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 /* Intended to be used by unlocked users, when concurrent qdisc release is 141 * possible. 142 */ 143 144 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc) 145 { 146 if (qdisc->flags & TCQ_F_BUILTIN) 147 return qdisc; 148 if (refcount_inc_not_zero(&qdisc->refcnt)) 149 return qdisc; 150 return NULL; 151 } 152 153 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc 154 * root_lock section, or provide their own memory barriers -- ordering 155 * against qdisc_run_begin/end() atomic bit operations. 156 */ 157 static inline bool qdisc_is_running(struct Qdisc *qdisc) 158 { 159 if (qdisc->flags & TCQ_F_NOLOCK) 160 return spin_is_locked(&qdisc->seqlock); 161 return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 162 } 163 164 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc) 165 { 166 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY); 167 } 168 169 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) 170 { 171 return q->flags & TCQ_F_CPUSTATS; 172 } 173 174 static inline bool qdisc_is_empty(const struct Qdisc *qdisc) 175 { 176 if (qdisc_is_percpu_stats(qdisc)) 177 return nolock_qdisc_is_empty(qdisc); 178 return !READ_ONCE(qdisc->q.qlen); 179 } 180 181 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with 182 * the qdisc root lock acquired. 183 */ 184 static inline bool qdisc_run_begin(struct Qdisc *qdisc) 185 { 186 if (qdisc->flags & TCQ_F_NOLOCK) { 187 if (spin_trylock(&qdisc->seqlock)) 188 return true; 189 190 /* No need to insist if the MISSED flag was already set. 191 * Note that test_and_set_bit() also gives us memory ordering 192 * guarantees wrt potential earlier enqueue() and below 193 * spin_trylock(), both of which are necessary to prevent races 194 */ 195 if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state)) 196 return false; 197 198 /* Try to take the lock again to make sure that we will either 199 * grab it or the CPU that still has it will see MISSED set 200 * when testing it in qdisc_run_end() 201 */ 202 return spin_trylock(&qdisc->seqlock); 203 } 204 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 205 } 206 207 static inline void qdisc_run_end(struct Qdisc *qdisc) 208 { 209 if (qdisc->flags & TCQ_F_NOLOCK) { 210 spin_unlock(&qdisc->seqlock); 211 212 /* spin_unlock() only has store-release semantic. The unlock 213 * and test_bit() ordering is a store-load ordering, so a full 214 * memory barrier is needed here. 215 */ 216 smp_mb(); 217 218 if (unlikely(test_bit(__QDISC_STATE_MISSED, 219 &qdisc->state))) 220 __netif_schedule(qdisc); 221 } else { 222 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 223 } 224 } 225 226 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 227 { 228 return qdisc->flags & TCQ_F_ONETXQUEUE; 229 } 230 231 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) 232 { 233 #ifdef CONFIG_BQL 234 /* Non-BQL migrated drivers will return 0, too. */ 235 return dql_avail(&txq->dql); 236 #else 237 return 0; 238 #endif 239 } 240 241 struct Qdisc_class_ops { 242 unsigned int flags; 243 /* Child qdisc manipulation */ 244 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 245 int (*graft)(struct Qdisc *, unsigned long cl, 246 struct Qdisc *, struct Qdisc **, 247 struct netlink_ext_ack *extack); 248 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 249 void (*qlen_notify)(struct Qdisc *, unsigned long); 250 251 /* Class manipulation routines */ 252 unsigned long (*find)(struct Qdisc *, u32 classid); 253 int (*change)(struct Qdisc *, u32, u32, 254 struct nlattr **, unsigned long *, 255 struct netlink_ext_ack *); 256 int (*delete)(struct Qdisc *, unsigned long, 257 struct netlink_ext_ack *); 258 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 259 260 /* Filter manipulation */ 261 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 262 unsigned long arg, 263 struct netlink_ext_ack *extack); 264 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 265 u32 classid); 266 void (*unbind_tcf)(struct Qdisc *, unsigned long); 267 268 /* rtnetlink specific */ 269 int (*dump)(struct Qdisc *, unsigned long, 270 struct sk_buff *skb, struct tcmsg*); 271 int (*dump_stats)(struct Qdisc *, unsigned long, 272 struct gnet_dump *); 273 }; 274 275 /* Qdisc_class_ops flag values */ 276 277 /* Implements API that doesn't require rtnl lock */ 278 enum qdisc_class_ops_flags { 279 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 280 }; 281 282 struct Qdisc_ops { 283 struct Qdisc_ops *next; 284 const struct Qdisc_class_ops *cl_ops; 285 char id[IFNAMSIZ]; 286 int priv_size; 287 unsigned int static_flags; 288 289 int (*enqueue)(struct sk_buff *skb, 290 struct Qdisc *sch, 291 struct sk_buff **to_free); 292 struct sk_buff * (*dequeue)(struct Qdisc *); 293 struct sk_buff * (*peek)(struct Qdisc *); 294 295 int (*init)(struct Qdisc *sch, struct nlattr *arg, 296 struct netlink_ext_ack *extack); 297 void (*reset)(struct Qdisc *); 298 void (*destroy)(struct Qdisc *); 299 int (*change)(struct Qdisc *sch, 300 struct nlattr *arg, 301 struct netlink_ext_ack *extack); 302 void (*attach)(struct Qdisc *sch); 303 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 304 void (*change_real_num_tx)(struct Qdisc *sch, 305 unsigned int new_real_tx); 306 307 int (*dump)(struct Qdisc *, struct sk_buff *); 308 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 309 310 void (*ingress_block_set)(struct Qdisc *sch, 311 u32 block_index); 312 void (*egress_block_set)(struct Qdisc *sch, 313 u32 block_index); 314 u32 (*ingress_block_get)(struct Qdisc *sch); 315 u32 (*egress_block_get)(struct Qdisc *sch); 316 317 struct module *owner; 318 }; 319 320 321 struct tcf_result { 322 union { 323 struct { 324 unsigned long class; 325 u32 classid; 326 }; 327 const struct tcf_proto *goto_tp; 328 329 }; 330 }; 331 332 struct tcf_chain; 333 334 struct tcf_proto_ops { 335 struct list_head head; 336 char kind[IFNAMSIZ]; 337 338 int (*classify)(struct sk_buff *, 339 const struct tcf_proto *, 340 struct tcf_result *); 341 int (*init)(struct tcf_proto*); 342 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 343 struct netlink_ext_ack *extack); 344 345 void* (*get)(struct tcf_proto*, u32 handle); 346 void (*put)(struct tcf_proto *tp, void *f); 347 int (*change)(struct net *net, struct sk_buff *, 348 struct tcf_proto*, unsigned long, 349 u32 handle, struct nlattr **, 350 void **, u32, 351 struct netlink_ext_ack *); 352 int (*delete)(struct tcf_proto *tp, void *arg, 353 bool *last, bool rtnl_held, 354 struct netlink_ext_ack *); 355 bool (*delete_empty)(struct tcf_proto *tp); 356 void (*walk)(struct tcf_proto *tp, 357 struct tcf_walker *arg, bool rtnl_held); 358 int (*reoffload)(struct tcf_proto *tp, bool add, 359 flow_setup_cb_t *cb, void *cb_priv, 360 struct netlink_ext_ack *extack); 361 void (*hw_add)(struct tcf_proto *tp, 362 void *type_data); 363 void (*hw_del)(struct tcf_proto *tp, 364 void *type_data); 365 void (*bind_class)(void *, u32, unsigned long, 366 void *, unsigned long); 367 void * (*tmplt_create)(struct net *net, 368 struct tcf_chain *chain, 369 struct nlattr **tca, 370 struct netlink_ext_ack *extack); 371 void (*tmplt_destroy)(void *tmplt_priv); 372 373 /* rtnetlink specific */ 374 int (*dump)(struct net*, struct tcf_proto*, void *, 375 struct sk_buff *skb, struct tcmsg*, 376 bool); 377 int (*terse_dump)(struct net *net, 378 struct tcf_proto *tp, void *fh, 379 struct sk_buff *skb, 380 struct tcmsg *t, bool rtnl_held); 381 int (*tmplt_dump)(struct sk_buff *skb, 382 struct net *net, 383 void *tmplt_priv); 384 385 struct module *owner; 386 int flags; 387 }; 388 389 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 390 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 391 * conditions can occur when filters are inserted/deleted simultaneously. 392 */ 393 enum tcf_proto_ops_flags { 394 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 395 }; 396 397 struct tcf_proto { 398 /* Fast access part */ 399 struct tcf_proto __rcu *next; 400 void __rcu *root; 401 402 /* called under RCU BH lock*/ 403 int (*classify)(struct sk_buff *, 404 const struct tcf_proto *, 405 struct tcf_result *); 406 __be16 protocol; 407 408 /* All the rest */ 409 u32 prio; 410 void *data; 411 const struct tcf_proto_ops *ops; 412 struct tcf_chain *chain; 413 /* Lock protects tcf_proto shared state and can be used by unlocked 414 * classifiers to protect their private data. 415 */ 416 spinlock_t lock; 417 bool deleting; 418 refcount_t refcnt; 419 struct rcu_head rcu; 420 struct hlist_node destroy_ht_node; 421 }; 422 423 struct qdisc_skb_cb { 424 struct { 425 unsigned int pkt_len; 426 u16 slave_dev_queue_mapping; 427 u16 tc_classid; 428 }; 429 #define QDISC_CB_PRIV_LEN 20 430 unsigned char data[QDISC_CB_PRIV_LEN]; 431 }; 432 433 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 434 435 struct tcf_chain { 436 /* Protects filter_chain. */ 437 struct mutex filter_chain_lock; 438 struct tcf_proto __rcu *filter_chain; 439 struct list_head list; 440 struct tcf_block *block; 441 u32 index; /* chain index */ 442 unsigned int refcnt; 443 unsigned int action_refcnt; 444 bool explicitly_created; 445 bool flushing; 446 const struct tcf_proto_ops *tmplt_ops; 447 void *tmplt_priv; 448 struct rcu_head rcu; 449 }; 450 451 struct tcf_block { 452 /* Lock protects tcf_block and lifetime-management data of chains 453 * attached to the block (refcnt, action_refcnt, explicitly_created). 454 */ 455 struct mutex lock; 456 struct list_head chain_list; 457 u32 index; /* block index for shared blocks */ 458 u32 classid; /* which class this block belongs to */ 459 refcount_t refcnt; 460 struct net *net; 461 struct Qdisc *q; 462 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 463 struct flow_block flow_block; 464 struct list_head owner_list; 465 bool keep_dst; 466 atomic_t offloadcnt; /* Number of oddloaded filters */ 467 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 468 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 469 struct { 470 struct tcf_chain *chain; 471 struct list_head filter_chain_list; 472 } chain0; 473 struct rcu_head rcu; 474 DECLARE_HASHTABLE(proto_destroy_ht, 7); 475 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 476 }; 477 478 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 479 { 480 return lockdep_is_held(&chain->filter_chain_lock); 481 } 482 483 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 484 { 485 return lockdep_is_held(&tp->lock); 486 } 487 488 #define tcf_chain_dereference(p, chain) \ 489 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 490 491 #define tcf_proto_dereference(p, tp) \ 492 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 493 494 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 495 { 496 struct qdisc_skb_cb *qcb; 497 498 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 499 BUILD_BUG_ON(sizeof(qcb->data) < sz); 500 } 501 502 static inline int qdisc_qlen(const struct Qdisc *q) 503 { 504 return q->q.qlen; 505 } 506 507 static inline int qdisc_qlen_sum(const struct Qdisc *q) 508 { 509 __u32 qlen = q->qstats.qlen; 510 int i; 511 512 if (qdisc_is_percpu_stats(q)) { 513 for_each_possible_cpu(i) 514 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 515 } else { 516 qlen += q->q.qlen; 517 } 518 519 return qlen; 520 } 521 522 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 523 { 524 return (struct qdisc_skb_cb *)skb->cb; 525 } 526 527 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 528 { 529 return &qdisc->q.lock; 530 } 531 532 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 533 { 534 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 535 536 return q; 537 } 538 539 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 540 { 541 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 542 } 543 544 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 545 { 546 return qdisc->dev_queue->qdisc_sleeping; 547 } 548 549 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 550 { 551 struct Qdisc *root = qdisc_root_sleeping(qdisc); 552 553 ASSERT_RTNL(); 554 return qdisc_lock(root); 555 } 556 557 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 558 { 559 return qdisc->dev_queue->dev; 560 } 561 562 static inline void sch_tree_lock(struct Qdisc *q) 563 { 564 if (q->flags & TCQ_F_MQROOT) 565 spin_lock_bh(qdisc_lock(q)); 566 else 567 spin_lock_bh(qdisc_root_sleeping_lock(q)); 568 } 569 570 static inline void sch_tree_unlock(struct Qdisc *q) 571 { 572 if (q->flags & TCQ_F_MQROOT) 573 spin_unlock_bh(qdisc_lock(q)); 574 else 575 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 576 } 577 578 extern struct Qdisc noop_qdisc; 579 extern struct Qdisc_ops noop_qdisc_ops; 580 extern struct Qdisc_ops pfifo_fast_ops; 581 extern struct Qdisc_ops mq_qdisc_ops; 582 extern struct Qdisc_ops noqueue_qdisc_ops; 583 extern const struct Qdisc_ops *default_qdisc_ops; 584 static inline const struct Qdisc_ops * 585 get_default_qdisc_ops(const struct net_device *dev, int ntx) 586 { 587 return ntx < dev->real_num_tx_queues ? 588 default_qdisc_ops : &pfifo_fast_ops; 589 } 590 591 struct Qdisc_class_common { 592 u32 classid; 593 struct hlist_node hnode; 594 }; 595 596 struct Qdisc_class_hash { 597 struct hlist_head *hash; 598 unsigned int hashsize; 599 unsigned int hashmask; 600 unsigned int hashelems; 601 }; 602 603 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 604 { 605 id ^= id >> 8; 606 id ^= id >> 4; 607 return id & mask; 608 } 609 610 static inline struct Qdisc_class_common * 611 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 612 { 613 struct Qdisc_class_common *cl; 614 unsigned int h; 615 616 if (!id) 617 return NULL; 618 619 h = qdisc_class_hash(id, hash->hashmask); 620 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 621 if (cl->classid == id) 622 return cl; 623 } 624 return NULL; 625 } 626 627 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 628 { 629 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 630 631 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 632 } 633 634 int qdisc_class_hash_init(struct Qdisc_class_hash *); 635 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 636 struct Qdisc_class_common *); 637 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 638 struct Qdisc_class_common *); 639 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 640 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 641 642 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 643 void dev_qdisc_change_real_num_tx(struct net_device *dev, 644 unsigned int new_real_tx); 645 void dev_init_scheduler(struct net_device *dev); 646 void dev_shutdown(struct net_device *dev); 647 void dev_activate(struct net_device *dev); 648 void dev_deactivate(struct net_device *dev); 649 void dev_deactivate_many(struct list_head *head); 650 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 651 struct Qdisc *qdisc); 652 void qdisc_reset(struct Qdisc *qdisc); 653 void qdisc_put(struct Qdisc *qdisc); 654 void qdisc_put_unlocked(struct Qdisc *qdisc); 655 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 656 #ifdef CONFIG_NET_SCHED 657 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 658 void *type_data); 659 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 660 struct Qdisc *new, struct Qdisc *old, 661 enum tc_setup_type type, void *type_data, 662 struct netlink_ext_ack *extack); 663 #else 664 static inline int 665 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 666 void *type_data) 667 { 668 q->flags &= ~TCQ_F_OFFLOADED; 669 return 0; 670 } 671 672 static inline void 673 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 674 struct Qdisc *new, struct Qdisc *old, 675 enum tc_setup_type type, void *type_data, 676 struct netlink_ext_ack *extack) 677 { 678 } 679 #endif 680 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 681 const struct Qdisc_ops *ops, 682 struct netlink_ext_ack *extack); 683 void qdisc_free(struct Qdisc *qdisc); 684 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 685 const struct Qdisc_ops *ops, u32 parentid, 686 struct netlink_ext_ack *extack); 687 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 688 const struct qdisc_size_table *stab); 689 int skb_do_redirect(struct sk_buff *); 690 691 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 692 { 693 #ifdef CONFIG_NET_CLS_ACT 694 return skb->tc_at_ingress; 695 #else 696 return false; 697 #endif 698 } 699 700 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 701 { 702 #ifdef CONFIG_NET_CLS_ACT 703 if (skb->tc_skip_classify) { 704 skb->tc_skip_classify = 0; 705 return true; 706 } 707 #endif 708 return false; 709 } 710 711 /* Reset all TX qdiscs greater than index of a device. */ 712 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 713 { 714 struct Qdisc *qdisc; 715 716 for (; i < dev->num_tx_queues; i++) { 717 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 718 if (qdisc) { 719 spin_lock_bh(qdisc_lock(qdisc)); 720 qdisc_reset(qdisc); 721 spin_unlock_bh(qdisc_lock(qdisc)); 722 } 723 } 724 } 725 726 /* Are all TX queues of the device empty? */ 727 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 728 { 729 unsigned int i; 730 731 rcu_read_lock(); 732 for (i = 0; i < dev->num_tx_queues; i++) { 733 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 734 const struct Qdisc *q = rcu_dereference(txq->qdisc); 735 736 if (!qdisc_is_empty(q)) { 737 rcu_read_unlock(); 738 return false; 739 } 740 } 741 rcu_read_unlock(); 742 return true; 743 } 744 745 /* Are any of the TX qdiscs changing? */ 746 static inline bool qdisc_tx_changing(const struct net_device *dev) 747 { 748 unsigned int i; 749 750 for (i = 0; i < dev->num_tx_queues; i++) { 751 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 752 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) 753 return true; 754 } 755 return false; 756 } 757 758 /* Is the device using the noop qdisc on all queues? */ 759 static inline bool qdisc_tx_is_noop(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 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 766 return false; 767 } 768 return true; 769 } 770 771 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 772 { 773 return qdisc_skb_cb(skb)->pkt_len; 774 } 775 776 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 777 enum net_xmit_qdisc_t { 778 __NET_XMIT_STOLEN = 0x00010000, 779 __NET_XMIT_BYPASS = 0x00020000, 780 }; 781 782 #ifdef CONFIG_NET_CLS_ACT 783 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 784 #else 785 #define net_xmit_drop_count(e) (1) 786 #endif 787 788 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 789 const struct Qdisc *sch) 790 { 791 #ifdef CONFIG_NET_SCHED 792 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 793 794 if (stab) 795 __qdisc_calculate_pkt_len(skb, stab); 796 #endif 797 } 798 799 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 800 struct sk_buff **to_free) 801 { 802 qdisc_calculate_pkt_len(skb, sch); 803 return sch->enqueue(skb, sch, to_free); 804 } 805 806 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats, 807 __u64 bytes, __u32 packets) 808 { 809 u64_stats_update_begin(&bstats->syncp); 810 u64_stats_add(&bstats->bytes, bytes); 811 u64_stats_add(&bstats->packets, packets); 812 u64_stats_update_end(&bstats->syncp); 813 } 814 815 static inline void bstats_update(struct gnet_stats_basic_sync *bstats, 816 const struct sk_buff *skb) 817 { 818 _bstats_update(bstats, 819 qdisc_pkt_len(skb), 820 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 821 } 822 823 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 824 const struct sk_buff *skb) 825 { 826 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb); 827 } 828 829 static inline void qdisc_bstats_update(struct Qdisc *sch, 830 const struct sk_buff *skb) 831 { 832 bstats_update(&sch->bstats, skb); 833 } 834 835 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 836 const struct sk_buff *skb) 837 { 838 sch->qstats.backlog -= qdisc_pkt_len(skb); 839 } 840 841 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 842 const struct sk_buff *skb) 843 { 844 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 845 } 846 847 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 848 const struct sk_buff *skb) 849 { 850 sch->qstats.backlog += qdisc_pkt_len(skb); 851 } 852 853 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 854 const struct sk_buff *skb) 855 { 856 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 857 } 858 859 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 860 { 861 this_cpu_inc(sch->cpu_qstats->qlen); 862 } 863 864 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 865 { 866 this_cpu_dec(sch->cpu_qstats->qlen); 867 } 868 869 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 870 { 871 this_cpu_inc(sch->cpu_qstats->requeues); 872 } 873 874 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 875 { 876 sch->qstats.drops += count; 877 } 878 879 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 880 { 881 qstats->drops++; 882 } 883 884 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 885 { 886 qstats->overlimits++; 887 } 888 889 static inline void qdisc_qstats_drop(struct Qdisc *sch) 890 { 891 qstats_drop_inc(&sch->qstats); 892 } 893 894 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 895 { 896 this_cpu_inc(sch->cpu_qstats->drops); 897 } 898 899 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 900 { 901 sch->qstats.overlimits++; 902 } 903 904 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 905 { 906 __u32 qlen = qdisc_qlen_sum(sch); 907 908 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 909 } 910 911 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 912 __u32 *backlog) 913 { 914 struct gnet_stats_queue qstats = { 0 }; 915 916 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats); 917 *qlen = qstats.qlen + qdisc_qlen(sch); 918 *backlog = qstats.backlog; 919 } 920 921 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 922 { 923 __u32 qlen, backlog; 924 925 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 926 qdisc_tree_reduce_backlog(sch, qlen, backlog); 927 } 928 929 static inline void qdisc_purge_queue(struct Qdisc *sch) 930 { 931 __u32 qlen, backlog; 932 933 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 934 qdisc_reset(sch); 935 qdisc_tree_reduce_backlog(sch, qlen, backlog); 936 } 937 938 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 939 struct qdisc_skb_head *qh) 940 { 941 struct sk_buff *last = qh->tail; 942 943 if (last) { 944 skb->next = NULL; 945 last->next = skb; 946 qh->tail = skb; 947 } else { 948 qh->tail = skb; 949 qh->head = skb; 950 } 951 qh->qlen++; 952 } 953 954 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 955 { 956 __qdisc_enqueue_tail(skb, &sch->q); 957 qdisc_qstats_backlog_inc(sch, skb); 958 return NET_XMIT_SUCCESS; 959 } 960 961 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 962 struct qdisc_skb_head *qh) 963 { 964 skb->next = qh->head; 965 966 if (!qh->head) 967 qh->tail = skb; 968 qh->head = skb; 969 qh->qlen++; 970 } 971 972 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 973 { 974 struct sk_buff *skb = qh->head; 975 976 if (likely(skb != NULL)) { 977 qh->head = skb->next; 978 qh->qlen--; 979 if (qh->head == NULL) 980 qh->tail = NULL; 981 skb->next = NULL; 982 } 983 984 return skb; 985 } 986 987 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 988 { 989 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 990 991 if (likely(skb != NULL)) { 992 qdisc_qstats_backlog_dec(sch, skb); 993 qdisc_bstats_update(sch, skb); 994 } 995 996 return skb; 997 } 998 999 /* Instead of calling kfree_skb() while root qdisc lock is held, 1000 * queue the skb for future freeing at end of __dev_xmit_skb() 1001 */ 1002 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1003 { 1004 skb->next = *to_free; 1005 *to_free = skb; 1006 } 1007 1008 static inline void __qdisc_drop_all(struct sk_buff *skb, 1009 struct sk_buff **to_free) 1010 { 1011 if (skb->prev) 1012 skb->prev->next = *to_free; 1013 else 1014 skb->next = *to_free; 1015 *to_free = skb; 1016 } 1017 1018 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1019 struct qdisc_skb_head *qh, 1020 struct sk_buff **to_free) 1021 { 1022 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1023 1024 if (likely(skb != NULL)) { 1025 unsigned int len = qdisc_pkt_len(skb); 1026 1027 qdisc_qstats_backlog_dec(sch, skb); 1028 __qdisc_drop(skb, to_free); 1029 return len; 1030 } 1031 1032 return 0; 1033 } 1034 1035 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1036 { 1037 const struct qdisc_skb_head *qh = &sch->q; 1038 1039 return qh->head; 1040 } 1041 1042 /* generic pseudo peek method for non-work-conserving qdisc */ 1043 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1044 { 1045 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1046 1047 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1048 if (!skb) { 1049 skb = sch->dequeue(sch); 1050 1051 if (skb) { 1052 __skb_queue_head(&sch->gso_skb, skb); 1053 /* it's still part of the queue */ 1054 qdisc_qstats_backlog_inc(sch, skb); 1055 sch->q.qlen++; 1056 } 1057 } 1058 1059 return skb; 1060 } 1061 1062 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1063 struct sk_buff *skb) 1064 { 1065 if (qdisc_is_percpu_stats(sch)) { 1066 qdisc_qstats_cpu_backlog_dec(sch, skb); 1067 qdisc_bstats_cpu_update(sch, skb); 1068 qdisc_qstats_cpu_qlen_dec(sch); 1069 } else { 1070 qdisc_qstats_backlog_dec(sch, skb); 1071 qdisc_bstats_update(sch, skb); 1072 sch->q.qlen--; 1073 } 1074 } 1075 1076 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1077 unsigned int pkt_len) 1078 { 1079 if (qdisc_is_percpu_stats(sch)) { 1080 qdisc_qstats_cpu_qlen_inc(sch); 1081 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1082 } else { 1083 sch->qstats.backlog += pkt_len; 1084 sch->q.qlen++; 1085 } 1086 } 1087 1088 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1089 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1090 { 1091 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1092 1093 if (skb) { 1094 skb = __skb_dequeue(&sch->gso_skb); 1095 if (qdisc_is_percpu_stats(sch)) { 1096 qdisc_qstats_cpu_backlog_dec(sch, skb); 1097 qdisc_qstats_cpu_qlen_dec(sch); 1098 } else { 1099 qdisc_qstats_backlog_dec(sch, skb); 1100 sch->q.qlen--; 1101 } 1102 } else { 1103 skb = sch->dequeue(sch); 1104 } 1105 1106 return skb; 1107 } 1108 1109 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1110 { 1111 /* 1112 * We do not know the backlog in bytes of this list, it 1113 * is up to the caller to correct it 1114 */ 1115 ASSERT_RTNL(); 1116 if (qh->qlen) { 1117 rtnl_kfree_skbs(qh->head, qh->tail); 1118 1119 qh->head = NULL; 1120 qh->tail = NULL; 1121 qh->qlen = 0; 1122 } 1123 } 1124 1125 static inline void qdisc_reset_queue(struct Qdisc *sch) 1126 { 1127 __qdisc_reset_queue(&sch->q); 1128 } 1129 1130 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1131 struct Qdisc **pold) 1132 { 1133 struct Qdisc *old; 1134 1135 sch_tree_lock(sch); 1136 old = *pold; 1137 *pold = new; 1138 if (old != NULL) 1139 qdisc_purge_queue(old); 1140 sch_tree_unlock(sch); 1141 1142 return old; 1143 } 1144 1145 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1146 { 1147 rtnl_kfree_skbs(skb, skb); 1148 qdisc_qstats_drop(sch); 1149 } 1150 1151 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1152 struct sk_buff **to_free) 1153 { 1154 __qdisc_drop(skb, to_free); 1155 qdisc_qstats_cpu_drop(sch); 1156 1157 return NET_XMIT_DROP; 1158 } 1159 1160 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1161 struct sk_buff **to_free) 1162 { 1163 __qdisc_drop(skb, to_free); 1164 qdisc_qstats_drop(sch); 1165 1166 return NET_XMIT_DROP; 1167 } 1168 1169 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1170 struct sk_buff **to_free) 1171 { 1172 __qdisc_drop_all(skb, to_free); 1173 qdisc_qstats_drop(sch); 1174 1175 return NET_XMIT_DROP; 1176 } 1177 1178 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1179 long it will take to send a packet given its size. 1180 */ 1181 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1182 { 1183 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1184 if (slot < 0) 1185 slot = 0; 1186 slot >>= rtab->rate.cell_log; 1187 if (slot > 255) 1188 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1189 return rtab->data[slot]; 1190 } 1191 1192 struct psched_ratecfg { 1193 u64 rate_bytes_ps; /* bytes per second */ 1194 u32 mult; 1195 u16 overhead; 1196 u16 mpu; 1197 u8 linklayer; 1198 u8 shift; 1199 }; 1200 1201 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1202 unsigned int len) 1203 { 1204 len += r->overhead; 1205 1206 if (len < r->mpu) 1207 len = r->mpu; 1208 1209 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1210 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1211 1212 return ((u64)len * r->mult) >> r->shift; 1213 } 1214 1215 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1216 const struct tc_ratespec *conf, 1217 u64 rate64); 1218 1219 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1220 const struct psched_ratecfg *r) 1221 { 1222 memset(res, 0, sizeof(*res)); 1223 1224 /* legacy struct tc_ratespec has a 32bit @rate field 1225 * Qdisc using 64bit rate should add new attributes 1226 * in order to maintain compatibility. 1227 */ 1228 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1229 1230 res->overhead = r->overhead; 1231 res->mpu = r->mpu; 1232 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1233 } 1234 1235 struct psched_pktrate { 1236 u64 rate_pkts_ps; /* packets per second */ 1237 u32 mult; 1238 u8 shift; 1239 }; 1240 1241 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1242 unsigned int pkt_num) 1243 { 1244 return ((u64)pkt_num * r->mult) >> r->shift; 1245 } 1246 1247 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1248 1249 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1250 * The fast path only needs to access filter list and to update stats 1251 */ 1252 struct mini_Qdisc { 1253 struct tcf_proto *filter_list; 1254 struct tcf_block *block; 1255 struct gnet_stats_basic_sync __percpu *cpu_bstats; 1256 struct gnet_stats_queue __percpu *cpu_qstats; 1257 unsigned long rcu_state; 1258 }; 1259 1260 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1261 const struct sk_buff *skb) 1262 { 1263 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1264 } 1265 1266 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1267 { 1268 this_cpu_inc(miniq->cpu_qstats->drops); 1269 } 1270 1271 struct mini_Qdisc_pair { 1272 struct mini_Qdisc miniq1; 1273 struct mini_Qdisc miniq2; 1274 struct mini_Qdisc __rcu **p_miniq; 1275 }; 1276 1277 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1278 struct tcf_proto *tp_head); 1279 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1280 struct mini_Qdisc __rcu **p_miniq); 1281 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1282 struct tcf_block *block); 1283 1284 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx); 1285 1286 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1287 1288 #endif 1289