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 44 /* HTB algorithm. 45 Author: devik@cdi.cz 46 ======================================================================== 47 HTB is like TBF with multiple classes. It is also similar to CBQ because 48 it allows to assign priority to each class in hierarchy. 49 In fact it is another implementation of Floyd's formal sharing. 50 51 Levels: 52 Each class is assigned level. Leaf has ALWAYS level 0 and root 53 classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level 54 one less than their parent. 55 */ 56 57 static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */ 58 #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */ 59 60 #if HTB_VER >> 16 != TC_HTB_PROTOVER 61 #error "Mismatched sch_htb.c and pkt_sch.h" 62 #endif 63 64 /* Module parameter and sysfs export */ 65 module_param (htb_hysteresis, int, 0640); 66 MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate"); 67 68 static int htb_rate_est = 0; /* htb classes have a default rate estimator */ 69 module_param(htb_rate_est, int, 0640); 70 MODULE_PARM_DESC(htb_rate_est, "setup a default rate estimator (4sec 16sec) for htb classes"); 71 72 /* used internaly to keep status of single class */ 73 enum htb_cmode { 74 HTB_CANT_SEND, /* class can't send and can't borrow */ 75 HTB_MAY_BORROW, /* class can't send but may borrow */ 76 HTB_CAN_SEND /* class can send */ 77 }; 78 79 struct htb_prio { 80 union { 81 struct rb_root row; 82 struct rb_root feed; 83 }; 84 struct rb_node *ptr; 85 /* When class changes from state 1->2 and disconnects from 86 * parent's feed then we lost ptr value and start from the 87 * first child again. Here we store classid of the 88 * last valid ptr (used when ptr is NULL). 89 */ 90 u32 last_ptr_id; 91 }; 92 93 /* interior & leaf nodes; props specific to leaves are marked L: 94 * To reduce false sharing, place mostly read fields at beginning, 95 * and mostly written ones at the end. 96 */ 97 struct htb_class { 98 struct Qdisc_class_common common; 99 struct psched_ratecfg rate; 100 struct psched_ratecfg ceil; 101 s64 buffer, cbuffer;/* token bucket depth/rate */ 102 s64 mbuffer; /* max wait time */ 103 u32 prio; /* these two are used only by leaves... */ 104 int quantum; /* but stored for parent-to-leaf return */ 105 106 struct tcf_proto __rcu *filter_list; /* class attached filters */ 107 int filter_cnt; 108 int refcnt; /* usage count of this class */ 109 110 int level; /* our level (see above) */ 111 unsigned int children; 112 struct htb_class *parent; /* parent class */ 113 114 struct gnet_stats_rate_est64 rate_est; 115 116 /* 117 * Written often fields 118 */ 119 struct gnet_stats_basic_packed bstats; 120 struct tc_htb_xstats xstats; /* our special stats */ 121 122 /* token bucket parameters */ 123 s64 tokens, ctokens;/* current number of tokens */ 124 s64 t_c; /* checkpoint time */ 125 126 union { 127 struct htb_class_leaf { 128 struct list_head drop_list; 129 int deficit[TC_HTB_MAXDEPTH]; 130 struct Qdisc *q; 131 } leaf; 132 struct htb_class_inner { 133 struct htb_prio clprio[TC_HTB_NUMPRIO]; 134 } inner; 135 } un; 136 s64 pq_key; 137 138 int prio_activity; /* for which prios are we active */ 139 enum htb_cmode cmode; /* current mode of the class */ 140 struct rb_node pq_node; /* node for event queue */ 141 struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ 142 143 unsigned int drops ____cacheline_aligned_in_smp; 144 }; 145 146 struct htb_level { 147 struct rb_root wait_pq; 148 struct htb_prio hprio[TC_HTB_NUMPRIO]; 149 }; 150 151 struct htb_sched { 152 struct Qdisc_class_hash clhash; 153 int defcls; /* class where unclassified flows go to */ 154 int rate2quantum; /* quant = rate / rate2quantum */ 155 156 /* filters for qdisc itself */ 157 struct tcf_proto __rcu *filter_list; 158 159 #define HTB_WARN_TOOMANYEVENTS 0x1 160 unsigned int warned; /* only one warning */ 161 int direct_qlen; 162 struct work_struct work; 163 164 /* non shaped skbs; let them go directly thru */ 165 struct sk_buff_head direct_queue; 166 long direct_pkts; 167 168 struct qdisc_watchdog watchdog; 169 170 s64 now; /* cached dequeue time */ 171 struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */ 172 173 /* time of nearest event per level (row) */ 174 s64 near_ev_cache[TC_HTB_MAXDEPTH]; 175 176 int row_mask[TC_HTB_MAXDEPTH]; 177 178 struct htb_level hlevel[TC_HTB_MAXDEPTH]; 179 }; 180 181 /* find class in global hash table using given handle */ 182 static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch) 183 { 184 struct htb_sched *q = qdisc_priv(sch); 185 struct Qdisc_class_common *clc; 186 187 clc = qdisc_class_find(&q->clhash, handle); 188 if (clc == NULL) 189 return NULL; 190 return container_of(clc, struct htb_class, common); 191 } 192 193 /** 194 * htb_classify - classify a packet into class 195 * 196 * It returns NULL if the packet should be dropped or -1 if the packet 197 * should be passed directly thru. In all other cases leaf class is returned. 198 * We allow direct class selection by classid in priority. The we examine 199 * filters in qdisc and in inner nodes (if higher filter points to the inner 200 * node). If we end up with classid MAJOR:0 we enqueue the skb into special 201 * internal fifo (direct). These packets then go directly thru. If we still 202 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful 203 * then finish and return direct queue. 204 */ 205 #define HTB_DIRECT ((struct htb_class *)-1L) 206 207 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, 208 int *qerr) 209 { 210 struct htb_sched *q = qdisc_priv(sch); 211 struct htb_class *cl; 212 struct tcf_result res; 213 struct tcf_proto *tcf; 214 int result; 215 216 /* allow to select class by setting skb->priority to valid classid; 217 * note that nfmark can be used too by attaching filter fw with no 218 * rules in it 219 */ 220 if (skb->priority == sch->handle) 221 return HTB_DIRECT; /* X:0 (direct flow) selected */ 222 cl = htb_find(skb->priority, sch); 223 if (cl) { 224 if (cl->level == 0) 225 return cl; 226 /* Start with inner filter chain if a non-leaf class is selected */ 227 tcf = rcu_dereference_bh(cl->filter_list); 228 } else { 229 tcf = rcu_dereference_bh(q->filter_list); 230 } 231 232 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 233 while (tcf && (result = tc_classify(skb, tcf, &res, false)) >= 0) { 234 #ifdef CONFIG_NET_CLS_ACT 235 switch (result) { 236 case TC_ACT_QUEUED: 237 case TC_ACT_STOLEN: 238 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 239 case TC_ACT_SHOT: 240 return NULL; 241 } 242 #endif 243 cl = (void *)res.class; 244 if (!cl) { 245 if (res.classid == sch->handle) 246 return HTB_DIRECT; /* X:0 (direct flow) */ 247 cl = htb_find(res.classid, sch); 248 if (!cl) 249 break; /* filter selected invalid classid */ 250 } 251 if (!cl->level) 252 return cl; /* we hit leaf; return it */ 253 254 /* we have got inner class; apply inner filter chain */ 255 tcf = rcu_dereference_bh(cl->filter_list); 256 } 257 /* classification failed; try to use default class */ 258 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch); 259 if (!cl || cl->level) 260 return HTB_DIRECT; /* bad default .. this is safe bet */ 261 return cl; 262 } 263 264 /** 265 * htb_add_to_id_tree - adds class to the round robin list 266 * 267 * Routine adds class to the list (actually tree) sorted by classid. 268 * Make sure that class is not already on such list for given prio. 269 */ 270 static void htb_add_to_id_tree(struct rb_root *root, 271 struct htb_class *cl, int prio) 272 { 273 struct rb_node **p = &root->rb_node, *parent = NULL; 274 275 while (*p) { 276 struct htb_class *c; 277 parent = *p; 278 c = rb_entry(parent, struct htb_class, node[prio]); 279 280 if (cl->common.classid > c->common.classid) 281 p = &parent->rb_right; 282 else 283 p = &parent->rb_left; 284 } 285 rb_link_node(&cl->node[prio], parent, p); 286 rb_insert_color(&cl->node[prio], root); 287 } 288 289 /** 290 * htb_add_to_wait_tree - adds class to the event queue with delay 291 * 292 * The class is added to priority event queue to indicate that class will 293 * change its mode in cl->pq_key microseconds. Make sure that class is not 294 * already in the queue. 295 */ 296 static void htb_add_to_wait_tree(struct htb_sched *q, 297 struct htb_class *cl, s64 delay) 298 { 299 struct rb_node **p = &q->hlevel[cl->level].wait_pq.rb_node, *parent = NULL; 300 301 cl->pq_key = q->now + delay; 302 if (cl->pq_key == q->now) 303 cl->pq_key++; 304 305 /* update the nearest event cache */ 306 if (q->near_ev_cache[cl->level] > cl->pq_key) 307 q->near_ev_cache[cl->level] = cl->pq_key; 308 309 while (*p) { 310 struct htb_class *c; 311 parent = *p; 312 c = rb_entry(parent, struct htb_class, pq_node); 313 if (cl->pq_key >= c->pq_key) 314 p = &parent->rb_right; 315 else 316 p = &parent->rb_left; 317 } 318 rb_link_node(&cl->pq_node, parent, p); 319 rb_insert_color(&cl->pq_node, &q->hlevel[cl->level].wait_pq); 320 } 321 322 /** 323 * htb_next_rb_node - finds next node in binary tree 324 * 325 * When we are past last key we return NULL. 326 * Average complexity is 2 steps per call. 327 */ 328 static inline void htb_next_rb_node(struct rb_node **n) 329 { 330 *n = rb_next(*n); 331 } 332 333 /** 334 * htb_add_class_to_row - add class to its row 335 * 336 * The class is added to row at priorities marked in mask. 337 * It does nothing if mask == 0. 338 */ 339 static inline void htb_add_class_to_row(struct htb_sched *q, 340 struct htb_class *cl, int mask) 341 { 342 q->row_mask[cl->level] |= mask; 343 while (mask) { 344 int prio = ffz(~mask); 345 mask &= ~(1 << prio); 346 htb_add_to_id_tree(&q->hlevel[cl->level].hprio[prio].row, cl, prio); 347 } 348 } 349 350 /* If this triggers, it is a bug in this code, but it need not be fatal */ 351 static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root) 352 { 353 if (RB_EMPTY_NODE(rb)) { 354 WARN_ON(1); 355 } else { 356 rb_erase(rb, root); 357 RB_CLEAR_NODE(rb); 358 } 359 } 360 361 362 /** 363 * htb_remove_class_from_row - removes class from its row 364 * 365 * The class is removed from row at priorities marked in mask. 366 * It does nothing if mask == 0. 367 */ 368 static inline void htb_remove_class_from_row(struct htb_sched *q, 369 struct htb_class *cl, int mask) 370 { 371 int m = 0; 372 struct htb_level *hlevel = &q->hlevel[cl->level]; 373 374 while (mask) { 375 int prio = ffz(~mask); 376 struct htb_prio *hprio = &hlevel->hprio[prio]; 377 378 mask &= ~(1 << prio); 379 if (hprio->ptr == cl->node + prio) 380 htb_next_rb_node(&hprio->ptr); 381 382 htb_safe_rb_erase(cl->node + prio, &hprio->row); 383 if (!hprio->row.rb_node) 384 m |= 1 << prio; 385 } 386 q->row_mask[cl->level] &= ~m; 387 } 388 389 /** 390 * htb_activate_prios - creates active classe's feed chain 391 * 392 * The class is connected to ancestors and/or appropriate rows 393 * for priorities it is participating on. cl->cmode must be new 394 * (activated) mode. It does nothing if cl->prio_activity == 0. 395 */ 396 static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl) 397 { 398 struct htb_class *p = cl->parent; 399 long m, mask = cl->prio_activity; 400 401 while (cl->cmode == HTB_MAY_BORROW && p && mask) { 402 m = mask; 403 while (m) { 404 int prio = ffz(~m); 405 m &= ~(1 << prio); 406 407 if (p->un.inner.clprio[prio].feed.rb_node) 408 /* parent already has its feed in use so that 409 * reset bit in mask as parent is already ok 410 */ 411 mask &= ~(1 << prio); 412 413 htb_add_to_id_tree(&p->un.inner.clprio[prio].feed, cl, prio); 414 } 415 p->prio_activity |= mask; 416 cl = p; 417 p = cl->parent; 418 419 } 420 if (cl->cmode == HTB_CAN_SEND && mask) 421 htb_add_class_to_row(q, cl, mask); 422 } 423 424 /** 425 * htb_deactivate_prios - remove class from feed chain 426 * 427 * cl->cmode must represent old mode (before deactivation). It does 428 * nothing if cl->prio_activity == 0. Class is removed from all feed 429 * chains and rows. 430 */ 431 static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) 432 { 433 struct htb_class *p = cl->parent; 434 long m, mask = cl->prio_activity; 435 436 while (cl->cmode == HTB_MAY_BORROW && p && mask) { 437 m = mask; 438 mask = 0; 439 while (m) { 440 int prio = ffz(~m); 441 m &= ~(1 << prio); 442 443 if (p->un.inner.clprio[prio].ptr == cl->node + prio) { 444 /* we are removing child which is pointed to from 445 * parent feed - forget the pointer but remember 446 * classid 447 */ 448 p->un.inner.clprio[prio].last_ptr_id = cl->common.classid; 449 p->un.inner.clprio[prio].ptr = NULL; 450 } 451 452 htb_safe_rb_erase(cl->node + prio, 453 &p->un.inner.clprio[prio].feed); 454 455 if (!p->un.inner.clprio[prio].feed.rb_node) 456 mask |= 1 << prio; 457 } 458 459 p->prio_activity &= ~mask; 460 cl = p; 461 p = cl->parent; 462 463 } 464 if (cl->cmode == HTB_CAN_SEND && mask) 465 htb_remove_class_from_row(q, cl, mask); 466 } 467 468 static inline s64 htb_lowater(const struct htb_class *cl) 469 { 470 if (htb_hysteresis) 471 return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0; 472 else 473 return 0; 474 } 475 static inline s64 htb_hiwater(const struct htb_class *cl) 476 { 477 if (htb_hysteresis) 478 return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0; 479 else 480 return 0; 481 } 482 483 484 /** 485 * htb_class_mode - computes and returns current class mode 486 * 487 * It computes cl's mode at time cl->t_c+diff and returns it. If mode 488 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference 489 * from now to time when cl will change its state. 490 * Also it is worth to note that class mode doesn't change simply 491 * at cl->{c,}tokens == 0 but there can rather be hysteresis of 492 * 0 .. -cl->{c,}buffer range. It is meant to limit number of 493 * mode transitions per time unit. The speed gain is about 1/6. 494 */ 495 static inline enum htb_cmode 496 htb_class_mode(struct htb_class *cl, s64 *diff) 497 { 498 s64 toks; 499 500 if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) { 501 *diff = -toks; 502 return HTB_CANT_SEND; 503 } 504 505 if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl)) 506 return HTB_CAN_SEND; 507 508 *diff = -toks; 509 return HTB_MAY_BORROW; 510 } 511 512 /** 513 * htb_change_class_mode - changes classe's mode 514 * 515 * This should be the only way how to change classe's mode under normal 516 * cirsumstances. Routine will update feed lists linkage, change mode 517 * and add class to the wait event queue if appropriate. New mode should 518 * be different from old one and cl->pq_key has to be valid if changing 519 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). 520 */ 521 static void 522 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, s64 *diff) 523 { 524 enum htb_cmode new_mode = htb_class_mode(cl, diff); 525 526 if (new_mode == cl->cmode) 527 return; 528 529 if (cl->prio_activity) { /* not necessary: speed optimization */ 530 if (cl->cmode != HTB_CANT_SEND) 531 htb_deactivate_prios(q, cl); 532 cl->cmode = new_mode; 533 if (new_mode != HTB_CANT_SEND) 534 htb_activate_prios(q, cl); 535 } else 536 cl->cmode = new_mode; 537 } 538 539 /** 540 * htb_activate - inserts leaf cl into appropriate active feeds 541 * 542 * Routine learns (new) priority of leaf and activates feed chain 543 * for the prio. It can be called on already active leaf safely. 544 * It also adds leaf into droplist. 545 */ 546 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl) 547 { 548 WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen); 549 550 if (!cl->prio_activity) { 551 cl->prio_activity = 1 << cl->prio; 552 htb_activate_prios(q, cl); 553 list_add_tail(&cl->un.leaf.drop_list, 554 q->drops + cl->prio); 555 } 556 } 557 558 /** 559 * htb_deactivate - remove leaf cl from active feeds 560 * 561 * Make sure that leaf is active. In the other words it can't be called 562 * with non-active leaf. It also removes class from the drop list. 563 */ 564 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl) 565 { 566 WARN_ON(!cl->prio_activity); 567 568 htb_deactivate_prios(q, cl); 569 cl->prio_activity = 0; 570 list_del_init(&cl->un.leaf.drop_list); 571 } 572 573 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch, 574 struct sk_buff **to_free) 575 { 576 int uninitialized_var(ret); 577 struct htb_sched *q = qdisc_priv(sch); 578 struct htb_class *cl = htb_classify(skb, sch, &ret); 579 580 if (cl == HTB_DIRECT) { 581 /* enqueue to helper queue */ 582 if (q->direct_queue.qlen < q->direct_qlen) { 583 __skb_queue_tail(&q->direct_queue, skb); 584 q->direct_pkts++; 585 } else { 586 return qdisc_drop(skb, sch, to_free); 587 } 588 #ifdef CONFIG_NET_CLS_ACT 589 } else if (!cl) { 590 if (ret & __NET_XMIT_BYPASS) 591 qdisc_qstats_drop(sch); 592 __qdisc_drop(skb, to_free); 593 return ret; 594 #endif 595 } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q, 596 to_free)) != NET_XMIT_SUCCESS) { 597 if (net_xmit_drop_count(ret)) { 598 qdisc_qstats_drop(sch); 599 cl->drops++; 600 } 601 return ret; 602 } else { 603 htb_activate(q, cl); 604 } 605 606 qdisc_qstats_backlog_inc(sch, skb); 607 sch->q.qlen++; 608 return NET_XMIT_SUCCESS; 609 } 610 611 static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, s64 diff) 612 { 613 s64 toks = diff + cl->tokens; 614 615 if (toks > cl->buffer) 616 toks = cl->buffer; 617 toks -= (s64) psched_l2t_ns(&cl->rate, bytes); 618 if (toks <= -cl->mbuffer) 619 toks = 1 - cl->mbuffer; 620 621 cl->tokens = toks; 622 } 623 624 static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, s64 diff) 625 { 626 s64 toks = diff + cl->ctokens; 627 628 if (toks > cl->cbuffer) 629 toks = cl->cbuffer; 630 toks -= (s64) psched_l2t_ns(&cl->ceil, bytes); 631 if (toks <= -cl->mbuffer) 632 toks = 1 - cl->mbuffer; 633 634 cl->ctokens = toks; 635 } 636 637 /** 638 * htb_charge_class - charges amount "bytes" to leaf and ancestors 639 * 640 * Routine assumes that packet "bytes" long was dequeued from leaf cl 641 * borrowing from "level". It accounts bytes to ceil leaky bucket for 642 * leaf and all ancestors and to rate bucket for ancestors at levels 643 * "level" and higher. It also handles possible change of mode resulting 644 * from the update. Note that mode can also increase here (MAY_BORROW to 645 * CAN_SEND) because we can use more precise clock that event queue here. 646 * In such case we remove class from event queue first. 647 */ 648 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, 649 int level, struct sk_buff *skb) 650 { 651 int bytes = qdisc_pkt_len(skb); 652 enum htb_cmode old_mode; 653 s64 diff; 654 655 while (cl) { 656 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); 657 if (cl->level >= level) { 658 if (cl->level == level) 659 cl->xstats.lends++; 660 htb_accnt_tokens(cl, bytes, diff); 661 } else { 662 cl->xstats.borrows++; 663 cl->tokens += diff; /* we moved t_c; update tokens */ 664 } 665 htb_accnt_ctokens(cl, bytes, diff); 666 cl->t_c = q->now; 667 668 old_mode = cl->cmode; 669 diff = 0; 670 htb_change_class_mode(q, cl, &diff); 671 if (old_mode != cl->cmode) { 672 if (old_mode != HTB_CAN_SEND) 673 htb_safe_rb_erase(&cl->pq_node, &q->hlevel[cl->level].wait_pq); 674 if (cl->cmode != HTB_CAN_SEND) 675 htb_add_to_wait_tree(q, cl, diff); 676 } 677 678 /* update basic stats except for leaves which are already updated */ 679 if (cl->level) 680 bstats_update(&cl->bstats, skb); 681 682 cl = cl->parent; 683 } 684 } 685 686 /** 687 * htb_do_events - make mode changes to classes at the level 688 * 689 * Scans event queue for pending events and applies them. Returns time of 690 * next pending event (0 for no event in pq, q->now for too many events). 691 * Note: Applied are events whose have cl->pq_key <= q->now. 692 */ 693 static s64 htb_do_events(struct htb_sched *q, const int level, 694 unsigned long start) 695 { 696 /* don't run for longer than 2 jiffies; 2 is used instead of 697 * 1 to simplify things when jiffy is going to be incremented 698 * too soon 699 */ 700 unsigned long stop_at = start + 2; 701 struct rb_root *wait_pq = &q->hlevel[level].wait_pq; 702 703 while (time_before(jiffies, stop_at)) { 704 struct htb_class *cl; 705 s64 diff; 706 struct rb_node *p = rb_first(wait_pq); 707 708 if (!p) 709 return 0; 710 711 cl = rb_entry(p, struct htb_class, pq_node); 712 if (cl->pq_key > q->now) 713 return cl->pq_key; 714 715 htb_safe_rb_erase(p, wait_pq); 716 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); 717 htb_change_class_mode(q, cl, &diff); 718 if (cl->cmode != HTB_CAN_SEND) 719 htb_add_to_wait_tree(q, cl, diff); 720 } 721 722 /* too much load - let's continue after a break for scheduling */ 723 if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) { 724 pr_warn("htb: too many events!\n"); 725 q->warned |= HTB_WARN_TOOMANYEVENTS; 726 } 727 728 return q->now; 729 } 730 731 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL 732 * is no such one exists. 733 */ 734 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n, 735 u32 id) 736 { 737 struct rb_node *r = NULL; 738 while (n) { 739 struct htb_class *cl = 740 rb_entry(n, struct htb_class, node[prio]); 741 742 if (id > cl->common.classid) { 743 n = n->rb_right; 744 } else if (id < cl->common.classid) { 745 r = n; 746 n = n->rb_left; 747 } else { 748 return n; 749 } 750 } 751 return r; 752 } 753 754 /** 755 * htb_lookup_leaf - returns next leaf class in DRR order 756 * 757 * Find leaf where current feed pointers points to. 758 */ 759 static struct htb_class *htb_lookup_leaf(struct htb_prio *hprio, const int prio) 760 { 761 int i; 762 struct { 763 struct rb_node *root; 764 struct rb_node **pptr; 765 u32 *pid; 766 } stk[TC_HTB_MAXDEPTH], *sp = stk; 767 768 BUG_ON(!hprio->row.rb_node); 769 sp->root = hprio->row.rb_node; 770 sp->pptr = &hprio->ptr; 771 sp->pid = &hprio->last_ptr_id; 772 773 for (i = 0; i < 65535; i++) { 774 if (!*sp->pptr && *sp->pid) { 775 /* ptr was invalidated but id is valid - try to recover 776 * the original or next ptr 777 */ 778 *sp->pptr = 779 htb_id_find_next_upper(prio, sp->root, *sp->pid); 780 } 781 *sp->pid = 0; /* ptr is valid now so that remove this hint as it 782 * can become out of date quickly 783 */ 784 if (!*sp->pptr) { /* we are at right end; rewind & go up */ 785 *sp->pptr = sp->root; 786 while ((*sp->pptr)->rb_left) 787 *sp->pptr = (*sp->pptr)->rb_left; 788 if (sp > stk) { 789 sp--; 790 if (!*sp->pptr) { 791 WARN_ON(1); 792 return NULL; 793 } 794 htb_next_rb_node(sp->pptr); 795 } 796 } else { 797 struct htb_class *cl; 798 struct htb_prio *clp; 799 800 cl = rb_entry(*sp->pptr, struct htb_class, node[prio]); 801 if (!cl->level) 802 return cl; 803 clp = &cl->un.inner.clprio[prio]; 804 (++sp)->root = clp->feed.rb_node; 805 sp->pptr = &clp->ptr; 806 sp->pid = &clp->last_ptr_id; 807 } 808 } 809 WARN_ON(1); 810 return NULL; 811 } 812 813 /* dequeues packet at given priority and level; call only if 814 * you are sure that there is active class at prio/level 815 */ 816 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, const int prio, 817 const int level) 818 { 819 struct sk_buff *skb = NULL; 820 struct htb_class *cl, *start; 821 struct htb_level *hlevel = &q->hlevel[level]; 822 struct htb_prio *hprio = &hlevel->hprio[prio]; 823 824 /* look initial class up in the row */ 825 start = cl = htb_lookup_leaf(hprio, prio); 826 827 do { 828 next: 829 if (unlikely(!cl)) 830 return NULL; 831 832 /* class can be empty - it is unlikely but can be true if leaf 833 * qdisc drops packets in enqueue routine or if someone used 834 * graft operation on the leaf since last dequeue; 835 * simply deactivate and skip such class 836 */ 837 if (unlikely(cl->un.leaf.q->q.qlen == 0)) { 838 struct htb_class *next; 839 htb_deactivate(q, cl); 840 841 /* row/level might become empty */ 842 if ((q->row_mask[level] & (1 << prio)) == 0) 843 return NULL; 844 845 next = htb_lookup_leaf(hprio, prio); 846 847 if (cl == start) /* fix start if we just deleted it */ 848 start = next; 849 cl = next; 850 goto next; 851 } 852 853 skb = cl->un.leaf.q->dequeue(cl->un.leaf.q); 854 if (likely(skb != NULL)) 855 break; 856 857 qdisc_warn_nonwc("htb", cl->un.leaf.q); 858 htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr: 859 &q->hlevel[0].hprio[prio].ptr); 860 cl = htb_lookup_leaf(hprio, prio); 861 862 } while (cl != start); 863 864 if (likely(skb != NULL)) { 865 bstats_update(&cl->bstats, skb); 866 cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb); 867 if (cl->un.leaf.deficit[level] < 0) { 868 cl->un.leaf.deficit[level] += cl->quantum; 869 htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr : 870 &q->hlevel[0].hprio[prio].ptr); 871 } 872 /* this used to be after charge_class but this constelation 873 * gives us slightly better performance 874 */ 875 if (!cl->un.leaf.q->q.qlen) 876 htb_deactivate(q, cl); 877 htb_charge_class(q, cl, level, skb); 878 } 879 return skb; 880 } 881 882 static struct sk_buff *htb_dequeue(struct Qdisc *sch) 883 { 884 struct sk_buff *skb; 885 struct htb_sched *q = qdisc_priv(sch); 886 int level; 887 s64 next_event; 888 unsigned long start_at; 889 890 /* try to dequeue direct packets as high prio (!) to minimize cpu work */ 891 skb = __skb_dequeue(&q->direct_queue); 892 if (skb != NULL) { 893 ok: 894 qdisc_bstats_update(sch, skb); 895 qdisc_qstats_backlog_dec(sch, skb); 896 sch->q.qlen--; 897 return skb; 898 } 899 900 if (!sch->q.qlen) 901 goto fin; 902 q->now = ktime_get_ns(); 903 start_at = jiffies; 904 905 next_event = q->now + 5LLU * NSEC_PER_SEC; 906 907 for (level = 0; level < TC_HTB_MAXDEPTH; level++) { 908 /* common case optimization - skip event handler quickly */ 909 int m; 910 s64 event = q->near_ev_cache[level]; 911 912 if (q->now >= event) { 913 event = htb_do_events(q, level, start_at); 914 if (!event) 915 event = q->now + NSEC_PER_SEC; 916 q->near_ev_cache[level] = event; 917 } 918 919 if (next_event > event) 920 next_event = event; 921 922 m = ~q->row_mask[level]; 923 while (m != (int)(-1)) { 924 int prio = ffz(m); 925 926 m |= 1 << prio; 927 skb = htb_dequeue_tree(q, prio, level); 928 if (likely(skb != NULL)) 929 goto ok; 930 } 931 } 932 qdisc_qstats_overlimit(sch); 933 if (likely(next_event > q->now)) 934 qdisc_watchdog_schedule_ns(&q->watchdog, next_event); 935 else 936 schedule_work(&q->work); 937 fin: 938 return skb; 939 } 940 941 /* reset all classes */ 942 /* always caled under BH & queue lock */ 943 static void htb_reset(struct Qdisc *sch) 944 { 945 struct htb_sched *q = qdisc_priv(sch); 946 struct htb_class *cl; 947 unsigned int i; 948 949 for (i = 0; i < q->clhash.hashsize; i++) { 950 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 951 if (cl->level) 952 memset(&cl->un.inner, 0, sizeof(cl->un.inner)); 953 else { 954 if (cl->un.leaf.q) 955 qdisc_reset(cl->un.leaf.q); 956 INIT_LIST_HEAD(&cl->un.leaf.drop_list); 957 } 958 cl->prio_activity = 0; 959 cl->cmode = HTB_CAN_SEND; 960 } 961 } 962 qdisc_watchdog_cancel(&q->watchdog); 963 __qdisc_reset_queue(&q->direct_queue); 964 sch->q.qlen = 0; 965 sch->qstats.backlog = 0; 966 memset(q->hlevel, 0, sizeof(q->hlevel)); 967 memset(q->row_mask, 0, sizeof(q->row_mask)); 968 for (i = 0; i < TC_HTB_NUMPRIO; i++) 969 INIT_LIST_HEAD(q->drops + i); 970 } 971 972 static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = { 973 [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) }, 974 [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) }, 975 [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, 976 [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, 977 [TCA_HTB_DIRECT_QLEN] = { .type = NLA_U32 }, 978 [TCA_HTB_RATE64] = { .type = NLA_U64 }, 979 [TCA_HTB_CEIL64] = { .type = NLA_U64 }, 980 }; 981 982 static void htb_work_func(struct work_struct *work) 983 { 984 struct htb_sched *q = container_of(work, struct htb_sched, work); 985 struct Qdisc *sch = q->watchdog.qdisc; 986 987 rcu_read_lock(); 988 __netif_schedule(qdisc_root(sch)); 989 rcu_read_unlock(); 990 } 991 992 static int htb_init(struct Qdisc *sch, struct nlattr *opt) 993 { 994 struct htb_sched *q = qdisc_priv(sch); 995 struct nlattr *tb[TCA_HTB_MAX + 1]; 996 struct tc_htb_glob *gopt; 997 int err; 998 int i; 999 1000 if (!opt) 1001 return -EINVAL; 1002 1003 err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy); 1004 if (err < 0) 1005 return err; 1006 1007 if (!tb[TCA_HTB_INIT]) 1008 return -EINVAL; 1009 1010 gopt = nla_data(tb[TCA_HTB_INIT]); 1011 if (gopt->version != HTB_VER >> 16) 1012 return -EINVAL; 1013 1014 err = qdisc_class_hash_init(&q->clhash); 1015 if (err < 0) 1016 return err; 1017 for (i = 0; i < TC_HTB_NUMPRIO; i++) 1018 INIT_LIST_HEAD(q->drops + i); 1019 1020 qdisc_watchdog_init(&q->watchdog, sch); 1021 INIT_WORK(&q->work, htb_work_func); 1022 __skb_queue_head_init(&q->direct_queue); 1023 1024 if (tb[TCA_HTB_DIRECT_QLEN]) 1025 q->direct_qlen = nla_get_u32(tb[TCA_HTB_DIRECT_QLEN]); 1026 else 1027 q->direct_qlen = qdisc_dev(sch)->tx_queue_len; 1028 1029 if ((q->rate2quantum = gopt->rate2quantum) < 1) 1030 q->rate2quantum = 1; 1031 q->defcls = gopt->defcls; 1032 1033 return 0; 1034 } 1035 1036 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) 1037 { 1038 struct htb_sched *q = qdisc_priv(sch); 1039 struct nlattr *nest; 1040 struct tc_htb_glob gopt; 1041 1042 /* Its safe to not acquire qdisc lock. As we hold RTNL, 1043 * no change can happen on the qdisc parameters. 1044 */ 1045 1046 gopt.direct_pkts = q->direct_pkts; 1047 gopt.version = HTB_VER; 1048 gopt.rate2quantum = q->rate2quantum; 1049 gopt.defcls = q->defcls; 1050 gopt.debug = 0; 1051 1052 nest = nla_nest_start(skb, TCA_OPTIONS); 1053 if (nest == NULL) 1054 goto nla_put_failure; 1055 if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt) || 1056 nla_put_u32(skb, TCA_HTB_DIRECT_QLEN, q->direct_qlen)) 1057 goto nla_put_failure; 1058 1059 return nla_nest_end(skb, nest); 1060 1061 nla_put_failure: 1062 nla_nest_cancel(skb, nest); 1063 return -1; 1064 } 1065 1066 static int htb_dump_class(struct Qdisc *sch, unsigned long arg, 1067 struct sk_buff *skb, struct tcmsg *tcm) 1068 { 1069 struct htb_class *cl = (struct htb_class *)arg; 1070 struct nlattr *nest; 1071 struct tc_htb_opt opt; 1072 1073 /* Its safe to not acquire qdisc lock. As we hold RTNL, 1074 * no change can happen on the class parameters. 1075 */ 1076 tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT; 1077 tcm->tcm_handle = cl->common.classid; 1078 if (!cl->level && cl->un.leaf.q) 1079 tcm->tcm_info = cl->un.leaf.q->handle; 1080 1081 nest = nla_nest_start(skb, TCA_OPTIONS); 1082 if (nest == NULL) 1083 goto nla_put_failure; 1084 1085 memset(&opt, 0, sizeof(opt)); 1086 1087 psched_ratecfg_getrate(&opt.rate, &cl->rate); 1088 opt.buffer = PSCHED_NS2TICKS(cl->buffer); 1089 psched_ratecfg_getrate(&opt.ceil, &cl->ceil); 1090 opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer); 1091 opt.quantum = cl->quantum; 1092 opt.prio = cl->prio; 1093 opt.level = cl->level; 1094 if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt)) 1095 goto nla_put_failure; 1096 if ((cl->rate.rate_bytes_ps >= (1ULL << 32)) && 1097 nla_put_u64_64bit(skb, TCA_HTB_RATE64, cl->rate.rate_bytes_ps, 1098 TCA_HTB_PAD)) 1099 goto nla_put_failure; 1100 if ((cl->ceil.rate_bytes_ps >= (1ULL << 32)) && 1101 nla_put_u64_64bit(skb, TCA_HTB_CEIL64, cl->ceil.rate_bytes_ps, 1102 TCA_HTB_PAD)) 1103 goto nla_put_failure; 1104 1105 return nla_nest_end(skb, nest); 1106 1107 nla_put_failure: 1108 nla_nest_cancel(skb, nest); 1109 return -1; 1110 } 1111 1112 static int 1113 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d) 1114 { 1115 struct htb_class *cl = (struct htb_class *)arg; 1116 struct gnet_stats_queue qs = { 1117 .drops = cl->drops, 1118 }; 1119 __u32 qlen = 0; 1120 1121 if (!cl->level && cl->un.leaf.q) { 1122 qlen = cl->un.leaf.q->q.qlen; 1123 qs.backlog = cl->un.leaf.q->qstats.backlog; 1124 } 1125 cl->xstats.tokens = clamp_t(s64, PSCHED_NS2TICKS(cl->tokens), 1126 INT_MIN, INT_MAX); 1127 cl->xstats.ctokens = clamp_t(s64, PSCHED_NS2TICKS(cl->ctokens), 1128 INT_MIN, INT_MAX); 1129 1130 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), 1131 d, NULL, &cl->bstats) < 0 || 1132 gnet_stats_copy_rate_est(d, NULL, &cl->rate_est) < 0 || 1133 gnet_stats_copy_queue(d, NULL, &qs, qlen) < 0) 1134 return -1; 1135 1136 return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); 1137 } 1138 1139 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 1140 struct Qdisc **old) 1141 { 1142 struct htb_class *cl = (struct htb_class *)arg; 1143 1144 if (cl->level) 1145 return -EINVAL; 1146 if (new == NULL && 1147 (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 1148 cl->common.classid)) == NULL) 1149 return -ENOBUFS; 1150 1151 *old = qdisc_replace(sch, new, &cl->un.leaf.q); 1152 return 0; 1153 } 1154 1155 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg) 1156 { 1157 struct htb_class *cl = (struct htb_class *)arg; 1158 return !cl->level ? cl->un.leaf.q : NULL; 1159 } 1160 1161 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg) 1162 { 1163 struct htb_class *cl = (struct htb_class *)arg; 1164 1165 if (cl->un.leaf.q->q.qlen == 0) 1166 htb_deactivate(qdisc_priv(sch), cl); 1167 } 1168 1169 static unsigned long htb_get(struct Qdisc *sch, u32 classid) 1170 { 1171 struct htb_class *cl = htb_find(classid, sch); 1172 if (cl) 1173 cl->refcnt++; 1174 return (unsigned long)cl; 1175 } 1176 1177 static inline int htb_parent_last_child(struct htb_class *cl) 1178 { 1179 if (!cl->parent) 1180 /* the root class */ 1181 return 0; 1182 if (cl->parent->children > 1) 1183 /* not the last child */ 1184 return 0; 1185 return 1; 1186 } 1187 1188 static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl, 1189 struct Qdisc *new_q) 1190 { 1191 struct htb_class *parent = cl->parent; 1192 1193 WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity); 1194 1195 if (parent->cmode != HTB_CAN_SEND) 1196 htb_safe_rb_erase(&parent->pq_node, 1197 &q->hlevel[parent->level].wait_pq); 1198 1199 parent->level = 0; 1200 memset(&parent->un.inner, 0, sizeof(parent->un.inner)); 1201 INIT_LIST_HEAD(&parent->un.leaf.drop_list); 1202 parent->un.leaf.q = new_q ? new_q : &noop_qdisc; 1203 parent->tokens = parent->buffer; 1204 parent->ctokens = parent->cbuffer; 1205 parent->t_c = ktime_get_ns(); 1206 parent->cmode = HTB_CAN_SEND; 1207 } 1208 1209 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl) 1210 { 1211 if (!cl->level) { 1212 WARN_ON(!cl->un.leaf.q); 1213 qdisc_destroy(cl->un.leaf.q); 1214 } 1215 gen_kill_estimator(&cl->bstats, &cl->rate_est); 1216 tcf_destroy_chain(&cl->filter_list); 1217 kfree(cl); 1218 } 1219 1220 static void htb_destroy(struct Qdisc *sch) 1221 { 1222 struct htb_sched *q = qdisc_priv(sch); 1223 struct hlist_node *next; 1224 struct htb_class *cl; 1225 unsigned int i; 1226 1227 cancel_work_sync(&q->work); 1228 qdisc_watchdog_cancel(&q->watchdog); 1229 /* This line used to be after htb_destroy_class call below 1230 * and surprisingly it worked in 2.4. But it must precede it 1231 * because filter need its target class alive to be able to call 1232 * unbind_filter on it (without Oops). 1233 */ 1234 tcf_destroy_chain(&q->filter_list); 1235 1236 for (i = 0; i < q->clhash.hashsize; i++) { 1237 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) 1238 tcf_destroy_chain(&cl->filter_list); 1239 } 1240 for (i = 0; i < q->clhash.hashsize; i++) { 1241 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i], 1242 common.hnode) 1243 htb_destroy_class(sch, cl); 1244 } 1245 qdisc_class_hash_destroy(&q->clhash); 1246 __qdisc_reset_queue(&q->direct_queue); 1247 } 1248 1249 static int htb_delete(struct Qdisc *sch, unsigned long arg) 1250 { 1251 struct htb_sched *q = qdisc_priv(sch); 1252 struct htb_class *cl = (struct htb_class *)arg; 1253 struct Qdisc *new_q = NULL; 1254 int last_child = 0; 1255 1256 /* TODO: why don't allow to delete subtree ? references ? does 1257 * tc subsys guarantee us that in htb_destroy it holds no class 1258 * refs so that we can remove children safely there ? 1259 */ 1260 if (cl->children || cl->filter_cnt) 1261 return -EBUSY; 1262 1263 if (!cl->level && htb_parent_last_child(cl)) { 1264 new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 1265 cl->parent->common.classid); 1266 last_child = 1; 1267 } 1268 1269 sch_tree_lock(sch); 1270 1271 if (!cl->level) { 1272 unsigned int qlen = cl->un.leaf.q->q.qlen; 1273 unsigned int backlog = cl->un.leaf.q->qstats.backlog; 1274 1275 qdisc_reset(cl->un.leaf.q); 1276 qdisc_tree_reduce_backlog(cl->un.leaf.q, qlen, backlog); 1277 } 1278 1279 /* delete from hash and active; remainder in destroy_class */ 1280 qdisc_class_hash_remove(&q->clhash, &cl->common); 1281 if (cl->parent) 1282 cl->parent->children--; 1283 1284 if (cl->prio_activity) 1285 htb_deactivate(q, cl); 1286 1287 if (cl->cmode != HTB_CAN_SEND) 1288 htb_safe_rb_erase(&cl->pq_node, 1289 &q->hlevel[cl->level].wait_pq); 1290 1291 if (last_child) 1292 htb_parent_to_leaf(q, cl, new_q); 1293 1294 BUG_ON(--cl->refcnt == 0); 1295 /* 1296 * This shouldn't happen: we "hold" one cops->get() when called 1297 * from tc_ctl_tclass; the destroy method is done from cops->put(). 1298 */ 1299 1300 sch_tree_unlock(sch); 1301 return 0; 1302 } 1303 1304 static void htb_put(struct Qdisc *sch, unsigned long arg) 1305 { 1306 struct htb_class *cl = (struct htb_class *)arg; 1307 1308 if (--cl->refcnt == 0) 1309 htb_destroy_class(sch, cl); 1310 } 1311 1312 static int htb_change_class(struct Qdisc *sch, u32 classid, 1313 u32 parentid, struct nlattr **tca, 1314 unsigned long *arg) 1315 { 1316 int err = -EINVAL; 1317 struct htb_sched *q = qdisc_priv(sch); 1318 struct htb_class *cl = (struct htb_class *)*arg, *parent; 1319 struct nlattr *opt = tca[TCA_OPTIONS]; 1320 struct nlattr *tb[TCA_HTB_MAX + 1]; 1321 struct tc_htb_opt *hopt; 1322 u64 rate64, ceil64; 1323 1324 /* extract all subattrs from opt attr */ 1325 if (!opt) 1326 goto failure; 1327 1328 err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy); 1329 if (err < 0) 1330 goto failure; 1331 1332 err = -EINVAL; 1333 if (tb[TCA_HTB_PARMS] == NULL) 1334 goto failure; 1335 1336 parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch); 1337 1338 hopt = nla_data(tb[TCA_HTB_PARMS]); 1339 if (!hopt->rate.rate || !hopt->ceil.rate) 1340 goto failure; 1341 1342 /* Keeping backward compatible with rate_table based iproute2 tc */ 1343 if (hopt->rate.linklayer == TC_LINKLAYER_UNAWARE) 1344 qdisc_put_rtab(qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB])); 1345 1346 if (hopt->ceil.linklayer == TC_LINKLAYER_UNAWARE) 1347 qdisc_put_rtab(qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB])); 1348 1349 if (!cl) { /* new class */ 1350 struct Qdisc *new_q; 1351 int prio; 1352 struct { 1353 struct nlattr nla; 1354 struct gnet_estimator opt; 1355 } est = { 1356 .nla = { 1357 .nla_len = nla_attr_size(sizeof(est.opt)), 1358 .nla_type = TCA_RATE, 1359 }, 1360 .opt = { 1361 /* 4s interval, 16s averaging constant */ 1362 .interval = 2, 1363 .ewma_log = 2, 1364 }, 1365 }; 1366 1367 /* check for valid classid */ 1368 if (!classid || TC_H_MAJ(classid ^ sch->handle) || 1369 htb_find(classid, sch)) 1370 goto failure; 1371 1372 /* check maximal depth */ 1373 if (parent && parent->parent && parent->parent->level < 2) { 1374 pr_err("htb: tree is too deep\n"); 1375 goto failure; 1376 } 1377 err = -ENOBUFS; 1378 cl = kzalloc(sizeof(*cl), GFP_KERNEL); 1379 if (!cl) 1380 goto failure; 1381 1382 if (htb_rate_est || tca[TCA_RATE]) { 1383 err = gen_new_estimator(&cl->bstats, NULL, 1384 &cl->rate_est, 1385 NULL, 1386 qdisc_root_sleeping_running(sch), 1387 tca[TCA_RATE] ? : &est.nla); 1388 if (err) { 1389 kfree(cl); 1390 goto failure; 1391 } 1392 } 1393 1394 cl->refcnt = 1; 1395 cl->children = 0; 1396 INIT_LIST_HEAD(&cl->un.leaf.drop_list); 1397 RB_CLEAR_NODE(&cl->pq_node); 1398 1399 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) 1400 RB_CLEAR_NODE(&cl->node[prio]); 1401 1402 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL) 1403 * so that can't be used inside of sch_tree_lock 1404 * -- thanks to Karlis Peisenieks 1405 */ 1406 new_q = qdisc_create_dflt(sch->dev_queue, 1407 &pfifo_qdisc_ops, classid); 1408 sch_tree_lock(sch); 1409 if (parent && !parent->level) { 1410 unsigned int qlen = parent->un.leaf.q->q.qlen; 1411 unsigned int backlog = parent->un.leaf.q->qstats.backlog; 1412 1413 /* turn parent into inner node */ 1414 qdisc_reset(parent->un.leaf.q); 1415 qdisc_tree_reduce_backlog(parent->un.leaf.q, qlen, backlog); 1416 qdisc_destroy(parent->un.leaf.q); 1417 if (parent->prio_activity) 1418 htb_deactivate(q, parent); 1419 1420 /* remove from evt list because of level change */ 1421 if (parent->cmode != HTB_CAN_SEND) { 1422 htb_safe_rb_erase(&parent->pq_node, &q->hlevel[0].wait_pq); 1423 parent->cmode = HTB_CAN_SEND; 1424 } 1425 parent->level = (parent->parent ? parent->parent->level 1426 : TC_HTB_MAXDEPTH) - 1; 1427 memset(&parent->un.inner, 0, sizeof(parent->un.inner)); 1428 } 1429 /* leaf (we) needs elementary qdisc */ 1430 cl->un.leaf.q = new_q ? new_q : &noop_qdisc; 1431 1432 cl->common.classid = classid; 1433 cl->parent = parent; 1434 1435 /* set class to be in HTB_CAN_SEND state */ 1436 cl->tokens = PSCHED_TICKS2NS(hopt->buffer); 1437 cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer); 1438 cl->mbuffer = 60ULL * NSEC_PER_SEC; /* 1min */ 1439 cl->t_c = ktime_get_ns(); 1440 cl->cmode = HTB_CAN_SEND; 1441 1442 /* attach to the hash list and parent's family */ 1443 qdisc_class_hash_insert(&q->clhash, &cl->common); 1444 if (parent) 1445 parent->children++; 1446 } else { 1447 if (tca[TCA_RATE]) { 1448 err = gen_replace_estimator(&cl->bstats, NULL, 1449 &cl->rate_est, 1450 NULL, 1451 qdisc_root_sleeping_running(sch), 1452 tca[TCA_RATE]); 1453 if (err) 1454 return err; 1455 } 1456 sch_tree_lock(sch); 1457 } 1458 1459 rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0; 1460 1461 ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0; 1462 1463 psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64); 1464 psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64); 1465 1466 /* it used to be a nasty bug here, we have to check that node 1467 * is really leaf before changing cl->un.leaf ! 1468 */ 1469 if (!cl->level) { 1470 u64 quantum = cl->rate.rate_bytes_ps; 1471 1472 do_div(quantum, q->rate2quantum); 1473 cl->quantum = min_t(u64, quantum, INT_MAX); 1474 1475 if (!hopt->quantum && cl->quantum < 1000) { 1476 pr_warn("HTB: quantum of class %X is small. Consider r2q change.\n", 1477 cl->common.classid); 1478 cl->quantum = 1000; 1479 } 1480 if (!hopt->quantum && cl->quantum > 200000) { 1481 pr_warn("HTB: quantum of class %X is big. Consider r2q change.\n", 1482 cl->common.classid); 1483 cl->quantum = 200000; 1484 } 1485 if (hopt->quantum) 1486 cl->quantum = hopt->quantum; 1487 if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO) 1488 cl->prio = TC_HTB_NUMPRIO - 1; 1489 } 1490 1491 cl->buffer = PSCHED_TICKS2NS(hopt->buffer); 1492 cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer); 1493 1494 sch_tree_unlock(sch); 1495 1496 qdisc_class_hash_grow(sch, &q->clhash); 1497 1498 *arg = (unsigned long)cl; 1499 return 0; 1500 1501 failure: 1502 return err; 1503 } 1504 1505 static struct tcf_proto __rcu **htb_find_tcf(struct Qdisc *sch, 1506 unsigned long arg) 1507 { 1508 struct htb_sched *q = qdisc_priv(sch); 1509 struct htb_class *cl = (struct htb_class *)arg; 1510 struct tcf_proto __rcu **fl = cl ? &cl->filter_list : &q->filter_list; 1511 1512 return fl; 1513 } 1514 1515 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, 1516 u32 classid) 1517 { 1518 struct htb_class *cl = htb_find(classid, sch); 1519 1520 /*if (cl && !cl->level) return 0; 1521 * The line above used to be there to prevent attaching filters to 1522 * leaves. But at least tc_index filter uses this just to get class 1523 * for other reasons so that we have to allow for it. 1524 * ---- 1525 * 19.6.2002 As Werner explained it is ok - bind filter is just 1526 * another way to "lock" the class - unlike "get" this lock can 1527 * be broken by class during destroy IIUC. 1528 */ 1529 if (cl) 1530 cl->filter_cnt++; 1531 return (unsigned long)cl; 1532 } 1533 1534 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) 1535 { 1536 struct htb_class *cl = (struct htb_class *)arg; 1537 1538 if (cl) 1539 cl->filter_cnt--; 1540 } 1541 1542 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) 1543 { 1544 struct htb_sched *q = qdisc_priv(sch); 1545 struct htb_class *cl; 1546 unsigned int i; 1547 1548 if (arg->stop) 1549 return; 1550 1551 for (i = 0; i < q->clhash.hashsize; i++) { 1552 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 1553 if (arg->count < arg->skip) { 1554 arg->count++; 1555 continue; 1556 } 1557 if (arg->fn(sch, (unsigned long)cl, arg) < 0) { 1558 arg->stop = 1; 1559 return; 1560 } 1561 arg->count++; 1562 } 1563 } 1564 } 1565 1566 static const struct Qdisc_class_ops htb_class_ops = { 1567 .graft = htb_graft, 1568 .leaf = htb_leaf, 1569 .qlen_notify = htb_qlen_notify, 1570 .get = htb_get, 1571 .put = htb_put, 1572 .change = htb_change_class, 1573 .delete = htb_delete, 1574 .walk = htb_walk, 1575 .tcf_chain = htb_find_tcf, 1576 .bind_tcf = htb_bind_filter, 1577 .unbind_tcf = htb_unbind_filter, 1578 .dump = htb_dump_class, 1579 .dump_stats = htb_dump_class_stats, 1580 }; 1581 1582 static struct Qdisc_ops htb_qdisc_ops __read_mostly = { 1583 .cl_ops = &htb_class_ops, 1584 .id = "htb", 1585 .priv_size = sizeof(struct htb_sched), 1586 .enqueue = htb_enqueue, 1587 .dequeue = htb_dequeue, 1588 .peek = qdisc_peek_dequeued, 1589 .init = htb_init, 1590 .reset = htb_reset, 1591 .destroy = htb_destroy, 1592 .dump = htb_dump, 1593 .owner = THIS_MODULE, 1594 }; 1595 1596 static int __init htb_module_init(void) 1597 { 1598 return register_qdisc(&htb_qdisc_ops); 1599 } 1600 static void __exit htb_module_exit(void) 1601 { 1602 unregister_qdisc(&htb_qdisc_ops); 1603 } 1604 1605 module_init(htb_module_init) 1606 module_exit(htb_module_exit) 1607 MODULE_LICENSE("GPL"); 1608