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 __QDISC_STATE_MISSED, 40 __QDISC_STATE_DRAINING, 41 }; 42 43 enum qdisc_state2_t { 44 /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly. 45 * Use qdisc_run_begin/end() or qdisc_is_running() instead. 46 */ 47 __QDISC_STATE2_RUNNING, 48 }; 49 50 #define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED) 51 #define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING) 52 53 #define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \ 54 QDISC_STATE_DRAINING) 55 56 struct qdisc_size_table { 57 struct rcu_head rcu; 58 struct list_head list; 59 struct tc_sizespec szopts; 60 int refcnt; 61 u16 data[]; 62 }; 63 64 /* similar to sk_buff_head, but skb->prev pointer is undefined. */ 65 struct qdisc_skb_head { 66 struct sk_buff *head; 67 struct sk_buff *tail; 68 __u32 qlen; 69 spinlock_t lock; 70 }; 71 72 struct Qdisc { 73 int (*enqueue)(struct sk_buff *skb, 74 struct Qdisc *sch, 75 struct sk_buff **to_free); 76 struct sk_buff * (*dequeue)(struct Qdisc *sch); 77 unsigned int flags; 78 #define TCQ_F_BUILTIN 1 79 #define TCQ_F_INGRESS 2 80 #define TCQ_F_CAN_BYPASS 4 81 #define TCQ_F_MQROOT 8 82 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for 83 * q->dev_queue : It can test 84 * netif_xmit_frozen_or_stopped() before 85 * dequeueing next packet. 86 * Its true for MQ/MQPRIO slaves, or non 87 * multiqueue device. 88 */ 89 #define TCQ_F_WARN_NONWC (1 << 16) 90 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */ 91 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy : 92 * qdisc_tree_decrease_qlen() should stop. 93 */ 94 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */ 95 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */ 96 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */ 97 u32 limit; 98 const struct Qdisc_ops *ops; 99 struct qdisc_size_table __rcu *stab; 100 struct hlist_node hash; 101 u32 handle; 102 u32 parent; 103 104 struct netdev_queue *dev_queue; 105 106 struct net_rate_estimator __rcu *rate_est; 107 struct gnet_stats_basic_sync __percpu *cpu_bstats; 108 struct gnet_stats_queue __percpu *cpu_qstats; 109 int pad; 110 refcount_t refcnt; 111 112 /* 113 * For performance sake on SMP, we put highly modified fields at the end 114 */ 115 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp; 116 struct qdisc_skb_head q; 117 struct gnet_stats_basic_sync bstats; 118 struct gnet_stats_queue qstats; 119 unsigned long state; 120 unsigned long state2; /* must be written under qdisc spinlock */ 121 struct Qdisc *next_sched; 122 struct sk_buff_head skb_bad_txq; 123 124 spinlock_t busylock ____cacheline_aligned_in_smp; 125 spinlock_t seqlock; 126 127 struct rcu_head rcu; 128 netdevice_tracker dev_tracker; 129 /* private data */ 130 long privdata[] ____cacheline_aligned; 131 }; 132 133 static inline void qdisc_refcount_inc(struct Qdisc *qdisc) 134 { 135 if (qdisc->flags & TCQ_F_BUILTIN) 136 return; 137 refcount_inc(&qdisc->refcnt); 138 } 139 140 static inline bool qdisc_refcount_dec_if_one(struct Qdisc *qdisc) 141 { 142 if (qdisc->flags & TCQ_F_BUILTIN) 143 return true; 144 return refcount_dec_if_one(&qdisc->refcnt); 145 } 146 147 /* Intended to be used by unlocked users, when concurrent qdisc release is 148 * possible. 149 */ 150 151 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc) 152 { 153 if (qdisc->flags & TCQ_F_BUILTIN) 154 return qdisc; 155 if (refcount_inc_not_zero(&qdisc->refcnt)) 156 return qdisc; 157 return NULL; 158 } 159 160 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc 161 * root_lock section, or provide their own memory barriers -- ordering 162 * against qdisc_run_begin/end() atomic bit operations. 163 */ 164 static inline bool qdisc_is_running(struct Qdisc *qdisc) 165 { 166 if (qdisc->flags & TCQ_F_NOLOCK) 167 return spin_is_locked(&qdisc->seqlock); 168 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 169 } 170 171 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc) 172 { 173 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY); 174 } 175 176 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) 177 { 178 return q->flags & TCQ_F_CPUSTATS; 179 } 180 181 static inline bool qdisc_is_empty(const struct Qdisc *qdisc) 182 { 183 if (qdisc_is_percpu_stats(qdisc)) 184 return nolock_qdisc_is_empty(qdisc); 185 return !READ_ONCE(qdisc->q.qlen); 186 } 187 188 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with 189 * the qdisc root lock acquired. 190 */ 191 static inline bool qdisc_run_begin(struct Qdisc *qdisc) 192 { 193 if (qdisc->flags & TCQ_F_NOLOCK) { 194 if (spin_trylock(&qdisc->seqlock)) 195 return true; 196 197 /* No need to insist if the MISSED flag was already set. 198 * Note that test_and_set_bit() also gives us memory ordering 199 * guarantees wrt potential earlier enqueue() and below 200 * spin_trylock(), both of which are necessary to prevent races 201 */ 202 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state)) 203 return false; 204 205 /* Try to take the lock again to make sure that we will either 206 * grab it or the CPU that still has it will see MISSED set 207 * when testing it in qdisc_run_end() 208 */ 209 return spin_trylock(&qdisc->seqlock); 210 } 211 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 212 } 213 214 static inline void qdisc_run_end(struct Qdisc *qdisc) 215 { 216 if (qdisc->flags & TCQ_F_NOLOCK) { 217 spin_unlock(&qdisc->seqlock); 218 219 /* spin_unlock() only has store-release semantic. The unlock 220 * and test_bit() ordering is a store-load ordering, so a full 221 * memory barrier is needed here. 222 */ 223 smp_mb(); 224 225 if (unlikely(test_bit(__QDISC_STATE_MISSED, 226 &qdisc->state))) 227 __netif_schedule(qdisc); 228 } else { 229 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 230 } 231 } 232 233 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 234 { 235 return qdisc->flags & TCQ_F_ONETXQUEUE; 236 } 237 238 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) 239 { 240 #ifdef CONFIG_BQL 241 /* Non-BQL migrated drivers will return 0, too. */ 242 return dql_avail(&txq->dql); 243 #else 244 return 0; 245 #endif 246 } 247 248 struct Qdisc_class_ops { 249 unsigned int flags; 250 /* Child qdisc manipulation */ 251 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 252 int (*graft)(struct Qdisc *, unsigned long cl, 253 struct Qdisc *, struct Qdisc **, 254 struct netlink_ext_ack *extack); 255 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 256 void (*qlen_notify)(struct Qdisc *, unsigned long); 257 258 /* Class manipulation routines */ 259 unsigned long (*find)(struct Qdisc *, u32 classid); 260 int (*change)(struct Qdisc *, u32, u32, 261 struct nlattr **, unsigned long *, 262 struct netlink_ext_ack *); 263 int (*delete)(struct Qdisc *, unsigned long, 264 struct netlink_ext_ack *); 265 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 266 267 /* Filter manipulation */ 268 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 269 unsigned long arg, 270 struct netlink_ext_ack *extack); 271 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 272 u32 classid); 273 void (*unbind_tcf)(struct Qdisc *, unsigned long); 274 275 /* rtnetlink specific */ 276 int (*dump)(struct Qdisc *, unsigned long, 277 struct sk_buff *skb, struct tcmsg*); 278 int (*dump_stats)(struct Qdisc *, unsigned long, 279 struct gnet_dump *); 280 }; 281 282 /* Qdisc_class_ops flag values */ 283 284 /* Implements API that doesn't require rtnl lock */ 285 enum qdisc_class_ops_flags { 286 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 287 }; 288 289 struct Qdisc_ops { 290 struct Qdisc_ops *next; 291 const struct Qdisc_class_ops *cl_ops; 292 char id[IFNAMSIZ]; 293 int priv_size; 294 unsigned int static_flags; 295 296 int (*enqueue)(struct sk_buff *skb, 297 struct Qdisc *sch, 298 struct sk_buff **to_free); 299 struct sk_buff * (*dequeue)(struct Qdisc *); 300 struct sk_buff * (*peek)(struct Qdisc *); 301 302 int (*init)(struct Qdisc *sch, struct nlattr *arg, 303 struct netlink_ext_ack *extack); 304 void (*reset)(struct Qdisc *); 305 void (*destroy)(struct Qdisc *); 306 int (*change)(struct Qdisc *sch, 307 struct nlattr *arg, 308 struct netlink_ext_ack *extack); 309 void (*attach)(struct Qdisc *sch); 310 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 311 void (*change_real_num_tx)(struct Qdisc *sch, 312 unsigned int new_real_tx); 313 314 int (*dump)(struct Qdisc *, struct sk_buff *); 315 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 316 317 void (*ingress_block_set)(struct Qdisc *sch, 318 u32 block_index); 319 void (*egress_block_set)(struct Qdisc *sch, 320 u32 block_index); 321 u32 (*ingress_block_get)(struct Qdisc *sch); 322 u32 (*egress_block_get)(struct Qdisc *sch); 323 324 struct module *owner; 325 }; 326 327 328 struct tcf_result { 329 union { 330 struct { 331 unsigned long class; 332 u32 classid; 333 }; 334 const struct tcf_proto *goto_tp; 335 336 }; 337 }; 338 339 struct tcf_chain; 340 341 struct tcf_proto_ops { 342 struct list_head head; 343 char kind[IFNAMSIZ]; 344 345 int (*classify)(struct sk_buff *, 346 const struct tcf_proto *, 347 struct tcf_result *); 348 int (*init)(struct tcf_proto*); 349 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 350 struct netlink_ext_ack *extack); 351 352 void* (*get)(struct tcf_proto*, u32 handle); 353 void (*put)(struct tcf_proto *tp, void *f); 354 int (*change)(struct net *net, struct sk_buff *, 355 struct tcf_proto*, unsigned long, 356 u32 handle, struct nlattr **, 357 void **, u32, 358 struct netlink_ext_ack *); 359 int (*delete)(struct tcf_proto *tp, void *arg, 360 bool *last, bool rtnl_held, 361 struct netlink_ext_ack *); 362 bool (*delete_empty)(struct tcf_proto *tp); 363 void (*walk)(struct tcf_proto *tp, 364 struct tcf_walker *arg, bool rtnl_held); 365 int (*reoffload)(struct tcf_proto *tp, bool add, 366 flow_setup_cb_t *cb, void *cb_priv, 367 struct netlink_ext_ack *extack); 368 void (*hw_add)(struct tcf_proto *tp, 369 void *type_data); 370 void (*hw_del)(struct tcf_proto *tp, 371 void *type_data); 372 void (*bind_class)(void *, u32, unsigned long, 373 void *, unsigned long); 374 void * (*tmplt_create)(struct net *net, 375 struct tcf_chain *chain, 376 struct nlattr **tca, 377 struct netlink_ext_ack *extack); 378 void (*tmplt_destroy)(void *tmplt_priv); 379 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp, 380 u32 handle); 381 382 /* rtnetlink specific */ 383 int (*dump)(struct net*, struct tcf_proto*, void *, 384 struct sk_buff *skb, struct tcmsg*, 385 bool); 386 int (*terse_dump)(struct net *net, 387 struct tcf_proto *tp, void *fh, 388 struct sk_buff *skb, 389 struct tcmsg *t, bool rtnl_held); 390 int (*tmplt_dump)(struct sk_buff *skb, 391 struct net *net, 392 void *tmplt_priv); 393 394 struct module *owner; 395 int flags; 396 }; 397 398 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 399 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 400 * conditions can occur when filters are inserted/deleted simultaneously. 401 */ 402 enum tcf_proto_ops_flags { 403 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 404 }; 405 406 struct tcf_proto { 407 /* Fast access part */ 408 struct tcf_proto __rcu *next; 409 void __rcu *root; 410 411 /* called under RCU BH lock*/ 412 int (*classify)(struct sk_buff *, 413 const struct tcf_proto *, 414 struct tcf_result *); 415 __be16 protocol; 416 417 /* All the rest */ 418 u32 prio; 419 void *data; 420 const struct tcf_proto_ops *ops; 421 struct tcf_chain *chain; 422 /* Lock protects tcf_proto shared state and can be used by unlocked 423 * classifiers to protect their private data. 424 */ 425 spinlock_t lock; 426 bool deleting; 427 refcount_t refcnt; 428 struct rcu_head rcu; 429 struct hlist_node destroy_ht_node; 430 }; 431 432 struct qdisc_skb_cb { 433 struct { 434 unsigned int pkt_len; 435 u16 slave_dev_queue_mapping; 436 u16 tc_classid; 437 }; 438 #define QDISC_CB_PRIV_LEN 20 439 unsigned char data[QDISC_CB_PRIV_LEN]; 440 }; 441 442 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 443 444 struct tcf_chain { 445 /* Protects filter_chain. */ 446 struct mutex filter_chain_lock; 447 struct tcf_proto __rcu *filter_chain; 448 struct list_head list; 449 struct tcf_block *block; 450 u32 index; /* chain index */ 451 unsigned int refcnt; 452 unsigned int action_refcnt; 453 bool explicitly_created; 454 bool flushing; 455 const struct tcf_proto_ops *tmplt_ops; 456 void *tmplt_priv; 457 struct rcu_head rcu; 458 }; 459 460 struct tcf_block { 461 /* Lock protects tcf_block and lifetime-management data of chains 462 * attached to the block (refcnt, action_refcnt, explicitly_created). 463 */ 464 struct mutex lock; 465 struct list_head chain_list; 466 u32 index; /* block index for shared blocks */ 467 u32 classid; /* which class this block belongs to */ 468 refcount_t refcnt; 469 struct net *net; 470 struct Qdisc *q; 471 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 472 struct flow_block flow_block; 473 struct list_head owner_list; 474 bool keep_dst; 475 atomic_t offloadcnt; /* Number of oddloaded filters */ 476 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 477 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 478 struct { 479 struct tcf_chain *chain; 480 struct list_head filter_chain_list; 481 } chain0; 482 struct rcu_head rcu; 483 DECLARE_HASHTABLE(proto_destroy_ht, 7); 484 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 485 }; 486 487 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 488 { 489 return lockdep_is_held(&chain->filter_chain_lock); 490 } 491 492 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 493 { 494 return lockdep_is_held(&tp->lock); 495 } 496 497 #define tcf_chain_dereference(p, chain) \ 498 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 499 500 #define tcf_proto_dereference(p, tp) \ 501 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 502 503 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 504 { 505 struct qdisc_skb_cb *qcb; 506 507 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 508 BUILD_BUG_ON(sizeof(qcb->data) < sz); 509 } 510 511 static inline int qdisc_qlen(const struct Qdisc *q) 512 { 513 return q->q.qlen; 514 } 515 516 static inline int qdisc_qlen_sum(const struct Qdisc *q) 517 { 518 __u32 qlen = q->qstats.qlen; 519 int i; 520 521 if (qdisc_is_percpu_stats(q)) { 522 for_each_possible_cpu(i) 523 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 524 } else { 525 qlen += q->q.qlen; 526 } 527 528 return qlen; 529 } 530 531 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 532 { 533 return (struct qdisc_skb_cb *)skb->cb; 534 } 535 536 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 537 { 538 return &qdisc->q.lock; 539 } 540 541 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 542 { 543 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 544 545 return q; 546 } 547 548 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 549 { 550 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 551 } 552 553 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 554 { 555 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping); 556 } 557 558 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 559 { 560 struct Qdisc *root = qdisc_root_sleeping(qdisc); 561 562 ASSERT_RTNL(); 563 return qdisc_lock(root); 564 } 565 566 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 567 { 568 return qdisc->dev_queue->dev; 569 } 570 571 static inline void sch_tree_lock(struct Qdisc *q) 572 { 573 if (q->flags & TCQ_F_MQROOT) 574 spin_lock_bh(qdisc_lock(q)); 575 else 576 spin_lock_bh(qdisc_root_sleeping_lock(q)); 577 } 578 579 static inline void sch_tree_unlock(struct Qdisc *q) 580 { 581 if (q->flags & TCQ_F_MQROOT) 582 spin_unlock_bh(qdisc_lock(q)); 583 else 584 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 585 } 586 587 extern struct Qdisc noop_qdisc; 588 extern struct Qdisc_ops noop_qdisc_ops; 589 extern struct Qdisc_ops pfifo_fast_ops; 590 extern struct Qdisc_ops mq_qdisc_ops; 591 extern struct Qdisc_ops noqueue_qdisc_ops; 592 extern const struct Qdisc_ops *default_qdisc_ops; 593 static inline const struct Qdisc_ops * 594 get_default_qdisc_ops(const struct net_device *dev, int ntx) 595 { 596 return ntx < dev->real_num_tx_queues ? 597 default_qdisc_ops : &pfifo_fast_ops; 598 } 599 600 struct Qdisc_class_common { 601 u32 classid; 602 struct hlist_node hnode; 603 }; 604 605 struct Qdisc_class_hash { 606 struct hlist_head *hash; 607 unsigned int hashsize; 608 unsigned int hashmask; 609 unsigned int hashelems; 610 }; 611 612 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 613 { 614 id ^= id >> 8; 615 id ^= id >> 4; 616 return id & mask; 617 } 618 619 static inline struct Qdisc_class_common * 620 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 621 { 622 struct Qdisc_class_common *cl; 623 unsigned int h; 624 625 if (!id) 626 return NULL; 627 628 h = qdisc_class_hash(id, hash->hashmask); 629 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 630 if (cl->classid == id) 631 return cl; 632 } 633 return NULL; 634 } 635 636 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 637 { 638 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 639 640 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 641 } 642 643 int qdisc_class_hash_init(struct Qdisc_class_hash *); 644 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 645 struct Qdisc_class_common *); 646 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 647 struct Qdisc_class_common *); 648 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 649 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 650 651 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 652 void dev_qdisc_change_real_num_tx(struct net_device *dev, 653 unsigned int new_real_tx); 654 void dev_init_scheduler(struct net_device *dev); 655 void dev_shutdown(struct net_device *dev); 656 void dev_activate(struct net_device *dev); 657 void dev_deactivate(struct net_device *dev); 658 void dev_deactivate_many(struct list_head *head); 659 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 660 struct Qdisc *qdisc); 661 void qdisc_reset(struct Qdisc *qdisc); 662 void qdisc_destroy(struct Qdisc *qdisc); 663 void qdisc_put(struct Qdisc *qdisc); 664 void qdisc_put_unlocked(struct Qdisc *qdisc); 665 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 666 #ifdef CONFIG_NET_SCHED 667 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 668 void *type_data); 669 void 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 #else 674 static inline int 675 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 676 void *type_data) 677 { 678 q->flags &= ~TCQ_F_OFFLOADED; 679 return 0; 680 } 681 682 static inline void 683 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 684 struct Qdisc *new, struct Qdisc *old, 685 enum tc_setup_type type, void *type_data, 686 struct netlink_ext_ack *extack) 687 { 688 } 689 #endif 690 void qdisc_offload_query_caps(struct net_device *dev, 691 enum tc_setup_type type, 692 void *caps, size_t caps_len); 693 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 694 const struct Qdisc_ops *ops, 695 struct netlink_ext_ack *extack); 696 void qdisc_free(struct Qdisc *qdisc); 697 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 698 const struct Qdisc_ops *ops, u32 parentid, 699 struct netlink_ext_ack *extack); 700 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 701 const struct qdisc_size_table *stab); 702 int skb_do_redirect(struct sk_buff *); 703 704 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 705 { 706 #ifdef CONFIG_NET_CLS_ACT 707 return skb->tc_at_ingress; 708 #else 709 return false; 710 #endif 711 } 712 713 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 714 { 715 #ifdef CONFIG_NET_CLS_ACT 716 if (skb->tc_skip_classify) { 717 skb->tc_skip_classify = 0; 718 return true; 719 } 720 #endif 721 return false; 722 } 723 724 /* Reset all TX qdiscs greater than index of a device. */ 725 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 726 { 727 struct Qdisc *qdisc; 728 729 for (; i < dev->num_tx_queues; i++) { 730 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 731 if (qdisc) { 732 spin_lock_bh(qdisc_lock(qdisc)); 733 qdisc_reset(qdisc); 734 spin_unlock_bh(qdisc_lock(qdisc)); 735 } 736 } 737 } 738 739 /* Are all TX queues of the device empty? */ 740 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 741 { 742 unsigned int i; 743 744 rcu_read_lock(); 745 for (i = 0; i < dev->num_tx_queues; i++) { 746 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 747 const struct Qdisc *q = rcu_dereference(txq->qdisc); 748 749 if (!qdisc_is_empty(q)) { 750 rcu_read_unlock(); 751 return false; 752 } 753 } 754 rcu_read_unlock(); 755 return true; 756 } 757 758 /* Are any of the TX qdiscs changing? */ 759 static inline bool qdisc_tx_changing(const struct net_device *dev) 760 { 761 unsigned int i; 762 763 for (i = 0; i < dev->num_tx_queues; i++) { 764 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 765 766 if (rcu_access_pointer(txq->qdisc) != 767 rcu_access_pointer(txq->qdisc_sleeping)) 768 return true; 769 } 770 return false; 771 } 772 773 /* Is the device using the noop qdisc on all queues? */ 774 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 775 { 776 unsigned int i; 777 778 for (i = 0; i < dev->num_tx_queues; i++) { 779 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 780 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 781 return false; 782 } 783 return true; 784 } 785 786 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 787 { 788 return qdisc_skb_cb(skb)->pkt_len; 789 } 790 791 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 792 enum net_xmit_qdisc_t { 793 __NET_XMIT_STOLEN = 0x00010000, 794 __NET_XMIT_BYPASS = 0x00020000, 795 }; 796 797 #ifdef CONFIG_NET_CLS_ACT 798 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 799 #else 800 #define net_xmit_drop_count(e) (1) 801 #endif 802 803 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 804 const struct Qdisc *sch) 805 { 806 #ifdef CONFIG_NET_SCHED 807 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 808 809 if (stab) 810 __qdisc_calculate_pkt_len(skb, stab); 811 #endif 812 } 813 814 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 815 struct sk_buff **to_free) 816 { 817 qdisc_calculate_pkt_len(skb, sch); 818 return sch->enqueue(skb, sch, to_free); 819 } 820 821 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats, 822 __u64 bytes, __u32 packets) 823 { 824 u64_stats_update_begin(&bstats->syncp); 825 u64_stats_add(&bstats->bytes, bytes); 826 u64_stats_add(&bstats->packets, packets); 827 u64_stats_update_end(&bstats->syncp); 828 } 829 830 static inline void bstats_update(struct gnet_stats_basic_sync *bstats, 831 const struct sk_buff *skb) 832 { 833 _bstats_update(bstats, 834 qdisc_pkt_len(skb), 835 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 836 } 837 838 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 839 const struct sk_buff *skb) 840 { 841 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb); 842 } 843 844 static inline void qdisc_bstats_update(struct Qdisc *sch, 845 const struct sk_buff *skb) 846 { 847 bstats_update(&sch->bstats, skb); 848 } 849 850 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 851 const struct sk_buff *skb) 852 { 853 sch->qstats.backlog -= qdisc_pkt_len(skb); 854 } 855 856 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 857 const struct sk_buff *skb) 858 { 859 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 860 } 861 862 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 863 const struct sk_buff *skb) 864 { 865 sch->qstats.backlog += qdisc_pkt_len(skb); 866 } 867 868 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 869 const struct sk_buff *skb) 870 { 871 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 872 } 873 874 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 875 { 876 this_cpu_inc(sch->cpu_qstats->qlen); 877 } 878 879 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 880 { 881 this_cpu_dec(sch->cpu_qstats->qlen); 882 } 883 884 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 885 { 886 this_cpu_inc(sch->cpu_qstats->requeues); 887 } 888 889 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 890 { 891 sch->qstats.drops += count; 892 } 893 894 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 895 { 896 qstats->drops++; 897 } 898 899 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 900 { 901 qstats->overlimits++; 902 } 903 904 static inline void qdisc_qstats_drop(struct Qdisc *sch) 905 { 906 qstats_drop_inc(&sch->qstats); 907 } 908 909 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 910 { 911 this_cpu_inc(sch->cpu_qstats->drops); 912 } 913 914 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 915 { 916 sch->qstats.overlimits++; 917 } 918 919 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 920 { 921 __u32 qlen = qdisc_qlen_sum(sch); 922 923 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 924 } 925 926 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 927 __u32 *backlog) 928 { 929 struct gnet_stats_queue qstats = { 0 }; 930 931 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats); 932 *qlen = qstats.qlen + qdisc_qlen(sch); 933 *backlog = qstats.backlog; 934 } 935 936 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 937 { 938 __u32 qlen, backlog; 939 940 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 941 qdisc_tree_reduce_backlog(sch, qlen, backlog); 942 } 943 944 static inline void qdisc_purge_queue(struct Qdisc *sch) 945 { 946 __u32 qlen, backlog; 947 948 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 949 qdisc_reset(sch); 950 qdisc_tree_reduce_backlog(sch, qlen, backlog); 951 } 952 953 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 954 struct qdisc_skb_head *qh) 955 { 956 struct sk_buff *last = qh->tail; 957 958 if (last) { 959 skb->next = NULL; 960 last->next = skb; 961 qh->tail = skb; 962 } else { 963 qh->tail = skb; 964 qh->head = skb; 965 } 966 qh->qlen++; 967 } 968 969 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 970 { 971 __qdisc_enqueue_tail(skb, &sch->q); 972 qdisc_qstats_backlog_inc(sch, skb); 973 return NET_XMIT_SUCCESS; 974 } 975 976 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 977 struct qdisc_skb_head *qh) 978 { 979 skb->next = qh->head; 980 981 if (!qh->head) 982 qh->tail = skb; 983 qh->head = skb; 984 qh->qlen++; 985 } 986 987 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 988 { 989 struct sk_buff *skb = qh->head; 990 991 if (likely(skb != NULL)) { 992 qh->head = skb->next; 993 qh->qlen--; 994 if (qh->head == NULL) 995 qh->tail = NULL; 996 skb->next = NULL; 997 } 998 999 return skb; 1000 } 1001 1002 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1003 { 1004 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1005 1006 if (likely(skb != NULL)) { 1007 qdisc_qstats_backlog_dec(sch, skb); 1008 qdisc_bstats_update(sch, skb); 1009 } 1010 1011 return skb; 1012 } 1013 1014 /* Instead of calling kfree_skb() while root qdisc lock is held, 1015 * queue the skb for future freeing at end of __dev_xmit_skb() 1016 */ 1017 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1018 { 1019 skb->next = *to_free; 1020 *to_free = skb; 1021 } 1022 1023 static inline void __qdisc_drop_all(struct sk_buff *skb, 1024 struct sk_buff **to_free) 1025 { 1026 if (skb->prev) 1027 skb->prev->next = *to_free; 1028 else 1029 skb->next = *to_free; 1030 *to_free = skb; 1031 } 1032 1033 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1034 struct qdisc_skb_head *qh, 1035 struct sk_buff **to_free) 1036 { 1037 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1038 1039 if (likely(skb != NULL)) { 1040 unsigned int len = qdisc_pkt_len(skb); 1041 1042 qdisc_qstats_backlog_dec(sch, skb); 1043 __qdisc_drop(skb, to_free); 1044 return len; 1045 } 1046 1047 return 0; 1048 } 1049 1050 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1051 { 1052 const struct qdisc_skb_head *qh = &sch->q; 1053 1054 return qh->head; 1055 } 1056 1057 /* generic pseudo peek method for non-work-conserving qdisc */ 1058 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1059 { 1060 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1061 1062 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1063 if (!skb) { 1064 skb = sch->dequeue(sch); 1065 1066 if (skb) { 1067 __skb_queue_head(&sch->gso_skb, skb); 1068 /* it's still part of the queue */ 1069 qdisc_qstats_backlog_inc(sch, skb); 1070 sch->q.qlen++; 1071 } 1072 } 1073 1074 return skb; 1075 } 1076 1077 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1078 struct sk_buff *skb) 1079 { 1080 if (qdisc_is_percpu_stats(sch)) { 1081 qdisc_qstats_cpu_backlog_dec(sch, skb); 1082 qdisc_bstats_cpu_update(sch, skb); 1083 qdisc_qstats_cpu_qlen_dec(sch); 1084 } else { 1085 qdisc_qstats_backlog_dec(sch, skb); 1086 qdisc_bstats_update(sch, skb); 1087 sch->q.qlen--; 1088 } 1089 } 1090 1091 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1092 unsigned int pkt_len) 1093 { 1094 if (qdisc_is_percpu_stats(sch)) { 1095 qdisc_qstats_cpu_qlen_inc(sch); 1096 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1097 } else { 1098 sch->qstats.backlog += pkt_len; 1099 sch->q.qlen++; 1100 } 1101 } 1102 1103 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1104 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1105 { 1106 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1107 1108 if (skb) { 1109 skb = __skb_dequeue(&sch->gso_skb); 1110 if (qdisc_is_percpu_stats(sch)) { 1111 qdisc_qstats_cpu_backlog_dec(sch, skb); 1112 qdisc_qstats_cpu_qlen_dec(sch); 1113 } else { 1114 qdisc_qstats_backlog_dec(sch, skb); 1115 sch->q.qlen--; 1116 } 1117 } else { 1118 skb = sch->dequeue(sch); 1119 } 1120 1121 return skb; 1122 } 1123 1124 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1125 { 1126 /* 1127 * We do not know the backlog in bytes of this list, it 1128 * is up to the caller to correct it 1129 */ 1130 ASSERT_RTNL(); 1131 if (qh->qlen) { 1132 rtnl_kfree_skbs(qh->head, qh->tail); 1133 1134 qh->head = NULL; 1135 qh->tail = NULL; 1136 qh->qlen = 0; 1137 } 1138 } 1139 1140 static inline void qdisc_reset_queue(struct Qdisc *sch) 1141 { 1142 __qdisc_reset_queue(&sch->q); 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 struct psched_ratecfg { 1194 u64 rate_bytes_ps; /* bytes per second */ 1195 u32 mult; 1196 u16 overhead; 1197 u16 mpu; 1198 u8 linklayer; 1199 u8 shift; 1200 }; 1201 1202 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1203 unsigned int len) 1204 { 1205 len += r->overhead; 1206 1207 if (len < r->mpu) 1208 len = r->mpu; 1209 1210 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1211 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1212 1213 return ((u64)len * r->mult) >> r->shift; 1214 } 1215 1216 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1217 const struct tc_ratespec *conf, 1218 u64 rate64); 1219 1220 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1221 const struct psched_ratecfg *r) 1222 { 1223 memset(res, 0, sizeof(*res)); 1224 1225 /* legacy struct tc_ratespec has a 32bit @rate field 1226 * Qdisc using 64bit rate should add new attributes 1227 * in order to maintain compatibility. 1228 */ 1229 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1230 1231 res->overhead = r->overhead; 1232 res->mpu = r->mpu; 1233 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1234 } 1235 1236 struct psched_pktrate { 1237 u64 rate_pkts_ps; /* packets per second */ 1238 u32 mult; 1239 u8 shift; 1240 }; 1241 1242 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1243 unsigned int pkt_num) 1244 { 1245 return ((u64)pkt_num * r->mult) >> r->shift; 1246 } 1247 1248 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1249 1250 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1251 * The fast path only needs to access filter list and to update stats 1252 */ 1253 struct mini_Qdisc { 1254 struct tcf_proto *filter_list; 1255 struct tcf_block *block; 1256 struct gnet_stats_basic_sync __percpu *cpu_bstats; 1257 struct gnet_stats_queue __percpu *cpu_qstats; 1258 unsigned long rcu_state; 1259 }; 1260 1261 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1262 const struct sk_buff *skb) 1263 { 1264 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1265 } 1266 1267 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1268 { 1269 this_cpu_inc(miniq->cpu_qstats->drops); 1270 } 1271 1272 struct mini_Qdisc_pair { 1273 struct mini_Qdisc miniq1; 1274 struct mini_Qdisc miniq2; 1275 struct mini_Qdisc __rcu **p_miniq; 1276 }; 1277 1278 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1279 struct tcf_proto *tp_head); 1280 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1281 struct mini_Qdisc __rcu **p_miniq); 1282 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1283 struct tcf_block *block); 1284 1285 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx); 1286 1287 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1288 1289 /* Make sure qdisc is no longer in SCHED state. */ 1290 static inline void qdisc_synchronize(const struct Qdisc *q) 1291 { 1292 while (test_bit(__QDISC_STATE_SCHED, &q->state)) 1293 msleep(1); 1294 } 1295 1296 #endif 1297