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 (q->ops->dump && q->ops->dump(q, skb) < 0) 799 goto nla_put_failure; 800 qlen = q->q.qlen; 801 802 stab = rtnl_dereference(q->stab); 803 if (stab && qdisc_dump_stab(skb, stab) < 0) 804 goto nla_put_failure; 805 806 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 807 NULL, &d, TCA_PAD) < 0) 808 goto nla_put_failure; 809 810 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 811 goto nla_put_failure; 812 813 if (qdisc_is_percpu_stats(q)) { 814 cpu_bstats = q->cpu_bstats; 815 cpu_qstats = q->cpu_qstats; 816 } 817 818 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q), 819 &d, cpu_bstats, &q->bstats) < 0 || 820 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || 821 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0) 822 goto nla_put_failure; 823 824 if (gnet_stats_finish_copy(&d) < 0) 825 goto nla_put_failure; 826 827 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 828 return skb->len; 829 830 out_nlmsg_trim: 831 nla_put_failure: 832 nlmsg_trim(skb, b); 833 return -1; 834 } 835 836 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible) 837 { 838 if (q->flags & TCQ_F_BUILTIN) 839 return true; 840 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible) 841 return true; 842 843 return false; 844 } 845 846 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 847 struct nlmsghdr *n, u32 clid, 848 struct Qdisc *old, struct Qdisc *new) 849 { 850 struct sk_buff *skb; 851 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 852 853 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 854 if (!skb) 855 return -ENOBUFS; 856 857 if (old && !tc_qdisc_dump_ignore(old, false)) { 858 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq, 859 0, RTM_DELQDISC) < 0) 860 goto err_out; 861 } 862 if (new && !tc_qdisc_dump_ignore(new, false)) { 863 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq, 864 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) 865 goto err_out; 866 } 867 868 if (skb->len) 869 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 870 n->nlmsg_flags & NLM_F_ECHO); 871 872 err_out: 873 kfree_skb(skb); 874 return -EINVAL; 875 } 876 877 static void notify_and_destroy(struct net *net, struct sk_buff *skb, 878 struct nlmsghdr *n, u32 clid, 879 struct Qdisc *old, struct Qdisc *new) 880 { 881 if (new || old) 882 qdisc_notify(net, skb, n, clid, old, new); 883 884 if (old) 885 qdisc_destroy(old); 886 } 887 888 /* Graft qdisc "new" to class "classid" of qdisc "parent" or 889 * to device "dev". 890 * 891 * When appropriate send a netlink notification using 'skb' 892 * and "n". 893 * 894 * On success, destroy old qdisc. 895 */ 896 897 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 898 struct sk_buff *skb, struct nlmsghdr *n, u32 classid, 899 struct Qdisc *new, struct Qdisc *old) 900 { 901 struct Qdisc *q = old; 902 struct net *net = dev_net(dev); 903 int err = 0; 904 905 if (parent == NULL) { 906 unsigned int i, num_q, ingress; 907 908 ingress = 0; 909 num_q = dev->num_tx_queues; 910 if ((q && q->flags & TCQ_F_INGRESS) || 911 (new && new->flags & TCQ_F_INGRESS)) { 912 num_q = 1; 913 ingress = 1; 914 if (!dev_ingress_queue(dev)) 915 return -ENOENT; 916 } 917 918 if (dev->flags & IFF_UP) 919 dev_deactivate(dev); 920 921 if (new && new->ops->attach) 922 goto skip; 923 924 for (i = 0; i < num_q; i++) { 925 struct netdev_queue *dev_queue = dev_ingress_queue(dev); 926 927 if (!ingress) 928 dev_queue = netdev_get_tx_queue(dev, i); 929 930 old = dev_graft_qdisc(dev_queue, new); 931 if (new && i > 0) 932 qdisc_refcount_inc(new); 933 934 if (!ingress) 935 qdisc_destroy(old); 936 } 937 938 skip: 939 if (!ingress) { 940 notify_and_destroy(net, skb, n, classid, 941 dev->qdisc, new); 942 if (new && !new->ops->attach) 943 qdisc_refcount_inc(new); 944 dev->qdisc = new ? : &noop_qdisc; 945 946 if (new && new->ops->attach) 947 new->ops->attach(new); 948 } else { 949 notify_and_destroy(net, skb, n, classid, old, new); 950 } 951 952 if (dev->flags & IFF_UP) 953 dev_activate(dev); 954 } else { 955 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 956 957 err = -EOPNOTSUPP; 958 if (cops && cops->graft) { 959 unsigned long cl = cops->find(parent, classid); 960 961 if (cl) 962 err = cops->graft(parent, cl, new, &old); 963 else 964 err = -ENOENT; 965 } 966 if (!err) 967 notify_and_destroy(net, skb, n, classid, old, new); 968 } 969 return err; 970 } 971 972 /* lockdep annotation is needed for ingress; egress gets it only for name */ 973 static struct lock_class_key qdisc_tx_lock; 974 static struct lock_class_key qdisc_rx_lock; 975 976 /* 977 Allocate and initialize new qdisc. 978 979 Parameters are passed via opt. 980 */ 981 982 static struct Qdisc *qdisc_create(struct net_device *dev, 983 struct netdev_queue *dev_queue, 984 struct Qdisc *p, u32 parent, u32 handle, 985 struct nlattr **tca, int *errp) 986 { 987 int err; 988 struct nlattr *kind = tca[TCA_KIND]; 989 struct Qdisc *sch; 990 struct Qdisc_ops *ops; 991 struct qdisc_size_table *stab; 992 993 ops = qdisc_lookup_ops(kind); 994 #ifdef CONFIG_MODULES 995 if (ops == NULL && kind != NULL) { 996 char name[IFNAMSIZ]; 997 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 998 /* We dropped the RTNL semaphore in order to 999 * perform the module load. So, even if we 1000 * succeeded in loading the module we have to 1001 * tell the caller to replay the request. We 1002 * indicate this using -EAGAIN. 1003 * We replay the request because the device may 1004 * go away in the mean time. 1005 */ 1006 rtnl_unlock(); 1007 request_module("sch_%s", name); 1008 rtnl_lock(); 1009 ops = qdisc_lookup_ops(kind); 1010 if (ops != NULL) { 1011 /* We will try again qdisc_lookup_ops, 1012 * so don't keep a reference. 1013 */ 1014 module_put(ops->owner); 1015 err = -EAGAIN; 1016 goto err_out; 1017 } 1018 } 1019 } 1020 #endif 1021 1022 err = -ENOENT; 1023 if (ops == NULL) 1024 goto err_out; 1025 1026 sch = qdisc_alloc(dev_queue, ops); 1027 if (IS_ERR(sch)) { 1028 err = PTR_ERR(sch); 1029 goto err_out2; 1030 } 1031 1032 sch->parent = parent; 1033 1034 if (handle == TC_H_INGRESS) { 1035 sch->flags |= TCQ_F_INGRESS; 1036 handle = TC_H_MAKE(TC_H_INGRESS, 0); 1037 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); 1038 } else { 1039 if (handle == 0) { 1040 handle = qdisc_alloc_handle(dev); 1041 err = -ENOMEM; 1042 if (handle == 0) 1043 goto err_out3; 1044 } 1045 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); 1046 if (!netif_is_multiqueue(dev)) 1047 sch->flags |= TCQ_F_ONETXQUEUE; 1048 } 1049 1050 sch->handle = handle; 1051 1052 /* This exist to keep backward compatible with a userspace 1053 * loophole, what allowed userspace to get IFF_NO_QUEUE 1054 * facility on older kernels by setting tx_queue_len=0 (prior 1055 * to qdisc init), and then forgot to reinit tx_queue_len 1056 * before again attaching a qdisc. 1057 */ 1058 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) { 1059 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 1060 netdev_info(dev, "Caught tx_queue_len zero misconfig\n"); 1061 } 1062 1063 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { 1064 if (qdisc_is_percpu_stats(sch)) { 1065 sch->cpu_bstats = 1066 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu); 1067 if (!sch->cpu_bstats) 1068 goto err_out4; 1069 1070 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue); 1071 if (!sch->cpu_qstats) 1072 goto err_out4; 1073 } 1074 1075 if (tca[TCA_STAB]) { 1076 stab = qdisc_get_stab(tca[TCA_STAB]); 1077 if (IS_ERR(stab)) { 1078 err = PTR_ERR(stab); 1079 goto err_out4; 1080 } 1081 rcu_assign_pointer(sch->stab, stab); 1082 } 1083 if (tca[TCA_RATE]) { 1084 seqcount_t *running; 1085 1086 err = -EOPNOTSUPP; 1087 if (sch->flags & TCQ_F_MQROOT) 1088 goto err_out4; 1089 1090 if ((sch->parent != TC_H_ROOT) && 1091 !(sch->flags & TCQ_F_INGRESS) && 1092 (!p || !(p->flags & TCQ_F_MQROOT))) 1093 running = qdisc_root_sleeping_running(sch); 1094 else 1095 running = &sch->running; 1096 1097 err = gen_new_estimator(&sch->bstats, 1098 sch->cpu_bstats, 1099 &sch->rate_est, 1100 NULL, 1101 running, 1102 tca[TCA_RATE]); 1103 if (err) 1104 goto err_out4; 1105 } 1106 1107 qdisc_hash_add(sch, false); 1108 1109 return sch; 1110 } 1111 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */ 1112 if (ops->destroy) 1113 ops->destroy(sch); 1114 err_out3: 1115 dev_put(dev); 1116 kfree((char *) sch - sch->padded); 1117 err_out2: 1118 module_put(ops->owner); 1119 err_out: 1120 *errp = err; 1121 return NULL; 1122 1123 err_out4: 1124 free_percpu(sch->cpu_bstats); 1125 free_percpu(sch->cpu_qstats); 1126 /* 1127 * Any broken qdiscs that would require a ops->reset() here? 1128 * The qdisc was never in action so it shouldn't be necessary. 1129 */ 1130 qdisc_put_stab(rtnl_dereference(sch->stab)); 1131 if (ops->destroy) 1132 ops->destroy(sch); 1133 goto err_out3; 1134 } 1135 1136 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) 1137 { 1138 struct qdisc_size_table *ostab, *stab = NULL; 1139 int err = 0; 1140 1141 if (tca[TCA_OPTIONS]) { 1142 if (sch->ops->change == NULL) 1143 return -EINVAL; 1144 err = sch->ops->change(sch, tca[TCA_OPTIONS]); 1145 if (err) 1146 return err; 1147 } 1148 1149 if (tca[TCA_STAB]) { 1150 stab = qdisc_get_stab(tca[TCA_STAB]); 1151 if (IS_ERR(stab)) 1152 return PTR_ERR(stab); 1153 } 1154 1155 ostab = rtnl_dereference(sch->stab); 1156 rcu_assign_pointer(sch->stab, stab); 1157 qdisc_put_stab(ostab); 1158 1159 if (tca[TCA_RATE]) { 1160 /* NB: ignores errors from replace_estimator 1161 because change can't be undone. */ 1162 if (sch->flags & TCQ_F_MQROOT) 1163 goto out; 1164 gen_replace_estimator(&sch->bstats, 1165 sch->cpu_bstats, 1166 &sch->rate_est, 1167 NULL, 1168 qdisc_root_sleeping_running(sch), 1169 tca[TCA_RATE]); 1170 } 1171 out: 1172 return 0; 1173 } 1174 1175 struct check_loop_arg { 1176 struct qdisc_walker w; 1177 struct Qdisc *p; 1178 int depth; 1179 }; 1180 1181 static int check_loop_fn(struct Qdisc *q, unsigned long cl, 1182 struct qdisc_walker *w); 1183 1184 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 1185 { 1186 struct check_loop_arg arg; 1187 1188 if (q->ops->cl_ops == NULL) 1189 return 0; 1190 1191 arg.w.stop = arg.w.skip = arg.w.count = 0; 1192 arg.w.fn = check_loop_fn; 1193 arg.depth = depth; 1194 arg.p = p; 1195 q->ops->cl_ops->walk(q, &arg.w); 1196 return arg.w.stop ? -ELOOP : 0; 1197 } 1198 1199 static int 1200 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 1201 { 1202 struct Qdisc *leaf; 1203 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1204 struct check_loop_arg *arg = (struct check_loop_arg *)w; 1205 1206 leaf = cops->leaf(q, cl); 1207 if (leaf) { 1208 if (leaf == arg->p || arg->depth > 7) 1209 return -ELOOP; 1210 return check_loop(leaf, arg->p, arg->depth + 1); 1211 } 1212 return 0; 1213 } 1214 1215 /* 1216 * Delete/get qdisc. 1217 */ 1218 1219 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, 1220 struct netlink_ext_ack *extack) 1221 { 1222 struct net *net = sock_net(skb->sk); 1223 struct tcmsg *tcm = nlmsg_data(n); 1224 struct nlattr *tca[TCA_MAX + 1]; 1225 struct net_device *dev; 1226 u32 clid; 1227 struct Qdisc *q = NULL; 1228 struct Qdisc *p = NULL; 1229 int err; 1230 1231 if ((n->nlmsg_type != RTM_GETQDISC) && 1232 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1233 return -EPERM; 1234 1235 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1236 if (err < 0) 1237 return err; 1238 1239 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1240 if (!dev) 1241 return -ENODEV; 1242 1243 clid = tcm->tcm_parent; 1244 if (clid) { 1245 if (clid != TC_H_ROOT) { 1246 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 1247 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1248 if (!p) 1249 return -ENOENT; 1250 q = qdisc_leaf(p, clid); 1251 } else if (dev_ingress_queue(dev)) { 1252 q = dev_ingress_queue(dev)->qdisc_sleeping; 1253 } 1254 } else { 1255 q = dev->qdisc; 1256 } 1257 if (!q) 1258 return -ENOENT; 1259 1260 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) 1261 return -EINVAL; 1262 } else { 1263 q = qdisc_lookup(dev, tcm->tcm_handle); 1264 if (!q) 1265 return -ENOENT; 1266 } 1267 1268 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1269 return -EINVAL; 1270 1271 if (n->nlmsg_type == RTM_DELQDISC) { 1272 if (!clid) 1273 return -EINVAL; 1274 if (q->handle == 0) 1275 return -ENOENT; 1276 err = qdisc_graft(dev, p, skb, n, clid, NULL, q); 1277 if (err != 0) 1278 return err; 1279 } else { 1280 qdisc_notify(net, skb, n, clid, NULL, q); 1281 } 1282 return 0; 1283 } 1284 1285 /* 1286 * Create/change qdisc. 1287 */ 1288 1289 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, 1290 struct netlink_ext_ack *extack) 1291 { 1292 struct net *net = sock_net(skb->sk); 1293 struct tcmsg *tcm; 1294 struct nlattr *tca[TCA_MAX + 1]; 1295 struct net_device *dev; 1296 u32 clid; 1297 struct Qdisc *q, *p; 1298 int err; 1299 1300 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1301 return -EPERM; 1302 1303 replay: 1304 /* Reinit, just in case something touches this. */ 1305 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1306 if (err < 0) 1307 return err; 1308 1309 tcm = nlmsg_data(n); 1310 clid = tcm->tcm_parent; 1311 q = p = NULL; 1312 1313 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1314 if (!dev) 1315 return -ENODEV; 1316 1317 1318 if (clid) { 1319 if (clid != TC_H_ROOT) { 1320 if (clid != TC_H_INGRESS) { 1321 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1322 if (!p) 1323 return -ENOENT; 1324 q = qdisc_leaf(p, clid); 1325 } else if (dev_ingress_queue_create(dev)) { 1326 q = dev_ingress_queue(dev)->qdisc_sleeping; 1327 } 1328 } else { 1329 q = dev->qdisc; 1330 } 1331 1332 /* It may be default qdisc, ignore it */ 1333 if (q && q->handle == 0) 1334 q = NULL; 1335 1336 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 1337 if (tcm->tcm_handle) { 1338 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) 1339 return -EEXIST; 1340 if (TC_H_MIN(tcm->tcm_handle)) 1341 return -EINVAL; 1342 q = qdisc_lookup(dev, tcm->tcm_handle); 1343 if (!q) 1344 goto create_n_graft; 1345 if (n->nlmsg_flags & NLM_F_EXCL) 1346 return -EEXIST; 1347 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1348 return -EINVAL; 1349 if (q == p || 1350 (p && check_loop(q, p, 0))) 1351 return -ELOOP; 1352 qdisc_refcount_inc(q); 1353 goto graft; 1354 } else { 1355 if (!q) 1356 goto create_n_graft; 1357 1358 /* This magic test requires explanation. 1359 * 1360 * We know, that some child q is already 1361 * attached to this parent and have choice: 1362 * either to change it or to create/graft new one. 1363 * 1364 * 1. We are allowed to create/graft only 1365 * if CREATE and REPLACE flags are set. 1366 * 1367 * 2. If EXCL is set, requestor wanted to say, 1368 * that qdisc tcm_handle is not expected 1369 * to exist, so that we choose create/graft too. 1370 * 1371 * 3. The last case is when no flags are set. 1372 * Alas, it is sort of hole in API, we 1373 * cannot decide what to do unambiguously. 1374 * For now we select create/graft, if 1375 * user gave KIND, which does not match existing. 1376 */ 1377 if ((n->nlmsg_flags & NLM_F_CREATE) && 1378 (n->nlmsg_flags & NLM_F_REPLACE) && 1379 ((n->nlmsg_flags & NLM_F_EXCL) || 1380 (tca[TCA_KIND] && 1381 nla_strcmp(tca[TCA_KIND], q->ops->id)))) 1382 goto create_n_graft; 1383 } 1384 } 1385 } else { 1386 if (!tcm->tcm_handle) 1387 return -EINVAL; 1388 q = qdisc_lookup(dev, tcm->tcm_handle); 1389 } 1390 1391 /* Change qdisc parameters */ 1392 if (q == NULL) 1393 return -ENOENT; 1394 if (n->nlmsg_flags & NLM_F_EXCL) 1395 return -EEXIST; 1396 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1397 return -EINVAL; 1398 err = qdisc_change(q, tca); 1399 if (err == 0) 1400 qdisc_notify(net, skb, n, clid, NULL, q); 1401 return err; 1402 1403 create_n_graft: 1404 if (!(n->nlmsg_flags & NLM_F_CREATE)) 1405 return -ENOENT; 1406 if (clid == TC_H_INGRESS) { 1407 if (dev_ingress_queue(dev)) 1408 q = qdisc_create(dev, dev_ingress_queue(dev), p, 1409 tcm->tcm_parent, tcm->tcm_parent, 1410 tca, &err); 1411 else 1412 err = -ENOENT; 1413 } else { 1414 struct netdev_queue *dev_queue; 1415 1416 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) 1417 dev_queue = p->ops->cl_ops->select_queue(p, tcm); 1418 else if (p) 1419 dev_queue = p->dev_queue; 1420 else 1421 dev_queue = netdev_get_tx_queue(dev, 0); 1422 1423 q = qdisc_create(dev, dev_queue, p, 1424 tcm->tcm_parent, tcm->tcm_handle, 1425 tca, &err); 1426 } 1427 if (q == NULL) { 1428 if (err == -EAGAIN) 1429 goto replay; 1430 return err; 1431 } 1432 1433 graft: 1434 err = qdisc_graft(dev, p, skb, n, clid, q, NULL); 1435 if (err) { 1436 if (q) 1437 qdisc_destroy(q); 1438 return err; 1439 } 1440 1441 return 0; 1442 } 1443 1444 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, 1445 struct netlink_callback *cb, 1446 int *q_idx_p, int s_q_idx, bool recur, 1447 bool dump_invisible) 1448 { 1449 int ret = 0, q_idx = *q_idx_p; 1450 struct Qdisc *q; 1451 int b; 1452 1453 if (!root) 1454 return 0; 1455 1456 q = root; 1457 if (q_idx < s_q_idx) { 1458 q_idx++; 1459 } else { 1460 if (!tc_qdisc_dump_ignore(q, dump_invisible) && 1461 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1462 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1463 RTM_NEWQDISC) <= 0) 1464 goto done; 1465 q_idx++; 1466 } 1467 1468 /* If dumping singletons, there is no qdisc_dev(root) and the singleton 1469 * itself has already been dumped. 1470 * 1471 * If we've already dumped the top-level (ingress) qdisc above and the global 1472 * qdisc hashtable, we don't want to hit it again 1473 */ 1474 if (!qdisc_dev(root) || !recur) 1475 goto out; 1476 1477 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1478 if (q_idx < s_q_idx) { 1479 q_idx++; 1480 continue; 1481 } 1482 if (!tc_qdisc_dump_ignore(q, dump_invisible) && 1483 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1484 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1485 RTM_NEWQDISC) <= 0) 1486 goto done; 1487 q_idx++; 1488 } 1489 1490 out: 1491 *q_idx_p = q_idx; 1492 return ret; 1493 done: 1494 ret = -1; 1495 goto out; 1496 } 1497 1498 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 1499 { 1500 struct net *net = sock_net(skb->sk); 1501 int idx, q_idx; 1502 int s_idx, s_q_idx; 1503 struct net_device *dev; 1504 const struct nlmsghdr *nlh = cb->nlh; 1505 struct nlattr *tca[TCA_MAX + 1]; 1506 int err; 1507 1508 s_idx = cb->args[0]; 1509 s_q_idx = q_idx = cb->args[1]; 1510 1511 idx = 0; 1512 ASSERT_RTNL(); 1513 1514 err = nlmsg_parse(nlh, sizeof(struct tcmsg), tca, TCA_MAX, NULL, NULL); 1515 if (err < 0) 1516 return err; 1517 1518 for_each_netdev(net, dev) { 1519 struct netdev_queue *dev_queue; 1520 1521 if (idx < s_idx) 1522 goto cont; 1523 if (idx > s_idx) 1524 s_q_idx = 0; 1525 q_idx = 0; 1526 1527 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx, 1528 true, tca[TCA_DUMP_INVISIBLE]) < 0) 1529 goto done; 1530 1531 dev_queue = dev_ingress_queue(dev); 1532 if (dev_queue && 1533 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, 1534 &q_idx, s_q_idx, false, 1535 tca[TCA_DUMP_INVISIBLE]) < 0) 1536 goto done; 1537 1538 cont: 1539 idx++; 1540 } 1541 1542 done: 1543 cb->args[0] = idx; 1544 cb->args[1] = q_idx; 1545 1546 return skb->len; 1547 } 1548 1549 1550 1551 /************************************************ 1552 * Traffic classes manipulation. * 1553 ************************************************/ 1554 1555 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1556 unsigned long cl, 1557 u32 portid, u32 seq, u16 flags, int event) 1558 { 1559 struct tcmsg *tcm; 1560 struct nlmsghdr *nlh; 1561 unsigned char *b = skb_tail_pointer(skb); 1562 struct gnet_dump d; 1563 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1564 1565 cond_resched(); 1566 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1567 if (!nlh) 1568 goto out_nlmsg_trim; 1569 tcm = nlmsg_data(nlh); 1570 tcm->tcm_family = AF_UNSPEC; 1571 tcm->tcm__pad1 = 0; 1572 tcm->tcm__pad2 = 0; 1573 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1574 tcm->tcm_parent = q->handle; 1575 tcm->tcm_handle = q->handle; 1576 tcm->tcm_info = 0; 1577 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1578 goto nla_put_failure; 1579 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1580 goto nla_put_failure; 1581 1582 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1583 NULL, &d, TCA_PAD) < 0) 1584 goto nla_put_failure; 1585 1586 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1587 goto nla_put_failure; 1588 1589 if (gnet_stats_finish_copy(&d) < 0) 1590 goto nla_put_failure; 1591 1592 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1593 return skb->len; 1594 1595 out_nlmsg_trim: 1596 nla_put_failure: 1597 nlmsg_trim(skb, b); 1598 return -1; 1599 } 1600 1601 static int tclass_notify(struct net *net, struct sk_buff *oskb, 1602 struct nlmsghdr *n, struct Qdisc *q, 1603 unsigned long cl, int event) 1604 { 1605 struct sk_buff *skb; 1606 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1607 1608 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1609 if (!skb) 1610 return -ENOBUFS; 1611 1612 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) { 1613 kfree_skb(skb); 1614 return -EINVAL; 1615 } 1616 1617 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1618 n->nlmsg_flags & NLM_F_ECHO); 1619 } 1620 1621 static int tclass_del_notify(struct net *net, 1622 const struct Qdisc_class_ops *cops, 1623 struct sk_buff *oskb, struct nlmsghdr *n, 1624 struct Qdisc *q, unsigned long cl) 1625 { 1626 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1627 struct sk_buff *skb; 1628 int err = 0; 1629 1630 if (!cops->delete) 1631 return -EOPNOTSUPP; 1632 1633 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1634 if (!skb) 1635 return -ENOBUFS; 1636 1637 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, 1638 RTM_DELTCLASS) < 0) { 1639 kfree_skb(skb); 1640 return -EINVAL; 1641 } 1642 1643 err = cops->delete(q, cl); 1644 if (err) { 1645 kfree_skb(skb); 1646 return err; 1647 } 1648 1649 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1650 n->nlmsg_flags & NLM_F_ECHO); 1651 } 1652 1653 #ifdef CONFIG_NET_CLS 1654 1655 struct tcf_bind_args { 1656 struct tcf_walker w; 1657 u32 classid; 1658 unsigned long cl; 1659 }; 1660 1661 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg) 1662 { 1663 struct tcf_bind_args *a = (void *)arg; 1664 1665 if (tp->ops->bind_class) { 1666 struct Qdisc *q = tcf_block_q(tp->chain->block); 1667 1668 sch_tree_lock(q); 1669 tp->ops->bind_class(n, a->classid, a->cl); 1670 sch_tree_unlock(q); 1671 } 1672 return 0; 1673 } 1674 1675 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid, 1676 unsigned long new_cl) 1677 { 1678 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1679 struct tcf_block *block; 1680 struct tcf_chain *chain; 1681 unsigned long cl; 1682 1683 cl = cops->find(q, portid); 1684 if (!cl) 1685 return; 1686 block = cops->tcf_block(q, cl); 1687 if (!block) 1688 return; 1689 list_for_each_entry(chain, &block->chain_list, list) { 1690 struct tcf_proto *tp; 1691 1692 for (tp = rtnl_dereference(chain->filter_chain); 1693 tp; tp = rtnl_dereference(tp->next)) { 1694 struct tcf_bind_args arg = {}; 1695 1696 arg.w.fn = tcf_node_bind; 1697 arg.classid = clid; 1698 arg.cl = new_cl; 1699 tp->ops->walk(tp, &arg.w); 1700 } 1701 } 1702 } 1703 1704 #else 1705 1706 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid, 1707 unsigned long new_cl) 1708 { 1709 } 1710 1711 #endif 1712 1713 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, 1714 struct netlink_ext_ack *extack) 1715 { 1716 struct net *net = sock_net(skb->sk); 1717 struct tcmsg *tcm = nlmsg_data(n); 1718 struct nlattr *tca[TCA_MAX + 1]; 1719 struct net_device *dev; 1720 struct Qdisc *q = NULL; 1721 const struct Qdisc_class_ops *cops; 1722 unsigned long cl = 0; 1723 unsigned long new_cl; 1724 u32 portid; 1725 u32 clid; 1726 u32 qid; 1727 int err; 1728 1729 if ((n->nlmsg_type != RTM_GETTCLASS) && 1730 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1731 return -EPERM; 1732 1733 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1734 if (err < 0) 1735 return err; 1736 1737 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1738 if (!dev) 1739 return -ENODEV; 1740 1741 /* 1742 parent == TC_H_UNSPEC - unspecified parent. 1743 parent == TC_H_ROOT - class is root, which has no parent. 1744 parent == X:0 - parent is root class. 1745 parent == X:Y - parent is a node in hierarchy. 1746 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 1747 1748 handle == 0:0 - generate handle from kernel pool. 1749 handle == 0:Y - class is X:Y, where X:0 is qdisc. 1750 handle == X:Y - clear. 1751 handle == X:0 - root class. 1752 */ 1753 1754 /* Step 1. Determine qdisc handle X:0 */ 1755 1756 portid = tcm->tcm_parent; 1757 clid = tcm->tcm_handle; 1758 qid = TC_H_MAJ(clid); 1759 1760 if (portid != TC_H_ROOT) { 1761 u32 qid1 = TC_H_MAJ(portid); 1762 1763 if (qid && qid1) { 1764 /* If both majors are known, they must be identical. */ 1765 if (qid != qid1) 1766 return -EINVAL; 1767 } else if (qid1) { 1768 qid = qid1; 1769 } else if (qid == 0) 1770 qid = dev->qdisc->handle; 1771 1772 /* Now qid is genuine qdisc handle consistent 1773 * both with parent and child. 1774 * 1775 * TC_H_MAJ(portid) still may be unspecified, complete it now. 1776 */ 1777 if (portid) 1778 portid = TC_H_MAKE(qid, portid); 1779 } else { 1780 if (qid == 0) 1781 qid = dev->qdisc->handle; 1782 } 1783 1784 /* OK. Locate qdisc */ 1785 q = qdisc_lookup(dev, qid); 1786 if (!q) 1787 return -ENOENT; 1788 1789 /* An check that it supports classes */ 1790 cops = q->ops->cl_ops; 1791 if (cops == NULL) 1792 return -EINVAL; 1793 1794 /* Now try to get class */ 1795 if (clid == 0) { 1796 if (portid == TC_H_ROOT) 1797 clid = qid; 1798 } else 1799 clid = TC_H_MAKE(qid, clid); 1800 1801 if (clid) 1802 cl = cops->find(q, clid); 1803 1804 if (cl == 0) { 1805 err = -ENOENT; 1806 if (n->nlmsg_type != RTM_NEWTCLASS || 1807 !(n->nlmsg_flags & NLM_F_CREATE)) 1808 goto out; 1809 } else { 1810 switch (n->nlmsg_type) { 1811 case RTM_NEWTCLASS: 1812 err = -EEXIST; 1813 if (n->nlmsg_flags & NLM_F_EXCL) 1814 goto out; 1815 break; 1816 case RTM_DELTCLASS: 1817 err = tclass_del_notify(net, cops, skb, n, q, cl); 1818 /* Unbind the class with flilters with 0 */ 1819 tc_bind_tclass(q, portid, clid, 0); 1820 goto out; 1821 case RTM_GETTCLASS: 1822 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS); 1823 goto out; 1824 default: 1825 err = -EINVAL; 1826 goto out; 1827 } 1828 } 1829 1830 new_cl = cl; 1831 err = -EOPNOTSUPP; 1832 if (cops->change) 1833 err = cops->change(q, clid, portid, tca, &new_cl); 1834 if (err == 0) { 1835 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS); 1836 /* We just create a new class, need to do reverse binding. */ 1837 if (cl != new_cl) 1838 tc_bind_tclass(q, portid, clid, new_cl); 1839 } 1840 out: 1841 return err; 1842 } 1843 1844 struct qdisc_dump_args { 1845 struct qdisc_walker w; 1846 struct sk_buff *skb; 1847 struct netlink_callback *cb; 1848 }; 1849 1850 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, 1851 struct qdisc_walker *arg) 1852 { 1853 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 1854 1855 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid, 1856 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 1857 RTM_NEWTCLASS); 1858 } 1859 1860 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, 1861 struct tcmsg *tcm, struct netlink_callback *cb, 1862 int *t_p, int s_t) 1863 { 1864 struct qdisc_dump_args arg; 1865 1866 if (tc_qdisc_dump_ignore(q, false) || 1867 *t_p < s_t || !q->ops->cl_ops || 1868 (tcm->tcm_parent && 1869 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 1870 (*t_p)++; 1871 return 0; 1872 } 1873 if (*t_p > s_t) 1874 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 1875 arg.w.fn = qdisc_class_dump; 1876 arg.skb = skb; 1877 arg.cb = cb; 1878 arg.w.stop = 0; 1879 arg.w.skip = cb->args[1]; 1880 arg.w.count = 0; 1881 q->ops->cl_ops->walk(q, &arg.w); 1882 cb->args[1] = arg.w.count; 1883 if (arg.w.stop) 1884 return -1; 1885 (*t_p)++; 1886 return 0; 1887 } 1888 1889 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, 1890 struct tcmsg *tcm, struct netlink_callback *cb, 1891 int *t_p, int s_t) 1892 { 1893 struct Qdisc *q; 1894 int b; 1895 1896 if (!root) 1897 return 0; 1898 1899 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) 1900 return -1; 1901 1902 if (!qdisc_dev(root)) 1903 return 0; 1904 1905 if (tcm->tcm_parent) { 1906 q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent)); 1907 if (q && tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1908 return -1; 1909 return 0; 1910 } 1911 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1912 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1913 return -1; 1914 } 1915 1916 return 0; 1917 } 1918 1919 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 1920 { 1921 struct tcmsg *tcm = nlmsg_data(cb->nlh); 1922 struct net *net = sock_net(skb->sk); 1923 struct netdev_queue *dev_queue; 1924 struct net_device *dev; 1925 int t, s_t; 1926 1927 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 1928 return 0; 1929 dev = dev_get_by_index(net, tcm->tcm_ifindex); 1930 if (!dev) 1931 return 0; 1932 1933 s_t = cb->args[0]; 1934 t = 0; 1935 1936 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) 1937 goto done; 1938 1939 dev_queue = dev_ingress_queue(dev); 1940 if (dev_queue && 1941 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, 1942 &t, s_t) < 0) 1943 goto done; 1944 1945 done: 1946 cb->args[0] = t; 1947 1948 dev_put(dev); 1949 return skb->len; 1950 } 1951 1952 #ifdef CONFIG_PROC_FS 1953 static int psched_show(struct seq_file *seq, void *v) 1954 { 1955 seq_printf(seq, "%08x %08x %08x %08x\n", 1956 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), 1957 1000000, 1958 (u32)NSEC_PER_SEC / hrtimer_resolution); 1959 1960 return 0; 1961 } 1962 1963 static int psched_open(struct inode *inode, struct file *file) 1964 { 1965 return single_open(file, psched_show, NULL); 1966 } 1967 1968 static const struct file_operations psched_fops = { 1969 .owner = THIS_MODULE, 1970 .open = psched_open, 1971 .read = seq_read, 1972 .llseek = seq_lseek, 1973 .release = single_release, 1974 }; 1975 1976 static int __net_init psched_net_init(struct net *net) 1977 { 1978 struct proc_dir_entry *e; 1979 1980 e = proc_create("psched", 0, net->proc_net, &psched_fops); 1981 if (e == NULL) 1982 return -ENOMEM; 1983 1984 return 0; 1985 } 1986 1987 static void __net_exit psched_net_exit(struct net *net) 1988 { 1989 remove_proc_entry("psched", net->proc_net); 1990 } 1991 #else 1992 static int __net_init psched_net_init(struct net *net) 1993 { 1994 return 0; 1995 } 1996 1997 static void __net_exit psched_net_exit(struct net *net) 1998 { 1999 } 2000 #endif 2001 2002 static struct pernet_operations psched_net_ops = { 2003 .init = psched_net_init, 2004 .exit = psched_net_exit, 2005 }; 2006 2007 static int __init pktsched_init(void) 2008 { 2009 int err; 2010 2011 err = register_pernet_subsys(&psched_net_ops); 2012 if (err) { 2013 pr_err("pktsched_init: " 2014 "cannot initialize per netns operations\n"); 2015 return err; 2016 } 2017 2018 register_qdisc(&pfifo_fast_ops); 2019 register_qdisc(&pfifo_qdisc_ops); 2020 register_qdisc(&bfifo_qdisc_ops); 2021 register_qdisc(&pfifo_head_drop_qdisc_ops); 2022 register_qdisc(&mq_qdisc_ops); 2023 register_qdisc(&noqueue_qdisc_ops); 2024 2025 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0); 2026 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0); 2027 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, 2028 0); 2029 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0); 2030 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0); 2031 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, 2032 0); 2033 2034 return 0; 2035 } 2036 2037 subsys_initcall(pktsched_init); 2038