1 /* 2 * net/sched/sch_api.c Packet scheduler API. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 * 11 * Fixes: 12 * 13 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired. 14 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support 15 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support 16 */ 17 18 #include <linux/module.h> 19 #include <linux/types.h> 20 #include <linux/kernel.h> 21 #include <linux/string.h> 22 #include <linux/errno.h> 23 #include <linux/skbuff.h> 24 #include <linux/init.h> 25 #include <linux/proc_fs.h> 26 #include <linux/seq_file.h> 27 #include <linux/kmod.h> 28 #include <linux/list.h> 29 #include <linux/hrtimer.h> 30 #include <linux/lockdep.h> 31 #include <linux/slab.h> 32 #include <linux/hashtable.h> 33 34 #include <net/net_namespace.h> 35 #include <net/sock.h> 36 #include <net/netlink.h> 37 #include <net/pkt_sched.h> 38 39 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 40 struct nlmsghdr *n, u32 clid, 41 struct Qdisc *old, struct Qdisc *new); 42 static int tclass_notify(struct net *net, struct sk_buff *oskb, 43 struct nlmsghdr *n, struct Qdisc *q, 44 unsigned long cl, int event); 45 46 /* 47 48 Short review. 49 ------------- 50 51 This file consists of two interrelated parts: 52 53 1. queueing disciplines manager frontend. 54 2. traffic classes manager frontend. 55 56 Generally, queueing discipline ("qdisc") is a black box, 57 which is able to enqueue packets and to dequeue them (when 58 device is ready to send something) in order and at times 59 determined by algorithm hidden in it. 60 61 qdisc's are divided to two categories: 62 - "queues", which have no internal structure visible from outside. 63 - "schedulers", which split all the packets to "traffic classes", 64 using "packet classifiers" (look at cls_api.c) 65 66 In turn, classes may have child qdiscs (as rule, queues) 67 attached to them etc. etc. etc. 68 69 The goal of the routines in this file is to translate 70 information supplied by user in the form of handles 71 to more intelligible for kernel form, to make some sanity 72 checks and part of work, which is common to all qdiscs 73 and to provide rtnetlink notifications. 74 75 All real intelligent work is done inside qdisc modules. 76 77 78 79 Every discipline has two major routines: enqueue and dequeue. 80 81 ---dequeue 82 83 dequeue usually returns a skb to send. It is allowed to return NULL, 84 but it does not mean that queue is empty, it just means that 85 discipline does not want to send anything this time. 86 Queue is really empty if q->q.qlen == 0. 87 For complicated disciplines with multiple queues q->q is not 88 real packet queue, but however q->q.qlen must be valid. 89 90 ---enqueue 91 92 enqueue returns 0, if packet was enqueued successfully. 93 If packet (this one or another one) was dropped, it returns 94 not zero error code. 95 NET_XMIT_DROP - this packet dropped 96 Expected action: do not backoff, but wait until queue will clear. 97 NET_XMIT_CN - probably this packet enqueued, but another one dropped. 98 Expected action: backoff or ignore 99 100 Auxiliary routines: 101 102 ---peek 103 104 like dequeue but without removing a packet from the queue 105 106 ---reset 107 108 returns qdisc to initial state: purge all buffers, clear all 109 timers, counters (except for statistics) etc. 110 111 ---init 112 113 initializes newly created qdisc. 114 115 ---destroy 116 117 destroys resources allocated by init and during lifetime of qdisc. 118 119 ---change 120 121 changes qdisc parameters. 122 */ 123 124 /* Protects list of registered TC modules. It is pure SMP lock. */ 125 static DEFINE_RWLOCK(qdisc_mod_lock); 126 127 128 /************************************************ 129 * Queueing disciplines manipulation. * 130 ************************************************/ 131 132 133 /* The list of all installed queueing disciplines. */ 134 135 static struct Qdisc_ops *qdisc_base; 136 137 /* Register/unregister queueing discipline */ 138 139 int register_qdisc(struct Qdisc_ops *qops) 140 { 141 struct Qdisc_ops *q, **qp; 142 int rc = -EEXIST; 143 144 write_lock(&qdisc_mod_lock); 145 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 146 if (!strcmp(qops->id, q->id)) 147 goto out; 148 149 if (qops->enqueue == NULL) 150 qops->enqueue = noop_qdisc_ops.enqueue; 151 if (qops->peek == NULL) { 152 if (qops->dequeue == NULL) 153 qops->peek = noop_qdisc_ops.peek; 154 else 155 goto out_einval; 156 } 157 if (qops->dequeue == NULL) 158 qops->dequeue = noop_qdisc_ops.dequeue; 159 160 if (qops->cl_ops) { 161 const struct Qdisc_class_ops *cops = qops->cl_ops; 162 163 if (!(cops->get && cops->put && cops->walk && cops->leaf)) 164 goto out_einval; 165 166 if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf)) 167 goto out_einval; 168 } 169 170 qops->next = NULL; 171 *qp = qops; 172 rc = 0; 173 out: 174 write_unlock(&qdisc_mod_lock); 175 return rc; 176 177 out_einval: 178 rc = -EINVAL; 179 goto out; 180 } 181 EXPORT_SYMBOL(register_qdisc); 182 183 int unregister_qdisc(struct Qdisc_ops *qops) 184 { 185 struct Qdisc_ops *q, **qp; 186 int err = -ENOENT; 187 188 write_lock(&qdisc_mod_lock); 189 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 190 if (q == qops) 191 break; 192 if (q) { 193 *qp = q->next; 194 q->next = NULL; 195 err = 0; 196 } 197 write_unlock(&qdisc_mod_lock); 198 return err; 199 } 200 EXPORT_SYMBOL(unregister_qdisc); 201 202 /* Get default qdisc if not otherwise specified */ 203 void qdisc_get_default(char *name, size_t len) 204 { 205 read_lock(&qdisc_mod_lock); 206 strlcpy(name, default_qdisc_ops->id, len); 207 read_unlock(&qdisc_mod_lock); 208 } 209 210 static struct Qdisc_ops *qdisc_lookup_default(const char *name) 211 { 212 struct Qdisc_ops *q = NULL; 213 214 for (q = qdisc_base; q; q = q->next) { 215 if (!strcmp(name, q->id)) { 216 if (!try_module_get(q->owner)) 217 q = NULL; 218 break; 219 } 220 } 221 222 return q; 223 } 224 225 /* Set new default qdisc to use */ 226 int qdisc_set_default(const char *name) 227 { 228 const struct Qdisc_ops *ops; 229 230 if (!capable(CAP_NET_ADMIN)) 231 return -EPERM; 232 233 write_lock(&qdisc_mod_lock); 234 ops = qdisc_lookup_default(name); 235 if (!ops) { 236 /* Not found, drop lock and try to load module */ 237 write_unlock(&qdisc_mod_lock); 238 request_module("sch_%s", name); 239 write_lock(&qdisc_mod_lock); 240 241 ops = qdisc_lookup_default(name); 242 } 243 244 if (ops) { 245 /* Set new default */ 246 module_put(default_qdisc_ops->owner); 247 default_qdisc_ops = ops; 248 } 249 write_unlock(&qdisc_mod_lock); 250 251 return ops ? 0 : -ENOENT; 252 } 253 254 #ifdef CONFIG_NET_SCH_DEFAULT 255 /* Set default value from kernel config */ 256 static int __init sch_default_qdisc(void) 257 { 258 return qdisc_set_default(CONFIG_DEFAULT_NET_SCH); 259 } 260 late_initcall(sch_default_qdisc); 261 #endif 262 263 /* We know handle. Find qdisc among all qdisc's attached to device 264 * (root qdisc, all its children, children of children etc.) 265 * Note: caller either uses rtnl or rcu_read_lock() 266 */ 267 268 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle) 269 { 270 struct Qdisc *q; 271 272 if (!qdisc_dev(root)) 273 return (root->handle == handle ? root : NULL); 274 275 if (!(root->flags & TCQ_F_BUILTIN) && 276 root->handle == handle) 277 return root; 278 279 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) { 280 if (q->handle == handle) 281 return q; 282 } 283 return NULL; 284 } 285 286 void qdisc_hash_add(struct Qdisc *q, bool invisible) 287 { 288 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 289 struct Qdisc *root = qdisc_dev(q)->qdisc; 290 291 WARN_ON_ONCE(root == &noop_qdisc); 292 ASSERT_RTNL(); 293 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle); 294 if (invisible) 295 q->flags |= TCQ_F_INVISIBLE; 296 } 297 } 298 EXPORT_SYMBOL(qdisc_hash_add); 299 300 void qdisc_hash_del(struct Qdisc *q) 301 { 302 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 303 ASSERT_RTNL(); 304 hash_del_rcu(&q->hash); 305 } 306 } 307 EXPORT_SYMBOL(qdisc_hash_del); 308 309 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) 310 { 311 struct Qdisc *q; 312 313 q = qdisc_match_from_root(dev->qdisc, handle); 314 if (q) 315 goto out; 316 317 if (dev_ingress_queue(dev)) 318 q = qdisc_match_from_root( 319 dev_ingress_queue(dev)->qdisc_sleeping, 320 handle); 321 out: 322 return q; 323 } 324 325 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) 326 { 327 unsigned long cl; 328 struct Qdisc *leaf; 329 const struct Qdisc_class_ops *cops = p->ops->cl_ops; 330 331 if (cops == NULL) 332 return NULL; 333 cl = cops->get(p, classid); 334 335 if (cl == 0) 336 return NULL; 337 leaf = cops->leaf(p, cl); 338 cops->put(p, cl); 339 return leaf; 340 } 341 342 /* Find queueing discipline by name */ 343 344 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) 345 { 346 struct Qdisc_ops *q = NULL; 347 348 if (kind) { 349 read_lock(&qdisc_mod_lock); 350 for (q = qdisc_base; q; q = q->next) { 351 if (nla_strcmp(kind, q->id) == 0) { 352 if (!try_module_get(q->owner)) 353 q = NULL; 354 break; 355 } 356 } 357 read_unlock(&qdisc_mod_lock); 358 } 359 return q; 360 } 361 362 /* The linklayer setting were not transferred from iproute2, in older 363 * versions, and the rate tables lookup systems have been dropped in 364 * the kernel. To keep backward compatible with older iproute2 tc 365 * utils, we detect the linklayer setting by detecting if the rate 366 * table were modified. 367 * 368 * For linklayer ATM table entries, the rate table will be aligned to 369 * 48 bytes, thus some table entries will contain the same value. The 370 * mpu (min packet unit) is also encoded into the old rate table, thus 371 * starting from the mpu, we find low and high table entries for 372 * mapping this cell. If these entries contain the same value, when 373 * the rate tables have been modified for linklayer ATM. 374 * 375 * This is done by rounding mpu to the nearest 48 bytes cell/entry, 376 * and then roundup to the next cell, calc the table entry one below, 377 * and compare. 378 */ 379 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab) 380 { 381 int low = roundup(r->mpu, 48); 382 int high = roundup(low+1, 48); 383 int cell_low = low >> r->cell_log; 384 int cell_high = (high >> r->cell_log) - 1; 385 386 /* rtab is too inaccurate at rates > 100Mbit/s */ 387 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) { 388 pr_debug("TC linklayer: Giving up ATM detection\n"); 389 return TC_LINKLAYER_ETHERNET; 390 } 391 392 if ((cell_high > cell_low) && (cell_high < 256) 393 && (rtab[cell_low] == rtab[cell_high])) { 394 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n", 395 cell_low, cell_high, rtab[cell_high]); 396 return TC_LINKLAYER_ATM; 397 } 398 return TC_LINKLAYER_ETHERNET; 399 } 400 401 static struct qdisc_rate_table *qdisc_rtab_list; 402 403 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, 404 struct nlattr *tab) 405 { 406 struct qdisc_rate_table *rtab; 407 408 if (tab == NULL || r->rate == 0 || r->cell_log == 0 || 409 nla_len(tab) != TC_RTAB_SIZE) 410 return NULL; 411 412 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { 413 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) && 414 !memcmp(&rtab->data, nla_data(tab), 1024)) { 415 rtab->refcnt++; 416 return rtab; 417 } 418 } 419 420 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); 421 if (rtab) { 422 rtab->rate = *r; 423 rtab->refcnt = 1; 424 memcpy(rtab->data, nla_data(tab), 1024); 425 if (r->linklayer == TC_LINKLAYER_UNAWARE) 426 r->linklayer = __detect_linklayer(r, rtab->data); 427 rtab->next = qdisc_rtab_list; 428 qdisc_rtab_list = rtab; 429 } 430 return rtab; 431 } 432 EXPORT_SYMBOL(qdisc_get_rtab); 433 434 void qdisc_put_rtab(struct qdisc_rate_table *tab) 435 { 436 struct qdisc_rate_table *rtab, **rtabp; 437 438 if (!tab || --tab->refcnt) 439 return; 440 441 for (rtabp = &qdisc_rtab_list; 442 (rtab = *rtabp) != NULL; 443 rtabp = &rtab->next) { 444 if (rtab == tab) { 445 *rtabp = rtab->next; 446 kfree(rtab); 447 return; 448 } 449 } 450 } 451 EXPORT_SYMBOL(qdisc_put_rtab); 452 453 static LIST_HEAD(qdisc_stab_list); 454 455 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { 456 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, 457 [TCA_STAB_DATA] = { .type = NLA_BINARY }, 458 }; 459 460 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) 461 { 462 struct nlattr *tb[TCA_STAB_MAX + 1]; 463 struct qdisc_size_table *stab; 464 struct tc_sizespec *s; 465 unsigned int tsize = 0; 466 u16 *tab = NULL; 467 int err; 468 469 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy, NULL); 470 if (err < 0) 471 return ERR_PTR(err); 472 if (!tb[TCA_STAB_BASE]) 473 return ERR_PTR(-EINVAL); 474 475 s = nla_data(tb[TCA_STAB_BASE]); 476 477 if (s->tsize > 0) { 478 if (!tb[TCA_STAB_DATA]) 479 return ERR_PTR(-EINVAL); 480 tab = nla_data(tb[TCA_STAB_DATA]); 481 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); 482 } 483 484 if (tsize != s->tsize || (!tab && tsize > 0)) 485 return ERR_PTR(-EINVAL); 486 487 list_for_each_entry(stab, &qdisc_stab_list, list) { 488 if (memcmp(&stab->szopts, s, sizeof(*s))) 489 continue; 490 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16))) 491 continue; 492 stab->refcnt++; 493 return stab; 494 } 495 496 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); 497 if (!stab) 498 return ERR_PTR(-ENOMEM); 499 500 stab->refcnt = 1; 501 stab->szopts = *s; 502 if (tsize > 0) 503 memcpy(stab->data, tab, tsize * sizeof(u16)); 504 505 list_add_tail(&stab->list, &qdisc_stab_list); 506 507 return stab; 508 } 509 510 static void stab_kfree_rcu(struct rcu_head *head) 511 { 512 kfree(container_of(head, struct qdisc_size_table, rcu)); 513 } 514 515 void qdisc_put_stab(struct qdisc_size_table *tab) 516 { 517 if (!tab) 518 return; 519 520 if (--tab->refcnt == 0) { 521 list_del(&tab->list); 522 call_rcu_bh(&tab->rcu, stab_kfree_rcu); 523 } 524 } 525 EXPORT_SYMBOL(qdisc_put_stab); 526 527 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) 528 { 529 struct nlattr *nest; 530 531 nest = nla_nest_start(skb, TCA_STAB); 532 if (nest == NULL) 533 goto nla_put_failure; 534 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts)) 535 goto nla_put_failure; 536 nla_nest_end(skb, nest); 537 538 return skb->len; 539 540 nla_put_failure: 541 return -1; 542 } 543 544 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 545 const struct qdisc_size_table *stab) 546 { 547 int pkt_len, slot; 548 549 pkt_len = skb->len + stab->szopts.overhead; 550 if (unlikely(!stab->szopts.tsize)) 551 goto out; 552 553 slot = pkt_len + stab->szopts.cell_align; 554 if (unlikely(slot < 0)) 555 slot = 0; 556 557 slot >>= stab->szopts.cell_log; 558 if (likely(slot < stab->szopts.tsize)) 559 pkt_len = stab->data[slot]; 560 else 561 pkt_len = stab->data[stab->szopts.tsize - 1] * 562 (slot / stab->szopts.tsize) + 563 stab->data[slot % stab->szopts.tsize]; 564 565 pkt_len <<= stab->szopts.size_log; 566 out: 567 if (unlikely(pkt_len < 1)) 568 pkt_len = 1; 569 qdisc_skb_cb(skb)->pkt_len = pkt_len; 570 } 571 EXPORT_SYMBOL(__qdisc_calculate_pkt_len); 572 573 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc) 574 { 575 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { 576 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n", 577 txt, qdisc->ops->id, qdisc->handle >> 16); 578 qdisc->flags |= TCQ_F_WARN_NONWC; 579 } 580 } 581 EXPORT_SYMBOL(qdisc_warn_nonwc); 582 583 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) 584 { 585 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, 586 timer); 587 588 rcu_read_lock(); 589 __netif_schedule(qdisc_root(wd->qdisc)); 590 rcu_read_unlock(); 591 592 return HRTIMER_NORESTART; 593 } 594 595 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 596 { 597 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED); 598 wd->timer.function = qdisc_watchdog; 599 wd->qdisc = qdisc; 600 } 601 EXPORT_SYMBOL(qdisc_watchdog_init); 602 603 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires) 604 { 605 if (test_bit(__QDISC_STATE_DEACTIVATED, 606 &qdisc_root_sleeping(wd->qdisc)->state)) 607 return; 608 609 if (wd->last_expires == expires) 610 return; 611 612 wd->last_expires = expires; 613 hrtimer_start(&wd->timer, 614 ns_to_ktime(expires), 615 HRTIMER_MODE_ABS_PINNED); 616 } 617 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns); 618 619 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) 620 { 621 hrtimer_cancel(&wd->timer); 622 } 623 EXPORT_SYMBOL(qdisc_watchdog_cancel); 624 625 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) 626 { 627 unsigned int size = n * sizeof(struct hlist_head), i; 628 struct hlist_head *h; 629 630 if (size <= PAGE_SIZE) 631 h = kmalloc(size, GFP_KERNEL); 632 else 633 h = (struct hlist_head *) 634 __get_free_pages(GFP_KERNEL, get_order(size)); 635 636 if (h != NULL) { 637 for (i = 0; i < n; i++) 638 INIT_HLIST_HEAD(&h[i]); 639 } 640 return h; 641 } 642 643 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) 644 { 645 unsigned int size = n * sizeof(struct hlist_head); 646 647 if (size <= PAGE_SIZE) 648 kfree(h); 649 else 650 free_pages((unsigned long)h, get_order(size)); 651 } 652 653 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) 654 { 655 struct Qdisc_class_common *cl; 656 struct hlist_node *next; 657 struct hlist_head *nhash, *ohash; 658 unsigned int nsize, nmask, osize; 659 unsigned int i, h; 660 661 /* Rehash when load factor exceeds 0.75 */ 662 if (clhash->hashelems * 4 <= clhash->hashsize * 3) 663 return; 664 nsize = clhash->hashsize * 2; 665 nmask = nsize - 1; 666 nhash = qdisc_class_hash_alloc(nsize); 667 if (nhash == NULL) 668 return; 669 670 ohash = clhash->hash; 671 osize = clhash->hashsize; 672 673 sch_tree_lock(sch); 674 for (i = 0; i < osize; i++) { 675 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) { 676 h = qdisc_class_hash(cl->classid, nmask); 677 hlist_add_head(&cl->hnode, &nhash[h]); 678 } 679 } 680 clhash->hash = nhash; 681 clhash->hashsize = nsize; 682 clhash->hashmask = nmask; 683 sch_tree_unlock(sch); 684 685 qdisc_class_hash_free(ohash, osize); 686 } 687 EXPORT_SYMBOL(qdisc_class_hash_grow); 688 689 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) 690 { 691 unsigned int size = 4; 692 693 clhash->hash = qdisc_class_hash_alloc(size); 694 if (clhash->hash == NULL) 695 return -ENOMEM; 696 clhash->hashsize = size; 697 clhash->hashmask = size - 1; 698 clhash->hashelems = 0; 699 return 0; 700 } 701 EXPORT_SYMBOL(qdisc_class_hash_init); 702 703 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) 704 { 705 qdisc_class_hash_free(clhash->hash, clhash->hashsize); 706 } 707 EXPORT_SYMBOL(qdisc_class_hash_destroy); 708 709 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, 710 struct Qdisc_class_common *cl) 711 { 712 unsigned int h; 713 714 INIT_HLIST_NODE(&cl->hnode); 715 h = qdisc_class_hash(cl->classid, clhash->hashmask); 716 hlist_add_head(&cl->hnode, &clhash->hash[h]); 717 clhash->hashelems++; 718 } 719 EXPORT_SYMBOL(qdisc_class_hash_insert); 720 721 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, 722 struct Qdisc_class_common *cl) 723 { 724 hlist_del(&cl->hnode); 725 clhash->hashelems--; 726 } 727 EXPORT_SYMBOL(qdisc_class_hash_remove); 728 729 /* Allocate an unique handle from space managed by kernel 730 * Possible range is [8000-FFFF]:0000 (0x8000 values) 731 */ 732 static u32 qdisc_alloc_handle(struct net_device *dev) 733 { 734 int i = 0x8000; 735 static u32 autohandle = TC_H_MAKE(0x80000000U, 0); 736 737 do { 738 autohandle += TC_H_MAKE(0x10000U, 0); 739 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) 740 autohandle = TC_H_MAKE(0x80000000U, 0); 741 if (!qdisc_lookup(dev, autohandle)) 742 return autohandle; 743 cond_resched(); 744 } while (--i > 0); 745 746 return 0; 747 } 748 749 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n, 750 unsigned int len) 751 { 752 const struct Qdisc_class_ops *cops; 753 unsigned long cl; 754 u32 parentid; 755 int drops; 756 757 if (n == 0 && len == 0) 758 return; 759 drops = max_t(int, n, 0); 760 rcu_read_lock(); 761 while ((parentid = sch->parent)) { 762 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) 763 break; 764 765 if (sch->flags & TCQ_F_NOPARENT) 766 break; 767 /* TODO: perform the search on a per txq basis */ 768 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); 769 if (sch == NULL) { 770 WARN_ON_ONCE(parentid != TC_H_ROOT); 771 break; 772 } 773 cops = sch->ops->cl_ops; 774 if (cops->qlen_notify) { 775 cl = cops->get(sch, parentid); 776 cops->qlen_notify(sch, cl); 777 cops->put(sch, cl); 778 } 779 sch->q.qlen -= n; 780 sch->qstats.backlog -= len; 781 __qdisc_qstats_drop(sch, drops); 782 } 783 rcu_read_unlock(); 784 } 785 EXPORT_SYMBOL(qdisc_tree_reduce_backlog); 786 787 static void notify_and_destroy(struct net *net, struct sk_buff *skb, 788 struct nlmsghdr *n, u32 clid, 789 struct Qdisc *old, struct Qdisc *new) 790 { 791 if (new || old) 792 qdisc_notify(net, skb, n, clid, old, new); 793 794 if (old) 795 qdisc_destroy(old); 796 } 797 798 /* Graft qdisc "new" to class "classid" of qdisc "parent" or 799 * to device "dev". 800 * 801 * When appropriate send a netlink notification using 'skb' 802 * and "n". 803 * 804 * On success, destroy old qdisc. 805 */ 806 807 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 808 struct sk_buff *skb, struct nlmsghdr *n, u32 classid, 809 struct Qdisc *new, struct Qdisc *old) 810 { 811 struct Qdisc *q = old; 812 struct net *net = dev_net(dev); 813 int err = 0; 814 815 if (parent == NULL) { 816 unsigned int i, num_q, ingress; 817 818 ingress = 0; 819 num_q = dev->num_tx_queues; 820 if ((q && q->flags & TCQ_F_INGRESS) || 821 (new && new->flags & TCQ_F_INGRESS)) { 822 num_q = 1; 823 ingress = 1; 824 if (!dev_ingress_queue(dev)) 825 return -ENOENT; 826 } 827 828 if (dev->flags & IFF_UP) 829 dev_deactivate(dev); 830 831 if (new && new->ops->attach) 832 goto skip; 833 834 for (i = 0; i < num_q; i++) { 835 struct netdev_queue *dev_queue = dev_ingress_queue(dev); 836 837 if (!ingress) 838 dev_queue = netdev_get_tx_queue(dev, i); 839 840 old = dev_graft_qdisc(dev_queue, new); 841 if (new && i > 0) 842 refcount_inc(&new->refcnt); 843 844 if (!ingress) 845 qdisc_destroy(old); 846 } 847 848 skip: 849 if (!ingress) { 850 notify_and_destroy(net, skb, n, classid, 851 dev->qdisc, new); 852 if (new && !new->ops->attach) 853 refcount_inc(&new->refcnt); 854 dev->qdisc = new ? : &noop_qdisc; 855 856 if (new && new->ops->attach) 857 new->ops->attach(new); 858 } else { 859 notify_and_destroy(net, skb, n, classid, old, new); 860 } 861 862 if (dev->flags & IFF_UP) 863 dev_activate(dev); 864 } else { 865 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 866 867 err = -EOPNOTSUPP; 868 if (cops && cops->graft) { 869 unsigned long cl = cops->get(parent, classid); 870 if (cl) { 871 err = cops->graft(parent, cl, new, &old); 872 cops->put(parent, cl); 873 } else 874 err = -ENOENT; 875 } 876 if (!err) 877 notify_and_destroy(net, skb, n, classid, old, new); 878 } 879 return err; 880 } 881 882 /* lockdep annotation is needed for ingress; egress gets it only for name */ 883 static struct lock_class_key qdisc_tx_lock; 884 static struct lock_class_key qdisc_rx_lock; 885 886 /* 887 Allocate and initialize new qdisc. 888 889 Parameters are passed via opt. 890 */ 891 892 static struct Qdisc *qdisc_create(struct net_device *dev, 893 struct netdev_queue *dev_queue, 894 struct Qdisc *p, u32 parent, u32 handle, 895 struct nlattr **tca, int *errp) 896 { 897 int err; 898 struct nlattr *kind = tca[TCA_KIND]; 899 struct Qdisc *sch; 900 struct Qdisc_ops *ops; 901 struct qdisc_size_table *stab; 902 903 ops = qdisc_lookup_ops(kind); 904 #ifdef CONFIG_MODULES 905 if (ops == NULL && kind != NULL) { 906 char name[IFNAMSIZ]; 907 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 908 /* We dropped the RTNL semaphore in order to 909 * perform the module load. So, even if we 910 * succeeded in loading the module we have to 911 * tell the caller to replay the request. We 912 * indicate this using -EAGAIN. 913 * We replay the request because the device may 914 * go away in the mean time. 915 */ 916 rtnl_unlock(); 917 request_module("sch_%s", name); 918 rtnl_lock(); 919 ops = qdisc_lookup_ops(kind); 920 if (ops != NULL) { 921 /* We will try again qdisc_lookup_ops, 922 * so don't keep a reference. 923 */ 924 module_put(ops->owner); 925 err = -EAGAIN; 926 goto err_out; 927 } 928 } 929 } 930 #endif 931 932 err = -ENOENT; 933 if (ops == NULL) 934 goto err_out; 935 936 sch = qdisc_alloc(dev_queue, ops); 937 if (IS_ERR(sch)) { 938 err = PTR_ERR(sch); 939 goto err_out2; 940 } 941 942 sch->parent = parent; 943 944 if (handle == TC_H_INGRESS) { 945 sch->flags |= TCQ_F_INGRESS; 946 handle = TC_H_MAKE(TC_H_INGRESS, 0); 947 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); 948 } else { 949 if (handle == 0) { 950 handle = qdisc_alloc_handle(dev); 951 err = -ENOMEM; 952 if (handle == 0) 953 goto err_out3; 954 } 955 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); 956 if (!netif_is_multiqueue(dev)) 957 sch->flags |= TCQ_F_ONETXQUEUE; 958 } 959 960 sch->handle = handle; 961 962 /* This exist to keep backward compatible with a userspace 963 * loophole, what allowed userspace to get IFF_NO_QUEUE 964 * facility on older kernels by setting tx_queue_len=0 (prior 965 * to qdisc init), and then forgot to reinit tx_queue_len 966 * before again attaching a qdisc. 967 */ 968 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) { 969 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 970 netdev_info(dev, "Caught tx_queue_len zero misconfig\n"); 971 } 972 973 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { 974 if (qdisc_is_percpu_stats(sch)) { 975 sch->cpu_bstats = 976 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu); 977 if (!sch->cpu_bstats) 978 goto err_out4; 979 980 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue); 981 if (!sch->cpu_qstats) 982 goto err_out4; 983 } 984 985 if (tca[TCA_STAB]) { 986 stab = qdisc_get_stab(tca[TCA_STAB]); 987 if (IS_ERR(stab)) { 988 err = PTR_ERR(stab); 989 goto err_out4; 990 } 991 rcu_assign_pointer(sch->stab, stab); 992 } 993 if (tca[TCA_RATE]) { 994 seqcount_t *running; 995 996 err = -EOPNOTSUPP; 997 if (sch->flags & TCQ_F_MQROOT) 998 goto err_out4; 999 1000 if ((sch->parent != TC_H_ROOT) && 1001 !(sch->flags & TCQ_F_INGRESS) && 1002 (!p || !(p->flags & TCQ_F_MQROOT))) 1003 running = qdisc_root_sleeping_running(sch); 1004 else 1005 running = &sch->running; 1006 1007 err = gen_new_estimator(&sch->bstats, 1008 sch->cpu_bstats, 1009 &sch->rate_est, 1010 NULL, 1011 running, 1012 tca[TCA_RATE]); 1013 if (err) 1014 goto err_out4; 1015 } 1016 1017 qdisc_hash_add(sch, false); 1018 1019 return sch; 1020 } 1021 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */ 1022 if (ops->destroy) 1023 ops->destroy(sch); 1024 err_out3: 1025 dev_put(dev); 1026 kfree((char *) sch - sch->padded); 1027 err_out2: 1028 module_put(ops->owner); 1029 err_out: 1030 *errp = err; 1031 return NULL; 1032 1033 err_out4: 1034 free_percpu(sch->cpu_bstats); 1035 free_percpu(sch->cpu_qstats); 1036 /* 1037 * Any broken qdiscs that would require a ops->reset() here? 1038 * The qdisc was never in action so it shouldn't be necessary. 1039 */ 1040 qdisc_put_stab(rtnl_dereference(sch->stab)); 1041 if (ops->destroy) 1042 ops->destroy(sch); 1043 goto err_out3; 1044 } 1045 1046 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) 1047 { 1048 struct qdisc_size_table *ostab, *stab = NULL; 1049 int err = 0; 1050 1051 if (tca[TCA_OPTIONS]) { 1052 if (sch->ops->change == NULL) 1053 return -EINVAL; 1054 err = sch->ops->change(sch, tca[TCA_OPTIONS]); 1055 if (err) 1056 return err; 1057 } 1058 1059 if (tca[TCA_STAB]) { 1060 stab = qdisc_get_stab(tca[TCA_STAB]); 1061 if (IS_ERR(stab)) 1062 return PTR_ERR(stab); 1063 } 1064 1065 ostab = rtnl_dereference(sch->stab); 1066 rcu_assign_pointer(sch->stab, stab); 1067 qdisc_put_stab(ostab); 1068 1069 if (tca[TCA_RATE]) { 1070 /* NB: ignores errors from replace_estimator 1071 because change can't be undone. */ 1072 if (sch->flags & TCQ_F_MQROOT) 1073 goto out; 1074 gen_replace_estimator(&sch->bstats, 1075 sch->cpu_bstats, 1076 &sch->rate_est, 1077 NULL, 1078 qdisc_root_sleeping_running(sch), 1079 tca[TCA_RATE]); 1080 } 1081 out: 1082 return 0; 1083 } 1084 1085 struct check_loop_arg { 1086 struct qdisc_walker w; 1087 struct Qdisc *p; 1088 int depth; 1089 }; 1090 1091 static int check_loop_fn(struct Qdisc *q, unsigned long cl, 1092 struct qdisc_walker *w); 1093 1094 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 1095 { 1096 struct check_loop_arg arg; 1097 1098 if (q->ops->cl_ops == NULL) 1099 return 0; 1100 1101 arg.w.stop = arg.w.skip = arg.w.count = 0; 1102 arg.w.fn = check_loop_fn; 1103 arg.depth = depth; 1104 arg.p = p; 1105 q->ops->cl_ops->walk(q, &arg.w); 1106 return arg.w.stop ? -ELOOP : 0; 1107 } 1108 1109 static int 1110 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 1111 { 1112 struct Qdisc *leaf; 1113 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1114 struct check_loop_arg *arg = (struct check_loop_arg *)w; 1115 1116 leaf = cops->leaf(q, cl); 1117 if (leaf) { 1118 if (leaf == arg->p || arg->depth > 7) 1119 return -ELOOP; 1120 return check_loop(leaf, arg->p, arg->depth + 1); 1121 } 1122 return 0; 1123 } 1124 1125 /* 1126 * Delete/get qdisc. 1127 */ 1128 1129 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, 1130 struct netlink_ext_ack *extack) 1131 { 1132 struct net *net = sock_net(skb->sk); 1133 struct tcmsg *tcm = nlmsg_data(n); 1134 struct nlattr *tca[TCA_MAX + 1]; 1135 struct net_device *dev; 1136 u32 clid; 1137 struct Qdisc *q = NULL; 1138 struct Qdisc *p = NULL; 1139 int err; 1140 1141 if ((n->nlmsg_type != RTM_GETQDISC) && 1142 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1143 return -EPERM; 1144 1145 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1146 if (err < 0) 1147 return err; 1148 1149 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1150 if (!dev) 1151 return -ENODEV; 1152 1153 clid = tcm->tcm_parent; 1154 if (clid) { 1155 if (clid != TC_H_ROOT) { 1156 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 1157 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1158 if (!p) 1159 return -ENOENT; 1160 q = qdisc_leaf(p, clid); 1161 } else if (dev_ingress_queue(dev)) { 1162 q = dev_ingress_queue(dev)->qdisc_sleeping; 1163 } 1164 } else { 1165 q = dev->qdisc; 1166 } 1167 if (!q) 1168 return -ENOENT; 1169 1170 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) 1171 return -EINVAL; 1172 } else { 1173 q = qdisc_lookup(dev, tcm->tcm_handle); 1174 if (!q) 1175 return -ENOENT; 1176 } 1177 1178 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1179 return -EINVAL; 1180 1181 if (n->nlmsg_type == RTM_DELQDISC) { 1182 if (!clid) 1183 return -EINVAL; 1184 if (q->handle == 0) 1185 return -ENOENT; 1186 err = qdisc_graft(dev, p, skb, n, clid, NULL, q); 1187 if (err != 0) 1188 return err; 1189 } else { 1190 qdisc_notify(net, skb, n, clid, NULL, q); 1191 } 1192 return 0; 1193 } 1194 1195 /* 1196 * Create/change qdisc. 1197 */ 1198 1199 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, 1200 struct netlink_ext_ack *extack) 1201 { 1202 struct net *net = sock_net(skb->sk); 1203 struct tcmsg *tcm; 1204 struct nlattr *tca[TCA_MAX + 1]; 1205 struct net_device *dev; 1206 u32 clid; 1207 struct Qdisc *q, *p; 1208 int err; 1209 1210 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1211 return -EPERM; 1212 1213 replay: 1214 /* Reinit, just in case something touches this. */ 1215 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1216 if (err < 0) 1217 return err; 1218 1219 tcm = nlmsg_data(n); 1220 clid = tcm->tcm_parent; 1221 q = p = NULL; 1222 1223 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1224 if (!dev) 1225 return -ENODEV; 1226 1227 1228 if (clid) { 1229 if (clid != TC_H_ROOT) { 1230 if (clid != TC_H_INGRESS) { 1231 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1232 if (!p) 1233 return -ENOENT; 1234 q = qdisc_leaf(p, clid); 1235 } else if (dev_ingress_queue_create(dev)) { 1236 q = dev_ingress_queue(dev)->qdisc_sleeping; 1237 } 1238 } else { 1239 q = dev->qdisc; 1240 } 1241 1242 /* It may be default qdisc, ignore it */ 1243 if (q && q->handle == 0) 1244 q = NULL; 1245 1246 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 1247 if (tcm->tcm_handle) { 1248 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) 1249 return -EEXIST; 1250 if (TC_H_MIN(tcm->tcm_handle)) 1251 return -EINVAL; 1252 q = qdisc_lookup(dev, tcm->tcm_handle); 1253 if (!q) 1254 goto create_n_graft; 1255 if (n->nlmsg_flags & NLM_F_EXCL) 1256 return -EEXIST; 1257 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1258 return -EINVAL; 1259 if (q == p || 1260 (p && check_loop(q, p, 0))) 1261 return -ELOOP; 1262 refcount_inc(&q->refcnt); 1263 goto graft; 1264 } else { 1265 if (!q) 1266 goto create_n_graft; 1267 1268 /* This magic test requires explanation. 1269 * 1270 * We know, that some child q is already 1271 * attached to this parent and have choice: 1272 * either to change it or to create/graft new one. 1273 * 1274 * 1. We are allowed to create/graft only 1275 * if CREATE and REPLACE flags are set. 1276 * 1277 * 2. If EXCL is set, requestor wanted to say, 1278 * that qdisc tcm_handle is not expected 1279 * to exist, so that we choose create/graft too. 1280 * 1281 * 3. The last case is when no flags are set. 1282 * Alas, it is sort of hole in API, we 1283 * cannot decide what to do unambiguously. 1284 * For now we select create/graft, if 1285 * user gave KIND, which does not match existing. 1286 */ 1287 if ((n->nlmsg_flags & NLM_F_CREATE) && 1288 (n->nlmsg_flags & NLM_F_REPLACE) && 1289 ((n->nlmsg_flags & NLM_F_EXCL) || 1290 (tca[TCA_KIND] && 1291 nla_strcmp(tca[TCA_KIND], q->ops->id)))) 1292 goto create_n_graft; 1293 } 1294 } 1295 } else { 1296 if (!tcm->tcm_handle) 1297 return -EINVAL; 1298 q = qdisc_lookup(dev, tcm->tcm_handle); 1299 } 1300 1301 /* Change qdisc parameters */ 1302 if (q == NULL) 1303 return -ENOENT; 1304 if (n->nlmsg_flags & NLM_F_EXCL) 1305 return -EEXIST; 1306 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1307 return -EINVAL; 1308 err = qdisc_change(q, tca); 1309 if (err == 0) 1310 qdisc_notify(net, skb, n, clid, NULL, q); 1311 return err; 1312 1313 create_n_graft: 1314 if (!(n->nlmsg_flags & NLM_F_CREATE)) 1315 return -ENOENT; 1316 if (clid == TC_H_INGRESS) { 1317 if (dev_ingress_queue(dev)) 1318 q = qdisc_create(dev, dev_ingress_queue(dev), p, 1319 tcm->tcm_parent, tcm->tcm_parent, 1320 tca, &err); 1321 else 1322 err = -ENOENT; 1323 } else { 1324 struct netdev_queue *dev_queue; 1325 1326 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) 1327 dev_queue = p->ops->cl_ops->select_queue(p, tcm); 1328 else if (p) 1329 dev_queue = p->dev_queue; 1330 else 1331 dev_queue = netdev_get_tx_queue(dev, 0); 1332 1333 q = qdisc_create(dev, dev_queue, p, 1334 tcm->tcm_parent, tcm->tcm_handle, 1335 tca, &err); 1336 } 1337 if (q == NULL) { 1338 if (err == -EAGAIN) 1339 goto replay; 1340 return err; 1341 } 1342 1343 graft: 1344 err = qdisc_graft(dev, p, skb, n, clid, q, NULL); 1345 if (err) { 1346 if (q) 1347 qdisc_destroy(q); 1348 return err; 1349 } 1350 1351 return 0; 1352 } 1353 1354 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, 1355 u32 portid, u32 seq, u16 flags, int event) 1356 { 1357 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL; 1358 struct gnet_stats_queue __percpu *cpu_qstats = NULL; 1359 struct tcmsg *tcm; 1360 struct nlmsghdr *nlh; 1361 unsigned char *b = skb_tail_pointer(skb); 1362 struct gnet_dump d; 1363 struct qdisc_size_table *stab; 1364 __u32 qlen; 1365 1366 cond_resched(); 1367 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1368 if (!nlh) 1369 goto out_nlmsg_trim; 1370 tcm = nlmsg_data(nlh); 1371 tcm->tcm_family = AF_UNSPEC; 1372 tcm->tcm__pad1 = 0; 1373 tcm->tcm__pad2 = 0; 1374 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1375 tcm->tcm_parent = clid; 1376 tcm->tcm_handle = q->handle; 1377 tcm->tcm_info = refcount_read(&q->refcnt); 1378 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1379 goto nla_put_failure; 1380 if (q->ops->dump && q->ops->dump(q, skb) < 0) 1381 goto nla_put_failure; 1382 qlen = q->q.qlen; 1383 1384 stab = rtnl_dereference(q->stab); 1385 if (stab && qdisc_dump_stab(skb, stab) < 0) 1386 goto nla_put_failure; 1387 1388 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1389 NULL, &d, TCA_PAD) < 0) 1390 goto nla_put_failure; 1391 1392 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 1393 goto nla_put_failure; 1394 1395 if (qdisc_is_percpu_stats(q)) { 1396 cpu_bstats = q->cpu_bstats; 1397 cpu_qstats = q->cpu_qstats; 1398 } 1399 1400 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q), 1401 &d, cpu_bstats, &q->bstats) < 0 || 1402 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || 1403 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0) 1404 goto nla_put_failure; 1405 1406 if (gnet_stats_finish_copy(&d) < 0) 1407 goto nla_put_failure; 1408 1409 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1410 return skb->len; 1411 1412 out_nlmsg_trim: 1413 nla_put_failure: 1414 nlmsg_trim(skb, b); 1415 return -1; 1416 } 1417 1418 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible) 1419 { 1420 if (q->flags & TCQ_F_BUILTIN) 1421 return true; 1422 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible) 1423 return true; 1424 1425 return false; 1426 } 1427 1428 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 1429 struct nlmsghdr *n, u32 clid, 1430 struct Qdisc *old, struct Qdisc *new) 1431 { 1432 struct sk_buff *skb; 1433 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1434 1435 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1436 if (!skb) 1437 return -ENOBUFS; 1438 1439 if (old && !tc_qdisc_dump_ignore(old, false)) { 1440 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq, 1441 0, RTM_DELQDISC) < 0) 1442 goto err_out; 1443 } 1444 if (new && !tc_qdisc_dump_ignore(new, false)) { 1445 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq, 1446 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) 1447 goto err_out; 1448 } 1449 1450 if (skb->len) 1451 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1452 n->nlmsg_flags & NLM_F_ECHO); 1453 1454 err_out: 1455 kfree_skb(skb); 1456 return -EINVAL; 1457 } 1458 1459 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, 1460 struct netlink_callback *cb, 1461 int *q_idx_p, int s_q_idx, bool recur, 1462 bool dump_invisible) 1463 { 1464 int ret = 0, q_idx = *q_idx_p; 1465 struct Qdisc *q; 1466 int b; 1467 1468 if (!root) 1469 return 0; 1470 1471 q = root; 1472 if (q_idx < s_q_idx) { 1473 q_idx++; 1474 } else { 1475 if (!tc_qdisc_dump_ignore(q, dump_invisible) && 1476 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1477 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1478 RTM_NEWQDISC) <= 0) 1479 goto done; 1480 q_idx++; 1481 } 1482 1483 /* If dumping singletons, there is no qdisc_dev(root) and the singleton 1484 * itself has already been dumped. 1485 * 1486 * If we've already dumped the top-level (ingress) qdisc above and the global 1487 * qdisc hashtable, we don't want to hit it again 1488 */ 1489 if (!qdisc_dev(root) || !recur) 1490 goto out; 1491 1492 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1493 if (q_idx < s_q_idx) { 1494 q_idx++; 1495 continue; 1496 } 1497 if (!tc_qdisc_dump_ignore(q, dump_invisible) && 1498 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1499 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1500 RTM_NEWQDISC) <= 0) 1501 goto done; 1502 q_idx++; 1503 } 1504 1505 out: 1506 *q_idx_p = q_idx; 1507 return ret; 1508 done: 1509 ret = -1; 1510 goto out; 1511 } 1512 1513 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 1514 { 1515 struct net *net = sock_net(skb->sk); 1516 int idx, q_idx; 1517 int s_idx, s_q_idx; 1518 struct net_device *dev; 1519 const struct nlmsghdr *nlh = cb->nlh; 1520 struct tcmsg *tcm = nlmsg_data(nlh); 1521 struct nlattr *tca[TCA_MAX + 1]; 1522 int err; 1523 1524 s_idx = cb->args[0]; 1525 s_q_idx = q_idx = cb->args[1]; 1526 1527 idx = 0; 1528 ASSERT_RTNL(); 1529 1530 err = nlmsg_parse(nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL); 1531 if (err < 0) 1532 return err; 1533 1534 for_each_netdev(net, dev) { 1535 struct netdev_queue *dev_queue; 1536 1537 if (idx < s_idx) 1538 goto cont; 1539 if (idx > s_idx) 1540 s_q_idx = 0; 1541 q_idx = 0; 1542 1543 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx, 1544 true, tca[TCA_DUMP_INVISIBLE]) < 0) 1545 goto done; 1546 1547 dev_queue = dev_ingress_queue(dev); 1548 if (dev_queue && 1549 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, 1550 &q_idx, s_q_idx, false, 1551 tca[TCA_DUMP_INVISIBLE]) < 0) 1552 goto done; 1553 1554 cont: 1555 idx++; 1556 } 1557 1558 done: 1559 cb->args[0] = idx; 1560 cb->args[1] = q_idx; 1561 1562 return skb->len; 1563 } 1564 1565 1566 1567 /************************************************ 1568 * Traffic classes manipulation. * 1569 ************************************************/ 1570 1571 1572 1573 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, 1574 struct netlink_ext_ack *extack) 1575 { 1576 struct net *net = sock_net(skb->sk); 1577 struct tcmsg *tcm = nlmsg_data(n); 1578 struct nlattr *tca[TCA_MAX + 1]; 1579 struct net_device *dev; 1580 struct Qdisc *q = NULL; 1581 const struct Qdisc_class_ops *cops; 1582 unsigned long cl = 0; 1583 unsigned long new_cl; 1584 u32 portid; 1585 u32 clid; 1586 u32 qid; 1587 int err; 1588 1589 if ((n->nlmsg_type != RTM_GETTCLASS) && 1590 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1591 return -EPERM; 1592 1593 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); 1594 if (err < 0) 1595 return err; 1596 1597 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1598 if (!dev) 1599 return -ENODEV; 1600 1601 /* 1602 parent == TC_H_UNSPEC - unspecified parent. 1603 parent == TC_H_ROOT - class is root, which has no parent. 1604 parent == X:0 - parent is root class. 1605 parent == X:Y - parent is a node in hierarchy. 1606 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 1607 1608 handle == 0:0 - generate handle from kernel pool. 1609 handle == 0:Y - class is X:Y, where X:0 is qdisc. 1610 handle == X:Y - clear. 1611 handle == X:0 - root class. 1612 */ 1613 1614 /* Step 1. Determine qdisc handle X:0 */ 1615 1616 portid = tcm->tcm_parent; 1617 clid = tcm->tcm_handle; 1618 qid = TC_H_MAJ(clid); 1619 1620 if (portid != TC_H_ROOT) { 1621 u32 qid1 = TC_H_MAJ(portid); 1622 1623 if (qid && qid1) { 1624 /* If both majors are known, they must be identical. */ 1625 if (qid != qid1) 1626 return -EINVAL; 1627 } else if (qid1) { 1628 qid = qid1; 1629 } else if (qid == 0) 1630 qid = dev->qdisc->handle; 1631 1632 /* Now qid is genuine qdisc handle consistent 1633 * both with parent and child. 1634 * 1635 * TC_H_MAJ(portid) still may be unspecified, complete it now. 1636 */ 1637 if (portid) 1638 portid = TC_H_MAKE(qid, portid); 1639 } else { 1640 if (qid == 0) 1641 qid = dev->qdisc->handle; 1642 } 1643 1644 /* OK. Locate qdisc */ 1645 q = qdisc_lookup(dev, qid); 1646 if (!q) 1647 return -ENOENT; 1648 1649 /* An check that it supports classes */ 1650 cops = q->ops->cl_ops; 1651 if (cops == NULL) 1652 return -EINVAL; 1653 1654 /* Now try to get class */ 1655 if (clid == 0) { 1656 if (portid == TC_H_ROOT) 1657 clid = qid; 1658 } else 1659 clid = TC_H_MAKE(qid, clid); 1660 1661 if (clid) 1662 cl = cops->get(q, clid); 1663 1664 if (cl == 0) { 1665 err = -ENOENT; 1666 if (n->nlmsg_type != RTM_NEWTCLASS || 1667 !(n->nlmsg_flags & NLM_F_CREATE)) 1668 goto out; 1669 } else { 1670 switch (n->nlmsg_type) { 1671 case RTM_NEWTCLASS: 1672 err = -EEXIST; 1673 if (n->nlmsg_flags & NLM_F_EXCL) 1674 goto out; 1675 break; 1676 case RTM_DELTCLASS: 1677 err = -EOPNOTSUPP; 1678 if (cops->delete) 1679 err = cops->delete(q, cl); 1680 if (err == 0) 1681 tclass_notify(net, skb, n, q, cl, 1682 RTM_DELTCLASS); 1683 goto out; 1684 case RTM_GETTCLASS: 1685 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS); 1686 goto out; 1687 default: 1688 err = -EINVAL; 1689 goto out; 1690 } 1691 } 1692 1693 new_cl = cl; 1694 err = -EOPNOTSUPP; 1695 if (cops->change) 1696 err = cops->change(q, clid, portid, tca, &new_cl); 1697 if (err == 0) 1698 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS); 1699 1700 out: 1701 if (cl) 1702 cops->put(q, cl); 1703 1704 return err; 1705 } 1706 1707 1708 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1709 unsigned long cl, 1710 u32 portid, u32 seq, u16 flags, int event) 1711 { 1712 struct tcmsg *tcm; 1713 struct nlmsghdr *nlh; 1714 unsigned char *b = skb_tail_pointer(skb); 1715 struct gnet_dump d; 1716 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1717 1718 cond_resched(); 1719 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1720 if (!nlh) 1721 goto out_nlmsg_trim; 1722 tcm = nlmsg_data(nlh); 1723 tcm->tcm_family = AF_UNSPEC; 1724 tcm->tcm__pad1 = 0; 1725 tcm->tcm__pad2 = 0; 1726 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1727 tcm->tcm_parent = q->handle; 1728 tcm->tcm_handle = q->handle; 1729 tcm->tcm_info = 0; 1730 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1731 goto nla_put_failure; 1732 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1733 goto nla_put_failure; 1734 1735 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1736 NULL, &d, TCA_PAD) < 0) 1737 goto nla_put_failure; 1738 1739 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1740 goto nla_put_failure; 1741 1742 if (gnet_stats_finish_copy(&d) < 0) 1743 goto nla_put_failure; 1744 1745 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1746 return skb->len; 1747 1748 out_nlmsg_trim: 1749 nla_put_failure: 1750 nlmsg_trim(skb, b); 1751 return -1; 1752 } 1753 1754 static int tclass_notify(struct net *net, struct sk_buff *oskb, 1755 struct nlmsghdr *n, struct Qdisc *q, 1756 unsigned long cl, int event) 1757 { 1758 struct sk_buff *skb; 1759 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1760 1761 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1762 if (!skb) 1763 return -ENOBUFS; 1764 1765 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) { 1766 kfree_skb(skb); 1767 return -EINVAL; 1768 } 1769 1770 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1771 n->nlmsg_flags & NLM_F_ECHO); 1772 } 1773 1774 struct qdisc_dump_args { 1775 struct qdisc_walker w; 1776 struct sk_buff *skb; 1777 struct netlink_callback *cb; 1778 }; 1779 1780 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, 1781 struct qdisc_walker *arg) 1782 { 1783 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 1784 1785 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid, 1786 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 1787 RTM_NEWTCLASS); 1788 } 1789 1790 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, 1791 struct tcmsg *tcm, struct netlink_callback *cb, 1792 int *t_p, int s_t) 1793 { 1794 struct qdisc_dump_args arg; 1795 1796 if (tc_qdisc_dump_ignore(q, false) || 1797 *t_p < s_t || !q->ops->cl_ops || 1798 (tcm->tcm_parent && 1799 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 1800 (*t_p)++; 1801 return 0; 1802 } 1803 if (*t_p > s_t) 1804 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 1805 arg.w.fn = qdisc_class_dump; 1806 arg.skb = skb; 1807 arg.cb = cb; 1808 arg.w.stop = 0; 1809 arg.w.skip = cb->args[1]; 1810 arg.w.count = 0; 1811 q->ops->cl_ops->walk(q, &arg.w); 1812 cb->args[1] = arg.w.count; 1813 if (arg.w.stop) 1814 return -1; 1815 (*t_p)++; 1816 return 0; 1817 } 1818 1819 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, 1820 struct tcmsg *tcm, struct netlink_callback *cb, 1821 int *t_p, int s_t) 1822 { 1823 struct Qdisc *q; 1824 int b; 1825 1826 if (!root) 1827 return 0; 1828 1829 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) 1830 return -1; 1831 1832 if (!qdisc_dev(root)) 1833 return 0; 1834 1835 if (tcm->tcm_parent) { 1836 q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent)); 1837 if (q && tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1838 return -1; 1839 return 0; 1840 } 1841 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1842 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1843 return -1; 1844 } 1845 1846 return 0; 1847 } 1848 1849 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 1850 { 1851 struct tcmsg *tcm = nlmsg_data(cb->nlh); 1852 struct net *net = sock_net(skb->sk); 1853 struct netdev_queue *dev_queue; 1854 struct net_device *dev; 1855 int t, s_t; 1856 1857 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 1858 return 0; 1859 dev = dev_get_by_index(net, tcm->tcm_ifindex); 1860 if (!dev) 1861 return 0; 1862 1863 s_t = cb->args[0]; 1864 t = 0; 1865 1866 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) 1867 goto done; 1868 1869 dev_queue = dev_ingress_queue(dev); 1870 if (dev_queue && 1871 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, 1872 &t, s_t) < 0) 1873 goto done; 1874 1875 done: 1876 cb->args[0] = t; 1877 1878 dev_put(dev); 1879 return skb->len; 1880 } 1881 1882 #ifdef CONFIG_PROC_FS 1883 static int psched_show(struct seq_file *seq, void *v) 1884 { 1885 seq_printf(seq, "%08x %08x %08x %08x\n", 1886 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), 1887 1000000, 1888 (u32)NSEC_PER_SEC / hrtimer_resolution); 1889 1890 return 0; 1891 } 1892 1893 static int psched_open(struct inode *inode, struct file *file) 1894 { 1895 return single_open(file, psched_show, NULL); 1896 } 1897 1898 static const struct file_operations psched_fops = { 1899 .owner = THIS_MODULE, 1900 .open = psched_open, 1901 .read = seq_read, 1902 .llseek = seq_lseek, 1903 .release = single_release, 1904 }; 1905 1906 static int __net_init psched_net_init(struct net *net) 1907 { 1908 struct proc_dir_entry *e; 1909 1910 e = proc_create("psched", 0, net->proc_net, &psched_fops); 1911 if (e == NULL) 1912 return -ENOMEM; 1913 1914 return 0; 1915 } 1916 1917 static void __net_exit psched_net_exit(struct net *net) 1918 { 1919 remove_proc_entry("psched", net->proc_net); 1920 } 1921 #else 1922 static int __net_init psched_net_init(struct net *net) 1923 { 1924 return 0; 1925 } 1926 1927 static void __net_exit psched_net_exit(struct net *net) 1928 { 1929 } 1930 #endif 1931 1932 static struct pernet_operations psched_net_ops = { 1933 .init = psched_net_init, 1934 .exit = psched_net_exit, 1935 }; 1936 1937 static int __init pktsched_init(void) 1938 { 1939 int err; 1940 1941 err = register_pernet_subsys(&psched_net_ops); 1942 if (err) { 1943 pr_err("pktsched_init: " 1944 "cannot initialize per netns operations\n"); 1945 return err; 1946 } 1947 1948 register_qdisc(&pfifo_fast_ops); 1949 register_qdisc(&pfifo_qdisc_ops); 1950 register_qdisc(&bfifo_qdisc_ops); 1951 register_qdisc(&pfifo_head_drop_qdisc_ops); 1952 register_qdisc(&mq_qdisc_ops); 1953 register_qdisc(&noqueue_qdisc_ops); 1954 1955 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL); 1956 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL); 1957 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, 1958 NULL); 1959 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL); 1960 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL); 1961 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, 1962 NULL); 1963 1964 return 0; 1965 } 1966 1967 subsys_initcall(pktsched_init); 1968