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 }; 40 41 struct qdisc_size_table { 42 struct rcu_head rcu; 43 struct list_head list; 44 struct tc_sizespec szopts; 45 int refcnt; 46 u16 data[]; 47 }; 48 49 /* similar to sk_buff_head, but skb->prev pointer is undefined. */ 50 struct qdisc_skb_head { 51 struct sk_buff *head; 52 struct sk_buff *tail; 53 __u32 qlen; 54 spinlock_t lock; 55 }; 56 57 struct Qdisc { 58 int (*enqueue)(struct sk_buff *skb, 59 struct Qdisc *sch, 60 struct sk_buff **to_free); 61 struct sk_buff * (*dequeue)(struct Qdisc *sch); 62 unsigned int flags; 63 #define TCQ_F_BUILTIN 1 64 #define TCQ_F_INGRESS 2 65 #define TCQ_F_CAN_BYPASS 4 66 #define TCQ_F_MQROOT 8 67 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for 68 * q->dev_queue : It can test 69 * netif_xmit_frozen_or_stopped() before 70 * dequeueing next packet. 71 * Its true for MQ/MQPRIO slaves, or non 72 * multiqueue device. 73 */ 74 #define TCQ_F_WARN_NONWC (1 << 16) 75 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */ 76 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy : 77 * qdisc_tree_decrease_qlen() should stop. 78 */ 79 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */ 80 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */ 81 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */ 82 u32 limit; 83 const struct Qdisc_ops *ops; 84 struct qdisc_size_table __rcu *stab; 85 struct hlist_node hash; 86 u32 handle; 87 u32 parent; 88 89 struct netdev_queue *dev_queue; 90 91 struct net_rate_estimator __rcu *rate_est; 92 struct gnet_stats_basic_cpu __percpu *cpu_bstats; 93 struct gnet_stats_queue __percpu *cpu_qstats; 94 int padded; 95 refcount_t refcnt; 96 97 /* 98 * For performance sake on SMP, we put highly modified fields at the end 99 */ 100 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp; 101 struct qdisc_skb_head q; 102 struct gnet_stats_basic_packed bstats; 103 seqcount_t running; 104 struct gnet_stats_queue qstats; 105 unsigned long state; 106 struct Qdisc *next_sched; 107 struct sk_buff_head skb_bad_txq; 108 109 spinlock_t busylock ____cacheline_aligned_in_smp; 110 spinlock_t seqlock; 111 112 /* for NOLOCK qdisc, true if there are no enqueued skbs */ 113 bool empty; 114 struct rcu_head rcu; 115 }; 116 117 static inline void qdisc_refcount_inc(struct Qdisc *qdisc) 118 { 119 if (qdisc->flags & TCQ_F_BUILTIN) 120 return; 121 refcount_inc(&qdisc->refcnt); 122 } 123 124 /* Intended to be used by unlocked users, when concurrent qdisc release is 125 * possible. 126 */ 127 128 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc) 129 { 130 if (qdisc->flags & TCQ_F_BUILTIN) 131 return qdisc; 132 if (refcount_inc_not_zero(&qdisc->refcnt)) 133 return qdisc; 134 return NULL; 135 } 136 137 static inline bool qdisc_is_running(struct Qdisc *qdisc) 138 { 139 if (qdisc->flags & TCQ_F_NOLOCK) 140 return spin_is_locked(&qdisc->seqlock); 141 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false; 142 } 143 144 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) 145 { 146 return q->flags & TCQ_F_CPUSTATS; 147 } 148 149 static inline bool qdisc_is_empty(const struct Qdisc *qdisc) 150 { 151 if (qdisc_is_percpu_stats(qdisc)) 152 return READ_ONCE(qdisc->empty); 153 return !READ_ONCE(qdisc->q.qlen); 154 } 155 156 static inline bool qdisc_run_begin(struct Qdisc *qdisc) 157 { 158 if (qdisc->flags & TCQ_F_NOLOCK) { 159 if (!spin_trylock(&qdisc->seqlock)) 160 return false; 161 WRITE_ONCE(qdisc->empty, false); 162 } else if (qdisc_is_running(qdisc)) { 163 return false; 164 } 165 /* Variant of write_seqcount_begin() telling lockdep a trylock 166 * was attempted. 167 */ 168 raw_write_seqcount_begin(&qdisc->running); 169 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_); 170 return true; 171 } 172 173 static inline void qdisc_run_end(struct Qdisc *qdisc) 174 { 175 write_seqcount_end(&qdisc->running); 176 if (qdisc->flags & TCQ_F_NOLOCK) 177 spin_unlock(&qdisc->seqlock); 178 } 179 180 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 181 { 182 return qdisc->flags & TCQ_F_ONETXQUEUE; 183 } 184 185 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) 186 { 187 #ifdef CONFIG_BQL 188 /* Non-BQL migrated drivers will return 0, too. */ 189 return dql_avail(&txq->dql); 190 #else 191 return 0; 192 #endif 193 } 194 195 struct Qdisc_class_ops { 196 unsigned int flags; 197 /* Child qdisc manipulation */ 198 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 199 int (*graft)(struct Qdisc *, unsigned long cl, 200 struct Qdisc *, struct Qdisc **, 201 struct netlink_ext_ack *extack); 202 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 203 void (*qlen_notify)(struct Qdisc *, unsigned long); 204 205 /* Class manipulation routines */ 206 unsigned long (*find)(struct Qdisc *, u32 classid); 207 int (*change)(struct Qdisc *, u32, u32, 208 struct nlattr **, unsigned long *, 209 struct netlink_ext_ack *); 210 int (*delete)(struct Qdisc *, unsigned long); 211 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 212 213 /* Filter manipulation */ 214 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 215 unsigned long arg, 216 struct netlink_ext_ack *extack); 217 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 218 u32 classid); 219 void (*unbind_tcf)(struct Qdisc *, unsigned long); 220 221 /* rtnetlink specific */ 222 int (*dump)(struct Qdisc *, unsigned long, 223 struct sk_buff *skb, struct tcmsg*); 224 int (*dump_stats)(struct Qdisc *, unsigned long, 225 struct gnet_dump *); 226 }; 227 228 /* Qdisc_class_ops flag values */ 229 230 /* Implements API that doesn't require rtnl lock */ 231 enum qdisc_class_ops_flags { 232 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 233 }; 234 235 struct Qdisc_ops { 236 struct Qdisc_ops *next; 237 const struct Qdisc_class_ops *cl_ops; 238 char id[IFNAMSIZ]; 239 int priv_size; 240 unsigned int static_flags; 241 242 int (*enqueue)(struct sk_buff *skb, 243 struct Qdisc *sch, 244 struct sk_buff **to_free); 245 struct sk_buff * (*dequeue)(struct Qdisc *); 246 struct sk_buff * (*peek)(struct Qdisc *); 247 248 int (*init)(struct Qdisc *sch, struct nlattr *arg, 249 struct netlink_ext_ack *extack); 250 void (*reset)(struct Qdisc *); 251 void (*destroy)(struct Qdisc *); 252 int (*change)(struct Qdisc *sch, 253 struct nlattr *arg, 254 struct netlink_ext_ack *extack); 255 void (*attach)(struct Qdisc *sch); 256 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 257 258 int (*dump)(struct Qdisc *, struct sk_buff *); 259 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 260 261 void (*ingress_block_set)(struct Qdisc *sch, 262 u32 block_index); 263 void (*egress_block_set)(struct Qdisc *sch, 264 u32 block_index); 265 u32 (*ingress_block_get)(struct Qdisc *sch); 266 u32 (*egress_block_get)(struct Qdisc *sch); 267 268 struct module *owner; 269 }; 270 271 272 struct tcf_result { 273 union { 274 struct { 275 unsigned long class; 276 u32 classid; 277 }; 278 const struct tcf_proto *goto_tp; 279 280 /* used in the skb_tc_reinsert function */ 281 struct { 282 bool ingress; 283 struct gnet_stats_queue *qstats; 284 }; 285 }; 286 }; 287 288 struct tcf_chain; 289 290 struct tcf_proto_ops { 291 struct list_head head; 292 char kind[IFNAMSIZ]; 293 294 int (*classify)(struct sk_buff *, 295 const struct tcf_proto *, 296 struct tcf_result *); 297 int (*init)(struct tcf_proto*); 298 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 299 struct netlink_ext_ack *extack); 300 301 void* (*get)(struct tcf_proto*, u32 handle); 302 void (*put)(struct tcf_proto *tp, void *f); 303 int (*change)(struct net *net, struct sk_buff *, 304 struct tcf_proto*, unsigned long, 305 u32 handle, struct nlattr **, 306 void **, bool, bool, 307 struct netlink_ext_ack *); 308 int (*delete)(struct tcf_proto *tp, void *arg, 309 bool *last, bool rtnl_held, 310 struct netlink_ext_ack *); 311 bool (*delete_empty)(struct tcf_proto *tp); 312 void (*walk)(struct tcf_proto *tp, 313 struct tcf_walker *arg, bool rtnl_held); 314 int (*reoffload)(struct tcf_proto *tp, bool add, 315 flow_setup_cb_t *cb, void *cb_priv, 316 struct netlink_ext_ack *extack); 317 void (*hw_add)(struct tcf_proto *tp, 318 void *type_data); 319 void (*hw_del)(struct tcf_proto *tp, 320 void *type_data); 321 void (*bind_class)(void *, u32, unsigned long, 322 void *, unsigned long); 323 void * (*tmplt_create)(struct net *net, 324 struct tcf_chain *chain, 325 struct nlattr **tca, 326 struct netlink_ext_ack *extack); 327 void (*tmplt_destroy)(void *tmplt_priv); 328 329 /* rtnetlink specific */ 330 int (*dump)(struct net*, struct tcf_proto*, void *, 331 struct sk_buff *skb, struct tcmsg*, 332 bool); 333 int (*terse_dump)(struct net *net, 334 struct tcf_proto *tp, void *fh, 335 struct sk_buff *skb, 336 struct tcmsg *t, bool rtnl_held); 337 int (*tmplt_dump)(struct sk_buff *skb, 338 struct net *net, 339 void *tmplt_priv); 340 341 struct module *owner; 342 int flags; 343 }; 344 345 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 346 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 347 * conditions can occur when filters are inserted/deleted simultaneously. 348 */ 349 enum tcf_proto_ops_flags { 350 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 351 }; 352 353 struct tcf_proto { 354 /* Fast access part */ 355 struct tcf_proto __rcu *next; 356 void __rcu *root; 357 358 /* called under RCU BH lock*/ 359 int (*classify)(struct sk_buff *, 360 const struct tcf_proto *, 361 struct tcf_result *); 362 __be16 protocol; 363 364 /* All the rest */ 365 u32 prio; 366 void *data; 367 const struct tcf_proto_ops *ops; 368 struct tcf_chain *chain; 369 /* Lock protects tcf_proto shared state and can be used by unlocked 370 * classifiers to protect their private data. 371 */ 372 spinlock_t lock; 373 bool deleting; 374 refcount_t refcnt; 375 struct rcu_head rcu; 376 struct hlist_node destroy_ht_node; 377 }; 378 379 struct qdisc_skb_cb { 380 struct { 381 unsigned int pkt_len; 382 u16 slave_dev_queue_mapping; 383 u16 tc_classid; 384 }; 385 #define QDISC_CB_PRIV_LEN 20 386 unsigned char data[QDISC_CB_PRIV_LEN]; 387 u16 mru; 388 }; 389 390 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 391 392 struct tcf_chain { 393 /* Protects filter_chain. */ 394 struct mutex filter_chain_lock; 395 struct tcf_proto __rcu *filter_chain; 396 struct list_head list; 397 struct tcf_block *block; 398 u32 index; /* chain index */ 399 unsigned int refcnt; 400 unsigned int action_refcnt; 401 bool explicitly_created; 402 bool flushing; 403 const struct tcf_proto_ops *tmplt_ops; 404 void *tmplt_priv; 405 struct rcu_head rcu; 406 }; 407 408 struct tcf_block { 409 /* Lock protects tcf_block and lifetime-management data of chains 410 * attached to the block (refcnt, action_refcnt, explicitly_created). 411 */ 412 struct mutex lock; 413 struct list_head chain_list; 414 u32 index; /* block index for shared blocks */ 415 u32 classid; /* which class this block belongs to */ 416 refcount_t refcnt; 417 struct net *net; 418 struct Qdisc *q; 419 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 420 struct flow_block flow_block; 421 struct list_head owner_list; 422 bool keep_dst; 423 atomic_t offloadcnt; /* Number of oddloaded filters */ 424 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 425 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 426 struct { 427 struct tcf_chain *chain; 428 struct list_head filter_chain_list; 429 } chain0; 430 struct rcu_head rcu; 431 DECLARE_HASHTABLE(proto_destroy_ht, 7); 432 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 433 }; 434 435 #ifdef CONFIG_PROVE_LOCKING 436 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 437 { 438 return lockdep_is_held(&chain->filter_chain_lock); 439 } 440 441 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 442 { 443 return lockdep_is_held(&tp->lock); 444 } 445 #else 446 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain) 447 { 448 return true; 449 } 450 451 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 452 { 453 return true; 454 } 455 #endif /* #ifdef CONFIG_PROVE_LOCKING */ 456 457 #define tcf_chain_dereference(p, chain) \ 458 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 459 460 #define tcf_proto_dereference(p, tp) \ 461 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 462 463 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 464 { 465 struct qdisc_skb_cb *qcb; 466 467 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 468 BUILD_BUG_ON(sizeof(qcb->data) < sz); 469 } 470 471 static inline int qdisc_qlen_cpu(const struct Qdisc *q) 472 { 473 return this_cpu_ptr(q->cpu_qstats)->qlen; 474 } 475 476 static inline int qdisc_qlen(const struct Qdisc *q) 477 { 478 return q->q.qlen; 479 } 480 481 static inline int qdisc_qlen_sum(const struct Qdisc *q) 482 { 483 __u32 qlen = q->qstats.qlen; 484 int i; 485 486 if (qdisc_is_percpu_stats(q)) { 487 for_each_possible_cpu(i) 488 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 489 } else { 490 qlen += q->q.qlen; 491 } 492 493 return qlen; 494 } 495 496 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 497 { 498 return (struct qdisc_skb_cb *)skb->cb; 499 } 500 501 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 502 { 503 return &qdisc->q.lock; 504 } 505 506 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 507 { 508 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 509 510 return q; 511 } 512 513 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 514 { 515 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 516 } 517 518 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 519 { 520 return qdisc->dev_queue->qdisc_sleeping; 521 } 522 523 /* The qdisc root lock is a mechanism by which to top level 524 * of a qdisc tree can be locked from any qdisc node in the 525 * forest. This allows changing the configuration of some 526 * aspect of the qdisc tree while blocking out asynchronous 527 * qdisc access in the packet processing paths. 528 * 529 * It is only legal to do this when the root will not change 530 * on us. Otherwise we'll potentially lock the wrong qdisc 531 * root. This is enforced by holding the RTNL semaphore, which 532 * all users of this lock accessor must do. 533 */ 534 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 535 { 536 struct Qdisc *root = qdisc_root(qdisc); 537 538 ASSERT_RTNL(); 539 return qdisc_lock(root); 540 } 541 542 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 543 { 544 struct Qdisc *root = qdisc_root_sleeping(qdisc); 545 546 ASSERT_RTNL(); 547 return qdisc_lock(root); 548 } 549 550 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc) 551 { 552 struct Qdisc *root = qdisc_root_sleeping(qdisc); 553 554 ASSERT_RTNL(); 555 return &root->running; 556 } 557 558 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 559 { 560 return qdisc->dev_queue->dev; 561 } 562 563 static inline void sch_tree_lock(const struct Qdisc *q) 564 { 565 spin_lock_bh(qdisc_root_sleeping_lock(q)); 566 } 567 568 static inline void sch_tree_unlock(const struct Qdisc *q) 569 { 570 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 571 } 572 573 extern struct Qdisc noop_qdisc; 574 extern struct Qdisc_ops noop_qdisc_ops; 575 extern struct Qdisc_ops pfifo_fast_ops; 576 extern struct Qdisc_ops mq_qdisc_ops; 577 extern struct Qdisc_ops noqueue_qdisc_ops; 578 extern const struct Qdisc_ops *default_qdisc_ops; 579 static inline const struct Qdisc_ops * 580 get_default_qdisc_ops(const struct net_device *dev, int ntx) 581 { 582 return ntx < dev->real_num_tx_queues ? 583 default_qdisc_ops : &pfifo_fast_ops; 584 } 585 586 struct Qdisc_class_common { 587 u32 classid; 588 struct hlist_node hnode; 589 }; 590 591 struct Qdisc_class_hash { 592 struct hlist_head *hash; 593 unsigned int hashsize; 594 unsigned int hashmask; 595 unsigned int hashelems; 596 }; 597 598 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 599 { 600 id ^= id >> 8; 601 id ^= id >> 4; 602 return id & mask; 603 } 604 605 static inline struct Qdisc_class_common * 606 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 607 { 608 struct Qdisc_class_common *cl; 609 unsigned int h; 610 611 if (!id) 612 return NULL; 613 614 h = qdisc_class_hash(id, hash->hashmask); 615 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 616 if (cl->classid == id) 617 return cl; 618 } 619 return NULL; 620 } 621 622 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 623 { 624 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 625 626 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 627 } 628 629 int qdisc_class_hash_init(struct Qdisc_class_hash *); 630 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 631 struct Qdisc_class_common *); 632 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 633 struct Qdisc_class_common *); 634 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 635 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 636 637 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 638 void dev_init_scheduler(struct net_device *dev); 639 void dev_shutdown(struct net_device *dev); 640 void dev_activate(struct net_device *dev); 641 void dev_deactivate(struct net_device *dev); 642 void dev_deactivate_many(struct list_head *head); 643 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 644 struct Qdisc *qdisc); 645 void qdisc_reset(struct Qdisc *qdisc); 646 void qdisc_put(struct Qdisc *qdisc); 647 void qdisc_put_unlocked(struct Qdisc *qdisc); 648 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 649 #ifdef CONFIG_NET_SCHED 650 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 651 void *type_data); 652 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 653 struct Qdisc *new, struct Qdisc *old, 654 enum tc_setup_type type, void *type_data, 655 struct netlink_ext_ack *extack); 656 #else 657 static inline int 658 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 659 void *type_data) 660 { 661 q->flags &= ~TCQ_F_OFFLOADED; 662 return 0; 663 } 664 665 static inline void 666 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 667 struct Qdisc *new, struct Qdisc *old, 668 enum tc_setup_type type, void *type_data, 669 struct netlink_ext_ack *extack) 670 { 671 } 672 #endif 673 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 674 const struct Qdisc_ops *ops, 675 struct netlink_ext_ack *extack); 676 void qdisc_free(struct Qdisc *qdisc); 677 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 678 const struct Qdisc_ops *ops, u32 parentid, 679 struct netlink_ext_ack *extack); 680 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 681 const struct qdisc_size_table *stab); 682 int skb_do_redirect(struct sk_buff *); 683 684 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 685 { 686 #ifdef CONFIG_NET_CLS_ACT 687 return skb->tc_at_ingress; 688 #else 689 return false; 690 #endif 691 } 692 693 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 694 { 695 #ifdef CONFIG_NET_CLS_ACT 696 if (skb->tc_skip_classify) { 697 skb->tc_skip_classify = 0; 698 return true; 699 } 700 #endif 701 return false; 702 } 703 704 /* Reset all TX qdiscs greater than index of a device. */ 705 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 706 { 707 struct Qdisc *qdisc; 708 709 for (; i < dev->num_tx_queues; i++) { 710 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 711 if (qdisc) { 712 spin_lock_bh(qdisc_lock(qdisc)); 713 qdisc_reset(qdisc); 714 spin_unlock_bh(qdisc_lock(qdisc)); 715 } 716 } 717 } 718 719 /* Are all TX queues of the device empty? */ 720 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 721 { 722 unsigned int i; 723 724 rcu_read_lock(); 725 for (i = 0; i < dev->num_tx_queues; i++) { 726 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 727 const struct Qdisc *q = rcu_dereference(txq->qdisc); 728 729 if (!qdisc_is_empty(q)) { 730 rcu_read_unlock(); 731 return false; 732 } 733 } 734 rcu_read_unlock(); 735 return true; 736 } 737 738 /* Are any of the TX qdiscs changing? */ 739 static inline bool qdisc_tx_changing(const struct net_device *dev) 740 { 741 unsigned int i; 742 743 for (i = 0; i < dev->num_tx_queues; i++) { 744 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 745 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) 746 return true; 747 } 748 return false; 749 } 750 751 /* Is the device using the noop qdisc on all queues? */ 752 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 753 { 754 unsigned int i; 755 756 for (i = 0; i < dev->num_tx_queues; i++) { 757 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 758 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 759 return false; 760 } 761 return true; 762 } 763 764 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 765 { 766 return qdisc_skb_cb(skb)->pkt_len; 767 } 768 769 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 770 enum net_xmit_qdisc_t { 771 __NET_XMIT_STOLEN = 0x00010000, 772 __NET_XMIT_BYPASS = 0x00020000, 773 }; 774 775 #ifdef CONFIG_NET_CLS_ACT 776 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 777 #else 778 #define net_xmit_drop_count(e) (1) 779 #endif 780 781 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 782 const struct Qdisc *sch) 783 { 784 #ifdef CONFIG_NET_SCHED 785 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 786 787 if (stab) 788 __qdisc_calculate_pkt_len(skb, stab); 789 #endif 790 } 791 792 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 793 struct sk_buff **to_free) 794 { 795 qdisc_calculate_pkt_len(skb, sch); 796 return sch->enqueue(skb, sch, to_free); 797 } 798 799 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats, 800 __u64 bytes, __u32 packets) 801 { 802 bstats->bytes += bytes; 803 bstats->packets += packets; 804 } 805 806 static inline void bstats_update(struct gnet_stats_basic_packed *bstats, 807 const struct sk_buff *skb) 808 { 809 _bstats_update(bstats, 810 qdisc_pkt_len(skb), 811 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 812 } 813 814 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 815 __u64 bytes, __u32 packets) 816 { 817 u64_stats_update_begin(&bstats->syncp); 818 _bstats_update(&bstats->bstats, bytes, packets); 819 u64_stats_update_end(&bstats->syncp); 820 } 821 822 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 823 const struct sk_buff *skb) 824 { 825 u64_stats_update_begin(&bstats->syncp); 826 bstats_update(&bstats->bstats, skb); 827 u64_stats_update_end(&bstats->syncp); 828 } 829 830 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 831 const struct sk_buff *skb) 832 { 833 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb); 834 } 835 836 static inline void qdisc_bstats_update(struct Qdisc *sch, 837 const struct sk_buff *skb) 838 { 839 bstats_update(&sch->bstats, skb); 840 } 841 842 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 843 const struct sk_buff *skb) 844 { 845 sch->qstats.backlog -= qdisc_pkt_len(skb); 846 } 847 848 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 849 const struct sk_buff *skb) 850 { 851 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 852 } 853 854 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 855 const struct sk_buff *skb) 856 { 857 sch->qstats.backlog += qdisc_pkt_len(skb); 858 } 859 860 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 861 const struct sk_buff *skb) 862 { 863 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 864 } 865 866 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 867 { 868 this_cpu_inc(sch->cpu_qstats->qlen); 869 } 870 871 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 872 { 873 this_cpu_dec(sch->cpu_qstats->qlen); 874 } 875 876 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 877 { 878 this_cpu_inc(sch->cpu_qstats->requeues); 879 } 880 881 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 882 { 883 sch->qstats.drops += count; 884 } 885 886 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 887 { 888 qstats->drops++; 889 } 890 891 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 892 { 893 qstats->overlimits++; 894 } 895 896 static inline void qdisc_qstats_drop(struct Qdisc *sch) 897 { 898 qstats_drop_inc(&sch->qstats); 899 } 900 901 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 902 { 903 this_cpu_inc(sch->cpu_qstats->drops); 904 } 905 906 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 907 { 908 sch->qstats.overlimits++; 909 } 910 911 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 912 { 913 __u32 qlen = qdisc_qlen_sum(sch); 914 915 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 916 } 917 918 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 919 __u32 *backlog) 920 { 921 struct gnet_stats_queue qstats = { 0 }; 922 __u32 len = qdisc_qlen_sum(sch); 923 924 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len); 925 *qlen = qstats.qlen; 926 *backlog = qstats.backlog; 927 } 928 929 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 930 { 931 __u32 qlen, backlog; 932 933 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 934 qdisc_tree_reduce_backlog(sch, qlen, backlog); 935 } 936 937 static inline void qdisc_purge_queue(struct Qdisc *sch) 938 { 939 __u32 qlen, backlog; 940 941 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 942 qdisc_reset(sch); 943 qdisc_tree_reduce_backlog(sch, qlen, backlog); 944 } 945 946 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh) 947 { 948 qh->head = NULL; 949 qh->tail = NULL; 950 qh->qlen = 0; 951 } 952 953 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 954 struct qdisc_skb_head *qh) 955 { 956 struct sk_buff *last = qh->tail; 957 958 if (last) { 959 skb->next = NULL; 960 last->next = skb; 961 qh->tail = skb; 962 } else { 963 qh->tail = skb; 964 qh->head = skb; 965 } 966 qh->qlen++; 967 } 968 969 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 970 { 971 __qdisc_enqueue_tail(skb, &sch->q); 972 qdisc_qstats_backlog_inc(sch, skb); 973 return NET_XMIT_SUCCESS; 974 } 975 976 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 977 struct qdisc_skb_head *qh) 978 { 979 skb->next = qh->head; 980 981 if (!qh->head) 982 qh->tail = skb; 983 qh->head = skb; 984 qh->qlen++; 985 } 986 987 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 988 { 989 struct sk_buff *skb = qh->head; 990 991 if (likely(skb != NULL)) { 992 qh->head = skb->next; 993 qh->qlen--; 994 if (qh->head == NULL) 995 qh->tail = NULL; 996 skb->next = NULL; 997 } 998 999 return skb; 1000 } 1001 1002 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1003 { 1004 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1005 1006 if (likely(skb != NULL)) { 1007 qdisc_qstats_backlog_dec(sch, skb); 1008 qdisc_bstats_update(sch, skb); 1009 } 1010 1011 return skb; 1012 } 1013 1014 /* Instead of calling kfree_skb() while root qdisc lock is held, 1015 * queue the skb for future freeing at end of __dev_xmit_skb() 1016 */ 1017 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1018 { 1019 skb->next = *to_free; 1020 *to_free = skb; 1021 } 1022 1023 static inline void __qdisc_drop_all(struct sk_buff *skb, 1024 struct sk_buff **to_free) 1025 { 1026 if (skb->prev) 1027 skb->prev->next = *to_free; 1028 else 1029 skb->next = *to_free; 1030 *to_free = skb; 1031 } 1032 1033 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1034 struct qdisc_skb_head *qh, 1035 struct sk_buff **to_free) 1036 { 1037 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1038 1039 if (likely(skb != NULL)) { 1040 unsigned int len = qdisc_pkt_len(skb); 1041 1042 qdisc_qstats_backlog_dec(sch, skb); 1043 __qdisc_drop(skb, to_free); 1044 return len; 1045 } 1046 1047 return 0; 1048 } 1049 1050 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch, 1051 struct sk_buff **to_free) 1052 { 1053 return __qdisc_queue_drop_head(sch, &sch->q, to_free); 1054 } 1055 1056 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1057 { 1058 const struct qdisc_skb_head *qh = &sch->q; 1059 1060 return qh->head; 1061 } 1062 1063 /* generic pseudo peek method for non-work-conserving qdisc */ 1064 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1065 { 1066 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1067 1068 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1069 if (!skb) { 1070 skb = sch->dequeue(sch); 1071 1072 if (skb) { 1073 __skb_queue_head(&sch->gso_skb, skb); 1074 /* it's still part of the queue */ 1075 qdisc_qstats_backlog_inc(sch, skb); 1076 sch->q.qlen++; 1077 } 1078 } 1079 1080 return skb; 1081 } 1082 1083 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1084 struct sk_buff *skb) 1085 { 1086 if (qdisc_is_percpu_stats(sch)) { 1087 qdisc_qstats_cpu_backlog_dec(sch, skb); 1088 qdisc_bstats_cpu_update(sch, skb); 1089 qdisc_qstats_cpu_qlen_dec(sch); 1090 } else { 1091 qdisc_qstats_backlog_dec(sch, skb); 1092 qdisc_bstats_update(sch, skb); 1093 sch->q.qlen--; 1094 } 1095 } 1096 1097 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1098 unsigned int pkt_len) 1099 { 1100 if (qdisc_is_percpu_stats(sch)) { 1101 qdisc_qstats_cpu_qlen_inc(sch); 1102 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1103 } else { 1104 sch->qstats.backlog += pkt_len; 1105 sch->q.qlen++; 1106 } 1107 } 1108 1109 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1110 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1111 { 1112 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1113 1114 if (skb) { 1115 skb = __skb_dequeue(&sch->gso_skb); 1116 if (qdisc_is_percpu_stats(sch)) { 1117 qdisc_qstats_cpu_backlog_dec(sch, skb); 1118 qdisc_qstats_cpu_qlen_dec(sch); 1119 } else { 1120 qdisc_qstats_backlog_dec(sch, skb); 1121 sch->q.qlen--; 1122 } 1123 } else { 1124 skb = sch->dequeue(sch); 1125 } 1126 1127 return skb; 1128 } 1129 1130 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1131 { 1132 /* 1133 * We do not know the backlog in bytes of this list, it 1134 * is up to the caller to correct it 1135 */ 1136 ASSERT_RTNL(); 1137 if (qh->qlen) { 1138 rtnl_kfree_skbs(qh->head, qh->tail); 1139 1140 qh->head = NULL; 1141 qh->tail = NULL; 1142 qh->qlen = 0; 1143 } 1144 } 1145 1146 static inline void qdisc_reset_queue(struct Qdisc *sch) 1147 { 1148 __qdisc_reset_queue(&sch->q); 1149 sch->qstats.backlog = 0; 1150 } 1151 1152 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1153 struct Qdisc **pold) 1154 { 1155 struct Qdisc *old; 1156 1157 sch_tree_lock(sch); 1158 old = *pold; 1159 *pold = new; 1160 if (old != NULL) 1161 qdisc_tree_flush_backlog(old); 1162 sch_tree_unlock(sch); 1163 1164 return old; 1165 } 1166 1167 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1168 { 1169 rtnl_kfree_skbs(skb, skb); 1170 qdisc_qstats_drop(sch); 1171 } 1172 1173 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1174 struct sk_buff **to_free) 1175 { 1176 __qdisc_drop(skb, to_free); 1177 qdisc_qstats_cpu_drop(sch); 1178 1179 return NET_XMIT_DROP; 1180 } 1181 1182 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1183 struct sk_buff **to_free) 1184 { 1185 __qdisc_drop(skb, to_free); 1186 qdisc_qstats_drop(sch); 1187 1188 return NET_XMIT_DROP; 1189 } 1190 1191 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1192 struct sk_buff **to_free) 1193 { 1194 __qdisc_drop_all(skb, to_free); 1195 qdisc_qstats_drop(sch); 1196 1197 return NET_XMIT_DROP; 1198 } 1199 1200 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1201 long it will take to send a packet given its size. 1202 */ 1203 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1204 { 1205 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1206 if (slot < 0) 1207 slot = 0; 1208 slot >>= rtab->rate.cell_log; 1209 if (slot > 255) 1210 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1211 return rtab->data[slot]; 1212 } 1213 1214 struct psched_ratecfg { 1215 u64 rate_bytes_ps; /* bytes per second */ 1216 u32 mult; 1217 u16 overhead; 1218 u8 linklayer; 1219 u8 shift; 1220 }; 1221 1222 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1223 unsigned int len) 1224 { 1225 len += r->overhead; 1226 1227 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1228 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1229 1230 return ((u64)len * r->mult) >> r->shift; 1231 } 1232 1233 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1234 const struct tc_ratespec *conf, 1235 u64 rate64); 1236 1237 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1238 const struct psched_ratecfg *r) 1239 { 1240 memset(res, 0, sizeof(*res)); 1241 1242 /* legacy struct tc_ratespec has a 32bit @rate field 1243 * Qdisc using 64bit rate should add new attributes 1244 * in order to maintain compatibility. 1245 */ 1246 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1247 1248 res->overhead = r->overhead; 1249 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1250 } 1251 1252 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1253 * The fast path only needs to access filter list and to update stats 1254 */ 1255 struct mini_Qdisc { 1256 struct tcf_proto *filter_list; 1257 struct tcf_block *block; 1258 struct gnet_stats_basic_cpu __percpu *cpu_bstats; 1259 struct gnet_stats_queue __percpu *cpu_qstats; 1260 struct rcu_head rcu; 1261 }; 1262 1263 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1264 const struct sk_buff *skb) 1265 { 1266 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1267 } 1268 1269 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1270 { 1271 this_cpu_inc(miniq->cpu_qstats->drops); 1272 } 1273 1274 struct mini_Qdisc_pair { 1275 struct mini_Qdisc miniq1; 1276 struct mini_Qdisc miniq2; 1277 struct mini_Qdisc __rcu **p_miniq; 1278 }; 1279 1280 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1281 struct tcf_proto *tp_head); 1282 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1283 struct mini_Qdisc __rcu **p_miniq); 1284 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1285 struct tcf_block *block); 1286 1287 static inline int skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res) 1288 { 1289 return res->ingress ? netif_receive_skb(skb) : dev_queue_xmit(skb); 1290 } 1291 1292 #endif 1293