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