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 #define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED) 44 #define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING) 45 46 #define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \ 47 QDISC_STATE_DRAINING) 48 49 struct qdisc_size_table { 50 struct rcu_head rcu; 51 struct list_head list; 52 struct tc_sizespec szopts; 53 int refcnt; 54 u16 data[]; 55 }; 56 57 /* similar to sk_buff_head, but skb->prev pointer is undefined. */ 58 struct qdisc_skb_head { 59 struct sk_buff *head; 60 struct sk_buff *tail; 61 __u32 qlen; 62 spinlock_t lock; 63 }; 64 65 struct Qdisc { 66 int (*enqueue)(struct sk_buff *skb, 67 struct Qdisc *sch, 68 struct sk_buff **to_free); 69 struct sk_buff * (*dequeue)(struct Qdisc *sch); 70 unsigned int flags; 71 #define TCQ_F_BUILTIN 1 72 #define TCQ_F_INGRESS 2 73 #define TCQ_F_CAN_BYPASS 4 74 #define TCQ_F_MQROOT 8 75 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for 76 * q->dev_queue : It can test 77 * netif_xmit_frozen_or_stopped() before 78 * dequeueing next packet. 79 * Its true for MQ/MQPRIO slaves, or non 80 * multiqueue device. 81 */ 82 #define TCQ_F_WARN_NONWC (1 << 16) 83 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */ 84 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy : 85 * qdisc_tree_decrease_qlen() should stop. 86 */ 87 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */ 88 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */ 89 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */ 90 u32 limit; 91 const struct Qdisc_ops *ops; 92 struct qdisc_size_table __rcu *stab; 93 struct hlist_node hash; 94 u32 handle; 95 u32 parent; 96 97 struct netdev_queue *dev_queue; 98 99 struct net_rate_estimator __rcu *rate_est; 100 struct gnet_stats_basic_cpu __percpu *cpu_bstats; 101 struct gnet_stats_queue __percpu *cpu_qstats; 102 int pad; 103 refcount_t refcnt; 104 105 /* 106 * For performance sake on SMP, we put highly modified fields at the end 107 */ 108 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp; 109 struct qdisc_skb_head q; 110 struct gnet_stats_basic_packed bstats; 111 seqcount_t running; 112 struct gnet_stats_queue qstats; 113 unsigned long state; 114 struct Qdisc *next_sched; 115 struct sk_buff_head skb_bad_txq; 116 117 spinlock_t busylock ____cacheline_aligned_in_smp; 118 spinlock_t seqlock; 119 120 struct rcu_head rcu; 121 122 /* private data */ 123 long privdata[] ____cacheline_aligned; 124 }; 125 126 static inline void qdisc_refcount_inc(struct Qdisc *qdisc) 127 { 128 if (qdisc->flags & TCQ_F_BUILTIN) 129 return; 130 refcount_inc(&qdisc->refcnt); 131 } 132 133 /* Intended to be used by unlocked users, when concurrent qdisc release is 134 * possible. 135 */ 136 137 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc) 138 { 139 if (qdisc->flags & TCQ_F_BUILTIN) 140 return qdisc; 141 if (refcount_inc_not_zero(&qdisc->refcnt)) 142 return qdisc; 143 return NULL; 144 } 145 146 static inline bool qdisc_is_running(struct Qdisc *qdisc) 147 { 148 if (qdisc->flags & TCQ_F_NOLOCK) 149 return spin_is_locked(&qdisc->seqlock); 150 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false; 151 } 152 153 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc) 154 { 155 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY); 156 } 157 158 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) 159 { 160 return q->flags & TCQ_F_CPUSTATS; 161 } 162 163 static inline bool qdisc_is_empty(const struct Qdisc *qdisc) 164 { 165 if (qdisc_is_percpu_stats(qdisc)) 166 return nolock_qdisc_is_empty(qdisc); 167 return !READ_ONCE(qdisc->q.qlen); 168 } 169 170 static inline bool qdisc_run_begin(struct Qdisc *qdisc) 171 { 172 if (qdisc->flags & TCQ_F_NOLOCK) { 173 if (spin_trylock(&qdisc->seqlock)) 174 return true; 175 176 /* If the MISSED flag is set, it means other thread has 177 * set the MISSED flag before second spin_trylock(), so 178 * we can return false here to avoid multi cpus doing 179 * the set_bit() and second spin_trylock() concurrently. 180 */ 181 if (test_bit(__QDISC_STATE_MISSED, &qdisc->state)) 182 return false; 183 184 /* Set the MISSED flag before the second spin_trylock(), 185 * if the second spin_trylock() return false, it means 186 * other cpu holding the lock will do dequeuing for us 187 * or it will see the MISSED flag set after releasing 188 * lock and reschedule the net_tx_action() to do the 189 * dequeuing. 190 */ 191 set_bit(__QDISC_STATE_MISSED, &qdisc->state); 192 193 /* Retry again in case other CPU may not see the new flag 194 * after it releases the lock at the end of qdisc_run_end(). 195 */ 196 return spin_trylock(&qdisc->seqlock); 197 } else if (qdisc_is_running(qdisc)) { 198 return false; 199 } 200 /* Variant of write_seqcount_begin() telling lockdep a trylock 201 * was attempted. 202 */ 203 raw_write_seqcount_begin(&qdisc->running); 204 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_); 205 return true; 206 } 207 208 static inline void qdisc_run_end(struct Qdisc *qdisc) 209 { 210 if (qdisc->flags & TCQ_F_NOLOCK) { 211 spin_unlock(&qdisc->seqlock); 212 213 if (unlikely(test_bit(__QDISC_STATE_MISSED, 214 &qdisc->state))) 215 __netif_schedule(qdisc); 216 } else { 217 write_seqcount_end(&qdisc->running); 218 } 219 } 220 221 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 222 { 223 return qdisc->flags & TCQ_F_ONETXQUEUE; 224 } 225 226 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) 227 { 228 #ifdef CONFIG_BQL 229 /* Non-BQL migrated drivers will return 0, too. */ 230 return dql_avail(&txq->dql); 231 #else 232 return 0; 233 #endif 234 } 235 236 struct Qdisc_class_ops { 237 unsigned int flags; 238 /* Child qdisc manipulation */ 239 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 240 int (*graft)(struct Qdisc *, unsigned long cl, 241 struct Qdisc *, struct Qdisc **, 242 struct netlink_ext_ack *extack); 243 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 244 void (*qlen_notify)(struct Qdisc *, unsigned long); 245 246 /* Class manipulation routines */ 247 unsigned long (*find)(struct Qdisc *, u32 classid); 248 int (*change)(struct Qdisc *, u32, u32, 249 struct nlattr **, unsigned long *, 250 struct netlink_ext_ack *); 251 int (*delete)(struct Qdisc *, unsigned long, 252 struct netlink_ext_ack *); 253 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 254 255 /* Filter manipulation */ 256 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 257 unsigned long arg, 258 struct netlink_ext_ack *extack); 259 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 260 u32 classid); 261 void (*unbind_tcf)(struct Qdisc *, unsigned long); 262 263 /* rtnetlink specific */ 264 int (*dump)(struct Qdisc *, unsigned long, 265 struct sk_buff *skb, struct tcmsg*); 266 int (*dump_stats)(struct Qdisc *, unsigned long, 267 struct gnet_dump *); 268 }; 269 270 /* Qdisc_class_ops flag values */ 271 272 /* Implements API that doesn't require rtnl lock */ 273 enum qdisc_class_ops_flags { 274 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 275 }; 276 277 struct Qdisc_ops { 278 struct Qdisc_ops *next; 279 const struct Qdisc_class_ops *cl_ops; 280 char id[IFNAMSIZ]; 281 int priv_size; 282 unsigned int static_flags; 283 284 int (*enqueue)(struct sk_buff *skb, 285 struct Qdisc *sch, 286 struct sk_buff **to_free); 287 struct sk_buff * (*dequeue)(struct Qdisc *); 288 struct sk_buff * (*peek)(struct Qdisc *); 289 290 int (*init)(struct Qdisc *sch, struct nlattr *arg, 291 struct netlink_ext_ack *extack); 292 void (*reset)(struct Qdisc *); 293 void (*destroy)(struct Qdisc *); 294 int (*change)(struct Qdisc *sch, 295 struct nlattr *arg, 296 struct netlink_ext_ack *extack); 297 void (*attach)(struct Qdisc *sch); 298 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 299 300 int (*dump)(struct Qdisc *, struct sk_buff *); 301 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 302 303 void (*ingress_block_set)(struct Qdisc *sch, 304 u32 block_index); 305 void (*egress_block_set)(struct Qdisc *sch, 306 u32 block_index); 307 u32 (*ingress_block_get)(struct Qdisc *sch); 308 u32 (*egress_block_get)(struct Qdisc *sch); 309 310 struct module *owner; 311 }; 312 313 314 struct tcf_result { 315 union { 316 struct { 317 unsigned long class; 318 u32 classid; 319 }; 320 const struct tcf_proto *goto_tp; 321 322 /* used in the skb_tc_reinsert function */ 323 struct { 324 bool ingress; 325 struct gnet_stats_queue *qstats; 326 }; 327 }; 328 }; 329 330 struct tcf_chain; 331 332 struct tcf_proto_ops { 333 struct list_head head; 334 char kind[IFNAMSIZ]; 335 336 int (*classify)(struct sk_buff *, 337 const struct tcf_proto *, 338 struct tcf_result *); 339 int (*init)(struct tcf_proto*); 340 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 341 struct netlink_ext_ack *extack); 342 343 void* (*get)(struct tcf_proto*, u32 handle); 344 void (*put)(struct tcf_proto *tp, void *f); 345 int (*change)(struct net *net, struct sk_buff *, 346 struct tcf_proto*, unsigned long, 347 u32 handle, struct nlattr **, 348 void **, bool, bool, 349 struct netlink_ext_ack *); 350 int (*delete)(struct tcf_proto *tp, void *arg, 351 bool *last, bool rtnl_held, 352 struct netlink_ext_ack *); 353 bool (*delete_empty)(struct tcf_proto *tp); 354 void (*walk)(struct tcf_proto *tp, 355 struct tcf_walker *arg, bool rtnl_held); 356 int (*reoffload)(struct tcf_proto *tp, bool add, 357 flow_setup_cb_t *cb, void *cb_priv, 358 struct netlink_ext_ack *extack); 359 void (*hw_add)(struct tcf_proto *tp, 360 void *type_data); 361 void (*hw_del)(struct tcf_proto *tp, 362 void *type_data); 363 void (*bind_class)(void *, u32, unsigned long, 364 void *, unsigned long); 365 void * (*tmplt_create)(struct net *net, 366 struct tcf_chain *chain, 367 struct nlattr **tca, 368 struct netlink_ext_ack *extack); 369 void (*tmplt_destroy)(void *tmplt_priv); 370 371 /* rtnetlink specific */ 372 int (*dump)(struct net*, struct tcf_proto*, void *, 373 struct sk_buff *skb, struct tcmsg*, 374 bool); 375 int (*terse_dump)(struct net *net, 376 struct tcf_proto *tp, void *fh, 377 struct sk_buff *skb, 378 struct tcmsg *t, bool rtnl_held); 379 int (*tmplt_dump)(struct sk_buff *skb, 380 struct net *net, 381 void *tmplt_priv); 382 383 struct module *owner; 384 int flags; 385 }; 386 387 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 388 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 389 * conditions can occur when filters are inserted/deleted simultaneously. 390 */ 391 enum tcf_proto_ops_flags { 392 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 393 }; 394 395 struct tcf_proto { 396 /* Fast access part */ 397 struct tcf_proto __rcu *next; 398 void __rcu *root; 399 400 /* called under RCU BH lock*/ 401 int (*classify)(struct sk_buff *, 402 const struct tcf_proto *, 403 struct tcf_result *); 404 __be16 protocol; 405 406 /* All the rest */ 407 u32 prio; 408 void *data; 409 const struct tcf_proto_ops *ops; 410 struct tcf_chain *chain; 411 /* Lock protects tcf_proto shared state and can be used by unlocked 412 * classifiers to protect their private data. 413 */ 414 spinlock_t lock; 415 bool deleting; 416 refcount_t refcnt; 417 struct rcu_head rcu; 418 struct hlist_node destroy_ht_node; 419 }; 420 421 struct qdisc_skb_cb { 422 struct { 423 unsigned int pkt_len; 424 u16 slave_dev_queue_mapping; 425 u16 tc_classid; 426 }; 427 #define QDISC_CB_PRIV_LEN 20 428 unsigned char data[QDISC_CB_PRIV_LEN]; 429 u16 mru; 430 bool post_ct; 431 }; 432 433 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 434 435 struct tcf_chain { 436 /* Protects filter_chain. */ 437 struct mutex filter_chain_lock; 438 struct tcf_proto __rcu *filter_chain; 439 struct list_head list; 440 struct tcf_block *block; 441 u32 index; /* chain index */ 442 unsigned int refcnt; 443 unsigned int action_refcnt; 444 bool explicitly_created; 445 bool flushing; 446 const struct tcf_proto_ops *tmplt_ops; 447 void *tmplt_priv; 448 struct rcu_head rcu; 449 }; 450 451 struct tcf_block { 452 /* Lock protects tcf_block and lifetime-management data of chains 453 * attached to the block (refcnt, action_refcnt, explicitly_created). 454 */ 455 struct mutex lock; 456 struct list_head chain_list; 457 u32 index; /* block index for shared blocks */ 458 u32 classid; /* which class this block belongs to */ 459 refcount_t refcnt; 460 struct net *net; 461 struct Qdisc *q; 462 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 463 struct flow_block flow_block; 464 struct list_head owner_list; 465 bool keep_dst; 466 atomic_t offloadcnt; /* Number of oddloaded filters */ 467 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 468 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 469 struct { 470 struct tcf_chain *chain; 471 struct list_head filter_chain_list; 472 } chain0; 473 struct rcu_head rcu; 474 DECLARE_HASHTABLE(proto_destroy_ht, 7); 475 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 476 }; 477 478 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 479 { 480 return lockdep_is_held(&chain->filter_chain_lock); 481 } 482 483 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 484 { 485 return lockdep_is_held(&tp->lock); 486 } 487 488 #define tcf_chain_dereference(p, chain) \ 489 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 490 491 #define tcf_proto_dereference(p, tp) \ 492 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 493 494 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 495 { 496 struct qdisc_skb_cb *qcb; 497 498 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 499 BUILD_BUG_ON(sizeof(qcb->data) < sz); 500 } 501 502 static inline int qdisc_qlen_cpu(const struct Qdisc *q) 503 { 504 return this_cpu_ptr(q->cpu_qstats)->qlen; 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 seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc) 582 { 583 struct Qdisc *root = qdisc_root_sleeping(qdisc); 584 585 ASSERT_RTNL(); 586 return &root->running; 587 } 588 589 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 590 { 591 return qdisc->dev_queue->dev; 592 } 593 594 static inline void sch_tree_lock(struct Qdisc *q) 595 { 596 if (q->flags & TCQ_F_MQROOT) 597 spin_lock_bh(qdisc_lock(q)); 598 else 599 spin_lock_bh(qdisc_root_sleeping_lock(q)); 600 } 601 602 static inline void sch_tree_unlock(struct Qdisc *q) 603 { 604 if (q->flags & TCQ_F_MQROOT) 605 spin_unlock_bh(qdisc_lock(q)); 606 else 607 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 608 } 609 610 extern struct Qdisc noop_qdisc; 611 extern struct Qdisc_ops noop_qdisc_ops; 612 extern struct Qdisc_ops pfifo_fast_ops; 613 extern struct Qdisc_ops mq_qdisc_ops; 614 extern struct Qdisc_ops noqueue_qdisc_ops; 615 extern const struct Qdisc_ops *default_qdisc_ops; 616 static inline const struct Qdisc_ops * 617 get_default_qdisc_ops(const struct net_device *dev, int ntx) 618 { 619 return ntx < dev->real_num_tx_queues ? 620 default_qdisc_ops : &pfifo_fast_ops; 621 } 622 623 struct Qdisc_class_common { 624 u32 classid; 625 struct hlist_node hnode; 626 }; 627 628 struct Qdisc_class_hash { 629 struct hlist_head *hash; 630 unsigned int hashsize; 631 unsigned int hashmask; 632 unsigned int hashelems; 633 }; 634 635 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 636 { 637 id ^= id >> 8; 638 id ^= id >> 4; 639 return id & mask; 640 } 641 642 static inline struct Qdisc_class_common * 643 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 644 { 645 struct Qdisc_class_common *cl; 646 unsigned int h; 647 648 if (!id) 649 return NULL; 650 651 h = qdisc_class_hash(id, hash->hashmask); 652 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 653 if (cl->classid == id) 654 return cl; 655 } 656 return NULL; 657 } 658 659 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 660 { 661 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 662 663 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 664 } 665 666 int qdisc_class_hash_init(struct Qdisc_class_hash *); 667 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 668 struct Qdisc_class_common *); 669 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 670 struct Qdisc_class_common *); 671 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 672 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 673 674 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 675 void dev_init_scheduler(struct net_device *dev); 676 void dev_shutdown(struct net_device *dev); 677 void dev_activate(struct net_device *dev); 678 void dev_deactivate(struct net_device *dev); 679 void dev_deactivate_many(struct list_head *head); 680 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 681 struct Qdisc *qdisc); 682 void qdisc_reset(struct Qdisc *qdisc); 683 void qdisc_put(struct Qdisc *qdisc); 684 void qdisc_put_unlocked(struct Qdisc *qdisc); 685 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 686 #ifdef CONFIG_NET_SCHED 687 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 688 void *type_data); 689 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 690 struct Qdisc *new, struct Qdisc *old, 691 enum tc_setup_type type, void *type_data, 692 struct netlink_ext_ack *extack); 693 #else 694 static inline int 695 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 696 void *type_data) 697 { 698 q->flags &= ~TCQ_F_OFFLOADED; 699 return 0; 700 } 701 702 static inline void 703 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 704 struct Qdisc *new, struct Qdisc *old, 705 enum tc_setup_type type, void *type_data, 706 struct netlink_ext_ack *extack) 707 { 708 } 709 #endif 710 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 711 const struct Qdisc_ops *ops, 712 struct netlink_ext_ack *extack); 713 void qdisc_free(struct Qdisc *qdisc); 714 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 715 const struct Qdisc_ops *ops, u32 parentid, 716 struct netlink_ext_ack *extack); 717 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 718 const struct qdisc_size_table *stab); 719 int skb_do_redirect(struct sk_buff *); 720 721 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 722 { 723 #ifdef CONFIG_NET_CLS_ACT 724 return skb->tc_at_ingress; 725 #else 726 return false; 727 #endif 728 } 729 730 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 731 { 732 #ifdef CONFIG_NET_CLS_ACT 733 if (skb->tc_skip_classify) { 734 skb->tc_skip_classify = 0; 735 return true; 736 } 737 #endif 738 return false; 739 } 740 741 /* Reset all TX qdiscs greater than index of a device. */ 742 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 743 { 744 struct Qdisc *qdisc; 745 746 for (; i < dev->num_tx_queues; i++) { 747 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 748 if (qdisc) { 749 spin_lock_bh(qdisc_lock(qdisc)); 750 qdisc_reset(qdisc); 751 spin_unlock_bh(qdisc_lock(qdisc)); 752 } 753 } 754 } 755 756 /* Are all TX queues of the device empty? */ 757 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 758 { 759 unsigned int i; 760 761 rcu_read_lock(); 762 for (i = 0; i < dev->num_tx_queues; i++) { 763 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 764 const struct Qdisc *q = rcu_dereference(txq->qdisc); 765 766 if (!qdisc_is_empty(q)) { 767 rcu_read_unlock(); 768 return false; 769 } 770 } 771 rcu_read_unlock(); 772 return true; 773 } 774 775 /* Are any of the TX qdiscs changing? */ 776 static inline bool qdisc_tx_changing(const struct net_device *dev) 777 { 778 unsigned int i; 779 780 for (i = 0; i < dev->num_tx_queues; i++) { 781 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 782 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) 783 return true; 784 } 785 return false; 786 } 787 788 /* Is the device using the noop qdisc on all queues? */ 789 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 790 { 791 unsigned int i; 792 793 for (i = 0; i < dev->num_tx_queues; i++) { 794 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 795 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 796 return false; 797 } 798 return true; 799 } 800 801 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 802 { 803 return qdisc_skb_cb(skb)->pkt_len; 804 } 805 806 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 807 enum net_xmit_qdisc_t { 808 __NET_XMIT_STOLEN = 0x00010000, 809 __NET_XMIT_BYPASS = 0x00020000, 810 }; 811 812 #ifdef CONFIG_NET_CLS_ACT 813 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 814 #else 815 #define net_xmit_drop_count(e) (1) 816 #endif 817 818 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 819 const struct Qdisc *sch) 820 { 821 #ifdef CONFIG_NET_SCHED 822 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 823 824 if (stab) 825 __qdisc_calculate_pkt_len(skb, stab); 826 #endif 827 } 828 829 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 830 struct sk_buff **to_free) 831 { 832 qdisc_calculate_pkt_len(skb, sch); 833 return sch->enqueue(skb, sch, to_free); 834 } 835 836 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats, 837 __u64 bytes, __u32 packets) 838 { 839 bstats->bytes += bytes; 840 bstats->packets += packets; 841 } 842 843 static inline void bstats_update(struct gnet_stats_basic_packed *bstats, 844 const struct sk_buff *skb) 845 { 846 _bstats_update(bstats, 847 qdisc_pkt_len(skb), 848 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 849 } 850 851 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 852 __u64 bytes, __u32 packets) 853 { 854 u64_stats_update_begin(&bstats->syncp); 855 _bstats_update(&bstats->bstats, bytes, packets); 856 u64_stats_update_end(&bstats->syncp); 857 } 858 859 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 860 const struct sk_buff *skb) 861 { 862 u64_stats_update_begin(&bstats->syncp); 863 bstats_update(&bstats->bstats, skb); 864 u64_stats_update_end(&bstats->syncp); 865 } 866 867 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 868 const struct sk_buff *skb) 869 { 870 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb); 871 } 872 873 static inline void qdisc_bstats_update(struct Qdisc *sch, 874 const struct sk_buff *skb) 875 { 876 bstats_update(&sch->bstats, skb); 877 } 878 879 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 880 const struct sk_buff *skb) 881 { 882 sch->qstats.backlog -= qdisc_pkt_len(skb); 883 } 884 885 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 886 const struct sk_buff *skb) 887 { 888 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 889 } 890 891 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 892 const struct sk_buff *skb) 893 { 894 sch->qstats.backlog += qdisc_pkt_len(skb); 895 } 896 897 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 898 const struct sk_buff *skb) 899 { 900 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 901 } 902 903 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 904 { 905 this_cpu_inc(sch->cpu_qstats->qlen); 906 } 907 908 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 909 { 910 this_cpu_dec(sch->cpu_qstats->qlen); 911 } 912 913 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 914 { 915 this_cpu_inc(sch->cpu_qstats->requeues); 916 } 917 918 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 919 { 920 sch->qstats.drops += count; 921 } 922 923 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 924 { 925 qstats->drops++; 926 } 927 928 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 929 { 930 qstats->overlimits++; 931 } 932 933 static inline void qdisc_qstats_drop(struct Qdisc *sch) 934 { 935 qstats_drop_inc(&sch->qstats); 936 } 937 938 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 939 { 940 this_cpu_inc(sch->cpu_qstats->drops); 941 } 942 943 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 944 { 945 sch->qstats.overlimits++; 946 } 947 948 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 949 { 950 __u32 qlen = qdisc_qlen_sum(sch); 951 952 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 953 } 954 955 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 956 __u32 *backlog) 957 { 958 struct gnet_stats_queue qstats = { 0 }; 959 __u32 len = qdisc_qlen_sum(sch); 960 961 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len); 962 *qlen = qstats.qlen; 963 *backlog = qstats.backlog; 964 } 965 966 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 967 { 968 __u32 qlen, backlog; 969 970 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 971 qdisc_tree_reduce_backlog(sch, qlen, backlog); 972 } 973 974 static inline void qdisc_purge_queue(struct Qdisc *sch) 975 { 976 __u32 qlen, backlog; 977 978 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 979 qdisc_reset(sch); 980 qdisc_tree_reduce_backlog(sch, qlen, backlog); 981 } 982 983 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh) 984 { 985 qh->head = NULL; 986 qh->tail = NULL; 987 qh->qlen = 0; 988 } 989 990 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 991 struct qdisc_skb_head *qh) 992 { 993 struct sk_buff *last = qh->tail; 994 995 if (last) { 996 skb->next = NULL; 997 last->next = skb; 998 qh->tail = skb; 999 } else { 1000 qh->tail = skb; 1001 qh->head = skb; 1002 } 1003 qh->qlen++; 1004 } 1005 1006 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 1007 { 1008 __qdisc_enqueue_tail(skb, &sch->q); 1009 qdisc_qstats_backlog_inc(sch, skb); 1010 return NET_XMIT_SUCCESS; 1011 } 1012 1013 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 1014 struct qdisc_skb_head *qh) 1015 { 1016 skb->next = qh->head; 1017 1018 if (!qh->head) 1019 qh->tail = skb; 1020 qh->head = skb; 1021 qh->qlen++; 1022 } 1023 1024 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 1025 { 1026 struct sk_buff *skb = qh->head; 1027 1028 if (likely(skb != NULL)) { 1029 qh->head = skb->next; 1030 qh->qlen--; 1031 if (qh->head == NULL) 1032 qh->tail = NULL; 1033 skb->next = NULL; 1034 } 1035 1036 return skb; 1037 } 1038 1039 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1040 { 1041 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1042 1043 if (likely(skb != NULL)) { 1044 qdisc_qstats_backlog_dec(sch, skb); 1045 qdisc_bstats_update(sch, skb); 1046 } 1047 1048 return skb; 1049 } 1050 1051 /* Instead of calling kfree_skb() while root qdisc lock is held, 1052 * queue the skb for future freeing at end of __dev_xmit_skb() 1053 */ 1054 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1055 { 1056 skb->next = *to_free; 1057 *to_free = skb; 1058 } 1059 1060 static inline void __qdisc_drop_all(struct sk_buff *skb, 1061 struct sk_buff **to_free) 1062 { 1063 if (skb->prev) 1064 skb->prev->next = *to_free; 1065 else 1066 skb->next = *to_free; 1067 *to_free = skb; 1068 } 1069 1070 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1071 struct qdisc_skb_head *qh, 1072 struct sk_buff **to_free) 1073 { 1074 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1075 1076 if (likely(skb != NULL)) { 1077 unsigned int len = qdisc_pkt_len(skb); 1078 1079 qdisc_qstats_backlog_dec(sch, skb); 1080 __qdisc_drop(skb, to_free); 1081 return len; 1082 } 1083 1084 return 0; 1085 } 1086 1087 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1088 { 1089 const struct qdisc_skb_head *qh = &sch->q; 1090 1091 return qh->head; 1092 } 1093 1094 /* generic pseudo peek method for non-work-conserving qdisc */ 1095 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1096 { 1097 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1098 1099 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1100 if (!skb) { 1101 skb = sch->dequeue(sch); 1102 1103 if (skb) { 1104 __skb_queue_head(&sch->gso_skb, skb); 1105 /* it's still part of the queue */ 1106 qdisc_qstats_backlog_inc(sch, skb); 1107 sch->q.qlen++; 1108 } 1109 } 1110 1111 return skb; 1112 } 1113 1114 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1115 struct sk_buff *skb) 1116 { 1117 if (qdisc_is_percpu_stats(sch)) { 1118 qdisc_qstats_cpu_backlog_dec(sch, skb); 1119 qdisc_bstats_cpu_update(sch, skb); 1120 qdisc_qstats_cpu_qlen_dec(sch); 1121 } else { 1122 qdisc_qstats_backlog_dec(sch, skb); 1123 qdisc_bstats_update(sch, skb); 1124 sch->q.qlen--; 1125 } 1126 } 1127 1128 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1129 unsigned int pkt_len) 1130 { 1131 if (qdisc_is_percpu_stats(sch)) { 1132 qdisc_qstats_cpu_qlen_inc(sch); 1133 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1134 } else { 1135 sch->qstats.backlog += pkt_len; 1136 sch->q.qlen++; 1137 } 1138 } 1139 1140 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1141 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1142 { 1143 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1144 1145 if (skb) { 1146 skb = __skb_dequeue(&sch->gso_skb); 1147 if (qdisc_is_percpu_stats(sch)) { 1148 qdisc_qstats_cpu_backlog_dec(sch, skb); 1149 qdisc_qstats_cpu_qlen_dec(sch); 1150 } else { 1151 qdisc_qstats_backlog_dec(sch, skb); 1152 sch->q.qlen--; 1153 } 1154 } else { 1155 skb = sch->dequeue(sch); 1156 } 1157 1158 return skb; 1159 } 1160 1161 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1162 { 1163 /* 1164 * We do not know the backlog in bytes of this list, it 1165 * is up to the caller to correct it 1166 */ 1167 ASSERT_RTNL(); 1168 if (qh->qlen) { 1169 rtnl_kfree_skbs(qh->head, qh->tail); 1170 1171 qh->head = NULL; 1172 qh->tail = NULL; 1173 qh->qlen = 0; 1174 } 1175 } 1176 1177 static inline void qdisc_reset_queue(struct Qdisc *sch) 1178 { 1179 __qdisc_reset_queue(&sch->q); 1180 sch->qstats.backlog = 0; 1181 } 1182 1183 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1184 struct Qdisc **pold) 1185 { 1186 struct Qdisc *old; 1187 1188 sch_tree_lock(sch); 1189 old = *pold; 1190 *pold = new; 1191 if (old != NULL) 1192 qdisc_purge_queue(old); 1193 sch_tree_unlock(sch); 1194 1195 return old; 1196 } 1197 1198 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1199 { 1200 rtnl_kfree_skbs(skb, skb); 1201 qdisc_qstats_drop(sch); 1202 } 1203 1204 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1205 struct sk_buff **to_free) 1206 { 1207 __qdisc_drop(skb, to_free); 1208 qdisc_qstats_cpu_drop(sch); 1209 1210 return NET_XMIT_DROP; 1211 } 1212 1213 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1214 struct sk_buff **to_free) 1215 { 1216 __qdisc_drop(skb, to_free); 1217 qdisc_qstats_drop(sch); 1218 1219 return NET_XMIT_DROP; 1220 } 1221 1222 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1223 struct sk_buff **to_free) 1224 { 1225 __qdisc_drop_all(skb, to_free); 1226 qdisc_qstats_drop(sch); 1227 1228 return NET_XMIT_DROP; 1229 } 1230 1231 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1232 long it will take to send a packet given its size. 1233 */ 1234 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1235 { 1236 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1237 if (slot < 0) 1238 slot = 0; 1239 slot >>= rtab->rate.cell_log; 1240 if (slot > 255) 1241 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1242 return rtab->data[slot]; 1243 } 1244 1245 struct psched_ratecfg { 1246 u64 rate_bytes_ps; /* bytes per second */ 1247 u32 mult; 1248 u16 overhead; 1249 u8 linklayer; 1250 u8 shift; 1251 }; 1252 1253 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1254 unsigned int len) 1255 { 1256 len += r->overhead; 1257 1258 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1259 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1260 1261 return ((u64)len * r->mult) >> r->shift; 1262 } 1263 1264 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1265 const struct tc_ratespec *conf, 1266 u64 rate64); 1267 1268 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1269 const struct psched_ratecfg *r) 1270 { 1271 memset(res, 0, sizeof(*res)); 1272 1273 /* legacy struct tc_ratespec has a 32bit @rate field 1274 * Qdisc using 64bit rate should add new attributes 1275 * in order to maintain compatibility. 1276 */ 1277 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1278 1279 res->overhead = r->overhead; 1280 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1281 } 1282 1283 struct psched_pktrate { 1284 u64 rate_pkts_ps; /* packets per second */ 1285 u32 mult; 1286 u8 shift; 1287 }; 1288 1289 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1290 unsigned int pkt_num) 1291 { 1292 return ((u64)pkt_num * r->mult) >> r->shift; 1293 } 1294 1295 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1296 1297 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1298 * The fast path only needs to access filter list and to update stats 1299 */ 1300 struct mini_Qdisc { 1301 struct tcf_proto *filter_list; 1302 struct tcf_block *block; 1303 struct gnet_stats_basic_cpu __percpu *cpu_bstats; 1304 struct gnet_stats_queue __percpu *cpu_qstats; 1305 struct rcu_head rcu; 1306 }; 1307 1308 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1309 const struct sk_buff *skb) 1310 { 1311 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1312 } 1313 1314 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1315 { 1316 this_cpu_inc(miniq->cpu_qstats->drops); 1317 } 1318 1319 struct mini_Qdisc_pair { 1320 struct mini_Qdisc miniq1; 1321 struct mini_Qdisc miniq2; 1322 struct mini_Qdisc __rcu **p_miniq; 1323 }; 1324 1325 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1326 struct tcf_proto *tp_head); 1327 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1328 struct mini_Qdisc __rcu **p_miniq); 1329 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1330 struct tcf_block *block); 1331 1332 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1333 1334 #endif 1335