1 /* 2 * net/sched/sch_api.c Packet scheduler API. 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: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 * 11 * Fixes: 12 * 13 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired. 14 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support 15 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support 16 */ 17 18 #include <linux/module.h> 19 #include <linux/types.h> 20 #include <linux/kernel.h> 21 #include <linux/string.h> 22 #include <linux/errno.h> 23 #include <linux/skbuff.h> 24 #include <linux/init.h> 25 #include <linux/proc_fs.h> 26 #include <linux/seq_file.h> 27 #include <linux/kmod.h> 28 #include <linux/list.h> 29 #include <linux/hrtimer.h> 30 #include <linux/lockdep.h> 31 #include <linux/slab.h> 32 #include <linux/hashtable.h> 33 34 #include <net/net_namespace.h> 35 #include <net/sock.h> 36 #include <net/netlink.h> 37 #include <net/pkt_sched.h> 38 39 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 40 struct nlmsghdr *n, u32 clid, 41 struct Qdisc *old, struct Qdisc *new); 42 static int tclass_notify(struct net *net, struct sk_buff *oskb, 43 struct nlmsghdr *n, struct Qdisc *q, 44 unsigned long cl, int event); 45 46 /* 47 48 Short review. 49 ------------- 50 51 This file consists of two interrelated parts: 52 53 1. queueing disciplines manager frontend. 54 2. traffic classes manager frontend. 55 56 Generally, queueing discipline ("qdisc") is a black box, 57 which is able to enqueue packets and to dequeue them (when 58 device is ready to send something) in order and at times 59 determined by algorithm hidden in it. 60 61 qdisc's are divided to two categories: 62 - "queues", which have no internal structure visible from outside. 63 - "schedulers", which split all the packets to "traffic classes", 64 using "packet classifiers" (look at cls_api.c) 65 66 In turn, classes may have child qdiscs (as rule, queues) 67 attached to them etc. etc. etc. 68 69 The goal of the routines in this file is to translate 70 information supplied by user in the form of handles 71 to more intelligible for kernel form, to make some sanity 72 checks and part of work, which is common to all qdiscs 73 and to provide rtnetlink notifications. 74 75 All real intelligent work is done inside qdisc modules. 76 77 78 79 Every discipline has two major routines: enqueue and dequeue. 80 81 ---dequeue 82 83 dequeue usually returns a skb to send. It is allowed to return NULL, 84 but it does not mean that queue is empty, it just means that 85 discipline does not want to send anything this time. 86 Queue is really empty if q->q.qlen == 0. 87 For complicated disciplines with multiple queues q->q is not 88 real packet queue, but however q->q.qlen must be valid. 89 90 ---enqueue 91 92 enqueue returns 0, if packet was enqueued successfully. 93 If packet (this one or another one) was dropped, it returns 94 not zero error code. 95 NET_XMIT_DROP - this packet dropped 96 Expected action: do not backoff, but wait until queue will clear. 97 NET_XMIT_CN - probably this packet enqueued, but another one dropped. 98 Expected action: backoff or ignore 99 100 Auxiliary routines: 101 102 ---peek 103 104 like dequeue but without removing a packet from the queue 105 106 ---reset 107 108 returns qdisc to initial state: purge all buffers, clear all 109 timers, counters (except for statistics) etc. 110 111 ---init 112 113 initializes newly created qdisc. 114 115 ---destroy 116 117 destroys resources allocated by init and during lifetime of qdisc. 118 119 ---change 120 121 changes qdisc parameters. 122 */ 123 124 /* Protects list of registered TC modules. It is pure SMP lock. */ 125 static DEFINE_RWLOCK(qdisc_mod_lock); 126 127 128 /************************************************ 129 * Queueing disciplines manipulation. * 130 ************************************************/ 131 132 133 /* The list of all installed queueing disciplines. */ 134 135 static struct Qdisc_ops *qdisc_base; 136 137 /* Register/unregister queueing discipline */ 138 139 int register_qdisc(struct Qdisc_ops *qops) 140 { 141 struct Qdisc_ops *q, **qp; 142 int rc = -EEXIST; 143 144 write_lock(&qdisc_mod_lock); 145 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 146 if (!strcmp(qops->id, q->id)) 147 goto out; 148 149 if (qops->enqueue == NULL) 150 qops->enqueue = noop_qdisc_ops.enqueue; 151 if (qops->peek == NULL) { 152 if (qops->dequeue == NULL) 153 qops->peek = noop_qdisc_ops.peek; 154 else 155 goto out_einval; 156 } 157 if (qops->dequeue == NULL) 158 qops->dequeue = noop_qdisc_ops.dequeue; 159 160 if (qops->cl_ops) { 161 const struct Qdisc_class_ops *cops = qops->cl_ops; 162 163 if (!(cops->get && cops->put && cops->walk && cops->leaf)) 164 goto out_einval; 165 166 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf)) 167 goto out_einval; 168 } 169 170 qops->next = NULL; 171 *qp = qops; 172 rc = 0; 173 out: 174 write_unlock(&qdisc_mod_lock); 175 return rc; 176 177 out_einval: 178 rc = -EINVAL; 179 goto out; 180 } 181 EXPORT_SYMBOL(register_qdisc); 182 183 int unregister_qdisc(struct Qdisc_ops *qops) 184 { 185 struct Qdisc_ops *q, **qp; 186 int err = -ENOENT; 187 188 write_lock(&qdisc_mod_lock); 189 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 190 if (q == qops) 191 break; 192 if (q) { 193 *qp = q->next; 194 q->next = NULL; 195 err = 0; 196 } 197 write_unlock(&qdisc_mod_lock); 198 return err; 199 } 200 EXPORT_SYMBOL(unregister_qdisc); 201 202 /* Get default qdisc if not otherwise specified */ 203 void qdisc_get_default(char *name, size_t len) 204 { 205 read_lock(&qdisc_mod_lock); 206 strlcpy(name, default_qdisc_ops->id, len); 207 read_unlock(&qdisc_mod_lock); 208 } 209 210 static struct Qdisc_ops *qdisc_lookup_default(const char *name) 211 { 212 struct Qdisc_ops *q = NULL; 213 214 for (q = qdisc_base; q; q = q->next) { 215 if (!strcmp(name, q->id)) { 216 if (!try_module_get(q->owner)) 217 q = NULL; 218 break; 219 } 220 } 221 222 return q; 223 } 224 225 /* Set new default qdisc to use */ 226 int qdisc_set_default(const char *name) 227 { 228 const struct Qdisc_ops *ops; 229 230 if (!capable(CAP_NET_ADMIN)) 231 return -EPERM; 232 233 write_lock(&qdisc_mod_lock); 234 ops = qdisc_lookup_default(name); 235 if (!ops) { 236 /* Not found, drop lock and try to load module */ 237 write_unlock(&qdisc_mod_lock); 238 request_module("sch_%s", name); 239 write_lock(&qdisc_mod_lock); 240 241 ops = qdisc_lookup_default(name); 242 } 243 244 if (ops) { 245 /* Set new default */ 246 module_put(default_qdisc_ops->owner); 247 default_qdisc_ops = ops; 248 } 249 write_unlock(&qdisc_mod_lock); 250 251 return ops ? 0 : -ENOENT; 252 } 253 254 /* We know handle. Find qdisc among all qdisc's attached to device 255 * (root qdisc, all its children, children of children etc.) 256 * Note: caller either uses rtnl or rcu_read_lock() 257 */ 258 259 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle) 260 { 261 struct Qdisc *q; 262 263 if (!qdisc_dev(root)) 264 return (root->handle == handle ? root : NULL); 265 266 if (!(root->flags & TCQ_F_BUILTIN) && 267 root->handle == handle) 268 return root; 269 270 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) { 271 if (q->handle == handle) 272 return q; 273 } 274 return NULL; 275 } 276 277 void qdisc_hash_add(struct Qdisc *q) 278 { 279 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 280 struct Qdisc *root = qdisc_dev(q)->qdisc; 281 282 WARN_ON_ONCE(root == &noop_qdisc); 283 ASSERT_RTNL(); 284 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle); 285 } 286 } 287 EXPORT_SYMBOL(qdisc_hash_add); 288 289 void qdisc_hash_del(struct Qdisc *q) 290 { 291 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 292 ASSERT_RTNL(); 293 hash_del_rcu(&q->hash); 294 } 295 } 296 EXPORT_SYMBOL(qdisc_hash_del); 297 298 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) 299 { 300 struct Qdisc *q; 301 302 q = qdisc_match_from_root(dev->qdisc, handle); 303 if (q) 304 goto out; 305 306 if (dev_ingress_queue(dev)) 307 q = qdisc_match_from_root( 308 dev_ingress_queue(dev)->qdisc_sleeping, 309 handle); 310 out: 311 return q; 312 } 313 314 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) 315 { 316 unsigned long cl; 317 struct Qdisc *leaf; 318 const struct Qdisc_class_ops *cops = p->ops->cl_ops; 319 320 if (cops == NULL) 321 return NULL; 322 cl = cops->get(p, classid); 323 324 if (cl == 0) 325 return NULL; 326 leaf = cops->leaf(p, cl); 327 cops->put(p, cl); 328 return leaf; 329 } 330 331 /* Find queueing discipline by name */ 332 333 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) 334 { 335 struct Qdisc_ops *q = NULL; 336 337 if (kind) { 338 read_lock(&qdisc_mod_lock); 339 for (q = qdisc_base; q; q = q->next) { 340 if (nla_strcmp(kind, q->id) == 0) { 341 if (!try_module_get(q->owner)) 342 q = NULL; 343 break; 344 } 345 } 346 read_unlock(&qdisc_mod_lock); 347 } 348 return q; 349 } 350 351 /* The linklayer setting were not transferred from iproute2, in older 352 * versions, and the rate tables lookup systems have been dropped in 353 * the kernel. To keep backward compatible with older iproute2 tc 354 * utils, we detect the linklayer setting by detecting if the rate 355 * table were modified. 356 * 357 * For linklayer ATM table entries, the rate table will be aligned to 358 * 48 bytes, thus some table entries will contain the same value. The 359 * mpu (min packet unit) is also encoded into the old rate table, thus 360 * starting from the mpu, we find low and high table entries for 361 * mapping this cell. If these entries contain the same value, when 362 * the rate tables have been modified for linklayer ATM. 363 * 364 * This is done by rounding mpu to the nearest 48 bytes cell/entry, 365 * and then roundup to the next cell, calc the table entry one below, 366 * and compare. 367 */ 368 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab) 369 { 370 int low = roundup(r->mpu, 48); 371 int high = roundup(low+1, 48); 372 int cell_low = low >> r->cell_log; 373 int cell_high = (high >> r->cell_log) - 1; 374 375 /* rtab is too inaccurate at rates > 100Mbit/s */ 376 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) { 377 pr_debug("TC linklayer: Giving up ATM detection\n"); 378 return TC_LINKLAYER_ETHERNET; 379 } 380 381 if ((cell_high > cell_low) && (cell_high < 256) 382 && (rtab[cell_low] == rtab[cell_high])) { 383 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n", 384 cell_low, cell_high, rtab[cell_high]); 385 return TC_LINKLAYER_ATM; 386 } 387 return TC_LINKLAYER_ETHERNET; 388 } 389 390 static struct qdisc_rate_table *qdisc_rtab_list; 391 392 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, 393 struct nlattr *tab) 394 { 395 struct qdisc_rate_table *rtab; 396 397 if (tab == NULL || r->rate == 0 || r->cell_log == 0 || 398 nla_len(tab) != TC_RTAB_SIZE) 399 return NULL; 400 401 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { 402 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) && 403 !memcmp(&rtab->data, nla_data(tab), 1024)) { 404 rtab->refcnt++; 405 return rtab; 406 } 407 } 408 409 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); 410 if (rtab) { 411 rtab->rate = *r; 412 rtab->refcnt = 1; 413 memcpy(rtab->data, nla_data(tab), 1024); 414 if (r->linklayer == TC_LINKLAYER_UNAWARE) 415 r->linklayer = __detect_linklayer(r, rtab->data); 416 rtab->next = qdisc_rtab_list; 417 qdisc_rtab_list = rtab; 418 } 419 return rtab; 420 } 421 EXPORT_SYMBOL(qdisc_get_rtab); 422 423 void qdisc_put_rtab(struct qdisc_rate_table *tab) 424 { 425 struct qdisc_rate_table *rtab, **rtabp; 426 427 if (!tab || --tab->refcnt) 428 return; 429 430 for (rtabp = &qdisc_rtab_list; 431 (rtab = *rtabp) != NULL; 432 rtabp = &rtab->next) { 433 if (rtab == tab) { 434 *rtabp = rtab->next; 435 kfree(rtab); 436 return; 437 } 438 } 439 } 440 EXPORT_SYMBOL(qdisc_put_rtab); 441 442 static LIST_HEAD(qdisc_stab_list); 443 444 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { 445 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, 446 [TCA_STAB_DATA] = { .type = NLA_BINARY }, 447 }; 448 449 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) 450 { 451 struct nlattr *tb[TCA_STAB_MAX + 1]; 452 struct qdisc_size_table *stab; 453 struct tc_sizespec *s; 454 unsigned int tsize = 0; 455 u16 *tab = NULL; 456 int err; 457 458 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy); 459 if (err < 0) 460 return ERR_PTR(err); 461 if (!tb[TCA_STAB_BASE]) 462 return ERR_PTR(-EINVAL); 463 464 s = nla_data(tb[TCA_STAB_BASE]); 465 466 if (s->tsize > 0) { 467 if (!tb[TCA_STAB_DATA]) 468 return ERR_PTR(-EINVAL); 469 tab = nla_data(tb[TCA_STAB_DATA]); 470 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); 471 } 472 473 if (tsize != s->tsize || (!tab && tsize > 0)) 474 return ERR_PTR(-EINVAL); 475 476 list_for_each_entry(stab, &qdisc_stab_list, list) { 477 if (memcmp(&stab->szopts, s, sizeof(*s))) 478 continue; 479 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16))) 480 continue; 481 stab->refcnt++; 482 return stab; 483 } 484 485 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); 486 if (!stab) 487 return ERR_PTR(-ENOMEM); 488 489 stab->refcnt = 1; 490 stab->szopts = *s; 491 if (tsize > 0) 492 memcpy(stab->data, tab, tsize * sizeof(u16)); 493 494 list_add_tail(&stab->list, &qdisc_stab_list); 495 496 return stab; 497 } 498 499 static void stab_kfree_rcu(struct rcu_head *head) 500 { 501 kfree(container_of(head, struct qdisc_size_table, rcu)); 502 } 503 504 void qdisc_put_stab(struct qdisc_size_table *tab) 505 { 506 if (!tab) 507 return; 508 509 if (--tab->refcnt == 0) { 510 list_del(&tab->list); 511 call_rcu_bh(&tab->rcu, stab_kfree_rcu); 512 } 513 } 514 EXPORT_SYMBOL(qdisc_put_stab); 515 516 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) 517 { 518 struct nlattr *nest; 519 520 nest = nla_nest_start(skb, TCA_STAB); 521 if (nest == NULL) 522 goto nla_put_failure; 523 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts)) 524 goto nla_put_failure; 525 nla_nest_end(skb, nest); 526 527 return skb->len; 528 529 nla_put_failure: 530 return -1; 531 } 532 533 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 534 const struct qdisc_size_table *stab) 535 { 536 int pkt_len, slot; 537 538 pkt_len = skb->len + stab->szopts.overhead; 539 if (unlikely(!stab->szopts.tsize)) 540 goto out; 541 542 slot = pkt_len + stab->szopts.cell_align; 543 if (unlikely(slot < 0)) 544 slot = 0; 545 546 slot >>= stab->szopts.cell_log; 547 if (likely(slot < stab->szopts.tsize)) 548 pkt_len = stab->data[slot]; 549 else 550 pkt_len = stab->data[stab->szopts.tsize - 1] * 551 (slot / stab->szopts.tsize) + 552 stab->data[slot % stab->szopts.tsize]; 553 554 pkt_len <<= stab->szopts.size_log; 555 out: 556 if (unlikely(pkt_len < 1)) 557 pkt_len = 1; 558 qdisc_skb_cb(skb)->pkt_len = pkt_len; 559 } 560 EXPORT_SYMBOL(__qdisc_calculate_pkt_len); 561 562 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc) 563 { 564 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { 565 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n", 566 txt, qdisc->ops->id, qdisc->handle >> 16); 567 qdisc->flags |= TCQ_F_WARN_NONWC; 568 } 569 } 570 EXPORT_SYMBOL(qdisc_warn_nonwc); 571 572 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) 573 { 574 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, 575 timer); 576 577 rcu_read_lock(); 578 __netif_schedule(qdisc_root(wd->qdisc)); 579 rcu_read_unlock(); 580 581 return HRTIMER_NORESTART; 582 } 583 584 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 585 { 586 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED); 587 wd->timer.function = qdisc_watchdog; 588 wd->qdisc = qdisc; 589 } 590 EXPORT_SYMBOL(qdisc_watchdog_init); 591 592 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires) 593 { 594 if (test_bit(__QDISC_STATE_DEACTIVATED, 595 &qdisc_root_sleeping(wd->qdisc)->state)) 596 return; 597 598 if (wd->last_expires == expires) 599 return; 600 601 wd->last_expires = expires; 602 hrtimer_start(&wd->timer, 603 ns_to_ktime(expires), 604 HRTIMER_MODE_ABS_PINNED); 605 } 606 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns); 607 608 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) 609 { 610 hrtimer_cancel(&wd->timer); 611 } 612 EXPORT_SYMBOL(qdisc_watchdog_cancel); 613 614 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) 615 { 616 unsigned int size = n * sizeof(struct hlist_head), i; 617 struct hlist_head *h; 618 619 if (size <= PAGE_SIZE) 620 h = kmalloc(size, GFP_KERNEL); 621 else 622 h = (struct hlist_head *) 623 __get_free_pages(GFP_KERNEL, get_order(size)); 624 625 if (h != NULL) { 626 for (i = 0; i < n; i++) 627 INIT_HLIST_HEAD(&h[i]); 628 } 629 return h; 630 } 631 632 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) 633 { 634 unsigned int size = n * sizeof(struct hlist_head); 635 636 if (size <= PAGE_SIZE) 637 kfree(h); 638 else 639 free_pages((unsigned long)h, get_order(size)); 640 } 641 642 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) 643 { 644 struct Qdisc_class_common *cl; 645 struct hlist_node *next; 646 struct hlist_head *nhash, *ohash; 647 unsigned int nsize, nmask, osize; 648 unsigned int i, h; 649 650 /* Rehash when load factor exceeds 0.75 */ 651 if (clhash->hashelems * 4 <= clhash->hashsize * 3) 652 return; 653 nsize = clhash->hashsize * 2; 654 nmask = nsize - 1; 655 nhash = qdisc_class_hash_alloc(nsize); 656 if (nhash == NULL) 657 return; 658 659 ohash = clhash->hash; 660 osize = clhash->hashsize; 661 662 sch_tree_lock(sch); 663 for (i = 0; i < osize; i++) { 664 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) { 665 h = qdisc_class_hash(cl->classid, nmask); 666 hlist_add_head(&cl->hnode, &nhash[h]); 667 } 668 } 669 clhash->hash = nhash; 670 clhash->hashsize = nsize; 671 clhash->hashmask = nmask; 672 sch_tree_unlock(sch); 673 674 qdisc_class_hash_free(ohash, osize); 675 } 676 EXPORT_SYMBOL(qdisc_class_hash_grow); 677 678 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) 679 { 680 unsigned int size = 4; 681 682 clhash->hash = qdisc_class_hash_alloc(size); 683 if (clhash->hash == NULL) 684 return -ENOMEM; 685 clhash->hashsize = size; 686 clhash->hashmask = size - 1; 687 clhash->hashelems = 0; 688 return 0; 689 } 690 EXPORT_SYMBOL(qdisc_class_hash_init); 691 692 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) 693 { 694 qdisc_class_hash_free(clhash->hash, clhash->hashsize); 695 } 696 EXPORT_SYMBOL(qdisc_class_hash_destroy); 697 698 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, 699 struct Qdisc_class_common *cl) 700 { 701 unsigned int h; 702 703 INIT_HLIST_NODE(&cl->hnode); 704 h = qdisc_class_hash(cl->classid, clhash->hashmask); 705 hlist_add_head(&cl->hnode, &clhash->hash[h]); 706 clhash->hashelems++; 707 } 708 EXPORT_SYMBOL(qdisc_class_hash_insert); 709 710 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, 711 struct Qdisc_class_common *cl) 712 { 713 hlist_del(&cl->hnode); 714 clhash->hashelems--; 715 } 716 EXPORT_SYMBOL(qdisc_class_hash_remove); 717 718 /* Allocate an unique handle from space managed by kernel 719 * Possible range is [8000-FFFF]:0000 (0x8000 values) 720 */ 721 static u32 qdisc_alloc_handle(struct net_device *dev) 722 { 723 int i = 0x8000; 724 static u32 autohandle = TC_H_MAKE(0x80000000U, 0); 725 726 do { 727 autohandle += TC_H_MAKE(0x10000U, 0); 728 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) 729 autohandle = TC_H_MAKE(0x80000000U, 0); 730 if (!qdisc_lookup(dev, autohandle)) 731 return autohandle; 732 cond_resched(); 733 } while (--i > 0); 734 735 return 0; 736 } 737 738 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n, 739 unsigned int len) 740 { 741 const struct Qdisc_class_ops *cops; 742 unsigned long cl; 743 u32 parentid; 744 int drops; 745 746 if (n == 0 && len == 0) 747 return; 748 drops = max_t(int, n, 0); 749 rcu_read_lock(); 750 while ((parentid = sch->parent)) { 751 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) 752 break; 753 754 if (sch->flags & TCQ_F_NOPARENT) 755 break; 756 /* TODO: perform the search on a per txq basis */ 757 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); 758 if (sch == NULL) { 759 WARN_ON_ONCE(parentid != TC_H_ROOT); 760 break; 761 } 762 cops = sch->ops->cl_ops; 763 if (cops->qlen_notify) { 764 cl = cops->get(sch, parentid); 765 cops->qlen_notify(sch, cl); 766 cops->put(sch, cl); 767 } 768 sch->q.qlen -= n; 769 sch->qstats.backlog -= len; 770 __qdisc_qstats_drop(sch, drops); 771 } 772 rcu_read_unlock(); 773 } 774 EXPORT_SYMBOL(qdisc_tree_reduce_backlog); 775 776 static void notify_and_destroy(struct net *net, struct sk_buff *skb, 777 struct nlmsghdr *n, u32 clid, 778 struct Qdisc *old, struct Qdisc *new) 779 { 780 if (new || old) 781 qdisc_notify(net, skb, n, clid, old, new); 782 783 if (old) 784 qdisc_destroy(old); 785 } 786 787 /* Graft qdisc "new" to class "classid" of qdisc "parent" or 788 * to device "dev". 789 * 790 * When appropriate send a netlink notification using 'skb' 791 * and "n". 792 * 793 * On success, destroy old qdisc. 794 */ 795 796 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 797 struct sk_buff *skb, struct nlmsghdr *n, u32 classid, 798 struct Qdisc *new, struct Qdisc *old) 799 { 800 struct Qdisc *q = old; 801 struct net *net = dev_net(dev); 802 int err = 0; 803 804 if (parent == NULL) { 805 unsigned int i, num_q, ingress; 806 807 ingress = 0; 808 num_q = dev->num_tx_queues; 809 if ((q && q->flags & TCQ_F_INGRESS) || 810 (new && new->flags & TCQ_F_INGRESS)) { 811 num_q = 1; 812 ingress = 1; 813 if (!dev_ingress_queue(dev)) 814 return -ENOENT; 815 } 816 817 if (dev->flags & IFF_UP) 818 dev_deactivate(dev); 819 820 if (new && new->ops->attach) 821 goto skip; 822 823 for (i = 0; i < num_q; i++) { 824 struct netdev_queue *dev_queue = dev_ingress_queue(dev); 825 826 if (!ingress) 827 dev_queue = netdev_get_tx_queue(dev, i); 828 829 old = dev_graft_qdisc(dev_queue, new); 830 if (new && i > 0) 831 atomic_inc(&new->refcnt); 832 833 if (!ingress) 834 qdisc_destroy(old); 835 } 836 837 skip: 838 if (!ingress) { 839 notify_and_destroy(net, skb, n, classid, 840 dev->qdisc, new); 841 if (new && !new->ops->attach) 842 atomic_inc(&new->refcnt); 843 dev->qdisc = new ? : &noop_qdisc; 844 845 if (new && new->ops->attach) 846 new->ops->attach(new); 847 } else { 848 notify_and_destroy(net, skb, n, classid, old, new); 849 } 850 851 if (dev->flags & IFF_UP) 852 dev_activate(dev); 853 } else { 854 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 855 856 err = -EOPNOTSUPP; 857 if (cops && cops->graft) { 858 unsigned long cl = cops->get(parent, classid); 859 if (cl) { 860 err = cops->graft(parent, cl, new, &old); 861 cops->put(parent, cl); 862 } else 863 err = -ENOENT; 864 } 865 if (!err) 866 notify_and_destroy(net, skb, n, classid, old, new); 867 } 868 return err; 869 } 870 871 /* lockdep annotation is needed for ingress; egress gets it only for name */ 872 static struct lock_class_key qdisc_tx_lock; 873 static struct lock_class_key qdisc_rx_lock; 874 875 /* 876 Allocate and initialize new qdisc. 877 878 Parameters are passed via opt. 879 */ 880 881 static struct Qdisc *qdisc_create(struct net_device *dev, 882 struct netdev_queue *dev_queue, 883 struct Qdisc *p, u32 parent, u32 handle, 884 struct nlattr **tca, int *errp) 885 { 886 int err; 887 struct nlattr *kind = tca[TCA_KIND]; 888 struct Qdisc *sch; 889 struct Qdisc_ops *ops; 890 struct qdisc_size_table *stab; 891 892 ops = qdisc_lookup_ops(kind); 893 #ifdef CONFIG_MODULES 894 if (ops == NULL && kind != NULL) { 895 char name[IFNAMSIZ]; 896 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 897 /* We dropped the RTNL semaphore in order to 898 * perform the module load. So, even if we 899 * succeeded in loading the module we have to 900 * tell the caller to replay the request. We 901 * indicate this using -EAGAIN. 902 * We replay the request because the device may 903 * go away in the mean time. 904 */ 905 rtnl_unlock(); 906 request_module("sch_%s", name); 907 rtnl_lock(); 908 ops = qdisc_lookup_ops(kind); 909 if (ops != NULL) { 910 /* We will try again qdisc_lookup_ops, 911 * so don't keep a reference. 912 */ 913 module_put(ops->owner); 914 err = -EAGAIN; 915 goto err_out; 916 } 917 } 918 } 919 #endif 920 921 err = -ENOENT; 922 if (ops == NULL) 923 goto err_out; 924 925 sch = qdisc_alloc(dev_queue, ops); 926 if (IS_ERR(sch)) { 927 err = PTR_ERR(sch); 928 goto err_out2; 929 } 930 931 sch->parent = parent; 932 933 if (handle == TC_H_INGRESS) { 934 sch->flags |= TCQ_F_INGRESS; 935 handle = TC_H_MAKE(TC_H_INGRESS, 0); 936 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); 937 } else { 938 if (handle == 0) { 939 handle = qdisc_alloc_handle(dev); 940 err = -ENOMEM; 941 if (handle == 0) 942 goto err_out3; 943 } 944 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); 945 if (!netif_is_multiqueue(dev)) 946 sch->flags |= TCQ_F_ONETXQUEUE; 947 } 948 949 sch->handle = handle; 950 951 /* This exist to keep backward compatible with a userspace 952 * loophole, what allowed userspace to get IFF_NO_QUEUE 953 * facility on older kernels by setting tx_queue_len=0 (prior 954 * to qdisc init), and then forgot to reinit tx_queue_len 955 * before again attaching a qdisc. 956 */ 957 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) { 958 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 959 netdev_info(dev, "Caught tx_queue_len zero misconfig\n"); 960 } 961 962 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { 963 if (qdisc_is_percpu_stats(sch)) { 964 sch->cpu_bstats = 965 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu); 966 if (!sch->cpu_bstats) 967 goto err_out4; 968 969 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue); 970 if (!sch->cpu_qstats) 971 goto err_out4; 972 } 973 974 if (tca[TCA_STAB]) { 975 stab = qdisc_get_stab(tca[TCA_STAB]); 976 if (IS_ERR(stab)) { 977 err = PTR_ERR(stab); 978 goto err_out4; 979 } 980 rcu_assign_pointer(sch->stab, stab); 981 } 982 if (tca[TCA_RATE]) { 983 seqcount_t *running; 984 985 err = -EOPNOTSUPP; 986 if (sch->flags & TCQ_F_MQROOT) 987 goto err_out4; 988 989 if ((sch->parent != TC_H_ROOT) && 990 !(sch->flags & TCQ_F_INGRESS) && 991 (!p || !(p->flags & TCQ_F_MQROOT))) 992 running = qdisc_root_sleeping_running(sch); 993 else 994 running = &sch->running; 995 996 err = gen_new_estimator(&sch->bstats, 997 sch->cpu_bstats, 998 &sch->rate_est, 999 NULL, 1000 running, 1001 tca[TCA_RATE]); 1002 if (err) 1003 goto err_out4; 1004 } 1005 1006 qdisc_hash_add(sch); 1007 1008 return sch; 1009 } 1010 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */ 1011 if (ops->destroy) 1012 ops->destroy(sch); 1013 err_out3: 1014 dev_put(dev); 1015 kfree((char *) sch - sch->padded); 1016 err_out2: 1017 module_put(ops->owner); 1018 err_out: 1019 *errp = err; 1020 return NULL; 1021 1022 err_out4: 1023 free_percpu(sch->cpu_bstats); 1024 free_percpu(sch->cpu_qstats); 1025 /* 1026 * Any broken qdiscs that would require a ops->reset() here? 1027 * The qdisc was never in action so it shouldn't be necessary. 1028 */ 1029 qdisc_put_stab(rtnl_dereference(sch->stab)); 1030 if (ops->destroy) 1031 ops->destroy(sch); 1032 goto err_out3; 1033 } 1034 1035 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) 1036 { 1037 struct qdisc_size_table *ostab, *stab = NULL; 1038 int err = 0; 1039 1040 if (tca[TCA_OPTIONS]) { 1041 if (sch->ops->change == NULL) 1042 return -EINVAL; 1043 err = sch->ops->change(sch, tca[TCA_OPTIONS]); 1044 if (err) 1045 return err; 1046 } 1047 1048 if (tca[TCA_STAB]) { 1049 stab = qdisc_get_stab(tca[TCA_STAB]); 1050 if (IS_ERR(stab)) 1051 return PTR_ERR(stab); 1052 } 1053 1054 ostab = rtnl_dereference(sch->stab); 1055 rcu_assign_pointer(sch->stab, stab); 1056 qdisc_put_stab(ostab); 1057 1058 if (tca[TCA_RATE]) { 1059 /* NB: ignores errors from replace_estimator 1060 because change can't be undone. */ 1061 if (sch->flags & TCQ_F_MQROOT) 1062 goto out; 1063 gen_replace_estimator(&sch->bstats, 1064 sch->cpu_bstats, 1065 &sch->rate_est, 1066 NULL, 1067 qdisc_root_sleeping_running(sch), 1068 tca[TCA_RATE]); 1069 } 1070 out: 1071 return 0; 1072 } 1073 1074 struct check_loop_arg { 1075 struct qdisc_walker w; 1076 struct Qdisc *p; 1077 int depth; 1078 }; 1079 1080 static int check_loop_fn(struct Qdisc *q, unsigned long cl, 1081 struct qdisc_walker *w); 1082 1083 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 1084 { 1085 struct check_loop_arg arg; 1086 1087 if (q->ops->cl_ops == NULL) 1088 return 0; 1089 1090 arg.w.stop = arg.w.skip = arg.w.count = 0; 1091 arg.w.fn = check_loop_fn; 1092 arg.depth = depth; 1093 arg.p = p; 1094 q->ops->cl_ops->walk(q, &arg.w); 1095 return arg.w.stop ? -ELOOP : 0; 1096 } 1097 1098 static int 1099 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 1100 { 1101 struct Qdisc *leaf; 1102 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1103 struct check_loop_arg *arg = (struct check_loop_arg *)w; 1104 1105 leaf = cops->leaf(q, cl); 1106 if (leaf) { 1107 if (leaf == arg->p || arg->depth > 7) 1108 return -ELOOP; 1109 return check_loop(leaf, arg->p, arg->depth + 1); 1110 } 1111 return 0; 1112 } 1113 1114 /* 1115 * Delete/get qdisc. 1116 */ 1117 1118 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n) 1119 { 1120 struct net *net = sock_net(skb->sk); 1121 struct tcmsg *tcm = nlmsg_data(n); 1122 struct nlattr *tca[TCA_MAX + 1]; 1123 struct net_device *dev; 1124 u32 clid; 1125 struct Qdisc *q = NULL; 1126 struct Qdisc *p = NULL; 1127 int err; 1128 1129 if ((n->nlmsg_type != RTM_GETQDISC) && 1130 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1131 return -EPERM; 1132 1133 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1134 if (err < 0) 1135 return err; 1136 1137 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1138 if (!dev) 1139 return -ENODEV; 1140 1141 clid = tcm->tcm_parent; 1142 if (clid) { 1143 if (clid != TC_H_ROOT) { 1144 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 1145 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1146 if (!p) 1147 return -ENOENT; 1148 q = qdisc_leaf(p, clid); 1149 } else if (dev_ingress_queue(dev)) { 1150 q = dev_ingress_queue(dev)->qdisc_sleeping; 1151 } 1152 } else { 1153 q = dev->qdisc; 1154 } 1155 if (!q) 1156 return -ENOENT; 1157 1158 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) 1159 return -EINVAL; 1160 } else { 1161 q = qdisc_lookup(dev, tcm->tcm_handle); 1162 if (!q) 1163 return -ENOENT; 1164 } 1165 1166 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1167 return -EINVAL; 1168 1169 if (n->nlmsg_type == RTM_DELQDISC) { 1170 if (!clid) 1171 return -EINVAL; 1172 if (q->handle == 0) 1173 return -ENOENT; 1174 err = qdisc_graft(dev, p, skb, n, clid, NULL, q); 1175 if (err != 0) 1176 return err; 1177 } else { 1178 qdisc_notify(net, skb, n, clid, NULL, q); 1179 } 1180 return 0; 1181 } 1182 1183 /* 1184 * Create/change qdisc. 1185 */ 1186 1187 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n) 1188 { 1189 struct net *net = sock_net(skb->sk); 1190 struct tcmsg *tcm; 1191 struct nlattr *tca[TCA_MAX + 1]; 1192 struct net_device *dev; 1193 u32 clid; 1194 struct Qdisc *q, *p; 1195 int err; 1196 1197 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1198 return -EPERM; 1199 1200 replay: 1201 /* Reinit, just in case something touches this. */ 1202 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1203 if (err < 0) 1204 return err; 1205 1206 tcm = nlmsg_data(n); 1207 clid = tcm->tcm_parent; 1208 q = p = NULL; 1209 1210 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1211 if (!dev) 1212 return -ENODEV; 1213 1214 1215 if (clid) { 1216 if (clid != TC_H_ROOT) { 1217 if (clid != TC_H_INGRESS) { 1218 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1219 if (!p) 1220 return -ENOENT; 1221 q = qdisc_leaf(p, clid); 1222 } else if (dev_ingress_queue_create(dev)) { 1223 q = dev_ingress_queue(dev)->qdisc_sleeping; 1224 } 1225 } else { 1226 q = dev->qdisc; 1227 } 1228 1229 /* It may be default qdisc, ignore it */ 1230 if (q && q->handle == 0) 1231 q = NULL; 1232 1233 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 1234 if (tcm->tcm_handle) { 1235 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) 1236 return -EEXIST; 1237 if (TC_H_MIN(tcm->tcm_handle)) 1238 return -EINVAL; 1239 q = qdisc_lookup(dev, tcm->tcm_handle); 1240 if (!q) 1241 goto create_n_graft; 1242 if (n->nlmsg_flags & NLM_F_EXCL) 1243 return -EEXIST; 1244 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1245 return -EINVAL; 1246 if (q == p || 1247 (p && check_loop(q, p, 0))) 1248 return -ELOOP; 1249 atomic_inc(&q->refcnt); 1250 goto graft; 1251 } else { 1252 if (!q) 1253 goto create_n_graft; 1254 1255 /* This magic test requires explanation. 1256 * 1257 * We know, that some child q is already 1258 * attached to this parent and have choice: 1259 * either to change it or to create/graft new one. 1260 * 1261 * 1. We are allowed to create/graft only 1262 * if CREATE and REPLACE flags are set. 1263 * 1264 * 2. If EXCL is set, requestor wanted to say, 1265 * that qdisc tcm_handle is not expected 1266 * to exist, so that we choose create/graft too. 1267 * 1268 * 3. The last case is when no flags are set. 1269 * Alas, it is sort of hole in API, we 1270 * cannot decide what to do unambiguously. 1271 * For now we select create/graft, if 1272 * user gave KIND, which does not match existing. 1273 */ 1274 if ((n->nlmsg_flags & NLM_F_CREATE) && 1275 (n->nlmsg_flags & NLM_F_REPLACE) && 1276 ((n->nlmsg_flags & NLM_F_EXCL) || 1277 (tca[TCA_KIND] && 1278 nla_strcmp(tca[TCA_KIND], q->ops->id)))) 1279 goto create_n_graft; 1280 } 1281 } 1282 } else { 1283 if (!tcm->tcm_handle) 1284 return -EINVAL; 1285 q = qdisc_lookup(dev, tcm->tcm_handle); 1286 } 1287 1288 /* Change qdisc parameters */ 1289 if (q == NULL) 1290 return -ENOENT; 1291 if (n->nlmsg_flags & NLM_F_EXCL) 1292 return -EEXIST; 1293 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1294 return -EINVAL; 1295 err = qdisc_change(q, tca); 1296 if (err == 0) 1297 qdisc_notify(net, skb, n, clid, NULL, q); 1298 return err; 1299 1300 create_n_graft: 1301 if (!(n->nlmsg_flags & NLM_F_CREATE)) 1302 return -ENOENT; 1303 if (clid == TC_H_INGRESS) { 1304 if (dev_ingress_queue(dev)) 1305 q = qdisc_create(dev, dev_ingress_queue(dev), p, 1306 tcm->tcm_parent, tcm->tcm_parent, 1307 tca, &err); 1308 else 1309 err = -ENOENT; 1310 } else { 1311 struct netdev_queue *dev_queue; 1312 1313 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) 1314 dev_queue = p->ops->cl_ops->select_queue(p, tcm); 1315 else if (p) 1316 dev_queue = p->dev_queue; 1317 else 1318 dev_queue = netdev_get_tx_queue(dev, 0); 1319 1320 q = qdisc_create(dev, dev_queue, p, 1321 tcm->tcm_parent, tcm->tcm_handle, 1322 tca, &err); 1323 } 1324 if (q == NULL) { 1325 if (err == -EAGAIN) 1326 goto replay; 1327 return err; 1328 } 1329 1330 graft: 1331 err = qdisc_graft(dev, p, skb, n, clid, q, NULL); 1332 if (err) { 1333 if (q) 1334 qdisc_destroy(q); 1335 return err; 1336 } 1337 1338 return 0; 1339 } 1340 1341 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, 1342 u32 portid, u32 seq, u16 flags, int event) 1343 { 1344 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL; 1345 struct gnet_stats_queue __percpu *cpu_qstats = NULL; 1346 struct tcmsg *tcm; 1347 struct nlmsghdr *nlh; 1348 unsigned char *b = skb_tail_pointer(skb); 1349 struct gnet_dump d; 1350 struct qdisc_size_table *stab; 1351 __u32 qlen; 1352 1353 cond_resched(); 1354 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1355 if (!nlh) 1356 goto out_nlmsg_trim; 1357 tcm = nlmsg_data(nlh); 1358 tcm->tcm_family = AF_UNSPEC; 1359 tcm->tcm__pad1 = 0; 1360 tcm->tcm__pad2 = 0; 1361 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1362 tcm->tcm_parent = clid; 1363 tcm->tcm_handle = q->handle; 1364 tcm->tcm_info = atomic_read(&q->refcnt); 1365 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1366 goto nla_put_failure; 1367 if (q->ops->dump && q->ops->dump(q, skb) < 0) 1368 goto nla_put_failure; 1369 qlen = q->q.qlen; 1370 1371 stab = rtnl_dereference(q->stab); 1372 if (stab && qdisc_dump_stab(skb, stab) < 0) 1373 goto nla_put_failure; 1374 1375 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1376 NULL, &d, TCA_PAD) < 0) 1377 goto nla_put_failure; 1378 1379 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 1380 goto nla_put_failure; 1381 1382 if (qdisc_is_percpu_stats(q)) { 1383 cpu_bstats = q->cpu_bstats; 1384 cpu_qstats = q->cpu_qstats; 1385 } 1386 1387 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q), 1388 &d, cpu_bstats, &q->bstats) < 0 || 1389 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || 1390 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0) 1391 goto nla_put_failure; 1392 1393 if (gnet_stats_finish_copy(&d) < 0) 1394 goto nla_put_failure; 1395 1396 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1397 return skb->len; 1398 1399 out_nlmsg_trim: 1400 nla_put_failure: 1401 nlmsg_trim(skb, b); 1402 return -1; 1403 } 1404 1405 static bool tc_qdisc_dump_ignore(struct Qdisc *q) 1406 { 1407 return (q->flags & TCQ_F_BUILTIN) ? true : false; 1408 } 1409 1410 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 1411 struct nlmsghdr *n, u32 clid, 1412 struct Qdisc *old, struct Qdisc *new) 1413 { 1414 struct sk_buff *skb; 1415 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1416 1417 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1418 if (!skb) 1419 return -ENOBUFS; 1420 1421 if (old && !tc_qdisc_dump_ignore(old)) { 1422 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq, 1423 0, RTM_DELQDISC) < 0) 1424 goto err_out; 1425 } 1426 if (new && !tc_qdisc_dump_ignore(new)) { 1427 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq, 1428 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) 1429 goto err_out; 1430 } 1431 1432 if (skb->len) 1433 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1434 n->nlmsg_flags & NLM_F_ECHO); 1435 1436 err_out: 1437 kfree_skb(skb); 1438 return -EINVAL; 1439 } 1440 1441 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, 1442 struct netlink_callback *cb, 1443 int *q_idx_p, int s_q_idx, bool recur) 1444 { 1445 int ret = 0, q_idx = *q_idx_p; 1446 struct Qdisc *q; 1447 int b; 1448 1449 if (!root) 1450 return 0; 1451 1452 q = root; 1453 if (q_idx < s_q_idx) { 1454 q_idx++; 1455 } else { 1456 if (!tc_qdisc_dump_ignore(q) && 1457 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1458 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1459 RTM_NEWQDISC) <= 0) 1460 goto done; 1461 q_idx++; 1462 } 1463 1464 /* If dumping singletons, there is no qdisc_dev(root) and the singleton 1465 * itself has already been dumped. 1466 * 1467 * If we've already dumped the top-level (ingress) qdisc above and the global 1468 * qdisc hashtable, we don't want to hit it again 1469 */ 1470 if (!qdisc_dev(root) || !recur) 1471 goto out; 1472 1473 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1474 if (q_idx < s_q_idx) { 1475 q_idx++; 1476 continue; 1477 } 1478 if (!tc_qdisc_dump_ignore(q) && 1479 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1480 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1481 RTM_NEWQDISC) <= 0) 1482 goto done; 1483 q_idx++; 1484 } 1485 1486 out: 1487 *q_idx_p = q_idx; 1488 return ret; 1489 done: 1490 ret = -1; 1491 goto out; 1492 } 1493 1494 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 1495 { 1496 struct net *net = sock_net(skb->sk); 1497 int idx, q_idx; 1498 int s_idx, s_q_idx; 1499 struct net_device *dev; 1500 1501 s_idx = cb->args[0]; 1502 s_q_idx = q_idx = cb->args[1]; 1503 1504 idx = 0; 1505 ASSERT_RTNL(); 1506 for_each_netdev(net, dev) { 1507 struct netdev_queue *dev_queue; 1508 1509 if (idx < s_idx) 1510 goto cont; 1511 if (idx > s_idx) 1512 s_q_idx = 0; 1513 q_idx = 0; 1514 1515 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx, 1516 true) < 0) 1517 goto done; 1518 1519 dev_queue = dev_ingress_queue(dev); 1520 if (dev_queue && 1521 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, 1522 &q_idx, s_q_idx, false) < 0) 1523 goto done; 1524 1525 cont: 1526 idx++; 1527 } 1528 1529 done: 1530 cb->args[0] = idx; 1531 cb->args[1] = q_idx; 1532 1533 return skb->len; 1534 } 1535 1536 1537 1538 /************************************************ 1539 * Traffic classes manipulation. * 1540 ************************************************/ 1541 1542 1543 1544 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n) 1545 { 1546 struct net *net = sock_net(skb->sk); 1547 struct tcmsg *tcm = nlmsg_data(n); 1548 struct nlattr *tca[TCA_MAX + 1]; 1549 struct net_device *dev; 1550 struct Qdisc *q = NULL; 1551 const struct Qdisc_class_ops *cops; 1552 unsigned long cl = 0; 1553 unsigned long new_cl; 1554 u32 portid; 1555 u32 clid; 1556 u32 qid; 1557 int err; 1558 1559 if ((n->nlmsg_type != RTM_GETTCLASS) && 1560 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1561 return -EPERM; 1562 1563 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1564 if (err < 0) 1565 return err; 1566 1567 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1568 if (!dev) 1569 return -ENODEV; 1570 1571 /* 1572 parent == TC_H_UNSPEC - unspecified parent. 1573 parent == TC_H_ROOT - class is root, which has no parent. 1574 parent == X:0 - parent is root class. 1575 parent == X:Y - parent is a node in hierarchy. 1576 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 1577 1578 handle == 0:0 - generate handle from kernel pool. 1579 handle == 0:Y - class is X:Y, where X:0 is qdisc. 1580 handle == X:Y - clear. 1581 handle == X:0 - root class. 1582 */ 1583 1584 /* Step 1. Determine qdisc handle X:0 */ 1585 1586 portid = tcm->tcm_parent; 1587 clid = tcm->tcm_handle; 1588 qid = TC_H_MAJ(clid); 1589 1590 if (portid != TC_H_ROOT) { 1591 u32 qid1 = TC_H_MAJ(portid); 1592 1593 if (qid && qid1) { 1594 /* If both majors are known, they must be identical. */ 1595 if (qid != qid1) 1596 return -EINVAL; 1597 } else if (qid1) { 1598 qid = qid1; 1599 } else if (qid == 0) 1600 qid = dev->qdisc->handle; 1601 1602 /* Now qid is genuine qdisc handle consistent 1603 * both with parent and child. 1604 * 1605 * TC_H_MAJ(portid) still may be unspecified, complete it now. 1606 */ 1607 if (portid) 1608 portid = TC_H_MAKE(qid, portid); 1609 } else { 1610 if (qid == 0) 1611 qid = dev->qdisc->handle; 1612 } 1613 1614 /* OK. Locate qdisc */ 1615 q = qdisc_lookup(dev, qid); 1616 if (!q) 1617 return -ENOENT; 1618 1619 /* An check that it supports classes */ 1620 cops = q->ops->cl_ops; 1621 if (cops == NULL) 1622 return -EINVAL; 1623 1624 /* Now try to get class */ 1625 if (clid == 0) { 1626 if (portid == TC_H_ROOT) 1627 clid = qid; 1628 } else 1629 clid = TC_H_MAKE(qid, clid); 1630 1631 if (clid) 1632 cl = cops->get(q, clid); 1633 1634 if (cl == 0) { 1635 err = -ENOENT; 1636 if (n->nlmsg_type != RTM_NEWTCLASS || 1637 !(n->nlmsg_flags & NLM_F_CREATE)) 1638 goto out; 1639 } else { 1640 switch (n->nlmsg_type) { 1641 case RTM_NEWTCLASS: 1642 err = -EEXIST; 1643 if (n->nlmsg_flags & NLM_F_EXCL) 1644 goto out; 1645 break; 1646 case RTM_DELTCLASS: 1647 err = -EOPNOTSUPP; 1648 if (cops->delete) 1649 err = cops->delete(q, cl); 1650 if (err == 0) 1651 tclass_notify(net, skb, n, q, cl, 1652 RTM_DELTCLASS); 1653 goto out; 1654 case RTM_GETTCLASS: 1655 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS); 1656 goto out; 1657 default: 1658 err = -EINVAL; 1659 goto out; 1660 } 1661 } 1662 1663 new_cl = cl; 1664 err = -EOPNOTSUPP; 1665 if (cops->change) 1666 err = cops->change(q, clid, portid, tca, &new_cl); 1667 if (err == 0) 1668 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS); 1669 1670 out: 1671 if (cl) 1672 cops->put(q, cl); 1673 1674 return err; 1675 } 1676 1677 1678 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1679 unsigned long cl, 1680 u32 portid, u32 seq, u16 flags, int event) 1681 { 1682 struct tcmsg *tcm; 1683 struct nlmsghdr *nlh; 1684 unsigned char *b = skb_tail_pointer(skb); 1685 struct gnet_dump d; 1686 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1687 1688 cond_resched(); 1689 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1690 if (!nlh) 1691 goto out_nlmsg_trim; 1692 tcm = nlmsg_data(nlh); 1693 tcm->tcm_family = AF_UNSPEC; 1694 tcm->tcm__pad1 = 0; 1695 tcm->tcm__pad2 = 0; 1696 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1697 tcm->tcm_parent = q->handle; 1698 tcm->tcm_handle = q->handle; 1699 tcm->tcm_info = 0; 1700 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1701 goto nla_put_failure; 1702 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1703 goto nla_put_failure; 1704 1705 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1706 NULL, &d, TCA_PAD) < 0) 1707 goto nla_put_failure; 1708 1709 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1710 goto nla_put_failure; 1711 1712 if (gnet_stats_finish_copy(&d) < 0) 1713 goto nla_put_failure; 1714 1715 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1716 return skb->len; 1717 1718 out_nlmsg_trim: 1719 nla_put_failure: 1720 nlmsg_trim(skb, b); 1721 return -1; 1722 } 1723 1724 static int tclass_notify(struct net *net, struct sk_buff *oskb, 1725 struct nlmsghdr *n, struct Qdisc *q, 1726 unsigned long cl, int event) 1727 { 1728 struct sk_buff *skb; 1729 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1730 1731 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1732 if (!skb) 1733 return -ENOBUFS; 1734 1735 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) { 1736 kfree_skb(skb); 1737 return -EINVAL; 1738 } 1739 1740 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1741 n->nlmsg_flags & NLM_F_ECHO); 1742 } 1743 1744 struct qdisc_dump_args { 1745 struct qdisc_walker w; 1746 struct sk_buff *skb; 1747 struct netlink_callback *cb; 1748 }; 1749 1750 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, 1751 struct qdisc_walker *arg) 1752 { 1753 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 1754 1755 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid, 1756 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 1757 RTM_NEWTCLASS); 1758 } 1759 1760 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, 1761 struct tcmsg *tcm, struct netlink_callback *cb, 1762 int *t_p, int s_t) 1763 { 1764 struct qdisc_dump_args arg; 1765 1766 if (tc_qdisc_dump_ignore(q) || 1767 *t_p < s_t || !q->ops->cl_ops || 1768 (tcm->tcm_parent && 1769 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 1770 (*t_p)++; 1771 return 0; 1772 } 1773 if (*t_p > s_t) 1774 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 1775 arg.w.fn = qdisc_class_dump; 1776 arg.skb = skb; 1777 arg.cb = cb; 1778 arg.w.stop = 0; 1779 arg.w.skip = cb->args[1]; 1780 arg.w.count = 0; 1781 q->ops->cl_ops->walk(q, &arg.w); 1782 cb->args[1] = arg.w.count; 1783 if (arg.w.stop) 1784 return -1; 1785 (*t_p)++; 1786 return 0; 1787 } 1788 1789 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, 1790 struct tcmsg *tcm, struct netlink_callback *cb, 1791 int *t_p, int s_t) 1792 { 1793 struct Qdisc *q; 1794 int b; 1795 1796 if (!root) 1797 return 0; 1798 1799 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) 1800 return -1; 1801 1802 if (!qdisc_dev(root)) 1803 return 0; 1804 1805 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1806 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1807 return -1; 1808 } 1809 1810 return 0; 1811 } 1812 1813 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 1814 { 1815 struct tcmsg *tcm = nlmsg_data(cb->nlh); 1816 struct net *net = sock_net(skb->sk); 1817 struct netdev_queue *dev_queue; 1818 struct net_device *dev; 1819 int t, s_t; 1820 1821 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 1822 return 0; 1823 dev = dev_get_by_index(net, tcm->tcm_ifindex); 1824 if (!dev) 1825 return 0; 1826 1827 s_t = cb->args[0]; 1828 t = 0; 1829 1830 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) 1831 goto done; 1832 1833 dev_queue = dev_ingress_queue(dev); 1834 if (dev_queue && 1835 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, 1836 &t, s_t) < 0) 1837 goto done; 1838 1839 done: 1840 cb->args[0] = t; 1841 1842 dev_put(dev); 1843 return skb->len; 1844 } 1845 1846 /* Main classifier routine: scans classifier chain attached 1847 * to this qdisc, (optionally) tests for protocol and asks 1848 * specific classifiers. 1849 */ 1850 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp, 1851 struct tcf_result *res, bool compat_mode) 1852 { 1853 __be16 protocol = tc_skb_protocol(skb); 1854 #ifdef CONFIG_NET_CLS_ACT 1855 const int max_reclassify_loop = 4; 1856 const struct tcf_proto *old_tp = tp; 1857 int limit = 0; 1858 1859 reclassify: 1860 #endif 1861 for (; tp; tp = rcu_dereference_bh(tp->next)) { 1862 int err; 1863 1864 if (tp->protocol != protocol && 1865 tp->protocol != htons(ETH_P_ALL)) 1866 continue; 1867 1868 err = tp->classify(skb, tp, res); 1869 #ifdef CONFIG_NET_CLS_ACT 1870 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) 1871 goto reset; 1872 #endif 1873 if (err >= 0) 1874 return err; 1875 } 1876 1877 return TC_ACT_UNSPEC; /* signal: continue lookup */ 1878 #ifdef CONFIG_NET_CLS_ACT 1879 reset: 1880 if (unlikely(limit++ >= max_reclassify_loop)) { 1881 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n", 1882 tp->q->ops->id, tp->prio & 0xffff, 1883 ntohs(tp->protocol)); 1884 return TC_ACT_SHOT; 1885 } 1886 1887 tp = old_tp; 1888 protocol = tc_skb_protocol(skb); 1889 goto reclassify; 1890 #endif 1891 } 1892 EXPORT_SYMBOL(tc_classify); 1893 1894 #ifdef CONFIG_PROC_FS 1895 static int psched_show(struct seq_file *seq, void *v) 1896 { 1897 seq_printf(seq, "%08x %08x %08x %08x\n", 1898 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), 1899 1000000, 1900 (u32)NSEC_PER_SEC / hrtimer_resolution); 1901 1902 return 0; 1903 } 1904 1905 static int psched_open(struct inode *inode, struct file *file) 1906 { 1907 return single_open(file, psched_show, NULL); 1908 } 1909 1910 static const struct file_operations psched_fops = { 1911 .owner = THIS_MODULE, 1912 .open = psched_open, 1913 .read = seq_read, 1914 .llseek = seq_lseek, 1915 .release = single_release, 1916 }; 1917 1918 static int __net_init psched_net_init(struct net *net) 1919 { 1920 struct proc_dir_entry *e; 1921 1922 e = proc_create("psched", 0, net->proc_net, &psched_fops); 1923 if (e == NULL) 1924 return -ENOMEM; 1925 1926 return 0; 1927 } 1928 1929 static void __net_exit psched_net_exit(struct net *net) 1930 { 1931 remove_proc_entry("psched", net->proc_net); 1932 } 1933 #else 1934 static int __net_init psched_net_init(struct net *net) 1935 { 1936 return 0; 1937 } 1938 1939 static void __net_exit psched_net_exit(struct net *net) 1940 { 1941 } 1942 #endif 1943 1944 static struct pernet_operations psched_net_ops = { 1945 .init = psched_net_init, 1946 .exit = psched_net_exit, 1947 }; 1948 1949 static int __init pktsched_init(void) 1950 { 1951 int err; 1952 1953 err = register_pernet_subsys(&psched_net_ops); 1954 if (err) { 1955 pr_err("pktsched_init: " 1956 "cannot initialize per netns operations\n"); 1957 return err; 1958 } 1959 1960 register_qdisc(&pfifo_fast_ops); 1961 register_qdisc(&pfifo_qdisc_ops); 1962 register_qdisc(&bfifo_qdisc_ops); 1963 register_qdisc(&pfifo_head_drop_qdisc_ops); 1964 register_qdisc(&mq_qdisc_ops); 1965 register_qdisc(&noqueue_qdisc_ops); 1966 1967 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL); 1968 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL); 1969 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, 1970 NULL); 1971 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL); 1972 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL); 1973 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, 1974 NULL); 1975 1976 return 0; 1977 } 1978 1979 subsys_initcall(pktsched_init); 1980
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.