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 }; 326 327 328 struct tcf_result { 329 union { 330 struct { 331 unsigned long class; 332 u32 classid; 333 }; 334 const struct tcf_proto *goto_tp; 335 336 }; 337 }; 338 339 struct tcf_chain; 340 341 struct 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 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 refcount_t refcnt; 428 struct rcu_head rcu; 429 struct hlist_node destroy_ht_node; 430 }; 431 432 struct qdisc_skb_cb { 433 struct { 434 unsigned int pkt_len; 435 u16 slave_dev_queue_mapping; 436 u16 tc_classid; 437 }; 438 #define QDISC_CB_PRIV_LEN 20 439 unsigned char data[QDISC_CB_PRIV_LEN]; 440 }; 441 442 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 443 444 struct tcf_chain { 445 /* Protects filter_chain. */ 446 struct mutex filter_chain_lock; 447 struct tcf_proto __rcu *filter_chain; 448 struct list_head list; 449 struct tcf_block *block; 450 u32 index; /* chain index */ 451 unsigned int refcnt; 452 unsigned int action_refcnt; 453 bool explicitly_created; 454 bool flushing; 455 const struct tcf_proto_ops *tmplt_ops; 456 void *tmplt_priv; 457 struct rcu_head rcu; 458 }; 459 460 struct tcf_block { 461 /* Lock protects tcf_block and lifetime-management data of chains 462 * attached to the block (refcnt, action_refcnt, explicitly_created). 463 */ 464 struct mutex lock; 465 struct list_head chain_list; 466 u32 index; /* block index for shared blocks */ 467 u32 classid; /* which class this block belongs to */ 468 refcount_t refcnt; 469 struct net *net; 470 struct Qdisc *q; 471 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 472 struct flow_block flow_block; 473 struct list_head owner_list; 474 bool keep_dst; 475 atomic_t offloadcnt; /* Number of oddloaded filters */ 476 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 477 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 478 struct { 479 struct tcf_chain *chain; 480 struct list_head filter_chain_list; 481 } chain0; 482 struct rcu_head rcu; 483 DECLARE_HASHTABLE(proto_destroy_ht, 7); 484 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 485 }; 486 487 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 488 { 489 return lockdep_is_held(&chain->filter_chain_lock); 490 } 491 492 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 493 { 494 return lockdep_is_held(&tp->lock); 495 } 496 497 #define tcf_chain_dereference(p, chain) \ 498 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 499 500 #define tcf_proto_dereference(p, tp) \ 501 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 502 503 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 504 { 505 struct qdisc_skb_cb *qcb; 506 507 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 508 BUILD_BUG_ON(sizeof(qcb->data) < sz); 509 } 510 511 static inline int qdisc_qlen(const struct Qdisc *q) 512 { 513 return q->q.qlen; 514 } 515 516 static inline int qdisc_qlen_sum(const struct Qdisc *q) 517 { 518 __u32 qlen = q->qstats.qlen; 519 int i; 520 521 if (qdisc_is_percpu_stats(q)) { 522 for_each_possible_cpu(i) 523 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 524 } else { 525 qlen += q->q.qlen; 526 } 527 528 return qlen; 529 } 530 531 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 532 { 533 return (struct qdisc_skb_cb *)skb->cb; 534 } 535 536 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 537 { 538 return &qdisc->q.lock; 539 } 540 541 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 542 { 543 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 544 545 return q; 546 } 547 548 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 549 { 550 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 551 } 552 553 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 554 { 555 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping); 556 } 557 558 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 559 { 560 struct Qdisc *root = qdisc_root_sleeping(qdisc); 561 562 ASSERT_RTNL(); 563 return qdisc_lock(root); 564 } 565 566 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 567 { 568 return qdisc->dev_queue->dev; 569 } 570 571 static inline void sch_tree_lock(struct Qdisc *q) 572 { 573 if (q->flags & TCQ_F_MQROOT) 574 spin_lock_bh(qdisc_lock(q)); 575 else 576 spin_lock_bh(qdisc_root_sleeping_lock(q)); 577 } 578 579 static inline void sch_tree_unlock(struct Qdisc *q) 580 { 581 if (q->flags & TCQ_F_MQROOT) 582 spin_unlock_bh(qdisc_lock(q)); 583 else 584 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 585 } 586 587 extern struct Qdisc noop_qdisc; 588 extern struct Qdisc_ops noop_qdisc_ops; 589 extern struct Qdisc_ops pfifo_fast_ops; 590 extern const u8 sch_default_prio2band[TC_PRIO_MAX + 1]; 591 extern struct Qdisc_ops mq_qdisc_ops; 592 extern struct Qdisc_ops noqueue_qdisc_ops; 593 extern const struct Qdisc_ops *default_qdisc_ops; 594 static inline const struct Qdisc_ops * 595 get_default_qdisc_ops(const struct net_device *dev, int ntx) 596 { 597 return ntx < dev->real_num_tx_queues ? 598 default_qdisc_ops : &pfifo_fast_ops; 599 } 600 601 struct Qdisc_class_common { 602 u32 classid; 603 unsigned int filter_cnt; 604 struct hlist_node hnode; 605 }; 606 607 struct Qdisc_class_hash { 608 struct hlist_head *hash; 609 unsigned int hashsize; 610 unsigned int hashmask; 611 unsigned int hashelems; 612 }; 613 614 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 615 { 616 id ^= id >> 8; 617 id ^= id >> 4; 618 return id & mask; 619 } 620 621 static inline struct Qdisc_class_common * 622 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 623 { 624 struct Qdisc_class_common *cl; 625 unsigned int h; 626 627 if (!id) 628 return NULL; 629 630 h = qdisc_class_hash(id, hash->hashmask); 631 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 632 if (cl->classid == id) 633 return cl; 634 } 635 return NULL; 636 } 637 638 static inline bool qdisc_class_in_use(const struct Qdisc_class_common *cl) 639 { 640 return cl->filter_cnt > 0; 641 } 642 643 static inline void qdisc_class_get(struct Qdisc_class_common *cl) 644 { 645 unsigned int res; 646 647 if (check_add_overflow(cl->filter_cnt, 1, &res)) 648 WARN(1, "Qdisc class overflow"); 649 650 cl->filter_cnt = res; 651 } 652 653 static inline void qdisc_class_put(struct Qdisc_class_common *cl) 654 { 655 unsigned int res; 656 657 if (check_sub_overflow(cl->filter_cnt, 1, &res)) 658 WARN(1, "Qdisc class underflow"); 659 660 cl->filter_cnt = res; 661 } 662 663 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 664 { 665 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 666 667 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 668 } 669 670 int qdisc_class_hash_init(struct Qdisc_class_hash *); 671 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 672 struct Qdisc_class_common *); 673 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 674 struct Qdisc_class_common *); 675 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 676 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 677 678 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 679 void dev_qdisc_change_real_num_tx(struct net_device *dev, 680 unsigned int new_real_tx); 681 void dev_init_scheduler(struct net_device *dev); 682 void dev_shutdown(struct net_device *dev); 683 void dev_activate(struct net_device *dev); 684 void dev_deactivate(struct net_device *dev); 685 void dev_deactivate_many(struct list_head *head); 686 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 687 struct Qdisc *qdisc); 688 void qdisc_reset(struct Qdisc *qdisc); 689 void qdisc_destroy(struct Qdisc *qdisc); 690 void qdisc_put(struct Qdisc *qdisc); 691 void qdisc_put_unlocked(struct Qdisc *qdisc); 692 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 693 #ifdef CONFIG_NET_SCHED 694 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 695 void *type_data); 696 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 697 struct Qdisc *new, struct Qdisc *old, 698 enum tc_setup_type type, void *type_data, 699 struct netlink_ext_ack *extack); 700 #else 701 static inline int 702 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 703 void *type_data) 704 { 705 q->flags &= ~TCQ_F_OFFLOADED; 706 return 0; 707 } 708 709 static inline void 710 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 711 struct Qdisc *new, struct Qdisc *old, 712 enum tc_setup_type type, void *type_data, 713 struct netlink_ext_ack *extack) 714 { 715 } 716 #endif 717 void qdisc_offload_query_caps(struct net_device *dev, 718 enum tc_setup_type type, 719 void *caps, size_t caps_len); 720 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 721 const struct Qdisc_ops *ops, 722 struct netlink_ext_ack *extack); 723 void qdisc_free(struct Qdisc *qdisc); 724 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 725 const struct Qdisc_ops *ops, u32 parentid, 726 struct netlink_ext_ack *extack); 727 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 728 const struct qdisc_size_table *stab); 729 int skb_do_redirect(struct sk_buff *); 730 731 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 732 { 733 #ifdef CONFIG_NET_XGRESS 734 return skb->tc_at_ingress; 735 #else 736 return false; 737 #endif 738 } 739 740 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 741 { 742 #ifdef CONFIG_NET_CLS_ACT 743 if (skb->tc_skip_classify) { 744 skb->tc_skip_classify = 0; 745 return true; 746 } 747 #endif 748 return false; 749 } 750 751 /* Reset all TX qdiscs greater than index of a device. */ 752 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 753 { 754 struct Qdisc *qdisc; 755 756 for (; i < dev->num_tx_queues; i++) { 757 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 758 if (qdisc) { 759 spin_lock_bh(qdisc_lock(qdisc)); 760 qdisc_reset(qdisc); 761 spin_unlock_bh(qdisc_lock(qdisc)); 762 } 763 } 764 } 765 766 /* Are all TX queues of the device empty? */ 767 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 768 { 769 unsigned int i; 770 771 rcu_read_lock(); 772 for (i = 0; i < dev->num_tx_queues; i++) { 773 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 774 const struct Qdisc *q = rcu_dereference(txq->qdisc); 775 776 if (!qdisc_is_empty(q)) { 777 rcu_read_unlock(); 778 return false; 779 } 780 } 781 rcu_read_unlock(); 782 return true; 783 } 784 785 /* Are any of the TX qdiscs changing? */ 786 static inline bool qdisc_tx_changing(const struct net_device *dev) 787 { 788 unsigned int i; 789 790 for (i = 0; i < dev->num_tx_queues; i++) { 791 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 792 793 if (rcu_access_pointer(txq->qdisc) != 794 rcu_access_pointer(txq->qdisc_sleeping)) 795 return true; 796 } 797 return false; 798 } 799 800 /* Is the device using the noop qdisc on all queues? */ 801 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 802 { 803 unsigned int i; 804 805 for (i = 0; i < dev->num_tx_queues; i++) { 806 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 807 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 808 return false; 809 } 810 return true; 811 } 812 813 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 814 { 815 return qdisc_skb_cb(skb)->pkt_len; 816 } 817 818 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 819 enum net_xmit_qdisc_t { 820 __NET_XMIT_STOLEN = 0x00010000, 821 __NET_XMIT_BYPASS = 0x00020000, 822 }; 823 824 #ifdef CONFIG_NET_CLS_ACT 825 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 826 #else 827 #define net_xmit_drop_count(e) (1) 828 #endif 829 830 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 831 const struct Qdisc *sch) 832 { 833 #ifdef CONFIG_NET_SCHED 834 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 835 836 if (stab) 837 __qdisc_calculate_pkt_len(skb, stab); 838 #endif 839 } 840 841 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 842 struct sk_buff **to_free) 843 { 844 qdisc_calculate_pkt_len(skb, sch); 845 return sch->enqueue(skb, sch, to_free); 846 } 847 848 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats, 849 __u64 bytes, __u32 packets) 850 { 851 u64_stats_update_begin(&bstats->syncp); 852 u64_stats_add(&bstats->bytes, bytes); 853 u64_stats_add(&bstats->packets, packets); 854 u64_stats_update_end(&bstats->syncp); 855 } 856 857 static inline void bstats_update(struct gnet_stats_basic_sync *bstats, 858 const struct sk_buff *skb) 859 { 860 _bstats_update(bstats, 861 qdisc_pkt_len(skb), 862 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 863 } 864 865 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 866 const struct sk_buff *skb) 867 { 868 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb); 869 } 870 871 static inline void qdisc_bstats_update(struct Qdisc *sch, 872 const struct sk_buff *skb) 873 { 874 bstats_update(&sch->bstats, skb); 875 } 876 877 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 878 const struct sk_buff *skb) 879 { 880 sch->qstats.backlog -= qdisc_pkt_len(skb); 881 } 882 883 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 884 const struct sk_buff *skb) 885 { 886 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 887 } 888 889 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 890 const struct sk_buff *skb) 891 { 892 sch->qstats.backlog += qdisc_pkt_len(skb); 893 } 894 895 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 896 const struct sk_buff *skb) 897 { 898 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 899 } 900 901 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 902 { 903 this_cpu_inc(sch->cpu_qstats->qlen); 904 } 905 906 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 907 { 908 this_cpu_dec(sch->cpu_qstats->qlen); 909 } 910 911 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 912 { 913 this_cpu_inc(sch->cpu_qstats->requeues); 914 } 915 916 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 917 { 918 sch->qstats.drops += count; 919 } 920 921 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 922 { 923 qstats->drops++; 924 } 925 926 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 927 { 928 qstats->overlimits++; 929 } 930 931 static inline void qdisc_qstats_drop(struct Qdisc *sch) 932 { 933 qstats_drop_inc(&sch->qstats); 934 } 935 936 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 937 { 938 this_cpu_inc(sch->cpu_qstats->drops); 939 } 940 941 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 942 { 943 sch->qstats.overlimits++; 944 } 945 946 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 947 { 948 __u32 qlen = qdisc_qlen_sum(sch); 949 950 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 951 } 952 953 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 954 __u32 *backlog) 955 { 956 struct gnet_stats_queue qstats = { 0 }; 957 958 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats); 959 *qlen = qstats.qlen + qdisc_qlen(sch); 960 *backlog = qstats.backlog; 961 } 962 963 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 964 { 965 __u32 qlen, backlog; 966 967 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 968 qdisc_tree_reduce_backlog(sch, qlen, backlog); 969 } 970 971 static inline void qdisc_purge_queue(struct Qdisc *sch) 972 { 973 __u32 qlen, backlog; 974 975 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 976 qdisc_reset(sch); 977 qdisc_tree_reduce_backlog(sch, qlen, backlog); 978 } 979 980 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 981 struct qdisc_skb_head *qh) 982 { 983 struct sk_buff *last = qh->tail; 984 985 if (last) { 986 skb->next = NULL; 987 last->next = skb; 988 qh->tail = skb; 989 } else { 990 qh->tail = skb; 991 qh->head = skb; 992 } 993 qh->qlen++; 994 } 995 996 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 997 { 998 __qdisc_enqueue_tail(skb, &sch->q); 999 qdisc_qstats_backlog_inc(sch, skb); 1000 return NET_XMIT_SUCCESS; 1001 } 1002 1003 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 1004 struct qdisc_skb_head *qh) 1005 { 1006 skb->next = qh->head; 1007 1008 if (!qh->head) 1009 qh->tail = skb; 1010 qh->head = skb; 1011 qh->qlen++; 1012 } 1013 1014 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 1015 { 1016 struct sk_buff *skb = qh->head; 1017 1018 if (likely(skb != NULL)) { 1019 qh->head = skb->next; 1020 qh->qlen--; 1021 if (qh->head == NULL) 1022 qh->tail = NULL; 1023 skb->next = NULL; 1024 } 1025 1026 return skb; 1027 } 1028 1029 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1030 { 1031 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1032 1033 if (likely(skb != NULL)) { 1034 qdisc_qstats_backlog_dec(sch, skb); 1035 qdisc_bstats_update(sch, skb); 1036 } 1037 1038 return skb; 1039 } 1040 1041 /* Instead of calling kfree_skb() while root qdisc lock is held, 1042 * queue the skb for future freeing at end of __dev_xmit_skb() 1043 */ 1044 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1045 { 1046 skb->next = *to_free; 1047 *to_free = skb; 1048 } 1049 1050 static inline void __qdisc_drop_all(struct sk_buff *skb, 1051 struct sk_buff **to_free) 1052 { 1053 if (skb->prev) 1054 skb->prev->next = *to_free; 1055 else 1056 skb->next = *to_free; 1057 *to_free = skb; 1058 } 1059 1060 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1061 struct qdisc_skb_head *qh, 1062 struct sk_buff **to_free) 1063 { 1064 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1065 1066 if (likely(skb != NULL)) { 1067 unsigned int len = qdisc_pkt_len(skb); 1068 1069 qdisc_qstats_backlog_dec(sch, skb); 1070 __qdisc_drop(skb, to_free); 1071 return len; 1072 } 1073 1074 return 0; 1075 } 1076 1077 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1078 { 1079 const struct qdisc_skb_head *qh = &sch->q; 1080 1081 return qh->head; 1082 } 1083 1084 /* generic pseudo peek method for non-work-conserving qdisc */ 1085 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1086 { 1087 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1088 1089 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1090 if (!skb) { 1091 skb = sch->dequeue(sch); 1092 1093 if (skb) { 1094 __skb_queue_head(&sch->gso_skb, skb); 1095 /* it's still part of the queue */ 1096 qdisc_qstats_backlog_inc(sch, skb); 1097 sch->q.qlen++; 1098 } 1099 } 1100 1101 return skb; 1102 } 1103 1104 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1105 struct sk_buff *skb) 1106 { 1107 if (qdisc_is_percpu_stats(sch)) { 1108 qdisc_qstats_cpu_backlog_dec(sch, skb); 1109 qdisc_bstats_cpu_update(sch, skb); 1110 qdisc_qstats_cpu_qlen_dec(sch); 1111 } else { 1112 qdisc_qstats_backlog_dec(sch, skb); 1113 qdisc_bstats_update(sch, skb); 1114 sch->q.qlen--; 1115 } 1116 } 1117 1118 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1119 unsigned int pkt_len) 1120 { 1121 if (qdisc_is_percpu_stats(sch)) { 1122 qdisc_qstats_cpu_qlen_inc(sch); 1123 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1124 } else { 1125 sch->qstats.backlog += pkt_len; 1126 sch->q.qlen++; 1127 } 1128 } 1129 1130 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1131 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1132 { 1133 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1134 1135 if (skb) { 1136 skb = __skb_dequeue(&sch->gso_skb); 1137 if (qdisc_is_percpu_stats(sch)) { 1138 qdisc_qstats_cpu_backlog_dec(sch, skb); 1139 qdisc_qstats_cpu_qlen_dec(sch); 1140 } else { 1141 qdisc_qstats_backlog_dec(sch, skb); 1142 sch->q.qlen--; 1143 } 1144 } else { 1145 skb = sch->dequeue(sch); 1146 } 1147 1148 return skb; 1149 } 1150 1151 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1152 { 1153 /* 1154 * We do not know the backlog in bytes of this list, it 1155 * is up to the caller to correct it 1156 */ 1157 ASSERT_RTNL(); 1158 if (qh->qlen) { 1159 rtnl_kfree_skbs(qh->head, qh->tail); 1160 1161 qh->head = NULL; 1162 qh->tail = NULL; 1163 qh->qlen = 0; 1164 } 1165 } 1166 1167 static inline void qdisc_reset_queue(struct Qdisc *sch) 1168 { 1169 __qdisc_reset_queue(&sch->q); 1170 } 1171 1172 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1173 struct Qdisc **pold) 1174 { 1175 struct Qdisc *old; 1176 1177 sch_tree_lock(sch); 1178 old = *pold; 1179 *pold = new; 1180 if (old != NULL) 1181 qdisc_purge_queue(old); 1182 sch_tree_unlock(sch); 1183 1184 return old; 1185 } 1186 1187 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1188 { 1189 rtnl_kfree_skbs(skb, skb); 1190 qdisc_qstats_drop(sch); 1191 } 1192 1193 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1194 struct sk_buff **to_free) 1195 { 1196 __qdisc_drop(skb, to_free); 1197 qdisc_qstats_cpu_drop(sch); 1198 1199 return NET_XMIT_DROP; 1200 } 1201 1202 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1203 struct sk_buff **to_free) 1204 { 1205 __qdisc_drop(skb, to_free); 1206 qdisc_qstats_drop(sch); 1207 1208 return NET_XMIT_DROP; 1209 } 1210 1211 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1212 struct sk_buff **to_free) 1213 { 1214 __qdisc_drop_all(skb, to_free); 1215 qdisc_qstats_drop(sch); 1216 1217 return NET_XMIT_DROP; 1218 } 1219 1220 struct psched_ratecfg { 1221 u64 rate_bytes_ps; /* bytes per second */ 1222 u32 mult; 1223 u16 overhead; 1224 u16 mpu; 1225 u8 linklayer; 1226 u8 shift; 1227 }; 1228 1229 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1230 unsigned int len) 1231 { 1232 len += r->overhead; 1233 1234 if (len < r->mpu) 1235 len = r->mpu; 1236 1237 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1238 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1239 1240 return ((u64)len * r->mult) >> r->shift; 1241 } 1242 1243 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1244 const struct tc_ratespec *conf, 1245 u64 rate64); 1246 1247 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1248 const struct psched_ratecfg *r) 1249 { 1250 memset(res, 0, sizeof(*res)); 1251 1252 /* legacy struct tc_ratespec has a 32bit @rate field 1253 * Qdisc using 64bit rate should add new attributes 1254 * in order to maintain compatibility. 1255 */ 1256 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1257 1258 res->overhead = r->overhead; 1259 res->mpu = r->mpu; 1260 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1261 } 1262 1263 struct psched_pktrate { 1264 u64 rate_pkts_ps; /* packets per second */ 1265 u32 mult; 1266 u8 shift; 1267 }; 1268 1269 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1270 unsigned int pkt_num) 1271 { 1272 return ((u64)pkt_num * r->mult) >> r->shift; 1273 } 1274 1275 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1276 1277 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1278 * The fast path only needs to access filter list and to update stats 1279 */ 1280 struct mini_Qdisc { 1281 struct tcf_proto *filter_list; 1282 struct tcf_block *block; 1283 struct gnet_stats_basic_sync __percpu *cpu_bstats; 1284 struct gnet_stats_queue __percpu *cpu_qstats; 1285 unsigned long rcu_state; 1286 }; 1287 1288 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1289 const struct sk_buff *skb) 1290 { 1291 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1292 } 1293 1294 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1295 { 1296 this_cpu_inc(miniq->cpu_qstats->drops); 1297 } 1298 1299 struct mini_Qdisc_pair { 1300 struct mini_Qdisc miniq1; 1301 struct mini_Qdisc miniq2; 1302 struct mini_Qdisc __rcu **p_miniq; 1303 }; 1304 1305 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1306 struct tcf_proto *tp_head); 1307 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1308 struct mini_Qdisc __rcu **p_miniq); 1309 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1310 struct tcf_block *block); 1311 1312 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx); 1313 1314 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1315 1316 /* Make sure qdisc is no longer in SCHED state. */ 1317 static inline void qdisc_synchronize(const struct Qdisc *q) 1318 { 1319 while (test_bit(__QDISC_STATE_SCHED, &q->state)) 1320 msleep(1); 1321 } 1322 1323 #endif 1324