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 struct netlink_ext_ack *); 215 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 216 217 /* Filter manipulation */ 218 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 219 unsigned long arg, 220 struct netlink_ext_ack *extack); 221 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 222 u32 classid); 223 void (*unbind_tcf)(struct Qdisc *, unsigned long); 224 225 /* rtnetlink specific */ 226 int (*dump)(struct Qdisc *, unsigned long, 227 struct sk_buff *skb, struct tcmsg*); 228 int (*dump_stats)(struct Qdisc *, unsigned long, 229 struct gnet_dump *); 230 }; 231 232 /* Qdisc_class_ops flag values */ 233 234 /* Implements API that doesn't require rtnl lock */ 235 enum qdisc_class_ops_flags { 236 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 237 }; 238 239 struct Qdisc_ops { 240 struct Qdisc_ops *next; 241 const struct Qdisc_class_ops *cl_ops; 242 char id[IFNAMSIZ]; 243 int priv_size; 244 unsigned int static_flags; 245 246 int (*enqueue)(struct sk_buff *skb, 247 struct Qdisc *sch, 248 struct sk_buff **to_free); 249 struct sk_buff * (*dequeue)(struct Qdisc *); 250 struct sk_buff * (*peek)(struct Qdisc *); 251 252 int (*init)(struct Qdisc *sch, struct nlattr *arg, 253 struct netlink_ext_ack *extack); 254 void (*reset)(struct Qdisc *); 255 void (*destroy)(struct Qdisc *); 256 int (*change)(struct Qdisc *sch, 257 struct nlattr *arg, 258 struct netlink_ext_ack *extack); 259 void (*attach)(struct Qdisc *sch); 260 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 261 262 int (*dump)(struct Qdisc *, struct sk_buff *); 263 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 264 265 void (*ingress_block_set)(struct Qdisc *sch, 266 u32 block_index); 267 void (*egress_block_set)(struct Qdisc *sch, 268 u32 block_index); 269 u32 (*ingress_block_get)(struct Qdisc *sch); 270 u32 (*egress_block_get)(struct Qdisc *sch); 271 272 struct module *owner; 273 }; 274 275 276 struct tcf_result { 277 union { 278 struct { 279 unsigned long class; 280 u32 classid; 281 }; 282 const struct tcf_proto *goto_tp; 283 284 /* used in the skb_tc_reinsert function */ 285 struct { 286 bool ingress; 287 struct gnet_stats_queue *qstats; 288 }; 289 }; 290 }; 291 292 struct tcf_chain; 293 294 struct tcf_proto_ops { 295 struct list_head head; 296 char kind[IFNAMSIZ]; 297 298 int (*classify)(struct sk_buff *, 299 const struct tcf_proto *, 300 struct tcf_result *); 301 int (*init)(struct tcf_proto*); 302 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 303 struct netlink_ext_ack *extack); 304 305 void* (*get)(struct tcf_proto*, u32 handle); 306 void (*put)(struct tcf_proto *tp, void *f); 307 int (*change)(struct net *net, struct sk_buff *, 308 struct tcf_proto*, unsigned long, 309 u32 handle, struct nlattr **, 310 void **, bool, bool, 311 struct netlink_ext_ack *); 312 int (*delete)(struct tcf_proto *tp, void *arg, 313 bool *last, bool rtnl_held, 314 struct netlink_ext_ack *); 315 bool (*delete_empty)(struct tcf_proto *tp); 316 void (*walk)(struct tcf_proto *tp, 317 struct tcf_walker *arg, bool rtnl_held); 318 int (*reoffload)(struct tcf_proto *tp, bool add, 319 flow_setup_cb_t *cb, void *cb_priv, 320 struct netlink_ext_ack *extack); 321 void (*hw_add)(struct tcf_proto *tp, 322 void *type_data); 323 void (*hw_del)(struct tcf_proto *tp, 324 void *type_data); 325 void (*bind_class)(void *, u32, unsigned long, 326 void *, unsigned long); 327 void * (*tmplt_create)(struct net *net, 328 struct tcf_chain *chain, 329 struct nlattr **tca, 330 struct netlink_ext_ack *extack); 331 void (*tmplt_destroy)(void *tmplt_priv); 332 333 /* rtnetlink specific */ 334 int (*dump)(struct net*, struct tcf_proto*, void *, 335 struct sk_buff *skb, struct tcmsg*, 336 bool); 337 int (*terse_dump)(struct net *net, 338 struct tcf_proto *tp, void *fh, 339 struct sk_buff *skb, 340 struct tcmsg *t, bool rtnl_held); 341 int (*tmplt_dump)(struct sk_buff *skb, 342 struct net *net, 343 void *tmplt_priv); 344 345 struct module *owner; 346 int flags; 347 }; 348 349 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 350 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 351 * conditions can occur when filters are inserted/deleted simultaneously. 352 */ 353 enum tcf_proto_ops_flags { 354 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 355 }; 356 357 struct tcf_proto { 358 /* Fast access part */ 359 struct tcf_proto __rcu *next; 360 void __rcu *root; 361 362 /* called under RCU BH lock*/ 363 int (*classify)(struct sk_buff *, 364 const struct tcf_proto *, 365 struct tcf_result *); 366 __be16 protocol; 367 368 /* All the rest */ 369 u32 prio; 370 void *data; 371 const struct tcf_proto_ops *ops; 372 struct tcf_chain *chain; 373 /* Lock protects tcf_proto shared state and can be used by unlocked 374 * classifiers to protect their private data. 375 */ 376 spinlock_t lock; 377 bool deleting; 378 refcount_t refcnt; 379 struct rcu_head rcu; 380 struct hlist_node destroy_ht_node; 381 }; 382 383 struct qdisc_skb_cb { 384 struct { 385 unsigned int pkt_len; 386 u16 slave_dev_queue_mapping; 387 u16 tc_classid; 388 }; 389 #define QDISC_CB_PRIV_LEN 20 390 unsigned char data[QDISC_CB_PRIV_LEN]; 391 u16 mru; 392 bool post_ct; 393 }; 394 395 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 396 397 struct tcf_chain { 398 /* Protects filter_chain. */ 399 struct mutex filter_chain_lock; 400 struct tcf_proto __rcu *filter_chain; 401 struct list_head list; 402 struct tcf_block *block; 403 u32 index; /* chain index */ 404 unsigned int refcnt; 405 unsigned int action_refcnt; 406 bool explicitly_created; 407 bool flushing; 408 const struct tcf_proto_ops *tmplt_ops; 409 void *tmplt_priv; 410 struct rcu_head rcu; 411 }; 412 413 struct tcf_block { 414 /* Lock protects tcf_block and lifetime-management data of chains 415 * attached to the block (refcnt, action_refcnt, explicitly_created). 416 */ 417 struct mutex lock; 418 struct list_head chain_list; 419 u32 index; /* block index for shared blocks */ 420 u32 classid; /* which class this block belongs to */ 421 refcount_t refcnt; 422 struct net *net; 423 struct Qdisc *q; 424 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 425 struct flow_block flow_block; 426 struct list_head owner_list; 427 bool keep_dst; 428 atomic_t offloadcnt; /* Number of oddloaded filters */ 429 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 430 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 431 struct { 432 struct tcf_chain *chain; 433 struct list_head filter_chain_list; 434 } chain0; 435 struct rcu_head rcu; 436 DECLARE_HASHTABLE(proto_destroy_ht, 7); 437 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 438 }; 439 440 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 441 { 442 return lockdep_is_held(&chain->filter_chain_lock); 443 } 444 445 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 446 { 447 return lockdep_is_held(&tp->lock); 448 } 449 450 #define tcf_chain_dereference(p, chain) \ 451 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 452 453 #define tcf_proto_dereference(p, tp) \ 454 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 455 456 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 457 { 458 struct qdisc_skb_cb *qcb; 459 460 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 461 BUILD_BUG_ON(sizeof(qcb->data) < sz); 462 } 463 464 static inline int qdisc_qlen_cpu(const struct Qdisc *q) 465 { 466 return this_cpu_ptr(q->cpu_qstats)->qlen; 467 } 468 469 static inline int qdisc_qlen(const struct Qdisc *q) 470 { 471 return q->q.qlen; 472 } 473 474 static inline int qdisc_qlen_sum(const struct Qdisc *q) 475 { 476 __u32 qlen = q->qstats.qlen; 477 int i; 478 479 if (qdisc_is_percpu_stats(q)) { 480 for_each_possible_cpu(i) 481 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 482 } else { 483 qlen += q->q.qlen; 484 } 485 486 return qlen; 487 } 488 489 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 490 { 491 return (struct qdisc_skb_cb *)skb->cb; 492 } 493 494 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 495 { 496 return &qdisc->q.lock; 497 } 498 499 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 500 { 501 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 502 503 return q; 504 } 505 506 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 507 { 508 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 509 } 510 511 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 512 { 513 return qdisc->dev_queue->qdisc_sleeping; 514 } 515 516 /* The qdisc root lock is a mechanism by which to top level 517 * of a qdisc tree can be locked from any qdisc node in the 518 * forest. This allows changing the configuration of some 519 * aspect of the qdisc tree while blocking out asynchronous 520 * qdisc access in the packet processing paths. 521 * 522 * It is only legal to do this when the root will not change 523 * on us. Otherwise we'll potentially lock the wrong qdisc 524 * root. This is enforced by holding the RTNL semaphore, which 525 * all users of this lock accessor must do. 526 */ 527 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 528 { 529 struct Qdisc *root = qdisc_root(qdisc); 530 531 ASSERT_RTNL(); 532 return qdisc_lock(root); 533 } 534 535 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 536 { 537 struct Qdisc *root = qdisc_root_sleeping(qdisc); 538 539 ASSERT_RTNL(); 540 return qdisc_lock(root); 541 } 542 543 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc) 544 { 545 struct Qdisc *root = qdisc_root_sleeping(qdisc); 546 547 ASSERT_RTNL(); 548 return &root->running; 549 } 550 551 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 552 { 553 return qdisc->dev_queue->dev; 554 } 555 556 static inline void sch_tree_lock(struct Qdisc *q) 557 { 558 if (q->flags & TCQ_F_MQROOT) 559 spin_lock_bh(qdisc_lock(q)); 560 else 561 spin_lock_bh(qdisc_root_sleeping_lock(q)); 562 } 563 564 static inline void sch_tree_unlock(struct Qdisc *q) 565 { 566 if (q->flags & TCQ_F_MQROOT) 567 spin_unlock_bh(qdisc_lock(q)); 568 else 569 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 570 } 571 572 extern struct Qdisc noop_qdisc; 573 extern struct Qdisc_ops noop_qdisc_ops; 574 extern struct Qdisc_ops pfifo_fast_ops; 575 extern struct Qdisc_ops mq_qdisc_ops; 576 extern struct Qdisc_ops noqueue_qdisc_ops; 577 extern const struct Qdisc_ops *default_qdisc_ops; 578 static inline const struct Qdisc_ops * 579 get_default_qdisc_ops(const struct net_device *dev, int ntx) 580 { 581 return ntx < dev->real_num_tx_queues ? 582 default_qdisc_ops : &pfifo_fast_ops; 583 } 584 585 struct Qdisc_class_common { 586 u32 classid; 587 struct hlist_node hnode; 588 }; 589 590 struct Qdisc_class_hash { 591 struct hlist_head *hash; 592 unsigned int hashsize; 593 unsigned int hashmask; 594 unsigned int hashelems; 595 }; 596 597 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 598 { 599 id ^= id >> 8; 600 id ^= id >> 4; 601 return id & mask; 602 } 603 604 static inline struct Qdisc_class_common * 605 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 606 { 607 struct Qdisc_class_common *cl; 608 unsigned int h; 609 610 if (!id) 611 return NULL; 612 613 h = qdisc_class_hash(id, hash->hashmask); 614 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 615 if (cl->classid == id) 616 return cl; 617 } 618 return NULL; 619 } 620 621 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 622 { 623 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 624 625 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 626 } 627 628 int qdisc_class_hash_init(struct Qdisc_class_hash *); 629 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 630 struct Qdisc_class_common *); 631 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 632 struct Qdisc_class_common *); 633 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 634 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 635 636 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 637 void dev_init_scheduler(struct net_device *dev); 638 void dev_shutdown(struct net_device *dev); 639 void dev_activate(struct net_device *dev); 640 void dev_deactivate(struct net_device *dev); 641 void dev_deactivate_many(struct list_head *head); 642 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 643 struct Qdisc *qdisc); 644 void qdisc_reset(struct Qdisc *qdisc); 645 void qdisc_put(struct Qdisc *qdisc); 646 void qdisc_put_unlocked(struct Qdisc *qdisc); 647 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 648 #ifdef CONFIG_NET_SCHED 649 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 650 void *type_data); 651 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 652 struct Qdisc *new, struct Qdisc *old, 653 enum tc_setup_type type, void *type_data, 654 struct netlink_ext_ack *extack); 655 #else 656 static inline int 657 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 658 void *type_data) 659 { 660 q->flags &= ~TCQ_F_OFFLOADED; 661 return 0; 662 } 663 664 static inline void 665 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 666 struct Qdisc *new, struct Qdisc *old, 667 enum tc_setup_type type, void *type_data, 668 struct netlink_ext_ack *extack) 669 { 670 } 671 #endif 672 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 673 const struct Qdisc_ops *ops, 674 struct netlink_ext_ack *extack); 675 void qdisc_free(struct Qdisc *qdisc); 676 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 677 const struct Qdisc_ops *ops, u32 parentid, 678 struct netlink_ext_ack *extack); 679 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 680 const struct qdisc_size_table *stab); 681 int skb_do_redirect(struct sk_buff *); 682 683 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 684 { 685 #ifdef CONFIG_NET_CLS_ACT 686 return skb->tc_at_ingress; 687 #else 688 return false; 689 #endif 690 } 691 692 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 693 { 694 #ifdef CONFIG_NET_CLS_ACT 695 if (skb->tc_skip_classify) { 696 skb->tc_skip_classify = 0; 697 return true; 698 } 699 #endif 700 return false; 701 } 702 703 /* Reset all TX qdiscs greater than index of a device. */ 704 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 705 { 706 struct Qdisc *qdisc; 707 708 for (; i < dev->num_tx_queues; i++) { 709 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 710 if (qdisc) { 711 spin_lock_bh(qdisc_lock(qdisc)); 712 qdisc_reset(qdisc); 713 spin_unlock_bh(qdisc_lock(qdisc)); 714 } 715 } 716 } 717 718 /* Are all TX queues of the device empty? */ 719 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 720 { 721 unsigned int i; 722 723 rcu_read_lock(); 724 for (i = 0; i < dev->num_tx_queues; i++) { 725 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 726 const struct Qdisc *q = rcu_dereference(txq->qdisc); 727 728 if (!qdisc_is_empty(q)) { 729 rcu_read_unlock(); 730 return false; 731 } 732 } 733 rcu_read_unlock(); 734 return true; 735 } 736 737 /* Are any of the TX qdiscs changing? */ 738 static inline bool qdisc_tx_changing(const struct net_device *dev) 739 { 740 unsigned int i; 741 742 for (i = 0; i < dev->num_tx_queues; i++) { 743 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 744 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) 745 return true; 746 } 747 return false; 748 } 749 750 /* Is the device using the noop qdisc on all queues? */ 751 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 752 { 753 unsigned int i; 754 755 for (i = 0; i < dev->num_tx_queues; i++) { 756 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 757 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 758 return false; 759 } 760 return true; 761 } 762 763 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 764 { 765 return qdisc_skb_cb(skb)->pkt_len; 766 } 767 768 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 769 enum net_xmit_qdisc_t { 770 __NET_XMIT_STOLEN = 0x00010000, 771 __NET_XMIT_BYPASS = 0x00020000, 772 }; 773 774 #ifdef CONFIG_NET_CLS_ACT 775 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 776 #else 777 #define net_xmit_drop_count(e) (1) 778 #endif 779 780 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 781 const struct Qdisc *sch) 782 { 783 #ifdef CONFIG_NET_SCHED 784 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 785 786 if (stab) 787 __qdisc_calculate_pkt_len(skb, stab); 788 #endif 789 } 790 791 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 792 struct sk_buff **to_free) 793 { 794 qdisc_calculate_pkt_len(skb, sch); 795 return sch->enqueue(skb, sch, to_free); 796 } 797 798 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats, 799 __u64 bytes, __u32 packets) 800 { 801 bstats->bytes += bytes; 802 bstats->packets += packets; 803 } 804 805 static inline void bstats_update(struct gnet_stats_basic_packed *bstats, 806 const struct sk_buff *skb) 807 { 808 _bstats_update(bstats, 809 qdisc_pkt_len(skb), 810 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 811 } 812 813 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 814 __u64 bytes, __u32 packets) 815 { 816 u64_stats_update_begin(&bstats->syncp); 817 _bstats_update(&bstats->bstats, bytes, packets); 818 u64_stats_update_end(&bstats->syncp); 819 } 820 821 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 822 const struct sk_buff *skb) 823 { 824 u64_stats_update_begin(&bstats->syncp); 825 bstats_update(&bstats->bstats, skb); 826 u64_stats_update_end(&bstats->syncp); 827 } 828 829 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 830 const struct sk_buff *skb) 831 { 832 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb); 833 } 834 835 static inline void qdisc_bstats_update(struct Qdisc *sch, 836 const struct sk_buff *skb) 837 { 838 bstats_update(&sch->bstats, skb); 839 } 840 841 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 842 const struct sk_buff *skb) 843 { 844 sch->qstats.backlog -= qdisc_pkt_len(skb); 845 } 846 847 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 848 const struct sk_buff *skb) 849 { 850 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 851 } 852 853 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 854 const struct sk_buff *skb) 855 { 856 sch->qstats.backlog += qdisc_pkt_len(skb); 857 } 858 859 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 860 const struct sk_buff *skb) 861 { 862 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 863 } 864 865 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 866 { 867 this_cpu_inc(sch->cpu_qstats->qlen); 868 } 869 870 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 871 { 872 this_cpu_dec(sch->cpu_qstats->qlen); 873 } 874 875 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 876 { 877 this_cpu_inc(sch->cpu_qstats->requeues); 878 } 879 880 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 881 { 882 sch->qstats.drops += count; 883 } 884 885 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 886 { 887 qstats->drops++; 888 } 889 890 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 891 { 892 qstats->overlimits++; 893 } 894 895 static inline void qdisc_qstats_drop(struct Qdisc *sch) 896 { 897 qstats_drop_inc(&sch->qstats); 898 } 899 900 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 901 { 902 this_cpu_inc(sch->cpu_qstats->drops); 903 } 904 905 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 906 { 907 sch->qstats.overlimits++; 908 } 909 910 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 911 { 912 __u32 qlen = qdisc_qlen_sum(sch); 913 914 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 915 } 916 917 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 918 __u32 *backlog) 919 { 920 struct gnet_stats_queue qstats = { 0 }; 921 __u32 len = qdisc_qlen_sum(sch); 922 923 __gnet_stats_copy_queue(&qstats, sch->cpu_qstats, &sch->qstats, len); 924 *qlen = qstats.qlen; 925 *backlog = qstats.backlog; 926 } 927 928 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 929 { 930 __u32 qlen, backlog; 931 932 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 933 qdisc_tree_reduce_backlog(sch, qlen, backlog); 934 } 935 936 static inline void qdisc_purge_queue(struct Qdisc *sch) 937 { 938 __u32 qlen, backlog; 939 940 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 941 qdisc_reset(sch); 942 qdisc_tree_reduce_backlog(sch, qlen, backlog); 943 } 944 945 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh) 946 { 947 qh->head = NULL; 948 qh->tail = NULL; 949 qh->qlen = 0; 950 } 951 952 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 953 struct qdisc_skb_head *qh) 954 { 955 struct sk_buff *last = qh->tail; 956 957 if (last) { 958 skb->next = NULL; 959 last->next = skb; 960 qh->tail = skb; 961 } else { 962 qh->tail = skb; 963 qh->head = skb; 964 } 965 qh->qlen++; 966 } 967 968 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 969 { 970 __qdisc_enqueue_tail(skb, &sch->q); 971 qdisc_qstats_backlog_inc(sch, skb); 972 return NET_XMIT_SUCCESS; 973 } 974 975 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 976 struct qdisc_skb_head *qh) 977 { 978 skb->next = qh->head; 979 980 if (!qh->head) 981 qh->tail = skb; 982 qh->head = skb; 983 qh->qlen++; 984 } 985 986 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 987 { 988 struct sk_buff *skb = qh->head; 989 990 if (likely(skb != NULL)) { 991 qh->head = skb->next; 992 qh->qlen--; 993 if (qh->head == NULL) 994 qh->tail = NULL; 995 skb->next = NULL; 996 } 997 998 return skb; 999 } 1000 1001 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1002 { 1003 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1004 1005 if (likely(skb != NULL)) { 1006 qdisc_qstats_backlog_dec(sch, skb); 1007 qdisc_bstats_update(sch, skb); 1008 } 1009 1010 return skb; 1011 } 1012 1013 /* Instead of calling kfree_skb() while root qdisc lock is held, 1014 * queue the skb for future freeing at end of __dev_xmit_skb() 1015 */ 1016 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1017 { 1018 skb->next = *to_free; 1019 *to_free = skb; 1020 } 1021 1022 static inline void __qdisc_drop_all(struct sk_buff *skb, 1023 struct sk_buff **to_free) 1024 { 1025 if (skb->prev) 1026 skb->prev->next = *to_free; 1027 else 1028 skb->next = *to_free; 1029 *to_free = skb; 1030 } 1031 1032 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1033 struct qdisc_skb_head *qh, 1034 struct sk_buff **to_free) 1035 { 1036 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1037 1038 if (likely(skb != NULL)) { 1039 unsigned int len = qdisc_pkt_len(skb); 1040 1041 qdisc_qstats_backlog_dec(sch, skb); 1042 __qdisc_drop(skb, to_free); 1043 return len; 1044 } 1045 1046 return 0; 1047 } 1048 1049 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1050 { 1051 const struct qdisc_skb_head *qh = &sch->q; 1052 1053 return qh->head; 1054 } 1055 1056 /* generic pseudo peek method for non-work-conserving qdisc */ 1057 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1058 { 1059 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1060 1061 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1062 if (!skb) { 1063 skb = sch->dequeue(sch); 1064 1065 if (skb) { 1066 __skb_queue_head(&sch->gso_skb, skb); 1067 /* it's still part of the queue */ 1068 qdisc_qstats_backlog_inc(sch, skb); 1069 sch->q.qlen++; 1070 } 1071 } 1072 1073 return skb; 1074 } 1075 1076 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1077 struct sk_buff *skb) 1078 { 1079 if (qdisc_is_percpu_stats(sch)) { 1080 qdisc_qstats_cpu_backlog_dec(sch, skb); 1081 qdisc_bstats_cpu_update(sch, skb); 1082 qdisc_qstats_cpu_qlen_dec(sch); 1083 } else { 1084 qdisc_qstats_backlog_dec(sch, skb); 1085 qdisc_bstats_update(sch, skb); 1086 sch->q.qlen--; 1087 } 1088 } 1089 1090 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1091 unsigned int pkt_len) 1092 { 1093 if (qdisc_is_percpu_stats(sch)) { 1094 qdisc_qstats_cpu_qlen_inc(sch); 1095 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1096 } else { 1097 sch->qstats.backlog += pkt_len; 1098 sch->q.qlen++; 1099 } 1100 } 1101 1102 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1103 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1104 { 1105 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1106 1107 if (skb) { 1108 skb = __skb_dequeue(&sch->gso_skb); 1109 if (qdisc_is_percpu_stats(sch)) { 1110 qdisc_qstats_cpu_backlog_dec(sch, skb); 1111 qdisc_qstats_cpu_qlen_dec(sch); 1112 } else { 1113 qdisc_qstats_backlog_dec(sch, skb); 1114 sch->q.qlen--; 1115 } 1116 } else { 1117 skb = sch->dequeue(sch); 1118 } 1119 1120 return skb; 1121 } 1122 1123 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1124 { 1125 /* 1126 * We do not know the backlog in bytes of this list, it 1127 * is up to the caller to correct it 1128 */ 1129 ASSERT_RTNL(); 1130 if (qh->qlen) { 1131 rtnl_kfree_skbs(qh->head, qh->tail); 1132 1133 qh->head = NULL; 1134 qh->tail = NULL; 1135 qh->qlen = 0; 1136 } 1137 } 1138 1139 static inline void qdisc_reset_queue(struct Qdisc *sch) 1140 { 1141 __qdisc_reset_queue(&sch->q); 1142 sch->qstats.backlog = 0; 1143 } 1144 1145 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1146 struct Qdisc **pold) 1147 { 1148 struct Qdisc *old; 1149 1150 sch_tree_lock(sch); 1151 old = *pold; 1152 *pold = new; 1153 if (old != NULL) 1154 qdisc_purge_queue(old); 1155 sch_tree_unlock(sch); 1156 1157 return old; 1158 } 1159 1160 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1161 { 1162 rtnl_kfree_skbs(skb, skb); 1163 qdisc_qstats_drop(sch); 1164 } 1165 1166 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1167 struct sk_buff **to_free) 1168 { 1169 __qdisc_drop(skb, to_free); 1170 qdisc_qstats_cpu_drop(sch); 1171 1172 return NET_XMIT_DROP; 1173 } 1174 1175 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1176 struct sk_buff **to_free) 1177 { 1178 __qdisc_drop(skb, to_free); 1179 qdisc_qstats_drop(sch); 1180 1181 return NET_XMIT_DROP; 1182 } 1183 1184 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1185 struct sk_buff **to_free) 1186 { 1187 __qdisc_drop_all(skb, to_free); 1188 qdisc_qstats_drop(sch); 1189 1190 return NET_XMIT_DROP; 1191 } 1192 1193 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1194 long it will take to send a packet given its size. 1195 */ 1196 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1197 { 1198 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1199 if (slot < 0) 1200 slot = 0; 1201 slot >>= rtab->rate.cell_log; 1202 if (slot > 255) 1203 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1204 return rtab->data[slot]; 1205 } 1206 1207 struct psched_ratecfg { 1208 u64 rate_bytes_ps; /* bytes per second */ 1209 u32 mult; 1210 u16 overhead; 1211 u8 linklayer; 1212 u8 shift; 1213 }; 1214 1215 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1216 unsigned int len) 1217 { 1218 len += r->overhead; 1219 1220 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1221 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1222 1223 return ((u64)len * r->mult) >> r->shift; 1224 } 1225 1226 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1227 const struct tc_ratespec *conf, 1228 u64 rate64); 1229 1230 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1231 const struct psched_ratecfg *r) 1232 { 1233 memset(res, 0, sizeof(*res)); 1234 1235 /* legacy struct tc_ratespec has a 32bit @rate field 1236 * Qdisc using 64bit rate should add new attributes 1237 * in order to maintain compatibility. 1238 */ 1239 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1240 1241 res->overhead = r->overhead; 1242 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1243 } 1244 1245 struct psched_pktrate { 1246 u64 rate_pkts_ps; /* packets per second */ 1247 u32 mult; 1248 u8 shift; 1249 }; 1250 1251 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1252 unsigned int pkt_num) 1253 { 1254 return ((u64)pkt_num * r->mult) >> r->shift; 1255 } 1256 1257 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1258 1259 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1260 * The fast path only needs to access filter list and to update stats 1261 */ 1262 struct mini_Qdisc { 1263 struct tcf_proto *filter_list; 1264 struct tcf_block *block; 1265 struct gnet_stats_basic_cpu __percpu *cpu_bstats; 1266 struct gnet_stats_queue __percpu *cpu_qstats; 1267 struct rcu_head rcu; 1268 }; 1269 1270 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1271 const struct sk_buff *skb) 1272 { 1273 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1274 } 1275 1276 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1277 { 1278 this_cpu_inc(miniq->cpu_qstats->drops); 1279 } 1280 1281 struct mini_Qdisc_pair { 1282 struct mini_Qdisc miniq1; 1283 struct mini_Qdisc miniq2; 1284 struct mini_Qdisc __rcu **p_miniq; 1285 }; 1286 1287 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1288 struct tcf_proto *tp_head); 1289 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1290 struct mini_Qdisc __rcu **p_miniq); 1291 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1292 struct tcf_block *block); 1293 1294 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1295 1296 #endif 1297