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 #include <net/pkt_cls.h> 39 40 /* 41 42 Short review. 43 ------------- 44 45 This file consists of two interrelated parts: 46 47 1. queueing disciplines manager frontend. 48 2. traffic classes manager frontend. 49 50 Generally, queueing discipline ("qdisc") is a black box, 51 which is able to enqueue packets and to dequeue them (when 52 device is ready to send something) in order and at times 53 determined by algorithm hidden in it. 54 55 qdisc's are divided to two categories: 56 - "queues", which have no internal structure visible from outside. 57 - "schedulers", which split all the packets to "traffic classes", 58 using "packet classifiers" (look at cls_api.c) 59 60 In turn, classes may have child qdiscs (as rule, queues) 61 attached to them etc. etc. etc. 62 63 The goal of the routines in this file is to translate 64 information supplied by user in the form of handles 65 to more intelligible for kernel form, to make some sanity 66 checks and part of work, which is common to all qdiscs 67 and to provide rtnetlink notifications. 68 69 All real intelligent work is done inside qdisc modules. 70 71 72 73 Every discipline has two major routines: enqueue and dequeue. 74 75 ---dequeue 76 77 dequeue usually returns a skb to send. It is allowed to return NULL, 78 but it does not mean that queue is empty, it just means that 79 discipline does not want to send anything this time. 80 Queue is really empty if q->q.qlen == 0. 81 For complicated disciplines with multiple queues q->q is not 82 real packet queue, but however q->q.qlen must be valid. 83 84 ---enqueue 85 86 enqueue returns 0, if packet was enqueued successfully. 87 If packet (this one or another one) was dropped, it returns 88 not zero error code. 89 NET_XMIT_DROP - this packet dropped 90 Expected action: do not backoff, but wait until queue will clear. 91 NET_XMIT_CN - probably this packet enqueued, but another one dropped. 92 Expected action: backoff or ignore 93 94 Auxiliary routines: 95 96 ---peek 97 98 like dequeue but without removing a packet from the queue 99 100 ---reset 101 102 returns qdisc to initial state: purge all buffers, clear all 103 timers, counters (except for statistics) etc. 104 105 ---init 106 107 initializes newly created qdisc. 108 109 ---destroy 110 111 destroys resources allocated by init and during lifetime of qdisc. 112 113 ---change 114 115 changes qdisc parameters. 116 */ 117 118 /* Protects list of registered TC modules. It is pure SMP lock. */ 119 static DEFINE_RWLOCK(qdisc_mod_lock); 120 121 122 /************************************************ 123 * Queueing disciplines manipulation. * 124 ************************************************/ 125 126 127 /* The list of all installed queueing disciplines. */ 128 129 static struct Qdisc_ops *qdisc_base; 130 131 /* Register/unregister queueing discipline */ 132 133 int register_qdisc(struct Qdisc_ops *qops) 134 { 135 struct Qdisc_ops *q, **qp; 136 int rc = -EEXIST; 137 138 write_lock(&qdisc_mod_lock); 139 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 140 if (!strcmp(qops->id, q->id)) 141 goto out; 142 143 if (qops->enqueue == NULL) 144 qops->enqueue = noop_qdisc_ops.enqueue; 145 if (qops->peek == NULL) { 146 if (qops->dequeue == NULL) 147 qops->peek = noop_qdisc_ops.peek; 148 else 149 goto out_einval; 150 } 151 if (qops->dequeue == NULL) 152 qops->dequeue = noop_qdisc_ops.dequeue; 153 154 if (qops->cl_ops) { 155 const struct Qdisc_class_ops *cops = qops->cl_ops; 156 157 if (!(cops->find && cops->walk && cops->leaf)) 158 goto out_einval; 159 160 if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf)) 161 goto out_einval; 162 } 163 164 qops->next = NULL; 165 *qp = qops; 166 rc = 0; 167 out: 168 write_unlock(&qdisc_mod_lock); 169 return rc; 170 171 out_einval: 172 rc = -EINVAL; 173 goto out; 174 } 175 EXPORT_SYMBOL(register_qdisc); 176 177 int unregister_qdisc(struct Qdisc_ops *qops) 178 { 179 struct Qdisc_ops *q, **qp; 180 int err = -ENOENT; 181 182 write_lock(&qdisc_mod_lock); 183 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 184 if (q == qops) 185 break; 186 if (q) { 187 *qp = q->next; 188 q->next = NULL; 189 err = 0; 190 } 191 write_unlock(&qdisc_mod_lock); 192 return err; 193 } 194 EXPORT_SYMBOL(unregister_qdisc); 195 196 /* Get default qdisc if not otherwise specified */ 197 void qdisc_get_default(char *name, size_t len) 198 { 199 read_lock(&qdisc_mod_lock); 200 strlcpy(name, default_qdisc_ops->id, len); 201 read_unlock(&qdisc_mod_lock); 202 } 203 204 static struct Qdisc_ops *qdisc_lookup_default(const char *name) 205 { 206 struct Qdisc_ops *q = NULL; 207 208 for (q = qdisc_base; q; q = q->next) { 209 if (!strcmp(name, q->id)) { 210 if (!try_module_get(q->owner)) 211 q = NULL; 212 break; 213 } 214 } 215 216 return q; 217 } 218 219 /* Set new default qdisc to use */ 220 int qdisc_set_default(const char *name) 221 { 222 const struct Qdisc_ops *ops; 223 224 if (!capable(CAP_NET_ADMIN)) 225 return -EPERM; 226 227 write_lock(&qdisc_mod_lock); 228 ops = qdisc_lookup_default(name); 229 if (!ops) { 230 /* Not found, drop lock and try to load module */ 231 write_unlock(&qdisc_mod_lock); 232 request_module("sch_%s", name); 233 write_lock(&qdisc_mod_lock); 234 235 ops = qdisc_lookup_default(name); 236 } 237 238 if (ops) { 239 /* Set new default */ 240 module_put(default_qdisc_ops->owner); 241 default_qdisc_ops = ops; 242 } 243 write_unlock(&qdisc_mod_lock); 244 245 return ops ? 0 : -ENOENT; 246 } 247 248 #ifdef CONFIG_NET_SCH_DEFAULT 249 /* Set default value from kernel config */ 250 static int __init sch_default_qdisc(void) 251 { 252 return qdisc_set_default(CONFIG_DEFAULT_NET_SCH); 253 } 254 late_initcall(sch_default_qdisc); 255 #endif 256 257 /* We know handle. Find qdisc among all qdisc's attached to device 258 * (root qdisc, all its children, children of children etc.) 259 * Note: caller either uses rtnl or rcu_read_lock() 260 */ 261 262 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle) 263 { 264 struct Qdisc *q; 265 266 if (!qdisc_dev(root)) 267 return (root->handle == handle ? root : NULL); 268 269 if (!(root->flags & TCQ_F_BUILTIN) && 270 root->handle == handle) 271 return root; 272 273 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) { 274 if (q->handle == handle) 275 return q; 276 } 277 return NULL; 278 } 279 280 void qdisc_hash_add(struct Qdisc *q, bool invisible) 281 { 282 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 283 ASSERT_RTNL(); 284 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle); 285 if (invisible) 286 q->flags |= TCQ_F_INVISIBLE; 287 } 288 } 289 EXPORT_SYMBOL(qdisc_hash_add); 290 291 void qdisc_hash_del(struct Qdisc *q) 292 { 293 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 294 ASSERT_RTNL(); 295 hash_del_rcu(&q->hash); 296 } 297 } 298 EXPORT_SYMBOL(qdisc_hash_del); 299 300 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) 301 { 302 struct Qdisc *q; 303 304 q = qdisc_match_from_root(dev->qdisc, handle); 305 if (q) 306 goto out; 307 308 if (dev_ingress_queue(dev)) 309 q = qdisc_match_from_root( 310 dev_ingress_queue(dev)->qdisc_sleeping, 311 handle); 312 out: 313 return q; 314 } 315 316 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) 317 { 318 unsigned long cl; 319 struct Qdisc *leaf; 320 const struct Qdisc_class_ops *cops = p->ops->cl_ops; 321 322 if (cops == NULL) 323 return NULL; 324 cl = cops->find(p, classid); 325 326 if (cl == 0) 327 return NULL; 328 leaf = cops->leaf(p, cl); 329 return leaf; 330 } 331 332 /* Find queueing discipline by name */ 333 334 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) 335 { 336 struct Qdisc_ops *q = NULL; 337 338 if (kind) { 339 read_lock(&qdisc_mod_lock); 340 for (q = qdisc_base; q; q = q->next) { 341 if (nla_strcmp(kind, q->id) == 0) { 342 if (!try_module_get(q->owner)) 343 q = NULL; 344 break; 345 } 346 } 347 read_unlock(&qdisc_mod_lock); 348 } 349 return q; 350 } 351 352 /* The linklayer setting were not transferred from iproute2, in older 353 * versions, and the rate tables lookup systems have been dropped in 354 * the kernel. To keep backward compatible with older iproute2 tc 355 * utils, we detect the linklayer setting by detecting if the rate 356 * table were modified. 357 * 358 * For linklayer ATM table entries, the rate table will be aligned to 359 * 48 bytes, thus some table entries will contain the same value. The 360 * mpu (min packet unit) is also encoded into the old rate table, thus 361 * starting from the mpu, we find low and high table entries for 362 * mapping this cell. If these entries contain the same value, when 363 * the rate tables have been modified for linklayer ATM. 364 * 365 * This is done by rounding mpu to the nearest 48 bytes cell/entry, 366 * and then roundup to the next cell, calc the table entry one below, 367 * and compare. 368 */ 369 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab) 370 { 371 int low = roundup(r->mpu, 48); 372 int high = roundup(low+1, 48); 373 int cell_low = low >> r->cell_log; 374 int cell_high = (high >> r->cell_log) - 1; 375 376 /* rtab is too inaccurate at rates > 100Mbit/s */ 377 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) { 378 pr_debug("TC linklayer: Giving up ATM detection\n"); 379 return TC_LINKLAYER_ETHERNET; 380 } 381 382 if ((cell_high > cell_low) && (cell_high < 256) 383 && (rtab[cell_low] == rtab[cell_high])) { 384 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n", 385 cell_low, cell_high, rtab[cell_high]); 386 return TC_LINKLAYER_ATM; 387 } 388 return TC_LINKLAYER_ETHERNET; 389 } 390 391 static struct qdisc_rate_table *qdisc_rtab_list; 392 393 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, 394 struct nlattr *tab) 395 { 396 struct qdisc_rate_table *rtab; 397 398 if (tab == NULL || r->rate == 0 || r->cell_log == 0 || 399 nla_len(tab) != TC_RTAB_SIZE) 400 return NULL; 401 402 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { 403 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) && 404 !memcmp(&rtab->data, nla_data(tab), 1024)) { 405 rtab->refcnt++; 406 return rtab; 407 } 408 } 409 410 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); 411 if (rtab) { 412 rtab->rate = *r; 413 rtab->refcnt = 1; 414 memcpy(rtab->data, nla_data(tab), 1024); 415 if (r->linklayer == TC_LINKLAYER_UNAWARE) 416 r->linklayer = __detect_linklayer(r, rtab->data); 417 rtab->next = qdisc_rtab_list; 418 qdisc_rtab_list = rtab; 419 } 420 return rtab; 421 } 422 EXPORT_SYMBOL(qdisc_get_rtab); 423 424 void qdisc_put_rtab(struct qdisc_rate_table *tab) 425 { 426 struct qdisc_rate_table *rtab, **rtabp; 427 428 if (!tab || --tab->refcnt) 429 return; 430 431 for (rtabp = &qdisc_rtab_list; 432 (rtab = *rtabp) != NULL; 433 rtabp = &rtab->next) { 434 if (rtab == tab) { 435 *rtabp = rtab->next; 436 kfree(rtab); 437 return; 438 } 439 } 440 } 441 EXPORT_SYMBOL(qdisc_put_rtab); 442 443 static LIST_HEAD(qdisc_stab_list); 444 445 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { 446 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, 447 [TCA_STAB_DATA] = { .type = NLA_BINARY }, 448 }; 449 450 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) 451 { 452 struct nlattr *tb[TCA_STAB_MAX + 1]; 453 struct qdisc_size_table *stab; 454 struct tc_sizespec *s; 455 unsigned int tsize = 0; 456 u16 *tab = NULL; 457 int err; 458 459 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy, NULL); 460 if (err < 0) 461 return ERR_PTR(err); 462 if (!tb[TCA_STAB_BASE]) 463 return ERR_PTR(-EINVAL); 464 465 s = nla_data(tb[TCA_STAB_BASE]); 466 467 if (s->tsize > 0) { 468 if (!tb[TCA_STAB_DATA]) 469 return ERR_PTR(-EINVAL); 470 tab = nla_data(tb[TCA_STAB_DATA]); 471 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); 472 } 473 474 if (tsize != s->tsize || (!tab && tsize > 0)) 475 return ERR_PTR(-EINVAL); 476 477 list_for_each_entry(stab, &qdisc_stab_list, list) { 478 if (memcmp(&stab->szopts, s, sizeof(*s))) 479 continue; 480 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16))) 481 continue; 482 stab->refcnt++; 483 return stab; 484 } 485 486 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); 487 if (!stab) 488 return ERR_PTR(-ENOMEM); 489 490 stab->refcnt = 1; 491 stab->szopts = *s; 492 if (tsize > 0) 493 memcpy(stab->data, tab, tsize * sizeof(u16)); 494 495 list_add_tail(&stab->list, &qdisc_stab_list); 496 497 return stab; 498 } 499 500 static void stab_kfree_rcu(struct rcu_head *head) 501 { 502 kfree(container_of(head, struct qdisc_size_table, rcu)); 503 } 504 505 void qdisc_put_stab(struct qdisc_size_table *tab) 506 { 507 if (!tab) 508 return; 509 510 if (--tab->refcnt == 0) { 511 list_del(&tab->list); 512 call_rcu_bh(&tab->rcu, stab_kfree_rcu); 513 } 514 } 515 EXPORT_SYMBOL(qdisc_put_stab); 516 517 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) 518 { 519 struct nlattr *nest; 520 521 nest = nla_nest_start(skb, TCA_STAB); 522 if (nest == NULL) 523 goto nla_put_failure; 524 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts)) 525 goto nla_put_failure; 526 nla_nest_end(skb, nest); 527 528 return skb->len; 529 530 nla_put_failure: 531 return -1; 532 } 533 534 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 535 const struct qdisc_size_table *stab) 536 { 537 int pkt_len, slot; 538 539 pkt_len = skb->len + stab->szopts.overhead; 540 if (unlikely(!stab->szopts.tsize)) 541 goto out; 542 543 slot = pkt_len + stab->szopts.cell_align; 544 if (unlikely(slot < 0)) 545 slot = 0; 546 547 slot >>= stab->szopts.cell_log; 548 if (likely(slot < stab->szopts.tsize)) 549 pkt_len = stab->data[slot]; 550 else 551 pkt_len = stab->data[stab->szopts.tsize - 1] * 552 (slot / stab->szopts.tsize) + 553 stab->data[slot % stab->szopts.tsize]; 554 555 pkt_len <<= stab->szopts.size_log; 556 out: 557 if (unlikely(pkt_len < 1)) 558 pkt_len = 1; 559 qdisc_skb_cb(skb)->pkt_len = pkt_len; 560 } 561 EXPORT_SYMBOL(__qdisc_calculate_pkt_len); 562 563 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc) 564 { 565 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { 566 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n", 567 txt, qdisc->ops->id, qdisc->handle >> 16); 568 qdisc->flags |= TCQ_F_WARN_NONWC; 569 } 570 } 571 EXPORT_SYMBOL(qdisc_warn_nonwc); 572 573 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) 574 { 575 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, 576 timer); 577 578 rcu_read_lock(); 579 __netif_schedule(qdisc_root(wd->qdisc)); 580 rcu_read_unlock(); 581 582 return HRTIMER_NORESTART; 583 } 584 585 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 586 { 587 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED); 588 wd->timer.function = qdisc_watchdog; 589 wd->qdisc = qdisc; 590 } 591 EXPORT_SYMBOL(qdisc_watchdog_init); 592 593 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires) 594 { 595 if (test_bit(__QDISC_STATE_DEACTIVATED, 596 &qdisc_root_sleeping(wd->qdisc)->state)) 597 return; 598 599 if (wd->last_expires == expires) 600 return; 601 602 wd->last_expires = expires; 603 hrtimer_start(&wd->timer, 604 ns_to_ktime(expires), 605 HRTIMER_MODE_ABS_PINNED); 606 } 607 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns); 608 609 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) 610 { 611 hrtimer_cancel(&wd->timer); 612 } 613 EXPORT_SYMBOL(qdisc_watchdog_cancel); 614 615 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) 616 { 617 struct hlist_head *h; 618 unsigned int i; 619 620 h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL); 621 622 if (h != NULL) { 623 for (i = 0; i < n; i++) 624 INIT_HLIST_HEAD(&h[i]); 625 } 626 return h; 627 } 628 629 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) 630 { 631 struct Qdisc_class_common *cl; 632 struct hlist_node *next; 633 struct hlist_head *nhash, *ohash; 634 unsigned int nsize, nmask, osize; 635 unsigned int i, h; 636 637 /* Rehash when load factor exceeds 0.75 */ 638 if (clhash->hashelems * 4 <= clhash->hashsize * 3) 639 return; 640 nsize = clhash->hashsize * 2; 641 nmask = nsize - 1; 642 nhash = qdisc_class_hash_alloc(nsize); 643 if (nhash == NULL) 644 return; 645 646 ohash = clhash->hash; 647 osize = clhash->hashsize; 648 649 sch_tree_lock(sch); 650 for (i = 0; i < osize; i++) { 651 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) { 652 h = qdisc_class_hash(cl->classid, nmask); 653 hlist_add_head(&cl->hnode, &nhash[h]); 654 } 655 } 656 clhash->hash = nhash; 657 clhash->hashsize = nsize; 658 clhash->hashmask = nmask; 659 sch_tree_unlock(sch); 660 661 kvfree(ohash); 662 } 663 EXPORT_SYMBOL(qdisc_class_hash_grow); 664 665 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) 666 { 667 unsigned int size = 4; 668 669 clhash->hash = qdisc_class_hash_alloc(size); 670 if (clhash->hash == NULL) 671 return -ENOMEM; 672 clhash->hashsize = size; 673 clhash->hashmask = size - 1; 674 clhash->hashelems = 0; 675 return 0; 676 } 677 EXPORT_SYMBOL(qdisc_class_hash_init); 678 679 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) 680 { 681 kvfree(clhash->hash); 682 } 683 EXPORT_SYMBOL(qdisc_class_hash_destroy); 684 685 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, 686 struct Qdisc_class_common *cl) 687 { 688 unsigned int h; 689 690 INIT_HLIST_NODE(&cl->hnode); 691 h = qdisc_class_hash(cl->classid, clhash->hashmask); 692 hlist_add_head(&cl->hnode, &clhash->hash[h]); 693 clhash->hashelems++; 694 } 695 EXPORT_SYMBOL(qdisc_class_hash_insert); 696 697 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, 698 struct Qdisc_class_common *cl) 699 { 700 hlist_del(&cl->hnode); 701 clhash->hashelems--; 702 } 703 EXPORT_SYMBOL(qdisc_class_hash_remove); 704 705 /* Allocate an unique handle from space managed by kernel 706 * Possible range is [8000-FFFF]:0000 (0x8000 values) 707 */ 708 static u32 qdisc_alloc_handle(struct net_device *dev) 709 { 710 int i = 0x8000; 711 static u32 autohandle = TC_H_MAKE(0x80000000U, 0); 712 713 do { 714 autohandle += TC_H_MAKE(0x10000U, 0); 715 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) 716 autohandle = TC_H_MAKE(0x80000000U, 0); 717 if (!qdisc_lookup(dev, autohandle)) 718 return autohandle; 719 cond_resched(); 720 } while (--i > 0); 721 722 return 0; 723 } 724 725 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n, 726 unsigned int len) 727 { 728 const struct Qdisc_class_ops *cops; 729 unsigned long cl; 730 u32 parentid; 731 bool notify; 732 int drops; 733 734 if (n == 0 && len == 0) 735 return; 736 drops = max_t(int, n, 0); 737 rcu_read_lock(); 738 while ((parentid = sch->parent)) { 739 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) 740 break; 741 742 if (sch->flags & TCQ_F_NOPARENT) 743 break; 744 /* Notify parent qdisc only if child qdisc becomes empty. 745 * 746 * If child was empty even before update then backlog 747 * counter is screwed and we skip notification because 748 * parent class is already passive. 749 */ 750 notify = !sch->q.qlen && !WARN_ON_ONCE(!n); 751 /* TODO: perform the search on a per txq basis */ 752 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); 753 if (sch == NULL) { 754 WARN_ON_ONCE(parentid != TC_H_ROOT); 755 break; 756 } 757 cops = sch->ops->cl_ops; 758 if (notify && cops->qlen_notify) { 759 cl = cops->find(sch, parentid); 760 cops->qlen_notify(sch, cl); 761 } 762 sch->q.qlen -= n; 763 sch->qstats.backlog -= len; 764 __qdisc_qstats_drop(sch, drops); 765 } 766 rcu_read_unlock(); 767 } 768 EXPORT_SYMBOL(qdisc_tree_reduce_backlog); 769 770 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, 771 u32 portid, u32 seq, u16 flags, int event) 772 { 773 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL; 774 struct gnet_stats_queue __percpu *cpu_qstats = NULL; 775 struct tcmsg *tcm; 776 struct nlmsghdr *nlh; 777 unsigned char *b = skb_tail_pointer(skb); 778 struct gnet_dump d; 779 struct qdisc_size_table *stab; 780 __u32 qlen; 781 782 cond_resched(); 783 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 784 if (!nlh) 785 goto out_nlmsg_trim; 786 tcm = nlmsg_data(nlh); 787 tcm->tcm_family = AF_UNSPEC; 788 tcm->tcm__pad1 = 0; 789 tcm->tcm__pad2 = 0; 790 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 791 tcm->tcm_parent = clid; 792 tcm->tcm_handle = q->handle; 793 tcm->tcm_info = refcount_read(&q->refcnt); 794 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 795 goto nla_put_failure; 796 if (q->ops->dump && q->ops->dump(q, skb) < 0) 797 goto nla_put_failure; 798 qlen = q->q.qlen; 799 800 stab = rtnl_dereference(q->stab); 801 if (stab && qdisc_dump_stab(skb, stab) < 0) 802 goto nla_put_failure; 803 804 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 805 NULL, &d, TCA_PAD) < 0) 806 goto nla_put_failure; 807 808 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 809 goto nla_put_failure; 810 811 if (qdisc_is_percpu_stats(q)) { 812 cpu_bstats = q->cpu_bstats; 813 cpu_qstats = q->cpu_qstats; 814 } 815 816 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q), 817 &d, cpu_bstats, &q->bstats) < 0 || 818 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || 819 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0) 820 goto nla_put_failure; 821 822 if (gnet_stats_finish_copy(&d) < 0) 823 goto nla_put_failure; 824 825 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 826 return skb->len; 827 828 out_nlmsg_trim: 829 nla_put_failure: 830 nlmsg_trim(skb, b); 831 return -1; 832 } 833 834 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible) 835 { 836 if (q->flags & TCQ_F_BUILTIN) 837 return true; 838 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible) 839 return true; 840 841 return false; 842 } 843 844 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 845 struct nlmsghdr *n, u32 clid, 846 struct Qdisc *old, struct Qdisc *new) 847 { 848 struct sk_buff *skb; 849 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 850 851 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 852 if (!skb) 853 return -ENOBUFS; 854 855 if (old && !tc_qdisc_dump_ignore(old, false)) { 856 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq, 857 0, RTM_DELQDISC) < 0) 858 goto err_out; 859 } 860 if (new && !tc_qdisc_dump_ignore(new, false)) { 861 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq, 862 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) 863 goto err_out; 864 } 865 866 if (skb->len) 867 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 868 n->nlmsg_flags & NLM_F_ECHO); 869 870 err_out: 871 kfree_skb(skb); 872 return -EINVAL; 873 } 874 875 static void notify_and_destroy(struct net *net, struct sk_buff *skb, 876 struct nlmsghdr *n, u32 clid, 877 struct Qdisc *old, struct Qdisc *new) 878 { 879 if (new || old) 880 qdisc_notify(net, skb, n, clid, old, new); 881 882 if (old) 883 qdisc_destroy(old); 884 } 885 886 /* Graft qdisc "new" to class "classid" of qdisc "parent" or 887 * to device "dev". 888 * 889 * When appropriate send a netlink notification using 'skb' 890 * and "n". 891 * 892 * On success, destroy old qdisc. 893 */ 894 895 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 896 struct sk_buff *skb, struct nlmsghdr *n, u32 classid, 897 struct Qdisc *new, struct Qdisc *old) 898 { 899 struct Qdisc *q = old; 900 struct net *net = dev_net(dev); 901 int err = 0; 902 903 if (parent == NULL) { 904 unsigned int i, num_q, ingress; 905 906 ingress = 0; 907 num_q = dev->num_tx_queues; 908 if ((q && q->flags & TCQ_F_INGRESS) || 909 (new && new->flags & TCQ_F_INGRESS)) { 910 num_q = 1; 911 ingress = 1; 912 if (!dev_ingress_queue(dev)) 913 return -ENOENT; 914 } 915 916 if (dev->flags & IFF_UP) 917 dev_deactivate(dev); 918 919 if (new && new->ops->attach) 920 goto skip; 921 922 for (i = 0; i < num_q; i++) { 923 struct netdev_queue *dev_queue = dev_ingress_queue(dev); 924 925 if (!ingress) 926 dev_queue = netdev_get_tx_queue(dev, i); 927 928 old = dev_graft_qdisc(dev_queue, new); 929 if (new && i > 0) 930 qdisc_refcount_inc(new); 931 932 if (!ingress) 933 qdisc_destroy(old); 934 } 935 936 skip: 937 if (!ingress) { 938 notify_and_destroy(net, skb, n, classid, 939 dev->qdisc, new); 940 if (new && !new->ops->attach) 941 qdisc_refcount_inc(new); 942 dev->qdisc = new ? : &noop_qdisc; 943 944 if (new && new->ops->attach) 945 new->ops->attach(new); 946 } else { 947 notify_and_destroy(net, skb, n, classid, old, new); 948 } 949 950 if (dev->flags & IFF_UP) 951 dev_activate(dev); 952 } else { 953 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 954 955 err = -EOPNOTSUPP; 956 if (cops && cops->graft) { 957 unsigned long cl = cops->find(parent, classid); 958 959 if (cl) 960 err = cops->graft(parent, cl, new, &old); 961 else 962 err = -ENOENT; 963 } 964 if (!err) 965 notify_and_destroy(net, skb, n, classid, old, new); 966 } 967 return err; 968 } 969 970 /* lockdep annotation is needed for ingress; egress gets it only for name */ 971 static struct lock_class_key qdisc_tx_lock; 972 static struct lock_class_key qdisc_rx_lock; 973 974 /* 975 Allocate and initialize new qdisc. 976 977 Parameters are passed via opt. 978 */ 979 980 static struct Qdisc *qdisc_create(struct net_device *dev, 981 struct netdev_queue *dev_queue, 982 struct Qdisc *p, u32 parent, u32 handle, 983 struct nlattr **tca, int *errp) 984 { 985 int err; 986 struct nlattr *kind = tca[TCA_KIND]; 987 struct Qdisc *sch; 988 struct Qdisc_ops *ops; 989 struct qdisc_size_table *stab; 990 991 ops = qdisc_lookup_ops(kind); 992 #ifdef CONFIG_MODULES 993 if (ops == NULL && kind != NULL) { 994 char name[IFNAMSIZ]; 995 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 996 /* We dropped the RTNL semaphore in order to 997 * perform the module load. So, even if we 998 * succeeded in loading the module we have to 999 * tell the caller to replay the request. We 1000 * indicate this using -EAGAIN. 1001 * We replay the request because the device may 1002 * go away in the mean time. 1003 */ 1004 rtnl_unlock(); 1005 request_module("sch_%s", name); 1006 rtnl_lock(); 1007 ops = qdisc_lookup_ops(kind); 1008 if (ops != NULL) { 1009 /* We will try again qdisc_lookup_ops, 1010 * so don't keep a reference. 1011 */ 1012 module_put(ops->owner); 1013 err = -EAGAIN; 1014 goto err_out; 1015 } 1016 } 1017 } 1018 #endif 1019 1020 err = -ENOENT; 1021 if (ops == NULL) 1022 goto err_out; 1023 1024 sch = qdisc_alloc(dev_queue, ops); 1025 if (IS_ERR(sch)) { 1026 err = PTR_ERR(sch); 1027 goto err_out2; 1028 } 1029 1030 sch->parent = parent; 1031 1032 if (handle == TC_H_INGRESS) { 1033 sch->flags |= TCQ_F_INGRESS; 1034 handle = TC_H_MAKE(TC_H_INGRESS, 0); 1035 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); 1036 } else { 1037 if (handle == 0) { 1038 handle = qdisc_alloc_handle(dev); 1039 err = -ENOMEM; 1040 if (handle == 0) 1041 goto err_out3; 1042 } 1043 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); 1044 if (!netif_is_multiqueue(dev)) 1045 sch->flags |= TCQ_F_ONETXQUEUE; 1046 } 1047 1048 sch->handle = handle; 1049 1050 /* This exist to keep backward compatible with a userspace 1051 * loophole, what allowed userspace to get IFF_NO_QUEUE 1052 * facility on older kernels by setting tx_queue_len=0 (prior 1053 * to qdisc init), and then forgot to reinit tx_queue_len 1054 * before again attaching a qdisc. 1055 */ 1056 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) { 1057 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 1058 netdev_info(dev, "Caught tx_queue_len zero misconfig\n"); 1059 } 1060 1061 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { 1062 if (qdisc_is_percpu_stats(sch)) { 1063 sch->cpu_bstats = 1064 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu); 1065 if (!sch->cpu_bstats) 1066 goto err_out4; 1067 1068 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue); 1069 if (!sch->cpu_qstats) 1070 goto err_out4; 1071 } 1072 1073 if (tca[TCA_STAB]) { 1074 stab = qdisc_get_stab(tca[TCA_STAB]); 1075 if (IS_ERR(stab)) { 1076 err = PTR_ERR(stab); 1077 goto err_out4; 1078 } 1079 rcu_assign_pointer(sch->stab, stab); 1080 } 1081 if (tca[TCA_RATE]) { 1082 seqcount_t *running; 1083 1084 err = -EOPNOTSUPP; 1085 if (sch->flags & TCQ_F_MQROOT) 1086 goto err_out4; 1087 1088 if ((sch->parent != TC_H_ROOT) && 1089 !(sch->flags & TCQ_F_INGRESS) && 1090 (!p || !(p->flags & TCQ_F_MQROOT))) 1091 running = qdisc_root_sleeping_running(sch); 1092 else 1093 running = &sch->running; 1094 1095 err = gen_new_estimator(&sch->bstats, 1096 sch->cpu_bstats, 1097 &sch->rate_est, 1098 NULL, 1099 running, 1100 tca[TCA_RATE]); 1101 if (err) 1102 goto err_out4; 1103 } 1104 1105 qdisc_hash_add(sch, false); 1106 1107 return sch; 1108 } 1109 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */ 1110 if (ops->destroy) 1111 ops->destroy(sch); 1112 err_out3: 1113 dev_put(dev); 1114 kfree((char *) sch - sch->padded); 1115 err_out2: 1116 module_put(ops->owner); 1117 err_out: 1118 *errp = err; 1119 return NULL; 1120 1121 err_out4: 1122 free_percpu(sch->cpu_bstats); 1123 free_percpu(sch->cpu_qstats); 1124 /* 1125 * Any broken qdiscs that would require a ops->reset() here? 1126 * The qdisc was never in action so it shouldn't be necessary. 1127 */ 1128 qdisc_put_stab(rtnl_dereference(sch->stab)); 1129 if (ops->destroy) 1130 ops->destroy(sch); 1131 goto err_out3; 1132 } 1133 1134 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) 1135 { 1136 struct qdisc_size_table *ostab, *stab = NULL; 1137 int err = 0; 1138 1139 if (tca[TCA_OPTIONS]) { 1140 if (sch->ops->change == NULL) 1141 return -EINVAL; 1142 err = sch->ops->change(sch, tca[TCA_OPTIONS]); 1143 if (err) 1144 return err; 1145 } 1146 1147 if (tca[TCA_STAB]) { 1148 stab = qdisc_get_stab(tca[TCA_STAB]); 1149 if (IS_ERR(stab)) 1150 return PTR_ERR(stab); 1151 } 1152 1153 ostab = rtnl_dereference(sch->stab); 1154 rcu_assign_pointer(sch->stab, stab); 1155 qdisc_put_stab(ostab); 1156 1157 if (tca[TCA_RATE]) { 1158 /* NB: ignores errors from replace_estimator 1159 because change can't be undone. */ 1160 if (sch->flags & TCQ_F_MQROOT) 1161 goto out; 1162 gen_replace_estimator(&sch->bstats, 1163 sch->cpu_bstats, 1164 &sch->rate_est, 1165 NULL, 1166 qdisc_root_sleeping_running(sch), 1167 tca[TCA_RATE]); 1168 } 1169 out: 1170 return 0; 1171 } 1172 1173 struct check_loop_arg { 1174 struct qdisc_walker w; 1175 struct Qdisc *p; 1176 int depth; 1177 }; 1178 1179 static int check_loop_fn(struct Qdisc *q, unsigned long cl, 1180 struct qdisc_walker *w); 1181 1182 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 1183 { 1184 struct check_loop_arg arg; 1185 1186 if (q->ops->cl_ops == NULL) 1187 return 0; 1188 1189 arg.w.stop = arg.w.skip = arg.w.count = 0; 1190 arg.w.fn = check_loop_fn; 1191 arg.depth = depth; 1192 arg.p = p; 1193 q->ops->cl_ops->walk(q, &arg.w); 1194 return arg.w.stop ? -ELOOP : 0; 1195 } 1196 1197 static int 1198 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 1199 { 1200 struct Qdisc *leaf; 1201 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1202 struct check_loop_arg *arg = (struct check_loop_arg *)w; 1203 1204 leaf = cops->leaf(q, cl); 1205 if (leaf) { 1206 if (leaf == arg->p || arg->depth > 7) 1207 return -ELOOP; 1208 return check_loop(leaf, arg->p, arg->depth + 1); 1209 } 1210 return 0; 1211 } 1212 1213 /* 1214 * Delete/get qdisc. 1215 */ 1216 1217 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, 1218 struct netlink_ext_ack *extack) 1219 { 1220 struct net *net = sock_net(skb->sk); 1221 struct tcmsg *tcm = nlmsg_data(n); 1222 struct nlattr *tca[TCA_MAX + 1]; 1223 struct net_device *dev; 1224 u32 clid; 1225 struct Qdisc *q = NULL; 1226 struct Qdisc *p = NULL; 1227 int err; 1228 1229 if ((n->nlmsg_type != RTM_GETQDISC) && 1230 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1231 return -EPERM; 1232 1233 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1234 if (err < 0) 1235 return err; 1236 1237 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1238 if (!dev) 1239 return -ENODEV; 1240 1241 clid = tcm->tcm_parent; 1242 if (clid) { 1243 if (clid != TC_H_ROOT) { 1244 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 1245 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1246 if (!p) 1247 return -ENOENT; 1248 q = qdisc_leaf(p, clid); 1249 } else if (dev_ingress_queue(dev)) { 1250 q = dev_ingress_queue(dev)->qdisc_sleeping; 1251 } 1252 } else { 1253 q = dev->qdisc; 1254 } 1255 if (!q) 1256 return -ENOENT; 1257 1258 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) 1259 return -EINVAL; 1260 } else { 1261 q = qdisc_lookup(dev, tcm->tcm_handle); 1262 if (!q) 1263 return -ENOENT; 1264 } 1265 1266 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1267 return -EINVAL; 1268 1269 if (n->nlmsg_type == RTM_DELQDISC) { 1270 if (!clid) 1271 return -EINVAL; 1272 if (q->handle == 0) 1273 return -ENOENT; 1274 err = qdisc_graft(dev, p, skb, n, clid, NULL, q); 1275 if (err != 0) 1276 return err; 1277 } else { 1278 qdisc_notify(net, skb, n, clid, NULL, q); 1279 } 1280 return 0; 1281 } 1282 1283 /* 1284 * Create/change qdisc. 1285 */ 1286 1287 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, 1288 struct netlink_ext_ack *extack) 1289 { 1290 struct net *net = sock_net(skb->sk); 1291 struct tcmsg *tcm; 1292 struct nlattr *tca[TCA_MAX + 1]; 1293 struct net_device *dev; 1294 u32 clid; 1295 struct Qdisc *q, *p; 1296 int err; 1297 1298 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1299 return -EPERM; 1300 1301 replay: 1302 /* Reinit, just in case something touches this. */ 1303 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1304 if (err < 0) 1305 return err; 1306 1307 tcm = nlmsg_data(n); 1308 clid = tcm->tcm_parent; 1309 q = p = NULL; 1310 1311 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1312 if (!dev) 1313 return -ENODEV; 1314 1315 1316 if (clid) { 1317 if (clid != TC_H_ROOT) { 1318 if (clid != TC_H_INGRESS) { 1319 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1320 if (!p) 1321 return -ENOENT; 1322 q = qdisc_leaf(p, clid); 1323 } else if (dev_ingress_queue_create(dev)) { 1324 q = dev_ingress_queue(dev)->qdisc_sleeping; 1325 } 1326 } else { 1327 q = dev->qdisc; 1328 } 1329 1330 /* It may be default qdisc, ignore it */ 1331 if (q && q->handle == 0) 1332 q = NULL; 1333 1334 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 1335 if (tcm->tcm_handle) { 1336 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) 1337 return -EEXIST; 1338 if (TC_H_MIN(tcm->tcm_handle)) 1339 return -EINVAL; 1340 q = qdisc_lookup(dev, tcm->tcm_handle); 1341 if (!q) 1342 goto create_n_graft; 1343 if (n->nlmsg_flags & NLM_F_EXCL) 1344 return -EEXIST; 1345 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1346 return -EINVAL; 1347 if (q == p || 1348 (p && check_loop(q, p, 0))) 1349 return -ELOOP; 1350 qdisc_refcount_inc(q); 1351 goto graft; 1352 } else { 1353 if (!q) 1354 goto create_n_graft; 1355 1356 /* This magic test requires explanation. 1357 * 1358 * We know, that some child q is already 1359 * attached to this parent and have choice: 1360 * either to change it or to create/graft new one. 1361 * 1362 * 1. We are allowed to create/graft only 1363 * if CREATE and REPLACE flags are set. 1364 * 1365 * 2. If EXCL is set, requestor wanted to say, 1366 * that qdisc tcm_handle is not expected 1367 * to exist, so that we choose create/graft too. 1368 * 1369 * 3. The last case is when no flags are set. 1370 * Alas, it is sort of hole in API, we 1371 * cannot decide what to do unambiguously. 1372 * For now we select create/graft, if 1373 * user gave KIND, which does not match existing. 1374 */ 1375 if ((n->nlmsg_flags & NLM_F_CREATE) && 1376 (n->nlmsg_flags & NLM_F_REPLACE) && 1377 ((n->nlmsg_flags & NLM_F_EXCL) || 1378 (tca[TCA_KIND] && 1379 nla_strcmp(tca[TCA_KIND], q->ops->id)))) 1380 goto create_n_graft; 1381 } 1382 } 1383 } else { 1384 if (!tcm->tcm_handle) 1385 return -EINVAL; 1386 q = qdisc_lookup(dev, tcm->tcm_handle); 1387 } 1388 1389 /* Change qdisc parameters */ 1390 if (q == NULL) 1391 return -ENOENT; 1392 if (n->nlmsg_flags & NLM_F_EXCL) 1393 return -EEXIST; 1394 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1395 return -EINVAL; 1396 err = qdisc_change(q, tca); 1397 if (err == 0) 1398 qdisc_notify(net, skb, n, clid, NULL, q); 1399 return err; 1400 1401 create_n_graft: 1402 if (!(n->nlmsg_flags & NLM_F_CREATE)) 1403 return -ENOENT; 1404 if (clid == TC_H_INGRESS) { 1405 if (dev_ingress_queue(dev)) 1406 q = qdisc_create(dev, dev_ingress_queue(dev), p, 1407 tcm->tcm_parent, tcm->tcm_parent, 1408 tca, &err); 1409 else 1410 err = -ENOENT; 1411 } else { 1412 struct netdev_queue *dev_queue; 1413 1414 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) 1415 dev_queue = p->ops->cl_ops->select_queue(p, tcm); 1416 else if (p) 1417 dev_queue = p->dev_queue; 1418 else 1419 dev_queue = netdev_get_tx_queue(dev, 0); 1420 1421 q = qdisc_create(dev, dev_queue, p, 1422 tcm->tcm_parent, tcm->tcm_handle, 1423 tca, &err); 1424 } 1425 if (q == NULL) { 1426 if (err == -EAGAIN) 1427 goto replay; 1428 return err; 1429 } 1430 1431 graft: 1432 err = qdisc_graft(dev, p, skb, n, clid, q, NULL); 1433 if (err) { 1434 if (q) 1435 qdisc_destroy(q); 1436 return err; 1437 } 1438 1439 return 0; 1440 } 1441 1442 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, 1443 struct netlink_callback *cb, 1444 int *q_idx_p, int s_q_idx, bool recur, 1445 bool dump_invisible) 1446 { 1447 int ret = 0, q_idx = *q_idx_p; 1448 struct Qdisc *q; 1449 int b; 1450 1451 if (!root) 1452 return 0; 1453 1454 q = root; 1455 if (q_idx < s_q_idx) { 1456 q_idx++; 1457 } else { 1458 if (!tc_qdisc_dump_ignore(q, dump_invisible) && 1459 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1460 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1461 RTM_NEWQDISC) <= 0) 1462 goto done; 1463 q_idx++; 1464 } 1465 1466 /* If dumping singletons, there is no qdisc_dev(root) and the singleton 1467 * itself has already been dumped. 1468 * 1469 * If we've already dumped the top-level (ingress) qdisc above and the global 1470 * qdisc hashtable, we don't want to hit it again 1471 */ 1472 if (!qdisc_dev(root) || !recur) 1473 goto out; 1474 1475 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1476 if (q_idx < s_q_idx) { 1477 q_idx++; 1478 continue; 1479 } 1480 if (!tc_qdisc_dump_ignore(q, dump_invisible) && 1481 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1482 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1483 RTM_NEWQDISC) <= 0) 1484 goto done; 1485 q_idx++; 1486 } 1487 1488 out: 1489 *q_idx_p = q_idx; 1490 return ret; 1491 done: 1492 ret = -1; 1493 goto out; 1494 } 1495 1496 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 1497 { 1498 struct net *net = sock_net(skb->sk); 1499 int idx, q_idx; 1500 int s_idx, s_q_idx; 1501 struct net_device *dev; 1502 const struct nlmsghdr *nlh = cb->nlh; 1503 struct nlattr *tca[TCA_MAX + 1]; 1504 int err; 1505 1506 s_idx = cb->args[0]; 1507 s_q_idx = q_idx = cb->args[1]; 1508 1509 idx = 0; 1510 ASSERT_RTNL(); 1511 1512 err = nlmsg_parse(nlh, sizeof(struct tcmsg), tca, TCA_MAX, NULL, NULL); 1513 if (err < 0) 1514 return err; 1515 1516 for_each_netdev(net, dev) { 1517 struct netdev_queue *dev_queue; 1518 1519 if (idx < s_idx) 1520 goto cont; 1521 if (idx > s_idx) 1522 s_q_idx = 0; 1523 q_idx = 0; 1524 1525 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx, 1526 true, tca[TCA_DUMP_INVISIBLE]) < 0) 1527 goto done; 1528 1529 dev_queue = dev_ingress_queue(dev); 1530 if (dev_queue && 1531 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, 1532 &q_idx, s_q_idx, false, 1533 tca[TCA_DUMP_INVISIBLE]) < 0) 1534 goto done; 1535 1536 cont: 1537 idx++; 1538 } 1539 1540 done: 1541 cb->args[0] = idx; 1542 cb->args[1] = q_idx; 1543 1544 return skb->len; 1545 } 1546 1547 1548 1549 /************************************************ 1550 * Traffic classes manipulation. * 1551 ************************************************/ 1552 1553 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1554 unsigned long cl, 1555 u32 portid, u32 seq, u16 flags, int event) 1556 { 1557 struct tcmsg *tcm; 1558 struct nlmsghdr *nlh; 1559 unsigned char *b = skb_tail_pointer(skb); 1560 struct gnet_dump d; 1561 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1562 1563 cond_resched(); 1564 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1565 if (!nlh) 1566 goto out_nlmsg_trim; 1567 tcm = nlmsg_data(nlh); 1568 tcm->tcm_family = AF_UNSPEC; 1569 tcm->tcm__pad1 = 0; 1570 tcm->tcm__pad2 = 0; 1571 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1572 tcm->tcm_parent = q->handle; 1573 tcm->tcm_handle = q->handle; 1574 tcm->tcm_info = 0; 1575 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1576 goto nla_put_failure; 1577 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1578 goto nla_put_failure; 1579 1580 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1581 NULL, &d, TCA_PAD) < 0) 1582 goto nla_put_failure; 1583 1584 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1585 goto nla_put_failure; 1586 1587 if (gnet_stats_finish_copy(&d) < 0) 1588 goto nla_put_failure; 1589 1590 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1591 return skb->len; 1592 1593 out_nlmsg_trim: 1594 nla_put_failure: 1595 nlmsg_trim(skb, b); 1596 return -1; 1597 } 1598 1599 static int tclass_notify(struct net *net, struct sk_buff *oskb, 1600 struct nlmsghdr *n, struct Qdisc *q, 1601 unsigned long cl, int event) 1602 { 1603 struct sk_buff *skb; 1604 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1605 1606 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1607 if (!skb) 1608 return -ENOBUFS; 1609 1610 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) { 1611 kfree_skb(skb); 1612 return -EINVAL; 1613 } 1614 1615 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1616 n->nlmsg_flags & NLM_F_ECHO); 1617 } 1618 1619 static int tclass_del_notify(struct net *net, 1620 const struct Qdisc_class_ops *cops, 1621 struct sk_buff *oskb, struct nlmsghdr *n, 1622 struct Qdisc *q, unsigned long cl) 1623 { 1624 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1625 struct sk_buff *skb; 1626 int err = 0; 1627 1628 if (!cops->delete) 1629 return -EOPNOTSUPP; 1630 1631 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1632 if (!skb) 1633 return -ENOBUFS; 1634 1635 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, 1636 RTM_DELTCLASS) < 0) { 1637 kfree_skb(skb); 1638 return -EINVAL; 1639 } 1640 1641 err = cops->delete(q, cl); 1642 if (err) { 1643 kfree_skb(skb); 1644 return err; 1645 } 1646 1647 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1648 n->nlmsg_flags & NLM_F_ECHO); 1649 } 1650 1651 #ifdef CONFIG_NET_CLS 1652 1653 struct tcf_bind_args { 1654 struct tcf_walker w; 1655 u32 classid; 1656 unsigned long cl; 1657 }; 1658 1659 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg) 1660 { 1661 struct tcf_bind_args *a = (void *)arg; 1662 1663 if (tp->ops->bind_class) { 1664 tcf_tree_lock(tp); 1665 tp->ops->bind_class(n, a->classid, a->cl); 1666 tcf_tree_unlock(tp); 1667 } 1668 return 0; 1669 } 1670 1671 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid, 1672 unsigned long new_cl) 1673 { 1674 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1675 struct tcf_block *block; 1676 struct tcf_chain *chain; 1677 unsigned long cl; 1678 1679 cl = cops->find(q, portid); 1680 if (!cl) 1681 return; 1682 block = cops->tcf_block(q, cl); 1683 if (!block) 1684 return; 1685 list_for_each_entry(chain, &block->chain_list, list) { 1686 struct tcf_proto *tp; 1687 1688 for (tp = rtnl_dereference(chain->filter_chain); 1689 tp; tp = rtnl_dereference(tp->next)) { 1690 struct tcf_bind_args arg = {}; 1691 1692 arg.w.fn = tcf_node_bind; 1693 arg.classid = clid; 1694 arg.cl = new_cl; 1695 tp->ops->walk(tp, &arg.w); 1696 } 1697 } 1698 } 1699 1700 #else 1701 1702 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid, 1703 unsigned long new_cl) 1704 { 1705 } 1706 1707 #endif 1708 1709 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, 1710 struct netlink_ext_ack *extack) 1711 { 1712 struct net *net = sock_net(skb->sk); 1713 struct tcmsg *tcm = nlmsg_data(n); 1714 struct nlattr *tca[TCA_MAX + 1]; 1715 struct net_device *dev; 1716 struct Qdisc *q = NULL; 1717 const struct Qdisc_class_ops *cops; 1718 unsigned long cl = 0; 1719 unsigned long new_cl; 1720 u32 portid; 1721 u32 clid; 1722 u32 qid; 1723 int err; 1724 1725 if ((n->nlmsg_type != RTM_GETTCLASS) && 1726 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1727 return -EPERM; 1728 1729 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1730 if (err < 0) 1731 return err; 1732 1733 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1734 if (!dev) 1735 return -ENODEV; 1736 1737 /* 1738 parent == TC_H_UNSPEC - unspecified parent. 1739 parent == TC_H_ROOT - class is root, which has no parent. 1740 parent == X:0 - parent is root class. 1741 parent == X:Y - parent is a node in hierarchy. 1742 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 1743 1744 handle == 0:0 - generate handle from kernel pool. 1745 handle == 0:Y - class is X:Y, where X:0 is qdisc. 1746 handle == X:Y - clear. 1747 handle == X:0 - root class. 1748 */ 1749 1750 /* Step 1. Determine qdisc handle X:0 */ 1751 1752 portid = tcm->tcm_parent; 1753 clid = tcm->tcm_handle; 1754 qid = TC_H_MAJ(clid); 1755 1756 if (portid != TC_H_ROOT) { 1757 u32 qid1 = TC_H_MAJ(portid); 1758 1759 if (qid && qid1) { 1760 /* If both majors are known, they must be identical. */ 1761 if (qid != qid1) 1762 return -EINVAL; 1763 } else if (qid1) { 1764 qid = qid1; 1765 } else if (qid == 0) 1766 qid = dev->qdisc->handle; 1767 1768 /* Now qid is genuine qdisc handle consistent 1769 * both with parent and child. 1770 * 1771 * TC_H_MAJ(portid) still may be unspecified, complete it now. 1772 */ 1773 if (portid) 1774 portid = TC_H_MAKE(qid, portid); 1775 } else { 1776 if (qid == 0) 1777 qid = dev->qdisc->handle; 1778 } 1779 1780 /* OK. Locate qdisc */ 1781 q = qdisc_lookup(dev, qid); 1782 if (!q) 1783 return -ENOENT; 1784 1785 /* An check that it supports classes */ 1786 cops = q->ops->cl_ops; 1787 if (cops == NULL) 1788 return -EINVAL; 1789 1790 /* Now try to get class */ 1791 if (clid == 0) { 1792 if (portid == TC_H_ROOT) 1793 clid = qid; 1794 } else 1795 clid = TC_H_MAKE(qid, clid); 1796 1797 if (clid) 1798 cl = cops->find(q, clid); 1799 1800 if (cl == 0) { 1801 err = -ENOENT; 1802 if (n->nlmsg_type != RTM_NEWTCLASS || 1803 !(n->nlmsg_flags & NLM_F_CREATE)) 1804 goto out; 1805 } else { 1806 switch (n->nlmsg_type) { 1807 case RTM_NEWTCLASS: 1808 err = -EEXIST; 1809 if (n->nlmsg_flags & NLM_F_EXCL) 1810 goto out; 1811 break; 1812 case RTM_DELTCLASS: 1813 err = tclass_del_notify(net, cops, skb, n, q, cl); 1814 /* Unbind the class with flilters with 0 */ 1815 tc_bind_tclass(q, portid, clid, 0); 1816 goto out; 1817 case RTM_GETTCLASS: 1818 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS); 1819 goto out; 1820 default: 1821 err = -EINVAL; 1822 goto out; 1823 } 1824 } 1825 1826 new_cl = cl; 1827 err = -EOPNOTSUPP; 1828 if (cops->change) 1829 err = cops->change(q, clid, portid, tca, &new_cl); 1830 if (err == 0) { 1831 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS); 1832 /* We just create a new class, need to do reverse binding. */ 1833 if (cl != new_cl) 1834 tc_bind_tclass(q, portid, clid, new_cl); 1835 } 1836 out: 1837 return err; 1838 } 1839 1840 struct qdisc_dump_args { 1841 struct qdisc_walker w; 1842 struct sk_buff *skb; 1843 struct netlink_callback *cb; 1844 }; 1845 1846 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, 1847 struct qdisc_walker *arg) 1848 { 1849 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 1850 1851 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid, 1852 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 1853 RTM_NEWTCLASS); 1854 } 1855 1856 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, 1857 struct tcmsg *tcm, struct netlink_callback *cb, 1858 int *t_p, int s_t) 1859 { 1860 struct qdisc_dump_args arg; 1861 1862 if (tc_qdisc_dump_ignore(q, false) || 1863 *t_p < s_t || !q->ops->cl_ops || 1864 (tcm->tcm_parent && 1865 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 1866 (*t_p)++; 1867 return 0; 1868 } 1869 if (*t_p > s_t) 1870 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 1871 arg.w.fn = qdisc_class_dump; 1872 arg.skb = skb; 1873 arg.cb = cb; 1874 arg.w.stop = 0; 1875 arg.w.skip = cb->args[1]; 1876 arg.w.count = 0; 1877 q->ops->cl_ops->walk(q, &arg.w); 1878 cb->args[1] = arg.w.count; 1879 if (arg.w.stop) 1880 return -1; 1881 (*t_p)++; 1882 return 0; 1883 } 1884 1885 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, 1886 struct tcmsg *tcm, struct netlink_callback *cb, 1887 int *t_p, int s_t) 1888 { 1889 struct Qdisc *q; 1890 int b; 1891 1892 if (!root) 1893 return 0; 1894 1895 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) 1896 return -1; 1897 1898 if (!qdisc_dev(root)) 1899 return 0; 1900 1901 if (tcm->tcm_parent) { 1902 q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent)); 1903 if (q && tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1904 return -1; 1905 return 0; 1906 } 1907 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1908 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1909 return -1; 1910 } 1911 1912 return 0; 1913 } 1914 1915 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 1916 { 1917 struct tcmsg *tcm = nlmsg_data(cb->nlh); 1918 struct net *net = sock_net(skb->sk); 1919 struct netdev_queue *dev_queue; 1920 struct net_device *dev; 1921 int t, s_t; 1922 1923 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 1924 return 0; 1925 dev = dev_get_by_index(net, tcm->tcm_ifindex); 1926 if (!dev) 1927 return 0; 1928 1929 s_t = cb->args[0]; 1930 t = 0; 1931 1932 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) 1933 goto done; 1934 1935 dev_queue = dev_ingress_queue(dev); 1936 if (dev_queue && 1937 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, 1938 &t, s_t) < 0) 1939 goto done; 1940 1941 done: 1942 cb->args[0] = t; 1943 1944 dev_put(dev); 1945 return skb->len; 1946 } 1947 1948 #ifdef CONFIG_PROC_FS 1949 static int psched_show(struct seq_file *seq, void *v) 1950 { 1951 seq_printf(seq, "%08x %08x %08x %08x\n", 1952 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), 1953 1000000, 1954 (u32)NSEC_PER_SEC / hrtimer_resolution); 1955 1956 return 0; 1957 } 1958 1959 static int psched_open(struct inode *inode, struct file *file) 1960 { 1961 return single_open(file, psched_show, NULL); 1962 } 1963 1964 static const struct file_operations psched_fops = { 1965 .owner = THIS_MODULE, 1966 .open = psched_open, 1967 .read = seq_read, 1968 .llseek = seq_lseek, 1969 .release = single_release, 1970 }; 1971 1972 static int __net_init psched_net_init(struct net *net) 1973 { 1974 struct proc_dir_entry *e; 1975 1976 e = proc_create("psched", 0, net->proc_net, &psched_fops); 1977 if (e == NULL) 1978 return -ENOMEM; 1979 1980 return 0; 1981 } 1982 1983 static void __net_exit psched_net_exit(struct net *net) 1984 { 1985 remove_proc_entry("psched", net->proc_net); 1986 } 1987 #else 1988 static int __net_init psched_net_init(struct net *net) 1989 { 1990 return 0; 1991 } 1992 1993 static void __net_exit psched_net_exit(struct net *net) 1994 { 1995 } 1996 #endif 1997 1998 static struct pernet_operations psched_net_ops = { 1999 .init = psched_net_init, 2000 .exit = psched_net_exit, 2001 }; 2002 2003 static int __init pktsched_init(void) 2004 { 2005 int err; 2006 2007 err = register_pernet_subsys(&psched_net_ops); 2008 if (err) { 2009 pr_err("pktsched_init: " 2010 "cannot initialize per netns operations\n"); 2011 return err; 2012 } 2013 2014 register_qdisc(&pfifo_fast_ops); 2015 register_qdisc(&pfifo_qdisc_ops); 2016 register_qdisc(&bfifo_qdisc_ops); 2017 register_qdisc(&pfifo_head_drop_qdisc_ops); 2018 register_qdisc(&mq_qdisc_ops); 2019 register_qdisc(&noqueue_qdisc_ops); 2020 2021 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0); 2022 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0); 2023 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, 2024 0); 2025 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0); 2026 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0); 2027 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, 2028 0); 2029 2030 return 0; 2031 } 2032 2033 subsys_initcall(pktsched_init); 2034