1 #ifndef __NET_SCHED_GENERIC_H 2 #define __NET_SCHED_GENERIC_H 3 4 #include <linux/netdevice.h> 5 #include <linux/types.h> 6 #include <linux/rcupdate.h> 7 #include <linux/pkt_sched.h> 8 #include <linux/pkt_cls.h> 9 #include <net/gen_stats.h> 10 #include <net/rtnetlink.h> 11 12 struct Qdisc_ops; 13 struct qdisc_walker; 14 struct tcf_walker; 15 struct module; 16 17 struct qdisc_rate_table { 18 struct tc_ratespec rate; 19 u32 data[256]; 20 struct qdisc_rate_table *next; 21 int refcnt; 22 }; 23 24 enum qdisc_state_t { 25 __QDISC_STATE_SCHED, 26 __QDISC_STATE_DEACTIVATED, 27 __QDISC_STATE_THROTTLED, 28 }; 29 30 /* 31 * following bits are only changed while qdisc lock is held 32 */ 33 enum qdisc___state_t { 34 __QDISC___STATE_RUNNING = 1, 35 }; 36 37 struct qdisc_size_table { 38 struct rcu_head rcu; 39 struct list_head list; 40 struct tc_sizespec szopts; 41 int refcnt; 42 u16 data[]; 43 }; 44 45 struct Qdisc { 46 int (*enqueue)(struct sk_buff *skb, struct Qdisc *dev); 47 struct sk_buff * (*dequeue)(struct Qdisc *dev); 48 unsigned int flags; 49 #define TCQ_F_BUILTIN 1 50 #define TCQ_F_INGRESS 2 51 #define TCQ_F_CAN_BYPASS 4 52 #define TCQ_F_MQROOT 8 53 #define TCQ_F_WARN_NONWC (1 << 16) 54 int padded; 55 const struct Qdisc_ops *ops; 56 struct qdisc_size_table __rcu *stab; 57 struct list_head list; 58 u32 handle; 59 u32 parent; 60 atomic_t refcnt; 61 struct gnet_stats_rate_est rate_est; 62 int (*reshape_fail)(struct sk_buff *skb, 63 struct Qdisc *q); 64 65 void *u32_node; 66 67 /* This field is deprecated, but it is still used by CBQ 68 * and it will live until better solution will be invented. 69 */ 70 struct Qdisc *__parent; 71 struct netdev_queue *dev_queue; 72 struct Qdisc *next_sched; 73 74 struct sk_buff *gso_skb; 75 /* 76 * For performance sake on SMP, we put highly modified fields at the end 77 */ 78 unsigned long state; 79 struct sk_buff_head q; 80 struct gnet_stats_basic_packed bstats; 81 unsigned int __state; 82 struct gnet_stats_queue qstats; 83 struct rcu_head rcu_head; 84 spinlock_t busylock; 85 u32 limit; 86 }; 87 88 static inline bool qdisc_is_running(const struct Qdisc *qdisc) 89 { 90 return (qdisc->__state & __QDISC___STATE_RUNNING) ? true : false; 91 } 92 93 static inline bool qdisc_run_begin(struct Qdisc *qdisc) 94 { 95 if (qdisc_is_running(qdisc)) 96 return false; 97 qdisc->__state |= __QDISC___STATE_RUNNING; 98 return true; 99 } 100 101 static inline void qdisc_run_end(struct Qdisc *qdisc) 102 { 103 qdisc->__state &= ~__QDISC___STATE_RUNNING; 104 } 105 106 static inline bool qdisc_is_throttled(const struct Qdisc *qdisc) 107 { 108 return test_bit(__QDISC_STATE_THROTTLED, &qdisc->state) ? true : false; 109 } 110 111 static inline void qdisc_throttled(struct Qdisc *qdisc) 112 { 113 set_bit(__QDISC_STATE_THROTTLED, &qdisc->state); 114 } 115 116 static inline void qdisc_unthrottled(struct Qdisc *qdisc) 117 { 118 clear_bit(__QDISC_STATE_THROTTLED, &qdisc->state); 119 } 120 121 struct Qdisc_class_ops { 122 /* Child qdisc manipulation */ 123 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 124 int (*graft)(struct Qdisc *, unsigned long cl, 125 struct Qdisc *, struct Qdisc **); 126 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 127 void (*qlen_notify)(struct Qdisc *, unsigned long); 128 129 /* Class manipulation routines */ 130 unsigned long (*get)(struct Qdisc *, u32 classid); 131 void (*put)(struct Qdisc *, unsigned long); 132 int (*change)(struct Qdisc *, u32, u32, 133 struct nlattr **, unsigned long *); 134 int (*delete)(struct Qdisc *, unsigned long); 135 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 136 137 /* Filter manipulation */ 138 struct tcf_proto ** (*tcf_chain)(struct Qdisc *, unsigned long); 139 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 140 u32 classid); 141 void (*unbind_tcf)(struct Qdisc *, unsigned long); 142 143 /* rtnetlink specific */ 144 int (*dump)(struct Qdisc *, unsigned long, 145 struct sk_buff *skb, struct tcmsg*); 146 int (*dump_stats)(struct Qdisc *, unsigned long, 147 struct gnet_dump *); 148 }; 149 150 struct Qdisc_ops { 151 struct Qdisc_ops *next; 152 const struct Qdisc_class_ops *cl_ops; 153 char id[IFNAMSIZ]; 154 int priv_size; 155 156 int (*enqueue)(struct sk_buff *, struct Qdisc *); 157 struct sk_buff * (*dequeue)(struct Qdisc *); 158 struct sk_buff * (*peek)(struct Qdisc *); 159 unsigned int (*drop)(struct Qdisc *); 160 161 int (*init)(struct Qdisc *, struct nlattr *arg); 162 void (*reset)(struct Qdisc *); 163 void (*destroy)(struct Qdisc *); 164 int (*change)(struct Qdisc *, struct nlattr *arg); 165 void (*attach)(struct Qdisc *); 166 167 int (*dump)(struct Qdisc *, struct sk_buff *); 168 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 169 170 struct module *owner; 171 }; 172 173 174 struct tcf_result { 175 unsigned long class; 176 u32 classid; 177 }; 178 179 struct tcf_proto_ops { 180 struct tcf_proto_ops *next; 181 char kind[IFNAMSIZ]; 182 183 int (*classify)(struct sk_buff *, 184 const struct tcf_proto *, 185 struct tcf_result *); 186 int (*init)(struct tcf_proto*); 187 void (*destroy)(struct tcf_proto*); 188 189 unsigned long (*get)(struct tcf_proto*, u32 handle); 190 void (*put)(struct tcf_proto*, unsigned long); 191 int (*change)(struct tcf_proto*, unsigned long, 192 u32 handle, struct nlattr **, 193 unsigned long *); 194 int (*delete)(struct tcf_proto*, unsigned long); 195 void (*walk)(struct tcf_proto*, struct tcf_walker *arg); 196 197 /* rtnetlink specific */ 198 int (*dump)(struct tcf_proto*, unsigned long, 199 struct sk_buff *skb, struct tcmsg*); 200 201 struct module *owner; 202 }; 203 204 struct tcf_proto { 205 /* Fast access part */ 206 struct tcf_proto *next; 207 void *root; 208 int (*classify)(struct sk_buff *, 209 const struct tcf_proto *, 210 struct tcf_result *); 211 __be16 protocol; 212 213 /* All the rest */ 214 u32 prio; 215 u32 classid; 216 struct Qdisc *q; 217 void *data; 218 const struct tcf_proto_ops *ops; 219 }; 220 221 struct qdisc_skb_cb { 222 unsigned int pkt_len; 223 unsigned char data[24]; 224 }; 225 226 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 227 { 228 struct qdisc_skb_cb *qcb; 229 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(unsigned int) + sz); 230 BUILD_BUG_ON(sizeof(qcb->data) < sz); 231 } 232 233 static inline int qdisc_qlen(const struct Qdisc *q) 234 { 235 return q->q.qlen; 236 } 237 238 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 239 { 240 return (struct qdisc_skb_cb *)skb->cb; 241 } 242 243 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 244 { 245 return &qdisc->q.lock; 246 } 247 248 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 249 { 250 return qdisc->dev_queue->qdisc; 251 } 252 253 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 254 { 255 return qdisc->dev_queue->qdisc_sleeping; 256 } 257 258 /* The qdisc root lock is a mechanism by which to top level 259 * of a qdisc tree can be locked from any qdisc node in the 260 * forest. This allows changing the configuration of some 261 * aspect of the qdisc tree while blocking out asynchronous 262 * qdisc access in the packet processing paths. 263 * 264 * It is only legal to do this when the root will not change 265 * on us. Otherwise we'll potentially lock the wrong qdisc 266 * root. This is enforced by holding the RTNL semaphore, which 267 * all users of this lock accessor must do. 268 */ 269 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 270 { 271 struct Qdisc *root = qdisc_root(qdisc); 272 273 ASSERT_RTNL(); 274 return qdisc_lock(root); 275 } 276 277 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 278 { 279 struct Qdisc *root = qdisc_root_sleeping(qdisc); 280 281 ASSERT_RTNL(); 282 return qdisc_lock(root); 283 } 284 285 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 286 { 287 return qdisc->dev_queue->dev; 288 } 289 290 static inline void sch_tree_lock(const struct Qdisc *q) 291 { 292 spin_lock_bh(qdisc_root_sleeping_lock(q)); 293 } 294 295 static inline void sch_tree_unlock(const struct Qdisc *q) 296 { 297 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 298 } 299 300 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q) 301 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q) 302 303 extern struct Qdisc noop_qdisc; 304 extern struct Qdisc_ops noop_qdisc_ops; 305 extern struct Qdisc_ops pfifo_fast_ops; 306 extern struct Qdisc_ops mq_qdisc_ops; 307 308 struct Qdisc_class_common { 309 u32 classid; 310 struct hlist_node hnode; 311 }; 312 313 struct Qdisc_class_hash { 314 struct hlist_head *hash; 315 unsigned int hashsize; 316 unsigned int hashmask; 317 unsigned int hashelems; 318 }; 319 320 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 321 { 322 id ^= id >> 8; 323 id ^= id >> 4; 324 return id & mask; 325 } 326 327 static inline struct Qdisc_class_common * 328 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 329 { 330 struct Qdisc_class_common *cl; 331 struct hlist_node *n; 332 unsigned int h; 333 334 h = qdisc_class_hash(id, hash->hashmask); 335 hlist_for_each_entry(cl, n, &hash->hash[h], hnode) { 336 if (cl->classid == id) 337 return cl; 338 } 339 return NULL; 340 } 341 342 extern int qdisc_class_hash_init(struct Qdisc_class_hash *); 343 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *); 344 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *); 345 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 346 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 347 348 extern void dev_init_scheduler(struct net_device *dev); 349 extern void dev_shutdown(struct net_device *dev); 350 extern void dev_activate(struct net_device *dev); 351 extern void dev_deactivate(struct net_device *dev); 352 extern void dev_deactivate_many(struct list_head *head); 353 extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 354 struct Qdisc *qdisc); 355 extern void qdisc_reset(struct Qdisc *qdisc); 356 extern void qdisc_destroy(struct Qdisc *qdisc); 357 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n); 358 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 359 struct Qdisc_ops *ops); 360 extern struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 361 struct Qdisc_ops *ops, u32 parentid); 362 extern void __qdisc_calculate_pkt_len(struct sk_buff *skb, 363 const struct qdisc_size_table *stab); 364 extern void tcf_destroy(struct tcf_proto *tp); 365 extern void tcf_destroy_chain(struct tcf_proto **fl); 366 367 /* Reset all TX qdiscs greater then index of a device. */ 368 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 369 { 370 struct Qdisc *qdisc; 371 372 for (; i < dev->num_tx_queues; i++) { 373 qdisc = netdev_get_tx_queue(dev, i)->qdisc; 374 if (qdisc) { 375 spin_lock_bh(qdisc_lock(qdisc)); 376 qdisc_reset(qdisc); 377 spin_unlock_bh(qdisc_lock(qdisc)); 378 } 379 } 380 } 381 382 static inline void qdisc_reset_all_tx(struct net_device *dev) 383 { 384 qdisc_reset_all_tx_gt(dev, 0); 385 } 386 387 /* Are all TX queues of the device empty? */ 388 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 389 { 390 unsigned int i; 391 for (i = 0; i < dev->num_tx_queues; i++) { 392 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 393 const struct Qdisc *q = txq->qdisc; 394 395 if (q->q.qlen) 396 return false; 397 } 398 return true; 399 } 400 401 /* Are any of the TX qdiscs changing? */ 402 static inline bool qdisc_tx_changing(const struct net_device *dev) 403 { 404 unsigned int i; 405 for (i = 0; i < dev->num_tx_queues; i++) { 406 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 407 if (txq->qdisc != txq->qdisc_sleeping) 408 return true; 409 } 410 return false; 411 } 412 413 /* Is the device using the noop qdisc on all queues? */ 414 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 415 { 416 unsigned int i; 417 for (i = 0; i < dev->num_tx_queues; i++) { 418 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 419 if (txq->qdisc != &noop_qdisc) 420 return false; 421 } 422 return true; 423 } 424 425 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 426 { 427 return qdisc_skb_cb(skb)->pkt_len; 428 } 429 430 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 431 enum net_xmit_qdisc_t { 432 __NET_XMIT_STOLEN = 0x00010000, 433 __NET_XMIT_BYPASS = 0x00020000, 434 }; 435 436 #ifdef CONFIG_NET_CLS_ACT 437 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 438 #else 439 #define net_xmit_drop_count(e) (1) 440 #endif 441 442 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 443 const struct Qdisc *sch) 444 { 445 #ifdef CONFIG_NET_SCHED 446 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 447 448 if (stab) 449 __qdisc_calculate_pkt_len(skb, stab); 450 #endif 451 } 452 453 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch) 454 { 455 qdisc_calculate_pkt_len(skb, sch); 456 return sch->enqueue(skb, sch); 457 } 458 459 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch) 460 { 461 qdisc_skb_cb(skb)->pkt_len = skb->len; 462 return qdisc_enqueue(skb, sch) & NET_XMIT_MASK; 463 } 464 465 466 static inline void bstats_update(struct gnet_stats_basic_packed *bstats, 467 const struct sk_buff *skb) 468 { 469 bstats->bytes += qdisc_pkt_len(skb); 470 bstats->packets += skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1; 471 } 472 473 static inline void qdisc_bstats_update(struct Qdisc *sch, 474 const struct sk_buff *skb) 475 { 476 bstats_update(&sch->bstats, skb); 477 } 478 479 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch, 480 struct sk_buff_head *list) 481 { 482 __skb_queue_tail(list, skb); 483 sch->qstats.backlog += qdisc_pkt_len(skb); 484 485 return NET_XMIT_SUCCESS; 486 } 487 488 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 489 { 490 return __qdisc_enqueue_tail(skb, sch, &sch->q); 491 } 492 493 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch, 494 struct sk_buff_head *list) 495 { 496 struct sk_buff *skb = __skb_dequeue(list); 497 498 if (likely(skb != NULL)) { 499 sch->qstats.backlog -= qdisc_pkt_len(skb); 500 qdisc_bstats_update(sch, skb); 501 } 502 503 return skb; 504 } 505 506 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 507 { 508 return __qdisc_dequeue_head(sch, &sch->q); 509 } 510 511 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 512 struct sk_buff_head *list) 513 { 514 struct sk_buff *skb = __skb_dequeue(list); 515 516 if (likely(skb != NULL)) { 517 unsigned int len = qdisc_pkt_len(skb); 518 sch->qstats.backlog -= len; 519 kfree_skb(skb); 520 return len; 521 } 522 523 return 0; 524 } 525 526 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch) 527 { 528 return __qdisc_queue_drop_head(sch, &sch->q); 529 } 530 531 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch, 532 struct sk_buff_head *list) 533 { 534 struct sk_buff *skb = __skb_dequeue_tail(list); 535 536 if (likely(skb != NULL)) 537 sch->qstats.backlog -= qdisc_pkt_len(skb); 538 539 return skb; 540 } 541 542 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch) 543 { 544 return __qdisc_dequeue_tail(sch, &sch->q); 545 } 546 547 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 548 { 549 return skb_peek(&sch->q); 550 } 551 552 /* generic pseudo peek method for non-work-conserving qdisc */ 553 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 554 { 555 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 556 if (!sch->gso_skb) { 557 sch->gso_skb = sch->dequeue(sch); 558 if (sch->gso_skb) 559 /* it's still part of the queue */ 560 sch->q.qlen++; 561 } 562 563 return sch->gso_skb; 564 } 565 566 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 567 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 568 { 569 struct sk_buff *skb = sch->gso_skb; 570 571 if (skb) { 572 sch->gso_skb = NULL; 573 sch->q.qlen--; 574 } else { 575 skb = sch->dequeue(sch); 576 } 577 578 return skb; 579 } 580 581 static inline void __qdisc_reset_queue(struct Qdisc *sch, 582 struct sk_buff_head *list) 583 { 584 /* 585 * We do not know the backlog in bytes of this list, it 586 * is up to the caller to correct it 587 */ 588 __skb_queue_purge(list); 589 } 590 591 static inline void qdisc_reset_queue(struct Qdisc *sch) 592 { 593 __qdisc_reset_queue(sch, &sch->q); 594 sch->qstats.backlog = 0; 595 } 596 597 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch, 598 struct sk_buff_head *list) 599 { 600 struct sk_buff *skb = __qdisc_dequeue_tail(sch, list); 601 602 if (likely(skb != NULL)) { 603 unsigned int len = qdisc_pkt_len(skb); 604 kfree_skb(skb); 605 return len; 606 } 607 608 return 0; 609 } 610 611 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch) 612 { 613 return __qdisc_queue_drop(sch, &sch->q); 614 } 615 616 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 617 { 618 kfree_skb(skb); 619 sch->qstats.drops++; 620 621 return NET_XMIT_DROP; 622 } 623 624 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch) 625 { 626 sch->qstats.drops++; 627 628 #ifdef CONFIG_NET_CLS_ACT 629 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch)) 630 goto drop; 631 632 return NET_XMIT_SUCCESS; 633 634 drop: 635 #endif 636 kfree_skb(skb); 637 return NET_XMIT_DROP; 638 } 639 640 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 641 long it will take to send a packet given its size. 642 */ 643 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 644 { 645 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 646 if (slot < 0) 647 slot = 0; 648 slot >>= rtab->rate.cell_log; 649 if (slot > 255) 650 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 651 return rtab->data[slot]; 652 } 653 654 #ifdef CONFIG_NET_CLS_ACT 655 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask, 656 int action) 657 { 658 struct sk_buff *n; 659 660 n = skb_clone(skb, gfp_mask); 661 662 if (n) { 663 n->tc_verd = SET_TC_VERD(n->tc_verd, 0); 664 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd); 665 n->tc_verd = CLR_TC_MUNGED(n->tc_verd); 666 } 667 return n; 668 } 669 #endif 670 671 #endif 672