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 <net/gen_stats.h> 17 #include <net/rtnetlink.h> 18 19 struct Qdisc_ops; 20 struct qdisc_walker; 21 struct tcf_walker; 22 struct module; 23 struct bpf_flow_keys; 24 25 typedef int tc_setup_cb_t(enum tc_setup_type type, 26 void *type_data, void *cb_priv); 27 28 typedef int tc_indr_block_bind_cb_t(struct net_device *dev, void *cb_priv, 29 enum tc_setup_type type, void *type_data); 30 31 struct qdisc_rate_table { 32 struct tc_ratespec rate; 33 u32 data[256]; 34 struct qdisc_rate_table *next; 35 int refcnt; 36 }; 37 38 enum qdisc_state_t { 39 __QDISC_STATE_SCHED, 40 __QDISC_STATE_DEACTIVATED, 41 }; 42 43 struct qdisc_size_table { 44 struct rcu_head rcu; 45 struct list_head list; 46 struct tc_sizespec szopts; 47 int refcnt; 48 u16 data[]; 49 }; 50 51 /* similar to sk_buff_head, but skb->prev pointer is undefined. */ 52 struct qdisc_skb_head { 53 struct sk_buff *head; 54 struct sk_buff *tail; 55 __u32 qlen; 56 spinlock_t lock; 57 }; 58 59 struct Qdisc { 60 int (*enqueue)(struct sk_buff *skb, 61 struct Qdisc *sch, 62 struct sk_buff **to_free); 63 struct sk_buff * (*dequeue)(struct Qdisc *sch); 64 unsigned int flags; 65 #define TCQ_F_BUILTIN 1 66 #define TCQ_F_INGRESS 2 67 #define TCQ_F_CAN_BYPASS 4 68 #define TCQ_F_MQROOT 8 69 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for 70 * q->dev_queue : It can test 71 * netif_xmit_frozen_or_stopped() before 72 * dequeueing next packet. 73 * Its true for MQ/MQPRIO slaves, or non 74 * multiqueue device. 75 */ 76 #define TCQ_F_WARN_NONWC (1 << 16) 77 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */ 78 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy : 79 * qdisc_tree_decrease_qlen() should stop. 80 */ 81 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */ 82 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */ 83 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */ 84 u32 limit; 85 const struct Qdisc_ops *ops; 86 struct qdisc_size_table __rcu *stab; 87 struct hlist_node hash; 88 u32 handle; 89 u32 parent; 90 91 struct netdev_queue *dev_queue; 92 93 struct net_rate_estimator __rcu *rate_est; 94 struct gnet_stats_basic_cpu __percpu *cpu_bstats; 95 struct gnet_stats_queue __percpu *cpu_qstats; 96 int padded; 97 refcount_t refcnt; 98 99 /* 100 * For performance sake on SMP, we put highly modified fields at the end 101 */ 102 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp; 103 struct qdisc_skb_head q; 104 struct gnet_stats_basic_packed bstats; 105 seqcount_t running; 106 struct gnet_stats_queue qstats; 107 unsigned long state; 108 struct Qdisc *next_sched; 109 struct sk_buff_head skb_bad_txq; 110 111 spinlock_t busylock ____cacheline_aligned_in_smp; 112 spinlock_t seqlock; 113 114 /* for NOLOCK qdisc, true if there are no enqueued skbs */ 115 bool empty; 116 struct rcu_head rcu; 117 }; 118 119 static inline void qdisc_refcount_inc(struct Qdisc *qdisc) 120 { 121 if (qdisc->flags & TCQ_F_BUILTIN) 122 return; 123 refcount_inc(&qdisc->refcnt); 124 } 125 126 /* Intended to be used by unlocked users, when concurrent qdisc release is 127 * possible. 128 */ 129 130 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc) 131 { 132 if (qdisc->flags & TCQ_F_BUILTIN) 133 return qdisc; 134 if (refcount_inc_not_zero(&qdisc->refcnt)) 135 return qdisc; 136 return NULL; 137 } 138 139 static inline bool qdisc_is_running(struct Qdisc *qdisc) 140 { 141 if (qdisc->flags & TCQ_F_NOLOCK) 142 return spin_is_locked(&qdisc->seqlock); 143 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false; 144 } 145 146 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) 147 { 148 return q->flags & TCQ_F_CPUSTATS; 149 } 150 151 static inline bool qdisc_is_empty(const struct Qdisc *qdisc) 152 { 153 if (qdisc_is_percpu_stats(qdisc)) 154 return qdisc->empty; 155 return !qdisc->q.qlen; 156 } 157 158 static inline bool qdisc_run_begin(struct Qdisc *qdisc) 159 { 160 if (qdisc->flags & TCQ_F_NOLOCK) { 161 if (!spin_trylock(&qdisc->seqlock)) 162 return false; 163 qdisc->empty = false; 164 } else if (qdisc_is_running(qdisc)) { 165 return false; 166 } 167 /* Variant of write_seqcount_begin() telling lockdep a trylock 168 * was attempted. 169 */ 170 raw_write_seqcount_begin(&qdisc->running); 171 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_); 172 return true; 173 } 174 175 static inline void qdisc_run_end(struct Qdisc *qdisc) 176 { 177 write_seqcount_end(&qdisc->running); 178 if (qdisc->flags & TCQ_F_NOLOCK) 179 spin_unlock(&qdisc->seqlock); 180 } 181 182 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 183 { 184 return qdisc->flags & TCQ_F_ONETXQUEUE; 185 } 186 187 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) 188 { 189 #ifdef CONFIG_BQL 190 /* Non-BQL migrated drivers will return 0, too. */ 191 return dql_avail(&txq->dql); 192 #else 193 return 0; 194 #endif 195 } 196 197 struct Qdisc_class_ops { 198 unsigned int flags; 199 /* Child qdisc manipulation */ 200 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 201 int (*graft)(struct Qdisc *, unsigned long cl, 202 struct Qdisc *, struct Qdisc **, 203 struct netlink_ext_ack *extack); 204 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 205 void (*qlen_notify)(struct Qdisc *, unsigned long); 206 207 /* Class manipulation routines */ 208 unsigned long (*find)(struct Qdisc *, u32 classid); 209 int (*change)(struct Qdisc *, u32, u32, 210 struct nlattr **, unsigned long *, 211 struct netlink_ext_ack *); 212 int (*delete)(struct Qdisc *, unsigned long); 213 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 214 215 /* Filter manipulation */ 216 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 217 unsigned long arg, 218 struct netlink_ext_ack *extack); 219 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 220 u32 classid); 221 void (*unbind_tcf)(struct Qdisc *, unsigned long); 222 223 /* rtnetlink specific */ 224 int (*dump)(struct Qdisc *, unsigned long, 225 struct sk_buff *skb, struct tcmsg*); 226 int (*dump_stats)(struct Qdisc *, unsigned long, 227 struct gnet_dump *); 228 }; 229 230 /* Qdisc_class_ops flag values */ 231 232 /* Implements API that doesn't require rtnl lock */ 233 enum qdisc_class_ops_flags { 234 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 235 }; 236 237 struct Qdisc_ops { 238 struct Qdisc_ops *next; 239 const struct Qdisc_class_ops *cl_ops; 240 char id[IFNAMSIZ]; 241 int priv_size; 242 unsigned int static_flags; 243 244 int (*enqueue)(struct sk_buff *skb, 245 struct Qdisc *sch, 246 struct sk_buff **to_free); 247 struct sk_buff * (*dequeue)(struct Qdisc *); 248 struct sk_buff * (*peek)(struct Qdisc *); 249 250 int (*init)(struct Qdisc *sch, struct nlattr *arg, 251 struct netlink_ext_ack *extack); 252 void (*reset)(struct Qdisc *); 253 void (*destroy)(struct Qdisc *); 254 int (*change)(struct Qdisc *sch, 255 struct nlattr *arg, 256 struct netlink_ext_ack *extack); 257 void (*attach)(struct Qdisc *sch); 258 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 259 260 int (*dump)(struct Qdisc *, struct sk_buff *); 261 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 262 263 void (*ingress_block_set)(struct Qdisc *sch, 264 u32 block_index); 265 void (*egress_block_set)(struct Qdisc *sch, 266 u32 block_index); 267 u32 (*ingress_block_get)(struct Qdisc *sch); 268 u32 (*egress_block_get)(struct Qdisc *sch); 269 270 struct module *owner; 271 }; 272 273 274 struct tcf_result { 275 union { 276 struct { 277 unsigned long class; 278 u32 classid; 279 }; 280 const struct tcf_proto *goto_tp; 281 282 /* used in the skb_tc_reinsert function */ 283 struct { 284 bool ingress; 285 struct gnet_stats_queue *qstats; 286 }; 287 }; 288 }; 289 290 struct tcf_chain; 291 292 struct tcf_proto_ops { 293 struct list_head head; 294 char kind[IFNAMSIZ]; 295 296 int (*classify)(struct sk_buff *, 297 const struct tcf_proto *, 298 struct tcf_result *); 299 int (*init)(struct tcf_proto*); 300 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 301 struct netlink_ext_ack *extack); 302 303 void* (*get)(struct tcf_proto*, u32 handle); 304 void (*put)(struct tcf_proto *tp, void *f); 305 int (*change)(struct net *net, struct sk_buff *, 306 struct tcf_proto*, unsigned long, 307 u32 handle, struct nlattr **, 308 void **, bool, bool, 309 struct netlink_ext_ack *); 310 int (*delete)(struct tcf_proto *tp, void *arg, 311 bool *last, bool rtnl_held, 312 struct netlink_ext_ack *); 313 void (*walk)(struct tcf_proto *tp, 314 struct tcf_walker *arg, bool rtnl_held); 315 int (*reoffload)(struct tcf_proto *tp, bool add, 316 tc_setup_cb_t *cb, void *cb_priv, 317 struct netlink_ext_ack *extack); 318 void (*bind_class)(void *, u32, unsigned long); 319 void * (*tmplt_create)(struct net *net, 320 struct tcf_chain *chain, 321 struct nlattr **tca, 322 struct netlink_ext_ack *extack); 323 void (*tmplt_destroy)(void *tmplt_priv); 324 325 /* rtnetlink specific */ 326 int (*dump)(struct net*, struct tcf_proto*, void *, 327 struct sk_buff *skb, struct tcmsg*, 328 bool); 329 int (*tmplt_dump)(struct sk_buff *skb, 330 struct net *net, 331 void *tmplt_priv); 332 333 struct module *owner; 334 int flags; 335 }; 336 337 enum tcf_proto_ops_flags { 338 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 339 }; 340 341 struct tcf_proto { 342 /* Fast access part */ 343 struct tcf_proto __rcu *next; 344 void __rcu *root; 345 346 /* called under RCU BH lock*/ 347 int (*classify)(struct sk_buff *, 348 const struct tcf_proto *, 349 struct tcf_result *); 350 __be16 protocol; 351 352 /* All the rest */ 353 u32 prio; 354 void *data; 355 const struct tcf_proto_ops *ops; 356 struct tcf_chain *chain; 357 /* Lock protects tcf_proto shared state and can be used by unlocked 358 * classifiers to protect their private data. 359 */ 360 spinlock_t lock; 361 bool deleting; 362 refcount_t refcnt; 363 struct rcu_head rcu; 364 }; 365 366 struct qdisc_skb_cb { 367 struct { 368 unsigned int pkt_len; 369 u16 slave_dev_queue_mapping; 370 u16 tc_classid; 371 }; 372 #define QDISC_CB_PRIV_LEN 20 373 unsigned char data[QDISC_CB_PRIV_LEN]; 374 }; 375 376 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 377 378 struct tcf_chain { 379 /* Protects filter_chain. */ 380 struct mutex filter_chain_lock; 381 struct tcf_proto __rcu *filter_chain; 382 struct list_head list; 383 struct tcf_block *block; 384 u32 index; /* chain index */ 385 unsigned int refcnt; 386 unsigned int action_refcnt; 387 bool explicitly_created; 388 bool flushing; 389 const struct tcf_proto_ops *tmplt_ops; 390 void *tmplt_priv; 391 struct rcu_head rcu; 392 }; 393 394 struct tcf_block { 395 /* Lock protects tcf_block and lifetime-management data of chains 396 * attached to the block (refcnt, action_refcnt, explicitly_created). 397 */ 398 struct mutex lock; 399 struct list_head chain_list; 400 u32 index; /* block index for shared blocks */ 401 refcount_t refcnt; 402 struct net *net; 403 struct Qdisc *q; 404 struct list_head cb_list; 405 struct list_head owner_list; 406 bool keep_dst; 407 unsigned int offloadcnt; /* Number of oddloaded filters */ 408 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 409 struct { 410 struct tcf_chain *chain; 411 struct list_head filter_chain_list; 412 } chain0; 413 struct rcu_head rcu; 414 }; 415 416 #ifdef CONFIG_PROVE_LOCKING 417 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 418 { 419 return lockdep_is_held(&chain->filter_chain_lock); 420 } 421 422 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 423 { 424 return lockdep_is_held(&tp->lock); 425 } 426 #else 427 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain) 428 { 429 return true; 430 } 431 432 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 433 { 434 return true; 435 } 436 #endif /* #ifdef CONFIG_PROVE_LOCKING */ 437 438 #define tcf_chain_dereference(p, chain) \ 439 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 440 441 #define tcf_proto_dereference(p, tp) \ 442 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 443 444 static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags) 445 { 446 if (*flags & TCA_CLS_FLAGS_IN_HW) 447 return; 448 *flags |= TCA_CLS_FLAGS_IN_HW; 449 block->offloadcnt++; 450 } 451 452 static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags) 453 { 454 if (!(*flags & TCA_CLS_FLAGS_IN_HW)) 455 return; 456 *flags &= ~TCA_CLS_FLAGS_IN_HW; 457 block->offloadcnt--; 458 } 459 460 static inline void 461 tc_cls_offload_cnt_update(struct tcf_block *block, u32 *cnt, 462 u32 *flags, bool add) 463 { 464 if (add) { 465 if (!*cnt) 466 tcf_block_offload_inc(block, flags); 467 (*cnt)++; 468 } else { 469 (*cnt)--; 470 if (!*cnt) 471 tcf_block_offload_dec(block, flags); 472 } 473 } 474 475 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 476 { 477 struct qdisc_skb_cb *qcb; 478 479 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz); 480 BUILD_BUG_ON(sizeof(qcb->data) < sz); 481 } 482 483 static inline int qdisc_qlen_cpu(const struct Qdisc *q) 484 { 485 return this_cpu_ptr(q->cpu_qstats)->qlen; 486 } 487 488 static inline int qdisc_qlen(const struct Qdisc *q) 489 { 490 return q->q.qlen; 491 } 492 493 static inline int qdisc_qlen_sum(const struct Qdisc *q) 494 { 495 __u32 qlen = q->qstats.qlen; 496 int i; 497 498 if (qdisc_is_percpu_stats(q)) { 499 for_each_possible_cpu(i) 500 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 501 } else { 502 qlen += q->q.qlen; 503 } 504 505 return qlen; 506 } 507 508 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 509 { 510 return (struct qdisc_skb_cb *)skb->cb; 511 } 512 513 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 514 { 515 return &qdisc->q.lock; 516 } 517 518 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 519 { 520 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 521 522 return q; 523 } 524 525 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 526 { 527 return qdisc->dev_queue->qdisc_sleeping; 528 } 529 530 /* The qdisc root lock is a mechanism by which to top level 531 * of a qdisc tree can be locked from any qdisc node in the 532 * forest. This allows changing the configuration of some 533 * aspect of the qdisc tree while blocking out asynchronous 534 * qdisc access in the packet processing paths. 535 * 536 * It is only legal to do this when the root will not change 537 * on us. Otherwise we'll potentially lock the wrong qdisc 538 * root. This is enforced by holding the RTNL semaphore, which 539 * all users of this lock accessor must do. 540 */ 541 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 542 { 543 struct Qdisc *root = qdisc_root(qdisc); 544 545 ASSERT_RTNL(); 546 return qdisc_lock(root); 547 } 548 549 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 550 { 551 struct Qdisc *root = qdisc_root_sleeping(qdisc); 552 553 ASSERT_RTNL(); 554 return qdisc_lock(root); 555 } 556 557 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc) 558 { 559 struct Qdisc *root = qdisc_root_sleeping(qdisc); 560 561 ASSERT_RTNL(); 562 return &root->running; 563 } 564 565 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 566 { 567 return qdisc->dev_queue->dev; 568 } 569 570 static inline void sch_tree_lock(const struct Qdisc *q) 571 { 572 spin_lock_bh(qdisc_root_sleeping_lock(q)); 573 } 574 575 static inline void sch_tree_unlock(const struct Qdisc *q) 576 { 577 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 578 } 579 580 extern struct Qdisc noop_qdisc; 581 extern struct Qdisc_ops noop_qdisc_ops; 582 extern struct Qdisc_ops pfifo_fast_ops; 583 extern struct Qdisc_ops mq_qdisc_ops; 584 extern struct Qdisc_ops noqueue_qdisc_ops; 585 extern const struct Qdisc_ops *default_qdisc_ops; 586 static inline const struct Qdisc_ops * 587 get_default_qdisc_ops(const struct net_device *dev, int ntx) 588 { 589 return ntx < dev->real_num_tx_queues ? 590 default_qdisc_ops : &pfifo_fast_ops; 591 } 592 593 struct Qdisc_class_common { 594 u32 classid; 595 struct hlist_node hnode; 596 }; 597 598 struct Qdisc_class_hash { 599 struct hlist_head *hash; 600 unsigned int hashsize; 601 unsigned int hashmask; 602 unsigned int hashelems; 603 }; 604 605 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 606 { 607 id ^= id >> 8; 608 id ^= id >> 4; 609 return id & mask; 610 } 611 612 static inline struct Qdisc_class_common * 613 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 614 { 615 struct Qdisc_class_common *cl; 616 unsigned int h; 617 618 if (!id) 619 return NULL; 620 621 h = qdisc_class_hash(id, hash->hashmask); 622 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 623 if (cl->classid == id) 624 return cl; 625 } 626 return NULL; 627 } 628 629 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 630 { 631 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 632 633 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 634 } 635 636 int qdisc_class_hash_init(struct Qdisc_class_hash *); 637 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 638 struct Qdisc_class_common *); 639 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 640 struct Qdisc_class_common *); 641 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 642 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 643 644 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 645 void dev_init_scheduler(struct net_device *dev); 646 void dev_shutdown(struct net_device *dev); 647 void dev_activate(struct net_device *dev); 648 void dev_deactivate(struct net_device *dev); 649 void dev_deactivate_many(struct list_head *head); 650 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 651 struct Qdisc *qdisc); 652 void qdisc_reset(struct Qdisc *qdisc); 653 void qdisc_put(struct Qdisc *qdisc); 654 void qdisc_put_unlocked(struct Qdisc *qdisc); 655 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 656 #ifdef CONFIG_NET_SCHED 657 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 658 void *type_data); 659 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 660 struct Qdisc *new, struct Qdisc *old, 661 enum tc_setup_type type, void *type_data, 662 struct netlink_ext_ack *extack); 663 #else 664 static inline int 665 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 666 void *type_data) 667 { 668 q->flags &= ~TCQ_F_OFFLOADED; 669 return 0; 670 } 671 672 static inline void 673 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 674 struct Qdisc *new, struct Qdisc *old, 675 enum tc_setup_type type, void *type_data, 676 struct netlink_ext_ack *extack) 677 { 678 } 679 #endif 680 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 681 const struct Qdisc_ops *ops, 682 struct netlink_ext_ack *extack); 683 void qdisc_free(struct Qdisc *qdisc); 684 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 685 const struct Qdisc_ops *ops, u32 parentid, 686 struct netlink_ext_ack *extack); 687 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 688 const struct qdisc_size_table *stab); 689 int skb_do_redirect(struct sk_buff *); 690 691 static inline void skb_reset_tc(struct sk_buff *skb) 692 { 693 #ifdef CONFIG_NET_CLS_ACT 694 skb->tc_redirected = 0; 695 #endif 696 } 697 698 static inline bool skb_is_tc_redirected(const struct sk_buff *skb) 699 { 700 #ifdef CONFIG_NET_CLS_ACT 701 return skb->tc_redirected; 702 #else 703 return false; 704 #endif 705 } 706 707 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 708 { 709 #ifdef CONFIG_NET_CLS_ACT 710 return skb->tc_at_ingress; 711 #else 712 return false; 713 #endif 714 } 715 716 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 717 { 718 #ifdef CONFIG_NET_CLS_ACT 719 if (skb->tc_skip_classify) { 720 skb->tc_skip_classify = 0; 721 return true; 722 } 723 #endif 724 return false; 725 } 726 727 /* Reset all TX qdiscs greater than index of a device. */ 728 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 729 { 730 struct Qdisc *qdisc; 731 732 for (; i < dev->num_tx_queues; i++) { 733 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 734 if (qdisc) { 735 spin_lock_bh(qdisc_lock(qdisc)); 736 qdisc_reset(qdisc); 737 spin_unlock_bh(qdisc_lock(qdisc)); 738 } 739 } 740 } 741 742 static inline void qdisc_reset_all_tx(struct net_device *dev) 743 { 744 qdisc_reset_all_tx_gt(dev, 0); 745 } 746 747 /* Are all TX queues of the device empty? */ 748 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 749 { 750 unsigned int i; 751 752 rcu_read_lock(); 753 for (i = 0; i < dev->num_tx_queues; i++) { 754 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 755 const struct Qdisc *q = rcu_dereference(txq->qdisc); 756 757 if (!qdisc_is_empty(q)) { 758 rcu_read_unlock(); 759 return false; 760 } 761 } 762 rcu_read_unlock(); 763 return true; 764 } 765 766 /* Are any of the TX qdiscs changing? */ 767 static inline bool qdisc_tx_changing(const struct net_device *dev) 768 { 769 unsigned int i; 770 771 for (i = 0; i < dev->num_tx_queues; i++) { 772 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 773 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) 774 return true; 775 } 776 return false; 777 } 778 779 /* Is the device using the noop qdisc on all queues? */ 780 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 781 { 782 unsigned int i; 783 784 for (i = 0; i < dev->num_tx_queues; i++) { 785 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 786 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 787 return false; 788 } 789 return true; 790 } 791 792 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 793 { 794 return qdisc_skb_cb(skb)->pkt_len; 795 } 796 797 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 798 enum net_xmit_qdisc_t { 799 __NET_XMIT_STOLEN = 0x00010000, 800 __NET_XMIT_BYPASS = 0x00020000, 801 }; 802 803 #ifdef CONFIG_NET_CLS_ACT 804 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 805 #else 806 #define net_xmit_drop_count(e) (1) 807 #endif 808 809 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 810 const struct Qdisc *sch) 811 { 812 #ifdef CONFIG_NET_SCHED 813 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 814 815 if (stab) 816 __qdisc_calculate_pkt_len(skb, stab); 817 #endif 818 } 819 820 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 821 struct sk_buff **to_free) 822 { 823 qdisc_calculate_pkt_len(skb, sch); 824 return sch->enqueue(skb, sch, to_free); 825 } 826 827 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats, 828 __u64 bytes, __u32 packets) 829 { 830 bstats->bytes += bytes; 831 bstats->packets += packets; 832 } 833 834 static inline void bstats_update(struct gnet_stats_basic_packed *bstats, 835 const struct sk_buff *skb) 836 { 837 _bstats_update(bstats, 838 qdisc_pkt_len(skb), 839 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 840 } 841 842 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 843 __u64 bytes, __u32 packets) 844 { 845 u64_stats_update_begin(&bstats->syncp); 846 _bstats_update(&bstats->bstats, bytes, packets); 847 u64_stats_update_end(&bstats->syncp); 848 } 849 850 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 851 const struct sk_buff *skb) 852 { 853 u64_stats_update_begin(&bstats->syncp); 854 bstats_update(&bstats->bstats, skb); 855 u64_stats_update_end(&bstats->syncp); 856 } 857 858 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 859 const struct sk_buff *skb) 860 { 861 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb); 862 } 863 864 static inline void qdisc_bstats_update(struct Qdisc *sch, 865 const struct sk_buff *skb) 866 { 867 bstats_update(&sch->bstats, skb); 868 } 869 870 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 871 const struct sk_buff *skb) 872 { 873 sch->qstats.backlog -= qdisc_pkt_len(skb); 874 } 875 876 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 877 const struct sk_buff *skb) 878 { 879 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 880 } 881 882 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 883 const struct sk_buff *skb) 884 { 885 sch->qstats.backlog += qdisc_pkt_len(skb); 886 } 887 888 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 889 const struct sk_buff *skb) 890 { 891 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 892 } 893 894 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 895 { 896 this_cpu_inc(sch->cpu_qstats->qlen); 897 } 898 899 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 900 { 901 this_cpu_dec(sch->cpu_qstats->qlen); 902 } 903 904 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 905 { 906 this_cpu_inc(sch->cpu_qstats->requeues); 907 } 908 909 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 910 { 911 sch->qstats.drops += count; 912 } 913 914 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 915 { 916 qstats->drops++; 917 } 918 919 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 920 { 921 qstats->overlimits++; 922 } 923 924 static inline void qdisc_qstats_drop(struct Qdisc *sch) 925 { 926 qstats_drop_inc(&sch->qstats); 927 } 928 929 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 930 { 931 this_cpu_inc(sch->cpu_qstats->drops); 932 } 933 934 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 935 { 936 sch->qstats.overlimits++; 937 } 938 939 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 940 { 941 __u32 qlen = qdisc_qlen_sum(sch); 942 943 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 944 } 945 946 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 947 __u32 *backlog) 948 { 949 struct gnet_stats_queue qstats = { 0 }; 950 __u32 len = qdisc_qlen_sum(sch); 951 952 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len); 953 *qlen = qstats.qlen; 954 *backlog = qstats.backlog; 955 } 956 957 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 958 { 959 __u32 qlen, backlog; 960 961 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 962 qdisc_tree_reduce_backlog(sch, qlen, backlog); 963 } 964 965 static inline void qdisc_purge_queue(struct Qdisc *sch) 966 { 967 __u32 qlen, backlog; 968 969 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 970 qdisc_reset(sch); 971 qdisc_tree_reduce_backlog(sch, qlen, backlog); 972 } 973 974 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh) 975 { 976 qh->head = NULL; 977 qh->tail = NULL; 978 qh->qlen = 0; 979 } 980 981 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 982 struct qdisc_skb_head *qh) 983 { 984 struct sk_buff *last = qh->tail; 985 986 if (last) { 987 skb->next = NULL; 988 last->next = skb; 989 qh->tail = skb; 990 } else { 991 qh->tail = skb; 992 qh->head = skb; 993 } 994 qh->qlen++; 995 } 996 997 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 998 { 999 __qdisc_enqueue_tail(skb, &sch->q); 1000 qdisc_qstats_backlog_inc(sch, skb); 1001 return NET_XMIT_SUCCESS; 1002 } 1003 1004 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 1005 struct qdisc_skb_head *qh) 1006 { 1007 skb->next = qh->head; 1008 1009 if (!qh->head) 1010 qh->tail = skb; 1011 qh->head = skb; 1012 qh->qlen++; 1013 } 1014 1015 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 1016 { 1017 struct sk_buff *skb = qh->head; 1018 1019 if (likely(skb != NULL)) { 1020 qh->head = skb->next; 1021 qh->qlen--; 1022 if (qh->head == NULL) 1023 qh->tail = NULL; 1024 skb->next = NULL; 1025 } 1026 1027 return skb; 1028 } 1029 1030 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1031 { 1032 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1033 1034 if (likely(skb != NULL)) { 1035 qdisc_qstats_backlog_dec(sch, skb); 1036 qdisc_bstats_update(sch, skb); 1037 } 1038 1039 return skb; 1040 } 1041 1042 /* Instead of calling kfree_skb() while root qdisc lock is held, 1043 * queue the skb for future freeing at end of __dev_xmit_skb() 1044 */ 1045 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1046 { 1047 skb->next = *to_free; 1048 *to_free = skb; 1049 } 1050 1051 static inline void __qdisc_drop_all(struct sk_buff *skb, 1052 struct sk_buff **to_free) 1053 { 1054 if (skb->prev) 1055 skb->prev->next = *to_free; 1056 else 1057 skb->next = *to_free; 1058 *to_free = skb; 1059 } 1060 1061 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1062 struct qdisc_skb_head *qh, 1063 struct sk_buff **to_free) 1064 { 1065 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1066 1067 if (likely(skb != NULL)) { 1068 unsigned int len = qdisc_pkt_len(skb); 1069 1070 qdisc_qstats_backlog_dec(sch, skb); 1071 __qdisc_drop(skb, to_free); 1072 return len; 1073 } 1074 1075 return 0; 1076 } 1077 1078 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch, 1079 struct sk_buff **to_free) 1080 { 1081 return __qdisc_queue_drop_head(sch, &sch->q, to_free); 1082 } 1083 1084 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1085 { 1086 const struct qdisc_skb_head *qh = &sch->q; 1087 1088 return qh->head; 1089 } 1090 1091 /* generic pseudo peek method for non-work-conserving qdisc */ 1092 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1093 { 1094 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1095 1096 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1097 if (!skb) { 1098 skb = sch->dequeue(sch); 1099 1100 if (skb) { 1101 __skb_queue_head(&sch->gso_skb, skb); 1102 /* it's still part of the queue */ 1103 qdisc_qstats_backlog_inc(sch, skb); 1104 sch->q.qlen++; 1105 } 1106 } 1107 1108 return skb; 1109 } 1110 1111 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1112 struct sk_buff *skb) 1113 { 1114 if (qdisc_is_percpu_stats(sch)) { 1115 qdisc_qstats_cpu_backlog_dec(sch, skb); 1116 qdisc_bstats_cpu_update(sch, skb); 1117 qdisc_qstats_cpu_qlen_dec(sch); 1118 } else { 1119 qdisc_qstats_backlog_dec(sch, skb); 1120 qdisc_bstats_update(sch, skb); 1121 sch->q.qlen--; 1122 } 1123 } 1124 1125 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1126 unsigned int pkt_len) 1127 { 1128 if (qdisc_is_percpu_stats(sch)) { 1129 qdisc_qstats_cpu_qlen_inc(sch); 1130 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1131 } else { 1132 sch->qstats.backlog += pkt_len; 1133 sch->q.qlen++; 1134 } 1135 } 1136 1137 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1138 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1139 { 1140 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1141 1142 if (skb) { 1143 skb = __skb_dequeue(&sch->gso_skb); 1144 if (qdisc_is_percpu_stats(sch)) { 1145 qdisc_qstats_cpu_backlog_dec(sch, skb); 1146 qdisc_qstats_cpu_qlen_dec(sch); 1147 } else { 1148 qdisc_qstats_backlog_dec(sch, skb); 1149 sch->q.qlen--; 1150 } 1151 } else { 1152 skb = sch->dequeue(sch); 1153 } 1154 1155 return skb; 1156 } 1157 1158 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1159 { 1160 /* 1161 * We do not know the backlog in bytes of this list, it 1162 * is up to the caller to correct it 1163 */ 1164 ASSERT_RTNL(); 1165 if (qh->qlen) { 1166 rtnl_kfree_skbs(qh->head, qh->tail); 1167 1168 qh->head = NULL; 1169 qh->tail = NULL; 1170 qh->qlen = 0; 1171 } 1172 } 1173 1174 static inline void qdisc_reset_queue(struct Qdisc *sch) 1175 { 1176 __qdisc_reset_queue(&sch->q); 1177 sch->qstats.backlog = 0; 1178 } 1179 1180 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1181 struct Qdisc **pold) 1182 { 1183 struct Qdisc *old; 1184 1185 sch_tree_lock(sch); 1186 old = *pold; 1187 *pold = new; 1188 if (old != NULL) 1189 qdisc_tree_flush_backlog(old); 1190 sch_tree_unlock(sch); 1191 1192 return old; 1193 } 1194 1195 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1196 { 1197 rtnl_kfree_skbs(skb, skb); 1198 qdisc_qstats_drop(sch); 1199 } 1200 1201 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1202 struct sk_buff **to_free) 1203 { 1204 __qdisc_drop(skb, to_free); 1205 qdisc_qstats_cpu_drop(sch); 1206 1207 return NET_XMIT_DROP; 1208 } 1209 1210 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1211 struct sk_buff **to_free) 1212 { 1213 __qdisc_drop(skb, to_free); 1214 qdisc_qstats_drop(sch); 1215 1216 return NET_XMIT_DROP; 1217 } 1218 1219 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1220 struct sk_buff **to_free) 1221 { 1222 __qdisc_drop_all(skb, to_free); 1223 qdisc_qstats_drop(sch); 1224 1225 return NET_XMIT_DROP; 1226 } 1227 1228 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1229 long it will take to send a packet given its size. 1230 */ 1231 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1232 { 1233 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1234 if (slot < 0) 1235 slot = 0; 1236 slot >>= rtab->rate.cell_log; 1237 if (slot > 255) 1238 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1239 return rtab->data[slot]; 1240 } 1241 1242 struct psched_ratecfg { 1243 u64 rate_bytes_ps; /* bytes per second */ 1244 u32 mult; 1245 u16 overhead; 1246 u8 linklayer; 1247 u8 shift; 1248 }; 1249 1250 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1251 unsigned int len) 1252 { 1253 len += r->overhead; 1254 1255 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1256 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1257 1258 return ((u64)len * r->mult) >> r->shift; 1259 } 1260 1261 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1262 const struct tc_ratespec *conf, 1263 u64 rate64); 1264 1265 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1266 const struct psched_ratecfg *r) 1267 { 1268 memset(res, 0, sizeof(*res)); 1269 1270 /* legacy struct tc_ratespec has a 32bit @rate field 1271 * Qdisc using 64bit rate should add new attributes 1272 * in order to maintain compatibility. 1273 */ 1274 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1275 1276 res->overhead = r->overhead; 1277 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1278 } 1279 1280 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1281 * The fast path only needs to access filter list and to update stats 1282 */ 1283 struct mini_Qdisc { 1284 struct tcf_proto *filter_list; 1285 struct gnet_stats_basic_cpu __percpu *cpu_bstats; 1286 struct gnet_stats_queue __percpu *cpu_qstats; 1287 struct rcu_head rcu; 1288 }; 1289 1290 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1291 const struct sk_buff *skb) 1292 { 1293 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1294 } 1295 1296 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1297 { 1298 this_cpu_inc(miniq->cpu_qstats->drops); 1299 } 1300 1301 struct mini_Qdisc_pair { 1302 struct mini_Qdisc miniq1; 1303 struct mini_Qdisc miniq2; 1304 struct mini_Qdisc __rcu **p_miniq; 1305 }; 1306 1307 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1308 struct tcf_proto *tp_head); 1309 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1310 struct mini_Qdisc __rcu **p_miniq); 1311 1312 static inline void skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res) 1313 { 1314 struct gnet_stats_queue *stats = res->qstats; 1315 int ret; 1316 1317 if (res->ingress) 1318 ret = netif_receive_skb(skb); 1319 else 1320 ret = dev_queue_xmit(skb); 1321 if (ret && stats) 1322 qstats_overlimit_inc(res->qstats); 1323 } 1324 1325 #endif 1326