len(skb); 896} 897 898static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 899 const struct sk_buff *skb) 900{ 901 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 902} 903 904static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 905{ 906 this_cpu_inc(sch->cpu_qstats->qlen); 907} 908 909static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 910{ 911 this_cpu_dec(sch->cpu_qstats->qlen); 912} 913 914static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 915{ 916 this_cpu_inc(sch->cpu_qstats->requeues); 917} 918 919static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 920{ 921 sch->qstats.drops += count; 922} 923 924static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 925{ 926 qstats->drops++; 927} 928 929static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 930{ 931 qstats->overlimits++; 932} 933 934static inline void qdisc_qstats_drop(struct Qdisc *sch) 935{ 936 qstats_drop_inc(&sch->qstats); 937} 938 939static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 940{ 941 this_cpu_inc(sch->cpu_qstats->drops); 942} 943 944static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 945{ 946 sch->qstats.overlimits++; 947} 948 949static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 950{ 951 __u32 qlen = qdisc_qlen_sum(sch); 952 953 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 954} 955 956static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 957 __u32 *backlog) 958{ 959 struct gnet_stats_queue qstats = { 0 }; 960 961 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats); 962 *qlen = qstats.qlen + qdisc_qlen(sch); 963 *backlog = qstats.backlog; 964} 965 966static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 967{ 968 __u32 qlen, backlog; 969 970 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 971 qdisc_tree_reduce_backlog(sch, qlen, backlog); 972} 973 974static inline void qdisc_purge_queue(struct Qdisc *sch) 975{ 976 __u32 qlen, backlog; 977 978 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 979 qdisc_reset(sch); 980 qdisc_tree_reduce_backlog(sch, qlen, backlog); 981} 982 983static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 984 struct qdisc_skb_head *qh) 985{ 986 struct sk_buff *last = qh->tail; 987 988 if (last) { 989 skb->next = NULL; 990 last->next = skb; 991 qh->tail = skb; 992 } else { 993 qh->tail = skb; 994 qh->head = skb; 995 } 996 qh->qlen++; 997} 998 999static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 1000{ 1001 __qdisc_enqueue_tail(skb, &sch->q); 1002 qdisc_qstats_backlog_inc(sch, skb); 1003 return NET_XMIT_SUCCESS; 1004} 1005 1006static inline void __qdisc_enqueue_head(struct sk_buff *skb, 1007 struct qdisc_skb_head *qh) 1008{ 1009 skb->next = qh->head; 1010 1011 if (!qh->head) 1012 qh->tail = skb; 1013 qh->head = skb; 1014 qh->qlen++; 1015} 1016 1017static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 1018{ 1019 struct sk_buff *skb = qh->head; 1020 1021 if (likely(skb != NULL)) { 1022 qh->head = skb->next; 1023 qh->qlen--; 1024 if (qh->head == NULL) 1025 qh->tail = NULL; 1026 skb->next = NULL; 1027 } 1028 1029 return skb; 1030} 1031 1032static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1033{ 1034 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1035 1036 if (likely(skb != NULL)) { 1037 qdisc_qstats_backlog_dec(sch, skb); 1038 qdisc_bstats_update(sch, skb); 1039 } 1040 1041 return skb; 1042} 1043 1044/* Instead of calling kfree_skb() while root qdisc lock is held, 1045 * queue the skb for future freeing at end of __dev_xmit_skb() 1046 */ 1047static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1048{ 1049 skb->next = *to_free; 1050 *to_free = skb; 1051} 1052 1053static inline void __qdisc_drop_all(struct sk_buff *skb, 1054 struct sk_buff **to_free) 1055{ 1056 if (skb->prev) 1057 skb->prev->next = *to_free; 1058 else 1059 skb->next = *to_free; 1060 *to_free = skb; 1061} 1062 1063static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1064 struct qdisc_skb_head *qh, 1065 struct sk_buff **to_free) 1066{ 1067 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1068 1069 if (likely(skb != NULL)) { 1070 unsigned int len = qdisc_pkt_len(skb); 1071 1072 qdisc_qstats_backlog_dec(sch, skb); 1073 __qdisc_drop(skb, to_free); 1074 return len; 1075 } 1076 1077 return 0; 1078} 1079 1080static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1081{ 1082 const struct qdisc_skb_head *qh = &sch->q; 1083 1084 return qh->head; 1085} 1086 1087/* generic pseudo peek method for non-work-conserving qdisc */ 1088static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1089{ 1090 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1091 1092 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1093 if (!skb) { 1094 skb = sch->dequeue(sch); 1095 1096 if (skb) { 1097 __skb_queue_head(&sch->gso_skb, skb); 1098 /* it's still part of the queue */ 1099 qdisc_qstats_backlog_inc(sch, skb); 1100 sch->q.qlen++; 1101 } 1102 } 1103 1104 return skb; 1105} 1106 1107static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1108 struct sk_buff *skb) 1109{ 1110 if (qdisc_is_percpu_stats(sch)) { 1111 qdisc_qstats_cpu_backlog_dec(sch, skb); 1112 qdisc_bstats_cpu_update(sch, skb); 1113 qdisc_qstats_cpu_qlen_dec(sch); 1114 } else { 1115 qdisc_qstats_backlog_dec(sch, skb); 1116 qdisc_bstats_update(sch, skb); 1117 sch->q.qlen--; 1118 } 1119} 1120 1121static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1122 unsigned int pkt_len) 1123{ 1124 if (qdisc_is_percpu_stats(sch)) { 1125 qdisc_qstats_cpu_qlen_inc(sch); 1126 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1127 } else { 1128 sch->qstats.backlog += pkt_len; 1129 sch->q.qlen++; 1130 } 1131} 1132 1133/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1134static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1135{ 1136 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1137 1138 if (skb) { 1139 skb = __skb_dequeue(&sch->gso_skb); 1140 if (qdisc_is_percpu_stats(sch)) { 1141 qdisc_qstats_cpu_backlog_dec(sch, skb); 1142 qdisc_qstats_cpu_qlen_dec(sch); 1143 } else { 1144 qdisc_qstats_backlog_dec(sch, skb); 1145 sch->q.qlen--; 1146 } 1147 } else { 1148 skb = sch->dequeue(sch); 1149 } 1150 1151 return skb; 1152} 1153 1154static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1155{ 1156 /* 1157 * We do not know the backlog in bytes of this list, it 1158 * is up to the caller to correct it 1159 */ 1160 ASSERT_RTNL(); 1161 if (qh->qlen) { 1162 rtnl_kfree_skbs(qh->head, qh->tail); 1163 1164 qh->head = NULL; 1165 qh->tail = NULL; 1166 qh->qlen = 0; 1167 } 1168} 1169 1170static inline void qdisc_reset_queue(struct Qdisc *sch) 1171{ 1172 __qdisc_reset_queue(&sch->q); 1173} 1174 1175static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1176 struct Qdisc **pold) 1177{ 1178 struct Qdisc *old; 1179 1180 sch_tree_lock(sch); 1181 old = *pold; 1182 *pold = new; 1183 if (old != NULL) 1184 qdisc_purge_queue(old); 1185 sch_tree_unlock(sch); 1186 1187 return old; 1188} 1189 1190static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1191{ 1192 rtnl_kfree_skbs(skb, skb); 1193 qdisc_qstats_drop(sch); 1194} 1195 1196static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1197 struct sk_buff **to_free) 1198{ 1199 __qdisc_drop(skb, to_free); 1200 qdisc_qstats_cpu_drop(sch); 1201 1202 return NET_XMIT_DROP; 1203} 1204 1205static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1206 struct sk_buff **to_free) 1207{ 1208 __qdisc_drop(skb, to_free); 1209 qdisc_qstats_drop(sch); 1210 1211 return NET_XMIT_DROP; 1212} 1213 1214static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1215 struct sk_buff **to_free) 1216{ 1217 __qdisc_drop_all(skb, to_free); 1218 qdisc_qstats_drop(sch); 1219 1220 return NET_XMIT_DROP; 1221} 1222 1223struct psched_ratecfg { 1224 u64 rate_bytes_ps; /* bytes per second */ 1225 u32 mult; 1226 u16 overhead; 1227 u16 mpu; 1228 u8 linklayer; 1229 u8 shift; 1230}; 1231 1232static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1233 unsigned int len) 1234{ 1235 len += r->overhead; 1236 1237 if (len < r->mpu) 1238 len = r->mpu; 1239 1240 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1241 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1242 1243 return ((u64)len * r->mult) >> r->shift; 1244} 1245 1246void psched_ratecfg_precompute(struct psched_ratecfg *r, 1247 const struct tc_ratespec *conf, 1248 u64 rate64); 1249 1250static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1251 const struct psched_ratecfg *r) 1252{ 1253 memset(res, 0, sizeof(*res)); 1254 1255 /* legacy struct tc_ratespec has a 32bit @rate field 1256 * Qdisc using 64bit rate should add new attributes 1257 * in order to maintain compatibility. 1258 */ 1259 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1260 1261 res->overhead = r->overhead; 1262 res->mpu = r->mpu; 1263 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1264} 1265 1266struct psched_pktrate { 1267 u64 rate_pkts_ps; /* packets per second */ 1268 u32 mult; 1269 u8 shift; 1270}; 1271 1272static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1273 unsigned int pkt_num) 1274{ 1275 return ((u64)pkt_num * r->mult) >> r->shift; 1276} 1277 1278void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1279 1280/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1281 * The fast path only needs to access filter list and to update stats 1282 */ 1283struct mini_Qdisc { 1284 struct tcf_proto *filter_list; 1285 struct tcf_block *block; 1286 struct gnet_stats_basic_sync __percpu *cpu_bstats; 1287 struct gnet_stats_queue __percpu *cpu_qstats; 1288 unsigned long rcu_state; 1289}; 1290 1291static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1292 const struct sk_buff *skb) 1293{ 1294 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1295} 1296 1297static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1298{ 1299 this_cpu_inc(miniq->cpu_qstats->drops); 1300} 1301 1302struct mini_Qdisc_pair { 1303 struct mini_Qdisc miniq1; 1304 struct mini_Qdisc miniq2; 1305 struct mini_Qdisc __rcu **p_miniq; 1306}; 1307 1308void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1309 struct tcf_proto *tp_head); 1310void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1311 struct mini_Qdisc __rcu **p_miniq); 1312void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1313 struct tcf_block *block); 1314 1315void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx); 1316 1317int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1318 1319/* Make sure qdisc is no longer in SCHED state. */ 1320static inline void qdisc_synchronize(const struct Qdisc *q) 1321{ 1322 while (test_bit(__QDISC_STATE_SCHED, &q->state)) 1323 msleep(1); 1324} 1325 1326#endif 1327