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 23 struct Qdisc_ops; 24 struct qdisc_walker; 25 struct tcf_walker; 26 struct module; 27 struct bpf_flow_keys; 28 29 struct qdisc_rate_table { 30 struct tc_ratespec rate; 31 u32 data[256]; 32 struct qdisc_rate_table *next; 33 int refcnt; 34 }; 35 36 enum qdisc_state_t { 37 __QDISC_STATE_SCHED, 38 __QDISC_STATE_DEACTIVATED, 39 __QDISC_STATE_MISSED, 40 __QDISC_STATE_DRAINING, 41 }; 42 43 enum 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 56 struct 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. */ 65 struct qdisc_skb_head { 66 struct sk_buff *head; 67 struct sk_buff *tail; 68 __u32 qlen; 69 spinlock_t lock; 70 }; 71 72 struct 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 }; 132 133 static inline void qdisc_refcount_inc(struct Qdisc *qdisc) 134 { 135 if (qdisc->flags & TCQ_F_BUILTIN) 136 return; 137 refcount_inc(&qdisc->refcnt); 138 } 139 140 static 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 */ 150 151 static 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 */ 164 static 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 } 170 171 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc) 172 { 173 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY); 174 } 175 176 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) 177 { 178 return q->flags & TCQ_F_CPUSTATS; 179 } 180 181 static 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 */ 191 static 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 } 213 214 static 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 } 232 233 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 234 { 235 return qdisc->flags & TCQ_F_ONETXQUEUE; 236 } 237 238 static 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 } 247 248 struct 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 */ 285 enum qdisc_class_ops_flags { 286 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 287 }; 288 289 struct 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 struct tcf_result { 328 union { 329 struct { 330 unsigned long class; 331 u32 classid; 332 }; 333 const struct tcf_proto *goto_tp; 334 }; 335 enum skb_drop_reason drop_reason; 336 }; 337 338 struct tcf_chain; 339 340 struct tcf_proto_ops { 341 struct list_head head; 342 char kind[IFNAMSIZ]; 343 344 int (*classify)(struct sk_buff *, 345 const struct tcf_proto *, 346 struct tcf_result *); 347 int (*init)(struct tcf_proto*); 348 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 349 struct netlink_ext_ack *extack); 350 351 void* (*get)(struct tcf_proto*, u32 handle); 352 void (*put)(struct tcf_proto *tp, void *f); 353 int (*change)(struct net *net, struct sk_buff *, 354 struct tcf_proto*, unsigned long, 355 u32 handle, struct nlattr **, 356 void **, u32, 357 struct netlink_ext_ack *); 358 int (*delete)(struct tcf_proto *tp, void *arg, 359 bool *last, bool rtnl_held, 360 struct netlink_ext_ack *); 361 bool (*delete_empty)(struct tcf_proto *tp); 362 void (*walk)(struct tcf_proto *tp, 363 struct tcf_walker *arg, bool rtnl_held); 364 int (*reoffload)(struct tcf_proto *tp, bool add, 365 flow_setup_cb_t *cb, void *cb_priv, 366 struct netlink_ext_ack *extack); 367 void (*hw_add)(struct tcf_proto *tp, 368 void *type_data); 369 void (*hw_del)(struct tcf_proto *tp, 370 void *type_data); 371 void (*bind_class)(void *, u32, unsigned long, 372 void *, unsigned long); 373 void * (*tmplt_create)(struct net *net, 374 struct tcf_chain *chain, 375 struct nlattr **tca, 376 struct netlink_ext_ack *extack); 377 void (*tmplt_destroy)(void *tmplt_priv); 378 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp, 379 u32 handle); 380 381 /* rtnetlink specific */ 382 int (*dump)(struct net*, struct tcf_proto*, void *, 383 struct sk_buff *skb, struct tcmsg*, 384 bool); 385 int (*terse_dump)(struct net *net, 386 struct tcf_proto *tp, void *fh, 387 struct sk_buff *skb, 388 struct tcmsg *t, bool rtnl_held); 389 int (*tmplt_dump)(struct sk_buff *skb, 390 struct net *net, 391 void *tmplt_priv); 392 393 struct module *owner; 394 int flags; 395 }; 396 397 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 398 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 399 * conditions can occur when filters are inserted/deleted simultaneously. 400 */ 401 enum tcf_proto_ops_flags { 402 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 403 }; 404 405 struct tcf_proto { 406 /* Fast access part */ 407 struct tcf_proto __rcu *next; 408 void __rcu *root; 409 410 /* called under RCU BH lock*/ 411 int (*classify)(struct sk_buff *, 412 const struct tcf_proto *, 413 struct tcf_result *); 414 __be16 protocol; 415 416 /* All the rest */ 417 u32 prio; 418 void *data; 419 const struct tcf_proto_ops *ops; 420 struct tcf_chain *chain; 421 /* Lock protects tcf_proto shared state and can be used by unlocked 422 * classifiers to protect their private data. 423 */ 424 spinlock_t lock; 425 bool deleting; 426 refcount_t refcnt; 427 struct rcu_head rcu; 428 struct hlist_node destroy_ht_node; 429 }; 430 431 struct qdisc_skb_cb { 432 struct { 433 unsigned int pkt_len; 434 u16 slave_dev_queue_mapping; 435 u16 tc_classid; 436 }; 437 #define QDISC_CB_PRIV_LEN 20 438 unsigned char data[QDISC_CB_PRIV_LEN]; 439 }; 440 441 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 442 443 struct tcf_chain { 444 /* Protects filter_chain. */ 445 struct mutex filter_chain_lock; 446 struct tcf_proto __rcu *filter_chain; 447 struct list_head list; 448 struct tcf_block *block; 449 u32 index; /* chain index */ 450 unsigned int refcnt; 451 unsigned int action_refcnt; 452 bool explicitly_created; 453 bool flushing; 454 const struct tcf_proto_ops *tmplt_ops; 455 void *tmplt_priv; 456 struct rcu_head rcu; 457 }; 458 459 struct tcf_block { 460 /* Lock protects tcf_block and lifetime-management data of chains 461 * attached to the block (refcnt, action_refcnt, explicitly_created). 462 */ 463 struct mutex lock; 464 struct list_head chain_list; 465 u32 index; /* block index for shared blocks */ 466 u32 classid; /* which class this block belongs to */ 467 refcount_t refcnt; 468 struct net *net; 469 struct Qdisc *q; 470 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 471 struct flow_block flow_block; 472 struct list_head owner_list; 473 bool keep_dst; 474 atomic_t offloadcnt; /* Number of oddloaded filters */ 475 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 476 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 477 struct { 478 struct tcf_chain *chain; 479 struct list_head filter_chain_list; 480 } chain0; 481 struct rcu_head rcu; 482 DECLARE_HASHTABLE(proto_destroy_ht, 7); 483 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 484 }; 485 486 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 487 { 488 return lockdep_is_held(&chain->filter_chain_lock); 489 } 490 491 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 492 { 493 return lockdep_is_held(&tp->lock); 494 } 495 496 #define tcf_chain_dereference(p, chain) \ 497 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 498 499 #define tcf_proto_dereference(p, tp) \ 500 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 501 502 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 503 { 504 struct qdisc_skb_cb *qcb; 505 506 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 507 BUILD_BUG_ON(sizeof(qcb->data) < sz); 508 } 509 510 static inline int qdisc_qlen(const struct Qdisc *q) 511 { 512 return q->q.qlen; 513 } 514 515 static inline int qdisc_qlen_sum(const struct Qdisc *q) 516 { 517 __u32 qlen = q->qstats.qlen; 518 int i; 519 520 if (qdisc_is_percpu_stats(q)) { 521 for_each_possible_cpu(i) 522 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 523 } else { 524 qlen += q->q.qlen; 525 } 526 527 return qlen; 528 } 529 530 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 531 { 532 return (struct qdisc_skb_cb *)skb->cb; 533 } 534 535 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 536 { 537 return &qdisc->q.lock; 538 } 539 540 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 541 { 542 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 543 544 return q; 545 } 546 547 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 548 { 549 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 550 } 551 552 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 553 { 554 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping); 555 } 556 557 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 558 { 559 struct Qdisc *root = qdisc_root_sleeping(qdisc); 560 561 ASSERT_RTNL(); 562 return qdisc_lock(root); 563 } 564 565 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 566 { 567 return qdisc->dev_queue->dev; 568 } 569 570 static inline void sch_tree_lock(struct Qdisc *q) 571 { 572 if (q->flags & TCQ_F_MQROOT) 573 spin_lock_bh(qdisc_lock(q)); 574 else 575 spin_lock_bh(qdisc_root_sleeping_lock(q)); 576 } 577 578 static inline void sch_tree_unlock(struct Qdisc *q) 579 { 580 if (q->flags & TCQ_F_MQROOT) 581 spin_unlock_bh(qdisc_lock(q)); 582 else 583 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 584 } 585 586 extern struct Qdisc noop_qdisc; 587 extern struct Qdisc_ops noop_qdisc_ops; 588 extern struct Qdisc_ops pfifo_fast_ops; 589 extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1]; 590 extern struct Qdisc_ops mq_qdisc_ops; 591 extern struct Qdisc_ops noqueue_qdisc_ops; 592 extern const struct Qdisc_ops *default_qdisc_ops; 593 static inline const struct Qdisc_ops * 594 get_default_qdisc_ops(const struct net_device *dev, int ntx) 595 { 596 return ntx < dev->real_num_tx_queues ? 597 default_qdisc_ops : &pfifo_fast_ops; 598 } 599 600 struct Qdisc_class_common { 601 u32 classid; 602 unsigned int filter_cnt; 603 struct hlist_node hnode; 604 }; 605 606 struct Qdisc_class_hash { 607 struct hlist_head *hash; 608 unsigned int hashsize; 609 unsigned int hashmask; 610 unsigned int hashelems; 611 }; 612 613 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 614 { 615 id ^= id >> 8; 616 id ^= id >> 4; 617 return id & mask; 618 } 619 620 static inline struct Qdisc_class_common * 621 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 622 { 623 struct Qdisc_class_common *cl; 624 unsigned int h; 625 626 if (!id) 627 return NULL; 628 629 h = qdisc_class_hash(id, hash->hashmask); 630 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 631 if (cl->classid == id) 632 return cl; 633 } 634 return NULL; 635 } 636 637 static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl) 638 { 639 return cl->filter_cnt > 0; 640 } 641 642 static inline void qdisc_class_get(struct Qdisc_class_common *cl) 643 { 644 unsigned int res; 645 646 if (check_add_overflow(cl->filter_cnt, 1, &res)) 647 WARN(1, "Qdisc class overflow"); 648 649 cl->filter_cnt = res; 650 } 651 652 static inline void qdisc_class_put(struct Qdisc_class_common *cl) 653 { 654 unsigned int res; 655 656 if (check_sub_overflow(cl->filter_cnt, 1, &res)) 657 WARN(1, "Qdisc class underflow"); 658 659 cl->filter_cnt = res; 660 } 661 662 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 663 { 664 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 665 666 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 667 } 668 669 int qdisc_class_hash_init(struct Qdisc_class_hash *); 670 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 671 struct Qdisc_class_common *); 672 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 673 struct Qdisc_class_common *); 674 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 675 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 676 677 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 678 void dev_qdisc_change_real_num_tx(struct net_device *dev, 679 unsigned int new_real_tx); 680 void dev_init_scheduler(struct net_device *dev); 681 void dev_shutdown(struct net_device *dev); 682 void dev_activate(struct net_device *dev); 683 void dev_deactivate(struct net_device *dev); 684 void dev_deactivate_many(struct list_head *head); 685 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 686 struct Qdisc *qdisc); 687 void qdisc_reset(struct Qdisc *qdisc); 688 void qdisc_destroy(struct Qdisc *qdisc); 689 void qdisc_put(struct Qdisc *qdisc); 690 void qdisc_put_unlocked(struct Qdisc *qdisc); 691 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 692 #ifdef CONFIG_NET_SCHED 693 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 694 void *type_data); 695 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 696 struct Qdisc *new, struct Qdisc *old, 697 enum tc_setup_type type, void *type_data, 698 struct netlink_ext_ack *extack); 699 #else 700 static inline int 701 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 702 void *type_data) 703 { 704 q->flags &= ~TCQ_F_OFFLOADED; 705 return 0; 706 } 707 708 static inline void 709 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 710 struct Qdisc *new, struct Qdisc *old, 711 enum tc_setup_type type, void *type_data, 712 struct netlink_ext_ack *extack) 713 { 714 } 715 #endif 716 void qdisc_offload_query_caps(struct net_device *dev, 717 enum tc_setup_type type, 718 void *caps, size_t caps_len); 719 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 720 const struct Qdisc_ops *ops, 721 struct netlink_ext_ack *extack); 722 void qdisc_free(struct Qdisc *qdisc); 723 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 724 const struct Qdisc_ops *ops, u32 parentid, 725 struct netlink_ext_ack *extack); 726 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 727 const struct qdisc_size_table *stab); 728 int skb_do_redirect(struct sk_buff *); 729 730 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 731 { 732 #ifdef CONFIG_NET_XGRESS 733 return skb->tc_at_ingress; 734 #else 735 return false; 736 #endif 737 } 738 739 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 740 { 741 #ifdef CONFIG_NET_CLS_ACT 742 if (skb->tc_skip_classify) { 743 skb->tc_skip_classify = 0; 744 return true; 745 } 746 #endif 747 return false; 748 } 749 750 /* Reset all TX qdiscs greater than index of a device. */ 751 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 752 { 753 struct Qdisc *qdisc; 754 755 for (; i < dev->num_tx_queues; i++) { 756 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 757 if (qdisc) { 758 spin_lock_bh(qdisc_lock(qdisc)); 759 qdisc_reset(qdisc); 760 spin_unlock_bh(qdisc_lock(qdisc)); 761 } 762 } 763 } 764 765 /* Are all TX queues of the device empty? */ 766 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 767 { 768 unsigned int i; 769 770 rcu_read_lock(); 771 for (i = 0; i < dev->num_tx_queues; i++) { 772 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 773 const struct Qdisc *q = rcu_dereference(txq->qdisc); 774 775 if (!qdisc_is_empty(q)) { 776 rcu_read_unlock(); 777 return false; 778 } 779 } 780 rcu_read_unlock(); 781 return true; 782 } 783 784 /* Are any of the TX qdiscs changing? */ 785 static inline bool qdisc_tx_changing(const struct net_device *dev) 786 { 787 unsigned int i; 788 789 for (i = 0; i < dev->num_tx_queues; i++) { 790 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 791 792 if (rcu_access_pointer(txq->qdisc) != 793 rcu_access_pointer(txq->qdisc_sleeping)) 794 return true; 795 } 796 return false; 797 } 798 799 /* Is the device using the noop qdisc on all queues? */ 800 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 801 { 802 unsigned int i; 803 804 for (i = 0; i < dev->num_tx_queues; i++) { 805 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 806 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 807 return false; 808 } 809 return true; 810 } 811 812 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 813 { 814 return qdisc_skb_cb(skb)->pkt_len; 815 } 816 817 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 818 enum net_xmit_qdisc_t { 819 __NET_XMIT_STOLEN = 0x00010000, 820 __NET_XMIT_BYPASS = 0x00020000, 821 }; 822 823 #ifdef CONFIG_NET_CLS_ACT 824 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 825 #else 826 #define net_xmit_drop_count(e) (1) 827 #endif 828 829 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 830 const struct Qdisc *sch) 831 { 832 #ifdef CONFIG_NET_SCHED 833 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 834 835 if (stab) 836 __qdisc_calculate_pkt_len(skb, stab); 837 #endif 838 } 839 840 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 841 struct sk_buff **to_free) 842 { 843 qdisc_calculate_pkt_len(skb, sch); 844 return sch->enqueue(skb, sch, to_free); 845 } 846 847 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats, 848 __u64 bytes, __u32 packets) 849 { 850 u64_stats_update_begin(&bstats->syncp); 851 u64_stats_add(&bstats->bytes, bytes); 852 u64_stats_add(&bstats->packets, packets); 853 u64_stats_update_end(&bstats->syncp); 854 } 855 856 static inline void bstats_update(struct gnet_stats_basic_sync *bstats, 857 const struct sk_buff *skb) 858 { 859 _bstats_update(bstats, 860 qdisc_pkt_len(skb), 861 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 862 } 863 864 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 865 const struct sk_buff *skb) 866 { 867 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb); 868 } 869 870 static inline void qdisc_bstats_update(struct Qdisc *sch, 871 const struct sk_buff *skb) 872 { 873 bstats_update(&sch->bstats, skb); 874 } 875 876 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 877 const struct sk_buff *skb) 878 { 879 sch->qstats.backlog -= qdisc_pkt_len(skb); 880 } 881 882 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 883 const struct sk_buff *skb) 884 { 885 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 886 } 887 888 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 889 const struct sk_buff *skb) 890 { 891 sch->qstats.backlog += qdisc_pkt_len(skb); 892 } 893 894 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 895 const struct sk_buff *skb) 896 { 897 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 898 } 899 900 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 901 { 902 this_cpu_inc(sch->cpu_qstats->qlen); 903 } 904 905 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 906 { 907 this_cpu_dec(sch->cpu_qstats->qlen); 908 } 909 910 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 911 { 912 this_cpu_inc(sch->cpu_qstats->requeues); 913 } 914 915 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 916 { 917 sch->qstats.drops += count; 918 } 919 920 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 921 { 922 qstats->drops++; 923 } 924 925 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 926 { 927 qstats->overlimits++; 928 } 929 930 static inline void qdisc_qstats_drop(struct Qdisc *sch) 931 { 932 qstats_drop_inc(&sch->qstats); 933 } 934 935 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 936 { 937 this_cpu_inc(sch->cpu_qstats->drops); 938 } 939 940 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 941 { 942 sch->qstats.overlimits++; 943 } 944 945 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 946 { 947 __u32 qlen = qdisc_qlen_sum(sch); 948 949 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 950 } 951 952 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 953 __u32 *backlog) 954 { 955 struct gnet_stats_queue qstats = { 0 }; 956 957 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats); 958 *qlen = qstats.qlen + qdisc_qlen(sch); 959 *backlog = qstats.backlog; 960 } 961 962 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 963 { 964 __u32 qlen, backlog; 965 966 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 967 qdisc_tree_reduce_backlog(sch, qlen, backlog); 968 } 969 970 static inline void qdisc_purge_queue(struct Qdisc *sch) 971 { 972 __u32 qlen, backlog; 973 974 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 975 qdisc_reset(sch); 976 qdisc_tree_reduce_backlog(sch, qlen, backlog); 977 } 978 979 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 980 struct qdisc_skb_head *qh) 981 { 982 struct sk_buff *last = qh->tail; 983 984 if (last) { 985 skb->next = NULL; 986 last->next = skb; 987 qh->tail = skb; 988 } else { 989 qh->tail = skb; 990 qh->head = skb; 991 } 992 qh->qlen++; 993 } 994 995 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 996 { 997 __qdisc_enqueue_tail(skb, &sch->q); 998 qdisc_qstats_backlog_inc(sch, skb); 999 return NET_XMIT_SUCCESS; 1000 } 1001 1002 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 1003 struct qdisc_skb_head *qh) 1004 { 1005 skb->next = qh->head; 1006 1007 if (!qh->head) 1008 qh->tail = skb; 1009 qh->head = skb; 1010 qh->qlen++; 1011 } 1012 1013 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 1014 { 1015 struct sk_buff *skb = qh->head; 1016 1017 if (likely(skb != NULL)) { 1018 qh->head = skb->next; 1019 qh->qlen--; 1020 if (qh->head == NULL) 1021 qh->tail = NULL; 1022 skb->next = NULL; 1023 } 1024 1025 return skb; 1026 } 1027 1028 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1029 { 1030 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1031 1032 if (likely(skb != NULL)) { 1033 qdisc_qstats_backlog_dec(sch, skb); 1034 qdisc_bstats_update(sch, skb); 1035 } 1036 1037 return skb; 1038 } 1039 1040 /* Instead of calling kfree_skb() while root qdisc lock is held, 1041 * queue the skb for future freeing at end of __dev_xmit_skb() 1042 */ 1043 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1044 { 1045 skb->next = *to_free; 1046 *to_free = skb; 1047 } 1048 1049 static inline void __qdisc_drop_all(struct sk_buff *skb, 1050 struct sk_buff **to_free) 1051 { 1052 if (skb->prev) 1053 skb->prev->next = *to_free; 1054 else 1055 skb->next = *to_free; 1056 *to_free = skb; 1057 } 1058 1059 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1060 struct qdisc_skb_head *qh, 1061 struct sk_buff **to_free) 1062 { 1063 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1064 1065 if (likely(skb != NULL)) { 1066 unsigned int len = qdisc_pkt_len(skb); 1067 1068 qdisc_qstats_backlog_dec(sch, skb); 1069 __qdisc_drop(skb, to_free); 1070 return len; 1071 } 1072 1073 return 0; 1074 } 1075 1076 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1077 { 1078 const struct qdisc_skb_head *qh = &sch->q; 1079 1080 return qh->head; 1081 } 1082 1083 /* generic pseudo peek method for non-work-conserving qdisc */ 1084 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1085 { 1086 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1087 1088 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1089 if (!skb) { 1090 skb = sch->dequeue(sch); 1091 1092 if (skb) { 1093 __skb_queue_head(&sch->gso_skb, skb); 1094 /* it's still part of the queue */ 1095 qdisc_qstats_backlog_inc(sch, skb); 1096 sch->q.qlen++; 1097 } 1098 } 1099 1100 return skb; 1101 } 1102 1103 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1104 struct sk_buff *skb) 1105 { 1106 if (qdisc_is_percpu_stats(sch)) { 1107 qdisc_qstats_cpu_backlog_dec(sch, skb); 1108 qdisc_bstats_cpu_update(sch, skb); 1109 qdisc_qstats_cpu_qlen_dec(sch); 1110 } else { 1111 qdisc_qstats_backlog_dec(sch, skb); 1112 qdisc_bstats_update(sch, skb); 1113 sch->q.qlen--; 1114 } 1115 } 1116 1117 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1118 unsigned int pkt_len) 1119 { 1120 if (qdisc_is_percpu_stats(sch)) { 1121 qdisc_qstats_cpu_qlen_inc(sch); 1122 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1123 } else { 1124 sch->qstats.backlog += pkt_len; 1125 sch->q.qlen++; 1126 } 1127 } 1128 1129 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1130 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1131 { 1132 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1133 1134 if (skb) { 1135 skb = __skb_dequeue(&sch->gso_skb); 1136 if (qdisc_is_percpu_stats(sch)) { 1137 qdisc_qstats_cpu_backlog_dec(sch, skb); 1138 qdisc_qstats_cpu_qlen_dec(sch); 1139 } else { 1140 qdisc_qstats_backlog_dec(sch, skb); 1141 sch->q.qlen--; 1142 } 1143 } else { 1144 skb = sch->dequeue(sch); 1145 } 1146 1147 return skb; 1148 } 1149 1150 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1151 { 1152 /* 1153 * We do not know the backlog in bytes of this list, it 1154 * is up to the caller to correct it 1155 */ 1156 ASSERT_RTNL(); 1157 if (qh->qlen) { 1158 rtnl_kfree_skbs(qh->head, qh->tail); 1159 1160 qh->head = NULL; 1161 qh->tail = NULL; 1162 qh->qlen = 0; 1163 } 1164 } 1165 1166 static inline void qdisc_reset_queue(struct Qdisc *sch) 1167 { 1168 __qdisc_reset_queue(&sch->q); 1169 } 1170 1171 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1172 struct Qdisc **pold) 1173 { 1174 struct Qdisc *old; 1175 1176 sch_tree_lock(sch); 1177 old = *pold; 1178 *pold = new; 1179 if (old != NULL) 1180 qdisc_purge_queue(old); 1181 sch_tree_unlock(sch); 1182 1183 return old; 1184 } 1185 1186 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1187 { 1188 rtnl_kfree_skbs(skb, skb); 1189 qdisc_qstats_drop(sch); 1190 } 1191 1192 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1193 struct sk_buff **to_free) 1194 { 1195 __qdisc_drop(skb, to_free); 1196 qdisc_qstats_cpu_drop(sch); 1197 1198 return NET_XMIT_DROP; 1199 } 1200 1201 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1202 struct sk_buff **to_free) 1203 { 1204 __qdisc_drop(skb, to_free); 1205 qdisc_qstats_drop(sch); 1206 1207 return NET_XMIT_DROP; 1208 } 1209 1210 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1211 struct sk_buff **to_free) 1212 { 1213 __qdisc_drop_all(skb, to_free); 1214 qdisc_qstats_drop(sch); 1215 1216 return NET_XMIT_DROP; 1217 } 1218 1219 struct psched_ratecfg { 1220 u64 rate_bytes_ps; /* bytes per second */ 1221 u32 mult; 1222 u16 overhead; 1223 u16 mpu; 1224 u8 linklayer; 1225 u8 shift; 1226 }; 1227 1228 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1229 unsigned int len) 1230 { 1231 len += r->overhead; 1232 1233 if (len < r->mpu) 1234 len = r->mpu; 1235 1236 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1237 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1238 1239 return ((u64)len * r->mult) >> r->shift; 1240 } 1241 1242 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1243 const struct tc_ratespec *conf, 1244 u64 rate64); 1245 1246 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1247 const struct psched_ratecfg *r) 1248 { 1249 memset(res, 0, sizeof(*res)); 1250 1251 /* legacy struct tc_ratespec has a 32bit @rate field 1252 * Qdisc using 64bit rate should add new attributes 1253 * in order to maintain compatibility. 1254 */ 1255 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1256 1257 res->overhead = r->overhead; 1258 res->mpu = r->mpu; 1259 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1260 } 1261 1262 struct psched_pktrate { 1263 u64 rate_pkts_ps; /* packets per second */ 1264 u32 mult; 1265 u8 shift; 1266 }; 1267 1268 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1269 unsigned int pkt_num) 1270 { 1271 return ((u64)pkt_num * r->mult) >> r->shift; 1272 } 1273 1274 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1275 1276 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1277 * The fast path only needs to access filter list and to update stats 1278 */ 1279 struct mini_Qdisc { 1280 struct tcf_proto *filter_list; 1281 struct tcf_block *block; 1282 struct gnet_stats_basic_sync __percpu *cpu_bstats; 1283 struct gnet_stats_queue __percpu *cpu_qstats; 1284 unsigned long rcu_state; 1285 }; 1286 1287 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1288 const struct sk_buff *skb) 1289 { 1290 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1291 } 1292 1293 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1294 { 1295 this_cpu_inc(miniq->cpu_qstats->drops); 1296 } 1297 1298 struct mini_Qdisc_pair { 1299 struct mini_Qdisc miniq1; 1300 struct mini_Qdisc miniq2; 1301 struct mini_Qdisc __rcu **p_miniq; 1302 }; 1303 1304 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1305 struct tcf_proto *tp_head); 1306 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1307 struct mini_Qdisc __rcu **p_miniq); 1308 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1309 struct tcf_block *block); 1310 1311 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx); 1312 1313 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1314 1315 /* Make sure qdisc is no longer in SCHED state. */ 1316 static inline void qdisc_synchronize(const struct Qdisc *q) 1317 { 1318 while (test_bit(__QDISC_STATE_SCHED, &q->state)) 1319 msleep(1); 1320 } 1321 1322 #endif 1323