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