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