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 <net/gen_stats.h> 17 #include <net/rtnetlink.h> 18 19 struct Qdisc_ops; 20 struct qdisc_walker; 21 struct tcf_walker; 22 struct module; 23 struct bpf_flow_keys; 24 25 typedef int tc_setup_cb_t(enum tc_setup_type type, 26 void *type_data, void *cb_priv); 27 28 typedef int tc_indr_block_bind_cb_t(struct net_device *dev, void *cb_priv, 29 enum tc_setup_type type, void *type_data); 30 31 struct qdisc_rate_table { 32 struct tc_ratespec rate; 33 u32 data[256]; 34 struct qdisc_rate_table *next; 35 int refcnt; 36 }; 37 38 enum qdisc_state_t { 39 __QDISC_STATE_SCHED, 40 __QDISC_STATE_DEACTIVATED, 41 }; 42 43 struct qdisc_size_table { 44 struct rcu_head rcu; 45 struct list_head list; 46 struct tc_sizespec szopts; 47 int refcnt; 48 u16 data[]; 49 }; 50 51 /* similar to sk_buff_head, but skb->prev pointer is undefined. */ 52 struct qdisc_skb_head { 53 struct sk_buff *head; 54 struct sk_buff *tail; 55 union { 56 u32 qlen; 57 atomic_t atomic_qlen; 58 }; 59 spinlock_t lock; 60 }; 61 62 struct Qdisc { 63 int (*enqueue)(struct sk_buff *skb, 64 struct Qdisc *sch, 65 struct sk_buff **to_free); 66 struct sk_buff * (*dequeue)(struct Qdisc *sch); 67 unsigned int flags; 68 #define TCQ_F_BUILTIN 1 69 #define TCQ_F_INGRESS 2 70 #define TCQ_F_CAN_BYPASS 4 71 #define TCQ_F_MQROOT 8 72 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for 73 * q->dev_queue : It can test 74 * netif_xmit_frozen_or_stopped() before 75 * dequeueing next packet. 76 * Its true for MQ/MQPRIO slaves, or non 77 * multiqueue device. 78 */ 79 #define TCQ_F_WARN_NONWC (1 << 16) 80 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */ 81 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy : 82 * qdisc_tree_decrease_qlen() should stop. 83 */ 84 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */ 85 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */ 86 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */ 87 u32 limit; 88 const struct Qdisc_ops *ops; 89 struct qdisc_size_table __rcu *stab; 90 struct hlist_node hash; 91 u32 handle; 92 u32 parent; 93 94 struct netdev_queue *dev_queue; 95 96 struct net_rate_estimator __rcu *rate_est; 97 struct gnet_stats_basic_cpu __percpu *cpu_bstats; 98 struct gnet_stats_queue __percpu *cpu_qstats; 99 int padded; 100 refcount_t refcnt; 101 102 /* 103 * For performance sake on SMP, we put highly modified fields at the end 104 */ 105 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp; 106 struct qdisc_skb_head q; 107 struct gnet_stats_basic_packed bstats; 108 seqcount_t running; 109 struct gnet_stats_queue qstats; 110 unsigned long state; 111 struct Qdisc *next_sched; 112 struct sk_buff_head skb_bad_txq; 113 114 spinlock_t busylock ____cacheline_aligned_in_smp; 115 spinlock_t seqlock; 116 struct rcu_head rcu; 117 }; 118 119 static inline void qdisc_refcount_inc(struct Qdisc *qdisc) 120 { 121 if (qdisc->flags & TCQ_F_BUILTIN) 122 return; 123 refcount_inc(&qdisc->refcnt); 124 } 125 126 /* Intended to be used by unlocked users, when concurrent qdisc release is 127 * possible. 128 */ 129 130 static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc) 131 { 132 if (qdisc->flags & TCQ_F_BUILTIN) 133 return qdisc; 134 if (refcount_inc_not_zero(&qdisc->refcnt)) 135 return qdisc; 136 return NULL; 137 } 138 139 static inline bool qdisc_is_running(struct Qdisc *qdisc) 140 { 141 if (qdisc->flags & TCQ_F_NOLOCK) 142 return spin_is_locked(&qdisc->seqlock); 143 return (raw_read_seqcount(&qdisc->running) & 1) ? true : false; 144 } 145 146 static inline bool qdisc_run_begin(struct Qdisc *qdisc) 147 { 148 if (qdisc->flags & TCQ_F_NOLOCK) { 149 if (!spin_trylock(&qdisc->seqlock)) 150 return false; 151 } else if (qdisc_is_running(qdisc)) { 152 return false; 153 } 154 /* Variant of write_seqcount_begin() telling lockdep a trylock 155 * was attempted. 156 */ 157 raw_write_seqcount_begin(&qdisc->running); 158 seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_); 159 return true; 160 } 161 162 static inline void qdisc_run_end(struct Qdisc *qdisc) 163 { 164 write_seqcount_end(&qdisc->running); 165 if (qdisc->flags & TCQ_F_NOLOCK) 166 spin_unlock(&qdisc->seqlock); 167 } 168 169 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) 170 { 171 return qdisc->flags & TCQ_F_ONETXQUEUE; 172 } 173 174 static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) 175 { 176 #ifdef CONFIG_BQL 177 /* Non-BQL migrated drivers will return 0, too. */ 178 return dql_avail(&txq->dql); 179 #else 180 return 0; 181 #endif 182 } 183 184 struct Qdisc_class_ops { 185 unsigned int flags; 186 /* Child qdisc manipulation */ 187 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 188 int (*graft)(struct Qdisc *, unsigned long cl, 189 struct Qdisc *, struct Qdisc **, 190 struct netlink_ext_ack *extack); 191 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 192 void (*qlen_notify)(struct Qdisc *, unsigned long); 193 194 /* Class manipulation routines */ 195 unsigned long (*find)(struct Qdisc *, u32 classid); 196 int (*change)(struct Qdisc *, u32, u32, 197 struct nlattr **, unsigned long *, 198 struct netlink_ext_ack *); 199 int (*delete)(struct Qdisc *, unsigned long); 200 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 201 202 /* Filter manipulation */ 203 struct tcf_block * (*tcf_block)(struct Qdisc *sch, 204 unsigned long arg, 205 struct netlink_ext_ack *extack); 206 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 207 u32 classid); 208 void (*unbind_tcf)(struct Qdisc *, unsigned long); 209 210 /* rtnetlink specific */ 211 int (*dump)(struct Qdisc *, unsigned long, 212 struct sk_buff *skb, struct tcmsg*); 213 int (*dump_stats)(struct Qdisc *, unsigned long, 214 struct gnet_dump *); 215 }; 216 217 /* Qdisc_class_ops flag values */ 218 219 /* Implements API that doesn't require rtnl lock */ 220 enum qdisc_class_ops_flags { 221 QDISC_CLASS_OPS_DOIT_UNLOCKED = 1, 222 }; 223 224 struct Qdisc_ops { 225 struct Qdisc_ops *next; 226 const struct Qdisc_class_ops *cl_ops; 227 char id[IFNAMSIZ]; 228 int priv_size; 229 unsigned int static_flags; 230 231 int (*enqueue)(struct sk_buff *skb, 232 struct Qdisc *sch, 233 struct sk_buff **to_free); 234 struct sk_buff * (*dequeue)(struct Qdisc *); 235 struct sk_buff * (*peek)(struct Qdisc *); 236 237 int (*init)(struct Qdisc *sch, struct nlattr *arg, 238 struct netlink_ext_ack *extack); 239 void (*reset)(struct Qdisc *); 240 void (*destroy)(struct Qdisc *); 241 int (*change)(struct Qdisc *sch, 242 struct nlattr *arg, 243 struct netlink_ext_ack *extack); 244 void (*attach)(struct Qdisc *sch); 245 int (*change_tx_queue_len)(struct Qdisc *, unsigned int); 246 247 int (*dump)(struct Qdisc *, struct sk_buff *); 248 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 249 250 void (*ingress_block_set)(struct Qdisc *sch, 251 u32 block_index); 252 void (*egress_block_set)(struct Qdisc *sch, 253 u32 block_index); 254 u32 (*ingress_block_get)(struct Qdisc *sch); 255 u32 (*egress_block_get)(struct Qdisc *sch); 256 257 struct module *owner; 258 }; 259 260 261 struct tcf_result { 262 union { 263 struct { 264 unsigned long class; 265 u32 classid; 266 }; 267 const struct tcf_proto *goto_tp; 268 269 /* used by the TC_ACT_REINSERT action */ 270 struct { 271 bool ingress; 272 struct gnet_stats_queue *qstats; 273 }; 274 }; 275 }; 276 277 struct tcf_chain; 278 279 struct tcf_proto_ops { 280 struct list_head head; 281 char kind[IFNAMSIZ]; 282 283 int (*classify)(struct sk_buff *, 284 const struct tcf_proto *, 285 struct tcf_result *); 286 int (*init)(struct tcf_proto*); 287 void (*destroy)(struct tcf_proto *tp, bool rtnl_held, 288 struct netlink_ext_ack *extack); 289 290 void* (*get)(struct tcf_proto*, u32 handle); 291 void (*put)(struct tcf_proto *tp, void *f); 292 int (*change)(struct net *net, struct sk_buff *, 293 struct tcf_proto*, unsigned long, 294 u32 handle, struct nlattr **, 295 void **, bool, bool, 296 struct netlink_ext_ack *); 297 int (*delete)(struct tcf_proto *tp, void *arg, 298 bool *last, bool rtnl_held, 299 struct netlink_ext_ack *); 300 void (*walk)(struct tcf_proto *tp, 301 struct tcf_walker *arg, bool rtnl_held); 302 int (*reoffload)(struct tcf_proto *tp, bool add, 303 tc_setup_cb_t *cb, void *cb_priv, 304 struct netlink_ext_ack *extack); 305 void (*bind_class)(void *, u32, unsigned long); 306 void * (*tmplt_create)(struct net *net, 307 struct tcf_chain *chain, 308 struct nlattr **tca, 309 struct netlink_ext_ack *extack); 310 void (*tmplt_destroy)(void *tmplt_priv); 311 312 /* rtnetlink specific */ 313 int (*dump)(struct net*, struct tcf_proto*, void *, 314 struct sk_buff *skb, struct tcmsg*, 315 bool); 316 int (*tmplt_dump)(struct sk_buff *skb, 317 struct net *net, 318 void *tmplt_priv); 319 320 struct module *owner; 321 int flags; 322 }; 323 324 enum tcf_proto_ops_flags { 325 TCF_PROTO_OPS_DOIT_UNLOCKED = 1, 326 }; 327 328 struct tcf_proto { 329 /* Fast access part */ 330 struct tcf_proto __rcu *next; 331 void __rcu *root; 332 333 /* called under RCU BH lock*/ 334 int (*classify)(struct sk_buff *, 335 const struct tcf_proto *, 336 struct tcf_result *); 337 __be16 protocol; 338 339 /* All the rest */ 340 u32 prio; 341 void *data; 342 const struct tcf_proto_ops *ops; 343 struct tcf_chain *chain; 344 /* Lock protects tcf_proto shared state and can be used by unlocked 345 * classifiers to protect their private data. 346 */ 347 spinlock_t lock; 348 bool deleting; 349 refcount_t refcnt; 350 struct rcu_head rcu; 351 }; 352 353 struct qdisc_skb_cb { 354 union { 355 struct { 356 unsigned int pkt_len; 357 u16 slave_dev_queue_mapping; 358 u16 tc_classid; 359 }; 360 struct bpf_flow_keys *flow_keys; 361 }; 362 #define QDISC_CB_PRIV_LEN 20 363 unsigned char data[QDISC_CB_PRIV_LEN]; 364 }; 365 366 typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); 367 368 struct tcf_chain { 369 /* Protects filter_chain. */ 370 struct mutex filter_chain_lock; 371 struct tcf_proto __rcu *filter_chain; 372 struct list_head list; 373 struct tcf_block *block; 374 u32 index; /* chain index */ 375 unsigned int refcnt; 376 unsigned int action_refcnt; 377 bool explicitly_created; 378 bool flushing; 379 const struct tcf_proto_ops *tmplt_ops; 380 void *tmplt_priv; 381 struct rcu_head rcu; 382 }; 383 384 struct tcf_block { 385 /* Lock protects tcf_block and lifetime-management data of chains 386 * attached to the block (refcnt, action_refcnt, explicitly_created). 387 */ 388 struct mutex lock; 389 struct list_head chain_list; 390 u32 index; /* block index for shared blocks */ 391 refcount_t refcnt; 392 struct net *net; 393 struct Qdisc *q; 394 struct list_head cb_list; 395 struct list_head owner_list; 396 bool keep_dst; 397 unsigned int offloadcnt; /* Number of oddloaded filters */ 398 unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ 399 struct { 400 struct tcf_chain *chain; 401 struct list_head filter_chain_list; 402 } chain0; 403 struct rcu_head rcu; 404 }; 405 406 #ifdef CONFIG_PROVE_LOCKING 407 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain) 408 { 409 return lockdep_is_held(&chain->filter_chain_lock); 410 } 411 412 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 413 { 414 return lockdep_is_held(&tp->lock); 415 } 416 #else 417 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block *chain) 418 { 419 return true; 420 } 421 422 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp) 423 { 424 return true; 425 } 426 #endif /* #ifdef CONFIG_PROVE_LOCKING */ 427 428 #define tcf_chain_dereference(p, chain) \ 429 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain)) 430 431 #define tcf_proto_dereference(p, tp) \ 432 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp)) 433 434 static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags) 435 { 436 if (*flags & TCA_CLS_FLAGS_IN_HW) 437 return; 438 *flags |= TCA_CLS_FLAGS_IN_HW; 439 block->offloadcnt++; 440 } 441 442 static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags) 443 { 444 if (!(*flags & TCA_CLS_FLAGS_IN_HW)) 445 return; 446 *flags &= ~TCA_CLS_FLAGS_IN_HW; 447 block->offloadcnt--; 448 } 449 450 static inline void 451 tc_cls_offload_cnt_update(struct tcf_block *block, u32 *cnt, 452 u32 *flags, bool add) 453 { 454 if (add) { 455 if (!*cnt) 456 tcf_block_offload_inc(block, flags); 457 (*cnt)++; 458 } else { 459 (*cnt)--; 460 if (!*cnt) 461 tcf_block_offload_dec(block, flags); 462 } 463 } 464 465 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 466 { 467 struct qdisc_skb_cb *qcb; 468 469 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz); 470 BUILD_BUG_ON(sizeof(qcb->data) < sz); 471 } 472 473 static inline int qdisc_qlen(const struct Qdisc *q) 474 { 475 return q->q.qlen; 476 } 477 478 static inline u32 qdisc_qlen_sum(const struct Qdisc *q) 479 { 480 u32 qlen = q->qstats.qlen; 481 482 if (q->flags & TCQ_F_NOLOCK) 483 qlen += atomic_read(&q->q.atomic_qlen); 484 else 485 qlen += q->q.qlen; 486 487 return qlen; 488 } 489 490 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 491 { 492 return (struct qdisc_skb_cb *)skb->cb; 493 } 494 495 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 496 { 497 return &qdisc->q.lock; 498 } 499 500 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 501 { 502 struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); 503 504 return q; 505 } 506 507 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 508 { 509 return qdisc->dev_queue->qdisc_sleeping; 510 } 511 512 /* The qdisc root lock is a mechanism by which to top level 513 * of a qdisc tree can be locked from any qdisc node in the 514 * forest. This allows changing the configuration of some 515 * aspect of the qdisc tree while blocking out asynchronous 516 * qdisc access in the packet processing paths. 517 * 518 * It is only legal to do this when the root will not change 519 * on us. Otherwise we'll potentially lock the wrong qdisc 520 * root. This is enforced by holding the RTNL semaphore, which 521 * all users of this lock accessor must do. 522 */ 523 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 524 { 525 struct Qdisc *root = qdisc_root(qdisc); 526 527 ASSERT_RTNL(); 528 return qdisc_lock(root); 529 } 530 531 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 532 { 533 struct Qdisc *root = qdisc_root_sleeping(qdisc); 534 535 ASSERT_RTNL(); 536 return qdisc_lock(root); 537 } 538 539 static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc) 540 { 541 struct Qdisc *root = qdisc_root_sleeping(qdisc); 542 543 ASSERT_RTNL(); 544 return &root->running; 545 } 546 547 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 548 { 549 return qdisc->dev_queue->dev; 550 } 551 552 static inline void sch_tree_lock(const struct Qdisc *q) 553 { 554 spin_lock_bh(qdisc_root_sleeping_lock(q)); 555 } 556 557 static inline void sch_tree_unlock(const struct Qdisc *q) 558 { 559 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 560 } 561 562 extern struct Qdisc noop_qdisc; 563 extern struct Qdisc_ops noop_qdisc_ops; 564 extern struct Qdisc_ops pfifo_fast_ops; 565 extern struct Qdisc_ops mq_qdisc_ops; 566 extern struct Qdisc_ops noqueue_qdisc_ops; 567 extern const struct Qdisc_ops *default_qdisc_ops; 568 static inline const struct Qdisc_ops * 569 get_default_qdisc_ops(const struct net_device *dev, int ntx) 570 { 571 return ntx < dev->real_num_tx_queues ? 572 default_qdisc_ops : &pfifo_fast_ops; 573 } 574 575 struct Qdisc_class_common { 576 u32 classid; 577 struct hlist_node hnode; 578 }; 579 580 struct Qdisc_class_hash { 581 struct hlist_head *hash; 582 unsigned int hashsize; 583 unsigned int hashmask; 584 unsigned int hashelems; 585 }; 586 587 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 588 { 589 id ^= id >> 8; 590 id ^= id >> 4; 591 return id & mask; 592 } 593 594 static inline struct Qdisc_class_common * 595 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 596 { 597 struct Qdisc_class_common *cl; 598 unsigned int h; 599 600 if (!id) 601 return NULL; 602 603 h = qdisc_class_hash(id, hash->hashmask); 604 hlist_for_each_entry(cl, &hash->hash[h], hnode) { 605 if (cl->classid == id) 606 return cl; 607 } 608 return NULL; 609 } 610 611 static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) 612 { 613 u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; 614 615 return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; 616 } 617 618 int qdisc_class_hash_init(struct Qdisc_class_hash *); 619 void qdisc_class_hash_insert(struct Qdisc_class_hash *, 620 struct Qdisc_class_common *); 621 void qdisc_class_hash_remove(struct Qdisc_class_hash *, 622 struct Qdisc_class_common *); 623 void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 624 void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 625 626 int dev_qdisc_change_tx_queue_len(struct net_device *dev); 627 void dev_init_scheduler(struct net_device *dev); 628 void dev_shutdown(struct net_device *dev); 629 void dev_activate(struct net_device *dev); 630 void dev_deactivate(struct net_device *dev); 631 void dev_deactivate_many(struct list_head *head); 632 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 633 struct Qdisc *qdisc); 634 void qdisc_reset(struct Qdisc *qdisc); 635 void qdisc_put(struct Qdisc *qdisc); 636 void qdisc_put_unlocked(struct Qdisc *qdisc); 637 void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len); 638 #ifdef CONFIG_NET_SCHED 639 int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 640 void *type_data); 641 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 642 struct Qdisc *new, struct Qdisc *old, 643 enum tc_setup_type type, void *type_data, 644 struct netlink_ext_ack *extack); 645 #else 646 static inline int 647 qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type, 648 void *type_data) 649 { 650 q->flags &= ~TCQ_F_OFFLOADED; 651 return 0; 652 } 653 654 static inline void 655 qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 656 struct Qdisc *new, struct Qdisc *old, 657 enum tc_setup_type type, void *type_data, 658 struct netlink_ext_ack *extack) 659 { 660 } 661 #endif 662 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 663 const struct Qdisc_ops *ops, 664 struct netlink_ext_ack *extack); 665 void qdisc_free(struct Qdisc *qdisc); 666 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 667 const struct Qdisc_ops *ops, u32 parentid, 668 struct netlink_ext_ack *extack); 669 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 670 const struct qdisc_size_table *stab); 671 int skb_do_redirect(struct sk_buff *); 672 673 static inline void skb_reset_tc(struct sk_buff *skb) 674 { 675 #ifdef CONFIG_NET_CLS_ACT 676 skb->tc_redirected = 0; 677 #endif 678 } 679 680 static inline bool skb_is_tc_redirected(const struct sk_buff *skb) 681 { 682 #ifdef CONFIG_NET_CLS_ACT 683 return skb->tc_redirected; 684 #else 685 return false; 686 #endif 687 } 688 689 static inline bool skb_at_tc_ingress(const struct sk_buff *skb) 690 { 691 #ifdef CONFIG_NET_CLS_ACT 692 return skb->tc_at_ingress; 693 #else 694 return false; 695 #endif 696 } 697 698 static inline bool skb_skip_tc_classify(struct sk_buff *skb) 699 { 700 #ifdef CONFIG_NET_CLS_ACT 701 if (skb->tc_skip_classify) { 702 skb->tc_skip_classify = 0; 703 return true; 704 } 705 #endif 706 return false; 707 } 708 709 /* Reset all TX qdiscs greater than index of a device. */ 710 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 711 { 712 struct Qdisc *qdisc; 713 714 for (; i < dev->num_tx_queues; i++) { 715 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); 716 if (qdisc) { 717 spin_lock_bh(qdisc_lock(qdisc)); 718 qdisc_reset(qdisc); 719 spin_unlock_bh(qdisc_lock(qdisc)); 720 } 721 } 722 } 723 724 static inline void qdisc_reset_all_tx(struct net_device *dev) 725 { 726 qdisc_reset_all_tx_gt(dev, 0); 727 } 728 729 /* Are all TX queues of the device empty? */ 730 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 731 { 732 unsigned int i; 733 734 rcu_read_lock(); 735 for (i = 0; i < dev->num_tx_queues; i++) { 736 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 737 const struct Qdisc *q = rcu_dereference(txq->qdisc); 738 739 if (q->q.qlen) { 740 rcu_read_unlock(); 741 return false; 742 } 743 } 744 rcu_read_unlock(); 745 return true; 746 } 747 748 /* Are any of the TX qdiscs changing? */ 749 static inline bool qdisc_tx_changing(const struct net_device *dev) 750 { 751 unsigned int i; 752 753 for (i = 0; i < dev->num_tx_queues; i++) { 754 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 755 if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) 756 return true; 757 } 758 return false; 759 } 760 761 /* Is the device using the noop qdisc on all queues? */ 762 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 763 { 764 unsigned int i; 765 766 for (i = 0; i < dev->num_tx_queues; i++) { 767 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 768 if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) 769 return false; 770 } 771 return true; 772 } 773 774 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 775 { 776 return qdisc_skb_cb(skb)->pkt_len; 777 } 778 779 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 780 enum net_xmit_qdisc_t { 781 __NET_XMIT_STOLEN = 0x00010000, 782 __NET_XMIT_BYPASS = 0x00020000, 783 }; 784 785 #ifdef CONFIG_NET_CLS_ACT 786 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 787 #else 788 #define net_xmit_drop_count(e) (1) 789 #endif 790 791 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 792 const struct Qdisc *sch) 793 { 794 #ifdef CONFIG_NET_SCHED 795 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 796 797 if (stab) 798 __qdisc_calculate_pkt_len(skb, stab); 799 #endif 800 } 801 802 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, 803 struct sk_buff **to_free) 804 { 805 qdisc_calculate_pkt_len(skb, sch); 806 return sch->enqueue(skb, sch, to_free); 807 } 808 809 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) 810 { 811 return q->flags & TCQ_F_CPUSTATS; 812 } 813 814 static inline void _bstats_update(struct gnet_stats_basic_packed *bstats, 815 __u64 bytes, __u32 packets) 816 { 817 bstats->bytes += bytes; 818 bstats->packets += packets; 819 } 820 821 static inline void bstats_update(struct gnet_stats_basic_packed *bstats, 822 const struct sk_buff *skb) 823 { 824 _bstats_update(bstats, 825 qdisc_pkt_len(skb), 826 skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); 827 } 828 829 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 830 __u64 bytes, __u32 packets) 831 { 832 u64_stats_update_begin(&bstats->syncp); 833 _bstats_update(&bstats->bstats, bytes, packets); 834 u64_stats_update_end(&bstats->syncp); 835 } 836 837 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, 838 const struct sk_buff *skb) 839 { 840 u64_stats_update_begin(&bstats->syncp); 841 bstats_update(&bstats->bstats, skb); 842 u64_stats_update_end(&bstats->syncp); 843 } 844 845 static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, 846 const struct sk_buff *skb) 847 { 848 bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb); 849 } 850 851 static inline void qdisc_bstats_update(struct Qdisc *sch, 852 const struct sk_buff *skb) 853 { 854 bstats_update(&sch->bstats, skb); 855 } 856 857 static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, 858 const struct sk_buff *skb) 859 { 860 sch->qstats.backlog -= qdisc_pkt_len(skb); 861 } 862 863 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, 864 const struct sk_buff *skb) 865 { 866 this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 867 } 868 869 static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, 870 const struct sk_buff *skb) 871 { 872 sch->qstats.backlog += qdisc_pkt_len(skb); 873 } 874 875 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, 876 const struct sk_buff *skb) 877 { 878 this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); 879 } 880 881 static inline void qdisc_qstats_atomic_qlen_inc(struct Qdisc *sch) 882 { 883 atomic_inc(&sch->q.atomic_qlen); 884 } 885 886 static inline void qdisc_qstats_atomic_qlen_dec(struct Qdisc *sch) 887 { 888 atomic_dec(&sch->q.atomic_qlen); 889 } 890 891 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) 892 { 893 this_cpu_inc(sch->cpu_qstats->requeues); 894 } 895 896 static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) 897 { 898 sch->qstats.drops += count; 899 } 900 901 static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) 902 { 903 qstats->drops++; 904 } 905 906 static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) 907 { 908 qstats->overlimits++; 909 } 910 911 static inline void qdisc_qstats_drop(struct Qdisc *sch) 912 { 913 qstats_drop_inc(&sch->qstats); 914 } 915 916 static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) 917 { 918 this_cpu_inc(sch->cpu_qstats->drops); 919 } 920 921 static inline void qdisc_qstats_overlimit(struct Qdisc *sch) 922 { 923 sch->qstats.overlimits++; 924 } 925 926 static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh) 927 { 928 qh->head = NULL; 929 qh->tail = NULL; 930 qh->qlen = 0; 931 } 932 933 static inline void __qdisc_enqueue_tail(struct sk_buff *skb, 934 struct qdisc_skb_head *qh) 935 { 936 struct sk_buff *last = qh->tail; 937 938 if (last) { 939 skb->next = NULL; 940 last->next = skb; 941 qh->tail = skb; 942 } else { 943 qh->tail = skb; 944 qh->head = skb; 945 } 946 qh->qlen++; 947 } 948 949 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 950 { 951 __qdisc_enqueue_tail(skb, &sch->q); 952 qdisc_qstats_backlog_inc(sch, skb); 953 return NET_XMIT_SUCCESS; 954 } 955 956 static inline void __qdisc_enqueue_head(struct sk_buff *skb, 957 struct qdisc_skb_head *qh) 958 { 959 skb->next = qh->head; 960 961 if (!qh->head) 962 qh->tail = skb; 963 qh->head = skb; 964 qh->qlen++; 965 } 966 967 static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) 968 { 969 struct sk_buff *skb = qh->head; 970 971 if (likely(skb != NULL)) { 972 qh->head = skb->next; 973 qh->qlen--; 974 if (qh->head == NULL) 975 qh->tail = NULL; 976 skb->next = NULL; 977 } 978 979 return skb; 980 } 981 982 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 983 { 984 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); 985 986 if (likely(skb != NULL)) { 987 qdisc_qstats_backlog_dec(sch, skb); 988 qdisc_bstats_update(sch, skb); 989 } 990 991 return skb; 992 } 993 994 /* Instead of calling kfree_skb() while root qdisc lock is held, 995 * queue the skb for future freeing at end of __dev_xmit_skb() 996 */ 997 static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) 998 { 999 skb->next = *to_free; 1000 *to_free = skb; 1001 } 1002 1003 static inline void __qdisc_drop_all(struct sk_buff *skb, 1004 struct sk_buff **to_free) 1005 { 1006 if (skb->prev) 1007 skb->prev->next = *to_free; 1008 else 1009 skb->next = *to_free; 1010 *to_free = skb; 1011 } 1012 1013 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 1014 struct qdisc_skb_head *qh, 1015 struct sk_buff **to_free) 1016 { 1017 struct sk_buff *skb = __qdisc_dequeue_head(qh); 1018 1019 if (likely(skb != NULL)) { 1020 unsigned int len = qdisc_pkt_len(skb); 1021 1022 qdisc_qstats_backlog_dec(sch, skb); 1023 __qdisc_drop(skb, to_free); 1024 return len; 1025 } 1026 1027 return 0; 1028 } 1029 1030 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch, 1031 struct sk_buff **to_free) 1032 { 1033 return __qdisc_queue_drop_head(sch, &sch->q, to_free); 1034 } 1035 1036 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 1037 { 1038 const struct qdisc_skb_head *qh = &sch->q; 1039 1040 return qh->head; 1041 } 1042 1043 /* generic pseudo peek method for non-work-conserving qdisc */ 1044 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 1045 { 1046 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1047 1048 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 1049 if (!skb) { 1050 skb = sch->dequeue(sch); 1051 1052 if (skb) { 1053 __skb_queue_head(&sch->gso_skb, skb); 1054 /* it's still part of the queue */ 1055 qdisc_qstats_backlog_inc(sch, skb); 1056 sch->q.qlen++; 1057 } 1058 } 1059 1060 return skb; 1061 } 1062 1063 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 1064 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 1065 { 1066 struct sk_buff *skb = skb_peek(&sch->gso_skb); 1067 1068 if (skb) { 1069 skb = __skb_dequeue(&sch->gso_skb); 1070 qdisc_qstats_backlog_dec(sch, skb); 1071 sch->q.qlen--; 1072 } else { 1073 skb = sch->dequeue(sch); 1074 } 1075 1076 return skb; 1077 } 1078 1079 static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) 1080 { 1081 /* 1082 * We do not know the backlog in bytes of this list, it 1083 * is up to the caller to correct it 1084 */ 1085 ASSERT_RTNL(); 1086 if (qh->qlen) { 1087 rtnl_kfree_skbs(qh->head, qh->tail); 1088 1089 qh->head = NULL; 1090 qh->tail = NULL; 1091 qh->qlen = 0; 1092 } 1093 } 1094 1095 static inline void qdisc_reset_queue(struct Qdisc *sch) 1096 { 1097 __qdisc_reset_queue(&sch->q); 1098 sch->qstats.backlog = 0; 1099 } 1100 1101 static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, 1102 struct Qdisc **pold) 1103 { 1104 struct Qdisc *old; 1105 1106 sch_tree_lock(sch); 1107 old = *pold; 1108 *pold = new; 1109 if (old != NULL) { 1110 unsigned int qlen = old->q.qlen; 1111 unsigned int backlog = old->qstats.backlog; 1112 1113 qdisc_reset(old); 1114 qdisc_tree_reduce_backlog(old, qlen, backlog); 1115 } 1116 sch_tree_unlock(sch); 1117 1118 return old; 1119 } 1120 1121 static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 1122 { 1123 rtnl_kfree_skbs(skb, skb); 1124 qdisc_qstats_drop(sch); 1125 } 1126 1127 static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, 1128 struct sk_buff **to_free) 1129 { 1130 __qdisc_drop(skb, to_free); 1131 qdisc_qstats_cpu_drop(sch); 1132 1133 return NET_XMIT_DROP; 1134 } 1135 1136 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, 1137 struct sk_buff **to_free) 1138 { 1139 __qdisc_drop(skb, to_free); 1140 qdisc_qstats_drop(sch); 1141 1142 return NET_XMIT_DROP; 1143 } 1144 1145 static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch, 1146 struct sk_buff **to_free) 1147 { 1148 __qdisc_drop_all(skb, to_free); 1149 qdisc_qstats_drop(sch); 1150 1151 return NET_XMIT_DROP; 1152 } 1153 1154 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 1155 long it will take to send a packet given its size. 1156 */ 1157 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 1158 { 1159 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 1160 if (slot < 0) 1161 slot = 0; 1162 slot >>= rtab->rate.cell_log; 1163 if (slot > 255) 1164 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 1165 return rtab->data[slot]; 1166 } 1167 1168 struct psched_ratecfg { 1169 u64 rate_bytes_ps; /* bytes per second */ 1170 u32 mult; 1171 u16 overhead; 1172 u8 linklayer; 1173 u8 shift; 1174 }; 1175 1176 static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, 1177 unsigned int len) 1178 { 1179 len += r->overhead; 1180 1181 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) 1182 return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; 1183 1184 return ((u64)len * r->mult) >> r->shift; 1185 } 1186 1187 void psched_ratecfg_precompute(struct psched_ratecfg *r, 1188 const struct tc_ratespec *conf, 1189 u64 rate64); 1190 1191 static inline void psched_ratecfg_getrate(struct tc_ratespec *res, 1192 const struct psched_ratecfg *r) 1193 { 1194 memset(res, 0, sizeof(*res)); 1195 1196 /* legacy struct tc_ratespec has a 32bit @rate field 1197 * Qdisc using 64bit rate should add new attributes 1198 * in order to maintain compatibility. 1199 */ 1200 res->rate = min_t(u64, r->rate_bytes_ps, ~0U); 1201 1202 res->overhead = r->overhead; 1203 res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); 1204 } 1205 1206 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. 1207 * The fast path only needs to access filter list and to update stats 1208 */ 1209 struct mini_Qdisc { 1210 struct tcf_proto *filter_list; 1211 struct gnet_stats_basic_cpu __percpu *cpu_bstats; 1212 struct gnet_stats_queue __percpu *cpu_qstats; 1213 struct rcu_head rcu; 1214 }; 1215 1216 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, 1217 const struct sk_buff *skb) 1218 { 1219 bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb); 1220 } 1221 1222 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) 1223 { 1224 this_cpu_inc(miniq->cpu_qstats->drops); 1225 } 1226 1227 struct mini_Qdisc_pair { 1228 struct mini_Qdisc miniq1; 1229 struct mini_Qdisc miniq2; 1230 struct mini_Qdisc __rcu **p_miniq; 1231 }; 1232 1233 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, 1234 struct tcf_proto *tp_head); 1235 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, 1236 struct mini_Qdisc __rcu **p_miniq); 1237 1238 static inline void skb_tc_reinsert(struct sk_buff *skb, struct tcf_result *res) 1239 { 1240 struct gnet_stats_queue *stats = res->qstats; 1241 int ret; 1242 1243 if (res->ingress) 1244 ret = netif_receive_skb(skb); 1245 else 1246 ret = dev_queue_xmit(skb); 1247 if (ret && stats) 1248 qstats_overlimit_inc(res->qstats); 1249 } 1250 1251 #endif 1252