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