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 if (!handle) 305 return NULL; 306 q = qdisc_match_from_root(dev->qdisc, handle); 307 if (q) 308 goto out; 309 310 if (dev_ingress_queue(dev)) 311 q = qdisc_match_from_root( 312 dev_ingress_queue(dev)->qdisc_sleeping, 313 handle); 314 out: 315 return q; 316 } 317 318 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) 319 { 320 unsigned long cl; 321 struct Qdisc *leaf; 322 const struct Qdisc_class_ops *cops = p->ops->cl_ops; 323 324 if (cops == NULL) 325 return NULL; 326 cl = cops->find(p, classid); 327 328 if (cl == 0) 329 return NULL; 330 leaf = cops->leaf(p, cl); 331 return leaf; 332 } 333 334 /* Find queueing discipline by name */ 335 336 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) 337 { 338 struct Qdisc_ops *q = NULL; 339 340 if (kind) { 341 read_lock(&qdisc_mod_lock); 342 for (q = qdisc_base; q; q = q->next) { 343 if (nla_strcmp(kind, q->id) == 0) { 344 if (!try_module_get(q->owner)) 345 q = NULL; 346 break; 347 } 348 } 349 read_unlock(&qdisc_mod_lock); 350 } 351 return q; 352 } 353 354 /* The linklayer setting were not transferred from iproute2, in older 355 * versions, and the rate tables lookup systems have been dropped in 356 * the kernel. To keep backward compatible with older iproute2 tc 357 * utils, we detect the linklayer setting by detecting if the rate 358 * table were modified. 359 * 360 * For linklayer ATM table entries, the rate table will be aligned to 361 * 48 bytes, thus some table entries will contain the same value. The 362 * mpu (min packet unit) is also encoded into the old rate table, thus 363 * starting from the mpu, we find low and high table entries for 364 * mapping this cell. If these entries contain the same value, when 365 * the rate tables have been modified for linklayer ATM. 366 * 367 * This is done by rounding mpu to the nearest 48 bytes cell/entry, 368 * and then roundup to the next cell, calc the table entry one below, 369 * and compare. 370 */ 371 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab) 372 { 373 int low = roundup(r->mpu, 48); 374 int high = roundup(low+1, 48); 375 int cell_low = low >> r->cell_log; 376 int cell_high = (high >> r->cell_log) - 1; 377 378 /* rtab is too inaccurate at rates > 100Mbit/s */ 379 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) { 380 pr_debug("TC linklayer: Giving up ATM detection\n"); 381 return TC_LINKLAYER_ETHERNET; 382 } 383 384 if ((cell_high > cell_low) && (cell_high < 256) 385 && (rtab[cell_low] == rtab[cell_high])) { 386 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n", 387 cell_low, cell_high, rtab[cell_high]); 388 return TC_LINKLAYER_ATM; 389 } 390 return TC_LINKLAYER_ETHERNET; 391 } 392 393 static struct qdisc_rate_table *qdisc_rtab_list; 394 395 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, 396 struct nlattr *tab) 397 { 398 struct qdisc_rate_table *rtab; 399 400 if (tab == NULL || r->rate == 0 || r->cell_log == 0 || 401 nla_len(tab) != TC_RTAB_SIZE) 402 return NULL; 403 404 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { 405 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) && 406 !memcmp(&rtab->data, nla_data(tab), 1024)) { 407 rtab->refcnt++; 408 return rtab; 409 } 410 } 411 412 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); 413 if (rtab) { 414 rtab->rate = *r; 415 rtab->refcnt = 1; 416 memcpy(rtab->data, nla_data(tab), 1024); 417 if (r->linklayer == TC_LINKLAYER_UNAWARE) 418 r->linklayer = __detect_linklayer(r, rtab->data); 419 rtab->next = qdisc_rtab_list; 420 qdisc_rtab_list = rtab; 421 } 422 return rtab; 423 } 424 EXPORT_SYMBOL(qdisc_get_rtab); 425 426 void qdisc_put_rtab(struct qdisc_rate_table *tab) 427 { 428 struct qdisc_rate_table *rtab, **rtabp; 429 430 if (!tab || --tab->refcnt) 431 return; 432 433 for (rtabp = &qdisc_rtab_list; 434 (rtab = *rtabp) != NULL; 435 rtabp = &rtab->next) { 436 if (rtab == tab) { 437 *rtabp = rtab->next; 438 kfree(rtab); 439 return; 440 } 441 } 442 } 443 EXPORT_SYMBOL(qdisc_put_rtab); 444 445 static LIST_HEAD(qdisc_stab_list); 446 447 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { 448 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, 449 [TCA_STAB_DATA] = { .type = NLA_BINARY }, 450 }; 451 452 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) 453 { 454 struct nlattr *tb[TCA_STAB_MAX + 1]; 455 struct qdisc_size_table *stab; 456 struct tc_sizespec *s; 457 unsigned int tsize = 0; 458 u16 *tab = NULL; 459 int err; 460 461 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy, NULL); 462 if (err < 0) 463 return ERR_PTR(err); 464 if (!tb[TCA_STAB_BASE]) 465 return ERR_PTR(-EINVAL); 466 467 s = nla_data(tb[TCA_STAB_BASE]); 468 469 if (s->tsize > 0) { 470 if (!tb[TCA_STAB_DATA]) 471 return ERR_PTR(-EINVAL); 472 tab = nla_data(tb[TCA_STAB_DATA]); 473 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); 474 } 475 476 if (tsize != s->tsize || (!tab && tsize > 0)) 477 return ERR_PTR(-EINVAL); 478 479 list_for_each_entry(stab, &qdisc_stab_list, list) { 480 if (memcmp(&stab->szopts, s, sizeof(*s))) 481 continue; 482 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16))) 483 continue; 484 stab->refcnt++; 485 return stab; 486 } 487 488 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); 489 if (!stab) 490 return ERR_PTR(-ENOMEM); 491 492 stab->refcnt = 1; 493 stab->szopts = *s; 494 if (tsize > 0) 495 memcpy(stab->data, tab, tsize * sizeof(u16)); 496 497 list_add_tail(&stab->list, &qdisc_stab_list); 498 499 return stab; 500 } 501 502 static void stab_kfree_rcu(struct rcu_head *head) 503 { 504 kfree(container_of(head, struct qdisc_size_table, rcu)); 505 } 506 507 void qdisc_put_stab(struct qdisc_size_table *tab) 508 { 509 if (!tab) 510 return; 511 512 if (--tab->refcnt == 0) { 513 list_del(&tab->list); 514 call_rcu_bh(&tab->rcu, stab_kfree_rcu); 515 } 516 } 517 EXPORT_SYMBOL(qdisc_put_stab); 518 519 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) 520 { 521 struct nlattr *nest; 522 523 nest = nla_nest_start(skb, TCA_STAB); 524 if (nest == NULL) 525 goto nla_put_failure; 526 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts)) 527 goto nla_put_failure; 528 nla_nest_end(skb, nest); 529 530 return skb->len; 531 532 nla_put_failure: 533 return -1; 534 } 535 536 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 537 const struct qdisc_size_table *stab) 538 { 539 int pkt_len, slot; 540 541 pkt_len = skb->len + stab->szopts.overhead; 542 if (unlikely(!stab->szopts.tsize)) 543 goto out; 544 545 slot = pkt_len + stab->szopts.cell_align; 546 if (unlikely(slot < 0)) 547 slot = 0; 548 549 slot >>= stab->szopts.cell_log; 550 if (likely(slot < stab->szopts.tsize)) 551 pkt_len = stab->data[slot]; 552 else 553 pkt_len = stab->data[stab->szopts.tsize - 1] * 554 (slot / stab->szopts.tsize) + 555 stab->data[slot % stab->szopts.tsize]; 556 557 pkt_len <<= stab->szopts.size_log; 558 out: 559 if (unlikely(pkt_len < 1)) 560 pkt_len = 1; 561 qdisc_skb_cb(skb)->pkt_len = pkt_len; 562 } 563 EXPORT_SYMBOL(__qdisc_calculate_pkt_len); 564 565 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc) 566 { 567 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { 568 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n", 569 txt, qdisc->ops->id, qdisc->handle >> 16); 570 qdisc->flags |= TCQ_F_WARN_NONWC; 571 } 572 } 573 EXPORT_SYMBOL(qdisc_warn_nonwc); 574 575 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) 576 { 577 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, 578 timer); 579 580 rcu_read_lock(); 581 __netif_schedule(qdisc_root(wd->qdisc)); 582 rcu_read_unlock(); 583 584 return HRTIMER_NORESTART; 585 } 586 587 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 588 { 589 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED); 590 wd->timer.function = qdisc_watchdog; 591 wd->qdisc = qdisc; 592 } 593 EXPORT_SYMBOL(qdisc_watchdog_init); 594 595 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires) 596 { 597 if (test_bit(__QDISC_STATE_DEACTIVATED, 598 &qdisc_root_sleeping(wd->qdisc)->state)) 599 return; 600 601 if (wd->last_expires == expires) 602 return; 603 604 wd->last_expires = expires; 605 hrtimer_start(&wd->timer, 606 ns_to_ktime(expires), 607 HRTIMER_MODE_ABS_PINNED); 608 } 609 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns); 610 611 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) 612 { 613 hrtimer_cancel(&wd->timer); 614 } 615 EXPORT_SYMBOL(qdisc_watchdog_cancel); 616 617 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) 618 { 619 struct hlist_head *h; 620 unsigned int i; 621 622 h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL); 623 624 if (h != NULL) { 625 for (i = 0; i < n; i++) 626 INIT_HLIST_HEAD(&h[i]); 627 } 628 return h; 629 } 630 631 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) 632 { 633 struct Qdisc_class_common *cl; 634 struct hlist_node *next; 635 struct hlist_head *nhash, *ohash; 636 unsigned int nsize, nmask, osize; 637 unsigned int i, h; 638 639 /* Rehash when load factor exceeds 0.75 */ 640 if (clhash->hashelems * 4 <= clhash->hashsize * 3) 641 return; 642 nsize = clhash->hashsize * 2; 643 nmask = nsize - 1; 644 nhash = qdisc_class_hash_alloc(nsize); 645 if (nhash == NULL) 646 return; 647 648 ohash = clhash->hash; 649 osize = clhash->hashsize; 650 651 sch_tree_lock(sch); 652 for (i = 0; i < osize; i++) { 653 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) { 654 h = qdisc_class_hash(cl->classid, nmask); 655 hlist_add_head(&cl->hnode, &nhash[h]); 656 } 657 } 658 clhash->hash = nhash; 659 clhash->hashsize = nsize; 660 clhash->hashmask = nmask; 661 sch_tree_unlock(sch); 662 663 kvfree(ohash); 664 } 665 EXPORT_SYMBOL(qdisc_class_hash_grow); 666 667 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) 668 { 669 unsigned int size = 4; 670 671 clhash->hash = qdisc_class_hash_alloc(size); 672 if (clhash->hash == NULL) 673 return -ENOMEM; 674 clhash->hashsize = size; 675 clhash->hashmask = size - 1; 676 clhash->hashelems = 0; 677 return 0; 678 } 679 EXPORT_SYMBOL(qdisc_class_hash_init); 680 681 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) 682 { 683 kvfree(clhash->hash); 684 } 685 EXPORT_SYMBOL(qdisc_class_hash_destroy); 686 687 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, 688 struct Qdisc_class_common *cl) 689 { 690 unsigned int h; 691 692 INIT_HLIST_NODE(&cl->hnode); 693 h = qdisc_class_hash(cl->classid, clhash->hashmask); 694 hlist_add_head(&cl->hnode, &clhash->hash[h]); 695 clhash->hashelems++; 696 } 697 EXPORT_SYMBOL(qdisc_class_hash_insert); 698 699 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, 700 struct Qdisc_class_common *cl) 701 { 702 hlist_del(&cl->hnode); 703 clhash->hashelems--; 704 } 705 EXPORT_SYMBOL(qdisc_class_hash_remove); 706 707 /* Allocate an unique handle from space managed by kernel 708 * Possible range is [8000-FFFF]:0000 (0x8000 values) 709 */ 710 static u32 qdisc_alloc_handle(struct net_device *dev) 711 { 712 int i = 0x8000; 713 static u32 autohandle = TC_H_MAKE(0x80000000U, 0); 714 715 do { 716 autohandle += TC_H_MAKE(0x10000U, 0); 717 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) 718 autohandle = TC_H_MAKE(0x80000000U, 0); 719 if (!qdisc_lookup(dev, autohandle)) 720 return autohandle; 721 cond_resched(); 722 } while (--i > 0); 723 724 return 0; 725 } 726 727 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n, 728 unsigned int len) 729 { 730 const struct Qdisc_class_ops *cops; 731 unsigned long cl; 732 u32 parentid; 733 bool notify; 734 int drops; 735 736 if (n == 0 && len == 0) 737 return; 738 drops = max_t(int, n, 0); 739 rcu_read_lock(); 740 while ((parentid = sch->parent)) { 741 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) 742 break; 743 744 if (sch->flags & TCQ_F_NOPARENT) 745 break; 746 /* Notify parent qdisc only if child qdisc becomes empty. 747 * 748 * If child was empty even before update then backlog 749 * counter is screwed and we skip notification because 750 * parent class is already passive. 751 */ 752 notify = !sch->q.qlen && !WARN_ON_ONCE(!n); 753 /* TODO: perform the search on a per txq basis */ 754 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); 755 if (sch == NULL) { 756 WARN_ON_ONCE(parentid != TC_H_ROOT); 757 break; 758 } 759 cops = sch->ops->cl_ops; 760 if (notify && cops->qlen_notify) { 761 cl = cops->find(sch, parentid); 762 cops->qlen_notify(sch, cl); 763 } 764 sch->q.qlen -= n; 765 sch->qstats.backlog -= len; 766 __qdisc_qstats_drop(sch, drops); 767 } 768 rcu_read_unlock(); 769 } 770 EXPORT_SYMBOL(qdisc_tree_reduce_backlog); 771 772 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, 773 u32 portid, u32 seq, u16 flags, int event) 774 { 775 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL; 776 struct gnet_stats_queue __percpu *cpu_qstats = NULL; 777 struct tcmsg *tcm; 778 struct nlmsghdr *nlh; 779 unsigned char *b = skb_tail_pointer(skb); 780 struct gnet_dump d; 781 struct qdisc_size_table *stab; 782 __u32 qlen; 783 784 cond_resched(); 785 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 786 if (!nlh) 787 goto out_nlmsg_trim; 788 tcm = nlmsg_data(nlh); 789 tcm->tcm_family = AF_UNSPEC; 790 tcm->tcm__pad1 = 0; 791 tcm->tcm__pad2 = 0; 792 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 793 tcm->tcm_parent = clid; 794 tcm->tcm_handle = q->handle; 795 tcm->tcm_info = refcount_read(&q->refcnt); 796 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 797 goto nla_put_failure; 798 if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED))) 799 goto nla_put_failure; 800 if (q->ops->dump && q->ops->dump(q, skb) < 0) 801 goto nla_put_failure; 802 qlen = q->q.qlen; 803 804 stab = rtnl_dereference(q->stab); 805 if (stab && qdisc_dump_stab(skb, stab) < 0) 806 goto nla_put_failure; 807 808 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 809 NULL, &d, TCA_PAD) < 0) 810 goto nla_put_failure; 811 812 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 813 goto nla_put_failure; 814 815 if (qdisc_is_percpu_stats(q)) { 816 cpu_bstats = q->cpu_bstats; 817 cpu_qstats = q->cpu_qstats; 818 } 819 820 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q), 821 &d, cpu_bstats, &q->bstats) < 0 || 822 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || 823 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0) 824 goto nla_put_failure; 825 826 if (gnet_stats_finish_copy(&d) < 0) 827 goto nla_put_failure; 828 829 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 830 return skb->len; 831 832 out_nlmsg_trim: 833 nla_put_failure: 834 nlmsg_trim(skb, b); 835 return -1; 836 } 837 838 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible) 839 { 840 if (q->flags & TCQ_F_BUILTIN) 841 return true; 842 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible) 843 return true; 844 845 return false; 846 } 847 848 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 849 struct nlmsghdr *n, u32 clid, 850 struct Qdisc *old, struct Qdisc *new) 851 { 852 struct sk_buff *skb; 853 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 854 855 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 856 if (!skb) 857 return -ENOBUFS; 858 859 if (old && !tc_qdisc_dump_ignore(old, false)) { 860 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq, 861 0, RTM_DELQDISC) < 0) 862 goto err_out; 863 } 864 if (new && !tc_qdisc_dump_ignore(new, false)) { 865 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq, 866 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) 867 goto err_out; 868 } 869 870 if (skb->len) 871 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 872 n->nlmsg_flags & NLM_F_ECHO); 873 874 err_out: 875 kfree_skb(skb); 876 return -EINVAL; 877 } 878 879 static void notify_and_destroy(struct net *net, struct sk_buff *skb, 880 struct nlmsghdr *n, u32 clid, 881 struct Qdisc *old, struct Qdisc *new) 882 { 883 if (new || old) 884 qdisc_notify(net, skb, n, clid, old, new); 885 886 if (old) 887 qdisc_destroy(old); 888 } 889 890 /* Graft qdisc "new" to class "classid" of qdisc "parent" or 891 * to device "dev". 892 * 893 * When appropriate send a netlink notification using 'skb' 894 * and "n". 895 * 896 * On success, destroy old qdisc. 897 */ 898 899 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 900 struct sk_buff *skb, struct nlmsghdr *n, u32 classid, 901 struct Qdisc *new, struct Qdisc *old) 902 { 903 struct Qdisc *q = old; 904 struct net *net = dev_net(dev); 905 int err = 0; 906 907 if (parent == NULL) { 908 unsigned int i, num_q, ingress; 909 910 ingress = 0; 911 num_q = dev->num_tx_queues; 912 if ((q && q->flags & TCQ_F_INGRESS) || 913 (new && new->flags & TCQ_F_INGRESS)) { 914 num_q = 1; 915 ingress = 1; 916 if (!dev_ingress_queue(dev)) 917 return -ENOENT; 918 } 919 920 if (dev->flags & IFF_UP) 921 dev_deactivate(dev); 922 923 if (new && new->ops->attach) 924 goto skip; 925 926 for (i = 0; i < num_q; i++) { 927 struct netdev_queue *dev_queue = dev_ingress_queue(dev); 928 929 if (!ingress) 930 dev_queue = netdev_get_tx_queue(dev, i); 931 932 old = dev_graft_qdisc(dev_queue, new); 933 if (new && i > 0) 934 qdisc_refcount_inc(new); 935 936 if (!ingress) 937 qdisc_destroy(old); 938 } 939 940 skip: 941 if (!ingress) { 942 notify_and_destroy(net, skb, n, classid, 943 dev->qdisc, new); 944 if (new && !new->ops->attach) 945 qdisc_refcount_inc(new); 946 dev->qdisc = new ? : &noop_qdisc; 947 948 if (new && new->ops->attach) 949 new->ops->attach(new); 950 } else { 951 notify_and_destroy(net, skb, n, classid, old, new); 952 } 953 954 if (dev->flags & IFF_UP) 955 dev_activate(dev); 956 } else { 957 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 958 959 err = -EOPNOTSUPP; 960 if (cops && cops->graft) { 961 unsigned long cl = cops->find(parent, classid); 962 963 if (cl) 964 err = cops->graft(parent, cl, new, &old); 965 else 966 err = -ENOENT; 967 } 968 if (!err) 969 notify_and_destroy(net, skb, n, classid, old, new); 970 } 971 return err; 972 } 973 974 /* lockdep annotation is needed for ingress; egress gets it only for name */ 975 static struct lock_class_key qdisc_tx_lock; 976 static struct lock_class_key qdisc_rx_lock; 977 978 /* 979 Allocate and initialize new qdisc. 980 981 Parameters are passed via opt. 982 */ 983 984 static struct Qdisc *qdisc_create(struct net_device *dev, 985 struct netdev_queue *dev_queue, 986 struct Qdisc *p, u32 parent, u32 handle, 987 struct nlattr **tca, int *errp) 988 { 989 int err; 990 struct nlattr *kind = tca[TCA_KIND]; 991 struct Qdisc *sch; 992 struct Qdisc_ops *ops; 993 struct qdisc_size_table *stab; 994 995 ops = qdisc_lookup_ops(kind); 996 #ifdef CONFIG_MODULES 997 if (ops == NULL && kind != NULL) { 998 char name[IFNAMSIZ]; 999 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 1000 /* We dropped the RTNL semaphore in order to 1001 * perform the module load. So, even if we 1002 * succeeded in loading the module we have to 1003 * tell the caller to replay the request. We 1004 * indicate this using -EAGAIN. 1005 * We replay the request because the device may 1006 * go away in the mean time. 1007 */ 1008 rtnl_unlock(); 1009 request_module("sch_%s", name); 1010 rtnl_lock(); 1011 ops = qdisc_lookup_ops(kind); 1012 if (ops != NULL) { 1013 /* We will try again qdisc_lookup_ops, 1014 * so don't keep a reference. 1015 */ 1016 module_put(ops->owner); 1017 err = -EAGAIN; 1018 goto err_out; 1019 } 1020 } 1021 } 1022 #endif 1023 1024 err = -ENOENT; 1025 if (ops == NULL) 1026 goto err_out; 1027 1028 sch = qdisc_alloc(dev_queue, ops); 1029 if (IS_ERR(sch)) { 1030 err = PTR_ERR(sch); 1031 goto err_out2; 1032 } 1033 1034 sch->parent = parent; 1035 1036 if (handle == TC_H_INGRESS) { 1037 sch->flags |= TCQ_F_INGRESS; 1038 handle = TC_H_MAKE(TC_H_INGRESS, 0); 1039 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); 1040 } else { 1041 if (handle == 0) { 1042 handle = qdisc_alloc_handle(dev); 1043 err = -ENOMEM; 1044 if (handle == 0) 1045 goto err_out3; 1046 } 1047 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); 1048 if (!netif_is_multiqueue(dev)) 1049 sch->flags |= TCQ_F_ONETXQUEUE; 1050 } 1051 1052 sch->handle = handle; 1053 1054 /* This exist to keep backward compatible with a userspace 1055 * loophole, what allowed userspace to get IFF_NO_QUEUE 1056 * facility on older kernels by setting tx_queue_len=0 (prior 1057 * to qdisc init), and then forgot to reinit tx_queue_len 1058 * before again attaching a qdisc. 1059 */ 1060 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) { 1061 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 1062 netdev_info(dev, "Caught tx_queue_len zero misconfig\n"); 1063 } 1064 1065 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { 1066 if (tca[TCA_STAB]) { 1067 stab = qdisc_get_stab(tca[TCA_STAB]); 1068 if (IS_ERR(stab)) { 1069 err = PTR_ERR(stab); 1070 goto err_out4; 1071 } 1072 rcu_assign_pointer(sch->stab, stab); 1073 } 1074 if (tca[TCA_RATE]) { 1075 seqcount_t *running; 1076 1077 err = -EOPNOTSUPP; 1078 if (sch->flags & TCQ_F_MQROOT) 1079 goto err_out4; 1080 1081 if ((sch->parent != TC_H_ROOT) && 1082 !(sch->flags & TCQ_F_INGRESS) && 1083 (!p || !(p->flags & TCQ_F_MQROOT))) 1084 running = qdisc_root_sleeping_running(sch); 1085 else 1086 running = &sch->running; 1087 1088 err = gen_new_estimator(&sch->bstats, 1089 sch->cpu_bstats, 1090 &sch->rate_est, 1091 NULL, 1092 running, 1093 tca[TCA_RATE]); 1094 if (err) 1095 goto err_out4; 1096 } 1097 1098 qdisc_hash_add(sch, false); 1099 1100 return sch; 1101 } 1102 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */ 1103 if (ops->destroy) 1104 ops->destroy(sch); 1105 err_out3: 1106 dev_put(dev); 1107 qdisc_free(sch); 1108 err_out2: 1109 module_put(ops->owner); 1110 err_out: 1111 *errp = err; 1112 return NULL; 1113 1114 err_out4: 1115 /* 1116 * Any broken qdiscs that would require a ops->reset() here? 1117 * The qdisc was never in action so it shouldn't be necessary. 1118 */ 1119 qdisc_put_stab(rtnl_dereference(sch->stab)); 1120 if (ops->destroy) 1121 ops->destroy(sch); 1122 goto err_out3; 1123 } 1124 1125 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) 1126 { 1127 struct qdisc_size_table *ostab, *stab = NULL; 1128 int err = 0; 1129 1130 if (tca[TCA_OPTIONS]) { 1131 if (sch->ops->change == NULL) 1132 return -EINVAL; 1133 err = sch->ops->change(sch, tca[TCA_OPTIONS]); 1134 if (err) 1135 return err; 1136 } 1137 1138 if (tca[TCA_STAB]) { 1139 stab = qdisc_get_stab(tca[TCA_STAB]); 1140 if (IS_ERR(stab)) 1141 return PTR_ERR(stab); 1142 } 1143 1144 ostab = rtnl_dereference(sch->stab); 1145 rcu_assign_pointer(sch->stab, stab); 1146 qdisc_put_stab(ostab); 1147 1148 if (tca[TCA_RATE]) { 1149 /* NB: ignores errors from replace_estimator 1150 because change can't be undone. */ 1151 if (sch->flags & TCQ_F_MQROOT) 1152 goto out; 1153 gen_replace_estimator(&sch->bstats, 1154 sch->cpu_bstats, 1155 &sch->rate_est, 1156 NULL, 1157 qdisc_root_sleeping_running(sch), 1158 tca[TCA_RATE]); 1159 } 1160 out: 1161 return 0; 1162 } 1163 1164 struct check_loop_arg { 1165 struct qdisc_walker w; 1166 struct Qdisc *p; 1167 int depth; 1168 }; 1169 1170 static int check_loop_fn(struct Qdisc *q, unsigned long cl, 1171 struct qdisc_walker *w); 1172 1173 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 1174 { 1175 struct check_loop_arg arg; 1176 1177 if (q->ops->cl_ops == NULL) 1178 return 0; 1179 1180 arg.w.stop = arg.w.skip = arg.w.count = 0; 1181 arg.w.fn = check_loop_fn; 1182 arg.depth = depth; 1183 arg.p = p; 1184 q->ops->cl_ops->walk(q, &arg.w); 1185 return arg.w.stop ? -ELOOP : 0; 1186 } 1187 1188 static int 1189 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 1190 { 1191 struct Qdisc *leaf; 1192 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1193 struct check_loop_arg *arg = (struct check_loop_arg *)w; 1194 1195 leaf = cops->leaf(q, cl); 1196 if (leaf) { 1197 if (leaf == arg->p || arg->depth > 7) 1198 return -ELOOP; 1199 return check_loop(leaf, arg->p, arg->depth + 1); 1200 } 1201 return 0; 1202 } 1203 1204 /* 1205 * Delete/get qdisc. 1206 */ 1207 1208 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, 1209 struct netlink_ext_ack *extack) 1210 { 1211 struct net *net = sock_net(skb->sk); 1212 struct tcmsg *tcm = nlmsg_data(n); 1213 struct nlattr *tca[TCA_MAX + 1]; 1214 struct net_device *dev; 1215 u32 clid; 1216 struct Qdisc *q = NULL; 1217 struct Qdisc *p = NULL; 1218 int err; 1219 1220 if ((n->nlmsg_type != RTM_GETQDISC) && 1221 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1222 return -EPERM; 1223 1224 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1225 if (err < 0) 1226 return err; 1227 1228 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1229 if (!dev) 1230 return -ENODEV; 1231 1232 clid = tcm->tcm_parent; 1233 if (clid) { 1234 if (clid != TC_H_ROOT) { 1235 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 1236 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1237 if (!p) 1238 return -ENOENT; 1239 q = qdisc_leaf(p, clid); 1240 } else if (dev_ingress_queue(dev)) { 1241 q = dev_ingress_queue(dev)->qdisc_sleeping; 1242 } 1243 } else { 1244 q = dev->qdisc; 1245 } 1246 if (!q) 1247 return -ENOENT; 1248 1249 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) 1250 return -EINVAL; 1251 } else { 1252 q = qdisc_lookup(dev, tcm->tcm_handle); 1253 if (!q) 1254 return -ENOENT; 1255 } 1256 1257 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1258 return -EINVAL; 1259 1260 if (n->nlmsg_type == RTM_DELQDISC) { 1261 if (!clid) 1262 return -EINVAL; 1263 if (q->handle == 0) 1264 return -ENOENT; 1265 err = qdisc_graft(dev, p, skb, n, clid, NULL, q); 1266 if (err != 0) 1267 return err; 1268 } else { 1269 qdisc_notify(net, skb, n, clid, NULL, q); 1270 } 1271 return 0; 1272 } 1273 1274 /* 1275 * Create/change qdisc. 1276 */ 1277 1278 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, 1279 struct netlink_ext_ack *extack) 1280 { 1281 struct net *net = sock_net(skb->sk); 1282 struct tcmsg *tcm; 1283 struct nlattr *tca[TCA_MAX + 1]; 1284 struct net_device *dev; 1285 u32 clid; 1286 struct Qdisc *q, *p; 1287 int err; 1288 1289 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1290 return -EPERM; 1291 1292 replay: 1293 /* Reinit, just in case something touches this. */ 1294 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1295 if (err < 0) 1296 return err; 1297 1298 tcm = nlmsg_data(n); 1299 clid = tcm->tcm_parent; 1300 q = p = NULL; 1301 1302 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1303 if (!dev) 1304 return -ENODEV; 1305 1306 1307 if (clid) { 1308 if (clid != TC_H_ROOT) { 1309 if (clid != TC_H_INGRESS) { 1310 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1311 if (!p) 1312 return -ENOENT; 1313 q = qdisc_leaf(p, clid); 1314 } else if (dev_ingress_queue_create(dev)) { 1315 q = dev_ingress_queue(dev)->qdisc_sleeping; 1316 } 1317 } else { 1318 q = dev->qdisc; 1319 } 1320 1321 /* It may be default qdisc, ignore it */ 1322 if (q && q->handle == 0) 1323 q = NULL; 1324 1325 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 1326 if (tcm->tcm_handle) { 1327 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) 1328 return -EEXIST; 1329 if (TC_H_MIN(tcm->tcm_handle)) 1330 return -EINVAL; 1331 q = qdisc_lookup(dev, tcm->tcm_handle); 1332 if (!q) 1333 goto create_n_graft; 1334 if (n->nlmsg_flags & NLM_F_EXCL) 1335 return -EEXIST; 1336 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1337 return -EINVAL; 1338 if (q == p || 1339 (p && check_loop(q, p, 0))) 1340 return -ELOOP; 1341 qdisc_refcount_inc(q); 1342 goto graft; 1343 } else { 1344 if (!q) 1345 goto create_n_graft; 1346 1347 /* This magic test requires explanation. 1348 * 1349 * We know, that some child q is already 1350 * attached to this parent and have choice: 1351 * either to change it or to create/graft new one. 1352 * 1353 * 1. We are allowed to create/graft only 1354 * if CREATE and REPLACE flags are set. 1355 * 1356 * 2. If EXCL is set, requestor wanted to say, 1357 * that qdisc tcm_handle is not expected 1358 * to exist, so that we choose create/graft too. 1359 * 1360 * 3. The last case is when no flags are set. 1361 * Alas, it is sort of hole in API, we 1362 * cannot decide what to do unambiguously. 1363 * For now we select create/graft, if 1364 * user gave KIND, which does not match existing. 1365 */ 1366 if ((n->nlmsg_flags & NLM_F_CREATE) && 1367 (n->nlmsg_flags & NLM_F_REPLACE) && 1368 ((n->nlmsg_flags & NLM_F_EXCL) || 1369 (tca[TCA_KIND] && 1370 nla_strcmp(tca[TCA_KIND], q->ops->id)))) 1371 goto create_n_graft; 1372 } 1373 } 1374 } else { 1375 if (!tcm->tcm_handle) 1376 return -EINVAL; 1377 q = qdisc_lookup(dev, tcm->tcm_handle); 1378 } 1379 1380 /* Change qdisc parameters */ 1381 if (q == NULL) 1382 return -ENOENT; 1383 if (n->nlmsg_flags & NLM_F_EXCL) 1384 return -EEXIST; 1385 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1386 return -EINVAL; 1387 err = qdisc_change(q, tca); 1388 if (err == 0) 1389 qdisc_notify(net, skb, n, clid, NULL, q); 1390 return err; 1391 1392 create_n_graft: 1393 if (!(n->nlmsg_flags & NLM_F_CREATE)) 1394 return -ENOENT; 1395 if (clid == TC_H_INGRESS) { 1396 if (dev_ingress_queue(dev)) 1397 q = qdisc_create(dev, dev_ingress_queue(dev), p, 1398 tcm->tcm_parent, tcm->tcm_parent, 1399 tca, &err); 1400 else 1401 err = -ENOENT; 1402 } else { 1403 struct netdev_queue *dev_queue; 1404 1405 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) 1406 dev_queue = p->ops->cl_ops->select_queue(p, tcm); 1407 else if (p) 1408 dev_queue = p->dev_queue; 1409 else 1410 dev_queue = netdev_get_tx_queue(dev, 0); 1411 1412 q = qdisc_create(dev, dev_queue, p, 1413 tcm->tcm_parent, tcm->tcm_handle, 1414 tca, &err); 1415 } 1416 if (q == NULL) { 1417 if (err == -EAGAIN) 1418 goto replay; 1419 return err; 1420 } 1421 1422 graft: 1423 err = qdisc_graft(dev, p, skb, n, clid, q, NULL); 1424 if (err) { 1425 if (q) 1426 qdisc_destroy(q); 1427 return err; 1428 } 1429 1430 return 0; 1431 } 1432 1433 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, 1434 struct netlink_callback *cb, 1435 int *q_idx_p, int s_q_idx, bool recur, 1436 bool dump_invisible) 1437 { 1438 int ret = 0, q_idx = *q_idx_p; 1439 struct Qdisc *q; 1440 int b; 1441 1442 if (!root) 1443 return 0; 1444 1445 q = root; 1446 if (q_idx < s_q_idx) { 1447 q_idx++; 1448 } else { 1449 if (!tc_qdisc_dump_ignore(q, dump_invisible) && 1450 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1451 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1452 RTM_NEWQDISC) <= 0) 1453 goto done; 1454 q_idx++; 1455 } 1456 1457 /* If dumping singletons, there is no qdisc_dev(root) and the singleton 1458 * itself has already been dumped. 1459 * 1460 * If we've already dumped the top-level (ingress) qdisc above and the global 1461 * qdisc hashtable, we don't want to hit it again 1462 */ 1463 if (!qdisc_dev(root) || !recur) 1464 goto out; 1465 1466 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1467 if (q_idx < s_q_idx) { 1468 q_idx++; 1469 continue; 1470 } 1471 if (!tc_qdisc_dump_ignore(q, dump_invisible) && 1472 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1473 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1474 RTM_NEWQDISC) <= 0) 1475 goto done; 1476 q_idx++; 1477 } 1478 1479 out: 1480 *q_idx_p = q_idx; 1481 return ret; 1482 done: 1483 ret = -1; 1484 goto out; 1485 } 1486 1487 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 1488 { 1489 struct net *net = sock_net(skb->sk); 1490 int idx, q_idx; 1491 int s_idx, s_q_idx; 1492 struct net_device *dev; 1493 const struct nlmsghdr *nlh = cb->nlh; 1494 struct nlattr *tca[TCA_MAX + 1]; 1495 int err; 1496 1497 s_idx = cb->args[0]; 1498 s_q_idx = q_idx = cb->args[1]; 1499 1500 idx = 0; 1501 ASSERT_RTNL(); 1502 1503 err = nlmsg_parse(nlh, sizeof(struct tcmsg), tca, TCA_MAX, NULL, NULL); 1504 if (err < 0) 1505 return err; 1506 1507 for_each_netdev(net, dev) { 1508 struct netdev_queue *dev_queue; 1509 1510 if (idx < s_idx) 1511 goto cont; 1512 if (idx > s_idx) 1513 s_q_idx = 0; 1514 q_idx = 0; 1515 1516 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx, 1517 true, tca[TCA_DUMP_INVISIBLE]) < 0) 1518 goto done; 1519 1520 dev_queue = dev_ingress_queue(dev); 1521 if (dev_queue && 1522 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, 1523 &q_idx, s_q_idx, false, 1524 tca[TCA_DUMP_INVISIBLE]) < 0) 1525 goto done; 1526 1527 cont: 1528 idx++; 1529 } 1530 1531 done: 1532 cb->args[0] = idx; 1533 cb->args[1] = q_idx; 1534 1535 return skb->len; 1536 } 1537 1538 1539 1540 /************************************************ 1541 * Traffic classes manipulation. * 1542 ************************************************/ 1543 1544 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1545 unsigned long cl, 1546 u32 portid, u32 seq, u16 flags, int event) 1547 { 1548 struct tcmsg *tcm; 1549 struct nlmsghdr *nlh; 1550 unsigned char *b = skb_tail_pointer(skb); 1551 struct gnet_dump d; 1552 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1553 1554 cond_resched(); 1555 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1556 if (!nlh) 1557 goto out_nlmsg_trim; 1558 tcm = nlmsg_data(nlh); 1559 tcm->tcm_family = AF_UNSPEC; 1560 tcm->tcm__pad1 = 0; 1561 tcm->tcm__pad2 = 0; 1562 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1563 tcm->tcm_parent = q->handle; 1564 tcm->tcm_handle = q->handle; 1565 tcm->tcm_info = 0; 1566 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1567 goto nla_put_failure; 1568 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1569 goto nla_put_failure; 1570 1571 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1572 NULL, &d, TCA_PAD) < 0) 1573 goto nla_put_failure; 1574 1575 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1576 goto nla_put_failure; 1577 1578 if (gnet_stats_finish_copy(&d) < 0) 1579 goto nla_put_failure; 1580 1581 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1582 return skb->len; 1583 1584 out_nlmsg_trim: 1585 nla_put_failure: 1586 nlmsg_trim(skb, b); 1587 return -1; 1588 } 1589 1590 static int tclass_notify(struct net *net, struct sk_buff *oskb, 1591 struct nlmsghdr *n, struct Qdisc *q, 1592 unsigned long cl, int event) 1593 { 1594 struct sk_buff *skb; 1595 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1596 1597 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1598 if (!skb) 1599 return -ENOBUFS; 1600 1601 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) { 1602 kfree_skb(skb); 1603 return -EINVAL; 1604 } 1605 1606 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1607 n->nlmsg_flags & NLM_F_ECHO); 1608 } 1609 1610 static int tclass_del_notify(struct net *net, 1611 const struct Qdisc_class_ops *cops, 1612 struct sk_buff *oskb, struct nlmsghdr *n, 1613 struct Qdisc *q, unsigned long cl) 1614 { 1615 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1616 struct sk_buff *skb; 1617 int err = 0; 1618 1619 if (!cops->delete) 1620 return -EOPNOTSUPP; 1621 1622 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1623 if (!skb) 1624 return -ENOBUFS; 1625 1626 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, 1627 RTM_DELTCLASS) < 0) { 1628 kfree_skb(skb); 1629 return -EINVAL; 1630 } 1631 1632 err = cops->delete(q, cl); 1633 if (err) { 1634 kfree_skb(skb); 1635 return err; 1636 } 1637 1638 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1639 n->nlmsg_flags & NLM_F_ECHO); 1640 } 1641 1642 #ifdef CONFIG_NET_CLS 1643 1644 struct tcf_bind_args { 1645 struct tcf_walker w; 1646 u32 classid; 1647 unsigned long cl; 1648 }; 1649 1650 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg) 1651 { 1652 struct tcf_bind_args *a = (void *)arg; 1653 1654 if (tp->ops->bind_class) { 1655 struct Qdisc *q = tcf_block_q(tp->chain->block); 1656 1657 sch_tree_lock(q); 1658 tp->ops->bind_class(n, a->classid, a->cl); 1659 sch_tree_unlock(q); 1660 } 1661 return 0; 1662 } 1663 1664 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid, 1665 unsigned long new_cl) 1666 { 1667 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1668 struct tcf_block *block; 1669 struct tcf_chain *chain; 1670 unsigned long cl; 1671 1672 cl = cops->find(q, portid); 1673 if (!cl) 1674 return; 1675 block = cops->tcf_block(q, cl); 1676 if (!block) 1677 return; 1678 list_for_each_entry(chain, &block->chain_list, list) { 1679 struct tcf_proto *tp; 1680 1681 for (tp = rtnl_dereference(chain->filter_chain); 1682 tp; tp = rtnl_dereference(tp->next)) { 1683 struct tcf_bind_args arg = {}; 1684 1685 arg.w.fn = tcf_node_bind; 1686 arg.classid = clid; 1687 arg.cl = new_cl; 1688 tp->ops->walk(tp, &arg.w); 1689 } 1690 } 1691 } 1692 1693 #else 1694 1695 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid, 1696 unsigned long new_cl) 1697 { 1698 } 1699 1700 #endif 1701 1702 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, 1703 struct netlink_ext_ack *extack) 1704 { 1705 struct net *net = sock_net(skb->sk); 1706 struct tcmsg *tcm = nlmsg_data(n); 1707 struct nlattr *tca[TCA_MAX + 1]; 1708 struct net_device *dev; 1709 struct Qdisc *q = NULL; 1710 const struct Qdisc_class_ops *cops; 1711 unsigned long cl = 0; 1712 unsigned long new_cl; 1713 u32 portid; 1714 u32 clid; 1715 u32 qid; 1716 int err; 1717 1718 if ((n->nlmsg_type != RTM_GETTCLASS) && 1719 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1720 return -EPERM; 1721 1722 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1723 if (err < 0) 1724 return err; 1725 1726 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1727 if (!dev) 1728 return -ENODEV; 1729 1730 /* 1731 parent == TC_H_UNSPEC - unspecified parent. 1732 parent == TC_H_ROOT - class is root, which has no parent. 1733 parent == X:0 - parent is root class. 1734 parent == X:Y - parent is a node in hierarchy. 1735 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 1736 1737 handle == 0:0 - generate handle from kernel pool. 1738 handle == 0:Y - class is X:Y, where X:0 is qdisc. 1739 handle == X:Y - clear. 1740 handle == X:0 - root class. 1741 */ 1742 1743 /* Step 1. Determine qdisc handle X:0 */ 1744 1745 portid = tcm->tcm_parent; 1746 clid = tcm->tcm_handle; 1747 qid = TC_H_MAJ(clid); 1748 1749 if (portid != TC_H_ROOT) { 1750 u32 qid1 = TC_H_MAJ(portid); 1751 1752 if (qid && qid1) { 1753 /* If both majors are known, they must be identical. */ 1754 if (qid != qid1) 1755 return -EINVAL; 1756 } else if (qid1) { 1757 qid = qid1; 1758 } else if (qid == 0) 1759 qid = dev->qdisc->handle; 1760 1761 /* Now qid is genuine qdisc handle consistent 1762 * both with parent and child. 1763 * 1764 * TC_H_MAJ(portid) still may be unspecified, complete it now. 1765 */ 1766 if (portid) 1767 portid = TC_H_MAKE(qid, portid); 1768 } else { 1769 if (qid == 0) 1770 qid = dev->qdisc->handle; 1771 } 1772 1773 /* OK. Locate qdisc */ 1774 q = qdisc_lookup(dev, qid); 1775 if (!q) 1776 return -ENOENT; 1777 1778 /* An check that it supports classes */ 1779 cops = q->ops->cl_ops; 1780 if (cops == NULL) 1781 return -EINVAL; 1782 1783 /* Now try to get class */ 1784 if (clid == 0) { 1785 if (portid == TC_H_ROOT) 1786 clid = qid; 1787 } else 1788 clid = TC_H_MAKE(qid, clid); 1789 1790 if (clid) 1791 cl = cops->find(q, clid); 1792 1793 if (cl == 0) { 1794 err = -ENOENT; 1795 if (n->nlmsg_type != RTM_NEWTCLASS || 1796 !(n->nlmsg_flags & NLM_F_CREATE)) 1797 goto out; 1798 } else { 1799 switch (n->nlmsg_type) { 1800 case RTM_NEWTCLASS: 1801 err = -EEXIST; 1802 if (n->nlmsg_flags & NLM_F_EXCL) 1803 goto out; 1804 break; 1805 case RTM_DELTCLASS: 1806 err = tclass_del_notify(net, cops, skb, n, q, cl); 1807 /* Unbind the class with flilters with 0 */ 1808 tc_bind_tclass(q, portid, clid, 0); 1809 goto out; 1810 case RTM_GETTCLASS: 1811 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS); 1812 goto out; 1813 default: 1814 err = -EINVAL; 1815 goto out; 1816 } 1817 } 1818 1819 new_cl = cl; 1820 err = -EOPNOTSUPP; 1821 if (cops->change) 1822 err = cops->change(q, clid, portid, tca, &new_cl); 1823 if (err == 0) { 1824 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS); 1825 /* We just create a new class, need to do reverse binding. */ 1826 if (cl != new_cl) 1827 tc_bind_tclass(q, portid, clid, new_cl); 1828 } 1829 out: 1830 return err; 1831 } 1832 1833 struct qdisc_dump_args { 1834 struct qdisc_walker w; 1835 struct sk_buff *skb; 1836 struct netlink_callback *cb; 1837 }; 1838 1839 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, 1840 struct qdisc_walker *arg) 1841 { 1842 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 1843 1844 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid, 1845 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 1846 RTM_NEWTCLASS); 1847 } 1848 1849 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, 1850 struct tcmsg *tcm, struct netlink_callback *cb, 1851 int *t_p, int s_t) 1852 { 1853 struct qdisc_dump_args arg; 1854 1855 if (tc_qdisc_dump_ignore(q, false) || 1856 *t_p < s_t || !q->ops->cl_ops || 1857 (tcm->tcm_parent && 1858 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 1859 (*t_p)++; 1860 return 0; 1861 } 1862 if (*t_p > s_t) 1863 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 1864 arg.w.fn = qdisc_class_dump; 1865 arg.skb = skb; 1866 arg.cb = cb; 1867 arg.w.stop = 0; 1868 arg.w.skip = cb->args[1]; 1869 arg.w.count = 0; 1870 q->ops->cl_ops->walk(q, &arg.w); 1871 cb->args[1] = arg.w.count; 1872 if (arg.w.stop) 1873 return -1; 1874 (*t_p)++; 1875 return 0; 1876 } 1877 1878 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, 1879 struct tcmsg *tcm, struct netlink_callback *cb, 1880 int *t_p, int s_t) 1881 { 1882 struct Qdisc *q; 1883 int b; 1884 1885 if (!root) 1886 return 0; 1887 1888 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) 1889 return -1; 1890 1891 if (!qdisc_dev(root)) 1892 return 0; 1893 1894 if (tcm->tcm_parent) { 1895 q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent)); 1896 if (q && tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1897 return -1; 1898 return 0; 1899 } 1900 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1901 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1902 return -1; 1903 } 1904 1905 return 0; 1906 } 1907 1908 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 1909 { 1910 struct tcmsg *tcm = nlmsg_data(cb->nlh); 1911 struct net *net = sock_net(skb->sk); 1912 struct netdev_queue *dev_queue; 1913 struct net_device *dev; 1914 int t, s_t; 1915 1916 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 1917 return 0; 1918 dev = dev_get_by_index(net, tcm->tcm_ifindex); 1919 if (!dev) 1920 return 0; 1921 1922 s_t = cb->args[0]; 1923 t = 0; 1924 1925 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) 1926 goto done; 1927 1928 dev_queue = dev_ingress_queue(dev); 1929 if (dev_queue && 1930 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, 1931 &t, s_t) < 0) 1932 goto done; 1933 1934 done: 1935 cb->args[0] = t; 1936 1937 dev_put(dev); 1938 return skb->len; 1939 } 1940 1941 #ifdef CONFIG_PROC_FS 1942 static int psched_show(struct seq_file *seq, void *v) 1943 { 1944 seq_printf(seq, "%08x %08x %08x %08x\n", 1945 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), 1946 1000000, 1947 (u32)NSEC_PER_SEC / hrtimer_resolution); 1948 1949 return 0; 1950 } 1951 1952 static int psched_open(struct inode *inode, struct file *file) 1953 { 1954 return single_open(file, psched_show, NULL); 1955 } 1956 1957 static const struct file_operations psched_fops = { 1958 .owner = THIS_MODULE, 1959 .open = psched_open, 1960 .read = seq_read, 1961 .llseek = seq_lseek, 1962 .release = single_release, 1963 }; 1964 1965 static int __net_init psched_net_init(struct net *net) 1966 { 1967 struct proc_dir_entry *e; 1968 1969 e = proc_create("psched", 0, net->proc_net, &psched_fops); 1970 if (e == NULL) 1971 return -ENOMEM; 1972 1973 return 0; 1974 } 1975 1976 static void __net_exit psched_net_exit(struct net *net) 1977 { 1978 remove_proc_entry("psched", net->proc_net); 1979 } 1980 #else 1981 static int __net_init psched_net_init(struct net *net) 1982 { 1983 return 0; 1984 } 1985 1986 static void __net_exit psched_net_exit(struct net *net) 1987 { 1988 } 1989 #endif 1990 1991 static struct pernet_operations psched_net_ops = { 1992 .init = psched_net_init, 1993 .exit = psched_net_exit, 1994 }; 1995 1996 static int __init pktsched_init(void) 1997 { 1998 int err; 1999 2000 err = register_pernet_subsys(&psched_net_ops); 2001 if (err) { 2002 pr_err("pktsched_init: " 2003 "cannot initialize per netns operations\n"); 2004 return err; 2005 } 2006 2007 register_qdisc(&pfifo_fast_ops); 2008 register_qdisc(&pfifo_qdisc_ops); 2009 register_qdisc(&bfifo_qdisc_ops); 2010 register_qdisc(&pfifo_head_drop_qdisc_ops); 2011 register_qdisc(&mq_qdisc_ops); 2012 register_qdisc(&noqueue_qdisc_ops); 2013 2014 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0); 2015 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0); 2016 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, 2017 0); 2018 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0); 2019 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0); 2020 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, 2021 0); 2022 2023 return 0; 2024 } 2025 2026 subsys_initcall(pktsched_init); 2027