1 /* 2 * net/sched/sch_api.c Packet scheduler API. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 * 11 * Fixes: 12 * 13 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired. 14 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support 15 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support 16 */ 17 18 #include <linux/module.h> 19 #include <linux/types.h> 20 #include <linux/kernel.h> 21 #include <linux/string.h> 22 #include <linux/errno.h> 23 #include <linux/skbuff.h> 24 #include <linux/init.h> 25 #include <linux/proc_fs.h> 26 #include <linux/seq_file.h> 27 #include <linux/kmod.h> 28 #include <linux/list.h> 29 #include <linux/hrtimer.h> 30 #include <linux/lockdep.h> 31 #include <linux/slab.h> 32 #include <linux/hashtable.h> 33 34 #include <net/net_namespace.h> 35 #include <net/sock.h> 36 #include <net/netlink.h> 37 #include <net/pkt_sched.h> 38 39 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 40 struct nlmsghdr *n, u32 clid, 41 struct Qdisc *old, struct Qdisc *new); 42 static int tclass_notify(struct net *net, struct sk_buff *oskb, 43 struct nlmsghdr *n, struct Qdisc *q, 44 unsigned long cl, int event); 45 46 /* 47 48 Short review. 49 ------------- 50 51 This file consists of two interrelated parts: 52 53 1. queueing disciplines manager frontend. 54 2. traffic classes manager frontend. 55 56 Generally, queueing discipline ("qdisc") is a black box, 57 which is able to enqueue packets and to dequeue them (when 58 device is ready to send something) in order and at times 59 determined by algorithm hidden in it. 60 61 qdisc's are divided to two categories: 62 - "queues", which have no internal structure visible from outside. 63 - "schedulers", which split all the packets to "traffic classes", 64 using "packet classifiers" (look at cls_api.c) 65 66 In turn, classes may have child qdiscs (as rule, queues) 67 attached to them etc. etc. etc. 68 69 The goal of the routines in this file is to translate 70 information supplied by user in the form of handles 71 to more intelligible for kernel form, to make some sanity 72 checks and part of work, which is common to all qdiscs 73 and to provide rtnetlink notifications. 74 75 All real intelligent work is done inside qdisc modules. 76 77 78 79 Every discipline has two major routines: enqueue and dequeue. 80 81 ---dequeue 82 83 dequeue usually returns a skb to send. It is allowed to return NULL, 84 but it does not mean that queue is empty, it just means that 85 discipline does not want to send anything this time. 86 Queue is really empty if q->q.qlen == 0. 87 For complicated disciplines with multiple queues q->q is not 88 real packet queue, but however q->q.qlen must be valid. 89 90 ---enqueue 91 92 enqueue returns 0, if packet was enqueued successfully. 93 If packet (this one or another one) was dropped, it returns 94 not zero error code. 95 NET_XMIT_DROP - this packet dropped 96 Expected action: do not backoff, but wait until queue will clear. 97 NET_XMIT_CN - probably this packet enqueued, but another one dropped. 98 Expected action: backoff or ignore 99 100 Auxiliary routines: 101 102 ---peek 103 104 like dequeue but without removing a packet from the queue 105 106 ---reset 107 108 returns qdisc to initial state: purge all buffers, clear all 109 timers, counters (except for statistics) etc. 110 111 ---init 112 113 initializes newly created qdisc. 114 115 ---destroy 116 117 destroys resources allocated by init and during lifetime of qdisc. 118 119 ---change 120 121 changes qdisc parameters. 122 */ 123 124 /* Protects list of registered TC modules. It is pure SMP lock. */ 125 static DEFINE_RWLOCK(qdisc_mod_lock); 126 127 128 /************************************************ 129 * Queueing disciplines manipulation. * 130 ************************************************/ 131 132 133 /* The list of all installed queueing disciplines. */ 134 135 static struct Qdisc_ops *qdisc_base; 136 137 /* Register/unregister queueing discipline */ 138 139 int register_qdisc(struct Qdisc_ops *qops) 140 { 141 struct Qdisc_ops *q, **qp; 142 int rc = -EEXIST; 143 144 write_lock(&qdisc_mod_lock); 145 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 146 if (!strcmp(qops->id, q->id)) 147 goto out; 148 149 if (qops->enqueue == NULL) 150 qops->enqueue = noop_qdisc_ops.enqueue; 151 if (qops->peek == NULL) { 152 if (qops->dequeue == NULL) 153 qops->peek = noop_qdisc_ops.peek; 154 else 155 goto out_einval; 156 } 157 if (qops->dequeue == NULL) 158 qops->dequeue = noop_qdisc_ops.dequeue; 159 160 if (qops->cl_ops) { 161 const struct Qdisc_class_ops *cops = qops->cl_ops; 162 163 if (!(cops->get && cops->put && cops->walk && cops->leaf)) 164 goto out_einval; 165 166 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf)) 167 goto out_einval; 168 } 169 170 qops->next = NULL; 171 *qp = qops; 172 rc = 0; 173 out: 174 write_unlock(&qdisc_mod_lock); 175 return rc; 176 177 out_einval: 178 rc = -EINVAL; 179 goto out; 180 } 181 EXPORT_SYMBOL(register_qdisc); 182 183 int unregister_qdisc(struct Qdisc_ops *qops) 184 { 185 struct Qdisc_ops *q, **qp; 186 int err = -ENOENT; 187 188 write_lock(&qdisc_mod_lock); 189 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 190 if (q == qops) 191 break; 192 if (q) { 193 *qp = q->next; 194 q->next = NULL; 195 err = 0; 196 } 197 write_unlock(&qdisc_mod_lock); 198 return err; 199 } 200 EXPORT_SYMBOL(unregister_qdisc); 201 202 /* Get default qdisc if not otherwise specified */ 203 void qdisc_get_default(char *name, size_t len) 204 { 205 read_lock(&qdisc_mod_lock); 206 strlcpy(name, default_qdisc_ops->id, len); 207 read_unlock(&qdisc_mod_lock); 208 } 209 210 static struct Qdisc_ops *qdisc_lookup_default(const char *name) 211 { 212 struct Qdisc_ops *q = NULL; 213 214 for (q = qdisc_base; q; q = q->next) { 215 if (!strcmp(name, q->id)) { 216 if (!try_module_get(q->owner)) 217 q = NULL; 218 break; 219 } 220 } 221 222 return q; 223 } 224 225 /* Set new default qdisc to use */ 226 int qdisc_set_default(const char *name) 227 { 228 const struct Qdisc_ops *ops; 229 230 if (!capable(CAP_NET_ADMIN)) 231 return -EPERM; 232 233 write_lock(&qdisc_mod_lock); 234 ops = qdisc_lookup_default(name); 235 if (!ops) { 236 /* Not found, drop lock and try to load module */ 237 write_unlock(&qdisc_mod_lock); 238 request_module("sch_%s", name); 239 write_lock(&qdisc_mod_lock); 240 241 ops = qdisc_lookup_default(name); 242 } 243 244 if (ops) { 245 /* Set new default */ 246 module_put(default_qdisc_ops->owner); 247 default_qdisc_ops = ops; 248 } 249 write_unlock(&qdisc_mod_lock); 250 251 return ops ? 0 : -ENOENT; 252 } 253 254 /* We know handle. Find qdisc among all qdisc's attached to device 255 * (root qdisc, all its children, children of children etc.) 256 * Note: caller either uses rtnl or rcu_read_lock() 257 */ 258 259 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle) 260 { 261 struct Qdisc *q; 262 263 if (!qdisc_dev(root)) 264 return (root->handle == handle ? root : NULL); 265 266 if (!(root->flags & TCQ_F_BUILTIN) && 267 root->handle == handle) 268 return root; 269 270 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) { 271 if (q->handle == handle) 272 return q; 273 } 274 return NULL; 275 } 276 277 void qdisc_hash_add(struct Qdisc *q) 278 { 279 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 280 struct Qdisc *root = qdisc_dev(q)->qdisc; 281 282 WARN_ON_ONCE(root == &noop_qdisc); 283 ASSERT_RTNL(); 284 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle); 285 } 286 } 287 EXPORT_SYMBOL(qdisc_hash_add); 288 289 void qdisc_hash_del(struct Qdisc *q) 290 { 291 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 292 ASSERT_RTNL(); 293 hash_del_rcu(&q->hash); 294 } 295 } 296 EXPORT_SYMBOL(qdisc_hash_del); 297 298 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) 299 { 300 struct Qdisc *q; 301 302 q = qdisc_match_from_root(dev->qdisc, handle); 303 if (q) 304 goto out; 305 306 if (dev_ingress_queue(dev)) 307 q = qdisc_match_from_root( 308 dev_ingress_queue(dev)->qdisc_sleeping, 309 handle); 310 out: 311 return q; 312 } 313 314 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) 315 { 316 unsigned long cl; 317 struct Qdisc *leaf; 318 const struct Qdisc_class_ops *cops = p->ops->cl_ops; 319 320 if (cops == NULL) 321 return NULL; 322 cl = cops->get(p, classid); 323 324 if (cl == 0) 325 return NULL; 326 leaf = cops->leaf(p, cl); 327 cops->put(p, cl); 328 return leaf; 329 } 330 331 /* Find queueing discipline by name */ 332 333 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) 334 { 335 struct Qdisc_ops *q = NULL; 336 337 if (kind) { 338 read_lock(&qdisc_mod_lock); 339 for (q = qdisc_base; q; q = q->next) { 340 if (nla_strcmp(kind, q->id) == 0) { 341 if (!try_module_get(q->owner)) 342 q = NULL; 343 break; 344 } 345 } 346 read_unlock(&qdisc_mod_lock); 347 } 348 return q; 349 } 350 351 /* The linklayer setting were not transferred from iproute2, in older 352 * versions, and the rate tables lookup systems have been dropped in 353 * the kernel. To keep backward compatible with older iproute2 tc 354 * utils, we detect the linklayer setting by detecting if the rate 355 * table were modified. 356 * 357 * For linklayer ATM table entries, the rate table will be aligned to 358 * 48 bytes, thus some table entries will contain the same value. The 359 * mpu (min packet unit) is also encoded into the old rate table, thus 360 * starting from the mpu, we find low and high table entries for 361 * mapping this cell. If these entries contain the same value, when 362 * the rate tables have been modified for linklayer ATM. 363 * 364 * This is done by rounding mpu to the nearest 48 bytes cell/entry, 365 * and then roundup to the next cell, calc the table entry one below, 366 * and compare. 367 */ 368 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab) 369 { 370 int low = roundup(r->mpu, 48); 371 int high = roundup(low+1, 48); 372 int cell_low = low >> r->cell_log; 373 int cell_high = (high >> r->cell_log) - 1; 374 375 /* rtab is too inaccurate at rates > 100Mbit/s */ 376 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) { 377 pr_debug("TC linklayer: Giving up ATM detection\n"); 378 return TC_LINKLAYER_ETHERNET; 379 } 380 381 if ((cell_high > cell_low) && (cell_high < 256) 382 && (rtab[cell_low] == rtab[cell_high])) { 383 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n", 384 cell_low, cell_high, rtab[cell_high]); 385 return TC_LINKLAYER_ATM; 386 } 387 return TC_LINKLAYER_ETHERNET; 388 } 389 390 static struct qdisc_rate_table *qdisc_rtab_list; 391 392 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, 393 struct nlattr *tab) 394 { 395 struct qdisc_rate_table *rtab; 396 397 if (tab == NULL || r->rate == 0 || r->cell_log == 0 || 398 nla_len(tab) != TC_RTAB_SIZE) 399 return NULL; 400 401 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { 402 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) && 403 !memcmp(&rtab->data, nla_data(tab), 1024)) { 404 rtab->refcnt++; 405 return rtab; 406 } 407 } 408 409 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); 410 if (rtab) { 411 rtab->rate = *r; 412 rtab->refcnt = 1; 413 memcpy(rtab->data, nla_data(tab), 1024); 414 if (r->linklayer == TC_LINKLAYER_UNAWARE) 415 r->linklayer = __detect_linklayer(r, rtab->data); 416 rtab->next = qdisc_rtab_list; 417 qdisc_rtab_list = rtab; 418 } 419 return rtab; 420 } 421 EXPORT_SYMBOL(qdisc_get_rtab); 422 423 void qdisc_put_rtab(struct qdisc_rate_table *tab) 424 { 425 struct qdisc_rate_table *rtab, **rtabp; 426 427 if (!tab || --tab->refcnt) 428 return; 429 430 for (rtabp = &qdisc_rtab_list; 431 (rtab = *rtabp) != NULL; 432 rtabp = &rtab->next) { 433 if (rtab == tab) { 434 *rtabp = rtab->next; 435 kfree(rtab); 436 return; 437 } 438 } 439 } 440 EXPORT_SYMBOL(qdisc_put_rtab); 441 442 static LIST_HEAD(qdisc_stab_list); 443 static DEFINE_SPINLOCK(qdisc_stab_lock); 444 445 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { 446 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, 447 [TCA_STAB_DATA] = { .type = NLA_BINARY }, 448 }; 449 450 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) 451 { 452 struct nlattr *tb[TCA_STAB_MAX + 1]; 453 struct qdisc_size_table *stab; 454 struct tc_sizespec *s; 455 unsigned int tsize = 0; 456 u16 *tab = NULL; 457 int err; 458 459 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy); 460 if (err < 0) 461 return ERR_PTR(err); 462 if (!tb[TCA_STAB_BASE]) 463 return ERR_PTR(-EINVAL); 464 465 s = nla_data(tb[TCA_STAB_BASE]); 466 467 if (s->tsize > 0) { 468 if (!tb[TCA_STAB_DATA]) 469 return ERR_PTR(-EINVAL); 470 tab = nla_data(tb[TCA_STAB_DATA]); 471 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); 472 } 473 474 if (tsize != s->tsize || (!tab && tsize > 0)) 475 return ERR_PTR(-EINVAL); 476 477 spin_lock(&qdisc_stab_lock); 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 spin_unlock(&qdisc_stab_lock); 486 return stab; 487 } 488 489 spin_unlock(&qdisc_stab_lock); 490 491 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); 492 if (!stab) 493 return ERR_PTR(-ENOMEM); 494 495 stab->refcnt = 1; 496 stab->szopts = *s; 497 if (tsize > 0) 498 memcpy(stab->data, tab, tsize * sizeof(u16)); 499 500 spin_lock(&qdisc_stab_lock); 501 list_add_tail(&stab->list, &qdisc_stab_list); 502 spin_unlock(&qdisc_stab_lock); 503 504 return stab; 505 } 506 507 static void stab_kfree_rcu(struct rcu_head *head) 508 { 509 kfree(container_of(head, struct qdisc_size_table, rcu)); 510 } 511 512 void qdisc_put_stab(struct qdisc_size_table *tab) 513 { 514 if (!tab) 515 return; 516 517 spin_lock(&qdisc_stab_lock); 518 519 if (--tab->refcnt == 0) { 520 list_del(&tab->list); 521 call_rcu_bh(&tab->rcu, stab_kfree_rcu); 522 } 523 524 spin_unlock(&qdisc_stab_lock); 525 } 526 EXPORT_SYMBOL(qdisc_put_stab); 527 528 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) 529 { 530 struct nlattr *nest; 531 532 nest = nla_nest_start(skb, TCA_STAB); 533 if (nest == NULL) 534 goto nla_put_failure; 535 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts)) 536 goto nla_put_failure; 537 nla_nest_end(skb, nest); 538 539 return skb->len; 540 541 nla_put_failure: 542 return -1; 543 } 544 545 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 546 const struct qdisc_size_table *stab) 547 { 548 int pkt_len, slot; 549 550 pkt_len = skb->len + stab->szopts.overhead; 551 if (unlikely(!stab->szopts.tsize)) 552 goto out; 553 554 slot = pkt_len + stab->szopts.cell_align; 555 if (unlikely(slot < 0)) 556 slot = 0; 557 558 slot >>= stab->szopts.cell_log; 559 if (likely(slot < stab->szopts.tsize)) 560 pkt_len = stab->data[slot]; 561 else 562 pkt_len = stab->data[stab->szopts.tsize - 1] * 563 (slot / stab->szopts.tsize) + 564 stab->data[slot % stab->szopts.tsize]; 565 566 pkt_len <<= stab->szopts.size_log; 567 out: 568 if (unlikely(pkt_len < 1)) 569 pkt_len = 1; 570 qdisc_skb_cb(skb)->pkt_len = pkt_len; 571 } 572 EXPORT_SYMBOL(__qdisc_calculate_pkt_len); 573 574 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc) 575 { 576 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { 577 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n", 578 txt, qdisc->ops->id, qdisc->handle >> 16); 579 qdisc->flags |= TCQ_F_WARN_NONWC; 580 } 581 } 582 EXPORT_SYMBOL(qdisc_warn_nonwc); 583 584 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) 585 { 586 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, 587 timer); 588 589 rcu_read_lock(); 590 __netif_schedule(qdisc_root(wd->qdisc)); 591 rcu_read_unlock(); 592 593 return HRTIMER_NORESTART; 594 } 595 596 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 597 { 598 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED); 599 wd->timer.function = qdisc_watchdog; 600 wd->qdisc = qdisc; 601 } 602 EXPORT_SYMBOL(qdisc_watchdog_init); 603 604 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires) 605 { 606 if (test_bit(__QDISC_STATE_DEACTIVATED, 607 &qdisc_root_sleeping(wd->qdisc)->state)) 608 return; 609 610 if (wd->last_expires == expires) 611 return; 612 613 wd->last_expires = expires; 614 hrtimer_start(&wd->timer, 615 ns_to_ktime(expires), 616 HRTIMER_MODE_ABS_PINNED); 617 } 618 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns); 619 620 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) 621 { 622 hrtimer_cancel(&wd->timer); 623 } 624 EXPORT_SYMBOL(qdisc_watchdog_cancel); 625 626 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) 627 { 628 unsigned int size = n * sizeof(struct hlist_head), i; 629 struct hlist_head *h; 630 631 if (size <= PAGE_SIZE) 632 h = kmalloc(size, GFP_KERNEL); 633 else 634 h = (struct hlist_head *) 635 __get_free_pages(GFP_KERNEL, get_order(size)); 636 637 if (h != NULL) { 638 for (i = 0; i < n; i++) 639 INIT_HLIST_HEAD(&h[i]); 640 } 641 return h; 642 } 643 644 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) 645 { 646 unsigned int size = n * sizeof(struct hlist_head); 647 648 if (size <= PAGE_SIZE) 649 kfree(h); 650 else 651 free_pages((unsigned long)h, get_order(size)); 652 } 653 654 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) 655 { 656 struct Qdisc_class_common *cl; 657 struct hlist_node *next; 658 struct hlist_head *nhash, *ohash; 659 unsigned int nsize, nmask, osize; 660 unsigned int i, h; 661 662 /* Rehash when load factor exceeds 0.75 */ 663 if (clhash->hashelems * 4 <= clhash->hashsize * 3) 664 return; 665 nsize = clhash->hashsize * 2; 666 nmask = nsize - 1; 667 nhash = qdisc_class_hash_alloc(nsize); 668 if (nhash == NULL) 669 return; 670 671 ohash = clhash->hash; 672 osize = clhash->hashsize; 673 674 sch_tree_lock(sch); 675 for (i = 0; i < osize; i++) { 676 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) { 677 h = qdisc_class_hash(cl->classid, nmask); 678 hlist_add_head(&cl->hnode, &nhash[h]); 679 } 680 } 681 clhash->hash = nhash; 682 clhash->hashsize = nsize; 683 clhash->hashmask = nmask; 684 sch_tree_unlock(sch); 685 686 qdisc_class_hash_free(ohash, osize); 687 } 688 EXPORT_SYMBOL(qdisc_class_hash_grow); 689 690 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) 691 { 692 unsigned int size = 4; 693 694 clhash->hash = qdisc_class_hash_alloc(size); 695 if (clhash->hash == NULL) 696 return -ENOMEM; 697 clhash->hashsize = size; 698 clhash->hashmask = size - 1; 699 clhash->hashelems = 0; 700 return 0; 701 } 702 EXPORT_SYMBOL(qdisc_class_hash_init); 703 704 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) 705 { 706 qdisc_class_hash_free(clhash->hash, clhash->hashsize); 707 } 708 EXPORT_SYMBOL(qdisc_class_hash_destroy); 709 710 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, 711 struct Qdisc_class_common *cl) 712 { 713 unsigned int h; 714 715 INIT_HLIST_NODE(&cl->hnode); 716 h = qdisc_class_hash(cl->classid, clhash->hashmask); 717 hlist_add_head(&cl->hnode, &clhash->hash[h]); 718 clhash->hashelems++; 719 } 720 EXPORT_SYMBOL(qdisc_class_hash_insert); 721 722 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, 723 struct Qdisc_class_common *cl) 724 { 725 hlist_del(&cl->hnode); 726 clhash->hashelems--; 727 } 728 EXPORT_SYMBOL(qdisc_class_hash_remove); 729 730 /* Allocate an unique handle from space managed by kernel 731 * Possible range is [8000-FFFF]:0000 (0x8000 values) 732 */ 733 static u32 qdisc_alloc_handle(struct net_device *dev) 734 { 735 int i = 0x8000; 736 static u32 autohandle = TC_H_MAKE(0x80000000U, 0); 737 738 do { 739 autohandle += TC_H_MAKE(0x10000U, 0); 740 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) 741 autohandle = TC_H_MAKE(0x80000000U, 0); 742 if (!qdisc_lookup(dev, autohandle)) 743 return autohandle; 744 cond_resched(); 745 } while (--i > 0); 746 747 return 0; 748 } 749 750 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n, 751 unsigned int len) 752 { 753 const struct Qdisc_class_ops *cops; 754 unsigned long cl; 755 u32 parentid; 756 int drops; 757 758 if (n == 0 && len == 0) 759 return; 760 drops = max_t(int, n, 0); 761 rcu_read_lock(); 762 while ((parentid = sch->parent)) { 763 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) 764 break; 765 766 if (sch->flags & TCQ_F_NOPARENT) 767 break; 768 /* TODO: perform the search on a per txq basis */ 769 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); 770 if (sch == NULL) { 771 WARN_ON_ONCE(parentid != TC_H_ROOT); 772 break; 773 } 774 cops = sch->ops->cl_ops; 775 if (cops->qlen_notify) { 776 cl = cops->get(sch, parentid); 777 cops->qlen_notify(sch, cl); 778 cops->put(sch, cl); 779 } 780 sch->q.qlen -= n; 781 sch->qstats.backlog -= len; 782 __qdisc_qstats_drop(sch, drops); 783 } 784 rcu_read_unlock(); 785 } 786 EXPORT_SYMBOL(qdisc_tree_reduce_backlog); 787 788 static void notify_and_destroy(struct net *net, struct sk_buff *skb, 789 struct nlmsghdr *n, u32 clid, 790 struct Qdisc *old, struct Qdisc *new) 791 { 792 if (new || old) 793 qdisc_notify(net, skb, n, clid, old, new); 794 795 if (old) 796 qdisc_destroy(old); 797 } 798 799 /* Graft qdisc "new" to class "classid" of qdisc "parent" or 800 * to device "dev". 801 * 802 * When appropriate send a netlink notification using 'skb' 803 * and "n". 804 * 805 * On success, destroy old qdisc. 806 */ 807 808 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 809 struct sk_buff *skb, struct nlmsghdr *n, u32 classid, 810 struct Qdisc *new, struct Qdisc *old) 811 { 812 struct Qdisc *q = old; 813 struct net *net = dev_net(dev); 814 int err = 0; 815 816 if (parent == NULL) { 817 unsigned int i, num_q, ingress; 818 819 ingress = 0; 820 num_q = dev->num_tx_queues; 821 if ((q && q->flags & TCQ_F_INGRESS) || 822 (new && new->flags & TCQ_F_INGRESS)) { 823 num_q = 1; 824 ingress = 1; 825 if (!dev_ingress_queue(dev)) 826 return -ENOENT; 827 } 828 829 if (dev->flags & IFF_UP) 830 dev_deactivate(dev); 831 832 if (new && new->ops->attach) 833 goto skip; 834 835 for (i = 0; i < num_q; i++) { 836 struct netdev_queue *dev_queue = dev_ingress_queue(dev); 837 838 if (!ingress) 839 dev_queue = netdev_get_tx_queue(dev, i); 840 841 old = dev_graft_qdisc(dev_queue, new); 842 if (new && i > 0) 843 atomic_inc(&new->refcnt); 844 845 if (!ingress) 846 qdisc_destroy(old); 847 } 848 849 skip: 850 if (!ingress) { 851 notify_and_destroy(net, skb, n, classid, 852 dev->qdisc, new); 853 if (new && !new->ops->attach) 854 atomic_inc(&new->refcnt); 855 dev->qdisc = new ? : &noop_qdisc; 856 857 if (new && new->ops->attach) 858 new->ops->attach(new); 859 } else { 860 notify_and_destroy(net, skb, n, classid, old, new); 861 } 862 863 if (dev->flags & IFF_UP) 864 dev_activate(dev); 865 } else { 866 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 867 868 err = -EOPNOTSUPP; 869 if (cops && cops->graft) { 870 unsigned long cl = cops->get(parent, classid); 871 if (cl) { 872 err = cops->graft(parent, cl, new, &old); 873 cops->put(parent, cl); 874 } else 875 err = -ENOENT; 876 } 877 if (!err) 878 notify_and_destroy(net, skb, n, classid, old, new); 879 } 880 return err; 881 } 882 883 /* lockdep annotation is needed for ingress; egress gets it only for name */ 884 static struct lock_class_key qdisc_tx_lock; 885 static struct lock_class_key qdisc_rx_lock; 886 887 /* 888 Allocate and initialize new qdisc. 889 890 Parameters are passed via opt. 891 */ 892 893 static struct Qdisc *qdisc_create(struct net_device *dev, 894 struct netdev_queue *dev_queue, 895 struct Qdisc *p, u32 parent, u32 handle, 896 struct nlattr **tca, int *errp) 897 { 898 int err; 899 struct nlattr *kind = tca[TCA_KIND]; 900 struct Qdisc *sch; 901 struct Qdisc_ops *ops; 902 struct qdisc_size_table *stab; 903 904 ops = qdisc_lookup_ops(kind); 905 #ifdef CONFIG_MODULES 906 if (ops == NULL && kind != NULL) { 907 char name[IFNAMSIZ]; 908 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 909 /* We dropped the RTNL semaphore in order to 910 * perform the module load. So, even if we 911 * succeeded in loading the module we have to 912 * tell the caller to replay the request. We 913 * indicate this using -EAGAIN. 914 * We replay the request because the device may 915 * go away in the mean time. 916 */ 917 rtnl_unlock(); 918 request_module("sch_%s", name); 919 rtnl_lock(); 920 ops = qdisc_lookup_ops(kind); 921 if (ops != NULL) { 922 /* We will try again qdisc_lookup_ops, 923 * so don't keep a reference. 924 */ 925 module_put(ops->owner); 926 err = -EAGAIN; 927 goto err_out; 928 } 929 } 930 } 931 #endif 932 933 err = -ENOENT; 934 if (ops == NULL) 935 goto err_out; 936 937 sch = qdisc_alloc(dev_queue, ops); 938 if (IS_ERR(sch)) { 939 err = PTR_ERR(sch); 940 goto err_out2; 941 } 942 943 sch->parent = parent; 944 945 if (handle == TC_H_INGRESS) { 946 sch->flags |= TCQ_F_INGRESS; 947 handle = TC_H_MAKE(TC_H_INGRESS, 0); 948 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); 949 } else { 950 if (handle == 0) { 951 handle = qdisc_alloc_handle(dev); 952 err = -ENOMEM; 953 if (handle == 0) 954 goto err_out3; 955 } 956 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); 957 if (!netif_is_multiqueue(dev)) 958 sch->flags |= TCQ_F_ONETXQUEUE; 959 } 960 961 sch->handle = handle; 962 963 /* This exist to keep backward compatible with a userspace 964 * loophole, what allowed userspace to get IFF_NO_QUEUE 965 * facility on older kernels by setting tx_queue_len=0 (prior 966 * to qdisc init), and then forgot to reinit tx_queue_len 967 * before again attaching a qdisc. 968 */ 969 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) { 970 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 971 netdev_info(dev, "Caught tx_queue_len zero misconfig\n"); 972 } 973 974 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { 975 if (qdisc_is_percpu_stats(sch)) { 976 sch->cpu_bstats = 977 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu); 978 if (!sch->cpu_bstats) 979 goto err_out4; 980 981 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue); 982 if (!sch->cpu_qstats) 983 goto err_out4; 984 } 985 986 if (tca[TCA_STAB]) { 987 stab = qdisc_get_stab(tca[TCA_STAB]); 988 if (IS_ERR(stab)) { 989 err = PTR_ERR(stab); 990 goto err_out4; 991 } 992 rcu_assign_pointer(sch->stab, stab); 993 } 994 if (tca[TCA_RATE]) { 995 seqcount_t *running; 996 997 err = -EOPNOTSUPP; 998 if (sch->flags & TCQ_F_MQROOT) 999 goto err_out4; 1000 1001 if ((sch->parent != TC_H_ROOT) && 1002 !(sch->flags & TCQ_F_INGRESS) && 1003 (!p || !(p->flags & TCQ_F_MQROOT))) 1004 running = qdisc_root_sleeping_running(sch); 1005 else 1006 running = &sch->running; 1007 1008 err = gen_new_estimator(&sch->bstats, 1009 sch->cpu_bstats, 1010 &sch->rate_est, 1011 NULL, 1012 running, 1013 tca[TCA_RATE]); 1014 if (err) 1015 goto err_out4; 1016 } 1017 1018 qdisc_hash_add(sch); 1019 1020 return sch; 1021 } 1022 err_out3: 1023 dev_put(dev); 1024 kfree((char *) sch - sch->padded); 1025 err_out2: 1026 module_put(ops->owner); 1027 err_out: 1028 *errp = err; 1029 return NULL; 1030 1031 err_out4: 1032 free_percpu(sch->cpu_bstats); 1033 free_percpu(sch->cpu_qstats); 1034 /* 1035 * Any broken qdiscs that would require a ops->reset() here? 1036 * The qdisc was never in action so it shouldn't be necessary. 1037 */ 1038 qdisc_put_stab(rtnl_dereference(sch->stab)); 1039 if (ops->destroy) 1040 ops->destroy(sch); 1041 goto err_out3; 1042 } 1043 1044 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) 1045 { 1046 struct qdisc_size_table *ostab, *stab = NULL; 1047 int err = 0; 1048 1049 if (tca[TCA_OPTIONS]) { 1050 if (sch->ops->change == NULL) 1051 return -EINVAL; 1052 err = sch->ops->change(sch, tca[TCA_OPTIONS]); 1053 if (err) 1054 return err; 1055 } 1056 1057 if (tca[TCA_STAB]) { 1058 stab = qdisc_get_stab(tca[TCA_STAB]); 1059 if (IS_ERR(stab)) 1060 return PTR_ERR(stab); 1061 } 1062 1063 ostab = rtnl_dereference(sch->stab); 1064 rcu_assign_pointer(sch->stab, stab); 1065 qdisc_put_stab(ostab); 1066 1067 if (tca[TCA_RATE]) { 1068 /* NB: ignores errors from replace_estimator 1069 because change can't be undone. */ 1070 if (sch->flags & TCQ_F_MQROOT) 1071 goto out; 1072 gen_replace_estimator(&sch->bstats, 1073 sch->cpu_bstats, 1074 &sch->rate_est, 1075 NULL, 1076 qdisc_root_sleeping_running(sch), 1077 tca[TCA_RATE]); 1078 } 1079 out: 1080 return 0; 1081 } 1082 1083 struct check_loop_arg { 1084 struct qdisc_walker w; 1085 struct Qdisc *p; 1086 int depth; 1087 }; 1088 1089 static int check_loop_fn(struct Qdisc *q, unsigned long cl, 1090 struct qdisc_walker *w); 1091 1092 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 1093 { 1094 struct check_loop_arg arg; 1095 1096 if (q->ops->cl_ops == NULL) 1097 return 0; 1098 1099 arg.w.stop = arg.w.skip = arg.w.count = 0; 1100 arg.w.fn = check_loop_fn; 1101 arg.depth = depth; 1102 arg.p = p; 1103 q->ops->cl_ops->walk(q, &arg.w); 1104 return arg.w.stop ? -ELOOP : 0; 1105 } 1106 1107 static int 1108 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 1109 { 1110 struct Qdisc *leaf; 1111 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1112 struct check_loop_arg *arg = (struct check_loop_arg *)w; 1113 1114 leaf = cops->leaf(q, cl); 1115 if (leaf) { 1116 if (leaf == arg->p || arg->depth > 7) 1117 return -ELOOP; 1118 return check_loop(leaf, arg->p, arg->depth + 1); 1119 } 1120 return 0; 1121 } 1122 1123 /* 1124 * Delete/get qdisc. 1125 */ 1126 1127 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n) 1128 { 1129 struct net *net = sock_net(skb->sk); 1130 struct tcmsg *tcm = nlmsg_data(n); 1131 struct nlattr *tca[TCA_MAX + 1]; 1132 struct net_device *dev; 1133 u32 clid; 1134 struct Qdisc *q = NULL; 1135 struct Qdisc *p = NULL; 1136 int err; 1137 1138 if ((n->nlmsg_type != RTM_GETQDISC) && 1139 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1140 return -EPERM; 1141 1142 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1143 if (err < 0) 1144 return err; 1145 1146 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1147 if (!dev) 1148 return -ENODEV; 1149 1150 clid = tcm->tcm_parent; 1151 if (clid) { 1152 if (clid != TC_H_ROOT) { 1153 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 1154 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1155 if (!p) 1156 return -ENOENT; 1157 q = qdisc_leaf(p, clid); 1158 } else if (dev_ingress_queue(dev)) { 1159 q = dev_ingress_queue(dev)->qdisc_sleeping; 1160 } 1161 } else { 1162 q = dev->qdisc; 1163 } 1164 if (!q) 1165 return -ENOENT; 1166 1167 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) 1168 return -EINVAL; 1169 } else { 1170 q = qdisc_lookup(dev, tcm->tcm_handle); 1171 if (!q) 1172 return -ENOENT; 1173 } 1174 1175 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1176 return -EINVAL; 1177 1178 if (n->nlmsg_type == RTM_DELQDISC) { 1179 if (!clid) 1180 return -EINVAL; 1181 if (q->handle == 0) 1182 return -ENOENT; 1183 err = qdisc_graft(dev, p, skb, n, clid, NULL, q); 1184 if (err != 0) 1185 return err; 1186 } else { 1187 qdisc_notify(net, skb, n, clid, NULL, q); 1188 } 1189 return 0; 1190 } 1191 1192 /* 1193 * Create/change qdisc. 1194 */ 1195 1196 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n) 1197 { 1198 struct net *net = sock_net(skb->sk); 1199 struct tcmsg *tcm; 1200 struct nlattr *tca[TCA_MAX + 1]; 1201 struct net_device *dev; 1202 u32 clid; 1203 struct Qdisc *q, *p; 1204 int err; 1205 1206 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1207 return -EPERM; 1208 1209 replay: 1210 /* Reinit, just in case something touches this. */ 1211 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1212 if (err < 0) 1213 return err; 1214 1215 tcm = nlmsg_data(n); 1216 clid = tcm->tcm_parent; 1217 q = p = NULL; 1218 1219 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1220 if (!dev) 1221 return -ENODEV; 1222 1223 1224 if (clid) { 1225 if (clid != TC_H_ROOT) { 1226 if (clid != TC_H_INGRESS) { 1227 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1228 if (!p) 1229 return -ENOENT; 1230 q = qdisc_leaf(p, clid); 1231 } else if (dev_ingress_queue_create(dev)) { 1232 q = dev_ingress_queue(dev)->qdisc_sleeping; 1233 } 1234 } else { 1235 q = dev->qdisc; 1236 } 1237 1238 /* It may be default qdisc, ignore it */ 1239 if (q && q->handle == 0) 1240 q = NULL; 1241 1242 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 1243 if (tcm->tcm_handle) { 1244 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) 1245 return -EEXIST; 1246 if (TC_H_MIN(tcm->tcm_handle)) 1247 return -EINVAL; 1248 q = qdisc_lookup(dev, tcm->tcm_handle); 1249 if (!q) 1250 goto create_n_graft; 1251 if (n->nlmsg_flags & NLM_F_EXCL) 1252 return -EEXIST; 1253 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1254 return -EINVAL; 1255 if (q == p || 1256 (p && check_loop(q, p, 0))) 1257 return -ELOOP; 1258 atomic_inc(&q->refcnt); 1259 goto graft; 1260 } else { 1261 if (!q) 1262 goto create_n_graft; 1263 1264 /* This magic test requires explanation. 1265 * 1266 * We know, that some child q is already 1267 * attached to this parent and have choice: 1268 * either to change it or to create/graft new one. 1269 * 1270 * 1. We are allowed to create/graft only 1271 * if CREATE and REPLACE flags are set. 1272 * 1273 * 2. If EXCL is set, requestor wanted to say, 1274 * that qdisc tcm_handle is not expected 1275 * to exist, so that we choose create/graft too. 1276 * 1277 * 3. The last case is when no flags are set. 1278 * Alas, it is sort of hole in API, we 1279 * cannot decide what to do unambiguously. 1280 * For now we select create/graft, if 1281 * user gave KIND, which does not match existing. 1282 */ 1283 if ((n->nlmsg_flags & NLM_F_CREATE) && 1284 (n->nlmsg_flags & NLM_F_REPLACE) && 1285 ((n->nlmsg_flags & NLM_F_EXCL) || 1286 (tca[TCA_KIND] && 1287 nla_strcmp(tca[TCA_KIND], q->ops->id)))) 1288 goto create_n_graft; 1289 } 1290 } 1291 } else { 1292 if (!tcm->tcm_handle) 1293 return -EINVAL; 1294 q = qdisc_lookup(dev, tcm->tcm_handle); 1295 } 1296 1297 /* Change qdisc parameters */ 1298 if (q == NULL) 1299 return -ENOENT; 1300 if (n->nlmsg_flags & NLM_F_EXCL) 1301 return -EEXIST; 1302 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1303 return -EINVAL; 1304 err = qdisc_change(q, tca); 1305 if (err == 0) 1306 qdisc_notify(net, skb, n, clid, NULL, q); 1307 return err; 1308 1309 create_n_graft: 1310 if (!(n->nlmsg_flags & NLM_F_CREATE)) 1311 return -ENOENT; 1312 if (clid == TC_H_INGRESS) { 1313 if (dev_ingress_queue(dev)) 1314 q = qdisc_create(dev, dev_ingress_queue(dev), p, 1315 tcm->tcm_parent, tcm->tcm_parent, 1316 tca, &err); 1317 else 1318 err = -ENOENT; 1319 } else { 1320 struct netdev_queue *dev_queue; 1321 1322 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) 1323 dev_queue = p->ops->cl_ops->select_queue(p, tcm); 1324 else if (p) 1325 dev_queue = p->dev_queue; 1326 else 1327 dev_queue = netdev_get_tx_queue(dev, 0); 1328 1329 q = qdisc_create(dev, dev_queue, p, 1330 tcm->tcm_parent, tcm->tcm_handle, 1331 tca, &err); 1332 } 1333 if (q == NULL) { 1334 if (err == -EAGAIN) 1335 goto replay; 1336 return err; 1337 } 1338 1339 graft: 1340 err = qdisc_graft(dev, p, skb, n, clid, q, NULL); 1341 if (err) { 1342 if (q) 1343 qdisc_destroy(q); 1344 return err; 1345 } 1346 1347 return 0; 1348 } 1349 1350 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, 1351 u32 portid, u32 seq, u16 flags, int event) 1352 { 1353 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL; 1354 struct gnet_stats_queue __percpu *cpu_qstats = NULL; 1355 struct tcmsg *tcm; 1356 struct nlmsghdr *nlh; 1357 unsigned char *b = skb_tail_pointer(skb); 1358 struct gnet_dump d; 1359 struct qdisc_size_table *stab; 1360 __u32 qlen; 1361 1362 cond_resched(); 1363 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1364 if (!nlh) 1365 goto out_nlmsg_trim; 1366 tcm = nlmsg_data(nlh); 1367 tcm->tcm_family = AF_UNSPEC; 1368 tcm->tcm__pad1 = 0; 1369 tcm->tcm__pad2 = 0; 1370 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1371 tcm->tcm_parent = clid; 1372 tcm->tcm_handle = q->handle; 1373 tcm->tcm_info = atomic_read(&q->refcnt); 1374 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1375 goto nla_put_failure; 1376 if (q->ops->dump && q->ops->dump(q, skb) < 0) 1377 goto nla_put_failure; 1378 qlen = q->q.qlen; 1379 1380 stab = rtnl_dereference(q->stab); 1381 if (stab && qdisc_dump_stab(skb, stab) < 0) 1382 goto nla_put_failure; 1383 1384 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1385 NULL, &d, TCA_PAD) < 0) 1386 goto nla_put_failure; 1387 1388 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 1389 goto nla_put_failure; 1390 1391 if (qdisc_is_percpu_stats(q)) { 1392 cpu_bstats = q->cpu_bstats; 1393 cpu_qstats = q->cpu_qstats; 1394 } 1395 1396 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q), 1397 &d, cpu_bstats, &q->bstats) < 0 || 1398 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || 1399 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0) 1400 goto nla_put_failure; 1401 1402 if (gnet_stats_finish_copy(&d) < 0) 1403 goto nla_put_failure; 1404 1405 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1406 return skb->len; 1407 1408 out_nlmsg_trim: 1409 nla_put_failure: 1410 nlmsg_trim(skb, b); 1411 return -1; 1412 } 1413 1414 static bool tc_qdisc_dump_ignore(struct Qdisc *q) 1415 { 1416 return (q->flags & TCQ_F_BUILTIN) ? true : false; 1417 } 1418 1419 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 1420 struct nlmsghdr *n, u32 clid, 1421 struct Qdisc *old, struct Qdisc *new) 1422 { 1423 struct sk_buff *skb; 1424 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1425 1426 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1427 if (!skb) 1428 return -ENOBUFS; 1429 1430 if (old && !tc_qdisc_dump_ignore(old)) { 1431 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq, 1432 0, RTM_DELQDISC) < 0) 1433 goto err_out; 1434 } 1435 if (new && !tc_qdisc_dump_ignore(new)) { 1436 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq, 1437 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) 1438 goto err_out; 1439 } 1440 1441 if (skb->len) 1442 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1443 n->nlmsg_flags & NLM_F_ECHO); 1444 1445 err_out: 1446 kfree_skb(skb); 1447 return -EINVAL; 1448 } 1449 1450 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, 1451 struct netlink_callback *cb, 1452 int *q_idx_p, int s_q_idx, bool recur) 1453 { 1454 int ret = 0, q_idx = *q_idx_p; 1455 struct Qdisc *q; 1456 int b; 1457 1458 if (!root) 1459 return 0; 1460 1461 q = root; 1462 if (q_idx < s_q_idx) { 1463 q_idx++; 1464 } else { 1465 if (!tc_qdisc_dump_ignore(q) && 1466 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1467 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1468 RTM_NEWQDISC) <= 0) 1469 goto done; 1470 q_idx++; 1471 } 1472 1473 /* If dumping singletons, there is no qdisc_dev(root) and the singleton 1474 * itself has already been dumped. 1475 * 1476 * If we've already dumped the top-level (ingress) qdisc above and the global 1477 * qdisc hashtable, we don't want to hit it again 1478 */ 1479 if (!qdisc_dev(root) || !recur) 1480 goto out; 1481 1482 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1483 if (q_idx < s_q_idx) { 1484 q_idx++; 1485 continue; 1486 } 1487 if (!tc_qdisc_dump_ignore(q) && 1488 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1489 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1490 RTM_NEWQDISC) <= 0) 1491 goto done; 1492 q_idx++; 1493 } 1494 1495 out: 1496 *q_idx_p = q_idx; 1497 return ret; 1498 done: 1499 ret = -1; 1500 goto out; 1501 } 1502 1503 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 1504 { 1505 struct net *net = sock_net(skb->sk); 1506 int idx, q_idx; 1507 int s_idx, s_q_idx; 1508 struct net_device *dev; 1509 1510 s_idx = cb->args[0]; 1511 s_q_idx = q_idx = cb->args[1]; 1512 1513 idx = 0; 1514 ASSERT_RTNL(); 1515 for_each_netdev(net, dev) { 1516 struct netdev_queue *dev_queue; 1517 1518 if (idx < s_idx) 1519 goto cont; 1520 if (idx > s_idx) 1521 s_q_idx = 0; 1522 q_idx = 0; 1523 1524 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx, 1525 true) < 0) 1526 goto done; 1527 1528 dev_queue = dev_ingress_queue(dev); 1529 if (dev_queue && 1530 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, 1531 &q_idx, s_q_idx, false) < 0) 1532 goto done; 1533 1534 cont: 1535 idx++; 1536 } 1537 1538 done: 1539 cb->args[0] = idx; 1540 cb->args[1] = q_idx; 1541 1542 return skb->len; 1543 } 1544 1545 1546 1547 /************************************************ 1548 * Traffic classes manipulation. * 1549 ************************************************/ 1550 1551 1552 1553 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n) 1554 { 1555 struct net *net = sock_net(skb->sk); 1556 struct tcmsg *tcm = nlmsg_data(n); 1557 struct nlattr *tca[TCA_MAX + 1]; 1558 struct net_device *dev; 1559 struct Qdisc *q = NULL; 1560 const struct Qdisc_class_ops *cops; 1561 unsigned long cl = 0; 1562 unsigned long new_cl; 1563 u32 portid; 1564 u32 clid; 1565 u32 qid; 1566 int err; 1567 1568 if ((n->nlmsg_type != RTM_GETTCLASS) && 1569 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1570 return -EPERM; 1571 1572 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1573 if (err < 0) 1574 return err; 1575 1576 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1577 if (!dev) 1578 return -ENODEV; 1579 1580 /* 1581 parent == TC_H_UNSPEC - unspecified parent. 1582 parent == TC_H_ROOT - class is root, which has no parent. 1583 parent == X:0 - parent is root class. 1584 parent == X:Y - parent is a node in hierarchy. 1585 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 1586 1587 handle == 0:0 - generate handle from kernel pool. 1588 handle == 0:Y - class is X:Y, where X:0 is qdisc. 1589 handle == X:Y - clear. 1590 handle == X:0 - root class. 1591 */ 1592 1593 /* Step 1. Determine qdisc handle X:0 */ 1594 1595 portid = tcm->tcm_parent; 1596 clid = tcm->tcm_handle; 1597 qid = TC_H_MAJ(clid); 1598 1599 if (portid != TC_H_ROOT) { 1600 u32 qid1 = TC_H_MAJ(portid); 1601 1602 if (qid && qid1) { 1603 /* If both majors are known, they must be identical. */ 1604 if (qid != qid1) 1605 return -EINVAL; 1606 } else if (qid1) { 1607 qid = qid1; 1608 } else if (qid == 0) 1609 qid = dev->qdisc->handle; 1610 1611 /* Now qid is genuine qdisc handle consistent 1612 * both with parent and child. 1613 * 1614 * TC_H_MAJ(portid) still may be unspecified, complete it now. 1615 */ 1616 if (portid) 1617 portid = TC_H_MAKE(qid, portid); 1618 } else { 1619 if (qid == 0) 1620 qid = dev->qdisc->handle; 1621 } 1622 1623 /* OK. Locate qdisc */ 1624 q = qdisc_lookup(dev, qid); 1625 if (!q) 1626 return -ENOENT; 1627 1628 /* An check that it supports classes */ 1629 cops = q->ops->cl_ops; 1630 if (cops == NULL) 1631 return -EINVAL; 1632 1633 /* Now try to get class */ 1634 if (clid == 0) { 1635 if (portid == TC_H_ROOT) 1636 clid = qid; 1637 } else 1638 clid = TC_H_MAKE(qid, clid); 1639 1640 if (clid) 1641 cl = cops->get(q, clid); 1642 1643 if (cl == 0) { 1644 err = -ENOENT; 1645 if (n->nlmsg_type != RTM_NEWTCLASS || 1646 !(n->nlmsg_flags & NLM_F_CREATE)) 1647 goto out; 1648 } else { 1649 switch (n->nlmsg_type) { 1650 case RTM_NEWTCLASS: 1651 err = -EEXIST; 1652 if (n->nlmsg_flags & NLM_F_EXCL) 1653 goto out; 1654 break; 1655 case RTM_DELTCLASS: 1656 err = -EOPNOTSUPP; 1657 if (cops->delete) 1658 err = cops->delete(q, cl); 1659 if (err == 0) 1660 tclass_notify(net, skb, n, q, cl, 1661 RTM_DELTCLASS); 1662 goto out; 1663 case RTM_GETTCLASS: 1664 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS); 1665 goto out; 1666 default: 1667 err = -EINVAL; 1668 goto out; 1669 } 1670 } 1671 1672 new_cl = cl; 1673 err = -EOPNOTSUPP; 1674 if (cops->change) 1675 err = cops->change(q, clid, portid, tca, &new_cl); 1676 if (err == 0) 1677 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS); 1678 1679 out: 1680 if (cl) 1681 cops->put(q, cl); 1682 1683 return err; 1684 } 1685 1686 1687 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1688 unsigned long cl, 1689 u32 portid, u32 seq, u16 flags, int event) 1690 { 1691 struct tcmsg *tcm; 1692 struct nlmsghdr *nlh; 1693 unsigned char *b = skb_tail_pointer(skb); 1694 struct gnet_dump d; 1695 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1696 1697 cond_resched(); 1698 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1699 if (!nlh) 1700 goto out_nlmsg_trim; 1701 tcm = nlmsg_data(nlh); 1702 tcm->tcm_family = AF_UNSPEC; 1703 tcm->tcm__pad1 = 0; 1704 tcm->tcm__pad2 = 0; 1705 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1706 tcm->tcm_parent = q->handle; 1707 tcm->tcm_handle = q->handle; 1708 tcm->tcm_info = 0; 1709 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1710 goto nla_put_failure; 1711 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1712 goto nla_put_failure; 1713 1714 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1715 NULL, &d, TCA_PAD) < 0) 1716 goto nla_put_failure; 1717 1718 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1719 goto nla_put_failure; 1720 1721 if (gnet_stats_finish_copy(&d) < 0) 1722 goto nla_put_failure; 1723 1724 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1725 return skb->len; 1726 1727 out_nlmsg_trim: 1728 nla_put_failure: 1729 nlmsg_trim(skb, b); 1730 return -1; 1731 } 1732 1733 static int tclass_notify(struct net *net, struct sk_buff *oskb, 1734 struct nlmsghdr *n, struct Qdisc *q, 1735 unsigned long cl, int event) 1736 { 1737 struct sk_buff *skb; 1738 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1739 1740 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1741 if (!skb) 1742 return -ENOBUFS; 1743 1744 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) { 1745 kfree_skb(skb); 1746 return -EINVAL; 1747 } 1748 1749 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1750 n->nlmsg_flags & NLM_F_ECHO); 1751 } 1752 1753 struct qdisc_dump_args { 1754 struct qdisc_walker w; 1755 struct sk_buff *skb; 1756 struct netlink_callback *cb; 1757 }; 1758 1759 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, 1760 struct qdisc_walker *arg) 1761 { 1762 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 1763 1764 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid, 1765 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 1766 RTM_NEWTCLASS); 1767 } 1768 1769 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, 1770 struct tcmsg *tcm, struct netlink_callback *cb, 1771 int *t_p, int s_t) 1772 { 1773 struct qdisc_dump_args arg; 1774 1775 if (tc_qdisc_dump_ignore(q) || 1776 *t_p < s_t || !q->ops->cl_ops || 1777 (tcm->tcm_parent && 1778 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 1779 (*t_p)++; 1780 return 0; 1781 } 1782 if (*t_p > s_t) 1783 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 1784 arg.w.fn = qdisc_class_dump; 1785 arg.skb = skb; 1786 arg.cb = cb; 1787 arg.w.stop = 0; 1788 arg.w.skip = cb->args[1]; 1789 arg.w.count = 0; 1790 q->ops->cl_ops->walk(q, &arg.w); 1791 cb->args[1] = arg.w.count; 1792 if (arg.w.stop) 1793 return -1; 1794 (*t_p)++; 1795 return 0; 1796 } 1797 1798 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, 1799 struct tcmsg *tcm, struct netlink_callback *cb, 1800 int *t_p, int s_t) 1801 { 1802 struct Qdisc *q; 1803 int b; 1804 1805 if (!root) 1806 return 0; 1807 1808 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) 1809 return -1; 1810 1811 if (!qdisc_dev(root)) 1812 return 0; 1813 1814 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1815 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1816 return -1; 1817 } 1818 1819 return 0; 1820 } 1821 1822 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 1823 { 1824 struct tcmsg *tcm = nlmsg_data(cb->nlh); 1825 struct net *net = sock_net(skb->sk); 1826 struct netdev_queue *dev_queue; 1827 struct net_device *dev; 1828 int t, s_t; 1829 1830 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 1831 return 0; 1832 dev = dev_get_by_index(net, tcm->tcm_ifindex); 1833 if (!dev) 1834 return 0; 1835 1836 s_t = cb->args[0]; 1837 t = 0; 1838 1839 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) 1840 goto done; 1841 1842 dev_queue = dev_ingress_queue(dev); 1843 if (dev_queue && 1844 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, 1845 &t, s_t) < 0) 1846 goto done; 1847 1848 done: 1849 cb->args[0] = t; 1850 1851 dev_put(dev); 1852 return skb->len; 1853 } 1854 1855 /* Main classifier routine: scans classifier chain attached 1856 * to this qdisc, (optionally) tests for protocol and asks 1857 * specific classifiers. 1858 */ 1859 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp, 1860 struct tcf_result *res, bool compat_mode) 1861 { 1862 __be16 protocol = tc_skb_protocol(skb); 1863 #ifdef CONFIG_NET_CLS_ACT 1864 const struct tcf_proto *old_tp = tp; 1865 int limit = 0; 1866 1867 reclassify: 1868 #endif 1869 for (; tp; tp = rcu_dereference_bh(tp->next)) { 1870 int err; 1871 1872 if (tp->protocol != protocol && 1873 tp->protocol != htons(ETH_P_ALL)) 1874 continue; 1875 1876 err = tp->classify(skb, tp, res); 1877 #ifdef CONFIG_NET_CLS_ACT 1878 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) 1879 goto reset; 1880 #endif 1881 if (err >= 0) 1882 return err; 1883 } 1884 1885 return TC_ACT_UNSPEC; /* signal: continue lookup */ 1886 #ifdef CONFIG_NET_CLS_ACT 1887 reset: 1888 if (unlikely(limit++ >= MAX_REC_LOOP)) { 1889 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n", 1890 tp->q->ops->id, tp->prio & 0xffff, 1891 ntohs(tp->protocol)); 1892 return TC_ACT_SHOT; 1893 } 1894 1895 tp = old_tp; 1896 protocol = tc_skb_protocol(skb); 1897 goto reclassify; 1898 #endif 1899 } 1900 EXPORT_SYMBOL(tc_classify); 1901 1902 bool tcf_destroy(struct tcf_proto *tp, bool force) 1903 { 1904 if (tp->ops->destroy(tp, force)) { 1905 module_put(tp->ops->owner); 1906 kfree_rcu(tp, rcu); 1907 return true; 1908 } 1909 1910 return false; 1911 } 1912 1913 void tcf_destroy_chain(struct tcf_proto __rcu **fl) 1914 { 1915 struct tcf_proto *tp; 1916 1917 while ((tp = rtnl_dereference(*fl)) != NULL) { 1918 RCU_INIT_POINTER(*fl, tp->next); 1919 tcf_destroy(tp, true); 1920 } 1921 } 1922 EXPORT_SYMBOL(tcf_destroy_chain); 1923 1924 #ifdef CONFIG_PROC_FS 1925 static int psched_show(struct seq_file *seq, void *v) 1926 { 1927 seq_printf(seq, "%08x %08x %08x %08x\n", 1928 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), 1929 1000000, 1930 (u32)NSEC_PER_SEC / hrtimer_resolution); 1931 1932 return 0; 1933 } 1934 1935 static int psched_open(struct inode *inode, struct file *file) 1936 { 1937 return single_open(file, psched_show, NULL); 1938 } 1939 1940 static const struct file_operations psched_fops = { 1941 .owner = THIS_MODULE, 1942 .open = psched_open, 1943 .read = seq_read, 1944 .llseek = seq_lseek, 1945 .release = single_release, 1946 }; 1947 1948 static int __net_init psched_net_init(struct net *net) 1949 { 1950 struct proc_dir_entry *e; 1951 1952 e = proc_create("psched", 0, net->proc_net, &psched_fops); 1953 if (e == NULL) 1954 return -ENOMEM; 1955 1956 return 0; 1957 } 1958 1959 static void __net_exit psched_net_exit(struct net *net) 1960 { 1961 remove_proc_entry("psched", net->proc_net); 1962 } 1963 #else 1964 static int __net_init psched_net_init(struct net *net) 1965 { 1966 return 0; 1967 } 1968 1969 static void __net_exit psched_net_exit(struct net *net) 1970 { 1971 } 1972 #endif 1973 1974 static struct pernet_operations psched_net_ops = { 1975 .init = psched_net_init, 1976 .exit = psched_net_exit, 1977 }; 1978 1979 static int __init pktsched_init(void) 1980 { 1981 int err; 1982 1983 err = register_pernet_subsys(&psched_net_ops); 1984 if (err) { 1985 pr_err("pktsched_init: " 1986 "cannot initialize per netns operations\n"); 1987 return err; 1988 } 1989 1990 register_qdisc(&pfifo_fast_ops); 1991 register_qdisc(&pfifo_qdisc_ops); 1992 register_qdisc(&bfifo_qdisc_ops); 1993 register_qdisc(&pfifo_head_drop_qdisc_ops); 1994 register_qdisc(&mq_qdisc_ops); 1995 register_qdisc(&noqueue_qdisc_ops); 1996 1997 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL); 1998 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL); 1999 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, 2000 NULL); 2001 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL); 2002 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL); 2003 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, 2004 NULL); 2005 2006 return 0; 2007 } 2008 2009 subsys_initcall(pktsched_init); 2010