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 struct tcf_exts * (*get_exts)(const struct tcf_proto *tp, 373 u32 handle); 374 375 /* rtnetlink specific */ 376 int (*dump)(struct net*, struct tcf_proto*, void *, 377 struct sk_buff *skb, struct tcmsg*, 378 bool); 379 int (*terse_dump)(struct net *net, 380 struct tcf_proto *tp, void *fh, 381 struct sk_buff *skb, 382 struct tcmsg *t, bool rtnl_held); 383 int (*tmplt_dump)(struct sk_buff *skb, 384 struct net *net, 385 void *tmplt_priv); 386 387 struct module *owner; 388 int flags; 389 }; 390 391 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 392 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 393 * conditions can occur when filters are inserted/deleted simultaneously. 394 */ 395 enum tcf_proto_ops_flags { 396 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 397 }; 398 399 struct tcf_proto { 400 /* Fast access part */ 401 struct tcf_proto __rcu *next; 402 void __rcu *root; 403 404 /* called under RCU BH lock*/ 405 int (*classify)(struct sk_buff *, 406 const struct tcf_proto *, 407 struct tcf_result *); 408 __be16 protocol; 409 410 /* All the rest */ 411 u32 prio; 412 void *data; 413 const struct tcf_proto_ops *ops; 414 struct tcf_chain *chain; 415 /* Lock protects tcf_proto shared state and can be used by unlocked 416 * classifiers to protect their private data. 417 */ 418 spinlock_t lock; 419 bool deleting; 420 refcount_t refcnt; 421 struct rcu_head rcu; 422 struct hlist_node destroy_ht_node; 423 }; 424 425 struct qdisc_skb_cb { 426 struct { 427 unsigned int pkt_len; 428 u16 slave_dev_queue_mapping; 429 u16 tc_classid; 430 }; 431 #define QDISC_CB_PRIV_LEN 20 432 unsigned char data[QDISC_CB_PRIV_LEN]; 433 }; 434 435 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 436 437 struct tcf_chain { 438 /* Protects filter_chain. */ 439 struct mutex filter_chain_lock; 440 struct tcf_proto __rcu *filter_chain; 441 struct list_head list; 442 struct tcf_block *block; 443 u32 index; /* chain index */ 444 unsigned int refcnt; 445 unsigned int action_refcnt; 446 bool explicitly_created; 447 bool flushing; 448 const struct tcf_proto_ops *tmplt_ops; 449 void *tmplt_priv; 450 struct rcu_head rcu; 451 }; 452 453 struct tcf_block { 454 /* Lock protects tcf_block and lifetime-management data of chains 455 * attached to the block (refcnt, action_refcnt, explicitly_created). 456 */ 457 struct mutex lock; 458 struct list_head chain_list; 459 u32 index; /* block index for shared blocks */ 460 u32 classid; /* which class this block belongs to */ 461 refcount_t refcnt; 462 struct net *net; 463 struct Qdisc *q; 464 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 465 struct flow_block flow_block; 466 struct list_head owner_list; 467 bool keep_dst; 468 atomic_t offloadcnt; /* Number of oddloaded filters */ 469 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 470 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 471 struct { 472 struct tcf_chain *chain; 473 struct list_head filter_chain_list; 474 } chain0; 475 struct rcu_head rcu; 476 DECLARE_HASHTABLE(proto_destroy_ht, 7); 477 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 478 }; 479 480 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 481 { 482 return lockdep_is_held(&chain->filter_chain_lock); 483 } 484 485 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 486 { 487 return lockdep_is_held(&tp->lock); 488 } 489 490 #define tcf_chain_dereference(p, chain) \ 491 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 492 493 #define tcf_proto_dereference(p, tp) \ 494 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 495 496 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 497 { 498 struct qdisc_skb_cb *qcb; 499 500 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 501 BUILD_BUG_ON(sizeof(qcb->data) < sz); 502 } 503 504 static inline int qdisc_qlen(const struct Qdisc *q) 505 { 506 return q->q.qlen; 507 } 508 509 static inline int qdisc_qlen_sum(const struct Qdisc *q) 510 { 511 __u32 qlen = q->qstats.qlen; 512 int i; 513 514 if (qdisc_is_percpu_stats(q)) { 515 for_each_possible_cpu(i) 516 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 517 } else { 518 qlen += q->q.qlen; 519 } 520 521 return qlen; 522 } 523 524 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 525 { 526 return (struct qdisc_skb_cb *)skb->cb; 527 } 528 529 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 530 { 531 return &qdisc->q.lock; 532 } 533 534 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 535 { 536 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 537 538 return q; 539 } 540 541 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 542 { 543 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 544 } 545 546 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 547 { 548 return rcu_dereference_rtnl(qdisc->dev_queue->qdisc_sleeping); 549 } 550 551 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 552 { 553 struct Qdisc *root = qdisc_root_sleeping(qdisc); 554 555 ASSERT_RTNL(); 556 return qdisc_lock(root); 557 } 558 559 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 560 { 561 return qdisc->dev_queue->dev; 562 } 563 564 static inline void sch_tree_lock(struct Qdisc *q) 565 { 566 if (q->flags & TCQ_F_MQROOT) 567 spin_lock_bh(qdisc_lock(q)); 568 else 569 spin_lock_bh(qdisc_root_sleeping_lock(q)); 570 } 571 572 static inline void sch_tree_unlock(struct Qdisc *q) 573 { 574 if (q->flags & TCQ_F_MQROOT) 575 spin_unlock_bh(qdisc_lock(q)); 576 else 577 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 578 } 579 580 extern struct Qdisc noop_qdisc; 581 extern struct Qdisc_ops noop_qdisc_ops; 582 extern struct Qdisc_ops pfifo_fast_ops; 583 extern struct Qdisc_ops mq_qdisc_ops; 584 extern struct Qdisc_ops noqueue_qdisc_ops; 585 extern const struct Qdisc_ops *default_qdisc_ops; 586 static inline const struct Qdisc_ops * 587 get_default_qdisc_ops(const struct net_device *dev, int ntx) 588 { 589 return ntx < dev->real_num_tx_queues ? 590 default_qdisc_ops : &pfifo_fast_ops; 591 } 592 593 struct Qdisc_class_common { 594 u32 classid; 595 struct hlist_node hnode; 596 }; 597 598 struct Qdisc_class_hash { 599 struct hlist_head *hash; 600 unsigned int hashsize; 601 unsigned int hashmask; 602 unsigned int hashelems; 603 }; 604 605 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 606 { 607 id ^= id >> 8; 608 id ^= id >> 4; 609 return id & mask; 610 } 611 612 static inline struct Qdisc_class_common * 613 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 614 { 615 struct Qdisc_class_common *cl; 616 unsigned int h; 617 618 if (!id) 619 return NULL; 620 621 h = qdisc_class_hash(id, hash->hashmask); 622 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 623 if (cl->classid == id) 624 return cl; 625 } 626 return NULL; 627 } 628 629 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 630 { 631 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 632 633 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 634 } 635 636 int qdisc_class_hash_init(struct Qdisc_class_hash *); 637 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 638 struct Qdisc_class_common *); 639 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 640 struct Qdisc_class_common *); 641 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 642 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 643 644 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 645 void dev_qdisc_change_real_num_tx(struct net_device *dev, 646 unsigned int new_real_tx); 647 void dev_init_scheduler(struct net_device *dev); 648 void dev_shutdown(struct net_device *dev); 649 void dev_activate(struct net_device *dev); 650 void dev_deactivate(struct net_device *dev); 651 void dev_deactivate_many(struct list_head *head); 652 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 653 struct Qdisc *qdisc); 654 void qdisc_reset(struct Qdisc *qdisc); 655 void qdisc_put(struct Qdisc *qdisc); 656 void qdisc_put_unlocked(struct Qdisc *qdisc); 657 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 658 #ifdef CONFIG_NET_SCHED 659 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 660 void *type_data); 661 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 662 struct Qdisc *new, struct Qdisc *old, 663 enum tc_setup_type type, void *type_data, 664 struct netlink_ext_ack *extack); 665 #else 666 static inline int 667 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 668 void *type_data) 669 { 670 q->flags &= ~TCQ_F_OFFLOADED; 671 return 0; 672 } 673 674 static inline void 675 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 676 struct Qdisc *new, struct Qdisc *old, 677 enum tc_setup_type type, void *type_data, 678 struct netlink_ext_ack *extack) 679 { 680 } 681 #endif 682 void qdisc_offload_query_caps(struct net_device *dev, 683 enum tc_setup_type type, 684 void *caps, size_t caps_len); 685 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 686 const struct Qdisc_ops *ops, 687 struct netlink_ext_ack *extack); 688 void qdisc_free(struct Qdisc *qdisc); 689 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 690 const struct Qdisc_ops *ops, u32 parentid, 691 struct netlink_ext_ack *extack); 692 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 693 const struct qdisc_size_table *stab); 694 int skb_do_redirect(struct sk_buff *); 695 696 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 697 { 698 #ifdef CONFIG_NET_CLS_ACT 699 return skb->tc_at_ingress; 700 #else 701 return false; 702 #endif 703 } 704 705 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 706 { 707 #ifdef CONFIG_NET_CLS_ACT 708 if (skb->tc_skip_classify) { 709 skb->tc_skip_classify = 0; 710 return true; 711 } 712 #endif 713 return false; 714 } 715 716 /* Reset all TX qdiscs greater than index of a device. */ 717 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 718 { 719 struct Qdisc *qdisc; 720 721 for (; i < dev->num_tx_queues; i++) { 722 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 723 if (qdisc) { 724 spin_lock_bh(qdisc_lock(qdisc)); 725 qdisc_reset(qdisc); 726 spin_unlock_bh(qdisc_lock(qdisc)); 727 } 728 } 729 } 730 731 /* Are all TX queues of the device empty? */ 732 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 733 { 734 unsigned int i; 735 736 rcu_read_lock(); 737 for (i = 0; i < dev->num_tx_queues; i++) { 738 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 739 const struct Qdisc *q = rcu_dereference(txq->qdisc); 740 741 if (!qdisc_is_empty(q)) { 742 rcu_read_unlock(); 743 return false; 744 } 745 } 746 rcu_read_unlock(); 747 return true; 748 } 749 750 /* Are any of the TX qdiscs changing? */ 751 static inline bool qdisc_tx_changing(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); 757 758 if (rcu_access_pointer(txq->qdisc) != 759 rcu_access_pointer(txq->qdisc_sleeping)) 760 return true; 761 } 762 return false; 763 } 764 765 /* Is the device using the noop qdisc on all queues? */ 766 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 767 { 768 unsigned int i; 769 770 for (i = 0; i < dev->num_tx_queues; i++) { 771 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 772 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 773 return false; 774 } 775 return true; 776 } 777 778 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 779 { 780 return qdisc_skb_cb(skb)->pkt_len; 781 } 782 783 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 784 enum net_xmit_qdisc_t { 785 __NET_XMIT_STOLEN = 0x00010000, 786 __NET_XMIT_BYPASS = 0x00020000, 787 }; 788 789 #ifdef CONFIG_NET_CLS_ACT 790 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 791 #else 792 #define net_xmit_drop_count(e) (1) 793 #endif 794 795 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 796 const struct Qdisc *sch) 797 { 798 #ifdef CONFIG_NET_SCHED 799 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 800 801 if (stab) 802 __qdisc_calculate_pkt_len(skb, stab); 803 #endif 804 } 805 806 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 807 struct sk_buff **to_free) 808 { 809 qdisc_calculate_pkt_len(skb, sch); 810 return sch->enqueue(skb, sch, to_free); 811 } 812 813 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats, 814 __u64 bytes, __u32 packets) 815 { 816 u64_stats_update_begin(&bstats->syncp); 817 u64_stats_add(&bstats->bytes, bytes); 818 u64_stats_add(&bstats->packets, packets); 819 u64_stats_update_end(&bstats->syncp); 820 } 821 822 static inline void bstats_update(struct gnet_stats_basic_sync *bstats, 823 const struct sk_buff *skb) 824 { 825 _bstats_update(bstats, 826 qdisc_pkt_len(skb), 827 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 828 } 829 830 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 831 const struct sk_buff *skb) 832 { 833 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb); 834 } 835 836 static inline void qdisc_bstats_update(struct Qdisc *sch, 837 const struct sk_buff *skb) 838 { 839 bstats_update(&sch->bstats, skb); 840 } 841 842 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 843 const struct sk_buff *skb) 844 { 845 sch->qstats.backlog -= qdisc_pkt_len(skb); 846 } 847 848 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 849 const struct sk_buff *skb) 850 { 851 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 852 } 853 854 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 855 const struct sk_buff *skb) 856 { 857 sch->qstats.backlog += qdisc_pkt_len(skb); 858 } 859 860 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 861 const struct sk_buff *skb) 862 { 863 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 864 } 865 866 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 867 { 868 this_cpu_inc(sch->cpu_qstats->qlen); 869 } 870 871 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 872 { 873 this_cpu_dec(sch->cpu_qstats->qlen); 874 } 875 876 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 877 { 878 this_cpu_inc(sch->cpu_qstats->requeues); 879 } 880 881 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 882 { 883 sch->qstats.drops += count; 884 } 885 886 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 887 { 888 qstats->drops++; 889 } 890 891 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 892 { 893 qstats->overlimits++; 894 } 895 896 static inline void qdisc_qstats_drop(struct Qdisc *sch) 897 { 898 qstats_drop_inc(&sch->qstats); 899 } 900 901 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 902 { 903 this_cpu_inc(sch->cpu_qstats->drops); 904 } 905 906 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 907 { 908 sch->qstats.overlimits++; 909 } 910 911 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 912 { 913 __u32 qlen = qdisc_qlen_sum(sch); 914 915 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 916 } 917 918 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 919 __u32 *backlog) 920 { 921 struct gnet_stats_queue qstats = { 0 }; 922 923 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats); 924 *qlen = qstats.qlen + qdisc_qlen(sch); 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_enqueue_tail(struct sk_buff *skb, 946 struct qdisc_skb_head *qh) 947 { 948 struct sk_buff *last = qh->tail; 949 950 if (last) { 951 skb->next = NULL; 952 last->next = skb; 953 qh->tail = skb; 954 } else { 955 qh->tail = skb; 956 qh->head = skb; 957 } 958 qh->qlen++; 959 } 960 961 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 962 { 963 __qdisc_enqueue_tail(skb, &sch->q); 964 qdisc_qstats_backlog_inc(sch, skb); 965 return NET_XMIT_SUCCESS; 966 } 967 968 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 969 struct qdisc_skb_head *qh) 970 { 971 skb->next = qh->head; 972 973 if (!qh->head) 974 qh->tail = skb; 975 qh->head = skb; 976 qh->qlen++; 977 } 978 979 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 980 { 981 struct sk_buff *skb = qh->head; 982 983 if (likely(skb != NULL)) { 984 qh->head = skb->next; 985 qh->qlen--; 986 if (qh->head == NULL) 987 qh->tail = NULL; 988 skb->next = NULL; 989 } 990 991 return skb; 992 } 993 994 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 995 { 996 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 997 998 if (likely(skb != NULL)) { 999 qdisc_qstats_backlog_dec(sch, skb); 1000 qdisc_bstats_update(sch, skb); 1001 } 1002 1003 return skb; 1004 } 1005 1006 /* Instead of calling kfree_skb() while root qdisc lock is held, 1007 * queue the skb for future freeing at end of __dev_xmit_skb() 1008 */ 1009 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1010 { 1011 skb->next = *to_free; 1012 *to_free = skb; 1013 } 1014 1015 static inline void __qdisc_drop_all(struct sk_buff *skb, 1016 struct sk_buff **to_free) 1017 { 1018 if (skb->prev) 1019 skb->prev->next = *to_free; 1020 else 1021 skb->next = *to_free; 1022 *to_free = skb; 1023 } 1024 1025 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1026 struct qdisc_skb_head *qh, 1027 struct sk_buff **to_free) 1028 { 1029 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1030 1031 if (likely(skb != NULL)) { 1032 unsigned int len = qdisc_pkt_len(skb); 1033 1034 qdisc_qstats_backlog_dec(sch, skb); 1035 __qdisc_drop(skb, to_free); 1036 return len; 1037 } 1038 1039 return 0; 1040 } 1041 1042 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1043 { 1044 const struct qdisc_skb_head *qh = &sch->q; 1045 1046 return qh->head; 1047 } 1048 1049 /* generic pseudo peek method for non-work-conserving qdisc */ 1050 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1051 { 1052 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1053 1054 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1055 if (!skb) { 1056 skb = sch->dequeue(sch); 1057 1058 if (skb) { 1059 __skb_queue_head(&sch->gso_skb, skb); 1060 /* it's still part of the queue */ 1061 qdisc_qstats_backlog_inc(sch, skb); 1062 sch->q.qlen++; 1063 } 1064 } 1065 1066 return skb; 1067 } 1068 1069 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1070 struct sk_buff *skb) 1071 { 1072 if (qdisc_is_percpu_stats(sch)) { 1073 qdisc_qstats_cpu_backlog_dec(sch, skb); 1074 qdisc_bstats_cpu_update(sch, skb); 1075 qdisc_qstats_cpu_qlen_dec(sch); 1076 } else { 1077 qdisc_qstats_backlog_dec(sch, skb); 1078 qdisc_bstats_update(sch, skb); 1079 sch->q.qlen--; 1080 } 1081 } 1082 1083 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1084 unsigned int pkt_len) 1085 { 1086 if (qdisc_is_percpu_stats(sch)) { 1087 qdisc_qstats_cpu_qlen_inc(sch); 1088 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1089 } else { 1090 sch->qstats.backlog += pkt_len; 1091 sch->q.qlen++; 1092 } 1093 } 1094 1095 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1096 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1097 { 1098 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1099 1100 if (skb) { 1101 skb = __skb_dequeue(&sch->gso_skb); 1102 if (qdisc_is_percpu_stats(sch)) { 1103 qdisc_qstats_cpu_backlog_dec(sch, skb); 1104 qdisc_qstats_cpu_qlen_dec(sch); 1105 } else { 1106 qdisc_qstats_backlog_dec(sch, skb); 1107 sch->q.qlen--; 1108 } 1109 } else { 1110 skb = sch->dequeue(sch); 1111 } 1112 1113 return skb; 1114 } 1115 1116 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1117 { 1118 /* 1119 * We do not know the backlog in bytes of this list, it 1120 * is up to the caller to correct it 1121 */ 1122 ASSERT_RTNL(); 1123 if (qh->qlen) { 1124 rtnl_kfree_skbs(qh->head, qh->tail); 1125 1126 qh->head = NULL; 1127 qh->tail = NULL; 1128 qh->qlen = 0; 1129 } 1130 } 1131 1132 static inline void qdisc_reset_queue(struct Qdisc *sch) 1133 { 1134 __qdisc_reset_queue(&sch->q); 1135 } 1136 1137 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1138 struct Qdisc **pold) 1139 { 1140 struct Qdisc *old; 1141 1142 sch_tree_lock(sch); 1143 old = *pold; 1144 *pold = new; 1145 if (old != NULL) 1146 qdisc_purge_queue(old); 1147 sch_tree_unlock(sch); 1148 1149 return old; 1150 } 1151 1152 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1153 { 1154 rtnl_kfree_skbs(skb, skb); 1155 qdisc_qstats_drop(sch); 1156 } 1157 1158 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1159 struct sk_buff **to_free) 1160 { 1161 __qdisc_drop(skb, to_free); 1162 qdisc_qstats_cpu_drop(sch); 1163 1164 return NET_XMIT_DROP; 1165 } 1166 1167 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1168 struct sk_buff **to_free) 1169 { 1170 __qdisc_drop(skb, to_free); 1171 qdisc_qstats_drop(sch); 1172 1173 return NET_XMIT_DROP; 1174 } 1175 1176 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1177 struct sk_buff **to_free) 1178 { 1179 __qdisc_drop_all(skb, to_free); 1180 qdisc_qstats_drop(sch); 1181 1182 return NET_XMIT_DROP; 1183 } 1184 1185 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1186 long it will take to send a packet given its size. 1187 */ 1188 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1189 { 1190 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1191 if (slot < 0) 1192 slot = 0; 1193 slot >>= rtab->rate.cell_log; 1194 if (slot > 255) 1195 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1196 return rtab->data[slot]; 1197 } 1198 1199 struct psched_ratecfg { 1200 u64 rate_bytes_ps; /* bytes per second */ 1201 u32 mult; 1202 u16 overhead; 1203 u16 mpu; 1204 u8 linklayer; 1205 u8 shift; 1206 }; 1207 1208 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1209 unsigned int len) 1210 { 1211 len += r->overhead; 1212 1213 if (len < r->mpu) 1214 len = r->mpu; 1215 1216 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1217 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1218 1219 return ((u64)len * r->mult) >> r->shift; 1220 } 1221 1222 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1223 const struct tc_ratespec *conf, 1224 u64 rate64); 1225 1226 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1227 const struct psched_ratecfg *r) 1228 { 1229 memset(res, 0, sizeof(*res)); 1230 1231 /* legacy struct tc_ratespec has a 32bit @rate field 1232 * Qdisc using 64bit rate should add new attributes 1233 * in order to maintain compatibility. 1234 */ 1235 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1236 1237 res->overhead = r->overhead; 1238 res->mpu = r->mpu; 1239 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1240 } 1241 1242 struct psched_pktrate { 1243 u64 rate_pkts_ps; /* packets per second */ 1244 u32 mult; 1245 u8 shift; 1246 }; 1247 1248 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1249 unsigned int pkt_num) 1250 { 1251 return ((u64)pkt_num * r->mult) >> r->shift; 1252 } 1253 1254 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1255 1256 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1257 * The fast path only needs to access filter list and to update stats 1258 */ 1259 struct mini_Qdisc { 1260 struct tcf_proto *filter_list; 1261 struct tcf_block *block; 1262 struct gnet_stats_basic_sync __percpu *cpu_bstats; 1263 struct gnet_stats_queue __percpu *cpu_qstats; 1264 unsigned long rcu_state; 1265 }; 1266 1267 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1268 const struct sk_buff *skb) 1269 { 1270 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1271 } 1272 1273 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1274 { 1275 this_cpu_inc(miniq->cpu_qstats->drops); 1276 } 1277 1278 struct mini_Qdisc_pair { 1279 struct mini_Qdisc miniq1; 1280 struct mini_Qdisc miniq2; 1281 struct mini_Qdisc __rcu **p_miniq; 1282 }; 1283 1284 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1285 struct tcf_proto *tp_head); 1286 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1287 struct mini_Qdisc __rcu **p_miniq); 1288 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1289 struct tcf_block *block); 1290 1291 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx); 1292 1293 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1294 1295 /* Make sure qdisc is no longer in SCHED state. */ 1296 static inline void qdisc_synchronize(const struct Qdisc *q) 1297 { 1298 while (test_bit(__QDISC_STATE_SCHED, &q->state)) 1299 msleep(1); 1300 } 1301 1302 #endif 1303