1 /* 2 * net/sched/sch_htb.c Hierarchical token bucket, feed tree version 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Authors: Martin Devera, <devik@cdi.cz> 10 * 11 * Credits (in time order) for older HTB versions: 12 * Stef Coene <stef.coene@docum.org> 13 * HTB support at LARTC mailing list 14 * Ondrej Kraus, <krauso@barr.cz> 15 * found missing INIT_QDISC(htb) 16 * Vladimir Smelhaus, Aamer Akhter, Bert Hubert 17 * helped a lot to locate nasty class stall bug 18 * Andi Kleen, Jamal Hadi, Bert Hubert 19 * code review and helpful comments on shaping 20 * Tomasz Wrona, <tw@eter.tym.pl> 21 * created test case so that I was able to fix nasty bug 22 * Wilfried Weissmann 23 * spotted bug in dequeue code and helped with fix 24 * Jiri Fojtasek 25 * fixed requeue routine 26 * and many others. thanks. 27 */ 28 #include <linux/module.h> 29 #include <linux/moduleparam.h> 30 #include <linux/types.h> 31 #include <linux/kernel.h> 32 #include <linux/string.h> 33 #include <linux/errno.h> 34 #include <linux/skbuff.h> 35 #include <linux/list.h> 36 #include <linux/compiler.h> 37 #include <linux/rbtree.h> 38 #include <linux/workqueue.h> 39 #include <linux/slab.h> 40 #include <net/netlink.h> 41 #include <net/sch_generic.h> 42 #include <net/pkt_sched.h> 43 #include <net/pkt_cls.h> 44 45 /* HTB algorithm. 46 Author: devik@cdi.cz 47 ======================================================================== 48 HTB is like TBF with multiple classes. It is also similar to CBQ because 49 it allows to assign priority to each class in hierarchy. 50 In fact it is another implementation of Floyd's formal sharing. 51 52 Levels: 53 Each class is assigned level. Leaf has ALWAYS level 0 and root 54 classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level 55 one less than their parent. 56 */ 57 58 static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */ 59 #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */ 60 61 #if HTB_VER >> 16 != TC_HTB_PROTOVER 62 #error "Mismatched sch_htb.c and pkt_sch.h" 63 #endif 64 65 /* Module parameter and sysfs export */ 66 module_param (htb_hysteresis, int, 0640); 67 MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate"); 68 69 static int htb_rate_est = 0; /* htb classes have a default rate estimator */ 70 module_param(htb_rate_est, int, 0640); 71 MODULE_PARM_DESC(htb_rate_est, "setup a default rate estimator (4sec 16sec) for htb classes"); 72 73 /* used internaly to keep status of single class */ 74 enum htb_cmode { 75 HTB_CANT_SEND, /* class can't send and can't borrow */ 76 HTB_MAY_BORROW, /* class can't send but may borrow */ 77 HTB_CAN_SEND /* class can send */ 78 }; 79 80 struct htb_prio { 81 union { 82 struct rb_root row; 83 struct rb_root feed; 84 }; 85 struct rb_node *ptr; 86 /* When class changes from state 1->2 and disconnects from 87 * parent's feed then we lost ptr value and start from the 88 * first child again. Here we store classid of the 89 * last valid ptr (used when ptr is NULL). 90 */ 91 u32 last_ptr_id; 92 }; 93 94 /* interior & leaf nodes; props specific to leaves are marked L: 95 * To reduce false sharing, place mostly read fields at beginning, 96 * and mostly written ones at the end. 97 */ 98 struct htb_class { 99 struct Qdisc_class_common common; 100 struct psched_ratecfg rate; 101 struct psched_ratecfg ceil; 102 s64 buffer, cbuffer;/* token bucket depth/rate */ 103 s64 mbuffer; /* max wait time */ 104 u32 prio; /* these two are used only by leaves... */ 105 int quantum; /* but stored for parent-to-leaf return */ 106 107 struct tcf_proto __rcu *filter_list; /* class attached filters */ 108 struct tcf_block *block; 109 int filter_cnt; 110 111 int level; /* our level (see above) */ 112 unsigned int children; 113 struct htb_class *parent; /* parent class */ 114 115 struct net_rate_estimator __rcu *rate_est; 116 117 /* 118 * Written often fields 119 */ 120 struct gnet_stats_basic_packed bstats; 121 struct tc_htb_xstats xstats; /* our special stats */ 122 123 /* token bucket parameters */ 124 s64 tokens, ctokens;/* current number of tokens */ 125 s64 t_c; /* checkpoint time */ 126 127 union { 128 struct htb_class_leaf { 129 struct list_head drop_list; 130 int deficit[TC_HTB_MAXDEPTH]; 131 struct Qdisc *q; 132 } leaf; 133 struct htb_class_inner { 134 struct htb_prio clprio[TC_HTB_NUMPRIO]; 135 } inner; 136 } un; 137 s64 pq_key; 138 139 int prio_activity; /* for which prios are we active */ 140 enum htb_cmode cmode; /* current mode of the class */ 141 struct rb_node pq_node; /* node for event queue */ 142 struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ 143 144 unsigned int drops ____cacheline_aligned_in_smp; 145 unsigned int overlimits; 146 }; 147 148 struct htb_level { 149 struct rb_root wait_pq; 150 struct htb_prio hprio[TC_HTB_NUMPRIO]; 151 }; 152 153 struct htb_sched { 154 struct Qdisc_class_hash clhash; 155 int defcls; /* class where unclassified flows go to */ 156 int rate2quantum; /* quant = rate / rate2quantum */ 157 158 /* filters for qdisc itself */ 159 struct tcf_proto __rcu *filter_list; 160 struct tcf_block *block; 161 162 #define HTB_WARN_TOOMANYEVENTS 0x1 163 unsigned int warned; /* only one warning */ 164 int direct_qlen; 165 struct work_struct work; 166 167 /* non shaped skbs; let them go directly thru */ 168 struct qdisc_skb_head direct_queue; 169 long direct_pkts; 170 171 struct qdisc_watchdog watchdog; 172 173 s64 now; /* cached dequeue time */ 174 struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */ 175 176 /* time of nearest event per level (row) */ 177 s64 near_ev_cache[TC_HTB_MAXDEPTH]; 178 179 int row_mask[TC_HTB_MAXDEPTH]; 180 181 struct htb_level hlevel[TC_HTB_MAXDEPTH]; 182 }; 183 184 /* find class in global hash table using given handle */ 185 static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch) 186 { 187 struct htb_sched *q = qdisc_priv(sch); 188 struct Qdisc_class_common *clc; 189 190 clc = qdisc_class_find(&q->clhash, handle); 191 if (clc == NULL) 192 return NULL; 193 return container_of(clc, struct htb_class, common); 194 } 195 196 static unsigned long htb_search(struct Qdisc *sch, u32 handle) 197 { 198 return (unsigned long)htb_find(handle, sch); 199 } 200 /** 201 * htb_classify - classify a packet into class 202 * 203 * It returns NULL if the packet should be dropped or -1 if the packet 204 * should be passed directly thru. In all other cases leaf class is returned. 205 * We allow direct class selection by classid in priority. The we examine 206 * filters in qdisc and in inner nodes (if higher filter points to the inner 207 * node). If we end up with classid MAJOR:0 we enqueue the skb into special 208 * internal fifo (direct). These packets then go directly thru. If we still 209 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful 210 * then finish and return direct queue. 211 */ 212 #define HTB_DIRECT ((struct htb_class *)-1L) 213 214 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, 215 int *qerr) 216 { 217 struct htb_sched *q = qdisc_priv(sch); 218 struct htb_class *cl; 219 struct tcf_result res; 220 struct tcf_proto *tcf; 221 int result; 222 223 /* allow to select class by setting skb->priority to valid classid; 224 * note that nfmark can be used too by attaching filter fw with no 225 * rules in it 226 */ 227 if (skb->priority == sch->handle) 228 return HTB_DIRECT; /* X:0 (direct flow) selected */ 229 cl = htb_find(skb->priority, sch); 230 if (cl) { 231 if (cl->level == 0) 232 return cl; 233 /* Start with inner filter chain if a non-leaf class is selected */ 234 tcf = rcu_dereference_bh(cl->filter_list); 235 } else { 236 tcf = rcu_dereference_bh(q->filter_list); 237 } 238 239 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 240 while (tcf && (result = tcf_classify(skb, tcf, &res, false)) >= 0) { 241 #ifdef CONFIG_NET_CLS_ACT 242 switch (result) { 243 case TC_ACT_QUEUED: 244 case TC_ACT_STOLEN: 245 case TC_ACT_TRAP: 246 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 247 case TC_ACT_SHOT: 248 return NULL; 249 } 250 #endif 251 cl = (void *)res.class; 252 if (!cl) { 253 if (res.classid == sch->handle) 254 return HTB_DIRECT; /* X:0 (direct flow) */ 255 cl = htb_find(res.classid, sch); 256 if (!cl) 257 break; /* filter selected invalid classid */ 258 } 259 if (!cl->level) 260 return cl; /* we hit leaf; return it */ 261 262 /* we have got inner class; apply inner filter chain */ 263 tcf = rcu_dereference_bh(cl->filter_list); 264 } 265 /* classification failed; try to use default class */ 266 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch); 267 if (!cl || cl->level) 268 return HTB_DIRECT; /* bad default .. this is safe bet */ 269 return cl; 270 } 271 272 /** 273 * htb_add_to_id_tree - adds class to the round robin list 274 * 275 * Routine adds class to the list (actually tree) sorted by classid. 276 * Make sure that class is not already on such list for given prio. 277 */ 278 static void htb_add_to_id_tree(struct rb_root *root, 279 struct htb_class *cl, int prio) 280 { 281 struct rb_node **p = &root->rb_node, *parent = NULL; 282 283 while (*p) { 284 struct htb_class *c; 285 parent = *p; 286 c = rb_entry(parent, struct htb_class, node[prio]); 287 288 if (cl->common.classid > c->common.classid) 289 p = &parent->rb_right; 290 else 291 p = &parent->rb_left; 292 } 293 rb_link_node(&cl->node[prio], parent, p); 294 rb_insert_color(&cl->node[prio], root); 295 } 296 297 /** 298 * htb_add_to_wait_tree - adds class to the event queue with delay 299 * 300 * The class is added to priority event queue to indicate that class will 301 * change its mode in cl->pq_key microseconds. Make sure that class is not 302 * already in the queue. 303 */ 304 static void htb_add_to_wait_tree(struct htb_sched *q, 305 struct htb_class *cl, s64 delay) 306 { 307 struct rb_node **p = &q->hlevel[cl->level].wait_pq.rb_node, *parent = NULL; 308 309 cl->pq_key = q->now + delay; 310 if (cl->pq_key == q->now) 311 cl->pq_key++; 312 313 /* update the nearest event cache */ 314 if (q->near_ev_cache[cl->level] > cl->pq_key) 315 q->near_ev_cache[cl->level] = cl->pq_key; 316 317 while (*p) { 318 struct htb_class *c; 319 parent = *p; 320 c = rb_entry(parent, struct htb_class, pq_node); 321 if (cl->pq_key >= c->pq_key) 322 p = &parent->rb_right; 323 else 324 p = &parent->rb_left; 325 } 326 rb_link_node(&cl->pq_node, parent, p); 327 rb_insert_color(&cl->pq_node, &q->hlevel[cl->level].wait_pq); 328 } 329 330 /** 331 * htb_next_rb_node - finds next node in binary tree 332 * 333 * When we are past last key we return NULL. 334 * Average complexity is 2 steps per call. 335 */ 336 static inline void htb_next_rb_node(struct rb_node **n) 337 { 338 *n = rb_next(*n); 339 } 340 341 /** 342 * htb_add_class_to_row - add class to its row 343 * 344 * The class is added to row at priorities marked in mask. 345 * It does nothing if mask == 0. 346 */ 347 static inline void htb_add_class_to_row(struct htb_sched *q, 348 struct htb_class *cl, int mask) 349 { 350 q->row_mask[cl->level] |= mask; 351 while (mask) { 352 int prio = ffz(~mask); 353 mask &= ~(1 << prio); 354 htb_add_to_id_tree(&q->hlevel[cl->level].hprio[prio].row, cl, prio); 355 } 356 } 357 358 /* If this triggers, it is a bug in this code, but it need not be fatal */ 359 static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root) 360 { 361 if (RB_EMPTY_NODE(rb)) { 362 WARN_ON(1); 363 } else { 364 rb_erase(rb, root); 365 RB_CLEAR_NODE(rb); 366 } 367 } 368 369 370 /** 371 * htb_remove_class_from_row - removes class from its row 372 * 373 * The class is removed from row at priorities marked in mask. 374 * It does nothing if mask == 0. 375 */ 376 static inline void htb_remove_class_from_row(struct htb_sched *q, 377 struct htb_class *cl, int mask) 378 { 379 int m = 0; 380 struct htb_level *hlevel = &q->hlevel[cl->level]; 381 382 while (mask) { 383 int prio = ffz(~mask); 384 struct htb_prio *hprio = &hlevel->hprio[prio]; 385 386 mask &= ~(1 << prio); 387 if (hprio->ptr == cl->node + prio) 388 htb_next_rb_node(&hprio->ptr); 389 390 htb_safe_rb_erase(cl->node + prio, &hprio->row); 391 if (!hprio->row.rb_node) 392 m |= 1 << prio; 393 } 394 q->row_mask[cl->level] &= ~m; 395 } 396 397 /** 398 * htb_activate_prios - creates active classe's feed chain 399 * 400 * The class is connected to ancestors and/or appropriate rows 401 * for priorities it is participating on. cl->cmode must be new 402 * (activated) mode. It does nothing if cl->prio_activity == 0. 403 */ 404 static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl) 405 { 406 struct htb_class *p = cl->parent; 407 long m, mask = cl->prio_activity; 408 409 while (cl->cmode == HTB_MAY_BORROW && p && mask) { 410 m = mask; 411 while (m) { 412 int prio = ffz(~m); 413 m &= ~(1 << prio); 414 415 if (p->un.inner.clprio[prio].feed.rb_node) 416 /* parent already has its feed in use so that 417 * reset bit in mask as parent is already ok 418 */ 419 mask &= ~(1 << prio); 420 421 htb_add_to_id_tree(&p->un.inner.clprio[prio].feed, cl, prio); 422 } 423 p->prio_activity |= mask; 424 cl = p; 425 p = cl->parent; 426 427 } 428 if (cl->cmode == HTB_CAN_SEND && mask) 429 htb_add_class_to_row(q, cl, mask); 430 } 431 432 /** 433 * htb_deactivate_prios - remove class from feed chain 434 * 435 * cl->cmode must represent old mode (before deactivation). It does 436 * nothing if cl->prio_activity == 0. Class is removed from all feed 437 * chains and rows. 438 */ 439 static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) 440 { 441 struct htb_class *p = cl->parent; 442 long m, mask = cl->prio_activity; 443 444 while (cl->cmode == HTB_MAY_BORROW && p && mask) { 445 m = mask; 446 mask = 0; 447 while (m) { 448 int prio = ffz(~m); 449 m &= ~(1 << prio); 450 451 if (p->un.inner.clprio[prio].ptr == cl->node + prio) { 452 /* we are removing child which is pointed to from 453 * parent feed - forget the pointer but remember 454 * classid 455 */ 456 p->un.inner.clprio[prio].last_ptr_id = cl->common.classid; 457 p->un.inner.clprio[prio].ptr = NULL; 458 } 459 460 htb_safe_rb_erase(cl->node + prio, 461 &p->un.inner.clprio[prio].feed); 462 463 if (!p->un.inner.clprio[prio].feed.rb_node) 464 mask |= 1 << prio; 465 } 466 467 p->prio_activity &= ~mask; 468 cl = p; 469 p = cl->parent; 470 471 } 472 if (cl->cmode == HTB_CAN_SEND && mask) 473 htb_remove_class_from_row(q, cl, mask); 474 } 475 476 static inline s64 htb_lowater(const struct htb_class *cl) 477 { 478 if (htb_hysteresis) 479 return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0; 480 else 481 return 0; 482 } 483 static inline s64 htb_hiwater(const struct htb_class *cl) 484 { 485 if (htb_hysteresis) 486 return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0; 487 else 488 return 0; 489 } 490 491 492 /** 493 * htb_class_mode - computes and returns current class mode 494 * 495 * It computes cl's mode at time cl->t_c+diff and returns it. If mode 496 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference 497 * from now to time when cl will change its state. 498 * Also it is worth to note that class mode doesn't change simply 499 * at cl->{c,}tokens == 0 but there can rather be hysteresis of 500 * 0 .. -cl->{c,}buffer range. It is meant to limit number of 501 * mode transitions per time unit. The speed gain is about 1/6. 502 */ 503 static inline enum htb_cmode 504 htb_class_mode(struct htb_class *cl, s64 *diff) 505 { 506 s64 toks; 507 508 if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) { 509 *diff = -toks; 510 return HTB_CANT_SEND; 511 } 512 513 if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl)) 514 return HTB_CAN_SEND; 515 516 *diff = -toks; 517 return HTB_MAY_BORROW; 518 } 519 520 /** 521 * htb_change_class_mode - changes classe's mode 522 * 523 * This should be the only way how to change classe's mode under normal 524 * cirsumstances. Routine will update feed lists linkage, change mode 525 * and add class to the wait event queue if appropriate. New mode should 526 * be different from old one and cl->pq_key has to be valid if changing 527 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). 528 */ 529 static void 530 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, s64 *diff) 531 { 532 enum htb_cmode new_mode = htb_class_mode(cl, diff); 533 534 if (new_mode == cl->cmode) 535 return; 536 537 if (new_mode == HTB_CANT_SEND) 538 cl->overlimits++; 539 540 if (cl->prio_activity) { /* not necessary: speed optimization */ 541 if (cl->cmode != HTB_CANT_SEND) 542 htb_deactivate_prios(q, cl); 543 cl->cmode = new_mode; 544 if (new_mode != HTB_CANT_SEND) 545 htb_activate_prios(q, cl); 546 } else 547 cl->cmode = new_mode; 548 } 549 550 /** 551 * htb_activate - inserts leaf cl into appropriate active feeds 552 * 553 * Routine learns (new) priority of leaf and activates feed chain 554 * for the prio. It can be called on already active leaf safely. 555 * It also adds leaf into droplist. 556 */ 557 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl) 558 { 559 WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen); 560 561 if (!cl->prio_activity) { 562 cl->prio_activity = 1 << cl->prio; 563 htb_activate_prios(q, cl); 564 list_add_tail(&cl->un.leaf.drop_list, 565 q->drops + cl->prio); 566 } 567 } 568 569 /** 570 * htb_deactivate - remove leaf cl from active feeds 571 * 572 * Make sure that leaf is active. In the other words it can't be called 573 * with non-active leaf. It also removes class from the drop list. 574 */ 575 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl) 576 { 577 WARN_ON(!cl->prio_activity); 578 579 htb_deactivate_prios(q, cl); 580 cl->prio_activity = 0; 581 list_del_init(&cl->un.leaf.drop_list); 582 } 583 584 static void htb_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch, 585 struct qdisc_skb_head *qh) 586 { 587 struct sk_buff *last = qh->tail; 588 589 if (last) { 590 skb->next = NULL; 591 last->next = skb; 592 qh->tail = skb; 593 } else { 594 qh->tail = skb; 595 qh->head = skb; 596 } 597 qh->qlen++; 598 } 599 600 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch, 601 struct sk_buff **to_free) 602 { 603 int uninitialized_var(ret); 604 struct htb_sched *q = qdisc_priv(sch); 605 struct htb_class *cl = htb_classify(skb, sch, &ret); 606 607 if (cl == HTB_DIRECT) { 608 /* enqueue to helper queue */ 609 if (q->direct_queue.qlen < q->direct_qlen) { 610 htb_enqueue_tail(skb, sch, &q->direct_queue); 611 q->direct_pkts++; 612 } else { 613 return qdisc_drop(skb, sch, to_free); 614 } 615 #ifdef CONFIG_NET_CLS_ACT 616 } else if (!cl) { 617 if (ret & __NET_XMIT_BYPASS) 618 qdisc_qstats_drop(sch); 619 __qdisc_drop(skb, to_free); 620 return ret; 621 #endif 622 } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q, 623 to_free)) != NET_XMIT_SUCCESS) { 624 if (net_xmit_drop_count(ret)) { 625 qdisc_qstats_drop(sch); 626 cl->drops++; 627 } 628 return ret; 629 } else { 630 htb_activate(q, cl); 631 } 632 633 qdisc_qstats_backlog_inc(sch, skb); 634 sch->q.qlen++; 635 return NET_XMIT_SUCCESS; 636 } 637 638 static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, s64 diff) 639 { 640 s64 toks = diff + cl->tokens; 641 642 if (toks > cl->buffer) 643 toks = cl->buffer; 644 toks -= (s64) psched_l2t_ns(&cl->rate, bytes); 645 if (toks <= -cl->mbuffer) 646 toks = 1 - cl->mbuffer; 647 648 cl->tokens = toks; 649 } 650 651 static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, s64 diff) 652 { 653 s64 toks = diff + cl->ctokens; 654 655 if (toks > cl->cbuffer) 656 toks = cl->cbuffer; 657 toks -= (s64) psched_l2t_ns(&cl->ceil, bytes); 658 if (toks <= -cl->mbuffer) 659 toks = 1 - cl->mbuffer; 660 661 cl->ctokens = toks; 662 } 663 664 /** 665 * htb_charge_class - charges amount "bytes" to leaf and ancestors 666 * 667 * Routine assumes that packet "bytes" long was dequeued from leaf cl 668 * borrowing from "level". It accounts bytes to ceil leaky bucket for 669 * leaf and all ancestors and to rate bucket for ancestors at levels 670 * "level" and higher. It also handles possible change of mode resulting 671 * from the update. Note that mode can also increase here (MAY_BORROW to 672 * CAN_SEND) because we can use more precise clock that event queue here. 673 * In such case we remove class from event queue first. 674 */ 675 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, 676 int level, struct sk_buff *skb) 677 { 678 int bytes = qdisc_pkt_len(skb); 679 enum htb_cmode old_mode; 680 s64 diff; 681 682 while (cl) { 683 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); 684 if (cl->level >= level) { 685 if (cl->level == level) 686 cl->xstats.lends++; 687 htb_accnt_tokens(cl, bytes, diff); 688 } else { 689 cl->xstats.borrows++; 690 cl->tokens += diff; /* we moved t_c; update tokens */ 691 } 692 htb_accnt_ctokens(cl, bytes, diff); 693 cl->t_c = q->now; 694 695 old_mode = cl->cmode; 696 diff = 0; 697 htb_change_class_mode(q, cl, &diff); 698 if (old_mode != cl->cmode) { 699 if (old_mode != HTB_CAN_SEND) 700 htb_safe_rb_erase(&cl->pq_node, &q->hlevel[cl->level].wait_pq); 701 if (cl->cmode != HTB_CAN_SEND) 702 htb_add_to_wait_tree(q, cl, diff); 703 } 704 705 /* update basic stats except for leaves which are already updated */ 706 if (cl->level) 707 bstats_update(&cl->bstats, skb); 708 709 cl = cl->parent; 710 } 711 } 712 713 /** 714 * htb_do_events - make mode changes to classes at the level 715 * 716 * Scans event queue for pending events and applies them. Returns time of 717 * next pending event (0 for no event in pq, q->now for too many events). 718 * Note: Applied are events whose have cl->pq_key <= q->now. 719 */ 720 static s64 htb_do_events(struct htb_sched *q, const int level, 721 unsigned long start) 722 { 723 /* don't run for longer than 2 jiffies; 2 is used instead of 724 * 1 to simplify things when jiffy is going to be incremented 725 * too soon 726 */ 727 unsigned long stop_at = start + 2; 728 struct rb_root *wait_pq = &q->hlevel[level].wait_pq; 729 730 while (time_before(jiffies, stop_at)) { 731 struct htb_class *cl; 732 s64 diff; 733 struct rb_node *p = rb_first(wait_pq); 734 735 if (!p) 736 return 0; 737 738 cl = rb_entry(p, struct htb_class, pq_node); 739 if (cl->pq_key > q->now) 740 return cl->pq_key; 741 742 htb_safe_rb_erase(p, wait_pq); 743 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); 744 htb_change_class_mode(q, cl, &diff); 745 if (cl->cmode != HTB_CAN_SEND) 746 htb_add_to_wait_tree(q, cl, diff); 747 } 748 749 /* too much load - let's continue after a break for scheduling */ 750 if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) { 751 pr_warn("htb: too many events!\n"); 752 q->warned |= HTB_WARN_TOOMANYEVENTS; 753 } 754 755 return q->now; 756 } 757 758 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL 759 * is no such one exists. 760 */ 761 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n, 762 u32 id) 763 { 764 struct rb_node *r = NULL; 765 while (n) { 766 struct htb_class *cl = 767 rb_entry(n, struct htb_class, node[prio]); 768 769 if (id > cl->common.classid) { 770 n = n->rb_right; 771 } else if (id < cl->common.classid) { 772 r = n; 773 n = n->rb_left; 774 } else { 775 return n; 776 } 777 } 778 return r; 779 } 780 781 /** 782 * htb_lookup_leaf - returns next leaf class in DRR order 783 * 784 * Find leaf where current feed pointers points to. 785 */ 786 static struct htb_class *htb_lookup_leaf(struct htb_prio *hprio, const int prio) 787 { 788 int i; 789 struct { 790 struct rb_node *root; 791 struct rb_node **pptr; 792 u32 *pid; 793 } stk[TC_HTB_MAXDEPTH], *sp = stk; 794 795 BUG_ON(!hprio->row.rb_node); 796 sp->root = hprio->row.rb_node; 797 sp->pptr = &hprio->ptr; 798 sp->pid = &hprio->last_ptr_id; 799 800 for (i = 0; i < 65535; i++) { 801 if (!*sp->pptr && *sp->pid) { 802 /* ptr was invalidated but id is valid - try to recover 803 * the original or next ptr 804 */ 805 *sp->pptr = 806 htb_id_find_next_upper(prio, sp->root, *sp->pid); 807 } 808 *sp->pid = 0; /* ptr is valid now so that remove this hint as it 809 * can become out of date quickly 810 */ 811 if (!*sp->pptr) { /* we are at right end; rewind & go up */ 812 *sp->pptr = sp->root; 813 while ((*sp->pptr)->rb_left) 814 *sp->pptr = (*sp->pptr)->rb_left; 815 if (sp > stk) { 816 sp--; 817 if (!*sp->pptr) { 818 WARN_ON(1); 819 return NULL; 820 } 821 htb_next_rb_node(sp->pptr); 822 } 823 } else { 824 struct htb_class *cl; 825 struct htb_prio *clp; 826 827 cl = rb_entry(*sp->pptr, struct htb_class, node[prio]); 828 if (!cl->level) 829 return cl; 830 clp = &cl->un.inner.clprio[prio]; 831 (++sp)->root = clp->feed.rb_node; 832 sp->pptr = &clp->ptr; 833 sp->pid = &clp->last_ptr_id; 834 } 835 } 836 WARN_ON(1); 837 return NULL; 838 } 839 840 /* dequeues packet at given priority and level; call only if 841 * you are sure that there is active class at prio/level 842 */ 843 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, const int prio, 844 const int level) 845 { 846 struct sk_buff *skb = NULL; 847 struct htb_class *cl, *start; 848 struct htb_level *hlevel = &q->hlevel[level]; 849 struct htb_prio *hprio = &hlevel->hprio[prio]; 850 851 /* look initial class up in the row */ 852 start = cl = htb_lookup_leaf(hprio, prio); 853 854 do { 855 next: 856 if (unlikely(!cl)) 857 return NULL; 858 859 /* class can be empty - it is unlikely but can be true if leaf 860 * qdisc drops packets in enqueue routine or if someone used 861 * graft operation on the leaf since last dequeue; 862 * simply deactivate and skip such class 863 */ 864 if (unlikely(cl->un.leaf.q->q.qlen == 0)) { 865 struct htb_class *next; 866 htb_deactivate(q, cl); 867 868 /* row/level might become empty */ 869 if ((q->row_mask[level] & (1 << prio)) == 0) 870 return NULL; 871 872 next = htb_lookup_leaf(hprio, prio); 873 874 if (cl == start) /* fix start if we just deleted it */ 875 start = next; 876 cl = next; 877 goto next; 878 } 879 880 skb = cl->un.leaf.q->dequeue(cl->un.leaf.q); 881 if (likely(skb != NULL)) 882 break; 883 884 qdisc_warn_nonwc("htb", cl->un.leaf.q); 885 htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr: 886 &q->hlevel[0].hprio[prio].ptr); 887 cl = htb_lookup_leaf(hprio, prio); 888 889 } while (cl != start); 890 891 if (likely(skb != NULL)) { 892 bstats_update(&cl->bstats, skb); 893 cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb); 894 if (cl->un.leaf.deficit[level] < 0) { 895 cl->un.leaf.deficit[level] += cl->quantum; 896 htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr : 897 &q->hlevel[0].hprio[prio].ptr); 898 } 899 /* this used to be after charge_class but this constelation 900 * gives us slightly better performance 901 */ 902 if (!cl->un.leaf.q->q.qlen) 903 htb_deactivate(q, cl); 904 htb_charge_class(q, cl, level, skb); 905 } 906 return skb; 907 } 908 909 static struct sk_buff *htb_dequeue(struct Qdisc *sch) 910 { 911 struct sk_buff *skb; 912 struct htb_sched *q = qdisc_priv(sch); 913 int level; 914 s64 next_event; 915 unsigned long start_at; 916 917 /* try to dequeue direct packets as high prio (!) to minimize cpu work */ 918 skb = __qdisc_dequeue_head(&q->direct_queue); 919 if (skb != NULL) { 920 ok: 921 qdisc_bstats_update(sch, skb); 922 qdisc_qstats_backlog_dec(sch, skb); 923 sch->q.qlen--; 924 return skb; 925 } 926 927 if (!sch->q.qlen) 928 goto fin; 929 q->now = ktime_get_ns(); 930 start_at = jiffies; 931 932 next_event = q->now + 5LLU * NSEC_PER_SEC; 933 934 for (level = 0; level < TC_HTB_MAXDEPTH; level++) { 935 /* common case optimization - skip event handler quickly */ 936 int m; 937 s64 event = q->near_ev_cache[level]; 938 939 if (q->now >= event) { 940 event = htb_do_events(q, level, start_at); 941 if (!event) 942 event = q->now + NSEC_PER_SEC; 943 q->near_ev_cache[level] = event; 944 } 945 946 if (next_event > event) 947 next_event = event; 948 949 m = ~q->row_mask[level]; 950 while (m != (int)(-1)) { 951 int prio = ffz(m); 952 953 m |= 1 << prio; 954 skb = htb_dequeue_tree(q, prio, level); 955 if (likely(skb != NULL)) 956 goto ok; 957 } 958 } 959 qdisc_qstats_overlimit(sch); 960 if (likely(next_event > q->now)) 961 qdisc_watchdog_schedule_ns(&q->watchdog, next_event); 962 else 963 schedule_work(&q->work); 964 fin: 965 return skb; 966 } 967 968 /* reset all classes */ 969 /* always caled under BH & queue lock */ 970 static void htb_reset(struct Qdisc *sch) 971 { 972 struct htb_sched *q = qdisc_priv(sch); 973 struct htb_class *cl; 974 unsigned int i; 975 976 for (i = 0; i < q->clhash.hashsize; i++) { 977 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 978 if (cl->level) 979 memset(&cl->un.inner, 0, sizeof(cl->un.inner)); 980 else { 981 if (cl->un.leaf.q) 982 qdisc_reset(cl->un.leaf.q); 983 INIT_LIST_HEAD(&cl->un.leaf.drop_list); 984 } 985 cl->prio_activity = 0; 986 cl->cmode = HTB_CAN_SEND; 987 } 988 } 989 qdisc_watchdog_cancel(&q->watchdog); 990 __qdisc_reset_queue(&q->direct_queue); 991 sch->q.qlen = 0; 992 sch->qstats.backlog = 0; 993 memset(q->hlevel, 0, sizeof(q->hlevel)); 994 memset(q->row_mask, 0, sizeof(q->row_mask)); 995 for (i = 0; i < TC_HTB_NUMPRIO; i++) 996 INIT_LIST_HEAD(q->drops + i); 997 } 998 999 static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = { 1000 [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) }, 1001 [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) }, 1002 [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, 1003 [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, 1004 [TCA_HTB_DIRECT_QLEN] = { .type = NLA_U32 }, 1005 [TCA_HTB_RATE64] = { .type = NLA_U64 }, 1006 [TCA_HTB_CEIL64] = { .type = NLA_U64 }, 1007 }; 1008 1009 static void htb_work_func(struct work_struct *work) 1010 { 1011 struct htb_sched *q = container_of(work, struct htb_sched, work); 1012 struct Qdisc *sch = q->watchdog.qdisc; 1013 1014 rcu_read_lock(); 1015 __netif_schedule(qdisc_root(sch)); 1016 rcu_read_unlock(); 1017 } 1018 1019 static int htb_init(struct Qdisc *sch, struct nlattr *opt) 1020 { 1021 struct htb_sched *q = qdisc_priv(sch); 1022 struct nlattr *tb[TCA_HTB_MAX + 1]; 1023 struct tc_htb_glob *gopt; 1024 int err; 1025 int i; 1026 1027 qdisc_watchdog_init(&q->watchdog, sch); 1028 INIT_WORK(&q->work, htb_work_func); 1029 1030 if (!opt) 1031 return -EINVAL; 1032 1033 err = tcf_block_get(&q->block, &q->filter_list); 1034 if (err) 1035 return err; 1036 1037 err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy, NULL); 1038 if (err < 0) 1039 return err; 1040 1041 if (!tb[TCA_HTB_INIT]) 1042 return -EINVAL; 1043 1044 gopt = nla_data(tb[TCA_HTB_INIT]); 1045 if (gopt->version != HTB_VER >> 16) 1046 return -EINVAL; 1047 1048 err = qdisc_class_hash_init(&q->clhash); 1049 if (err < 0) 1050 return err; 1051 for (i = 0; i < TC_HTB_NUMPRIO; i++) 1052 INIT_LIST_HEAD(q->drops + i); 1053 1054 qdisc_skb_head_init(&q->direct_queue); 1055 1056 if (tb[TCA_HTB_DIRECT_QLEN]) 1057 q->direct_qlen = nla_get_u32(tb[TCA_HTB_DIRECT_QLEN]); 1058 else 1059 q->direct_qlen = qdisc_dev(sch)->tx_queue_len; 1060 1061 if ((q->rate2quantum = gopt->rate2quantum) < 1) 1062 q->rate2quantum = 1; 1063 q->defcls = gopt->defcls; 1064 1065 return 0; 1066 } 1067 1068 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) 1069 { 1070 struct htb_sched *q = qdisc_priv(sch); 1071 struct nlattr *nest; 1072 struct tc_htb_glob gopt; 1073 1074 /* Its safe to not acquire qdisc lock. As we hold RTNL, 1075 * no change can happen on the qdisc parameters. 1076 */ 1077 1078 gopt.direct_pkts = q->direct_pkts; 1079 gopt.version = HTB_VER; 1080 gopt.rate2quantum = q->rate2quantum; 1081 gopt.defcls = q->defcls; 1082 gopt.debug = 0; 1083 1084 nest = nla_nest_start(skb, TCA_OPTIONS); 1085 if (nest == NULL) 1086 goto nla_put_failure; 1087 if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt) || 1088 nla_put_u32(skb, TCA_HTB_DIRECT_QLEN, q->direct_qlen)) 1089 goto nla_put_failure; 1090 1091 return nla_nest_end(skb, nest); 1092 1093 nla_put_failure: 1094 nla_nest_cancel(skb, nest); 1095 return -1; 1096 } 1097 1098 static int htb_dump_class(struct Qdisc *sch, unsigned long arg, 1099 struct sk_buff *skb, struct tcmsg *tcm) 1100 { 1101 struct htb_class *cl = (struct htb_class *)arg; 1102 struct nlattr *nest; 1103 struct tc_htb_opt opt; 1104 1105 /* Its safe to not acquire qdisc lock. As we hold RTNL, 1106 * no change can happen on the class parameters. 1107 */ 1108 tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT; 1109 tcm->tcm_handle = cl->common.classid; 1110 if (!cl->level && cl->un.leaf.q) 1111 tcm->tcm_info = cl->un.leaf.q->handle; 1112 1113 nest = nla_nest_start(skb, TCA_OPTIONS); 1114 if (nest == NULL) 1115 goto nla_put_failure; 1116 1117 memset(&opt, 0, sizeof(opt)); 1118 1119 psched_ratecfg_getrate(&opt.rate, &cl->rate); 1120 opt.buffer = PSCHED_NS2TICKS(cl->buffer); 1121 psched_ratecfg_getrate(&opt.ceil, &cl->ceil); 1122 opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer); 1123 opt.quantum = cl->quantum; 1124 opt.prio = cl->prio; 1125 opt.level = cl->level; 1126 if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt)) 1127 goto nla_put_failure; 1128 if ((cl->rate.rate_bytes_ps >= (1ULL << 32)) && 1129 nla_put_u64_64bit(skb, TCA_HTB_RATE64, cl->rate.rate_bytes_ps, 1130 TCA_HTB_PAD)) 1131 goto nla_put_failure; 1132 if ((cl->ceil.rate_bytes_ps >= (1ULL << 32)) && 1133 nla_put_u64_64bit(skb, TCA_HTB_CEIL64, cl->ceil.rate_bytes_ps, 1134 TCA_HTB_PAD)) 1135 goto nla_put_failure; 1136 1137 return nla_nest_end(skb, nest); 1138 1139 nla_put_failure: 1140 nla_nest_cancel(skb, nest); 1141 return -1; 1142 } 1143 1144 static int 1145 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d) 1146 { 1147 struct htb_class *cl = (struct htb_class *)arg; 1148 struct gnet_stats_queue qs = { 1149 .drops = cl->drops, 1150 .overlimits = cl->overlimits, 1151 }; 1152 __u32 qlen = 0; 1153 1154 if (!cl->level && cl->un.leaf.q) { 1155 qlen = cl->un.leaf.q->q.qlen; 1156 qs.backlog = cl->un.leaf.q->qstats.backlog; 1157 } 1158 cl->xstats.tokens = clamp_t(s64, PSCHED_NS2TICKS(cl->tokens), 1159 INT_MIN, INT_MAX); 1160 cl->xstats.ctokens = clamp_t(s64, PSCHED_NS2TICKS(cl->ctokens), 1161 INT_MIN, INT_MAX); 1162 1163 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), 1164 d, NULL, &cl->bstats) < 0 || 1165 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || 1166 gnet_stats_copy_queue(d, NULL, &qs, qlen) < 0) 1167 return -1; 1168 1169 return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); 1170 } 1171 1172 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 1173 struct Qdisc **old) 1174 { 1175 struct htb_class *cl = (struct htb_class *)arg; 1176 1177 if (cl->level) 1178 return -EINVAL; 1179 if (new == NULL && 1180 (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 1181 cl->common.classid)) == NULL) 1182 return -ENOBUFS; 1183 1184 *old = qdisc_replace(sch, new, &cl->un.leaf.q); 1185 return 0; 1186 } 1187 1188 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg) 1189 { 1190 struct htb_class *cl = (struct htb_class *)arg; 1191 return !cl->level ? cl->un.leaf.q : NULL; 1192 } 1193 1194 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg) 1195 { 1196 struct htb_class *cl = (struct htb_class *)arg; 1197 1198 htb_deactivate(qdisc_priv(sch), cl); 1199 } 1200 1201 static inline int htb_parent_last_child(struct htb_class *cl) 1202 { 1203 if (!cl->parent) 1204 /* the root class */ 1205 return 0; 1206 if (cl->parent->children > 1) 1207 /* not the last child */ 1208 return 0; 1209 return 1; 1210 } 1211 1212 static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl, 1213 struct Qdisc *new_q) 1214 { 1215 struct htb_class *parent = cl->parent; 1216 1217 WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity); 1218 1219 if (parent->cmode != HTB_CAN_SEND) 1220 htb_safe_rb_erase(&parent->pq_node, 1221 &q->hlevel[parent->level].wait_pq); 1222 1223 parent->level = 0; 1224 memset(&parent->un.inner, 0, sizeof(parent->un.inner)); 1225 INIT_LIST_HEAD(&parent->un.leaf.drop_list); 1226 parent->un.leaf.q = new_q ? new_q : &noop_qdisc; 1227 parent->tokens = parent->buffer; 1228 parent->ctokens = parent->cbuffer; 1229 parent->t_c = ktime_get_ns(); 1230 parent->cmode = HTB_CAN_SEND; 1231 } 1232 1233 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl) 1234 { 1235 if (!cl->level) { 1236 WARN_ON(!cl->un.leaf.q); 1237 qdisc_destroy(cl->un.leaf.q); 1238 } 1239 gen_kill_estimator(&cl->rate_est); 1240 tcf_block_put(cl->block); 1241 kfree(cl); 1242 } 1243 1244 static void htb_destroy(struct Qdisc *sch) 1245 { 1246 struct htb_sched *q = qdisc_priv(sch); 1247 struct hlist_node *next; 1248 struct htb_class *cl; 1249 unsigned int i; 1250 1251 cancel_work_sync(&q->work); 1252 qdisc_watchdog_cancel(&q->watchdog); 1253 /* This line used to be after htb_destroy_class call below 1254 * and surprisingly it worked in 2.4. But it must precede it 1255 * because filter need its target class alive to be able to call 1256 * unbind_filter on it (without Oops). 1257 */ 1258 tcf_block_put(q->block); 1259 1260 for (i = 0; i < q->clhash.hashsize; i++) { 1261 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 1262 tcf_block_put(cl->block); 1263 cl->block = NULL; 1264 } 1265 } 1266 for (i = 0; i < q->clhash.hashsize; i++) { 1267 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i], 1268 common.hnode) 1269 htb_destroy_class(sch, cl); 1270 } 1271 qdisc_class_hash_destroy(&q->clhash); 1272 __qdisc_reset_queue(&q->direct_queue); 1273 } 1274 1275 static int htb_delete(struct Qdisc *sch, unsigned long arg) 1276 { 1277 struct htb_sched *q = qdisc_priv(sch); 1278 struct htb_class *cl = (struct htb_class *)arg; 1279 struct Qdisc *new_q = NULL; 1280 int last_child = 0; 1281 1282 /* TODO: why don't allow to delete subtree ? references ? does 1283 * tc subsys guarantee us that in htb_destroy it holds no class 1284 * refs so that we can remove children safely there ? 1285 */ 1286 if (cl->children || cl->filter_cnt) 1287 return -EBUSY; 1288 1289 if (!cl->level && htb_parent_last_child(cl)) { 1290 new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 1291 cl->parent->common.classid); 1292 last_child = 1; 1293 } 1294 1295 sch_tree_lock(sch); 1296 1297 if (!cl->level) { 1298 unsigned int qlen = cl->un.leaf.q->q.qlen; 1299 unsigned int backlog = cl->un.leaf.q->qstats.backlog; 1300 1301 qdisc_reset(cl->un.leaf.q); 1302 qdisc_tree_reduce_backlog(cl->un.leaf.q, qlen, backlog); 1303 } 1304 1305 /* delete from hash and active; remainder in destroy_class */ 1306 qdisc_class_hash_remove(&q->clhash, &cl->common); 1307 if (cl->parent) 1308 cl->parent->children--; 1309 1310 if (cl->prio_activity) 1311 htb_deactivate(q, cl); 1312 1313 if (cl->cmode != HTB_CAN_SEND) 1314 htb_safe_rb_erase(&cl->pq_node, 1315 &q->hlevel[cl->level].wait_pq); 1316 1317 if (last_child) 1318 htb_parent_to_leaf(q, cl, new_q); 1319 1320 sch_tree_unlock(sch); 1321 1322 htb_destroy_class(sch, cl); 1323 return 0; 1324 } 1325 1326 static int htb_change_class(struct Qdisc *sch, u32 classid, 1327 u32 parentid, struct nlattr **tca, 1328 unsigned long *arg) 1329 { 1330 int err = -EINVAL; 1331 struct htb_sched *q = qdisc_priv(sch); 1332 struct htb_class *cl = (struct htb_class *)*arg, *parent; 1333 struct nlattr *opt = tca[TCA_OPTIONS]; 1334 struct nlattr *tb[TCA_HTB_MAX + 1]; 1335 struct tc_htb_opt *hopt; 1336 u64 rate64, ceil64; 1337 1338 /* extract all subattrs from opt attr */ 1339 if (!opt) 1340 goto failure; 1341 1342 err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy, NULL); 1343 if (err < 0) 1344 goto failure; 1345 1346 err = -EINVAL; 1347 if (tb[TCA_HTB_PARMS] == NULL) 1348 goto failure; 1349 1350 parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch); 1351 1352 hopt = nla_data(tb[TCA_HTB_PARMS]); 1353 if (!hopt->rate.rate || !hopt->ceil.rate) 1354 goto failure; 1355 1356 /* Keeping backward compatible with rate_table based iproute2 tc */ 1357 if (hopt->rate.linklayer == TC_LINKLAYER_UNAWARE) 1358 qdisc_put_rtab(qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB])); 1359 1360 if (hopt->ceil.linklayer == TC_LINKLAYER_UNAWARE) 1361 qdisc_put_rtab(qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB])); 1362 1363 if (!cl) { /* new class */ 1364 struct Qdisc *new_q; 1365 int prio; 1366 struct { 1367 struct nlattr nla; 1368 struct gnet_estimator opt; 1369 } est = { 1370 .nla = { 1371 .nla_len = nla_attr_size(sizeof(est.opt)), 1372 .nla_type = TCA_RATE, 1373 }, 1374 .opt = { 1375 /* 4s interval, 16s averaging constant */ 1376 .interval = 2, 1377 .ewma_log = 2, 1378 }, 1379 }; 1380 1381 /* check for valid classid */ 1382 if (!classid || TC_H_MAJ(classid ^ sch->handle) || 1383 htb_find(classid, sch)) 1384 goto failure; 1385 1386 /* check maximal depth */ 1387 if (parent && parent->parent && parent->parent->level < 2) { 1388 pr_err("htb: tree is too deep\n"); 1389 goto failure; 1390 } 1391 err = -ENOBUFS; 1392 cl = kzalloc(sizeof(*cl), GFP_KERNEL); 1393 if (!cl) 1394 goto failure; 1395 1396 err = tcf_block_get(&cl->block, &cl->filter_list); 1397 if (err) { 1398 kfree(cl); 1399 goto failure; 1400 } 1401 if (htb_rate_est || tca[TCA_RATE]) { 1402 err = gen_new_estimator(&cl->bstats, NULL, 1403 &cl->rate_est, 1404 NULL, 1405 qdisc_root_sleeping_running(sch), 1406 tca[TCA_RATE] ? : &est.nla); 1407 if (err) { 1408 tcf_block_put(cl->block); 1409 kfree(cl); 1410 goto failure; 1411 } 1412 } 1413 1414 cl->children = 0; 1415 INIT_LIST_HEAD(&cl->un.leaf.drop_list); 1416 RB_CLEAR_NODE(&cl->pq_node); 1417 1418 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) 1419 RB_CLEAR_NODE(&cl->node[prio]); 1420 1421 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL) 1422 * so that can't be used inside of sch_tree_lock 1423 * -- thanks to Karlis Peisenieks 1424 */ 1425 new_q = qdisc_create_dflt(sch->dev_queue, 1426 &pfifo_qdisc_ops, classid); 1427 sch_tree_lock(sch); 1428 if (parent && !parent->level) { 1429 unsigned int qlen = parent->un.leaf.q->q.qlen; 1430 unsigned int backlog = parent->un.leaf.q->qstats.backlog; 1431 1432 /* turn parent into inner node */ 1433 qdisc_reset(parent->un.leaf.q); 1434 qdisc_tree_reduce_backlog(parent->un.leaf.q, qlen, backlog); 1435 qdisc_destroy(parent->un.leaf.q); 1436 if (parent->prio_activity) 1437 htb_deactivate(q, parent); 1438 1439 /* remove from evt list because of level change */ 1440 if (parent->cmode != HTB_CAN_SEND) { 1441 htb_safe_rb_erase(&parent->pq_node, &q->hlevel[0].wait_pq); 1442 parent->cmode = HTB_CAN_SEND; 1443 } 1444 parent->level = (parent->parent ? parent->parent->level 1445 : TC_HTB_MAXDEPTH) - 1; 1446 memset(&parent->un.inner, 0, sizeof(parent->un.inner)); 1447 } 1448 /* leaf (we) needs elementary qdisc */ 1449 cl->un.leaf.q = new_q ? new_q : &noop_qdisc; 1450 1451 cl->common.classid = classid; 1452 cl->parent = parent; 1453 1454 /* set class to be in HTB_CAN_SEND state */ 1455 cl->tokens = PSCHED_TICKS2NS(hopt->buffer); 1456 cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer); 1457 cl->mbuffer = 60ULL * NSEC_PER_SEC; /* 1min */ 1458 cl->t_c = ktime_get_ns(); 1459 cl->cmode = HTB_CAN_SEND; 1460 1461 /* attach to the hash list and parent's family */ 1462 qdisc_class_hash_insert(&q->clhash, &cl->common); 1463 if (parent) 1464 parent->children++; 1465 if (cl->un.leaf.q != &noop_qdisc) 1466 qdisc_hash_add(cl->un.leaf.q, true); 1467 } else { 1468 if (tca[TCA_RATE]) { 1469 err = gen_replace_estimator(&cl->bstats, NULL, 1470 &cl->rate_est, 1471 NULL, 1472 qdisc_root_sleeping_running(sch), 1473 tca[TCA_RATE]); 1474 if (err) 1475 return err; 1476 } 1477 sch_tree_lock(sch); 1478 } 1479 1480 rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0; 1481 1482 ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0; 1483 1484 psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64); 1485 psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64); 1486 1487 /* it used to be a nasty bug here, we have to check that node 1488 * is really leaf before changing cl->un.leaf ! 1489 */ 1490 if (!cl->level) { 1491 u64 quantum = cl->rate.rate_bytes_ps; 1492 1493 do_div(quantum, q->rate2quantum); 1494 cl->quantum = min_t(u64, quantum, INT_MAX); 1495 1496 if (!hopt->quantum && cl->quantum < 1000) { 1497 pr_warn("HTB: quantum of class %X is small. Consider r2q change.\n", 1498 cl->common.classid); 1499 cl->quantum = 1000; 1500 } 1501 if (!hopt->quantum && cl->quantum > 200000) { 1502 pr_warn("HTB: quantum of class %X is big. Consider r2q change.\n", 1503 cl->common.classid); 1504 cl->quantum = 200000; 1505 } 1506 if (hopt->quantum) 1507 cl->quantum = hopt->quantum; 1508 if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO) 1509 cl->prio = TC_HTB_NUMPRIO - 1; 1510 } 1511 1512 cl->buffer = PSCHED_TICKS2NS(hopt->buffer); 1513 cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer); 1514 1515 sch_tree_unlock(sch); 1516 1517 qdisc_class_hash_grow(sch, &q->clhash); 1518 1519 *arg = (unsigned long)cl; 1520 return 0; 1521 1522 failure: 1523 return err; 1524 } 1525 1526 static struct tcf_block *htb_tcf_block(struct Qdisc *sch, unsigned long arg) 1527 { 1528 struct htb_sched *q = qdisc_priv(sch); 1529 struct htb_class *cl = (struct htb_class *)arg; 1530 1531 return cl ? cl->block : q->block; 1532 } 1533 1534 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, 1535 u32 classid) 1536 { 1537 struct htb_class *cl = htb_find(classid, sch); 1538 1539 /*if (cl && !cl->level) return 0; 1540 * The line above used to be there to prevent attaching filters to 1541 * leaves. But at least tc_index filter uses this just to get class 1542 * for other reasons so that we have to allow for it. 1543 * ---- 1544 * 19.6.2002 As Werner explained it is ok - bind filter is just 1545 * another way to "lock" the class - unlike "get" this lock can 1546 * be broken by class during destroy IIUC. 1547 */ 1548 if (cl) 1549 cl->filter_cnt++; 1550 return (unsigned long)cl; 1551 } 1552 1553 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) 1554 { 1555 struct htb_class *cl = (struct htb_class *)arg; 1556 1557 if (cl) 1558 cl->filter_cnt--; 1559 } 1560 1561 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) 1562 { 1563 struct htb_sched *q = qdisc_priv(sch); 1564 struct htb_class *cl; 1565 unsigned int i; 1566 1567 if (arg->stop) 1568 return; 1569 1570 for (i = 0; i < q->clhash.hashsize; i++) { 1571 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 1572 if (arg->count < arg->skip) { 1573 arg->count++; 1574 continue; 1575 } 1576 if (arg->fn(sch, (unsigned long)cl, arg) < 0) { 1577 arg->stop = 1; 1578 return; 1579 } 1580 arg->count++; 1581 } 1582 } 1583 } 1584 1585 static const struct Qdisc_class_ops htb_class_ops = { 1586 .graft = htb_graft, 1587 .leaf = htb_leaf, 1588 .qlen_notify = htb_qlen_notify, 1589 .find = htb_search, 1590 .change = htb_change_class, 1591 .delete = htb_delete, 1592 .walk = htb_walk, 1593 .tcf_block = htb_tcf_block, 1594 .bind_tcf = htb_bind_filter, 1595 .unbind_tcf = htb_unbind_filter, 1596 .dump = htb_dump_class, 1597 .dump_stats = htb_dump_class_stats, 1598 }; 1599 1600 static struct Qdisc_ops htb_qdisc_ops __read_mostly = { 1601 .cl_ops = &htb_class_ops, 1602 .id = "htb", 1603 .priv_size = sizeof(struct htb_sched), 1604 .enqueue = htb_enqueue, 1605 .dequeue = htb_dequeue, 1606 .peek = qdisc_peek_dequeued, 1607 .init = htb_init, 1608 .reset = htb_reset, 1609 .destroy = htb_destroy, 1610 .dump = htb_dump, 1611 .owner = THIS_MODULE, 1612 }; 1613 1614 static int __init htb_module_init(void) 1615 { 1616 return register_qdisc(&htb_qdisc_ops); 1617 } 1618 static void __exit htb_module_exit(void) 1619 { 1620 unregister_qdisc(&htb_qdisc_ops); 1621 } 1622 1623 module_init(htb_module_init) 1624 module_exit(htb_module_exit) 1625 MODULE_LICENSE("GPL"); 1626