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 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 439 { 440 return lockdep_is_held(&chain->filter_chain_lock); 441 } 442 443 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 444 { 445 return lockdep_is_held(&tp->lock); 446 } 447 448 #define tcf_chain_dereference(p, chain) \ 449 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 450 451 #define tcf_proto_dereference(p, tp) \ 452 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 453 454 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 455 { 456 struct qdisc_skb_cb *qcb; 457 458 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 459 BUILD_BUG_ON(sizeof(qcb->data) < sz); 460 } 461 462 static inline int qdisc_qlen_cpu(const struct Qdisc *q) 463 { 464 return this_cpu_ptr(q->cpu_qstats)->qlen; 465 } 466 467 static inline int qdisc_qlen(const struct Qdisc *q) 468 { 469 return q->q.qlen; 470 } 471 472 static inline int qdisc_qlen_sum(const struct Qdisc *q) 473 { 474 __u32 qlen = q->qstats.qlen; 475 int i; 476 477 if (qdisc_is_percpu_stats(q)) { 478 for_each_possible_cpu(i) 479 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 480 } else { 481 qlen += q->q.qlen; 482 } 483 484 return qlen; 485 } 486 487 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 488 { 489 return (struct qdisc_skb_cb *)skb->cb; 490 } 491 492 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 493 { 494 return &qdisc->q.lock; 495 } 496 497 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 498 { 499 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 500 501 return q; 502 } 503 504 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 505 { 506 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 507 } 508 509 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 510 { 511 return qdisc->dev_queue->qdisc_sleeping; 512 } 513 514 /* The qdisc root lock is a mechanism by which to top level 515 * of a qdisc tree can be locked from any qdisc node in the 516 * forest. This allows changing the configuration of some 517 * aspect of the qdisc tree while blocking out asynchronous 518 * qdisc access in the packet processing paths. 519 * 520 * It is only legal to do this when the root will not change 521 * on us. Otherwise we'll potentially lock the wrong qdisc 522 * root. This is enforced by holding the RTNL semaphore, which 523 * all users of this lock accessor must do. 524 */ 525 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 526 { 527 struct Qdisc *root = qdisc_root(qdisc); 528 529 ASSERT_RTNL(); 530 return qdisc_lock(root); 531 } 532 533 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 534 { 535 struct Qdisc *root = qdisc_root_sleeping(qdisc); 536 537 ASSERT_RTNL(); 538 return qdisc_lock(root); 539 } 540 541 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc) 542 { 543 struct Qdisc *root = qdisc_root_sleeping(qdisc); 544 545 ASSERT_RTNL(); 546 return &root->running; 547 } 548 549 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 550 { 551 return qdisc->dev_queue->dev; 552 } 553 554 static inline void sch_tree_lock(const struct Qdisc *q) 555 { 556 spin_lock_bh(qdisc_root_sleeping_lock(q)); 557 } 558 559 static inline void sch_tree_unlock(const struct Qdisc *q) 560 { 561 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 562 } 563 564 extern struct Qdisc noop_qdisc; 565 extern struct Qdisc_ops noop_qdisc_ops; 566 extern struct Qdisc_ops pfifo_fast_ops; 567 extern struct Qdisc_ops mq_qdisc_ops; 568 extern struct Qdisc_ops noqueue_qdisc_ops; 569 extern const struct Qdisc_ops *default_qdisc_ops; 570 static inline const struct Qdisc_ops * 571 get_default_qdisc_ops(const struct net_device *dev, int ntx) 572 { 573 return ntx < dev->real_num_tx_queues ? 574 default_qdisc_ops : &pfifo_fast_ops; 575 } 576 577 struct Qdisc_class_common { 578 u32 classid; 579 struct hlist_node hnode; 580 }; 581 582 struct Qdisc_class_hash { 583 struct hlist_head *hash; 584 unsigned int hashsize; 585 unsigned int hashmask; 586 unsigned int hashelems; 587 }; 588 589 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 590 { 591 id ^= id >> 8; 592 id ^= id >> 4; 593 return id & mask; 594 } 595 596 static inline struct Qdisc_class_common * 597 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 598 { 599 struct Qdisc_class_common *cl; 600 unsigned int h; 601 602 if (!id) 603 return NULL; 604 605 h = qdisc_class_hash(id, hash->hashmask); 606 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 607 if (cl->classid == id) 608 return cl; 609 } 610 return NULL; 611 } 612 613 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 614 { 615 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 616 617 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 618 } 619 620 int qdisc_class_hash_init(struct Qdisc_class_hash *); 621 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 622 struct Qdisc_class_common *); 623 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 624 struct Qdisc_class_common *); 625 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 626 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 627 628 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 629 void dev_init_scheduler(struct net_device *dev); 630 void dev_shutdown(struct net_device *dev); 631 void dev_activate(struct net_device *dev); 632 void dev_deactivate(struct net_device *dev); 633 void dev_deactivate_many(struct list_head *head); 634 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 635 struct Qdisc *qdisc); 636 void qdisc_reset(struct Qdisc *qdisc); 637 void qdisc_put(struct Qdisc *qdisc); 638 void qdisc_put_unlocked(struct Qdisc *qdisc); 639 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 640 #ifdef CONFIG_NET_SCHED 641 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 642 void *type_data); 643 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 644 struct Qdisc *new, struct Qdisc *old, 645 enum tc_setup_type type, void *type_data, 646 struct netlink_ext_ack *extack); 647 #else 648 static inline int 649 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 650 void *type_data) 651 { 652 q->flags &= ~TCQ_F_OFFLOADED; 653 return 0; 654 } 655 656 static inline void 657 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 658 struct Qdisc *new, struct Qdisc *old, 659 enum tc_setup_type type, void *type_data, 660 struct netlink_ext_ack *extack) 661 { 662 } 663 #endif 664 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 665 const struct Qdisc_ops *ops, 666 struct netlink_ext_ack *extack); 667 void qdisc_free(struct Qdisc *qdisc); 668 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 669 const struct Qdisc_ops *ops, u32 parentid, 670 struct netlink_ext_ack *extack); 671 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 672 const struct qdisc_size_table *stab); 673 int skb_do_redirect(struct sk_buff *); 674 675 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 676 { 677 #ifdef CONFIG_NET_CLS_ACT 678 return skb->tc_at_ingress; 679 #else 680 return false; 681 #endif 682 } 683 684 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 685 { 686 #ifdef CONFIG_NET_CLS_ACT 687 if (skb->tc_skip_classify) { 688 skb->tc_skip_classify = 0; 689 return true; 690 } 691 #endif 692 return false; 693 } 694 695 /* Reset all TX qdiscs greater than index of a device. */ 696 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 697 { 698 struct Qdisc *qdisc; 699 700 for (; i < dev->num_tx_queues; i++) { 701 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 702 if (qdisc) { 703 spin_lock_bh(qdisc_lock(qdisc)); 704 qdisc_reset(qdisc); 705 spin_unlock_bh(qdisc_lock(qdisc)); 706 } 707 } 708 } 709 710 /* Are all TX queues of the device empty? */ 711 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 712 { 713 unsigned int i; 714 715 rcu_read_lock(); 716 for (i = 0; i < dev->num_tx_queues; i++) { 717 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 718 const struct Qdisc *q = rcu_dereference(txq->qdisc); 719 720 if (!qdisc_is_empty(q)) { 721 rcu_read_unlock(); 722 return false; 723 } 724 } 725 rcu_read_unlock(); 726 return true; 727 } 728 729 /* Are any of the TX qdiscs changing? */ 730 static inline bool qdisc_tx_changing(const struct net_device *dev) 731 { 732 unsigned int i; 733 734 for (i = 0; i < dev->num_tx_queues; i++) { 735 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 736 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) 737 return true; 738 } 739 return false; 740 } 741 742 /* Is the device using the noop qdisc on all queues? */ 743 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 744 { 745 unsigned int i; 746 747 for (i = 0; i < dev->num_tx_queues; i++) { 748 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 749 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 750 return false; 751 } 752 return true; 753 } 754 755 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 756 { 757 return qdisc_skb_cb(skb)->pkt_len; 758 } 759 760 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 761 enum net_xmit_qdisc_t { 762 __NET_XMIT_STOLEN = 0x00010000, 763 __NET_XMIT_BYPASS = 0x00020000, 764 }; 765 766 #ifdef CONFIG_NET_CLS_ACT 767 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 768 #else 769 #define net_xmit_drop_count(e) (1) 770 #endif 771 772 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 773 const struct Qdisc *sch) 774 { 775 #ifdef CONFIG_NET_SCHED 776 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 777 778 if (stab) 779 __qdisc_calculate_pkt_len(skb, stab); 780 #endif 781 } 782 783 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 784 struct sk_buff **to_free) 785 { 786 qdisc_calculate_pkt_len(skb, sch); 787 return sch->enqueue(skb, sch, to_free); 788 } 789 790 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats, 791 __u64 bytes, __u32 packets) 792 { 793 bstats->bytes += bytes; 794 bstats->packets += packets; 795 } 796 797 static inline void bstats_update(struct gnet_stats_basic_packed *bstats, 798 const struct sk_buff *skb) 799 { 800 _bstats_update(bstats, 801 qdisc_pkt_len(skb), 802 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 803 } 804 805 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 806 __u64 bytes, __u32 packets) 807 { 808 u64_stats_update_begin(&bstats->syncp); 809 _bstats_update(&bstats->bstats, bytes, packets); 810 u64_stats_update_end(&bstats->syncp); 811 } 812 813 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 814 const struct sk_buff *skb) 815 { 816 u64_stats_update_begin(&bstats->syncp); 817 bstats_update(&bstats->bstats, skb); 818 u64_stats_update_end(&bstats->syncp); 819 } 820 821 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 822 const struct sk_buff *skb) 823 { 824 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb); 825 } 826 827 static inline void qdisc_bstats_update(struct Qdisc *sch, 828 const struct sk_buff *skb) 829 { 830 bstats_update(&sch->bstats, skb); 831 } 832 833 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 834 const struct sk_buff *skb) 835 { 836 sch->qstats.backlog -= qdisc_pkt_len(skb); 837 } 838 839 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 840 const struct sk_buff *skb) 841 { 842 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 843 } 844 845 static inline void qdisc_qstats_backlog_inc(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_inc(struct Qdisc *sch, 852 const struct sk_buff *skb) 853 { 854 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 855 } 856 857 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 858 { 859 this_cpu_inc(sch->cpu_qstats->qlen); 860 } 861 862 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 863 { 864 this_cpu_dec(sch->cpu_qstats->qlen); 865 } 866 867 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 868 { 869 this_cpu_inc(sch->cpu_qstats->requeues); 870 } 871 872 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 873 { 874 sch->qstats.drops += count; 875 } 876 877 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 878 { 879 qstats->drops++; 880 } 881 882 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 883 { 884 qstats->overlimits++; 885 } 886 887 static inline void qdisc_qstats_drop(struct Qdisc *sch) 888 { 889 qstats_drop_inc(&sch->qstats); 890 } 891 892 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 893 { 894 this_cpu_inc(sch->cpu_qstats->drops); 895 } 896 897 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 898 { 899 sch->qstats.overlimits++; 900 } 901 902 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 903 { 904 __u32 qlen = qdisc_qlen_sum(sch); 905 906 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 907 } 908 909 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 910 __u32 *backlog) 911 { 912 struct gnet_stats_queue qstats = { 0 }; 913 __u32 len = qdisc_qlen_sum(sch); 914 915 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len); 916 *qlen = qstats.qlen; 917 *backlog = qstats.backlog; 918 } 919 920 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 921 { 922 __u32 qlen, backlog; 923 924 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 925 qdisc_tree_reduce_backlog(sch, qlen, backlog); 926 } 927 928 static inline void qdisc_purge_queue(struct Qdisc *sch) 929 { 930 __u32 qlen, backlog; 931 932 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 933 qdisc_reset(sch); 934 qdisc_tree_reduce_backlog(sch, qlen, backlog); 935 } 936 937 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh) 938 { 939 qh->head = NULL; 940 qh->tail = NULL; 941 qh->qlen = 0; 942 } 943 944 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 945 struct qdisc_skb_head *qh) 946 { 947 struct sk_buff *last = qh->tail; 948 949 if (last) { 950 skb->next = NULL; 951 last->next = skb; 952 qh->tail = skb; 953 } else { 954 qh->tail = skb; 955 qh->head = skb; 956 } 957 qh->qlen++; 958 } 959 960 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 961 { 962 __qdisc_enqueue_tail(skb, &sch->q); 963 qdisc_qstats_backlog_inc(sch, skb); 964 return NET_XMIT_SUCCESS; 965 } 966 967 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 968 struct qdisc_skb_head *qh) 969 { 970 skb->next = qh->head; 971 972 if (!qh->head) 973 qh->tail = skb; 974 qh->head = skb; 975 qh->qlen++; 976 } 977 978 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 979 { 980 struct sk_buff *skb = qh->head; 981 982 if (likely(skb != NULL)) { 983 qh->head = skb->next; 984 qh->qlen--; 985 if (qh->head == NULL) 986 qh->tail = NULL; 987 skb->next = NULL; 988 } 989 990 return skb; 991 } 992 993 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 994 { 995 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 996 997 if (likely(skb != NULL)) { 998 qdisc_qstats_backlog_dec(sch, skb); 999 qdisc_bstats_update(sch, skb); 1000 } 1001 1002 return skb; 1003 } 1004 1005 /* Instead of calling kfree_skb() while root qdisc lock is held, 1006 * queue the skb for future freeing at end of __dev_xmit_skb() 1007 */ 1008 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1009 { 1010 skb->next = *to_free; 1011 *to_free = skb; 1012 } 1013 1014 static inline void __qdisc_drop_all(struct sk_buff *skb, 1015 struct sk_buff **to_free) 1016 { 1017 if (skb->prev) 1018 skb->prev->next = *to_free; 1019 else 1020 skb->next = *to_free; 1021 *to_free = skb; 1022 } 1023 1024 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1025 struct qdisc_skb_head *qh, 1026 struct sk_buff **to_free) 1027 { 1028 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1029 1030 if (likely(skb != NULL)) { 1031 unsigned int len = qdisc_pkt_len(skb); 1032 1033 qdisc_qstats_backlog_dec(sch, skb); 1034 __qdisc_drop(skb, to_free); 1035 return len; 1036 } 1037 1038 return 0; 1039 } 1040 1041 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1042 { 1043 const struct qdisc_skb_head *qh = &sch->q; 1044 1045 return qh->head; 1046 } 1047 1048 /* generic pseudo peek method for non-work-conserving qdisc */ 1049 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1050 { 1051 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1052 1053 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1054 if (!skb) { 1055 skb = sch->dequeue(sch); 1056 1057 if (skb) { 1058 __skb_queue_head(&sch->gso_skb, skb); 1059 /* it's still part of the queue */ 1060 qdisc_qstats_backlog_inc(sch, skb); 1061 sch->q.qlen++; 1062 } 1063 } 1064 1065 return skb; 1066 } 1067 1068 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1069 struct sk_buff *skb) 1070 { 1071 if (qdisc_is_percpu_stats(sch)) { 1072 qdisc_qstats_cpu_backlog_dec(sch, skb); 1073 qdisc_bstats_cpu_update(sch, skb); 1074 qdisc_qstats_cpu_qlen_dec(sch); 1075 } else { 1076 qdisc_qstats_backlog_dec(sch, skb); 1077 qdisc_bstats_update(sch, skb); 1078 sch->q.qlen--; 1079 } 1080 } 1081 1082 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1083 unsigned int pkt_len) 1084 { 1085 if (qdisc_is_percpu_stats(sch)) { 1086 qdisc_qstats_cpu_qlen_inc(sch); 1087 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1088 } else { 1089 sch->qstats.backlog += pkt_len; 1090 sch->q.qlen++; 1091 } 1092 } 1093 1094 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1095 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1096 { 1097 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1098 1099 if (skb) { 1100 skb = __skb_dequeue(&sch->gso_skb); 1101 if (qdisc_is_percpu_stats(sch)) { 1102 qdisc_qstats_cpu_backlog_dec(sch, skb); 1103 qdisc_qstats_cpu_qlen_dec(sch); 1104 } else { 1105 qdisc_qstats_backlog_dec(sch, skb); 1106 sch->q.qlen--; 1107 } 1108 } else { 1109 skb = sch->dequeue(sch); 1110 } 1111 1112 return skb; 1113 } 1114 1115 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1116 { 1117 /* 1118 * We do not know the backlog in bytes of this list, it 1119 * is up to the caller to correct it 1120 */ 1121 ASSERT_RTNL(); 1122 if (qh->qlen) { 1123 rtnl_kfree_skbs(qh->head, qh->tail); 1124 1125 qh->head = NULL; 1126 qh->tail = NULL; 1127 qh->qlen = 0; 1128 } 1129 } 1130 1131 static inline void qdisc_reset_queue(struct Qdisc *sch) 1132 { 1133 __qdisc_reset_queue(&sch->q); 1134 sch->qstats.backlog = 0; 1135 } 1136 1137 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1138 struct Qdisc **pold) 1139 { 1140 struct Qdisc *old; 1141 1142 sch_tree_lock(sch); 1143 old = *pold; 1144 *pold = new; 1145 if (old != NULL) 1146 qdisc_tree_flush_backlog(old); 1147 sch_tree_unlock(sch); 1148 1149 return old; 1150 } 1151 1152 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1153 { 1154 rtnl_kfree_skbs(skb, skb); 1155 qdisc_qstats_drop(sch); 1156 } 1157 1158 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1159 struct sk_buff **to_free) 1160 { 1161 __qdisc_drop(skb, to_free); 1162 qdisc_qstats_cpu_drop(sch); 1163 1164 return NET_XMIT_DROP; 1165 } 1166 1167 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1168 struct sk_buff **to_free) 1169 { 1170 __qdisc_drop(skb, to_free); 1171 qdisc_qstats_drop(sch); 1172 1173 return NET_XMIT_DROP; 1174 } 1175 1176 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1177 struct sk_buff **to_free) 1178 { 1179 __qdisc_drop_all(skb, to_free); 1180 qdisc_qstats_drop(sch); 1181 1182 return NET_XMIT_DROP; 1183 } 1184 1185 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1186 long it will take to send a packet given its size. 1187 */ 1188 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1189 { 1190 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1191 if (slot < 0) 1192 slot = 0; 1193 slot >>= rtab->rate.cell_log; 1194 if (slot > 255) 1195 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1196 return rtab->data[slot]; 1197 } 1198 1199 struct psched_ratecfg { 1200 u64 rate_bytes_ps; /* bytes per second */ 1201 u32 mult; 1202 u16 overhead; 1203 u8 linklayer; 1204 u8 shift; 1205 }; 1206 1207 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1208 unsigned int len) 1209 { 1210 len += r->overhead; 1211 1212 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1213 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1214 1215 return ((u64)len * r->mult) >> r->shift; 1216 } 1217 1218 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1219 const struct tc_ratespec *conf, 1220 u64 rate64); 1221 1222 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1223 const struct psched_ratecfg *r) 1224 { 1225 memset(res, 0, sizeof(*res)); 1226 1227 /* legacy struct tc_ratespec has a 32bit @rate field 1228 * Qdisc using 64bit rate should add new attributes 1229 * in order to maintain compatibility. 1230 */ 1231 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1232 1233 res->overhead = r->overhead; 1234 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1235 } 1236 1237 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1238 * The fast path only needs to access filter list and to update stats 1239 */ 1240 struct mini_Qdisc { 1241 struct tcf_proto *filter_list; 1242 struct tcf_block *block; 1243 struct gnet_stats_basic_cpu __percpu *cpu_bstats; 1244 struct gnet_stats_queue __percpu *cpu_qstats; 1245 struct rcu_head rcu; 1246 }; 1247 1248 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1249 const struct sk_buff *skb) 1250 { 1251 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1252 } 1253 1254 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1255 { 1256 this_cpu_inc(miniq->cpu_qstats->drops); 1257 } 1258 1259 struct mini_Qdisc_pair { 1260 struct mini_Qdisc miniq1; 1261 struct mini_Qdisc miniq2; 1262 struct mini_Qdisc __rcu **p_miniq; 1263 }; 1264 1265 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1266 struct tcf_proto *tp_head); 1267 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1268 struct mini_Qdisc __rcu **p_miniq); 1269 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1270 struct tcf_block *block); 1271 1272 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1273 1274 #endif 1275