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 12struct Qdisc_ops; 13struct qdisc_walker; 14struct tcf_walker; 15struct module; 16 17struct qdisc_rate_table { 18 struct tc_ratespec rate; 19 u32 data[256]; 20 struct qdisc_rate_table *next; 21 int refcnt; 22}; 23 24enum 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 */ 33enum qdisc___state_t { 34 __QDISC___STATE_RUNNING = 1, 35}; 36 37struct 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 45struct 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 88static inline bool qdisc_is_running(const struct Qdisc *qdisc) 89{ 90 return (qdisc->__state & __QDISC___STATE_RUNNING) ? true : false; 91} 92 93static 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 101static inline void qdisc_run_end(struct Qdisc *qdisc) 102{ 103 qdisc->__state &= ~__QDISC___STATE_RUNNING; 104} 105 106static inline bool qdisc_is_throttled(const struct Qdisc *qdisc) 107{ 108 return test_bit(__QDISC_STATE_THROTTLED, &qdisc->state) ? true : false; 109} 110 111static inline void qdisc_throttled(struct Qdisc *qdisc) 112{ 113 set_bit(__QDISC_STATE_THROTTLED, &qdisc->state); 114} 115 116static inline void qdisc_unthrottled(struct Qdisc *qdisc) 117{ 118 clear_bit(__QDISC_STATE_THROTTLED, &qdisc->state); 119} 120 121struct 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 150struct 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 174struct tcf_result { 175 unsigned long class; 176 u32 classid; 177}; 178 179struct 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 204struct 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 221struct qdisc_skb_cb { 222 unsigned int pkt_len; 223 u16 bond_queue_mapping; 224 u16 _pad; 225 unsigned char data[20]; 226}; 227 228static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 229{ 230 struct qdisc_skb_cb *qcb; 231 232 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz); 233 BUILD_BUG_ON(sizeof(qcb->data) < sz); 234} 235 236static inline int qdisc_qlen(const struct Qdisc *q) 237{ 238 return q->q.qlen; 239} 240 241static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 242{ 243 return (struct qdisc_skb_cb *)skb->cb; 244} 245 246static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 247{ 248 return &qdisc->q.lock; 249} 250 251static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 252{ 253 return qdisc->dev_queue->qdisc; 254} 255 256static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 257{ 258 return qdisc->dev_queue->qdisc_sleeping; 259} 260 261/* The qdisc root lock is a mechanism by which to top level 262 * of a qdisc tree can be locked from any qdisc node in the 263 * forest. This allows changing the configuration of some 264 * aspect of the qdisc tree while blocking out asynchronous 265 * qdisc access in the packet processing paths. 266 * 267 * It is only legal to do this when the root will not change 268 * on us. Otherwise we'll potentially lock the wrong qdisc 269 * root. This is enforced by holding the RTNL semaphore, which 270 * all users of this lock accessor must do. 271 */ 272static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 273{ 274 struct Qdisc *root = qdisc_root(qdisc); 275 276 ASSERT_RTNL(); 277 return qdisc_lock(root); 278} 279 280static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 281{ 282 struct Qdisc *root = qdisc_root_sleeping(qdisc); 283 284 ASSERT_RTNL(); 285 return qdisc_lock(root); 286} 287 288static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 289{ 290 return qdisc->dev_queue->dev; 291} 292 293static inline void sch_tree_lock(const struct Qdisc *q) 294{ 295 spin_lock_bh(qdisc_root_sleeping_lock(q)); 296} 297 298static inline void sch_tree_unlock(const struct Qdisc *q) 299{ 300 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 301} 302 303#define tcf_tree_lock(tp) sch_tree_lock((tp)->q) 304#define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q) 305 306extern struct Qdisc noop_qdisc; 307extern struct Qdisc_ops noop_qdisc_ops; 308extern struct Qdisc_ops pfifo_fast_ops; 309extern struct Qdisc_ops mq_qdisc_ops; 310 311struct Qdisc_class_common { 312 u32 classid; 313 struct hlist_node hnode; 314}; 315 316struct Qdisc_class_hash { 317 struct hlist_head *hash; 318 unsigned int hashsize; 319 unsigned int hashmask; 320 unsigned int hashelems; 321}; 322 323static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 324{ 325 id ^= id >> 8; 326 id ^= id >> 4; 327 return id & mask; 328} 329 330static inline struct Qdisc_class_common * 331qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 332{ 333 struct Qdisc_class_common *cl; 334 struct hlist_node *n; 335 unsigned int h; 336 337 h = qdisc_class_hash(id, hash->hashmask); 338 hlist_for_each_entry(cl, n, &hash->hash[h], hnode) { 339 if (cl->classid == id) 340 return cl; 341 } 342 return NULL; 343} 344 345extern int qdisc_class_hash_init(struct Qdisc_class_hash *); 346extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *); 347extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *); 348extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 349extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 350 351extern void dev_init_scheduler(struct net_device *dev); 352extern void dev_shutdown(struct net_device *dev); 353extern void dev_activate(struct net_device *dev); 354extern void dev_deactivate(struct net_device *dev); 355extern void dev_deactivate_many(struct list_head *head); 356extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 357 struct Qdisc *qdisc); 358extern void qdisc_reset(struct Qdisc *qdisc); 359extern void qdisc_destroy(struct Qdisc *qdisc); 360extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n); 361extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 362 struct Qdisc_ops *ops); 363extern struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 364 struct Qdisc_ops *ops, u32 parentid); 365extern void __qdisc_calculate_pkt_len(struct sk_buff *skb, 366 const struct qdisc_size_table *stab); 367extern void tcf_destroy(struct tcf_proto *tp); 368extern void tcf_destroy_chain(struct tcf_proto **fl); 369 370/* Reset all TX qdiscs greater then index of a device. */ 371static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 372{ 373 struct Qdisc *qdisc; 374 375 for (; i < dev->num_tx_queues; i++) { 376 qdisc = netdev_get_tx_queue(dev, i)->qdisc; 377 if (qdisc) { 378 spin_lock_bh(qdisc_lock(qdisc)); 379 qdisc_reset(qdisc); 380 spin_unlock_bh(qdisc_lock(qdisc)); 381 } 382 } 383} 384 385static inline void qdisc_reset_all_tx(struct net_device *dev) 386{ 387 qdisc_reset_all_tx_gt(dev, 0); 388} 389 390/* Are all TX queues of the device empty? */ 391static inline bool qdisc_all_tx_empty(const struct net_device *dev) 392{ 393 unsigned int i; 394 for (i = 0; i < dev->num_tx_queues; i++) { 395 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 396 const struct Qdisc *q = txq->qdisc; 397 398 if (q->q.qlen) 399 return false; 400 } 401 return true; 402} 403 404/* Are any of the TX qdiscs changing? */ 405static inline bool qdisc_tx_changing(const struct net_device *dev) 406{ 407 unsigned int i; 408 for (i = 0; i < dev->num_tx_queues; i++) { 409 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 410 if (txq->qdisc != txq->qdisc_sleeping) 411 return true; 412 } 413 return false; 414} 415 416/* Is the device using the noop qdisc on all queues? */ 417static inline bool qdisc_tx_is_noop(const struct net_device *dev) 418{ 419 unsigned int i; 420 for (i = 0; i < dev->num_tx_queues; i++) { 421 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 422 if (txq->qdisc != &noop_qdisc) 423 return false; 424 } 425 return true; 426} 427 428static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 429{ 430 return qdisc_skb_cb(skb)->pkt_len; 431} 432 433/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 434enum net_xmit_qdisc_t { 435 __NET_XMIT_STOLEN = 0x00010000, 436 __NET_XMIT_BYPASS = 0x00020000, 437}; 438 439#ifdef CONFIG_NET_CLS_ACT 440#define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 441#else 442#define net_xmit_drop_count(e) (1) 443#endif 444 445static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 446 const struct Qdisc *sch) 447{ 448#ifdef CONFIG_NET_SCHED 449 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 450 451 if (stab) 452 __qdisc_calculate_pkt_len(skb, stab); 453#endif 454} 455 456static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch) 457{ 458 qdisc_calculate_pkt_len(skb, sch); 459 return sch->enqueue(skb, sch); 460} 461 462static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch) 463{ 464 qdisc_skb_cb(skb)->pkt_len = skb->len; 465 return qdisc_enqueue(skb, sch) & NET_XMIT_MASK; 466} 467 468 469static inline void bstats_update(struct gnet_stats_basic_packed *bstats, 470 const struct sk_buff *skb) 471{ 472 bstats->bytes += qdisc_pkt_len(skb); 473 bstats->packets += skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1; 474} 475 476static inline void qdisc_bstats_update(struct Qdisc *sch, 477 const struct sk_buff *skb) 478{ 479 bstats_update(&sch->bstats, skb); 480} 481 482static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch, 483 struct sk_buff_head *list) 484{ 485 __skb_queue_tail(list, skb); 486 sch->qstats.backlog += qdisc_pkt_len(skb); 487 488 return NET_XMIT_SUCCESS; 489} 490 491static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 492{ 493 return __qdisc_enqueue_tail(skb, sch, &sch->q); 494} 495 496static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch, 497 struct sk_buff_head *list) 498{ 499 struct sk_buff *skb = __skb_dequeue(list); 500 501 if (likely(skb != NULL)) { 502 sch->qstats.backlog -= qdisc_pkt_len(skb); 503 qdisc_bstats_update(sch, skb); 504 } 505 506 return skb; 507} 508 509static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 510{ 511 return __qdisc_dequeue_head(sch, &sch->q); 512} 513 514static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 515 struct sk_buff_head *list) 516{ 517 struct sk_buff *skb = __skb_dequeue(list); 518 519 if (likely(skb != NULL)) { 520 unsigned int len = qdisc_pkt_len(skb); 521 sch->qstats.backlog -= len; 522 kfree_skb(skb); 523 return len; 524 } 525 526 return 0; 527} 528 529static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch) 530{ 531 return __qdisc_queue_drop_head(sch, &sch->q); 532} 533 534static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch, 535 struct sk_buff_head *list) 536{ 537 struct sk_buff *skb = __skb_dequeue_tail(list); 538 539 if (likely(skb != NULL)) 540 sch->qstats.backlog -= qdisc_pkt_len(skb); 541 542 return skb; 543} 544 545static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch) 546{ 547 return __qdisc_dequeue_tail(sch, &sch->q); 548} 549 550static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 551{ 552 return skb_peek(&sch->q); 553} 554 555/* generic pseudo peek method for non-work-conserving qdisc */ 556static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 557{ 558 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 559 if (!sch->gso_skb) { 560 sch->gso_skb = sch->dequeue(sch); 561 if (sch->gso_skb) 562 /* it's still part of the queue */ 563 sch->q.qlen++; 564 } 565 566 return sch->gso_skb; 567} 568 569/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 570static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 571{ 572 struct sk_buff *skb = sch->gso_skb; 573 574 if (skb) { 575 sch->gso_skb = NULL; 576 sch->q.qlen--; 577 } else { 578 skb = sch->dequeue(sch); 579 } 580 581 return skb; 582} 583 584static inline void __qdisc_reset_queue(struct Qdisc *sch, 585 struct sk_buff_head *list) 586{ 587 /* 588 * We do not know the backlog in bytes of this list, it 589 * is up to the caller to correct it 590 */ 591 __skb_queue_purge(list); 592} 593 594static inline void qdisc_reset_queue(struct Qdisc *sch) 595{ 596 __qdisc_reset_queue(sch, &sch->q); 597 sch->qstats.backlog = 0; 598} 599 600static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch, 601 struct sk_buff_head *list) 602{ 603 struct sk_buff *skb = __qdisc_dequeue_tail(sch, list); 604 605 if (likely(skb != NULL)) { 606 unsigned int len = qdisc_pkt_len(skb); 607 kfree_skb(skb); 608 return len; 609 } 610 611 return 0; 612} 613 614static inline unsigned int qdisc_queue_drop(struct Qdisc *sch) 615{ 616 return __qdisc_queue_drop(sch, &sch->q); 617} 618 619static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 620{ 621 kfree_skb(skb); 622 sch->qstats.drops++; 623 624 return NET_XMIT_DROP; 625} 626 627static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch) 628{ 629 sch->qstats.drops++; 630 631#ifdef CONFIG_NET_CLS_ACT 632 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch)) 633 goto drop; 634 635 return NET_XMIT_SUCCESS; 636 637drop: 638#endif 639 kfree_skb(skb); 640 return NET_XMIT_DROP; 641} 642 643/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 644 long it will take to send a packet given its size. 645 */ 646static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 647{ 648 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 649 if (slot < 0) 650 slot = 0; 651 slot >>= rtab->rate.cell_log; 652 if (slot > 255) 653 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 654 return rtab->data[slot]; 655} 656 657#ifdef CONFIG_NET_CLS_ACT 658static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask, 659 int action) 660{ 661 struct sk_buff *n; 662 663 n = skb_clone(skb, gfp_mask); 664 665 if (n) { 666 n->tc_verd = SET_TC_VERD(n->tc_verd, 0); 667 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd); 668 n->tc_verd = CLR_TC_MUNGED(n->tc_verd); 669 } 670 return n; 671} 672#endif 673 674#endif 675