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 /* Only for !TCQ_F_NOLOCK qdisc. Never access it directly. 42 * Use qdisc_run_begin/end() or qdisc_is_running() instead. 43 */ 44 __QDISC_STATE_RUNNING, 45 }; 46 47 #define QDISC_STATE_MISSED BIT(__QDISC_STATE_MISSED) 48 #define QDISC_STATE_DRAINING BIT(__QDISC_STATE_DRAINING) 49 50 #define QDISC_STATE_NON_EMPTY (QDISC_STATE_MISSED | \ 51 QDISC_STATE_DRAINING) 52 53 struct qdisc_size_table { 54 struct rcu_head rcu; 55 struct list_head list; 56 struct tc_sizespec szopts; 57 int refcnt; 58 u16 data[]; 59 }; 60 61 /* similar to sk_buff_head, but skb->prev pointer is undefined. */ 62 struct qdisc_skb_head { 63 struct sk_buff *head; 64 struct sk_buff *tail; 65 __u32 qlen; 66 spinlock_t lock; 67 }; 68 69 struct Qdisc { 70 int (*enqueue)(struct sk_buff *skb, 71 struct Qdisc *sch, 72 struct sk_buff **to_free); 73 struct sk_buff * (*dequeue)(struct Qdisc *sch); 74 unsigned int flags; 75 #define TCQ_F_BUILTIN 1 76 #define TCQ_F_INGRESS 2 77 #define TCQ_F_CAN_BYPASS 4 78 #define TCQ_F_MQROOT 8 79 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for 80 * q->dev_queue : It can test 81 * netif_xmit_frozen_or_stopped() before 82 * dequeueing next packet. 83 * Its true for MQ/MQPRIO slaves, or non 84 * multiqueue device. 85 */ 86 #define TCQ_F_WARN_NONWC (1 << 16) 87 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */ 88 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy : 89 * qdisc_tree_decrease_qlen() should stop. 90 */ 91 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */ 92 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */ 93 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */ 94 u32 limit; 95 const struct Qdisc_ops *ops; 96 struct qdisc_size_table __rcu *stab; 97 struct hlist_node hash; 98 u32 handle; 99 u32 parent; 100 101 struct netdev_queue *dev_queue; 102 103 struct net_rate_estimator __rcu *rate_est; 104 struct gnet_stats_basic_sync __percpu *cpu_bstats; 105 struct gnet_stats_queue __percpu *cpu_qstats; 106 int pad; 107 refcount_t refcnt; 108 109 /* 110 * For performance sake on SMP, we put highly modified fields at the end 111 */ 112 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp; 113 struct qdisc_skb_head q; 114 struct gnet_stats_basic_sync bstats; 115 struct gnet_stats_queue qstats; 116 unsigned long state; 117 struct Qdisc *next_sched; 118 struct sk_buff_head skb_bad_txq; 119 120 spinlock_t busylock ____cacheline_aligned_in_smp; 121 spinlock_t seqlock; 122 123 struct rcu_head rcu; 124 125 /* private data */ 126 long privdata[] ____cacheline_aligned; 127 }; 128 129 static inline void qdisc_refcount_inc(struct Qdisc *qdisc) 130 { 131 if (qdisc->flags & TCQ_F_BUILTIN) 132 return; 133 refcount_inc(&qdisc->refcnt); 134 } 135 136 /* Intended to be used by unlocked users, when concurrent qdisc release is 137 * possible. 138 */ 139 140 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc) 141 { 142 if (qdisc->flags & TCQ_F_BUILTIN) 143 return qdisc; 144 if (refcount_inc_not_zero(&qdisc->refcnt)) 145 return qdisc; 146 return NULL; 147 } 148 149 /* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc 150 * root_lock section, or provide their own memory barriers -- ordering 151 * against qdisc_run_begin/end() atomic bit operations. 152 */ 153 static inline bool qdisc_is_running(struct Qdisc *qdisc) 154 { 155 if (qdisc->flags & TCQ_F_NOLOCK) 156 return spin_is_locked(&qdisc->seqlock); 157 return test_bit(__QDISC_STATE_RUNNING, &qdisc->state); 158 } 159 160 static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc) 161 { 162 return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY); 163 } 164 165 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) 166 { 167 return q->flags & TCQ_F_CPUSTATS; 168 } 169 170 static inline bool qdisc_is_empty(const struct Qdisc *qdisc) 171 { 172 if (qdisc_is_percpu_stats(qdisc)) 173 return nolock_qdisc_is_empty(qdisc); 174 return !READ_ONCE(qdisc->q.qlen); 175 } 176 177 /* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with 178 * the qdisc root lock acquired. 179 */ 180 static inline bool qdisc_run_begin(struct Qdisc *qdisc) 181 { 182 if (qdisc->flags & TCQ_F_NOLOCK) { 183 if (spin_trylock(&qdisc->seqlock)) 184 return true; 185 186 /* Paired with smp_mb__after_atomic() to make sure 187 * STATE_MISSED checking is synchronized with clearing 188 * in pfifo_fast_dequeue(). 189 */ 190 smp_mb__before_atomic(); 191 192 /* If the MISSED flag is set, it means other thread has 193 * set the MISSED flag before second spin_trylock(), so 194 * we can return false here to avoid multi cpus doing 195 * the set_bit() and second spin_trylock() concurrently. 196 */ 197 if (test_bit(__QDISC_STATE_MISSED, &qdisc->state)) 198 return false; 199 200 /* Set the MISSED flag before the second spin_trylock(), 201 * if the second spin_trylock() return false, it means 202 * other cpu holding the lock will do dequeuing for us 203 * or it will see the MISSED flag set after releasing 204 * lock and reschedule the net_tx_action() to do the 205 * dequeuing. 206 */ 207 set_bit(__QDISC_STATE_MISSED, &qdisc->state); 208 209 /* spin_trylock() only has load-acquire semantic, so use 210 * smp_mb__after_atomic() to ensure STATE_MISSED is set 211 * before doing the second spin_trylock(). 212 */ 213 smp_mb__after_atomic(); 214 215 /* Retry again in case other CPU may not see the new flag 216 * after it releases the lock at the end of qdisc_run_end(). 217 */ 218 return spin_trylock(&qdisc->seqlock); 219 } 220 return !test_and_set_bit(__QDISC_STATE_RUNNING, &qdisc->state); 221 } 222 223 static inline void qdisc_run_end(struct Qdisc *qdisc) 224 { 225 if (qdisc->flags & TCQ_F_NOLOCK) { 226 spin_unlock(&qdisc->seqlock); 227 228 if (unlikely(test_bit(__QDISC_STATE_MISSED, 229 &qdisc->state))) 230 __netif_schedule(qdisc); 231 } else { 232 clear_bit(__QDISC_STATE_RUNNING, &qdisc->state); 233 } 234 } 235 236 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 237 { 238 return qdisc->flags & TCQ_F_ONETXQUEUE; 239 } 240 241 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) 242 { 243 #ifdef CONFIG_BQL 244 /* Non-BQL migrated drivers will return 0, too. */ 245 return dql_avail(&txq->dql); 246 #else 247 return 0; 248 #endif 249 } 250 251 struct Qdisc_class_ops { 252 unsigned int flags; 253 /* Child qdisc manipulation */ 254 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 255 int (*graft)(struct Qdisc *, unsigned long cl, 256 struct Qdisc *, struct Qdisc **, 257 struct netlink_ext_ack *extack); 258 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 259 void (*qlen_notify)(struct Qdisc *, unsigned long); 260 261 /* Class manipulation routines */ 262 unsigned long (*find)(struct Qdisc *, u32 classid); 263 int (*change)(struct Qdisc *, u32, u32, 264 struct nlattr **, unsigned long *, 265 struct netlink_ext_ack *); 266 int (*delete)(struct Qdisc *, unsigned long, 267 struct netlink_ext_ack *); 268 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 269 270 /* Filter manipulation */ 271 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 272 unsigned long arg, 273 struct netlink_ext_ack *extack); 274 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 275 u32 classid); 276 void (*unbind_tcf)(struct Qdisc *, unsigned long); 277 278 /* rtnetlink specific */ 279 int (*dump)(struct Qdisc *, unsigned long, 280 struct sk_buff *skb, struct tcmsg*); 281 int (*dump_stats)(struct Qdisc *, unsigned long, 282 struct gnet_dump *); 283 }; 284 285 /* Qdisc_class_ops flag values */ 286 287 /* Implements API that doesn't require rtnl lock */ 288 enum qdisc_class_ops_flags { 289 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 290 }; 291 292 struct Qdisc_ops { 293 struct Qdisc_ops *next; 294 const struct Qdisc_class_ops *cl_ops; 295 char id[IFNAMSIZ]; 296 int priv_size; 297 unsigned int static_flags; 298 299 int (*enqueue)(struct sk_buff *skb, 300 struct Qdisc *sch, 301 struct sk_buff **to_free); 302 struct sk_buff * (*dequeue)(struct Qdisc *); 303 struct sk_buff * (*peek)(struct Qdisc *); 304 305 int (*init)(struct Qdisc *sch, struct nlattr *arg, 306 struct netlink_ext_ack *extack); 307 void (*reset)(struct Qdisc *); 308 void (*destroy)(struct Qdisc *); 309 int (*change)(struct Qdisc *sch, 310 struct nlattr *arg, 311 struct netlink_ext_ack *extack); 312 void (*attach)(struct Qdisc *sch); 313 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 314 void (*change_real_num_tx)(struct Qdisc *sch, 315 unsigned int new_real_tx); 316 317 int (*dump)(struct Qdisc *, struct sk_buff *); 318 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 319 320 void (*ingress_block_set)(struct Qdisc *sch, 321 u32 block_index); 322 void (*egress_block_set)(struct Qdisc *sch, 323 u32 block_index); 324 u32 (*ingress_block_get)(struct Qdisc *sch); 325 u32 (*egress_block_get)(struct Qdisc *sch); 326 327 struct module *owner; 328 }; 329 330 331 struct tcf_result { 332 union { 333 struct { 334 unsigned long class; 335 u32 classid; 336 }; 337 const struct tcf_proto *goto_tp; 338 339 /* used in the skb_tc_reinsert function */ 340 struct { 341 bool ingress; 342 struct gnet_stats_queue *qstats; 343 }; 344 }; 345 }; 346 347 struct tcf_chain; 348 349 struct tcf_proto_ops { 350 struct list_head head; 351 char kind[IFNAMSIZ]; 352 353 int (*classify)(struct sk_buff *, 354 const struct tcf_proto *, 355 struct tcf_result *); 356 int (*init)(struct tcf_proto*); 357 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 358 struct netlink_ext_ack *extack); 359 360 void* (*get)(struct tcf_proto*, u32 handle); 361 void (*put)(struct tcf_proto *tp, void *f); 362 int (*change)(struct net *net, struct sk_buff *, 363 struct tcf_proto*, unsigned long, 364 u32 handle, struct nlattr **, 365 void **, u32, 366 struct netlink_ext_ack *); 367 int (*delete)(struct tcf_proto *tp, void *arg, 368 bool *last, bool rtnl_held, 369 struct netlink_ext_ack *); 370 bool (*delete_empty)(struct tcf_proto *tp); 371 void (*walk)(struct tcf_proto *tp, 372 struct tcf_walker *arg, bool rtnl_held); 373 int (*reoffload)(struct tcf_proto *tp, bool add, 374 flow_setup_cb_t *cb, void *cb_priv, 375 struct netlink_ext_ack *extack); 376 void (*hw_add)(struct tcf_proto *tp, 377 void *type_data); 378 void (*hw_del)(struct tcf_proto *tp, 379 void *type_data); 380 void (*bind_class)(void *, u32, unsigned long, 381 void *, unsigned long); 382 void * (*tmplt_create)(struct net *net, 383 struct tcf_chain *chain, 384 struct nlattr **tca, 385 struct netlink_ext_ack *extack); 386 void (*tmplt_destroy)(void *tmplt_priv); 387 388 /* rtnetlink specific */ 389 int (*dump)(struct net*, struct tcf_proto*, void *, 390 struct sk_buff *skb, struct tcmsg*, 391 bool); 392 int (*terse_dump)(struct net *net, 393 struct tcf_proto *tp, void *fh, 394 struct sk_buff *skb, 395 struct tcmsg *t, bool rtnl_held); 396 int (*tmplt_dump)(struct sk_buff *skb, 397 struct net *net, 398 void *tmplt_priv); 399 400 struct module *owner; 401 int flags; 402 }; 403 404 /* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags 405 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race 406 * conditions can occur when filters are inserted/deleted simultaneously. 407 */ 408 enum tcf_proto_ops_flags { 409 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 410 }; 411 412 struct tcf_proto { 413 /* Fast access part */ 414 struct tcf_proto __rcu *next; 415 void __rcu *root; 416 417 /* called under RCU BH lock*/ 418 int (*classify)(struct sk_buff *, 419 const struct tcf_proto *, 420 struct tcf_result *); 421 __be16 protocol; 422 423 /* All the rest */ 424 u32 prio; 425 void *data; 426 const struct tcf_proto_ops *ops; 427 struct tcf_chain *chain; 428 /* Lock protects tcf_proto shared state and can be used by unlocked 429 * classifiers to protect their private data. 430 */ 431 spinlock_t lock; 432 bool deleting; 433 refcount_t refcnt; 434 struct rcu_head rcu; 435 struct hlist_node destroy_ht_node; 436 }; 437 438 struct qdisc_skb_cb { 439 struct { 440 unsigned int pkt_len; 441 u16 slave_dev_queue_mapping; 442 u16 tc_classid; 443 }; 444 #define QDISC_CB_PRIV_LEN 20 445 unsigned char data[QDISC_CB_PRIV_LEN]; 446 u16 mru; 447 bool post_ct; 448 }; 449 450 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 451 452 struct tcf_chain { 453 /* Protects filter_chain. */ 454 struct mutex filter_chain_lock; 455 struct tcf_proto __rcu *filter_chain; 456 struct list_head list; 457 struct tcf_block *block; 458 u32 index; /* chain index */ 459 unsigned int refcnt; 460 unsigned int action_refcnt; 461 bool explicitly_created; 462 bool flushing; 463 const struct tcf_proto_ops *tmplt_ops; 464 void *tmplt_priv; 465 struct rcu_head rcu; 466 }; 467 468 struct tcf_block { 469 /* Lock protects tcf_block and lifetime-management data of chains 470 * attached to the block (refcnt, action_refcnt, explicitly_created). 471 */ 472 struct mutex lock; 473 struct list_head chain_list; 474 u32 index; /* block index for shared blocks */ 475 u32 classid; /* which class this block belongs to */ 476 refcount_t refcnt; 477 struct net *net; 478 struct Qdisc *q; 479 struct rw_semaphore cb_lock; /* protects cb_list and offload counters */ 480 struct flow_block flow_block; 481 struct list_head owner_list; 482 bool keep_dst; 483 atomic_t offloadcnt; /* Number of oddloaded filters */ 484 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 485 unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */ 486 struct { 487 struct tcf_chain *chain; 488 struct list_head filter_chain_list; 489 } chain0; 490 struct rcu_head rcu; 491 DECLARE_HASHTABLE(proto_destroy_ht, 7); 492 struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */ 493 }; 494 495 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 496 { 497 return lockdep_is_held(&chain->filter_chain_lock); 498 } 499 500 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 501 { 502 return lockdep_is_held(&tp->lock); 503 } 504 505 #define tcf_chain_dereference(p, chain) \ 506 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 507 508 #define tcf_proto_dereference(p, tp) \ 509 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 510 511 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 512 { 513 struct qdisc_skb_cb *qcb; 514 515 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb)); 516 BUILD_BUG_ON(sizeof(qcb->data) < sz); 517 } 518 519 static inline int qdisc_qlen_cpu(const struct Qdisc *q) 520 { 521 return this_cpu_ptr(q->cpu_qstats)->qlen; 522 } 523 524 static inline int qdisc_qlen(const struct Qdisc *q) 525 { 526 return q->q.qlen; 527 } 528 529 static inline int qdisc_qlen_sum(const struct Qdisc *q) 530 { 531 __u32 qlen = q->qstats.qlen; 532 int i; 533 534 if (qdisc_is_percpu_stats(q)) { 535 for_each_possible_cpu(i) 536 qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; 537 } else { 538 qlen += q->q.qlen; 539 } 540 541 return qlen; 542 } 543 544 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 545 { 546 return (struct qdisc_skb_cb *)skb->cb; 547 } 548 549 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 550 { 551 return &qdisc->q.lock; 552 } 553 554 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 555 { 556 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 557 558 return q; 559 } 560 561 static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc) 562 { 563 return rcu_dereference_bh(qdisc->dev_queue->qdisc); 564 } 565 566 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 567 { 568 return qdisc->dev_queue->qdisc_sleeping; 569 } 570 571 /* The qdisc root lock is a mechanism by which to top level 572 * of a qdisc tree can be locked from any qdisc node in the 573 * forest. This allows changing the configuration of some 574 * aspect of the qdisc tree while blocking out asynchronous 575 * qdisc access in the packet processing paths. 576 * 577 * It is only legal to do this when the root will not change 578 * on us. Otherwise we'll potentially lock the wrong qdisc 579 * root. This is enforced by holding the RTNL semaphore, which 580 * all users of this lock accessor must do. 581 */ 582 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 583 { 584 struct Qdisc *root = qdisc_root(qdisc); 585 586 ASSERT_RTNL(); 587 return qdisc_lock(root); 588 } 589 590 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 591 { 592 struct Qdisc *root = qdisc_root_sleeping(qdisc); 593 594 ASSERT_RTNL(); 595 return qdisc_lock(root); 596 } 597 598 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 599 { 600 return qdisc->dev_queue->dev; 601 } 602 603 static inline void sch_tree_lock(struct Qdisc *q) 604 { 605 if (q->flags & TCQ_F_MQROOT) 606 spin_lock_bh(qdisc_lock(q)); 607 else 608 spin_lock_bh(qdisc_root_sleeping_lock(q)); 609 } 610 611 static inline void sch_tree_unlock(struct Qdisc *q) 612 { 613 if (q->flags & TCQ_F_MQROOT) 614 spin_unlock_bh(qdisc_lock(q)); 615 else 616 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 617 } 618 619 extern struct Qdisc noop_qdisc; 620 extern struct Qdisc_ops noop_qdisc_ops; 621 extern struct Qdisc_ops pfifo_fast_ops; 622 extern struct Qdisc_ops mq_qdisc_ops; 623 extern struct Qdisc_ops noqueue_qdisc_ops; 624 extern const struct Qdisc_ops *default_qdisc_ops; 625 static inline const struct Qdisc_ops * 626 get_default_qdisc_ops(const struct net_device *dev, int ntx) 627 { 628 return ntx < dev->real_num_tx_queues ? 629 default_qdisc_ops : &pfifo_fast_ops; 630 } 631 632 struct Qdisc_class_common { 633 u32 classid; 634 struct hlist_node hnode; 635 }; 636 637 struct Qdisc_class_hash { 638 struct hlist_head *hash; 639 unsigned int hashsize; 640 unsigned int hashmask; 641 unsigned int hashelems; 642 }; 643 644 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 645 { 646 id ^= id >> 8; 647 id ^= id >> 4; 648 return id & mask; 649 } 650 651 static inline struct Qdisc_class_common * 652 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 653 { 654 struct Qdisc_class_common *cl; 655 unsigned int h; 656 657 if (!id) 658 return NULL; 659 660 h = qdisc_class_hash(id, hash->hashmask); 661 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 662 if (cl->classid == id) 663 return cl; 664 } 665 return NULL; 666 } 667 668 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 669 { 670 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 671 672 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 673 } 674 675 int qdisc_class_hash_init(struct Qdisc_class_hash *); 676 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 677 struct Qdisc_class_common *); 678 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 679 struct Qdisc_class_common *); 680 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 681 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 682 683 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 684 void dev_qdisc_change_real_num_tx(struct net_device *dev, 685 unsigned int new_real_tx); 686 void dev_init_scheduler(struct net_device *dev); 687 void dev_shutdown(struct net_device *dev); 688 void dev_activate(struct net_device *dev); 689 void dev_deactivate(struct net_device *dev); 690 void dev_deactivate_many(struct list_head *head); 691 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 692 struct Qdisc *qdisc); 693 void qdisc_reset(struct Qdisc *qdisc); 694 void qdisc_put(struct Qdisc *qdisc); 695 void qdisc_put_unlocked(struct Qdisc *qdisc); 696 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 697 #ifdef CONFIG_NET_SCHED 698 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 699 void *type_data); 700 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 701 struct Qdisc *new, struct Qdisc *old, 702 enum tc_setup_type type, void *type_data, 703 struct netlink_ext_ack *extack); 704 #else 705 static inline int 706 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 707 void *type_data) 708 { 709 q->flags &= ~TCQ_F_OFFLOADED; 710 return 0; 711 } 712 713 static inline void 714 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 715 struct Qdisc *new, struct Qdisc *old, 716 enum tc_setup_type type, void *type_data, 717 struct netlink_ext_ack *extack) 718 { 719 } 720 #endif 721 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 722 const struct Qdisc_ops *ops, 723 struct netlink_ext_ack *extack); 724 void qdisc_free(struct Qdisc *qdisc); 725 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 726 const struct Qdisc_ops *ops, u32 parentid, 727 struct netlink_ext_ack *extack); 728 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 729 const struct qdisc_size_table *stab); 730 int skb_do_redirect(struct sk_buff *); 731 732 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 733 { 734 #ifdef CONFIG_NET_CLS_ACT 735 return skb->tc_at_ingress; 736 #else 737 return false; 738 #endif 739 } 740 741 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 742 { 743 #ifdef CONFIG_NET_CLS_ACT 744 if (skb->tc_skip_classify) { 745 skb->tc_skip_classify = 0; 746 return true; 747 } 748 #endif 749 return false; 750 } 751 752 /* Reset all TX qdiscs greater than index of a device. */ 753 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 754 { 755 struct Qdisc *qdisc; 756 757 for (; i < dev->num_tx_queues; i++) { 758 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 759 if (qdisc) { 760 spin_lock_bh(qdisc_lock(qdisc)); 761 qdisc_reset(qdisc); 762 spin_unlock_bh(qdisc_lock(qdisc)); 763 } 764 } 765 } 766 767 /* Are all TX queues of the device empty? */ 768 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 769 { 770 unsigned int i; 771 772 rcu_read_lock(); 773 for (i = 0; i < dev->num_tx_queues; i++) { 774 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 775 const struct Qdisc *q = rcu_dereference(txq->qdisc); 776 777 if (!qdisc_is_empty(q)) { 778 rcu_read_unlock(); 779 return false; 780 } 781 } 782 rcu_read_unlock(); 783 return true; 784 } 785 786 /* Are any of the TX qdiscs changing? */ 787 static inline bool qdisc_tx_changing(const struct net_device *dev) 788 { 789 unsigned int i; 790 791 for (i = 0; i < dev->num_tx_queues; i++) { 792 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 793 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) 794 return true; 795 } 796 return false; 797 } 798 799 /* Is the device using the noop qdisc on all queues? */ 800 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 801 { 802 unsigned int i; 803 804 for (i = 0; i < dev->num_tx_queues; i++) { 805 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 806 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 807 return false; 808 } 809 return true; 810 } 811 812 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 813 { 814 return qdisc_skb_cb(skb)->pkt_len; 815 } 816 817 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 818 enum net_xmit_qdisc_t { 819 __NET_XMIT_STOLEN = 0x00010000, 820 __NET_XMIT_BYPASS = 0x00020000, 821 }; 822 823 #ifdef CONFIG_NET_CLS_ACT 824 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 825 #else 826 #define net_xmit_drop_count(e) (1) 827 #endif 828 829 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 830 const struct Qdisc *sch) 831 { 832 #ifdef CONFIG_NET_SCHED 833 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 834 835 if (stab) 836 __qdisc_calculate_pkt_len(skb, stab); 837 #endif 838 } 839 840 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 841 struct sk_buff **to_free) 842 { 843 qdisc_calculate_pkt_len(skb, sch); 844 return sch->enqueue(skb, sch, to_free); 845 } 846 847 static inline void _bstats_update(struct gnet_stats_basic_sync *bstats, 848 __u64 bytes, __u32 packets) 849 { 850 u64_stats_update_begin(&bstats->syncp); 851 u64_stats_add(&bstats->bytes, bytes); 852 u64_stats_add(&bstats->packets, packets); 853 u64_stats_update_end(&bstats->syncp); 854 } 855 856 static inline void bstats_update(struct gnet_stats_basic_sync *bstats, 857 const struct sk_buff *skb) 858 { 859 _bstats_update(bstats, 860 qdisc_pkt_len(skb), 861 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 862 } 863 864 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 865 const struct sk_buff *skb) 866 { 867 bstats_update(this_cpu_ptr(sch->cpu_bstats), skb); 868 } 869 870 static inline void qdisc_bstats_update(struct Qdisc *sch, 871 const struct sk_buff *skb) 872 { 873 bstats_update(&sch->bstats, skb); 874 } 875 876 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 877 const struct sk_buff *skb) 878 { 879 sch->qstats.backlog -= qdisc_pkt_len(skb); 880 } 881 882 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 883 const struct sk_buff *skb) 884 { 885 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 886 } 887 888 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 889 const struct sk_buff *skb) 890 { 891 sch->qstats.backlog += qdisc_pkt_len(skb); 892 } 893 894 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 895 const struct sk_buff *skb) 896 { 897 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 898 } 899 900 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) 901 { 902 this_cpu_inc(sch->cpu_qstats->qlen); 903 } 904 905 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) 906 { 907 this_cpu_dec(sch->cpu_qstats->qlen); 908 } 909 910 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 911 { 912 this_cpu_inc(sch->cpu_qstats->requeues); 913 } 914 915 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 916 { 917 sch->qstats.drops += count; 918 } 919 920 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 921 { 922 qstats->drops++; 923 } 924 925 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 926 { 927 qstats->overlimits++; 928 } 929 930 static inline void qdisc_qstats_drop(struct Qdisc *sch) 931 { 932 qstats_drop_inc(&sch->qstats); 933 } 934 935 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 936 { 937 this_cpu_inc(sch->cpu_qstats->drops); 938 } 939 940 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 941 { 942 sch->qstats.overlimits++; 943 } 944 945 static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch) 946 { 947 __u32 qlen = qdisc_qlen_sum(sch); 948 949 return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen); 950 } 951 952 static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch, __u32 *qlen, 953 __u32 *backlog) 954 { 955 struct gnet_stats_queue qstats = { 0 }; 956 957 gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats); 958 *qlen = qstats.qlen + qdisc_qlen(sch); 959 *backlog = qstats.backlog; 960 } 961 962 static inline void qdisc_tree_flush_backlog(struct Qdisc *sch) 963 { 964 __u32 qlen, backlog; 965 966 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 967 qdisc_tree_reduce_backlog(sch, qlen, backlog); 968 } 969 970 static inline void qdisc_purge_queue(struct Qdisc *sch) 971 { 972 __u32 qlen, backlog; 973 974 qdisc_qstats_qlen_backlog(sch, &qlen, &backlog); 975 qdisc_reset(sch); 976 qdisc_tree_reduce_backlog(sch, qlen, backlog); 977 } 978 979 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh) 980 { 981 qh->head = NULL; 982 qh->tail = NULL; 983 qh->qlen = 0; 984 } 985 986 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 987 struct qdisc_skb_head *qh) 988 { 989 struct sk_buff *last = qh->tail; 990 991 if (last) { 992 skb->next = NULL; 993 last->next = skb; 994 qh->tail = skb; 995 } else { 996 qh->tail = skb; 997 qh->head = skb; 998 } 999 qh->qlen++; 1000 } 1001 1002 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 1003 { 1004 __qdisc_enqueue_tail(skb, &sch->q); 1005 qdisc_qstats_backlog_inc(sch, skb); 1006 return NET_XMIT_SUCCESS; 1007 } 1008 1009 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 1010 struct qdisc_skb_head *qh) 1011 { 1012 skb->next = qh->head; 1013 1014 if (!qh->head) 1015 qh->tail = skb; 1016 qh->head = skb; 1017 qh->qlen++; 1018 } 1019 1020 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 1021 { 1022 struct sk_buff *skb = qh->head; 1023 1024 if (likely(skb != NULL)) { 1025 qh->head = skb->next; 1026 qh->qlen--; 1027 if (qh->head == NULL) 1028 qh->tail = NULL; 1029 skb->next = NULL; 1030 } 1031 1032 return skb; 1033 } 1034 1035 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 1036 { 1037 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 1038 1039 if (likely(skb != NULL)) { 1040 qdisc_qstats_backlog_dec(sch, skb); 1041 qdisc_bstats_update(sch, skb); 1042 } 1043 1044 return skb; 1045 } 1046 1047 /* Instead of calling kfree_skb() while root qdisc lock is held, 1048 * queue the skb for future freeing at end of __dev_xmit_skb() 1049 */ 1050 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 1051 { 1052 skb->next = *to_free; 1053 *to_free = skb; 1054 } 1055 1056 static inline void __qdisc_drop_all(struct sk_buff *skb, 1057 struct sk_buff **to_free) 1058 { 1059 if (skb->prev) 1060 skb->prev->next = *to_free; 1061 else 1062 skb->next = *to_free; 1063 *to_free = skb; 1064 } 1065 1066 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1067 struct qdisc_skb_head *qh, 1068 struct sk_buff **to_free) 1069 { 1070 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1071 1072 if (likely(skb != NULL)) { 1073 unsigned int len = qdisc_pkt_len(skb); 1074 1075 qdisc_qstats_backlog_dec(sch, skb); 1076 __qdisc_drop(skb, to_free); 1077 return len; 1078 } 1079 1080 return 0; 1081 } 1082 1083 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1084 { 1085 const struct qdisc_skb_head *qh = &sch->q; 1086 1087 return qh->head; 1088 } 1089 1090 /* generic pseudo peek method for non-work-conserving qdisc */ 1091 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1092 { 1093 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1094 1095 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1096 if (!skb) { 1097 skb = sch->dequeue(sch); 1098 1099 if (skb) { 1100 __skb_queue_head(&sch->gso_skb, skb); 1101 /* it's still part of the queue */ 1102 qdisc_qstats_backlog_inc(sch, skb); 1103 sch->q.qlen++; 1104 } 1105 } 1106 1107 return skb; 1108 } 1109 1110 static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch, 1111 struct sk_buff *skb) 1112 { 1113 if (qdisc_is_percpu_stats(sch)) { 1114 qdisc_qstats_cpu_backlog_dec(sch, skb); 1115 qdisc_bstats_cpu_update(sch, skb); 1116 qdisc_qstats_cpu_qlen_dec(sch); 1117 } else { 1118 qdisc_qstats_backlog_dec(sch, skb); 1119 qdisc_bstats_update(sch, skb); 1120 sch->q.qlen--; 1121 } 1122 } 1123 1124 static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch, 1125 unsigned int pkt_len) 1126 { 1127 if (qdisc_is_percpu_stats(sch)) { 1128 qdisc_qstats_cpu_qlen_inc(sch); 1129 this_cpu_add(sch->cpu_qstats->backlog, pkt_len); 1130 } else { 1131 sch->qstats.backlog += pkt_len; 1132 sch->q.qlen++; 1133 } 1134 } 1135 1136 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1137 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1138 { 1139 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1140 1141 if (skb) { 1142 skb = __skb_dequeue(&sch->gso_skb); 1143 if (qdisc_is_percpu_stats(sch)) { 1144 qdisc_qstats_cpu_backlog_dec(sch, skb); 1145 qdisc_qstats_cpu_qlen_dec(sch); 1146 } else { 1147 qdisc_qstats_backlog_dec(sch, skb); 1148 sch->q.qlen--; 1149 } 1150 } else { 1151 skb = sch->dequeue(sch); 1152 } 1153 1154 return skb; 1155 } 1156 1157 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1158 { 1159 /* 1160 * We do not know the backlog in bytes of this list, it 1161 * is up to the caller to correct it 1162 */ 1163 ASSERT_RTNL(); 1164 if (qh->qlen) { 1165 rtnl_kfree_skbs(qh->head, qh->tail); 1166 1167 qh->head = NULL; 1168 qh->tail = NULL; 1169 qh->qlen = 0; 1170 } 1171 } 1172 1173 static inline void qdisc_reset_queue(struct Qdisc *sch) 1174 { 1175 __qdisc_reset_queue(&sch->q); 1176 sch->qstats.backlog = 0; 1177 } 1178 1179 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1180 struct Qdisc **pold) 1181 { 1182 struct Qdisc *old; 1183 1184 sch_tree_lock(sch); 1185 old = *pold; 1186 *pold = new; 1187 if (old != NULL) 1188 qdisc_purge_queue(old); 1189 sch_tree_unlock(sch); 1190 1191 return old; 1192 } 1193 1194 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1195 { 1196 rtnl_kfree_skbs(skb, skb); 1197 qdisc_qstats_drop(sch); 1198 } 1199 1200 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1201 struct sk_buff **to_free) 1202 { 1203 __qdisc_drop(skb, to_free); 1204 qdisc_qstats_cpu_drop(sch); 1205 1206 return NET_XMIT_DROP; 1207 } 1208 1209 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1210 struct sk_buff **to_free) 1211 { 1212 __qdisc_drop(skb, to_free); 1213 qdisc_qstats_drop(sch); 1214 1215 return NET_XMIT_DROP; 1216 } 1217 1218 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1219 struct sk_buff **to_free) 1220 { 1221 __qdisc_drop_all(skb, to_free); 1222 qdisc_qstats_drop(sch); 1223 1224 return NET_XMIT_DROP; 1225 } 1226 1227 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1228 long it will take to send a packet given its size. 1229 */ 1230 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1231 { 1232 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1233 if (slot < 0) 1234 slot = 0; 1235 slot >>= rtab->rate.cell_log; 1236 if (slot > 255) 1237 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1238 return rtab->data[slot]; 1239 } 1240 1241 struct psched_ratecfg { 1242 u64 rate_bytes_ps; /* bytes per second */ 1243 u32 mult; 1244 u16 overhead; 1245 u8 linklayer; 1246 u8 shift; 1247 }; 1248 1249 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1250 unsigned int len) 1251 { 1252 len += r->overhead; 1253 1254 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1255 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1256 1257 return ((u64)len * r->mult) >> r->shift; 1258 } 1259 1260 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1261 const struct tc_ratespec *conf, 1262 u64 rate64); 1263 1264 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1265 const struct psched_ratecfg *r) 1266 { 1267 memset(res, 0, sizeof(*res)); 1268 1269 /* legacy struct tc_ratespec has a 32bit @rate field 1270 * Qdisc using 64bit rate should add new attributes 1271 * in order to maintain compatibility. 1272 */ 1273 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1274 1275 res->overhead = r->overhead; 1276 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1277 } 1278 1279 struct psched_pktrate { 1280 u64 rate_pkts_ps; /* packets per second */ 1281 u32 mult; 1282 u8 shift; 1283 }; 1284 1285 static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r, 1286 unsigned int pkt_num) 1287 { 1288 return ((u64)pkt_num * r->mult) >> r->shift; 1289 } 1290 1291 void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64); 1292 1293 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1294 * The fast path only needs to access filter list and to update stats 1295 */ 1296 struct mini_Qdisc { 1297 struct tcf_proto *filter_list; 1298 struct tcf_block *block; 1299 struct gnet_stats_basic_sync __percpu *cpu_bstats; 1300 struct gnet_stats_queue __percpu *cpu_qstats; 1301 struct rcu_head rcu; 1302 }; 1303 1304 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1305 const struct sk_buff *skb) 1306 { 1307 bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1308 } 1309 1310 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1311 { 1312 this_cpu_inc(miniq->cpu_qstats->drops); 1313 } 1314 1315 struct mini_Qdisc_pair { 1316 struct mini_Qdisc miniq1; 1317 struct mini_Qdisc miniq2; 1318 struct mini_Qdisc __rcu **p_miniq; 1319 }; 1320 1321 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1322 struct tcf_proto *tp_head); 1323 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1324 struct mini_Qdisc __rcu **p_miniq); 1325 void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp, 1326 struct tcf_block *block); 1327 1328 void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx); 1329 1330 int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb)); 1331 1332 #endif 1333