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 /* Paired with smp_mb__after_atomic() to make sure 191 * STATE_MISSED checking is synchronized with clearing 192 * in pfifo_fast_dequeue(). 193 */ 194 smp_mb__before_atomic(); 195 196 /* If the MISSED flag is set, it means other thread has 197 * set the MISSED flag before second spin_trylock(), so 198 * we can return false here to avoid multi cpus doing 199 * the set_bit() and second spin_trylock() concurrently. 200 */ 201 if (test_bit(__QDISC_STATE_MISSED, &qdisc->state)) 202 return false; 203 204 /* Set the MISSED flag before the second spin_trylock(), 205 * if the second spin_trylock() return false, it means 206 * other cpu holding the lock will do dequeuing for us 207 * or it will see the MISSED flag set after releasing 208 * lock and reschedule the net_tx_action() to do the 209 * dequeuing. 210 */ 211 set_bit(__QDISC_STATE_MISSED, &qdisc->state); 212 213 /* spin_trylock() only has load-acquire semantic, so use 214 * smp_mb__after_atomic() to ensure STATE_MISSED is set 215 * before doing the second spin_trylock(). 216 */ 217 smp_mb__after_atomic(); 218 219 /* Retry again in case other CPU may not see the new flag 220 * after it releases the lock at the end of qdisc_run_end(). 221 */ 222 return spin_trylock(&qdisc->seqlock); 223 } 224 return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 225 } 226 227 static inline void qdisc_run_end(struct Qdisc *qdisc) 228 { 229 if (qdisc->flags & TCQ_F_NOLOCK) { 230 spin_unlock(&qdisc->seqlock); 231 232 if (unlikely(test_bit(__QDISC_STATE_MISSED, 233 &qdisc->state))) 234 __netif_schedule(qdisc); 235 } else { 236 __clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2); 237 } 238 } 239 240 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 241 { 242 return qdisc->flags & TCQ_F_ONETXQUEUE; 243 } 244 245 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) 246 { 247 #ifdef CONFIG_BQL 248 /* Non-BQL migrated drivers will return 0, too. */ 249 return dql_avail(&txq->dql); 250 #else 251 return 0; 252 #endif 253 } 254 255 struct Qdisc_class_ops { 256 unsigned int flags; 257 /* Child qdisc manipulation */ 258 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 259 int (*graft)(struct Qdisc *, unsigned long cl, 260 struct Qdisc *, struct Qdisc **, 261 struct netlink_ext_ack *extack); 262 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 263 void (*qlen_notify)(struct Qdisc *, unsigned long); 264 265 /* Class manipulation routines */ 266 unsigned long (*find)(struct Qdisc *, u32 classid); 267 int (*change)(struct Qdisc *, u32, u32, 268 struct nlattr **, unsigned long *, 269 struct netlink_ext_ack *); 270 int (*delete)(struct Qdisc *, unsigned long, 271 struct netlink_ext_ack *); 272 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 273 274 /* Filter manipulation */ 275 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 276 unsigned long arg, 277 struct netlink_ext_ack *extack); 278 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 279 u32 classid); 280 void (*unbind_tcf)(struct Qdisc *, unsigned long); 281 282 /* rtnetlink specific */ 283 int (*dump)(struct Qdisc *, unsigned long, 284 struct sk_buff *skb, struct tcmsg*); 285 int (*dump_stats)(struct Qdisc *, unsigned long, 286 struct gnet_dump *); 287 }; 288 289 /* Qdisc_class_ops flag values */ 290 291 /* Implements API that doesn't require rtnl lock */ 292 enum qdisc_class_ops_flags { 293 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 294 }; 295 296 struct Qdisc_ops { 297 struct Qdisc_ops *next; 298 const struct Qdisc_class_ops *cl_ops; 299 char id[IFNAMSIZ]; 300 int priv_size; 301 unsigned int static_flags; 302 303 int (*enqueue)(struct sk_buff *skb, 304 struct Qdisc *sch, 305 struct sk_buff **to_free); 306 struct sk_buff * (*dequeue)(struct Qdisc *); 307 struct sk_buff * (*peek)(struct Qdisc *); 308 309 int (*init)(struct Qdisc *sch, struct nlattr *arg, 310 struct netlink_ext_ack *extack); 311 void (*reset)(struct Qdisc *); 312 void (*destroy)(struct Qdisc *); 313 int (*change)(struct Qdisc *sch, 314 struct nlattr *arg, 315 struct netlink_ext_ack *extack); 316 void (*attach)(struct Qdisc *sch); 317 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 318 void (*change_real_num_tx)(struct Qdisc *sch, 319 unsigned int new_real_tx); 320 321 int (*dump)(struct Qdisc *, struct sk_buff *); 322 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 323 324 void (*ingress_block_set)(struct Qdisc *sch, 325 u32 block_index); 326 void (*egress_block_set)(struct Qdisc *sch, 327 u32 block_index); 328 u32 (*ingress_block_get)(struct Qdisc *sch); 329 u32 (*egress_block_get)(struct Qdisc *sch); 330 331 struct module *owner; 332 }; 333 334 335 struct tcf_result { 336 union { 337 struct { 338 unsigned long class; 339 u32 classid; 340 }; 341 const struct tcf_proto *goto_tp; 342 343 /* used in the skb_tc_reinsert function */ 344 struct { 345 bool ingress; 346 struct gnet_stats_queue *qstats; 347 }; 348 }; 349 }; 350 351 struct tcf_chain; 352 353 struct tcf_proto_ops { 354 struct list_head head; 355 char kind[IFNAMSIZ]; 356 357 int (*classify)(struct sk_buff *, 358 const struct tcf_proto *, 359 struct tcf_result *); 360 int (*init)(struct tcf_proto*); 361 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 362 struct netlink_ext_ack *extack); 363 364 void* (*get)(struct tcf_proto*, u32 handle); 365 void (*put)(struct tcf_proto *tp, void *f); 366 int (*change)(struct net *net, struct sk_buff *, 367 struct tcf_proto*, unsigned long, 368 u32 handle, struct nlattr **, 369 void **, u32, 370 struct netlink_ext_ack *); 371 int (*delete)(struct tcf_proto *tp, void *arg, 372 bool *last, bool rtnl_held, 373 struct netlink_ext_ack *); 374 bool (*delete_empty)(struct tcf_proto *tp); 375 void (*walk)(struct tcf_proto *tp, 376 struct tcf_walker *arg, bool rtnl_held); 377 int (*reoffload)(struct tcf_proto *tp, bool add, 378 flow_setup_cb_t *cb, void *cb_priv, 379 struct netlink_ext_ack *extack); 380 void (*hw_add)(struct tcf_proto *tp, 381 void *type_data); 382 void (*hw_del)(struct tcf_proto *tp, 383 void *type_data); 384 void (*bind_class)(void *, u32, unsigned long, 385 void *, unsigned long); 386 void * (*tmplt_create)(struct net *net, 387 struct tcf_chain *chain, 388 struct nlattr **tca, 389 struct netlink_ext_ack *extack); 390 void (*tmplt_destroy)(void *tmplt_priv); 391 392 /* rtnetlink specific */ 393 int (*dump)(struct net*, struct tcf_proto*, void *, 394 struct sk_buff *skb, struct tcmsg*, 395 bool); 396 int (*terse_dump)(struct net *net, 397 struct tcf_proto *tp, void *fh, 398 struct sk_buff *skb, 399 struct tcmsg *t, bool rtnl_held); 400 int (*tmplt_dump)(struct sk_buff *skb, 401 struct net *net, 402 void *tmplt_priv); 403 404 struct module *owner; 405 int flags; 406 }; 407 408 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 409 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 410 * conditions can occur when filters are inserted/deleted simultaneously. 411 */ 412 enum tcf_proto_ops_flags { 413 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 414 }; 415 416 struct tcf_proto { 417 /* Fast access part */ 418 struct tcf_proto __rcu *next; 419 void __rcu *root; 420 421 /* called under RCU BH lock*/ 422 int (*classify)(struct sk_buff *, 423 const struct tcf_proto *, 424 struct tcf_result *); 425 __be16 protocol; 426 427 /* All the rest */ 428 u32 prio; 429 void *data; 430 const struct tcf_proto_ops *ops; 431 struct tcf_chain *chain; 432 /* Lock protects tcf_proto shared state and can be used by unlocked 433 * classifiers to protect their private data. 434 */ 435 spinlock_t lock; 436 bool deleting; 437 refcount_t refcnt; 438 struct rcu_head rcu; 439 struct hlist_node destroy_ht_node; 440 }; 441 442 struct qdisc_skb_cb { 443 struct { 444 unsigned int pkt_len; 445 u16 slave_dev_queue_mapping; 446 u16 tc_classid; 447 }; 448 #define QDISC_CB_PRIV_LEN 20 449 unsigned char data[QDISC_CB_PRIV_LEN]; 450 }; 451 452 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 453 454 struct tcf_chain { 455 /* Protects filter_chain. */ 456 struct mutex filter_chain_lock; 457 struct tcf_proto __rcu *filter_chain; 458 struct list_head list; 459 struct tcf_block *block; 460 u32 index; /* chain index */ 461 unsigned int refcnt; 462 unsigned int action_refcnt; 463 bool explicitly_created; 464 bool flushing; 465 const struct tcf_proto_ops *tmplt_ops; 466 void *tmplt_priv; 467 struct rcu_head rcu; 468 }; 469 470 struct tcf_block { 471 /* Lock protects tcf_block and lifetime-management data of chains 472 * attached to the block (refcnt, action_refcnt, explicitly_created). 473 */ 474 struct mutex lock; 475 struct list_head chain_list; 476 u32 index; /* block index for shared blocks */ 477 u32 classid; /* which class this block belongs to */ 478 refcount_t refcnt; 479 struct net *net; 480 struct Qdisc *q; 481 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 482 struct flow_block flow_block; 483 struct list_head owner_list; 484 bool keep_dst; 485 atomic_t offloadcnt; /* Number of oddloaded filters */ 486 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 487 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 488 struct { 489 struct tcf_chain *chain; 490 struct list_head filter_chain_list; 491 } chain0; 492 struct rcu_head rcu; 493 DECLARE_HASHTABLE(proto_destroy_ht, 7); 494 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 495 }; 496 497 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 498 { 499 return lockdep_is_held(&chain->filter_chain_lock); 500 } 501 502 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 503 { 504 return lockdep_is_held(&tp->lock); 505 } 506 507 #define tcf_chain_dereference(p, chain) \ 508 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 509 510 #define tcf_proto_dereference(p, tp) \ 511 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 512 513 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 514 { 515 struct qdisc_skb_cb *qcb; 516 517 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 518 BUILD_BUG_ON(sizeof(qcb->data) < sz); 519 } 520 521 static inline int qdisc_qlen(const struct Qdisc *q) 522 { 523 return q->q.qlen; 524 } 525 526 static inline int qdisc_qlen_sum(const struct Qdisc *q) 527 { 528 __u32 qlen = q->qstats.qlen; 529 int i; 530 531 if (qdisc_is_percpu_stats(q)) { 532 for_each_possible_cpu(i) 533 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 534 } else { 535 qlen += q->q.qlen; 536 } 537 538 return qlen; 539 } 540 541 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 542 { 543 return (struct qdisc_skb_cb *)skb->cb; 544 } 545 546 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 547 { 548 return &qdisc->q.lock; 549 } 550 551 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 552 { 553 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 554 555 return q; 556 } 557 558 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 559 { 560 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 561 } 562 563 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 564 { 565 return qdisc->dev_queue->qdisc_sleeping; 566 } 567 568 /* The qdisc root lock is a mechanism by which to top level 569 * of a qdisc tree can be locked from any qdisc node in the 570 * forest. This allows changing the configuration of some 571 * aspect of the qdisc tree while blocking out asynchronous 572 * qdisc access in the packet processing paths. 573 * 574 * It is only legal to do this when the root will not change 575 * on us. Otherwise we'll potentially lock the wrong qdisc 576 * root. This is enforced by holding the RTNL semaphore, which 577 * all users of this lock accessor must do. 578 */ 579 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 580 { 581 struct Qdisc *root = qdisc_root(qdisc); 582 583 ASSERT_RTNL(); 584 return qdisc_lock(root); 585 } 586 587 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 588 { 589 struct Qdisc *root = qdisc_root_sleeping(qdisc); 590 591 ASSERT_RTNL(); 592 return qdisc_lock(root); 593 } 594 595 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 596 { 597 return qdisc->dev_queue->dev; 598 } 599 600 static inline void sch_tree_lock(struct Qdisc *q) 601 { 602 if (q->flags & TCQ_F_MQROOT) 603 spin_lock_bh(qdisc_lock(q)); 604 else 605 spin_lock_bh(qdisc_root_sleeping_lock(q)); 606 } 607 608 static inline void sch_tree_unlock(struct Qdisc *q) 609 { 610 if (q->flags & TCQ_F_MQROOT) 611 spin_unlock_bh(qdisc_lock(q)); 612 else 613 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 614 } 615 616 extern struct Qdisc noop_qdisc; 617 extern struct Qdisc_ops noop_qdisc_ops; 618 extern struct Qdisc_ops pfifo_fast_ops; 619 extern struct Qdisc_ops mq_qdisc_ops; 620 extern struct Qdisc_ops noqueue_qdisc_ops; 621 extern const struct Qdisc_ops *default_qdisc_ops; 622 static inline const struct Qdisc_ops * 623 get_default_qdisc_ops(const struct net_device *dev, int ntx) 624 { 625 return ntx < dev->real_num_tx_queues ? 626 default_qdisc_ops : &pfifo_fast_ops; 627 } 628 629 struct Qdisc_class_common { 630 u32 classid; 631 struct hlist_node hnode; 632 }; 633 634 struct Qdisc_class_hash { 635 struct hlist_head *hash; 636 unsigned int hashsize; 637 unsigned int hashmask; 638 unsigned int hashelems; 639 }; 640 641 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 642 { 643 id ^= id >> 8; 644 id ^= id >> 4; 645 return id & mask; 646 } 647 648 static inline struct Qdisc_class_common * 649 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 650 { 651 struct Qdisc_class_common *cl; 652 unsigned int h; 653 654 if (!id) 655 return NULL; 656 657 h = qdisc_class_hash(id, hash->hashmask); 658 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 659 if (cl->classid == id) 660 return cl; 661 } 662 return NULL; 663 } 664 665 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 666 { 667 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 668 669 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 670 } 671 672 int qdisc_class_hash_init(struct Qdisc_class_hash *); 673 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 674 struct Qdisc_class_common *); 675 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 676 struct Qdisc_class_common *); 677 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 678 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 679 680 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 681 void dev_qdisc_change_real_num_tx(struct net_device *dev, 682 unsigned int new_real_tx); 683 void dev_init_scheduler(struct net_device *dev); 684 void dev_shutdown(struct net_device *dev); 685 void dev_activate(struct net_device *dev); 686 void dev_deactivate(struct net_device *dev); 687 void dev_deactivate_many(struct list_head *head); 688 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 689 struct Qdisc *qdisc); 690 void qdisc_reset(struct Qdisc *qdisc); 691 void qdisc_put(struct Qdisc *qdisc); 692 void qdisc_put_unlocked(struct Qdisc *qdisc); 693 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 694 #ifdef CONFIG_NET_SCHED 695 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 696 void *type_data); 697 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 698 struct Qdisc *new, struct Qdisc *old, 699 enum tc_setup_type type, void *type_data, 700 struct netlink_ext_ack *extack); 701 #else 702 static inline int 703 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 704 void *type_data) 705 { 706 q->flags &= ~TCQ_F_OFFLOADED; 707 return 0; 708 } 709 710 static inline void 711 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 712 struct Qdisc *new, struct Qdisc *old, 713 enum tc_setup_type type, void *type_data, 714 struct netlink_ext_ack *extack) 715 { 716 } 717 #endif 718 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 719 const struct Qdisc_ops *ops, 720 struct netlink_ext_ack *extack); 721 void qdisc_free(struct Qdisc *qdisc); 722 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 723 const struct Qdisc_ops *ops, u32 parentid, 724 struct netlink_ext_ack *extack); 725 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 726 const struct qdisc_size_table *stab); 727 int skb_do_redirect(struct sk_buff *); 728 729 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 730 { 731 #ifdef CONFIG_NET_CLS_ACT 732 return skb->tc_at_ingress; 733 #else 734 return false; 735 #endif 736 } 737 738 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 739 { 740 #ifdef CONFIG_NET_CLS_ACT 741 if (skb->tc_skip_classify) { 742 skb->tc_skip_classify = 0; 743 return true; 744 } 745 #endif 746 return false; 747 } 748 749 /* Reset all TX qdiscs greater than index of a device. */ 750 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 751 { 752 struct Qdisc *qdisc; 753 754 for (; i < dev->num_tx_queues; i++) { 755 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 756 if (qdisc) { 757 spin_lock_bh(qdisc_lock(qdisc)); 758 qdisc_reset(qdisc); 759 spin_unlock_bh(qdisc_lock(qdisc)); 760 } 761 } 762 } 763 764 /* Are all TX queues of the device empty? */ 765 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 766 { 767 unsigned int i; 768 769 rcu_read_lock(); 770 for (i = 0; i < dev->num_tx_queues; i++) { 771 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 772 const struct Qdisc *q = rcu_dereference(txq->qdisc); 773 774 if (!qdisc_is_empty(q)) { 775 rcu_read_unlock(); 776 return false; 777 } 778 } 779 rcu_read_unlock(); 780 return true; 781 } 782 783 /* Are any of the TX qdiscs changing? */ 784 static inline bool qdisc_tx_changing(const struct net_device *dev) 785 { 786 unsigned int i; 787 788 for (i = 0; i < dev->num_tx_queues; i++) { 789 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 790 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) 791 return true; 792 } 793 return false; 794 } 795 796 /* Is the device using the noop qdisc on all queues? */ 797 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 798 { 799 unsigned int i; 800 801 for (i = 0; i < dev->num_tx_queues; i++) { 802 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 803 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 804 return false; 805 } 806 return true; 807 } 808 809 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 810 { 811 return qdisc_skb_cb(skb)->pkt_len; 812 } 813 814 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 815 enum net_xmit_qdisc_t { 816 __NET_XMIT_STOLEN = 0x00010000, 817 __NET_XMIT_BYPASS = 0x00020000, 818 }; 819 820 #ifdef CONFIG_NET_CLS_ACT 821 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 822 #else 823 #define net_xmit_drop_count(e) (1) 824 #endif 825 826 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 827 const struct Qdisc *sch) 828 { 829 #ifdef CONFIG_NET_SCHED 830 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 831 832 if (stab) 833 __qdisc_calculate_pkt_len(skb, stab); 834 #endif 835 } 836 837 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 838 struct sk_buff **to_free) 839 { 840 qdisc_calculate_pkt_len(skb, sch); 841 return sch->enqueue(skb, sch, to_free); 842 } 843 844 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats, 845 __u64 bytes, __u32 packets) 846 { 847 u64_stats_update_begin(&bstats->syncp); 848 u64_stats_add(&bstats->bytes, bytes); 849 u64_stats_add(&bstats->packets, packets); 850 u64_stats_update_end(&bstats->syncp); 851 } 852 853 static inline void bstats_update(struct gnet_stats_basic_sync *bstats, 854 const struct sk_buff *skb) 855 { 856 _bstats_update(bstats, 857 qdisc_pkt_len(skb), 858 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 859 } 860 861 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 862 const struct sk_buff *skb) 863 { 864 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb); 865 } 866 867 static inline void qdisc_bstats_update(struct Qdisc *sch, 868 const struct sk_buff *skb) 869 { 870 bstats_update(&sch->bstats, skb); 871 } 872 873 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 874 const struct sk_buff *skb) 875 { 876 sch->qstats.backlog -= qdisc_pkt_len(skb); 877 } 878 879 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 880 const struct sk_buff *skb) 881 { 882 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 883 } 884 885 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 886 const struct sk_buff *skb) 887 { 888 sch->qstats.backlog += qdisc_pkt_len(skb); 889 } 890 891 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 892 const struct sk_buff *skb) 893 { 894 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 895 } 896 897 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 898 { 899 this_cpu_inc(sch->cpu_qstats->qlen); 900 } 901 902 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 903 { 904 this_cpu_dec(sch->cpu_qstats->qlen); 905 } 906 907 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 908 { 909 this_cpu_inc(sch->cpu_qstats->requeues); 910 } 911 912 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 913 { 914 sch->qstats.drops += count; 915 } 916 917 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 918 { 919 qstats->drops++; 920 } 921 922 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 923 { 924 qstats->overlimits++; 925 } 926 927 static inline void qdisc_qstats_drop(struct Qdisc *sch) 928 { 929 qstats_drop_inc(&sch->qstats); 930 } 931 932 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 933 { 934 this_cpu_inc(sch->cpu_qstats->drops); 935 } 936 937 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 938 { 939 sch->qstats.overlimits++; 940 } 941 942 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 943 { 944 __u32 qlen = qdisc_qlen_sum(sch); 945 946 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 947 } 948 949 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 950 __u32 *backlog) 951 { 952 struct gnet_stats_queue qstats = { 0 }; 953 954 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats); 955 *qlen = qstats.qlen + qdisc_qlen(sch); 956 *backlog = qstats.backlog; 957 } 958 959 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 960 { 961 __u32 qlen, backlog; 962 963 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 964 qdisc_tree_reduce_backlog(sch, qlen, backlog); 965 } 966 967 static inline void qdisc_purge_queue(struct Qdisc *sch) 968 { 969 __u32 qlen, backlog; 970 971 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 972 qdisc_reset(sch); 973 qdisc_tree_reduce_backlog(sch, qlen, backlog); 974 } 975 976 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh) 977 { 978 qh->head = NULL; 979 qh->tail = NULL; 980 qh->qlen = 0; 981 } 982 983 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 984 struct qdisc_skb_head *qh) 985 { 986 struct sk_buff *last = qh->tail; 987 988 if (last) { 989 skb->next = NULL; 990 last->next = skb; 991 qh->tail = skb; 992 } else { 993 qh->tail = skb; 994 qh->head = skb; 995 } 996 qh->qlen++; 997 } 998 999 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 1000 { 1001 __qdisc_enqueue_tail(skb, &sch->q); 1002 qdisc_qstats_backlog_inc(sch, skb); 1003 return NET_XMIT_SUCCESS; 1004 } 1005 1006 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 1007 struct qdisc_skb_head *qh) 1008 { 1009 skb->next = qh->head; 1010 1011 if (!qh->head) 1012 qh->tail = skb; 1013 qh->head = skb; 1014 qh->qlen++; 1015 } 1016 1017 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 1018 { 1019 struct sk_buff *skb = qh->head; 1020 1021 if (likely(skb != NULL)) { 1022 qh->head = skb->next; 1023 qh->qlen--; 1024 if (qh->head == NULL) 1025 qh->tail = NULL; 1026 skb->next = NULL; 1027 } 1028 1029 return skb; 1030 } 1031 1032 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1033 { 1034 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1035 1036 if (likely(skb != NULL)) { 1037 qdisc_qstats_backlog_dec(sch, skb); 1038 qdisc_bstats_update(sch, skb); 1039 } 1040 1041 return skb; 1042 } 1043 1044 /* Instead of calling kfree_skb() while root qdisc lock is held, 1045 * queue the skb for future freeing at end of __dev_xmit_skb() 1046 */ 1047 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1048 { 1049 skb->next = *to_free; 1050 *to_free = skb; 1051 } 1052 1053 static inline void __qdisc_drop_all(struct sk_buff *skb, 1054 struct sk_buff **to_free) 1055 { 1056 if (skb->prev) 1057 skb->prev->next = *to_free; 1058 else 1059 skb->next = *to_free; 1060 *to_free = skb; 1061 } 1062 1063 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1064 struct qdisc_skb_head *qh, 1065 struct sk_buff **to_free) 1066 { 1067 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1068 1069 if (likely(skb != NULL)) { 1070 unsigned int len = qdisc_pkt_len(skb); 1071 1072 qdisc_qstats_backlog_dec(sch, skb); 1073 __qdisc_drop(skb, to_free); 1074 return len; 1075 } 1076 1077 return 0; 1078 } 1079 1080 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1081 { 1082 const struct qdisc_skb_head *qh = &sch->q; 1083 1084 return qh->head; 1085 } 1086 1087 /* generic pseudo peek method for non-work-conserving qdisc */ 1088 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1089 { 1090 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1091 1092 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1093 if (!skb) { 1094 skb = sch->dequeue(sch); 1095 1096 if (skb) { 1097 __skb_queue_head(&sch->gso_skb, skb); 1098 /* it's still part of the queue */ 1099 qdisc_qstats_backlog_inc(sch, skb); 1100 sch->q.qlen++; 1101 } 1102 } 1103 1104 return skb; 1105 } 1106 1107 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1108 struct sk_buff *skb) 1109 { 1110 if (qdisc_is_percpu_stats(sch)) { 1111 qdisc_qstats_cpu_backlog_dec(sch, skb); 1112 qdisc_bstats_cpu_update(sch, skb); 1113 qdisc_qstats_cpu_qlen_dec(sch); 1114 } else { 1115 qdisc_qstats_backlog_dec(sch, skb); 1116 qdisc_bstats_update(sch, skb); 1117 sch->q.qlen--; 1118 } 1119 } 1120 1121 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1122 unsigned int pkt_len) 1123 { 1124 if (qdisc_is_percpu_stats(sch)) { 1125 qdisc_qstats_cpu_qlen_inc(sch); 1126 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1127 } else { 1128 sch->qstats.backlog += pkt_len; 1129 sch->q.qlen++; 1130 } 1131 } 1132 1133 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1134 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1135 { 1136 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1137 1138 if (skb) { 1139 skb = __skb_dequeue(&sch->gso_skb); 1140 if (qdisc_is_percpu_stats(sch)) { 1141 qdisc_qstats_cpu_backlog_dec(sch, skb); 1142 qdisc_qstats_cpu_qlen_dec(sch); 1143 } else { 1144 qdisc_qstats_backlog_dec(sch, skb); 1145 sch->q.qlen--; 1146 } 1147 } else { 1148 skb = sch->dequeue(sch); 1149 } 1150 1151 return skb; 1152 } 1153 1154 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1155 { 1156 /* 1157 * We do not know the backlog in bytes of this list, it 1158 * is up to the caller to correct it 1159 */ 1160 ASSERT_RTNL(); 1161 if (qh->qlen) { 1162 rtnl_kfree_skbs(qh->head, qh->tail); 1163 1164 qh->head = NULL; 1165 qh->tail = NULL; 1166 qh->qlen = 0; 1167 } 1168 } 1169 1170 static inline void qdisc_reset_queue(struct Qdisc *sch) 1171 { 1172 __qdisc_reset_queue(&sch->q); 1173 sch->qstats.backlog = 0; 1174 } 1175 1176 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1177 struct Qdisc **pold) 1178 { 1179 struct Qdisc *old; 1180 1181 sch_tree_lock(sch); 1182 old = *pold; 1183 *pold = new; 1184 if (old != NULL) 1185 qdisc_purge_queue(old); 1186 sch_tree_unlock(sch); 1187 1188 return old; 1189 } 1190 1191 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1192 { 1193 rtnl_kfree_skbs(skb, skb); 1194 qdisc_qstats_drop(sch); 1195 } 1196 1197 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1198 struct sk_buff **to_free) 1199 { 1200 __qdisc_drop(skb, to_free); 1201 qdisc_qstats_cpu_drop(sch); 1202 1203 return NET_XMIT_DROP; 1204 } 1205 1206 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1207 struct sk_buff **to_free) 1208 { 1209 __qdisc_drop(skb, to_free); 1210 qdisc_qstats_drop(sch); 1211 1212 return NET_XMIT_DROP; 1213 } 1214 1215 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1216 struct sk_buff **to_free) 1217 { 1218 __qdisc_drop_all(skb, to_free); 1219 qdisc_qstats_drop(sch); 1220 1221 return NET_XMIT_DROP; 1222 } 1223 1224 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1225 long it will take to send a packet given its size. 1226 */ 1227 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1228 { 1229 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1230 if (slot < 0) 1231 slot = 0; 1232 slot >>= rtab->rate.cell_log; 1233 if (slot > 255) 1234 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1235 return rtab->data[slot]; 1236 } 1237 1238 struct psched_ratecfg { 1239 u64 rate_bytes_ps; /* bytes per second */ 1240 u32 mult; 1241 u16 overhead; 1242 u16 mpu; 1243 u8 linklayer; 1244 u8 shift; 1245 }; 1246 1247 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1248 unsigned int len) 1249 { 1250 len += r->overhead; 1251 1252 if (len < r->mpu) 1253 len = r->mpu; 1254 1255 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1256 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1257 1258 return ((u64)len * r->mult) >> r->shift; 1259 } 1260 1261 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1262 const struct tc_ratespec *conf, 1263 u64 rate64); 1264 1265 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1266 const struct psched_ratecfg *r) 1267 { 1268 memset(res, 0, sizeof(*res)); 1269 1270 /* legacy struct tc_ratespec has a 32bit @rate field 1271 * Qdisc using 64bit rate should add new attributes 1272 * in order to maintain compatibility. 1273 */ 1274 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1275 1276 res->overhead = r->overhead; 1277 res->mpu = r->mpu; 1278 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1279 } 1280 1281 struct psched_pktrate { 1282 u64 rate_pkts_ps; /* packets per second */ 1283 u32 mult; 1284 u8 shift; 1285 }; 1286 1287 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1288 unsigned int pkt_num) 1289 { 1290 return ((u64)pkt_num * r->mult) >> r->shift; 1291 } 1292 1293 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1294 1295 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1296 * The fast path only needs to access filter list and to update stats 1297 */ 1298 struct mini_Qdisc { 1299 struct tcf_proto *filter_list; 1300 struct tcf_block *block; 1301 struct gnet_stats_basic_sync __percpu *cpu_bstats; 1302 struct gnet_stats_queue __percpu *cpu_qstats; 1303 unsigned long rcu_state; 1304 }; 1305 1306 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1307 const struct sk_buff *skb) 1308 { 1309 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1310 } 1311 1312 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1313 { 1314 this_cpu_inc(miniq->cpu_qstats->drops); 1315 } 1316 1317 struct mini_Qdisc_pair { 1318 struct mini_Qdisc miniq1; 1319 struct mini_Qdisc miniq2; 1320 struct mini_Qdisc __rcu **p_miniq; 1321 }; 1322 1323 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1324 struct tcf_proto *tp_head); 1325 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1326 struct mini_Qdisc __rcu **p_miniq); 1327 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1328 struct tcf_block *block); 1329 1330 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx); 1331 1332 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1333 1334 #endif 1335