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 /* Intended to be used by unlocked users, when concurrent qdisc release is 141 * possible. 142 */ 143 144 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc) 145 { 146 if (qdisc->flags & TCQ_F_BUILTIN) 147 return qdisc; 148 if (refcount_inc_not_zero(&qdisc->refcnt)) 149 return qdisc; 150 return NULL; 151 } 152 153 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc 154 * root_lock section, or provide their own memory barriers -- ordering 155 * against qdisc_run_begin/end() atomic bit operations. 156 */ 157 static inline bool qdisc_is_running(struct Qdisc *qdisc) 158 { 159 if (qdisc->flags & TCQ_F_NOLOCK) 160 return spin_is_locked(&qdisc->seqlock); 161 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 162 } 163 164 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc) 165 { 166 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY); 167 } 168 169 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) 170 { 171 return q->flags & TCQ_F_CPUSTATS; 172 } 173 174 static inline bool qdisc_is_empty(const struct Qdisc *qdisc) 175 { 176 if (qdisc_is_percpu_stats(qdisc)) 177 return nolock_qdisc_is_empty(qdisc); 178 return !READ_ONCE(qdisc->q.qlen); 179 } 180 181 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with 182 * the qdisc root lock acquired. 183 */ 184 static inline bool qdisc_run_begin(struct Qdisc *qdisc) 185 { 186 if (qdisc->flags & TCQ_F_NOLOCK) { 187 if (spin_trylock(&qdisc->seqlock)) 188 return true; 189 190 /* No need to insist if the MISSED flag was already set. 191 * Note that test_and_set_bit() also gives us memory ordering 192 * guarantees wrt potential earlier enqueue() and below 193 * spin_trylock(), both of which are necessary to prevent races 194 */ 195 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state)) 196 return false; 197 198 /* Try to take the lock again to make sure that we will either 199 * grab it or the CPU that still has it will see MISSED set 200 * when testing it in qdisc_run_end() 201 */ 202 return spin_trylock(&qdisc->seqlock); 203 } 204 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 205 } 206 207 static inline void qdisc_run_end(struct Qdisc *qdisc) 208 { 209 if (qdisc->flags & TCQ_F_NOLOCK) { 210 spin_unlock(&qdisc->seqlock); 211 212 /* spin_unlock() only has store-release semantic. The unlock 213 * and test_bit() ordering is a store-load ordering, so a full 214 * memory barrier is needed here. 215 */ 216 smp_mb(); 217 218 if (unlikely(test_bit(__QDISC_STATE_MISSED, 219 &qdisc->state))) 220 __netif_schedule(qdisc); 221 } else { 222 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 223 } 224 } 225 226 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 227 { 228 return qdisc->flags & TCQ_F_ONETXQUEUE; 229 } 230 231 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) 232 { 233 #ifdef CONFIG_BQL 234 /* Non-BQL migrated drivers will return 0, too. */ 235 return dql_avail(&txq->dql); 236 #else 237 return 0; 238 #endif 239 } 240 241 struct Qdisc_class_ops { 242 unsigned int flags; 243 /* Child qdisc manipulation */ 244 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 245 int (*graft)(struct Qdisc *, unsigned long cl, 246 struct Qdisc *, struct Qdisc **, 247 struct netlink_ext_ack *extack); 248 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 249 void (*qlen_notify)(struct Qdisc *, unsigned long); 250 251 /* Class manipulation routines */ 252 unsigned long (*find)(struct Qdisc *, u32 classid); 253 int (*change)(struct Qdisc *, u32, u32, 254 struct nlattr **, unsigned long *, 255 struct netlink_ext_ack *); 256 int (*delete)(struct Qdisc *, unsigned long, 257 struct netlink_ext_ack *); 258 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 259 260 /* Filter manipulation */ 261 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 262 unsigned long arg, 263 struct netlink_ext_ack *extack); 264 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 265 u32 classid); 266 void (*unbind_tcf)(struct Qdisc *, unsigned long); 267 268 /* rtnetlink specific */ 269 int (*dump)(struct Qdisc *, unsigned long, 270 struct sk_buff *skb, struct tcmsg*); 271 int (*dump_stats)(struct Qdisc *, unsigned long, 272 struct gnet_dump *); 273 }; 274 275 /* Qdisc_class_ops flag values */ 276 277 /* Implements API that doesn't require rtnl lock */ 278 enum qdisc_class_ops_flags { 279 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 280 }; 281 282 struct Qdisc_ops { 283 struct Qdisc_ops *next; 284 const struct Qdisc_class_ops *cl_ops; 285 char id[IFNAMSIZ]; 286 int priv_size; 287 unsigned int static_flags; 288 289 int (*enqueue)(struct sk_buff *skb, 290 struct Qdisc *sch, 291 struct sk_buff **to_free); 292 struct sk_buff * (*dequeue)(struct Qdisc *); 293 struct sk_buff * (*peek)(struct Qdisc *); 294 295 int (*init)(struct Qdisc *sch, struct nlattr *arg, 296 struct netlink_ext_ack *extack); 297 void (*reset)(struct Qdisc *); 298 void (*destroy)(struct Qdisc *); 299 int (*change)(struct Qdisc *sch, 300 struct nlattr *arg, 301 struct netlink_ext_ack *extack); 302 void (*attach)(struct Qdisc *sch); 303 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 304 void (*change_real_num_tx)(struct Qdisc *sch, 305 unsigned int new_real_tx); 306 307 int (*dump)(struct Qdisc *, struct sk_buff *); 308 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 309 310 void (*ingress_block_set)(struct Qdisc *sch, 311 u32 block_index); 312 void (*egress_block_set)(struct Qdisc *sch, 313 u32 block_index); 314 u32 (*ingress_block_get)(struct Qdisc *sch); 315 u32 (*egress_block_get)(struct Qdisc *sch); 316 317 struct module *owner; 318 }; 319 320 321 struct tcf_result { 322 union { 323 struct { 324 unsigned long class; 325 u32 classid; 326 }; 327 const struct tcf_proto *goto_tp; 328 329 /* used in the skb_tc_reinsert function */ 330 struct { 331 bool ingress; 332 struct gnet_stats_queue *qstats; 333 }; 334 }; 335 }; 336 337 struct tcf_chain; 338 339 struct tcf_proto_ops { 340 struct list_head head; 341 char kind[IFNAMSIZ]; 342 343 int (*classify)(struct sk_buff *, 344 const struct tcf_proto *, 345 struct tcf_result *); 346 int (*init)(struct tcf_proto*); 347 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 348 struct netlink_ext_ack *extack); 349 350 void* (*get)(struct tcf_proto*, u32 handle); 351 void (*put)(struct tcf_proto *tp, void *f); 352 int (*change)(struct net *net, struct sk_buff *, 353 struct tcf_proto*, unsigned long, 354 u32 handle, struct nlattr **, 355 void **, u32, 356 struct netlink_ext_ack *); 357 int (*delete)(struct tcf_proto *tp, void *arg, 358 bool *last, bool rtnl_held, 359 struct netlink_ext_ack *); 360 bool (*delete_empty)(struct tcf_proto *tp); 361 void (*walk)(struct tcf_proto *tp, 362 struct tcf_walker *arg, bool rtnl_held); 363 int (*reoffload)(struct tcf_proto *tp, bool add, 364 flow_setup_cb_t *cb, void *cb_priv, 365 struct netlink_ext_ack *extack); 366 void (*hw_add)(struct tcf_proto *tp, 367 void *type_data); 368 void (*hw_del)(struct tcf_proto *tp, 369 void *type_data); 370 void (*bind_class)(void *, u32, unsigned long, 371 void *, unsigned long); 372 void * (*tmplt_create)(struct net *net, 373 struct tcf_chain *chain, 374 struct nlattr **tca, 375 struct netlink_ext_ack *extack); 376 void (*tmplt_destroy)(void *tmplt_priv); 377 378 /* rtnetlink specific */ 379 int (*dump)(struct net*, struct tcf_proto*, void *, 380 struct sk_buff *skb, struct tcmsg*, 381 bool); 382 int (*terse_dump)(struct net *net, 383 struct tcf_proto *tp, void *fh, 384 struct sk_buff *skb, 385 struct tcmsg *t, bool rtnl_held); 386 int (*tmplt_dump)(struct sk_buff *skb, 387 struct net *net, 388 void *tmplt_priv); 389 390 struct module *owner; 391 int flags; 392 }; 393 394 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 395 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 396 * conditions can occur when filters are inserted/deleted simultaneously. 397 */ 398 enum tcf_proto_ops_flags { 399 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 400 }; 401 402 struct tcf_proto { 403 /* Fast access part */ 404 struct tcf_proto __rcu *next; 405 void __rcu *root; 406 407 /* called under RCU BH lock*/ 408 int (*classify)(struct sk_buff *, 409 const struct tcf_proto *, 410 struct tcf_result *); 411 __be16 protocol; 412 413 /* All the rest */ 414 u32 prio; 415 void *data; 416 const struct tcf_proto_ops *ops; 417 struct tcf_chain *chain; 418 /* Lock protects tcf_proto shared state and can be used by unlocked 419 * classifiers to protect their private data. 420 */ 421 spinlock_t lock; 422 bool deleting; 423 refcount_t refcnt; 424 struct rcu_head rcu; 425 struct hlist_node destroy_ht_node; 426 }; 427 428 struct qdisc_skb_cb { 429 struct { 430 unsigned int pkt_len; 431 u16 slave_dev_queue_mapping; 432 u16 tc_classid; 433 }; 434 #define QDISC_CB_PRIV_LEN 20 435 unsigned char data[QDISC_CB_PRIV_LEN]; 436 }; 437 438 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 439 440 struct tcf_chain { 441 /* Protects filter_chain. */ 442 struct mutex filter_chain_lock; 443 struct tcf_proto __rcu *filter_chain; 444 struct list_head list; 445 struct tcf_block *block; 446 u32 index; /* chain index */ 447 unsigned int refcnt; 448 unsigned int action_refcnt; 449 bool explicitly_created; 450 bool flushing; 451 const struct tcf_proto_ops *tmplt_ops; 452 void *tmplt_priv; 453 struct rcu_head rcu; 454 }; 455 456 struct tcf_block { 457 /* Lock protects tcf_block and lifetime-management data of chains 458 * attached to the block (refcnt, action_refcnt, explicitly_created). 459 */ 460 struct mutex lock; 461 struct list_head chain_list; 462 u32 index; /* block index for shared blocks */ 463 u32 classid; /* which class this block belongs to */ 464 refcount_t refcnt; 465 struct net *net; 466 struct Qdisc *q; 467 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 468 struct flow_block flow_block; 469 struct list_head owner_list; 470 bool keep_dst; 471 atomic_t offloadcnt; /* Number of oddloaded filters */ 472 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 473 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 474 struct { 475 struct tcf_chain *chain; 476 struct list_head filter_chain_list; 477 } chain0; 478 struct rcu_head rcu; 479 DECLARE_HASHTABLE(proto_destroy_ht, 7); 480 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 481 }; 482 483 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 484 { 485 return lockdep_is_held(&chain->filter_chain_lock); 486 } 487 488 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 489 { 490 return lockdep_is_held(&tp->lock); 491 } 492 493 #define tcf_chain_dereference(p, chain) \ 494 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 495 496 #define tcf_proto_dereference(p, tp) \ 497 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 498 499 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 500 { 501 struct qdisc_skb_cb *qcb; 502 503 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 504 BUILD_BUG_ON(sizeof(qcb->data) < sz); 505 } 506 507 static inline int qdisc_qlen(const struct Qdisc *q) 508 { 509 return q->q.qlen; 510 } 511 512 static inline int qdisc_qlen_sum(const struct Qdisc *q) 513 { 514 __u32 qlen = q->qstats.qlen; 515 int i; 516 517 if (qdisc_is_percpu_stats(q)) { 518 for_each_possible_cpu(i) 519 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 520 } else { 521 qlen += q->q.qlen; 522 } 523 524 return qlen; 525 } 526 527 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 528 { 529 return (struct qdisc_skb_cb *)skb->cb; 530 } 531 532 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 533 { 534 return &qdisc->q.lock; 535 } 536 537 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 538 { 539 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 540 541 return q; 542 } 543 544 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 545 { 546 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 547 } 548 549 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 550 { 551 return qdisc->dev_queue->qdisc_sleeping; 552 } 553 554 /* The qdisc root lock is a mechanism by which to top level 555 * of a qdisc tree can be locked from any qdisc node in the 556 * forest. This allows changing the configuration of some 557 * aspect of the qdisc tree while blocking out asynchronous 558 * qdisc access in the packet processing paths. 559 * 560 * It is only legal to do this when the root will not change 561 * on us. Otherwise we'll potentially lock the wrong qdisc 562 * root. This is enforced by holding the RTNL semaphore, which 563 * all users of this lock accessor must do. 564 */ 565 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 566 { 567 struct Qdisc *root = qdisc_root(qdisc); 568 569 ASSERT_RTNL(); 570 return qdisc_lock(root); 571 } 572 573 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 574 { 575 struct Qdisc *root = qdisc_root_sleeping(qdisc); 576 577 ASSERT_RTNL(); 578 return qdisc_lock(root); 579 } 580 581 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 582 { 583 return qdisc->dev_queue->dev; 584 } 585 586 static inline void sch_tree_lock(struct Qdisc *q) 587 { 588 if (q->flags & TCQ_F_MQROOT) 589 spin_lock_bh(qdisc_lock(q)); 590 else 591 spin_lock_bh(qdisc_root_sleeping_lock(q)); 592 } 593 594 static inline void sch_tree_unlock(struct Qdisc *q) 595 { 596 if (q->flags & TCQ_F_MQROOT) 597 spin_unlock_bh(qdisc_lock(q)); 598 else 599 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 600 } 601 602 extern struct Qdisc noop_qdisc; 603 extern struct Qdisc_ops noop_qdisc_ops; 604 extern struct Qdisc_ops pfifo_fast_ops; 605 extern struct Qdisc_ops mq_qdisc_ops; 606 extern struct Qdisc_ops noqueue_qdisc_ops; 607 extern const struct Qdisc_ops *default_qdisc_ops; 608 static inline const struct Qdisc_ops * 609 get_default_qdisc_ops(const struct net_device *dev, int ntx) 610 { 611 return ntx < dev->real_num_tx_queues ? 612 default_qdisc_ops : &pfifo_fast_ops; 613 } 614 615 struct Qdisc_class_common { 616 u32 classid; 617 struct hlist_node hnode; 618 }; 619 620 struct Qdisc_class_hash { 621 struct hlist_head *hash; 622 unsigned int hashsize; 623 unsigned int hashmask; 624 unsigned int hashelems; 625 }; 626 627 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 628 { 629 id ^= id >> 8; 630 id ^= id >> 4; 631 return id & mask; 632 } 633 634 static inline struct Qdisc_class_common * 635 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 636 { 637 struct Qdisc_class_common *cl; 638 unsigned int h; 639 640 if (!id) 641 return NULL; 642 643 h = qdisc_class_hash(id, hash->hashmask); 644 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 645 if (cl->classid == id) 646 return cl; 647 } 648 return NULL; 649 } 650 651 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 652 { 653 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 654 655 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 656 } 657 658 int qdisc_class_hash_init(struct Qdisc_class_hash *); 659 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 660 struct Qdisc_class_common *); 661 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 662 struct Qdisc_class_common *); 663 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 664 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 665 666 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 667 void dev_qdisc_change_real_num_tx(struct net_device *dev, 668 unsigned int new_real_tx); 669 void dev_init_scheduler(struct net_device *dev); 670 void dev_shutdown(struct net_device *dev); 671 void dev_activate(struct net_device *dev); 672 void dev_deactivate(struct net_device *dev); 673 void dev_deactivate_many(struct list_head *head); 674 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 675 struct Qdisc *qdisc); 676 void qdisc_reset(struct Qdisc *qdisc); 677 void qdisc_put(struct Qdisc *qdisc); 678 void qdisc_put_unlocked(struct Qdisc *qdisc); 679 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 680 #ifdef CONFIG_NET_SCHED 681 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 682 void *type_data); 683 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 684 struct Qdisc *new, struct Qdisc *old, 685 enum tc_setup_type type, void *type_data, 686 struct netlink_ext_ack *extack); 687 #else 688 static inline int 689 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 690 void *type_data) 691 { 692 q->flags &= ~TCQ_F_OFFLOADED; 693 return 0; 694 } 695 696 static inline void 697 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 698 struct Qdisc *new, struct Qdisc *old, 699 enum tc_setup_type type, void *type_data, 700 struct netlink_ext_ack *extack) 701 { 702 } 703 #endif 704 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 705 const struct Qdisc_ops *ops, 706 struct netlink_ext_ack *extack); 707 void qdisc_free(struct Qdisc *qdisc); 708 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 709 const struct Qdisc_ops *ops, u32 parentid, 710 struct netlink_ext_ack *extack); 711 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 712 const struct qdisc_size_table *stab); 713 int skb_do_redirect(struct sk_buff *); 714 715 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 716 { 717 #ifdef CONFIG_NET_CLS_ACT 718 return skb->tc_at_ingress; 719 #else 720 return false; 721 #endif 722 } 723 724 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 725 { 726 #ifdef CONFIG_NET_CLS_ACT 727 if (skb->tc_skip_classify) { 728 skb->tc_skip_classify = 0; 729 return true; 730 } 731 #endif 732 return false; 733 } 734 735 /* Reset all TX qdiscs greater than index of a device. */ 736 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 737 { 738 struct Qdisc *qdisc; 739 740 for (; i < dev->num_tx_queues; i++) { 741 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 742 if (qdisc) { 743 spin_lock_bh(qdisc_lock(qdisc)); 744 qdisc_reset(qdisc); 745 spin_unlock_bh(qdisc_lock(qdisc)); 746 } 747 } 748 } 749 750 /* Are all TX queues of the device empty? */ 751 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 752 { 753 unsigned int i; 754 755 rcu_read_lock(); 756 for (i = 0; i < dev->num_tx_queues; i++) { 757 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 758 const struct Qdisc *q = rcu_dereference(txq->qdisc); 759 760 if (!qdisc_is_empty(q)) { 761 rcu_read_unlock(); 762 return false; 763 } 764 } 765 rcu_read_unlock(); 766 return true; 767 } 768 769 /* Are any of the TX qdiscs changing? */ 770 static inline bool qdisc_tx_changing(const struct net_device *dev) 771 { 772 unsigned int i; 773 774 for (i = 0; i < dev->num_tx_queues; i++) { 775 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 776 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) 777 return true; 778 } 779 return false; 780 } 781 782 /* Is the device using the noop qdisc on all queues? */ 783 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 784 { 785 unsigned int i; 786 787 for (i = 0; i < dev->num_tx_queues; i++) { 788 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 789 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 790 return false; 791 } 792 return true; 793 } 794 795 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 796 { 797 return qdisc_skb_cb(skb)->pkt_len; 798 } 799 800 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 801 enum net_xmit_qdisc_t { 802 __NET_XMIT_STOLEN = 0x00010000, 803 __NET_XMIT_BYPASS = 0x00020000, 804 }; 805 806 #ifdef CONFIG_NET_CLS_ACT 807 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 808 #else 809 #define net_xmit_drop_count(e) (1) 810 #endif 811 812 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 813 const struct Qdisc *sch) 814 { 815 #ifdef CONFIG_NET_SCHED 816 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 817 818 if (stab) 819 __qdisc_calculate_pkt_len(skb, stab); 820 #endif 821 } 822 823 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 824 struct sk_buff **to_free) 825 { 826 qdisc_calculate_pkt_len(skb, sch); 827 return sch->enqueue(skb, sch, to_free); 828 } 829 830 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats, 831 __u64 bytes, __u32 packets) 832 { 833 u64_stats_update_begin(&bstats->syncp); 834 u64_stats_add(&bstats->bytes, bytes); 835 u64_stats_add(&bstats->packets, packets); 836 u64_stats_update_end(&bstats->syncp); 837 } 838 839 static inline void bstats_update(struct gnet_stats_basic_sync *bstats, 840 const struct sk_buff *skb) 841 { 842 _bstats_update(bstats, 843 qdisc_pkt_len(skb), 844 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 845 } 846 847 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 848 const struct sk_buff *skb) 849 { 850 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb); 851 } 852 853 static inline void qdisc_bstats_update(struct Qdisc *sch, 854 const struct sk_buff *skb) 855 { 856 bstats_update(&sch->bstats, skb); 857 } 858 859 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 860 const struct sk_buff *skb) 861 { 862 sch->qstats.backlog -= qdisc_pkt_len(skb); 863 } 864 865 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 866 const struct sk_buff *skb) 867 { 868 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 869 } 870 871 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 872 const struct sk_buff *skb) 873 { 874 sch->qstats.backlog += qdisc_pkt_len(skb); 875 } 876 877 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 878 const struct sk_buff *skb) 879 { 880 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 881 } 882 883 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 884 { 885 this_cpu_inc(sch->cpu_qstats->qlen); 886 } 887 888 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 889 { 890 this_cpu_dec(sch->cpu_qstats->qlen); 891 } 892 893 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 894 { 895 this_cpu_inc(sch->cpu_qstats->requeues); 896 } 897 898 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 899 { 900 sch->qstats.drops += count; 901 } 902 903 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 904 { 905 qstats->drops++; 906 } 907 908 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 909 { 910 qstats->overlimits++; 911 } 912 913 static inline void qdisc_qstats_drop(struct Qdisc *sch) 914 { 915 qstats_drop_inc(&sch->qstats); 916 } 917 918 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 919 { 920 this_cpu_inc(sch->cpu_qstats->drops); 921 } 922 923 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 924 { 925 sch->qstats.overlimits++; 926 } 927 928 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 929 { 930 __u32 qlen = qdisc_qlen_sum(sch); 931 932 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 933 } 934 935 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 936 __u32 *backlog) 937 { 938 struct gnet_stats_queue qstats = { 0 }; 939 940 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats); 941 *qlen = qstats.qlen + qdisc_qlen(sch); 942 *backlog = qstats.backlog; 943 } 944 945 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 946 { 947 __u32 qlen, backlog; 948 949 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 950 qdisc_tree_reduce_backlog(sch, qlen, backlog); 951 } 952 953 static inline void qdisc_purge_queue(struct Qdisc *sch) 954 { 955 __u32 qlen, backlog; 956 957 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 958 qdisc_reset(sch); 959 qdisc_tree_reduce_backlog(sch, qlen, backlog); 960 } 961 962 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh) 963 { 964 qh->head = NULL; 965 qh->tail = NULL; 966 qh->qlen = 0; 967 } 968 969 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 970 struct qdisc_skb_head *qh) 971 { 972 struct sk_buff *last = qh->tail; 973 974 if (last) { 975 skb->next = NULL; 976 last->next = skb; 977 qh->tail = skb; 978 } else { 979 qh->tail = skb; 980 qh->head = skb; 981 } 982 qh->qlen++; 983 } 984 985 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 986 { 987 __qdisc_enqueue_tail(skb, &sch->q); 988 qdisc_qstats_backlog_inc(sch, skb); 989 return NET_XMIT_SUCCESS; 990 } 991 992 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 993 struct qdisc_skb_head *qh) 994 { 995 skb->next = qh->head; 996 997 if (!qh->head) 998 qh->tail = skb; 999 qh->head = skb; 1000 qh->qlen++; 1001 } 1002 1003 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 1004 { 1005 struct sk_buff *skb = qh->head; 1006 1007 if (likely(skb != NULL)) { 1008 qh->head = skb->next; 1009 qh->qlen--; 1010 if (qh->head == NULL) 1011 qh->tail = NULL; 1012 skb->next = NULL; 1013 } 1014 1015 return skb; 1016 } 1017 1018 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1019 { 1020 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1021 1022 if (likely(skb != NULL)) { 1023 qdisc_qstats_backlog_dec(sch, skb); 1024 qdisc_bstats_update(sch, skb); 1025 } 1026 1027 return skb; 1028 } 1029 1030 /* Instead of calling kfree_skb() while root qdisc lock is held, 1031 * queue the skb for future freeing at end of __dev_xmit_skb() 1032 */ 1033 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1034 { 1035 skb->next = *to_free; 1036 *to_free = skb; 1037 } 1038 1039 static inline void __qdisc_drop_all(struct sk_buff *skb, 1040 struct sk_buff **to_free) 1041 { 1042 if (skb->prev) 1043 skb->prev->next = *to_free; 1044 else 1045 skb->next = *to_free; 1046 *to_free = skb; 1047 } 1048 1049 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1050 struct qdisc_skb_head *qh, 1051 struct sk_buff **to_free) 1052 { 1053 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1054 1055 if (likely(skb != NULL)) { 1056 unsigned int len = qdisc_pkt_len(skb); 1057 1058 qdisc_qstats_backlog_dec(sch, skb); 1059 __qdisc_drop(skb, to_free); 1060 return len; 1061 } 1062 1063 return 0; 1064 } 1065 1066 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1067 { 1068 const struct qdisc_skb_head *qh = &sch->q; 1069 1070 return qh->head; 1071 } 1072 1073 /* generic pseudo peek method for non-work-conserving qdisc */ 1074 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1075 { 1076 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1077 1078 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1079 if (!skb) { 1080 skb = sch->dequeue(sch); 1081 1082 if (skb) { 1083 __skb_queue_head(&sch->gso_skb, skb); 1084 /* it's still part of the queue */ 1085 qdisc_qstats_backlog_inc(sch, skb); 1086 sch->q.qlen++; 1087 } 1088 } 1089 1090 return skb; 1091 } 1092 1093 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1094 struct sk_buff *skb) 1095 { 1096 if (qdisc_is_percpu_stats(sch)) { 1097 qdisc_qstats_cpu_backlog_dec(sch, skb); 1098 qdisc_bstats_cpu_update(sch, skb); 1099 qdisc_qstats_cpu_qlen_dec(sch); 1100 } else { 1101 qdisc_qstats_backlog_dec(sch, skb); 1102 qdisc_bstats_update(sch, skb); 1103 sch->q.qlen--; 1104 } 1105 } 1106 1107 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1108 unsigned int pkt_len) 1109 { 1110 if (qdisc_is_percpu_stats(sch)) { 1111 qdisc_qstats_cpu_qlen_inc(sch); 1112 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1113 } else { 1114 sch->qstats.backlog += pkt_len; 1115 sch->q.qlen++; 1116 } 1117 } 1118 1119 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1120 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1121 { 1122 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1123 1124 if (skb) { 1125 skb = __skb_dequeue(&sch->gso_skb); 1126 if (qdisc_is_percpu_stats(sch)) { 1127 qdisc_qstats_cpu_backlog_dec(sch, skb); 1128 qdisc_qstats_cpu_qlen_dec(sch); 1129 } else { 1130 qdisc_qstats_backlog_dec(sch, skb); 1131 sch->q.qlen--; 1132 } 1133 } else { 1134 skb = sch->dequeue(sch); 1135 } 1136 1137 return skb; 1138 } 1139 1140 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1141 { 1142 /* 1143 * We do not know the backlog in bytes of this list, it 1144 * is up to the caller to correct it 1145 */ 1146 ASSERT_RTNL(); 1147 if (qh->qlen) { 1148 rtnl_kfree_skbs(qh->head, qh->tail); 1149 1150 qh->head = NULL; 1151 qh->tail = NULL; 1152 qh->qlen = 0; 1153 } 1154 } 1155 1156 static inline void qdisc_reset_queue(struct Qdisc *sch) 1157 { 1158 __qdisc_reset_queue(&sch->q); 1159 sch->qstats.backlog = 0; 1160 } 1161 1162 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1163 struct Qdisc **pold) 1164 { 1165 struct Qdisc *old; 1166 1167 sch_tree_lock(sch); 1168 old = *pold; 1169 *pold = new; 1170 if (old != NULL) 1171 qdisc_purge_queue(old); 1172 sch_tree_unlock(sch); 1173 1174 return old; 1175 } 1176 1177 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1178 { 1179 rtnl_kfree_skbs(skb, skb); 1180 qdisc_qstats_drop(sch); 1181 } 1182 1183 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1184 struct sk_buff **to_free) 1185 { 1186 __qdisc_drop(skb, to_free); 1187 qdisc_qstats_cpu_drop(sch); 1188 1189 return NET_XMIT_DROP; 1190 } 1191 1192 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1193 struct sk_buff **to_free) 1194 { 1195 __qdisc_drop(skb, to_free); 1196 qdisc_qstats_drop(sch); 1197 1198 return NET_XMIT_DROP; 1199 } 1200 1201 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1202 struct sk_buff **to_free) 1203 { 1204 __qdisc_drop_all(skb, to_free); 1205 qdisc_qstats_drop(sch); 1206 1207 return NET_XMIT_DROP; 1208 } 1209 1210 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1211 long it will take to send a packet given its size. 1212 */ 1213 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1214 { 1215 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1216 if (slot < 0) 1217 slot = 0; 1218 slot >>= rtab->rate.cell_log; 1219 if (slot > 255) 1220 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1221 return rtab->data[slot]; 1222 } 1223 1224 struct psched_ratecfg { 1225 u64 rate_bytes_ps; /* bytes per second */ 1226 u32 mult; 1227 u16 overhead; 1228 u16 mpu; 1229 u8 linklayer; 1230 u8 shift; 1231 }; 1232 1233 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1234 unsigned int len) 1235 { 1236 len += r->overhead; 1237 1238 if (len < r->mpu) 1239 len = r->mpu; 1240 1241 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1242 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1243 1244 return ((u64)len * r->mult) >> r->shift; 1245 } 1246 1247 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1248 const struct tc_ratespec *conf, 1249 u64 rate64); 1250 1251 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1252 const struct psched_ratecfg *r) 1253 { 1254 memset(res, 0, sizeof(*res)); 1255 1256 /* legacy struct tc_ratespec has a 32bit @rate field 1257 * Qdisc using 64bit rate should add new attributes 1258 * in order to maintain compatibility. 1259 */ 1260 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1261 1262 res->overhead = r->overhead; 1263 res->mpu = r->mpu; 1264 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1265 } 1266 1267 struct psched_pktrate { 1268 u64 rate_pkts_ps; /* packets per second */ 1269 u32 mult; 1270 u8 shift; 1271 }; 1272 1273 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1274 unsigned int pkt_num) 1275 { 1276 return ((u64)pkt_num * r->mult) >> r->shift; 1277 } 1278 1279 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1280 1281 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1282 * The fast path only needs to access filter list and to update stats 1283 */ 1284 struct mini_Qdisc { 1285 struct tcf_proto *filter_list; 1286 struct tcf_block *block; 1287 struct gnet_stats_basic_sync __percpu *cpu_bstats; 1288 struct gnet_stats_queue __percpu *cpu_qstats; 1289 unsigned long rcu_state; 1290 }; 1291 1292 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1293 const struct sk_buff *skb) 1294 { 1295 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1296 } 1297 1298 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1299 { 1300 this_cpu_inc(miniq->cpu_qstats->drops); 1301 } 1302 1303 struct mini_Qdisc_pair { 1304 struct mini_Qdisc miniq1; 1305 struct mini_Qdisc miniq2; 1306 struct mini_Qdisc __rcu **p_miniq; 1307 }; 1308 1309 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1310 struct tcf_proto *tp_head); 1311 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1312 struct mini_Qdisc __rcu **p_miniq); 1313 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1314 struct tcf_block *block); 1315 1316 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx); 1317 1318 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1319 1320 #endif 1321